Proteomics, Human Environmental Exposure, and Cardiometabolic Risk

Abstract

Background:

The biological mechanisms linking environmental exposures with cardiovascular disease (CVD) pathobiology are incompletely understood. We sought to identify circulating proteomic signatures of environmental exposures, and examine their associations with cardiometabolic and respiratory disease (CMD) in observational cohort studies.

Methods:

We tested the relations of >6500 circulating proteins with 29 environmental exposures across the built environment, green space, air pollution, temperature, and social vulnerability indicators in nearly 3000 participants of the Coronary Artery Risk Development in Young Adults (CARDIA) study across 4 centers using penalized and ordinary linear regression. In >3500 participants from the Framingham Heart Study (FHS) and Jackson Heart Study (JHS), we evaluated the prospective relations of proteomic signatures of the envirome with CVD and mortality using Cox models.

Results:

Proteomic signatures of the envirome identified novel/established CVD-relevant pathways including DNA damage, fibrosis, inflammation, and mitochondrial function. The proteomic signatures of the envirome were broadly related to CMD and respiratory phenotypes (e.g., body mass index, lipids, and left ventricular mass) in CARDIA, with replication in FHS/JHS. A proteomic signature of social vulnerability was associated with a composite of CVD/mortality (1428 events; FHS: HR=1.16, 95% CI 1.08–1.24, P=1.77e-05; JHS: HR=1.25 95% CI 1.13–1.38, P=6.38e-06; HR expressed as per 1 standard deviation increase in proteomic signature), robust to adjustment for known clinical risk factors.

Conclusions:

Environmental exposures are related to an inflammatory-metabolic proteome, which identifies individuals with CMD and respiratory phenotypes and outcomes. Future work examining the dynamic impact of the environment on human cardiometabolic health is warranted.

Graphical Abstract

INTRODUCTION

For the last two decades, environmental exposures (built structures, air pollution, neighborhood resources, food availability, etc.) have been linked to the risk of cardiometabolic disease (CMD) defined as obesity, insulin resistance, hypertension, and vascular disease^1–7. Alongside epidemiological investigations, studies in model systems have implicated certain exposures (e.g., fine particulate matter with median diameter < 2.5 μm, PM_2.5) in canonical mechanisms of cardiometabolic disease, including endothelial homeostasis and vascular function^8–10. These exposures are highly complex and integrative, with exposures with beneficial associations with health (e.g., proximity to greenspaces) potentially balancing effects from exposures considered more “adverse” (e.g., air pollution)¹¹. These inherent complexities present challenges for the study of environmental exposure in large human populations, specifically limiting the ability to assign a relation between exposures across individuals and geographic locations and the host biochemical state. Furthermore, linkage of the various external exposures (termed the “envirome”) with molecular phenotypes and CMD mechanisms has been limited by the availability of concurrent data on human CMD phenotypes, exposures, risk factors (e.g., obesity), and molecular markers across broad, relevant pathways. Uniting these elements represents a critical next step in understanding the complex interactions between the environment, biochemistry, and human CMD.

We studied 2961 individuals from the Coronary Artery Risk Development in Young Adults study (CARDIA) across 4 geographically distinct field centers in the United States (Birmingham, AL; Oakland, CA; Minneapolis, MN; Chicago, IL) with concurrent quantification of select human environmental exposures and a broad circulating proteome (≈7000 proteins). Our objectives were three-fold: (1) to investigate the plasma proteomic correlates of a multi-dimensional human envirome including measures of the built environment, social vulnerability, temperature trends, and air pollution; (2) to study the relation of this “environmental proteome” with risk factors, phenotypes, and pathways central to CMD; (3) to study whether alteration of levels of these proteins presage cardiovascular disease (CVD) among participants in the Framingham Heart Study (FHS) and Jackson Heart Study (JHS). The geographic diversity across CARDIA centers allowed this “natural” experiment to begin to investigate biochemical correlates of a human envirome. Our overall aim was to connect the measurable envirome with variations in the circulating proteome at an epidemiologic scale, a critical next step in understanding environmental-individual interactions relevant to CMD.

METHODS

Data Availability.

Data are publicly available through the respective coordinating centers^12,13, including data from CARDIA (www.cardia.dopm.uab.edu), the Framingham Heart Study (FHS; https://www.framinghamheartstudy.org/fhs-for-researchers/), and Jackson Heart Study (JHS; https://www.jacksonheartstudy.org/Research/Study-Data/Data-Access). Please see the Major Resources Table in the Supplemental Materials. The Institutional Review Board at each institution approved each study.

Study population.

CARDIA:

CARDIA is a prospective, cohort study of young White and Black adults (age 18–30 years recruited in 1985–1986 from 4 field centers in the United States: Birmingham, AL; Chicago, IL; Minneapolis, MN; and Oakland, CA) with a primary aim of studying risk factors for CVD in young adulthood^14–17. For the present analysis, we included 2961 individuals with available measures of the human envirome and circulating proteome at Year 25 (2010–2011) after enrollment. We utilized standard CMD clinical and biochemical risk factors, imaging indices of cardiovascular and metabolic risk, and demographic data, the details of which have been reported (Supplemental Table 1)^18–23.

Framingham Heart Study (FHS):

The enrollment and characteristics of the FHS Generation 2 (“Offspring”) cohort are detailed elsewhere²⁴. For the present investigation, 1777 participants who underwent protein quantification at the 5^th examination cycle without prevalent CVD who had complete data on covariates, incident CVD, and mortality were eligible for inclusion²⁵. Definitions of CMD clinical and biochemical risk factors have been reported previously^26–28. Cardiovascular disease events were defined as myocardial infarction, coronary death, angina pectoris, coronary insufficiency, heart failure, stroke or transient ischemic attack, or intermittent claudication²⁹. All participants provided informed written consent, and all study protocols were approved by the Boston University Medical Center Institutional Review Board.

Jackson Heart Study (JHS):

JHS recruited 5306 Black adults (age 35–84 years old except in the family study which recruited age ≥21 years) from the Jackson, MS metropolitan area (Hinds, Madison, Rankin counties) between 2000–2004, with ongoing surveillance and adjudication for CVD risk factors and outcomes. In this analysis we included 1795 individuals with proteomics performed at the first examination in JHS (see Proteomics, below) and complete data on covariates, incident CVD, and mortality³⁰. For this analysis, we defined incident CVD as fatal and non-fatal myocardial infarction, cardiac procedure, and stroke, as defined previously³¹.

Characterization of human environmental exposures.

We quantified different elements of the human envirome based on residential history, including measures of air pollution, temperature, built environment, food availability, green spaces, and community-level measures (Table 1). We included 5-year averages for pollutants and temperature exposure variables before each visit month. Methodology for quantification of environmental exposures is reproduced with minimal modification from CARDIA for scientific rigor and reproducibility (with attribution provided by this statement).

Table 1.

Definition and data sources of exposures in CARDIA.

