Immune Impacts of Fire Smoke Exposure

Abstract

Fire smoke exposure has become a global health concern and is associated with increased morbidity and mortality. There is a lack of understanding of the specific immune mechanisms involved in smoke exposure with preventative and targeted interventions needed. After exposure to fire smoke, which includes PM_2.5, toxic metals and per- and polyfluoroalkyl substances (PFAS), epidemiology-based studies demonstrate increases in respiratory (e.g. asthma exacerbations), cardiac (e.g. myocardial infarctions, arrhythmias), neurologic (e.g. stroke) and pregnancy-related (e.g. low birthweight, premature births) outcomes. However, mechanistic studies exploring how smoke exposure disrupts cellular homeostasis are lacking. Therefore, we collected blood from smoke-exposed individuals (n=31) and age- and sex-matched non smoke-exposed controls (n=29) and investigated these complex interactions using a single-cell exposomic approach based on both methylation and mass cytometry outcomes. Overall, our data demonstrated a strong association between smoke exposure and methylation at 133 disease-relevant gene loci, and immunophenotyping revealed increased homing and activation biomarkers. We developed an application of mass cytometry to analyze single cell/metal binding and found, for example, increased levels of mercury in dead cells and cadmium in the live and dead cell populations. Moreover, mercury levels were associated with years of smoke exposure. Several epigenetic sites across multiple chromosomes were associated with individual toxic metal isotopes in single immune cells. Our methods for detecting the effect of smoke exposure at the single-cell level and study results may help to determine the timing of exposures and identify specific molecular targets that could be modified to prevent and manage exposure to smoke.

1. Introduction

Exposure to fire smoke has become a timely global issue, impacting nearly every part of the world throughout each season¹. In 2020 alone, almost 25 million people were exposed to smoke with a PM_2.5 above 100 μg/m^{3 2}, resulting in adverse health outcomes³. Fire smoke, comprised of particulate matter, gases, materials from buildings (including per- and polyfluoroalkyl substances (PFAS) compounds), toxic metals, and carcinogenic compounds, is hazardous to the public. In 2022 the International Agency for Research on Cancer classified firefighting as a Group 1 human carcinogen, which is the highest carcinogenic hazard category, based on science to date⁴. Epigenetic mechanisms are thought to be involved in carcinogenesis, including DNA methylation^5,6, miRNA regulation^7,8, and histone posttranslational modifications⁹. Smoke from fires may contain different constituents based on their chemistry¹⁰ and often contains perfluorinated compounds and toxic metals, so we focused on these two common components of smoke^11–13 since they are measurable and can be quantified in vivo.

PFAS are a group of more than 4000 fluorinated compounds that do not degrade, i.e. “forever chemicals”¹⁴. Epidemiological studies have shown an association between both perfluorooctanoic acid (PFOA) and perfluorooctane sulfonate (PFOS) exposure and epigenetic changes in both adults¹⁵ and birth cohorts^16,17. However, how PFAS in fire smoke results in specific mechanistic changes in the human immune system is not known; our study is the first to begin to dissect and define the molecular pathways of smoke associated PFAS-mediated immune disruption in humans.

Toxic metal exposure (e.g. lead, mercury and cadmium) is detrimental to health, even at low concentrations. Given that toxic metals occur naturally in soils and plants, and are also present in manmade structures, smoke exposure from both wildfires and structural fires is of concern. Here, we developed methods to investigate toxic metal effects on single immune cells using mass cytometry while simultaneously determining the immunophenotype. Studies to discern how toxic metals impact the immune system at a single-cell resolution have heretofore not been done. Our approach using mass cytometry provides insights into how toxic metals affect immune cells in humans exposed to fire smoke and could apply to toxic metal exposures more broadly, outside of fire smoke.

To advance our mechanistic understanding of smoke exposure and to ultimately develop preventative and therapeutic interventions, we studied the blood of smoke-exposed (SE) and non-smoke exposed (non-SE) individuals at the single-cell level using a multi-omics approach. We hypothesized that smoke exposure would be associated with immune modulation, epigenetic modifications of key immunoregulatory genes, and immune dysfunction. To test this hypothesis, we used innovative single-cell omic techniques (epigenetic assays and mass cytometry using 27 cellular markers and 21 cell-coupled toxic metal isotopes) to study immune cells in human peripheral blood and used rigorous integrative bioinformatic analytical tools. Overall, we found previously unidentified pathways associated with smoke exposure, including differentially methylated CpG sites from 133 gene loci enriched for a wide variety of functions, increased memory CD8+ T cells and elevated activation and chemokine receptor biomarkers in multiple cell types. In addition, our data demonstrate that it is possible to identify toxic metal isotopes bound to immune cells at the single-cell level using mass cytometry.

2. Results

2.1. Cohort characteristics

We recruited two cohorts of participants: smoke-exposed (SE) adults (n=31, including both firefighters and smoke-exposed civilians) and non-smoke exposed (non-SE) adults (n=29). Cohorts were age- and sex-matched and there were no significant differences between the cohorts for age, sex, race/ethnicity, body-mass index or socioeconomic status (all middle class participants). Table 1 provides details on the demographic characteristics of the subjects, including further details on smoke exposure history (see also Supplementary Table 1). We had rigorous eligibility criteria for our cohorts. None of the participants were taking immunomodulatory drugs at or before the timing of the blood draw, which was collected within a month after smoke exposure. No sample was taken from an individual with an acute or chronic condition, except one participant had a history of fully treated, non-active thyroid disease. None of the participants were exposed to second-hand tobacco smoke or had a history of smoking.

Table 1.

Sample Demographics

	Smoke Exposed	Non-Smoke Exposed
Sample Size (n):	31	29
Age:	p-value = 0.06
Mean (SD)	40.64 (10.39)	35.34 (11.39)
Median	38	33
Sex:	p-value = 0.46
Female	9 (29.0%)	12 (41.3%)
Male	22 (70.9%)	17 (58.6%)
Race/Ethnicity:	p-value = 0.92
White	17 (56.6%)	18 (62.0%)
Asian	7 (23.3%)	7 (24.1%)
Black	2 (6.6%)	1 (3.4%)
Multi-racial	4 (13.3%)	3 (10.3%)
SES-middle:	31 (100%)	29 (100%)
BMI:	p-value =0.08
Mean (SD)	27.44 (3.4)	25.27 (5.1)
Median	26.53	24.2
Years smoke exposure:
Mean (SD)	9.3 (8.6)	N/A
Median	7	N/A

To identify molecular changes driven by these environmental toxins, we designed a multi-omics study with integrative analysis of epigenetic DNA methylation and single-cell immunophenotyping including toxic metal quantification using mass-cytometry (Figure 1).

