A multilevel computational and experimental study published in Frontiers in Cell and Developmental Biology has identified nitric oxide synthase 2, or NOS2, as a potential candidate molecular link between exposure to common air pollutants and biological pathways involved in depression. Conducted by researchers Huizheng Zhu, Bei Yin, Jingning Du, Siwei Zhang, Liyuan Su, and Shengyu Cai, the research investigated how seven major ambient pollutants might intersect with gene networks expressed in depressive states. By integrating network toxicology, transcriptomic co-expression analysis, ensemble machine learning, Mendelian randomization, single-cell virtual knockout modeling, molecular docking, molecular dynamics simulations, and in vitro cellular assays, the authors sought to uncover possible molecular intermediaries connecting air pollution exposure with depressive disorders.
Ambient air contamination represents a major worldwide public health burden, with the 2019 Global Burden of Disease study identifying air pollution as the fourth-ranked risk factor for global mortality, responsible for approximately 2.92 million deaths among males and 3.75 million deaths among females. Previous meta-analyses have reported positive associations between indoor air pollution exposure and depressive symptoms, including a risk ratio of 1.22 based on 12 studies comprising 61,217 participants. While clinical and epidemiological surveys have documented statistical correlations between exposure to pollutants and elevated depression risk, the underlying biological and molecular processes remain insufficiently characterized. The investigators designed their multilevel framework to address this specific knowledge gap, focusing on seven widespread chemical air pollutants: sulfur dioxide, nitric oxide, benzene, carbon monoxide, ozone, nitrogen dioxide, and toluene.
To map potential molecular targets of these environmental contaminants, the team interrogated the Target Net, Swiss Target Prediction, and SEA databases, compiling an initial set of 176 potential target genes associated with the seven pollutants. Concurrently, the researchers obtained peripheral blood gene expression profiles from two Gene Expression Omnibus datasets, GSE32280 and GSE76826. The GSE32280 cohort contributed 37 samples spanning healthy controls, major depressive disorder samples evaluated both before and after clinical treatment, and subsyndromal symptomatic depression samples. GSE76826 contributed 32 samples from healthy donors, individuals experiencing an active depressive episode, and patients in remission. Merging these cohorts yielded an integrated cohort of 20 healthy control profiles and 49 combined non-control profiles across the depression spectrum, which were harmonized using the empirical-Bayes ComBat algorithm to adjust for cohort and technical platform batch effects.
Using weighted gene co-expression network analysis across 2,341 genes with standard deviations exceeding 0.5, the authors identified 11 co-expression modules, among which the turquoise module showed the most significant statistical association with the non-control depression group, encompassing 760 candidate genes. When the 176 air pollution target genes were intersected with these 760 depression-associated module genes, nine shared candidate factors emerged: NOS2, HRH4, PNMT, GRM2, PTGES, ABCG2, CA3, F2, and CA7. Functional enrichment analysis indicated that these overlapping target genes were enriched in Gene Ontology terms relating to carboxylic acid transport, organic acid transport, the external surface of the plasma membrane, nuclear envelope lumen, carbonate dehydratase activity, and hydrolyase activity. KEGG pathway analysis further highlighted significant enrichment in neuroactive ligand-receptor interactions and nitrogen metabolism pathways.
To evaluate the discriminatory power of these nine candidate genes, the investigators developed machine learning binary classification pipelines across 87 eligible algorithmic configurations. The combined GSE32280 and GSE76826 dataset served as the training platform, while two separate clinical cohorts, GSE44593 with 28 samples and GSE53987 with 35 samples, were reserved for independent external validation. An ensemble pipeline pairing component-wise gradient boosting with ridge-penalized logistic regression emerged as the highest-performing architecture, achieving an average area under the receiver-operating-characteristic curve of 0.692 across datasets. When the model screened individual predictors, five genes were prioritized: CA7, F2, GRM2, NOS2, and PTGES. Among these five factors, NOS2 demonstrated the highest individual AUC value at 0.659, and differential expression comparisons confirmed that NOS2 was significantly upregulated in depression samples compared with healthy controls.
To evaluate whether this correlation reflected a potential genetic causal relationship, the authors conducted a two-sample Mendelian randomization analysis using summary statistics. Genetic instruments for NOS2 whole-blood gene expression were derived from the eQTLGen Consortium, encompassing sample sizes ranging from 20,989 to 29,285 predominantly European-ancestry participants, while outcome associations were obtained from an independent depression genome-wide association study comprising 27,568 cases and 457,030 controls. Eight independent cis-eQTL single-nucleotide polymorphisms within 500 kilobases of NOS2 met the genome-wide significance threshold and passed instrument strength filtering with F statistics exceeding 10. The primary inverse-variance weighted estimator indicated a statistically significant positive genetic association between higher genetically predicted NOS2 expression and depression, yielding an odds ratio of 1.004 with a 95 percent confidence interval of 1.000 to 1.008. Sensitivity analyses, including MR-Egger regression, weighted-median estimation, Cochran Q test, and MR-PRESSO testing, detected no evidence of horizontal pleiotropy or influential single outlier variants.