Exposure	Description	Source
CDC Social Vulnerability Index	CDC Social Vulnerability Index	CDC37
Normalized difference vegetation index (NDVI)	NDVI is an indicator of vegetation. Defined within 5km buffer	NASA (https://search.earthdata.nasa.gov/search)
Dist. to river	Distance to nearest river from participant’s address	National Hydrography dataset (https://www.sciencebase.gov/catalog/item/5ea068ae82cefae35a12a120)
Dist. to shoreline	Distance to nearest shoreline from participant’s address	NOAA (https://shoreline.noaa.gov/data/datasheets/medres.html)
Dist. to park	Distance to nearest major park from participant’s address	USA parks (https://www.arcgis.com/home/item.html?id=578968f975774d3fab79fe56c8c90941)
Fast food within 3km	Number of fast-food restaurants within 3 km of participant’s residence	Dun and Bradstreet, a commercial data set of US businesses38
Grocer within 3km	Number of grocery stores within 3km of participant’s residence
Slow food within 3km	Number of slow food restaurants within 3km of participant’s residence
Mini-mart within 3km	Number of convenience stores within 3km of participant’s residence
Any food within 3km	Access to any food facilities within 3km of participant’s residence
School within 3km	Number of educational facilities within 3km of participant’s residence
Public rec. within 3km	Number of public recreation facilities within 3km of participant’s residence
Road length within 3km	Total road length within 3km of participant’s residence	Tele Atlas StreetMap Premium for ArcGIS - North America (https://hub.arcgis.com/content/d3c77c670f924bd189befa4af4a9ca3c/about)
PM10	5-year average concentration prior to a visit estimated at a participant’s address.	U.S. EPA Air Quality System34
Ozone
NO2
SO2
CO
PM2.5 (5-yr avg.)		Dalhousie University Atmospheric Composition Analysis Group / GEOS-Chem32,33
Black carbon in PM2.5 (5-yr avg.)
Ammonium in PM2.5 (5-yr avg.)
Nitrate in PM2.5 (5-yr avg.)
Organic matter in PM2.5 (5-yr avg.)
Sulfate in PM2.5 (5-yr avg.)
Mineral dust in PM2.5 (5-yr avg.)
Sea salt in PM2.5 (5-yr avg.)
Summer temp. (5-yr avg.)	Average temperature of the past 5 summers prior to a visit	North America Land Data Assimilation System36
Winter temp. (5-yr avg.)	Average temperature of the past 5 winters prior to a visit
Temperature (5-yr avg.)	Average temperature of the past 5 years prior to a visit

Air pollution:

We utilize a validated product by the Dalhousie University Atmospheric Composition Analysis Group³² for PM_2.5 and PM_2.5 species in CARDIA, which employs a hybrid approach (satellite remote-sensing data, MODIS and MISR), chemical transport modeling, and ground monitors) to predict PM_2.5 concentrations at 1-kilometer resolution³². For PM_2.5 species (black and organic carbon, sulfate, nitrate, ammonium, mineral dust, sea salt particles) from 2000 onwards, GEOS-Chem (a global 3D chemical transport modeling) simulation integrated additional ground-based constraints via geographic weighted regression (GWR)³³. For other pollutants (PM₁₀, ozone [O₃], nitrogen dioxide [NO₂], sulfur dioxide [SO₂], and carbon monoxide [CO]), data since the 1980s were obtained from the U.S. EPA Air Quality System³⁴. To assign exposures to each CARDIA participant, we utilized annual averages of interpolated values derived from a spatiotemporal kriging method³⁵ based on residential history that use both spatial and temporal information to predict exposures at participants’ residential addresses.

Weather:

We obtained weather variable data from the North America Land Data Assimilation System (NLDAS), which integrates the most reliable weather observations and model outputs³⁶. This comprehensive system offers high-resolution weather parameters spanning from 1979 to the present. In our study, we specifically derived annual, summer, and winter averages to represent temperature exposures.

Green and blue spaces:

We derived two metrics for exposure to greenspaces: (1) Normalized Difference Vegetation Index (NDVI; downloaded from NASA; https://search.earthdata.nasa.gov/search) and (2) distance to the nearest major park. For NDVI data using in this study, we used the Moderate Resolution Imaging Spectroradiometer (MODIS). We also derived distance to the nearest river (a natural flowing watercourse) and shore (a boundary where a body of water (e.g., lakes) meets the land) from participants’ residential address.

Neighborhood-level contextual variables:

We used the Social Vulnerability Index (SVI) developed by the Centers for Disease Control and Prevention (CDC). The SVI is a comprehensive measure that takes into account factors such as socioeconomic status, household composition, minority status, language proficiency, and housing type³⁷.

Food environments:

We finally considered proximity and availability of various types of food stores near their residential locations. We specifically used the counts of different types of food stores within a 3-kilometer radius of each respondent previously developed in CARDIA³⁸, including fast food stores, “slow service” (sit-down service) food stores, grocery stores, and convenience stores. We incorporated other factors related to the built environment and community resources, including presence of educational facilities, public recreation amenities, and the total length of roads.

Quantification of the circulating proteome

Proteomic quantification was performed using the aptamer-based technology for all studies included in this report (SomaScan; Somalogic, Boulder, CO). The CARDIA study utilized the SomaScan 7k platform, which quantified 7291 aptamers. Sixty-eight participants had >1 sample run for proteomics, protein data for these subjects was averaged for use in models. Protein data were examined for batch effects, of which none were detected. We excluded 61 aptamers with a coefficient of variation >20%. No participant outliers were identified in multivariate approaches using principal components analysis. Proteomics data were log-transformed and standardized (mean 0, variance 1) for analysis.

Protein quantification in FHS and JHS was performed with earlier versions of the SomaScan platform (≈1000–1300 aptamers). We used previously described methods to normalize protein levels in FHS and JHS^25,39. Briefly, given differences in collection batch in FHS and JHS, protein levels were standardized within each batch, pooled, and rank normalized across all samples, with subsequent residualization against assay plate before analysis to remove plate-based effects, as described²⁵.

Statistical analysis

An overview of the study design and statistical methods is shown in Figure 1A. We subset the CARDIA sample into derivation (70%) and validation (30%) sets that were balanced by center for each exposure. Given the differences in missingness of exposure variables, unique derivation and validation sets were created for each exposure’s models. Independent and dependent variables were standardized (mean 0, variance 1), separately in derivation and validation samples. We constructed two sets of models: (1) linear models with each protein as a function of each environmental exposure with adjustments for age, sex, and race (to understand the association of exposures with proteins); (2) penalized (least absolute shrinkage and selection operator [LASSO]) regression models for exposures as a function of the proteome (to develop signatures of each exposure as a function of the proteome). A 5% false discovery rate (FDR, Benjamini-Hochberg method) was used to address type 1 error across all non-penalized models. All models were estimated separately in the derivation and validation set (with proteins passing 5% FDR in derivation tested in the validation set). Of note, we purposefully did not adjust for clinical center in models relating proteins to exposures given that broad exposure variability by center was essential to define proteomic correlates. For each exposure, we selected proteins with FDR < 5% and performed pathway analysis respectively. We used R package clusterProfiler⁴⁰ and performed enrichment analysis based on selected proteins’ gene symbol IDs on WikiPathways database using all genes in WikiPathways as the background⁴¹ (prioritizing an approach that assigns proteins to pathways, recognizing potential bias of enrichment P value). The top 5 most enriched pathways mapped with at least 5 significant proteins in each exposure were selected for data visualization by heatmap (Figure 4). Penalized regression models (LASSO; R package caret⁴²) were trained using the derivation sample and model fit was tested in the validation sample. Results from penalized regression were used to create protein “scores” of environmental exposures by summing the products of a participant’s standardized protein levels by the model coefficients.