Figure 1.

Schematic workflow of the multi-omics-based study of SE vs non-SE cohorts. Illustrations were created using BioRender.com with some image elements sourced from Adobe Stock (Adobe Education License). Created in BioRender. Kaushik, A. (2025) https://BioRender.com/u30w9b9.

2.2. Differential methylation suggests immune dysregulation

Methylation sequencing of 37,258 CpG sites specifically targeted for asthma and allergy was performed using published methods via Illumina Infinium on PBMCs from SE (n=31) and non-SE (n=29) cohorts¹⁸ (Supplementary Figure 1A–B). These sites covered a broad region of the genome (GRCh37), i.e., CpG Islands, promoter regions, intergenic regions, introns, exons, 3’UTRs, and 5’ UTRs. The raw data for the corresponding methylation profiles in these regions were processed and checked for quality (see Methods). M-values per site were used for further methylation analysis. The Principal Component Analysis (PCA) revealed associations between the CpG methylation profiles and Batch/Race of the study participants (Supplementary Figure 2). Therefore, both race and batch number were statistically adjusted during differential methylation analysis using the limma method. The analysis identified 94 CpG sites differentially methylated between SE vs. non-SE cohorts (log2FC > 0.50 and q < 0.20) (Figure 2A). These 94 CpG regions were associated with 133 genes (see Methods & Supplementary Table 2). Among these 94 CpG regions, 30 (31.90%) were associated with one gene and 64 (68.00%) were associated with two genes (Supplementary Figure 3A). The majority of these regions lied within a −5kb to +50kb distance to the Transcription Start Site (TSS), which indicates the potential regulatory nature of these epigenetic modifications (Supplementary Figure 3B). Based on the M-values of differentially methylated sites, the Partial Least Squares Discriminant Analysis (PLS-DA) based clustering analysis of the SE and non-SE cohorts further showed the distinct clustering of the SE group (Figure 2B). Here, we also observed a higher silhouette width in the SE group, which suggests that genetic profiles are clustered together more among SE group as compared to the non-SE.

Figure 2.

(A) The volcano plot representing differentially methylated sites. The differential methylation status of each site was calculated across SE (n = 31) versus non-SE (n = 29; control) cohorts. Here, red dots represent significant sites (adj. p-value (q) < 0.20 & absolute logFC > 0.50). (B) Scatter plot of PLS-DA results showing the separation of samples from SE and non-SE cohorts based on differentially methylated CpG sites. Each point represents an individual sample, plotted according to the first two PLS-DA dimensions (DIM-1 and DIM-2), which capture the largest variance between the groups. The plot illustrates the distinct methylation profiles associated with SE and non-SE cohorts. The plot also indicates that SE (silhouette width = 0.32) may possess similar epigenetic profile as compared to non-SE (silhouette width = 0.29) and therefore are clustered close to each other. (C) Significantly enriched pathways predicted from the genes associated with differentially methylated CpG sites. Metascape webserver was used to predict the significantly enriched pathways (p < 0.05), with default parameters. (D) Disease enrichment analysis results from the genes associated with differentially methylated CpG sites. STRING webserver was used for disease enrichment analysis. (E) The epigenetic profile in SE individuals (n = 23) with respect to their total PFOA and total PFOS abundance in their blood. For visualization, only the topmost randomly selected correlations with Total PFOA and Total PFOS were used. Data are presented as the mean ± SE. Plots display the correlation coefficient (ρ) and linear regression line with a 95% confidence interval shading. (F) Circos plot representing the edge list of significant correlation predicted between epigenetic sites and total PFOA/PFOS levels in SE individuals (n = 23). For enrichment analysis using metascape, p-values were calculated using the hypergeometric test and adjusted for multiple comparisons using the Benjamini–Hochberg method.

Furthermore, the 133 genes associated with the differentially methylated regions included several histone proteins, transcription factors, zinc finger proteins, and genes related to immunological pathways, i.e., FOXA3, HLA-B, IL-6, IL-1 and CXCR5 (Supplementary Table 2). The pathway enrichment analysis of the 133 genes using Metascape highlighted 20 enriched terms (p < 0.05; Figure 2C), which included pathways related to chromatin modification, cytokine signaling, and regulation of immune response.

To explore biological pathways affected by fire exposure, we identified 629 additional differentially methylated regions with q < 0.005 without including fold change as a filtration criterion. These 629 regions corresponded to 733 genes (Supplementary Data File 01), wherein 467 (74.24%) regions were associated with two genes and 153 (24.32%) regions were associated with only one gene (Supplementary Figure 4A–B). The genomic region-based pathway enrichment analysis suggests that these genes were significantly enriched in pathways related to cell activation, interleukin signaling, and regulation of immune responses, including TCR signaling and T-cell activation. (Supplementary Figure 4C). Most of these pathways are involved in gene regulation and assist in maintaining cell function. Phenotype enrichment revealed pathways related to asthma, carcinoma, liver disease and immune system disorders (Figure 2D). We then identified differentially methylated CpG sites that are significantly associated with years of fire smoke exposure in a subset of the SE group (i.e. active-duty firefighters). We compared the genomic regions of the 77 (8.20%) differentially methylated CpG sites versus years of firefighting exposure. Most of the CpG sites (n = 44) were associated with two genes and often located close to the transcription Start Site (Supplementary Figure 5A–B). Metascape enrichment analysis of 103 genes found several enriched immunological pathways, such as leukocyte differentiation, transcriptional mis-regulation in cancer, histone acetylation and B-cell activation (Supplementary Figure 5C). Moreover, the disease gene enrichment analysis revealed gene sets related to clinical outcomes, such as allergic reactions and colorectal cancer susceptibility (Supplementary Figure 5D).