The biological context of NOS2 expression in neural tissue was examined using single-nucleus RNA sequencing data from the human dorsolateral prefrontal cortex in dataset GSE201687, comprising 14 biological specimens divided between depression-like and healthy control donors. Across nine major annotated brain cell types, NOS2 expression was preferentially concentrated within microglia, which were identified using canonical markers such as CX3CR1 and TMEM119. In silico virtual knockout modeling using the scTenifoldKnk algorithm revealed that perturbing NOS2 in microglia induced significant expression shifts across multiple downstream genes, including NHSL1, BAIAP2L2, VCAN, CPZ, SERPINE2, TGM2, and STK32A. Functional pathway analysis of the resulting differentially regulated genes demonstrated pronounced enrichment in biological processes governing cellular responses to oxidative stress, toxic substance response, cellular oxidant detoxification, and inflammatory cascades, including the tumor necrosis factor, NF-kappa-B, and interleukin 17 pathways.
Structural interactions between the seven air pollutants and the human inducible nitric oxide synthase protein were assessed computationally through blind molecular docking against the crystallographic structure of the NOS2 oxygenase domain from PDB entry 2NSI. Docking calculations via AutoDock Vina predicted negative binding free energies for all seven chemical entities, with sulfur dioxide yielding negative 2.5 kilocalories per mole, nitric oxide negative 2.9, carbon monoxide negative 1.9, nitrogen dioxide negative 3.5, ozone negative 3.5, benzene negative 5.4, and toluene negative 5.7. Subsequent 100-nanosecond all-atom molecular dynamics simulations using Gromacs 2024.4 compared the dynamic stability of NOS2 bound to benzene versus toluene. Root-mean-square deviation and fluctuation curves confirmed structural integrity, while free energy landscape analyses revealed a single, concentrated low-energy basin for the NOS2-toluene complex compared to multiple dispersed basins for benzene. Furthermore, MM/GBSA calculations demonstrated an average binding free energy of negative 12.75 kilocalories per mole for toluene, compared to negative 9.27 kilocalories per mole for benzene.
To evaluate these computational predictions in a living cellular system, the researchers conducted in vitro experiments using the BV2 mouse microglial cell line exposed to nominal toluene concentrations of 20, 50, and 100 micromolar. Cell viability assessed by CCK-8 assays and cytotoxicity measured by lactate dehydrogenase release showed no significant damage after 24 hours of 20 or 50 micromolar exposure, whereas 100 micromolar toluene significantly reduced viability and elevated LDH leakage at both 24 and 48 hours. When cells were incubated with 100 micromolar toluene for 24 hours, quantitative RT-PCR and Western blotting revealed significant upregulation of NOS2 mRNA and protein levels compared with controls. This molecular response coincided with marked increases in stable nitrite accumulation, elevated intracellular reactive oxygen species detected by flow cytometry, and increased formation of 3-nitrotyrosine detected by immunofluorescence, accompanied by significant transcriptional upregulation of inflammatory markers including TNF-alpha, IL-1beta, IL-6, and NF-kappa-B p65.
The study authors emphasize that these multilevel observations provide hypothesis-generating clues rather than definitive clinical proof that NOS2 functions as a direct causal mediator of air pollution-induced depression. Important methodological limitations temper the findings. The in vitro cellular responses occurred under exposure conditions that caused significant cytotoxicity, meaning that observed inflammatory and oxidative signals could be partly confounded by cell death pathways. Furthermore, the public transcriptomic datasets utilized reflect diverse sample collections and technical platforms, the molecular docking models rely on algorithms designed primarily for organic molecules rather than small inorganic gases, and the genetic causal estimates from Mendelian randomization, while statistically significant, reflect modest odds ratios and depend on specific instrument assumptions without colocalization testing. The authors conclude that further validation in vivo, using animal models and targeted genetic knockdown or pharmacological inhibition of NOS2 across subcytotoxic exposure ranges, will be required to confirm these biological pathways.
Subject of Research: Air pollution exposure and depression molecular mechanisms
Article Title: NOS2 as a candidate molecular link between air pollutants and depression: an integrated computational and experimental study
Article References: Zhu, H., Yin, B., Du, J., Zhang, S., Su, L., & Cai, S. (2026). NOS2 as a candidate molecular link between air pollutants and depression: an integrated computational and experimental study. Frontiers in Cell and Developmental Biology, 14, Article 1962251. https://doi.org/10.3389/fcell.2026.1962251
Image Credits: AI Generated
DOI: 10.3389/fcell.2026.1962251
Keywords: air pollution, depression, nitric oxide synthase 2, machine learning, Mendelian randomization, molecular docking, neuroinflammation, oxidative stress, NOS2, candidate, molecular, link
News Source: Glenn Wilkins. (November 8, 2026). Computational and Cellular Study Links Air Pollutants to NOS2 in Depression. Scienmag.