Figure 1: Study design and correlations between environmental exposures.

(A) We derived circulating proteomic signatures of a variety of environmental exposures. These proteomic signatures were then examined for their relation with cardiorespiratory and metabolic outcomes in 3 separate cohorts. (B) Heatmap of Spearman correlation coefficients between all the environmental exposures. CDC = Centers for Disease Control and Prevention; NDVI = normalized difference vegetation index; Dist. = distance; NO2 = nitrogen dioxide; SO2 = sulfur dioxide; CO = carbon monoxide; PM10 = particulate matter with median diameter < 10 μm; PM2.5 = particulate matter with median diameter < 2.5 μm; avg. = average; temp. = temperature.

Figure 4: Pathway analyses plots.

Pathway enrichment analysis was performed on WikiPathways for significant proteins (FDR ≤ 0.05) in each exposure. Heatmaps were made based on negative log₁₀ transformed adjusted p values for the top 5 most enriched pathways mapped with at least 5 significant proteins in each exposure. Rows represent pathways and columns represent exposures and the white to red colors in the heatmap represents the level of enrichment. Grey color means none of the significant proteins in the exposure can be mapped to the pathway.

We tested the LASSO model-predicted environmental exposures (e.g., proteomic signatures of each exposure) and the parent environmental phenotypes themselves in relation to cardiometabolic and respiratory outcomes (Supplementary Table 1). To mitigate confounding effects of traditional risk factors, these models were adjusted for age, sex, race, center (to account for differences in outcomes across centers), body mass index (BMI), systolic and diastolic blood pressure, use of anti-hypertensive medication, diabetes, lifetime pack-years of smoking, estimated glomerular filtration rate, total cholesterol, and high-density lipoprotein (HDL). In models where the phenotype was part of the adjustments (e.g., total cholesterol), the phenotype was not included as part of the adjustments. Models for blood pressure and cholesterol/lipoproteins were additionally adjusted for anti-hypertensive or lipid-lowering medication use, respectively. We selected canonical scores within each exposure category (e.g., PM_2.5 from pollutants) based on their relations with cardiometabolic and respiratory phenotypes for translation into JHS and FHS.

To assess the clinical importance of the envirome-prioritized proteome, we tested proteomic signatures of environmental exposure for association with incident CVD and mortality in JHS and FHS in survival (Cox) models. Use of protein scores in JHS and FHS required recalibration as not all proteins measured in CARDIA were available. We identified the intersection of proteins among the 3 cohorts through matching on UniProt identifier (not SomaScan identifier as these may change between generations of SomaScan panels). We used recursive feature elimination (R package caret⁴²) with a 5% tolerance to recalibrate protein scores in CARDIA using the reduced set of proteins. Protein scores were then examined for relations with cardiometabolic and respiratory phenotypes in JHS and FHS along with relations with incident cardiovascular disease (defined above for JHS and FHS) and all-cause death. Cox models were adjusted for age, sex, BMI, smoking, diabetes, systolic blood pressure, use of anti-hypertensive medication, total cholesterol, and HDL. Participants who did not experience the outcome were censored at the time of last contact.

RESULTS

Study populations.

Our study included 2961 CARDIA study participants from the Year 25 exam (Table 2). The CARDIA sample had a median age of 51 years (56% women; 46% Black individuals), with a moderate prevalence of obesity, hypertension, diabetes, and coronary artery calcification. As expected, environmental exposures differed by CARDIA center (Table 3; Supplemental Figure 1). For example, participants from Birmingham had on average greater indices of social vulnerability, distances to green spaces, exposure to PM_2.5 and summer temperature. We observed modest correlations among exposures, with blocks of higher correlation for measures of the built environment and some measures of air pollution (Figure 1B). Characteristics of the JHS and FHS samples are shown in Table 4. Overall, we studied 3572 participants with a median age in the mid-50s (58% women; JHS comprised of all Black individuals, FHS comprised of predominantly White individuals), with a moderate to high 10-year predicted risk of CVD events.

Table 2:

Baseline characteristics of the CARDIA study population.