To investigate the epigenetic modification associated with measurable smoke constituents (i.e. PFAS level in the firefighters serum only; see Methods for further explanation) and given that we measured PFAS compounds from blood drawn at the same timepoint, we performed an additional Bayesian correlation analysis (see Methods) to predict those epigenetic modifications that were strongly correlated (absolute r > 0.60) with both total PFOA and PFOS levels (Figure 2E; Supplementary Data File 02 & file-03). We observed 213 epigenetic regions strongly correlated with total PFOS levels, whereas only 29 regions were strongly correlated with total PFOA levels (Figure 2F). For PFOS, the 213 regions were associated with 281 genes (Supplementary Data File 04) that were enriched with 20 significant pathway terms (p < 0.05; Supplementary Figure 6C). The 20 terms included several pathways related to signaling pathways, transcriptional regulation, Th₁₇-cell differentiation, hemopoiesis, and transcriptional mis-regulation in cancer. Similarly, for PFOA, 29 epigenetic modifications were associated with 47 genes (Supplementary Data File 05). These genes, when subjected to Metascape, revealed 10 significant pathway terms (p <0.05), including T-cell receptor signaling and protein ubiquitination (Supplementary Figure 7C).

2.3. Differences in activation markers and chemokine receptors

Mass Cytometry (Cytometry by Time of Flight or CyTOF) was performed on the Peripheral Blood Mononuclear Cells (PBMCs) from both cohorts. We profiled the immune cell subsets using a panel of 27 cell surface protein markers which included both lineage and functional markers (Supplementary Table S3). Using machine learning and unsupervised clustering we identified 10 unique cell clusters (Supplementary Figure 8A). The comparative analysis of cell clusters revealed no statistically significant differences (p > 0.05) in their total cell count proportions among the SE vs non-SE cohorts (Supplementary Figure 8B). We then manually gated five canonical immune cell types: CD8+ T cells, CD4+ T cells, NK cells, CD11c+ cells and CD19+ B cells (Supplementary Figures 9–11). The proportion analysis reveals that the gated cell types were in expected proportions among the total live cells, with the majority being CD4+ T cells (Supplementary Figure 9B).

However, given prior evidence that other environmental exposures, such as cigarette smoke impair T cell functioning and proliferation¹⁹, we further examined naive and memory CD8+ T cells. There were no differences in the proportion of total, naïve or memory CD8+ T cells when comparing SE versus non-SE cohorts (See Supplementary Figure 12A–B). However, there was a significantly greater expression of CXCR3 in memory CD8+ T cells of the SE vs non-SE cohort (Supplementary Figure 12C). To gain further insight we used unsupervised clustering of the manually gated total CD8+ population (Supplementary Figure 12D–E). The analysis highlighted two clusters (#2 and #9) that comprised the terminal effector memory (TEMRA) cell population (CCR7– CD45RA+) and represented 15.50% of the total CD8+ T cells. CCR6^hi cells in the TEMRA CD8+ T cell cluster #9 were found to be significantly elevated (q = 0.14), in the SE subjects compared to non-SE subjects (Supplementary Figure 9F). There was a higher proportion of CXCR3^hi (q=0.12) and PD-1^hi (q=0.20) cells from memory CD8+ T cell cluster #1 (Supplementary Figure 12G) and a higher proportion of CRTH2^hi (q=0.14) and CCR6^hi (q=0.09) cells from memory CD8+ T cell cluster #6 in SE vs non-SE (Supplementary Figure 12H). We also observed elevated levels of CD28 expression (q = 0.12) on memory CD8+ T cell cluster #7 in SE vs non-SE (Supplementary Figure 12I). Memory CD8+ T cell clusters #5 and #7 showed a lower proportion of CCR7^hi (q = 0.04 and q = 0.05) cells, and cluster #1 showed fewer CD154^hi cells (q=0.19) in the SE group as compared to non-SE group (Supplementary Figure 12J). Naïve CD8+ T cells constituted 48.50% of the total CD8+ T cells and there were no differences in cell count proportion or any marker expression when comparing SE versus non-SE groups.

Similarly, we gated naive and memory CD4+ T cells and then compared their proportion across the SE versus non-SE cohorts (Figure 3A). The manually gated memory CD4+ cells showed an elevated level of CCR6^hi cells (q=0.08) and a reduced level of CCR7^hi cells (q=0.15), along with the lower median signal intensity (MSI) of CCR7 in SE cohort (Figure 3B). We also performed unsupervised clustering on manually gated CD4+ T cells and identified 15 cell clusters (Figure 3C & Supplementary Figure 13A). For the total live CD4+ T cells, we observed 58.7% naïve CD4+ T cells, 0.20% TEMRA cells and 41.20% memory CD4+ T cells. Among these 15 unsupervised CD4+ T cell clusters (Figure 3C), memory CD4+ T cell cluster #4 had significantly elevated levels of CCR6^hi cells (q = 0.12) and lower expression of CCR4^hi cells (q = 0.04) and CCR7^hi cells (q = 0.09) in the SE subjects (Figure 3D & Supplementary Figure 13B). In memory CD4+ T cell cluster #6 (12.80% of total CD4+ T cells), there were a significantly elevated number of CCR6^hi cells (q = 0.14) and fewer CXCR5^hi the cells (q = 0.22) in SE versus non-SE (Figure 3E). There was also an elevated expression of CCR6 (q=0.14) in cluster #10, and significantly lower proportion of CRTH2^hi cells in clusters #4 (q = 0.12) and #12 (q = 0.19; Supplementary Figure 13C–D). In Cluster #8 there were fewer CCR4^hi cells (q=0.12) and a higher expression of CXCR3 (q=0.14) in SE cohort (Figure 3F).

Figure 3. CD4+ T cell differential analysis.

(A) Comparison of manually gated memory and naïve CD4+ T cells in SE (n=31) versus non-SE (n=29) groups. The p-value for the interaction term from the generalized linear model tests whether the difference in cell count proportions between CD4+ memory T cells and naïve T cells vary significantly between the SE and non-SE groups. This indicates whether the effect of cell type on cell count proportion is dependent on group status. (B) Boxplot representing significant differences in the proportion of CCR6^hi, CCR7^hi cells and CCR7 median expression in manually gated memory CD4+ T cells (SE versus non-SE) with respect to the parent cell population (CD3+ T cells). (C) Heatmap representing 15 unsupervised CD4+ T cell clusters. The Y-axis (left) represents cluster number and Y-axis (right) represents cell proportion with respect to total CD4+ T cells. (D-F) Boxplot representing differential profile of CCR4^hi, CCR6^hi, CXCR5^hi cells and median expression of CXCR3 across SE versus non-SE groups in cluster #4, #6 and #8. In the boxplots, medians are shown, and the “hinges” represent the first and third quartile. The whiskers are the smallest and largest values after exclusion of outliers (greater than the 75th percentile plus 1.5 times the interquartile range (IQR), or less than 25th percentile minus 1.5 times the IQR). q-values were calculated using the MAST algorithm implemented in the Seurat R package, accounting for batch effects as a latent variable. For MSI, the p-value was calculated using generalized linear modeling, adjusted for batches. Multiple hypothesis testing corrections were applied using the false discovery rate (FDR) method. For manually gated cell subsets, unadjusted p-values were used due to the limited number of hypotheses tested. SE (n = 31) and non-SE (n = 29; control) represent individual samples; no technical replicates included in analysis. MSI: Median Signal Intensity.