Characteristic	Overall N = 2,961	Birmingham, AL N = 688	Chicago, IL N = 691	Minneapolis, MN N = 795	Oakland, CA N = 787	p-value
Age	51.0 (47.0, 53.0); 0%	50.0 (47.0, 53.0); 0%	50.0 (47.0, 53.0); 0%	51.0 (48.0, 53.5); 0%	51.0 (47.0, 53.0); 0%	0.001
Female	1,655 (56%); 0%	377 (55%); 0%	388 (56%); 0%	422 (53%); 0%	468 (59%); 0%	0.073
Race						7.20e-15
Black	1,375 (46%); 0%	387 (56%); 0%	314 (45%); 0%	282 (35%); 0%	392 (50%); 0%
White	1,586 (54%); 0%	301 (44%); 0%	377 (55%); 0%	513 (65%); 0%	395 (50%); 0%
Body mass index (kg/m2)	29 (25, 34); 0.2%	30 (26, 35); 0.1%	28 (25, 34); 0.1%	29 (25, 33); 0.4%	28 (25, 34); 0%	1.50e-06
Systolic blood pressure (mmHg)	117 (108, 127); 0.1%	118 (109, 128); 0.1%	115 (106, 127); 0%	117 (108, 126); 0.1%	117 (108, 127); 0.3%	0.026
Diastolic blood pressure (mmHg)	73 (66, 81); 0.2%	73 (66, 80); 0.1%	73 (65, 81); 0%	73 (66, 80); 0.1%	74 (67, 82); 0.4%	0.111
Treated for hypertension	816 (28%); 0%	277 (40%); 0%	155 (22%); 0%	163 (21%); 0%	221 (28%); 0%	3.10e-18
Diabetes	437 (15%); 0%	141 (20%); 0%	94 (14%); 0%	87 (11%); 0%	115 (15%); 0%	3.75e-06
Lifetime pack-years of cigarettes	0 (0, 7); 0%	0 (0, 6); 0%	0 (0, 6); 0%	1 (0, 13); 0%	0 (0, 3); 0%	1.47e-15
eGFR (ml/min/1.73m2)	94 (82, 108); <0.1%	96 (83, 110); 0%	93 (80, 107); 0%	95 (84, 109); 0.1%	93 (83, 107); 0.1%	0.019
Total cholesterol (mg/dL)	189 (166, 214); 0%	184 (161, 207); 0%	189 (169, 218); 0%	190 (166, 215); 0%	194 (170, 217); 0%	6.13e-06
High-density lipoprotein (mg/dL)	55 (45, 67); 0%	53 (44, 63); 0%	56 (46, 70); 0%	54 (44, 67); 0%	57 (46, 68); 0%	3.22e-05
On lipid lowering medication	478 (16%); 0%	147 (21%); 0%	100 (14%); 0%	115 (14%); 0%	116 (15%); 0%	4.22e-04
LV mass index (2D) (g/m)	33 (28, 38); 4.0%	34 (29, 40); 4.7%	32 (27, 37); 3.0%	32 (28, 37); 4.5%	33 (28, 38); 3.8%	6.13e-06
LV mass index (M-mode) (g/m)	39 (32, 46); 7.9%	41 (34, 48); 5.4%	37 (31, 44); 4.1%	39 (33, 46); 14%	39 (32, 46); 7.1%	3.31e-08
LV GLS (%)	−15.07 (−16.70, −13.51); 13%	−14.60 (−16.35, −13.05); 20%	−15.34 (−16.89, −13.80); 7.4%	−15.03 (−16.60, −13.38); 11%	−15.15 (−16.67, −13.68); 13%	1.37e-04
Total CAC (HU)	0 (0, 5); 0.2%	0 (0, 3); 0%	0 (0, 6); 0.7%	0 (0, 7); 0%	0 (0, 4); 0.1%	0.407
Total AAC (HU)	2 (0, 95); 2.5%	2 (0, 104); 0.6%	1 (0, 77); 2.0%	4 (0, 142); 3.8%	1 (0, 65); 3.3%	0.009
Visceral fat vol. (cm3)	120 (78, 172); 1.0%	118 (81, 165); 0.1%	119 (80, 171); 0.7%	123 (79, 181); 2.8%	118 (71, 171); 0.1%	0.250
Subcutaneous fat vol. (cm3)	303 (208, 442); 1.0%	341 (231, 493); 0.1%	282 (193, 408); 0.7%	292 (203, 427); 2.8%	302 (211, 436); 0.1%	4.07e-09
VAT:SAT ratio	0.38 (0.26, 0.57); 1.0%	0.35 (0.24, 0.52); 0.1%	0.41 (0.28, 0.61); 0.7%	0.42 (0.28, 0.62); 2.8%	0.36 (0.25, 0.55); 0.1%	1.02e-11
Hemoglobin A1c (%)	5.50 (5.30, 5.80); 1.2%	5.60 (5.30, 6.00); 1.6%	5.50 (5.30, 5.90); 0.7%	5.50 (5.20, 5.80); 1.0%	5.50 (5.30, 5.80); 1.4%	3.05e-05
FEV1 (L)	2.64 (2.19, 3.18); 15%	2.51 (2.12, 3.09); 16%	2.65 (2.17, 3.29); 18%	2.73 (2.30, 3.23); 18%	2.64 (2.17, 3.11); 9.8%	4.74e-05
FVC (L)	3.45 (2.83, 4.19); 15%	3.25 (2.65, 3.98); 16%	3.52 (2.81, 4.34); 18%	3.65 (3.02, 4.39); 18%	3.38 (2.83, 4.03); 9.8%	1.75e-11
FEV1/FVC	0.78 (0.73, 0.81); 16%	0.79 (0.75, 0.83); 16%	0.77 (0.73, 0.81); 18%	0.76 (0.72, 0.80); 19%	0.79 (0.75, 0.82); 9.8%	5.94e-19
Sum of AHA Life Simple 7	9.00 (7.00, 10.00); 22%	8.00 (7.00, 10.00); 16%	9.00 (7.00, 11.00); 13%	9.00 (7.00, 10.00); 27%	9.00 (8.00, 11.00); 31%	6.39e-13
Pooled cohort equation risk score	0.03 (0.01, 0.06); 1.5%	0.03 (0.01, 0.07); 0.9%	0.03 (0.01, 0.05); 1.0%	0.03 (0.01, 0.05); 1.8%	0.03 (0.01, 0.05); 2.0%	1.02e-05

Baseline demographics and characteristics of the CARDIA study population. Measures at the Year 25 exam are reported, except for FEV1, FVC and FEV1/FVC ratio which were reported at Year 30. Continuous variables are reported as median (interquartile range); missingness. Categorical variables are reported as n (%); missingness. P-values are for the Kruskal-Wallis rank sum test for continuous variables and Chi-squared test for categorical variables. For pairwise testing between centers, please refer to Supplemental Table 2. eGFR = estimated glomerular filtration rate; LVEDD = left ventricular end diastolic diameter; LV = left ventricular; GLS = global longitudinal strain; CAC = coronary artery calcification; AAC = abdominal aortic calcification; vol. = volume; VAT = visceral abdominal tissue; SAT = subcutaneous abdominal tissue; FEV1 = forced expiratory volume in 1 second; FVC = forced vital capacity; AHA = American Heart Association; CVD = cardiovascular disease.

Table 3:

Distribution of environmental exposures across CARDIA center.

Exposure	Birmingham, AL N = 688	Chicago, IL N = 691	Minneapolis, MN N = 795	Oakland, CA N = 787	p-value
CDC Social Vulnerability Index	0.53 (0.21, 0.79); 0%	0.40 (0.17, 0.75); 0%	0.39 (0.13, 0.68); 0%	0.43 (0.21, 0.71); 0%	5.68e-06
NDVI	0.52 (0.43, 0.60); 0%	0.34 (0.22, 0.43); 0%	0.42 (0.18, 0.54); 0%	0.33 (0.27, 0.42); 0%	8.23e-125
Dist. to river (km)	0.33 (0.18, 0.68); 0%	1.81 (0.49, 5.36); 0%	1.14 (0.47, 1.98); 0%	0.57 (0.26, 1.32); 0%	1.72e-113
Dist. to shoreline (km)	308 (297, 321); 0%	7 (3, 19); 0%	210 (202, 217); 0%	4 (2, 12); 0%	<2.2e-308
Dist. to park (km)	7,965 (4,902, 12,416); 0%	4,086 (1,788, 6,820); 0%	2,527 (1,532, 4,647); 0%	2,251 (1,322, 3,987); 0%	2.26e-134
Fast food within 3km	15 (6, 26); 21%	55 (32, 83); 24%	33 (21, 57); 20%	44 (24, 70); 20%	8.77e-119
Grocer within 3km	8 (4, 16); 21%	27 (10, 92); 24%	17 (7, 35); 20%	27 (10, 49); 20%	8.44e-75
Slow food within 3km	13 (7, 24); 21%	54 (29, 91); 24%	26 (14, 54); 20%	57 (24, 106); 20%	2.79e-117
Mini-mart within 3km	16 (8, 29); 21%	34 (17, 60); 24%	17 (10, 25); 20%	23 (13, 35); 20%	1.88e-55
Any food within 3km	74 (39, 119); 21%	250 (140, 451); 24%	133 (81, 272); 20%	233 (122, 414); 20%	2.29e-117
School within 3km	14 (7, 25); 21%	66 (32, 102); 24%	38 (21, 65); 20%	54 (27, 97); 20%	5.86e-135
Public rec. within 3km	1.0 (0.0, 3.0); 21%	6.0 (3.0, 15.0); 24%	4.0 (1.0, 7.0); 20%	6.0 (2.0, 12.0); 20%	3.95e-111
Road length within 3km (km)	211,597 (140,014, 293,747); 0%	305,810 (225,600, 416,752); 0%	325,532 (210,848, 382,564); 0%	311,189 (205,385, 385,450); 0%	8.74e-58
PM10 (μg/m3)	22.2 (18.2, 24.1); 1.2%	21.5 (19.0, 26.1); 0.6%	19.0 (16.7, 23.6); 2.3%	19.4 (15.9, 21.4); 1.9%	1.30e-42
Ozone (ppm)	0.030 (0.027, 0.031); 20%	0.031 (0.024, 0.033); 0.4%	0.029 (0.024, 0.032); 28%	0.031 (0.025, 0.034); 0.6%	2.45e-07
NO2 (ppm)	8.9 (7.1, 9.9); 26%	12.6 (11.0, 17.2); 4.8%	8.5 (6.1, 10.5); 86%	7.9 (5.9, 11.9); 6.9%	1.21e-108
SO2 (ppm)	2.31 (1.86, 2.61); 2.9%	1.91 (1.69, 2.11); 2.9%	0.36 (0.23, 0.73); 4.0%	0.75 (0.70, 0.89); 5.8%	<2.2e-308
CO (ppm)	0.35 (0.33, 0.40); 4.8%	0.38 (0.35, 0.41); 2.9%	0.30 (0.28, 0.32); 59%	0.32 (0.29, 0.41); 5.3%	2.26e-102
PM2.5 (5-yr avg.; μg/m3)	13.74 (13.28, 13.96); 0%	12.84 (11.64, 13.52); 0.1%	10.07 (9.79, 10.27); 0.1%	10.25 (9.00, 10.66); 0%	<2.2e-308
PM2.5-Black carbon (5-yr avg.; μg/m3)	1.51 (1.46, 1.54); 0%	0.93 (0.83, 1.01); 0.1%	0.77 (0.69, 0.82); 0.1%	1.02 (0.74, 1.12); 0%	9.81e-300
PM2.5-Ammonium (5-yr avg.; μg/m3)	1.25 (1.22, 1.31); 0%	1.67 (1.47, 1.75); 0.1%	1.26 (1.24, 1.28); 0.1%	0.64 (0.60, 0.74); 0%	1.89e-313
PM2.5-Nitrate (5-yr avg.; μg/m3)	0.70 (0.66, 0.74); 0%	2.38 (2.20, 2.54); 0.1%	2.11 (1.99, 2.24); 0.1%	1.92 (1.38, 2.12); 0%	6.91e-301
PM2.5-Organic matter (5-yr avg.; μg/m3)	5.28 (5.12, 5.37); 0%	3.45 (3.00, 3.76); 0.1%	3.25 (2.52, 3.76); 0.1%	4.76 (3.49, 5.15); 0%	2.60e-235
PM2.5-Sulfate (5-yr avg.; μg/m3)	3.93 (3.78, 4.07); 0%	2.77 (2.58, 2.94); 0.1%	2.01 (1.94, 2.09); 0.1%	0.98 (0.89, 1.02); 0%	<2.2e-308
PM2.5-Mineral dust (5-yr avg.; μg/m3)	1.09 (1.04, 1.11); 0%	0.58 (0.51, 0.62); 0.1%	0.46 (0.44, 0.47); 0.1%	0.41 (0.37, 0.45); 0%	<2.2e-308
PM2.5-Sea salt PM2.5 (5-yr avg.; μg/m3)	0.23 (0.22, 0.24); 0%	0.26 (0.22, 0.28); 0.1%	0.16 (0.15, 0.17); 0.1%	1.44 (0.77, 1.75); 0%	<2.2e-308
Summer temp. (5-yr avg.; Celsius)	27.9 (27.8, 28.0); 0%	22.7 (22.5, 22.7); 0.1%	22.4 (22.2, 22.6); 0.1%	19.6 (17.7, 21.9); 0%	<2.2e-308
Winter temp. (5-yr avg.; Celsius)	7 (7, 7); 0%	−4 (−4, −3); 0.1%	−9 (−10, −9); 0.1%	11 (9, 11); 0%	<2.2e-308
Temperature (5-yr avg.; Celsius)	17.8 (17.6, 17.9); 0%	10.3 (10.1, 10.5); 0.1%	7.7 (7.5, 7.9); 0.1%	14.6 (14.0, 15.4); 0%	<2.2e-308

Distribution of environmental exposures across centers in CARDIA. Variables are reported as median (interquartile range); missingness. P-values are for the Kruskal-Wallis rank sum test. For pairwise testing between centers, please refer to Supplemental Table 3. CDC = Centers for Disease Control and Prevention; NDVI = normalized difference vegetation index; Dist. = distance; NO2 = nitrogen dioxide; SO2 = sulfur dioxide; CO = carbon monoxide; PM10 = particulate matter with median diameter < 10 μm; PM2.5 = particulate matter with median diameter < 2.5 μm; avg. = average; temp. = temperature; IQR = interquartile range.

Table 4:

Baseline Characteristics of FHS and JHS.

Characteristic	JHS N = 1795	FHS N = 1777
Age (years)	55 (45, 64); 0%	54 (47, 62); 0%
Women	1,106 (62%); 0%	978 (55%); 0%
Ever smoker	589 (33%); 0%	349 (20%); 0%*
Body mass index (kg/m2)	30 (27, 35); 0%	27 (24, 30); 0%
Treated for hypertension	906 (50%); 0%	300 (17%); 0%
Systolic blood pressure (mmHg)	126 (116, 136); 0%	124 (112, 138); 0%
Diastolic blood pressure (mmHg)	76 (70, 82); 0%	74 (68, 81); 0%
Hemoglobin A1c (%)	5.70 (5.30, 6.10); 0%%	NA
Fasting glucose (mg/dL)	NA	95 (89, 103); 0.8%
Total cholesterol (mg/dL)	197 (172, 224); 0%	204 (181, 227); 0%
High-density lipoprotein (md/dL)	49 (41, 60); 0%	48 (39, 59); 0%
Pooled cohort equation 10-yr risk	0.09 (0.04, 0.17); 15%	0.05 (0.02, 0.10); <0.1%
CVD events	460 (26%); 0%	905 (51%); 0%
Years to CVD event or censoring	13.6 (12.1, 14.6); 0%	23.3 (13.8, 26.5); 0%

Baseline characteristics of the FHS and JHS study population. Continuous variables are reported as median (interquartile range); missingness. Categorical variables are reported as n (%); missingness. CVD = cardiovascular disease; HF = heart failure; NA = not available.

Association between human proteome and the envirome and its relation to CMD.