Manual gating of the CD19+ B cells revealed elevated total CD19+ B cells (p= 0.10; Supplementary Fig. 14A), naïve and memory (Supplementary Figure 14B) B cells in SE versus non-SE cohort. We also performed unsupervised clustering of CD19+ B cells and analyzed the resulting clusters (Supplementary Figure 14C). HLA-DR^hiCD27+CD19+ B cells Cluster #6 showed an elevated expression of CCR5 (q=0.16) in the SE cohort, but lower proportion of PD-1^hi (q=0.18) and CXCR5^hi cells (q=0.22) (Supplementary Figure 14D).

Unsupervised clustering of manually gated CD11c+ myeloid cells (Supplementary Figure 15A) revealed differences in the proportion of three CD11c+ cell clusters- #1 (p=0.09), #4 (p=0.02) and #7 (p=0.02) in SE cohort (Supplementary Figure 15B), which represented 68.4% of total manually gated CD11c+ cells. Cluster #1 and #7 were CCR5+ and elevated in SE vs non-SE. CCR5+ clusters #4 and #5 also had elevated levels of CCR5^hi cells (q = 0.04 & q=0.12) in the SE vs non-SE cohorts (Supplementary Figure 15C). However, we observed a lower proportion of CD28^hi (q = 0.18) and CD27^hi (q = 0.21) cells in clusters #1 (Supplementary Figure 15D).

NK cells subsets CD56^dimCD16⁺ and CD56^bright CD16^Low were also gated (Supplementary Figure 16A–B). CD56^dimCD16⁺ NK cells constituted 93.3% of the total NK cells, and CCR6^hi (q=0.09) and CCR5^hi (q=0.20) cells were higher in SE cohort, whereas CD56^bright CD16^Low NK cell had lower proportions of CCR7^hi (q=0.18) and PD-1^hi (q=0.06) cells.

2.4. Toxic metals are bound to single immune cells

Using a novel approach to detect toxic metals in PBMCs using mass cytometry (Figure 1; see Methods), we measured the toxic metals that are bound to single cells by recording open channels between atomic masses 75 and 209 in which no metal-tagged antibodies were stained. Specifically, Bromide (Br), cadmium (Cd), antimony (Sb), tantalum (Ta), tungsten (W), mercury (Hg), and lead (Pb) isotopes were not used in the metal-tagged antibody staining but were recorded during cell acquisition. This approach not only allowed us to identify immune cell types at the single-cell resolution (using mass cytometry with standard approach), but also simultaneously quantify toxic metal content for each cell.

We first investigated differences in toxic metal levels in the total dead cell population between the SE and non-SE cohorts. There were significantly increased levels of two mercury isotopes (¹⁹⁸Hg and ¹⁹⁶Hg) and two cadmium isotopes (¹¹⁴Cd and ¹¹²Cd) in the dead immune cells of SE as compared to the non-SE cohorts (Supplementary Figure 17A), suggesting that the SE group was possibly exposed to more Hg and Cd and that this level of exposure was highly toxic to the cells. Furthermore, in the SE cohort the dead cell population bound to ¹⁹⁸Hg and ¹⁹⁶Hg were significantly correlated with years of smoke exposure (Supplementary Figure 17B), suggesting a cumulative impact of mercury. We also analyzed the toxic metals bound to the total live cell populations and the various immune cell types. We found that most of the toxic metal isotopes (e.g., ²⁰⁶Pb, ¹⁹⁶Hg, ¹⁰⁴W) were detected in a large set of live cells (Figure 4A), whereas cadmium isotopes (¹⁰⁶Cd, ¹⁰⁸Cd, ¹¹⁰Cd, ¹¹²Cd, ¹¹⁴Cd) were detected in a low number of live cells for both SE and non-SE groups. Among the toxic metals, there was a significant increase in the level of Cd, Ta and Br isotopes bound to live cells in the SE versus non-SE cohorts (Figure 4B). Specifically, there were significantly greater isotopes of ¹⁰⁶Cd (q = 0.10), ¹⁰⁸Cd (q = 0.04) and ¹¹⁰Cd (q = 0.16), ⁷⁹Br (q = 0.02) and ¹⁸⁰Ta (q=0.18) in the live cells of SE vs non-SE cohorts. Furthermore, analysis of manually gated naïve and memory CD8+ T cells subsets show elevated levels of the isotope ¹¹⁰Cd (q = 0.36 and q = 0.10), respectively. The NK cell subsets had an elevated level of ⁷⁹Br (q = 0.11 and q = 0.13).

Figure 4. Single-cell Toxic metal profiling.

(A) Distribution of different metals, as stacked density plots, in live single cells. (B-F) Boxplot representing toxic metals that are detected significantly differently across SE versus non-SE in manually gated CD8+ T cells, and NK cells subsets. In the boxplots, medians are shown, and the “hinges” represent the first and third quartile. The whiskers are the smallest and largest values after exclusion of outliers (greater than the 75th percentile plus 1.5 times the interquartile range (IQR), or less than 25th percentile minus 1.5 times the IQR). q-values were calculated using the MAST algorithm implemented in the Seurat R package, accounting for batch effects as a latent variable. Multiple hypothesis testing corrections were applied using the false discovery rate (FDR) method. SE (n = 31) and non-SE (n = 29; control) represent individual samples; no technical replicates included in analysis.