Our approach relies on understanding the joint variation between an exposure and circulating protein levels in a geographically diverse population, with the hypothesis that such conjoint variation may begin to represent a host biochemical profile reflective of the environment. Our first approach was to construct linear models for individual proteins as a function of each environmental exposure (adjusted for age, sex, and race; Data Supplement SD03 and SD04 for full regression results). Across 29 environmental exposure measures in our discovery sample, a broad range of the human proteome (N=6685 proteins) was related to the envirome (most frequent associations with distance to shore, N=5915; least with PM₁₀; N=4).

Given the complex inter-relations within the proteome, we next fit multi-protein models for each environmental exposure (LASSO) in our derivation sample (N=1412–2074 across all exposures) and tested its fit (R²) in our validation sample (N=602–887 across all exposures; Figure 2, Supplemental Table 4). We observed a broad range of model fits (R²=0.02–0.26), with a more modest fit for measures of the built environment and the best fit for temperature and social vulnerability (for details regarding the composition of these proteomic signatures please refer to Data Supplement SD05). The relation of these proteomic “signatures” of human envirome (and parent exposure variables) are shown in Figure 3. All models were adjusted for age, sex, race, center, and CMD risk factors (as noted in legend), given potential differences in phenotype across center and by CMD risk factors. Overall, while the environmental exposure phenotype had limited relations after adjustment (Figure 3A), we observed consistent relations between proteomics of all four environmental exposure categories with LV mass, obesity/regional adiposity, pulmonary function, and lipids (Figure 3B), even with adjustment for traditional risk factors, BMI, and CARDIA field center. Notably, regression estimates for models of CMD phenotypes (in Figure 3) vs. envirome exposure measures were modestly correlated with those for proteomic signatures of these exposures (Pearson r=0.53), though with significant variation across phenotypes (Figure 3C).

Figure 2. Model fit of protein scores for exposures.

(left) Barplot of the R² from linear models of exposure (outcome) as a function of its protein score (predictor) in the validation sample, which was not used to train the model. (right) Barplot of the number of proteins included in each protein score. CDC = Centers for Disease Control and Prevention; NDVI = normalized difference vegetation index; Dist. = distance; NO2 = nitrogen dioxide; SO2 = sulfur dioxide; CO = carbon monoxide; PM10 = particulate matter with median diameter < 10 μm; PM2.5 = particulate matter with median diameter < 2.5 μm; avg. = average; temp. = temperature.

Figure 3: Relation of environmental exposures and their corresponding protein scores with cardiometabolic and respiratory phenotypes in CARDIA.

Heatmap of beta coefficients for cardiometabolic and respiratory phenotypes as a function of the environmental exposure (A) or protein score (B) in the entire CARDIA dataset. Phenotype measures from the Year 25 exam are used, except for FEV1, FVC and FEV1/FVC ratio which were measured at Year 30. Protein scores and continuous phenotypes were standardized to mean 0, and variance 1, for use in models. Models were adjusted for age, sex, race, center, BMI, systolic and diastolic blood pressure, anti-hypertensive medication use, diabetes, lifetime pack-years of smoking, estimated glomerular filtration rate, total cholesterol and high-density lipoprotein. Models for cholesterol/lipoproteins were also adjusted for lipid-lowering medication use. For models where the outcome was part of the adjustments (e.g., BMI) the outcome was removed from the adjustments. (C) Comparison of regression coefficients for CMD-exposure vs. CMD-proteomic signature of exposure across the entire CARDIA sample, demonstrating significant variability in association magnitude. Color represents the model outcome: cardiac (LV mass, GLS), vascular (CAC, AAC, SBP, DBP), metabolic (BMI, VAT, SAT, VAT:SAT, HbA1c, total cholesterol, HDL), respiratory (FEV1, FVC, FEV1/FVC), and CVD risk scores (AHA Life Simple 7, pooled cohort equation). CDC = Centers for Disease Control and Prevention; NDVI = normalized difference vegetation index; Dist. = distance; NO2 = nitrogen dioxide; SO2 = sulfur dioxide; CO = carbon monoxide; PM10 = particulate matter with median diameter < 10 μm; PM2.5 = particulate matter with median diameter < 2.5 μm; avg. = average; temp. = temperature. LV = left ventricular; GLS = global longitudinal strain; CAC = coronary artery calcification; AAC = abdominal aortic calcification; vol. = volume; VAT = visceral abdominal tissue; SAT = subcutaneous abdominal tissue; SBP = systolic blood pressure; DBP = diastolic blood pressure; BMI = body mass index; HDL = high-density lipoprotein; FEV1 = forced expiratory volume in 1 second; FVC = forced vital capacity; AHA = American Heart Association; PCE = pooled cohort equation.

*false discovery rate <5% (Benjamini-Hochberg

Human proteome-envirome relations implicate broad pathways relevant to CMD

We next explored molecular pathways associated with environmental exposures (via their proteomic relations; Figure 4). Overall, proteomic association with air pollution, climate, or built environment included broad pathways of angiogenesis and endothelial survival (VEGFA-VEGFR2), regulation of cell growth (EGF-EGFR, PI3K-Akt signaling pathways), and inflammatory and adaptive immunity (Figure 4). While a deep survey per-protein exposition is out of scope for this report, we did note overlap across environmental exposures in pathways relevant to CMD that were consistent with the pathway analysis. For example, proteins related to the built environment included downstream mediators of broad fibrosis and immune signaling pathways, including TGF-beta signaling (SMAD2/3⁴³), PI3K signaling (GSK3A/B⁴⁴), innate immunity (STAT3⁴⁵), MAP kinase signaling (MAPKAPK2⁴⁶, MAPK10), and mitochondrial function and oxidative stress (PRDX3⁴⁷). Moreover, top proteins related to pollution exposures (specifically PM_2.5) were in pathways of DNA damage repair (APEX1⁴⁸, a main AP endonuclease; RECQL⁴⁹), mitochondrial metabolism (COQ9⁵⁰), proteins at the lung-pollutant interface (pulmonary surfactant-associated protein D, STFPD⁵¹), inflammation, and transcriptional/translational processing (ribosomal and nuclear proteins). Both unbiased analysis and per-protein review did not suggest a predominant pathway signature of the social vulnerability proteome but did include several mediators previously implicated in CMD pathogenesis (GDF15⁵², RARRES2⁵³, and THBS2⁵⁴, among them). Molecular pathways implicated by temperature appeared analogous to the pollution and built environmental proteome, likely reflecting some correlation between measures of temperature, elements of pollution, and built structures (Figure 1B).