2.5. Toxic metal exposure alters the epigenomic profile

To evaluate whether the toxic metal exposure may alter the epigenomic profiles of different immune cell types in SE versus non-SE cohorts, we performed a differential correlation analysis (r > 0.60 & FDR < 0.10; see Methods) between toxic metal abundance and DNA methylation levels. There were significantly different associations between CpG methylation and toxic metal abundance in the SE vs non-SE cohorts (Figure 5A). We found five epigenetic sites associated with toxic metals in the total live cells of the SE vs non-SE cohort. To gain further insight and test whether there was an association between metal abundance within distinct immune cell types (e.g., memory CD4+ T cells) and DNA methylation, we performed differential correlation analysis with respect to each cell type. Overall, 25 unique CpG sites were significantly differentially associated with eight cell subsets (Figure 5B). For instance, ¹⁸⁶W isotope was correlated with CpG sites in SE vs non-SE across four different cell subsets, whereas in memory CD8+ T cells ²⁰⁹Bi isotope was associated with five CpG sites. These significantly associated CpG sites are mainly located in chromosomes 1, 3, 6 and 17 (Figure 5C). Interestingly, the ¹⁹⁸Hg isotope bound to total live cells was also significantly associated with years of smoke exposure in SE participants (Supplementary Figure 18). Our data analysis indicated that the immunotoxicity of metal ions may occur in a cell-type specific manner in part, by modulating the cell’s epigenome.

Figure 5. Epigenetic modifications versus toxic metal profiling.

(A) Scatter plot representing the significant differential correlation between methylation profile (b-values) of a given CpG site and cellular toxic metal (²⁰⁶Pb) abundance in total live cells and manually gated cell types. Data are presented as the mean ± SE. Plots display the correlation coefficient (ρ) and linear regression line with a 95% confidence interval shading. (B) Bar plots to show number of significant CpG sites differentially correlating (SE versus non-SE) with toxic metal abundance profile for a given cell type. (C) Manhattan plot representing chromosomal number of CpG sites (X-axis) and their p-value significance (-log₁₀ p-value; Y-axis) representing the significant differential correlation between methylation profile (b-value) and cellular toxic metal abundance in total live cells and manually gated cell type. The p-values were calculated using generalized linear modeling including an interaction term, with batch adjustment. SE (n = 31) and non-SE (n = 29; control) represent individual samples; no technical replicates included in analysis.

3. Discussion

Given the proliferation of fire smoke exposure throughout the world, we investigated the mechanistic interactions between exposure to fire smoke and the immune system in humans. We applied a novel mass cytometry-based technique to measure individual toxic metal isotopes bound to single immune cells. Our findings identify molecular targets related to smoke-associated exposures. The SE cohort was represented by healthy individuals who were exposed to wildfire smoke (either through work or through public exposure); there are many people in the public exposed frequently to wildfire smoke so understanding how wildfire smoke affects the immune system and associated genes is an important first step towards developing better interventions to improve human health after fire smoke exposure.

We hypothesized that smoke exposure is linked to elevated levels of toxic metals and PFAS in the blood, leading to dysregulated immune cell expression profiles and increased methylation of key immunoregulatory genes. Overall, we found that the SE vs non-SE cohorts had a distinct methylation profile involving multiple chromosomes over 133 genes. The differentially methylated sites were enriched for a wide variety of functions, from chromatin modifications to protein stabilization. The toxic metals we measured had substantially more associations with methylation than PFAS levels, suggesting that they independently impact the epigenome via specific DNA methylation exposure signatures. This is consistent with a study by Goodrich, et al (2021)¹⁹, in which 50 firefighters had blood drawn as new recruits and then had a follow-up blood draw approximately two years later. DNA methylation analysis found 60 CpG sites with at least a 5% methylation difference after two years of firefighting and the genes were enriched for immune pathways related to cancer, cell signaling and transcription regulation.

Here, we determined that total PFOS was associated with more epigenetic modifications than total PFOA. Using a stringent Bayesian significance criterion of absolute r > 0.60, BF > 2, pd > 90%, and ROPE < 0.10 (see Methods) we determined that 213 epigenetic regions (0.57%) were strongly correlated with total PFOS levels, whereas only 29 regions (0.07%) were strongly correlated with total PFOA levels. Overall, our findings suggest that PFOS is relatively more toxic to the immune system than PFOA due to the stronger correlation with PFOS levels, rather than the PFOA levels. Involved pathways also differed between PFOS (cytokine signaling pathways, Th17 Cell differentiation and transcriptional dysregulation in cancer) versus PFOA (mainly T cell receptor signaling, glycerolipid metabolism, and death receptor signaling pathways). These findings are consistent with prior findings. For example, a gestational PFAS exposure study performed methylation analyses at birth and age 12 and found the same patterns at both ages with specific non-overlapping CpG sites for PFOS (n=2), PFOA (n=12), PFHxS (n=8), but the most for PFNA (n=413)²⁰.

We determined that there was greater expression of several biomarkers in memory CD8+ T cells (CXCR3, CRTH2, CD28) in the SE vs non-SE cohorts. CXCR3 has been shown to be involved in the localization of memory CD8+ T cells to the airways²¹ and chronic PM_2.5 exposure increases the homing of CXCR3+CD8+ T effector cells to the lung²².

CRTH2, also increased in the CD8+ memory T cells of the SE group, is a G-protein coupled receptor that binds to the ligand prostaglandin D2 (PGD2) and is suggested to be involved in allergic and eosinophilic inflammation through the activation and chemotaxis of type-2 cytokine-producing cells, such as Th2 cells^23,24 and ILC2^25,26. A recent study also demonstrated that CRTH2 is highly expressed on type 2 cytokine producing cells (e.g., IL-5 and IL-13) in human peripheral blood CD8+ T cells. CD8+CRTH2+ (Tc2) cells are enriched in the blood and airways of individuals with severe eosinophilic asthma compared to those with severe non-eosinophilic asthma and healthy controls^27,28.

The SE cohort also had increased CD28 in the CD8+ memory T cells in comparison to the non-SE group. CD28 provides essential costimulatory signals for T-cell activation, influencing the activation, proliferation, and effector functions of memory T cells²⁹. A previous study found that human memory T cells are more sensitive to CD28 costimulation than naive T cells, and single nucleotide polymorphisms associated with immune diseases (e.g., allergies, asthma, coeliac disease, etc.), identified through genome-wide association studies, are enriched in CD28 sensitive genes³⁰. Our results showed increased CD28 expression in memory CD8+ cells in SE subjects compared to non-SE subjects (Supplementary Figure 12I), suggesting that CD28 may be involved in regulating memory CD8+ T cell responses to smoke exposure. These findings are significant because memory CD8+ cells mediate responses to infection and modifications of CD8+ phenotypes and function can result in human immune dysfunction and subsequent disease^31,32. It is interesting to note that wildfire smoke exposure to the public has been demonstrated to be associated with increased infections through epidemiological studies^33–35.

For memory CD4+ cells, we also found differences between the cohorts with increased CCR6 and reduced CCR4 and CCR7 in the SE vs non-SE cohorts. CCR6+CD4+ T cells have been suggested to produce inflammatory cytokines such as IL-17 and IL-21^36–38. CCR6-expressing memory T cells can migrate in response to CCL20, which is secreted by airway epithelial cells³⁹, and its activation is closely associated with IL-1β-mediated inflammatory responses⁴⁰. Given that our results showed increased CCR6 expression in memory CD4+ and CD8+ T cells in SE vs non-SE subjects (Supplementary Figures 12F, 12H, Figures 3B, 3D, and 3E), this data suggest that smoke exposure may induce the migration of memory T cells to inflammatory sites through increased expression of CCR6. Overall, the upregulation of the activation and chemokine receptor markers suggest that the immune system has been activated in response to wildfire smoke pollutants. Smoke exposure may represent a persistent inflammatory stimulus capable of driving T cell exhaustion or terminal differentiation, such as in chronic infection or cancer⁴¹.

In addition, we demonstrated a significant association between DNA methylation and toxic metal abundance in total live immune cells. While metal-induced carcinogenicity is well-established, the immune mechanisms by which toxic metals lead to cancer are not fully understood. Therefore, the immune profiling method developed here is particularly informative due to the increasing exposure to toxic metals from environmental sources such as wildfire smoke. This approach may be important in frequently smoke-exposed populations, such as firefighters, who are increasingly recognized to have greater susceptibility to long-term health consequences, including chronic respiratory disease and cancer. Little is known about the impact on the immune system in this context, making CyTOF a highly useful tool for further investigation.

Interestingly, we discovered that the amount of Hg bound to dead cells was higher in the SE vs non-SE cohorts. Mercury is pervasive in the environment, with most individuals having some level of Hg in their blood from various environmental sources (e.g. contaminated fish or rice). Hg is also naturally contained in soil and vegetation⁴² and is mobilized during wildfires during the combustion process. The increase in Hg exposure from wildfire smoke may increase the cell’s toxicity threshold given that toxic metals can bind to a cell^43,44, and/or cause cell death with the apoptotic and autophagic pathways for Hg being dose dependent⁴⁵. Importantly, we determined that the level of ¹⁹⁸Hg in both total dead and total live cells correlate with years of smoke exposure.

Similarly, cadmium is also widely present in the environment and is a component of wildfire smoke (e.g. plants uptake Cd)⁴⁶. Once inhaled or ingested, Cd accumulates in the body over time and is associated with both cancer⁴⁷ and smoke-related toxicity⁴⁸. In the live cells, additional metals (Br and Ta) were elevated in the SE cohort. There were also toxic metal isotopes that were differentially bound to specific naïve versus memory immune cells. For example, ¹¹⁰Cd was bound to memory CD8+ cells in SE vs non-SE cohorts. It has previously been shown that certain toxic metals can bind to the major histocompatibility complex (MHC), and cause metal-induced cytotoxicity and inflammatory responses⁴⁹. Moreover, since T-cell receptor-MHC signaling is a key factor in producing robust immune responses, these metals may cause immune system dysfunction. Our data also demonstrate that ⁷⁹Br was bound to NK cells (CD56^dim CD16⁺ and CD56^BrightCD16^Low NK) in SE vs non-SE cohorts. NK cells originate in the bone marrow and are important for fighting infection and cancer⁵⁰, both outcomes which are impacted by exposure to wildfire smoke. In sum, these toxic metal findings may decrease the functioning of the immune system. This could explain the relationship between exposure to wildfires and immune dysfunction (e.g. the increase in COVID morbidity and mortality associated with wildfire smoke exposure³⁴).

Additionally, in future studies our toxic metals methodology may allow estimation of the timing of exposure by relying on the known average life span of immune cells. For example, given that CD8+ T cells have a lifespan of 32 days in circulation⁵¹, and because we determined that CD8+ T cells from our exposed individuals contained elevated levels of Cd, we can infer that the exposure to Cd in the environment likely occurred within the month prior to blood collection, as consistent with much of the timing of the SE cohort’s exposures to wildfire smoke. This method of measuring toxic metal exposure on the single cell level demonstrates the utility of toxic metal isotope analysis in immune cells as a possible biomarker for tracing relatively recent toxic exposures.

There are several limitations of this study. We have tried to mitigate concerns of confounders and interpretability by designing our cohorts to control for external variables (for example, we excluded anyone with an acute infection or on immunomodulatory drugs or who previously smoked tobacco products). We specifically excluded those with acute or chronic diseases. We focused on molecular mechanisms using quantifiable assays that are validated and standardized. To this end, we focused on toxic metals and PFOS/PFOA measurements since they can be quantified in blood samples and are often seen in those that are exposed to smoke^52–54. We understand that these components of wildfire smoke (PFOS/PFOA and toxic metals) could be found in other environmental exposures outside of wildfire smoke exposure; this is one of the reasons why we chose controls to compare against a known exposed group. We also acknowledge that we did not measure levels of vitamins or folic acid in the blood, all of which may influence DNA methylation levels. Future methylation studies could include these measurements if possible. Our sample size calculations showed that, based on a high dimensional single-cell omic approach⁵⁵, n=11 in each group could have been sufficient to yield significant results. However, we chose to increase the sample size consistent with other single cell omic studies in humans to account for unspecified variations^56–59. In addition, we purposefully chose smoke-exposed individuals as our source of blood samples versus age-matched, sex-matched controls to optimize our ability to detect a significant signal for our mechanistic analysis, especially given this was the first study of its kind. We found that other studies with groups exposed to smoke had similar targets^60,61; therefore, it is possible that our findings can be broadly applied those generally exposed to wildfire smoke. In the future, longitudinal studies with larger cohorts among different populations will be important to consider for mechanistic follow up studies.