Proteomics of environmental exposure and CMD in two geographically distinct centers

We selected representative proteomic signatures from each exposure category based on its relations with CMD phenotypes in CARDIA (Figure 3B) and tested their relations with CMD and respiratory phenotypes and long-term outcomes in 2 external cohorts. As the proteomic coverage in FHS and JHS differed from CARDIA, translation of the proteomic scores required recalibration (for details regarding the composition of these recalibrated proteomic signatures please refer to Data Supplement SD08). After comprehensive adjustment in both FHS and JHS, we observed similar patterns as in CARDIA, specifically with the proteomic correlates of social vulnerability and distances to green spaces related to CMD and respiratory phenotypes (FEV₁, FVC) (Figure 5). Over a median follow-up of 13.6 years (25^th-75^th percentile 12.1–14.6 years), we observed 248 deaths and 212 CVD events in JHS; in FHS, over a median follow up of 23.3 years (25^th-75^th percentile 13.8–26.5 years) there were 683 deaths and 573 CVD events. Models with age and sex adjustment (Model 1) demonstrated the relations of protein scores of greenspaces and social vulnerability with CVD and its composite with mortality in both geographies, with directionally consistent associations with air pollution metrics and higher summer temperatures (Figure 5; Supplemental Table 5). The association between a protein score of social vulnerability and outcome was robust to multivariable adjustment (BMI, smoking, diabetes, anti-hypertensive medication use, systolic blood pressure, total and high-density lipoprotein cholesterol; Model 2), with distances to greenspaces, higher summer temperatures, and air pollution exhibiting directionally consistent effect size.

Figure 5: Relations of protein scores with cardiometabolic and respiratory phenotypes, and outcomes in FHS and JHS.

(A) Representative protein scores (predictors) from each exposure category were examined in FHS and JHS for their relations with cardiometabolic and respiratory phenotypes (outcomes) in linear models adjusted for age, sex, BMI, smoking, diabetes, use of anti-hypertensive medications, systolic blood pressure, total cholesterol, and HDL. (B) Protein scores were entered into Cox regression models for incident CVD (left) and incident CVD plus all-cause death (right). The CDC Social Vulnerability Index (SVI) protein score was consistently related to outcomes in both FHS and JHS, robust to adjustment for traditional CVD risk factors. Definitions of incident CVD for FHS and JHS are reported in Methods. Model 1 is adjusted for age and sex. Model 2 is further adjusted for BMI, smoking, diabetes, use of anti-hypertensive medication, systolic blood pressure, total cholesterol, and HDL. Protein score composition is reported in Data Supplement SD08, and full Cox model results are reported in Supplemental Table 5. CDC = Centers for Disease Control and Prevention; dist. = distance; avg. = average; temp. = temperature; LV = left ventricular; CAC = coronary artery calcification; AAC = abdominal aortic calcification; BMI = body mass index; VAT = visceral abdominal tissue; SAT = subcutaneous abdominal tissue; HDL = high-density lipoprotein; FEV1 = forced expiratory volume in 1 second; FVC = forced vital capacity; PCE = pooled cohort equation.

*false discovery rate <5% (Benjamini-Hochberg)

The robust association of the proteomic score of social vulnerability with outcomes in JHS led us to explore its relations with neighborhood-level factors uniquely captured in that study (Figure 6). At the population level, a protein score for social vulnerability was associated with more neighborhood violence, poverty, neighborhood problems (e.g., excessive noise, heavy traffic, and litter), lower socioeconomic status, and social cohesion.

Figure 6: Relations of neighborhood factors with the protein score of CDC Social Vulnerability Index.

Provided the relation of SVI protein score with incident CVD and death, we further explored the relationship between the protein score and neighborhood factors available in JHS. The figure presents beta coefficients from models where the SVI protein score is the outcome and the neighborhood factor is the predictor, adjusted for age and sex. Dot size reflects the variance explained in the SVI protein score by the neighborhood factor using a type 1 ANOVA test with age and sex having priority in accounting for variance. Error bars are 1.96 x the standard error of the beta estimate. Faded items have a p value ≥0.05.

DISCUSSION

Here, we examined the relations between a human envirome composed of built, natural, and social environments across four American cities with a broad circulating proteome to define host biochemical profiles of environmental exposure and their links to CMD. Proteins related to measured environmental exposures specified shared pathways of cellular resilience, including angiogenesis, cell growth and turnover, and inflammation. Multivariable proteomic signatures of tested environmental exposures exhibited a wide range of fit (social vulnerability and select measures of air pollution displayed best fit) and they were consistently associated with key CMD phenotypes and subclinical measures of CVD, including cardiac hypertrophy, adiposity, and pulmonary function, robust to adjustment for standard risk factors and geography. Furthermore, in two geographically distinct populations (JHS and FHS), we observed relations between protein scores of each environmental exposure category (specifically social vulnerability) with CMD phenotypes and long-term CVD and mortality. Collectively, these results provide insight into joint variation between the natural, social, and built environment with the host proteome and its ramification on long-term cardiac health.

The human envirome has expanded over the last two decades to include a growing number of factors linked to human disease (stress/lifestyle, infectious disease, diet/activity, built and social environments, pollution, climate, among others)^55,56. There has been substantial progress in linking the envirome to CVD risk at a population level, highlighting the fundamental importance of and complexity in how the environment affects biochemical underpinnings of cardiometabolic risk. It is increasingly recognized that structural factors (e.g., rurality, social determinants of health linked to healthcare disparity), geography, policy, and community investment may influence environmental exposures and underlying risk. Certainly, environmental exposures do not act in isolation, contributing complexity to disentangling their joint effects. Identifying biochemical and phenotypic features that contribute to “allostasis”—an individual’s metabolic resilience to environmental stress—has been offered to approach an actionable endophenotype of external exposures⁵⁷ (e.g., with epidemiologic, epigenetic, and transcriptional studies of air pollution^58–60 and heavy metal studies of obesity and hypertension^61–64). In addition, most studies have focused on molecules in biofluids that directly reflect external exposure (e.g., microplastics, disinfectants, pesticides)^65–72, not necessarily capturing a broad host biological response to these exposures. Finally, model system studies have offered important insights into potential mechanisms⁷³, though whether their findings apply to broader geographically diverse populations in the real world remains less unclear.

The results of our study address these opportunities by providing a biological context to several key epidemiologic observations of the environment and CMD. From the broadest perspective, we observed pathway overlap across multiple domains of environmental exposure, principally involving axes of inflammation and immune activation, tissue remodeling and angiogenesis, mitochondrial function, and cell growth and signaling, consistent with phenotypic associations (fully adjusted) observed here that implicated proteomics of each environmental exposure in CMD. Indeed, these provide some biological rationale for prior observations linking certain exposures (e.g., greenness, air pollution) to CMD and its underlying mechanisms^74,75. Strikingly, PM_2.5 exposure was linked to pathways of DNA damage and transcriptional and protein turnover, consistent with data from model systems suggesting the impact of PM_2.5 exposure on DNA damage via increased oxidative stress and free radical production⁷³. Moreover, exposure to air pollution has been linked to pre- and post-transcriptional gene regulation (via rmethylation⁷⁶ or non-coding RNA⁷⁷), consistent with its association with proteins involved in transcription observed here. In this context, we observed the association of pulmonary surfactant-associated protein D (STFPD) with air pollution measures, consistent with the potential modulation of surfactant biology directly by air pollution⁷⁸. While these findings are observational and subject to confounding, the biological consistency between epidemiologic findings and biological observations (e.g., DNA damage) and pathway consistency between CMD pathogenesis and the environmental proteome is strongly suggestive of direct links warranting further study.