Our study presents a new approach to evaluating the effects of smoke exposure on human health by analyzing blood for toxic metals, PFAS, differential methylation, and immune cell phenotyping at both the single-cell and DNA levels. The study also introduces a novel use of mass cytometry to measure individual toxic metal isotopes bound to single immune cells. Our findings provide preliminary evidence suggesting that toxic metal exposure affects the epigenetic profile of immune cells in a cell-type-specific manner, which also contributes to the growing field of metallomics. These findings could be particularly relevant for populations who are at increased health risks due to chronic smoke exposure. Overall, our findings lay the foundation for advancements in detection, prevention, and management of interventions for smoke exposed populations.

Methods

Cohort recruitment:

IRB approval was provided by the Stanford University IRB review committee. After signing a written consent, participants had a venous blood draw. We chose blood as the biospecimen of choice because we planned to test the relationship between immune cells and toxins which can be measured at the single-cell level. Furthermore, studying the blood allows the investigation of the impacts of smoke exposure on methylation of key immunoregulatory genes. Peripheral blood mononuclear cells (PBMCs) and plasma were collected as per standard methods⁶². SE adults (n=31, including both firefighters and smoke-exposed civilians) and non-SE adults (n=29; control) underwent methylation and Mass Cytometry (CyTOF) analyses based on an age and sex-matched convenience sample (i.e. sufficient quantity of cells drawn at blood draw). The initial sample size calculation was based on our previous CyTOF study⁶³. By analyzing the proportion of CD8+ T cells among total live cells, we had estimated that an effect size of 1.24 and 11 participants per group were needed to detect a difference with an α value of 0.05 and power = 0.80. Exposure was determined by self-report and verification of location during fire smoke events (Supplementary Table 1). We also confirmed that the controls had not been exposed to fire smoke events by self-report and validation of location.

Sample collection:

Based on rigorous eligibility criteria based on self-report, neither the SE individuals nor controls were allowed to participate if they smoked tobacco, were taking any immunomodulatory therapies, or had ongoing chronic conditions or acute conditions, with the exception of participants having mild allergies and one female had fully treated, non-active thyroid disease. Sex and race of individuals were identified by self-report and were found to be balanced in the two cohorts for our studies (Table 1). Blood samples were collected within approximately a month of their last exposure to fire smoke.

Blood collection and sample processing:

We collected peripheral blood and isolated peripheral blood mononuclear cells (PBMCs) by Ficoll-Plaque per standard methods using EDTA collection tubes and no dilution of whole blood prior to processing⁶⁴. PBMCs were stored in liquid nitrogen and thawed once prior to use. We also collected plasma that was centrifuged, aliquoted, and stored in −80° F. The stored samples from smoke-exposed individuals were used for the measurement of PFAS analytes, including total PFOS (mean 3.63 μg/L; SD 1.91) and total PFOA (mean 1.48 μg/L; SD 0.65) (Eurofins, USA).

Per- and polyfluoroalkyl substances measurement:

Samples were sent for PFOS and PFOA analysis to Eurofins Discovery Service in Sacramento, CA. They used standardized and validated assays; specifically, they conducted an on-line solid phase extraction coupled with high-performance liquid chromatography - isotope dilution tandem mass spectrometry method. Low-concentration quality control materials (QCs) and high-concentration QCs, prepared from a calf serum pool, were analyzed with the study samples and with reagents and serum blanks to ensure the accuracy and reliability of the data. PFOS and PFOA measurements were only performed in firefighters.

Mass cytometry:

As per published method⁶⁵, cells were thawed and incubated in RMPI-10% FBS media at 37°C overnight. Next, cells were centrifuged at 300g for 10 minutes at room temperature. The cell pellet was resuspended in CyFACs buffer (PBS + 2% FBS + 0.1% sodium azide). Cell counts were adjusted to 1.5 × 10^6 viable cells/mL in CyFACs buffer. 100 μl of the diluted Cisplatin (5 μM) were added and cells incubated for 5 mins followed by 1 ml CyFACS buffer to stop the reaction. Cells were then washed in CyFACS buffer twice at 400g for 6 mins. Cells were then stained with Fc blocker (BioLegend, 422302) at room temperature for 10 minutes. Cells were then incubated with metal-tagged surface antibodies (Supplementary Table 3) for 30 minutes at room temperature, followed by 2 CyFACS washes (400g for 6 mins) and fixation in 2% PFA at room temperature for 30 mins. Cells were washed in 0.5 ml 1x perm buffer (EBioscience (00-8333-56) in 0.5 ml CyFACS twice to permeabilize. Diluted Ir-intercalator solution (2% PFA; 0.25uM Ir) was added and incubated for 20 minutes, followed by washing twice with 500 μl CyFACS buffer (700 g for 6 min) and then washed twice in 500 μl CyPBS (700 g for 6 min). Cells were stored overnight until run on the mass cytometer. Samples were then acquired on the mass cytometer (Helios, Fluidigm) after standard instrument setup procedures. We recorded from the standard antibody/metal cell-stained channels (Supplementary Table 3) and additionally, from open channels in which no metal-conjugated antibody was used to stain the cell (Supplementary Table 4). This allowed measurement of toxic metals including, bromide (isotopes 79 and 81), cadmium (isotopes 106, 108, 110,112, and 114), antimony (isotopes 121 and 123), tantalum (isotopes 180 and 181), tungsten (isotopes 182,184, and 186), mercury (isotopes 196,198, and 202), lead (isotopes 204, 206 and 208), and bismuth (isotope 209).

4.3. Mass cytometry data analysis:

Following normalization, FCS files underwent standard pre-processing to remove debris and doublets. Live immune cell types were analyzed by manual gating and unsupervised computational methods in parallel. Manual gating was used to identify major immune cell populations. We used FlowJo v10 (FlowJo LLC., Ashland, OR; version V10.6) to manually gate normalized, de-barcoded data files. Intact cells were first selected based on ¹⁹³Ir and ¹⁹⁵Ir staining. Doublets were removed along with ¹⁴⁰Ce beads. Live/dead cells were gated using ¹¹⁵IN staining. Samples were stained in 7 batches. Batch effects were evaluated by profiling and comparing two common samples used as internal controls (Supplementary Figure 19). These effects were minimized through arcsine transformation and landmark normalization and further adjusted during statistical modeling when comparing SE and non-SE groups. The gating schemes that were used to identify CD3+ and CD3− immune cell subsets are depicted in (Supplementary Fig. 10–11). FlowSOM-based unsupervised computational analysis was initially performed before the manual gating on all the live cells using default parameters using the following steps⁶⁶. First, expression values of lineage markers were transformed using the arcsine function (inverse hyperbolic sine; cofactor of 5), from the flowStats R package (version 4.12.0)⁶⁷. The complete set of lineage and functional markers were used in our CyTOF panel is available in Supplementary Table 3. Next, a random set of 80,000 live cells was selected from each sample, and a combined data frame of all the cells was subjected to FlowSOM-based clustering with a fixed number of 10 clusters. FlowSOM clusters were represented as a heatmap to show the median expression value of each marker within every cluster. Using the heatmap cell clusters were manually annotated to respective cell types. Next, 10,000 randomly selected cells from all 10 clusters were subjected to the non-linear dimensionality reduction technique- Uniform Manifold Approximation and Projection (UMAP) for cluster/cell type visualization⁶⁸. Statistical differences in cell proportions (or abundances) relative to their parent populations were evaluated between SE and non-SE groups using the MAST algorithm implemented in the Seurat R package⁶⁹. This analysis was performed on both manually gated and unsupervised cell clusters, incorporating batch effects as a latent variable while accounting for the single-cell nature of the data. Outcomes with p value < 0.05 and FDR adjusted p-value (q-value) $\le$ 0.20 were considered as significant. The same steps were repeated independently on each manually gated cell type, i.e., CD4+ T cells, CD8+ T cells, CD19+ B cells, NK cells and CD11c+ cells, where the number of fixed clusters varied for each cell type. Cells significantly expressing higher marker (e.g. CCR4^hi cells) were predicted computationally. For each marker, the distribution of its expression values in a given sample was used to estimate the 90th percentile, which served as the threshold. Cells expressing the marker above this threshold were classified as cells with high marker expression.

4.4. Toxic metal detection in immune cells using mass cytometry:

we recorded channels between atomic masses 75 and 209 in which no metal-tagged antibodies were stained. This allowed us to measure the toxic metals coupled to single cells. Computational methods were used to identify cells containing toxic metals in our dataset. For each toxic metal isotope, the signal intensity was used to cluster cells into two groups via Gaussian mixture modeling using Mclust algorithm⁷⁰. This clustering approach allowed us to distinguish the cell cluster containing the toxic metal isotopes versus those cells without isotopes. This step was repeated for each toxic metal, and only cells with at least one toxic metal were retained for further analysis. Subsequently, cell count proportion for each toxic metal per sample was calculated and were compared between SE and non-SE groups. Wherein, statistical differences in cell proportions with toxic metals were evaluated between SE and non-SE groups using the MAST algorithm implemented in the Seurat R package⁶⁹. This analysis was performed on both manually gated, incorporating batch effects as a latent variable while accounting for the single-cell nature of the data. This differential analysis aimed to identify any toxic metals exhibiting different levels between the two cohorts. Outcomes with p value < 0.05 and FDR adjusted p-value (q-value) ≤ 0.20 were considered significant.

Epigenetic profiling using methylation array:

Based on established methodologies from Morin et al.¹⁸, DNA methylation was assessed using the Asthma&Allergy array (Illumina).

Statistical Analysis of methylation data:

DNA methylation was assessed using the Illumina Allergy & Asthma Custom BeadChip. Methylation data from the array was processed using minfi (version 1.44.1). For the Custom array, probes that failed (detection P < 0.01 in at least 25% of samples), with underlying SNPs, mapped to the X chromosome, or were missing genomic coordinates were removed. The annotation and manifest packages for the Allergy & asthma array can be found at⁷¹. Raw probe values were background corrected using preprocessIllumina (bg. correct= “TRUE”, normalize= “no”), and quantile normalization was performed using ENmix (version 1.34.02)⁷². We converted methylation Beta-value to M-value using lumi R package (version 2.50.0)⁷³. Limma R package (version 3.56.2) was used to perform differential methylation analysis⁷⁴. The lmfit function was used to fit the linear model between methylation and SE/Non-SE group, adjusted for age and batch (~Group + Race + Batch). Association between epigenetic site and corresponding gene was performed using GREAT (using default parameters). It performs enrichment analysis using genomic coordinates by first associating genomic regions with nearby genes, followed by applying the gene annotations to the regions using GREAT webserver⁷⁵. The pathway enrichment analysis of annotated genes was performed using Metscape webserver⁷⁶.

Bayesian statistical analysis:

Bayesian statistics was performed using BayestestR package (version 0.13.1)⁷⁷. The correlation between a set of variables was computed using Bayesian correlation analysis. The Bayesian approach allowed us to enumerate the probability that the correlation or coefficient falls within a certain range or distribution, rather than a fixed value. Those correlations with Bayes factor (BF) > 2, probability of direction (pd) > 90% and median correlation coefficient (r) > 0.60 were considered as significant. Whereas for statistical modeling, coefficients with BF > 2, pd > 90%, and ROPE < 0.10 were considered significant. The default parameters, including weakly informative priors and Markov Chain Monte Carlo settings (chains = 4, iterations per chain = 2000; warm-up = 1000 iterations per chain and no thinning) were used.

Data visualization:

Circular plots (chord diagram) were plotted using R package circlize (version 0.4.15)⁷⁸. UMAP, density plots, scatterplots, Manhattan plot and boxplots were plotted using R package ggplot2 (version 3.4.4)⁷⁹. EnhancedVolcano R package (version 1.18.0) was used for a volcano plot visualization⁸⁰. Heatmaps were plotted using the R package pheatmap (version 1.0.12)⁸¹. UMAP, density plots, scatterplots, Manhattan plot and boxplots were plotted using R package ggplot2 (version 3.4.4)⁷⁹. EnhancedVolcano R package (version 1.18.0) was used for a volcano plot visualization⁸⁰. Heatmaps were plotted using the R package pheatmap (version 1.0.12)⁸¹. BioRender and Adobe stock images were used to create workflow illustrations.

Data availability

The DNA methylation and mass cytometry datasets (including manually gated live single cells and cell subsets) generated during the current study are available in the Harvard Dataverse data repository (https://doi.org/10.7910/DVN/5PWZJM).

Code availability

The R code to reproduce the results of the study are available at https://github.com/knadeaulab/Smoke_Exposure_Study