A key limitation of this study is the assessment of the proteome and the envirome at the same exam cycle in CARDIA, limiting any inference related to causality (“reverse” causation). Social determinants, community-individual resources, and access to healthy living spaces exhibit complex, critical interactions (e.g., rurality, social vulnerability, availability of health care) broadly linked to risk factor development⁷⁹, CVD, and mortality^80,81, even from a young age⁸². Indeed, a bidirectional relation between community-level social vulnerability and an individual’s health within that community is likely to exist, such that individuals residing in less resourced and more vulnerable areas may suffer from more adverse environmental exposure and higher chronic disease burden⁸³. Proteomics of the environmental exposure, in this case, may reflect the exposure and greater disease susceptibility. Mediation approaches have been advanced to begin to address these issues, though they are complex in the setting of multivariate omics data. In addition, pathway analyses here may be limited given joint variation in exposures across CARDIA center and clinical risk factors (not necessarily related to exposures), introducing the potential for confounding. Certainly, as noted above for PM_2.5, some proteomic associations may indeed be biologically plausible. Nevertheless, while “reverse causation” may challenge directional claims of causality among exposure, health, and the proteome, it does not detract from the notion that proteomics (1) provides a precision biomarker related to both the exposure and the health outcome at an individual level and (2) begins to underscore the joint importance of social determinants and the environment in CVD risk. Indeed, epidemiology has been a rich source of data in the exposome field⁸⁴, and future studies that map molecular phenotypes to the exposome an early stage in development—at a point where significant confounding (by decades of aging and exposure) may not supervene—will be a critical next step.

Of note, we observed best fit in proteomic models for social vulnerability and the greatest transportability of this proteomic signature for outcomes in both JHS and FHS, highlighting its importance in CMD. Higher social vulnerability—reflecting decreased community resourcing and resilience to natural, biological, or man-made emergencies (e.g., extreme weather, pandemic, floods, etc.)—has been linked to increased mortality and chronic disease burden⁸⁵, obesity (as early as childhood⁸⁶), and CVD death⁸¹. The association between the social vulnerability index with proteins reflecting inflammation, oxidative stress, and fibrosis—and the downstream relation of this proteome to phenotypes and/or outcomes in CARDIA, JHS, and FHS—suggest the importance of social determinants and community resourcing in CMD development. Importantly, in JHS, we observed associations between host proteomics of social vulnerability and neighborhood-level variables of cohesion, violence, noise, and poverty. These associations are consistent with reports linking this category of community-level exposures to heart disease³⁰ and support calls for enhanced community investment to interrupt CMD in at-risk communities.

The implications of our work should be viewed in the context of its design. Our study was a cross-sectional analysis of host proteomics and environmental exposure across four field centers. Variability across geographically distinct field centers was a unique feature that enabled variability in the envirome to facilitate regression with proteomics. Nevertheless, we attempted to address potential confounding due to differences in sociodemographic and clinical indices across centers in regression, though residual confounding remains possible. We used broad measures of the envirome which facilitates replication which are crude estimates of individual-level exposure. Future studies with serial measures and the inclusion of personal-level metrics (e.g., activity and dietary trends) would be helpful. Use of seasonal averages for temperature (e.g., 5-year summer average) may dilute extreme exposures and reduce meaningful differences in exposure to extreme weather. In addition, inclusion of additional environmental exposures (e.g., water pollution) would likely yield meaningful insights. With adjustment for sources of confounding thought to influence the proteome, we interpreted proteome-envirome relations as reflecting a “host profile” related to the environment. We are careful not to claim this is a true “response” to the environment, as there is no perturbation and pre- and post-proteomic/environmental measures. Nevertheless, it is important to note that randomized studies at large scale with perturbation in the environment are challenging in humans, and the pathways observed here are biologically consistent with smaller studies (e.g., air pollution). The pathway analysis is limited by the breadth of proteins surveyed, may be confounded by differences in clinical characteristics across center, and those results should be viewed as exploratory but does support the hypothesis that exposures are related to inflammation, angiogenesis, and cell cycle regulation. While the multivariable fit of proteomics to the environment is weak, it is in the range of genomic phenotypic predictors (≈10% variance explained), and the relations of each proteomic signature to phenotypes were stronger than the parent environmental exposure. The breadth of circulating proteins quantified in FHS and JHS (≈1300) was considerably smaller than in CARDIA (≈7000), requiring model recalibration. Importantly, our goal was not to identify surrogate measures of the envirome but rather to identify potential protein correlates. In that regard, there are concerns about the specificity of aptamer-based methods⁸⁷. Future studies incorporating a range of circulating molecules (proteomics, metabolomics, transcriptomics etc.) with precision exposure phenotyping in very large populations are necessary to elucidate molecular mediators, and their complex interactions, along the pathway of exposure to disease.

In conclusion, in a large community-based cohort of American adults, we employed broad host proteomics to define correlates of broad environmental exposures (the human “envirome”), demonstrating their relation to CMD phenotypes, biology, and long-term CMD outcomes in companion studies. These studies suggest a plausible framework for a serial tissue and circulating multi-omic study that integrates a broader human exposome with comprehensive biochemical readouts of exposure and disease susceptibility over time to examine the dynamic impact of the environment on human cardiometabolic health.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• Environmental exposures can have a physiologic impact on metabolism and heart disease.

• Broad population-based studies linking environmental exposure, high-dimensional molecular profiling, and cardiometabolic disease are not widely reported.

What New Information Does This Article Contribute?

• The relation between host cardiometabolic disease and the human “envirome” may be reflected in the circulating proteome, with pathways plausibly related to cardiometabolic disease.

• Proteomics of exposure were broadly related to cardiometabolic phenotypes, incident cardiovascular disease, and mortality.

While it is widely known that environmental exposures can impact human health, molecular markers to characterize these exposures are not well defined. Integrating broad molecular profiling with environmental exposures in 4 geographically distinct areas, we identified new and established pathways implicated in cardiometabolic disease development. Proteomic signatures of environmental exposures were related to cardiometabolic phenotypes and development of cardiovascular disease.

Nonstandard Abbreviations and Acronyms

CMD

cardiometabolic disease

PM_2.5

fine particulate matter with median diameter < 2.5 μm

PM₁₀

fine particulate matter with median diameter < 10 μm

CARDIA

Coronary Artery Risk Development in Young Adults

CVD

cardiovascular disease

FHS

Framingham Heart Study

JHS

Jackson Heart Study

O₃

ozone

NO₂

nitrogen dioxide

SO₂

sulfur dioxide

CO

carbon monoxide

EPA

Environmental Protection Agency

NLDAS

North America Land Data Assimilation System

NDVI

Normalized Difference Vegetation Index

NASA

National Aeronautics and Space Administration

MODIS

Moderate Resolution Imaging Spectroradiometer

SVI

Social Vulnerability Index

CDC

Centers for Disease Control and Prevention

LASSO

least absolute shrinkage and selection operato

FDR

false discovery rate

HDL

high-density lipoprotein

BMI

body mass index