Dynamic Human Environmental Exposome Revealed by Longitudinal Personal Monitoring

Abstract

Human health is dependent upon environmental exposures, yet the diversity and variation in exposures are poorly understood. We developed a sensitive method to monitor personal airborne biological and chemical exposures and followed the personal exposomes of 15 individuals for up to 890 days and over 66 distinct geographical locations. We found that individuals are potentially exposed to thousands of pan-domain species and chemical compounds, including insecticides and carcinogens. Personal biological and chemical exposomes are highly dynamic and vary spatiotemporally, even for individuals located in the same general geographical region. Integrated analysis of biological and chemical exposomes revealed strong location-dependent relationships. Finally, construction of an exposome interaction network demonstrated the presence of distinct yet interconnected human- and environment-centric clouds, comprised of interacting ecosystems such as human, flora, pets, and arthropods. Overall, we demonstrate that human exposomes are diverse, dynamic, spatiotemporally-driven interaction networks with the potential to impact human health.

Fig. 1.

Overview of the environmental exposome study. (A) A wearable device was modified to collect biotic (biological) and abiotic (chemical) compounds simultaneously from environmental airborne exposures, which were analyzed by NGS and LC-MS technologies, respectively. The schematic depicts the size of the employed filter in our study (red dot), relative to a conventional filter (grey). (B) The yearly trend of four variables measured by the device. Colored arrows denote the 3-month calendar seasons. (C) The sampling scheme for this study. P1 (green), P2 (dark blue), and P3 (red) were the most tracked. “Others” included samples from several individuals. (D) The sampling locations of P1, P2, and P3. (E) Representative SEM images of the sample filters. Top left, control filter; the rest are various particles identified based on morphology. Scale bars are indicated. (F) Total diversity of biotics from all samples. The pie chart indicates the total relative abundance of respective kingdoms/subkingdoms. The pan-domain phylogenetic tree is constructed from all identified species in the dataset. The total CPMs of individual species are plotted in the outer ring. (G) DNA and RNA viruses, including dsDNA, ssDNA, ssRNA, dsRNA, and retro-transcribing viruses were identified. Colored outer arc and the branches denote their respective natural hosts.

Fig. 2.

The highly dynamic and diverse human environmental exposome. (A) Timeline and (B) hierarchical clustered exposome profiles of samples from P1, P2, and P3 (top row) and all samples (bottom row), respectively. Proportions are calculated at the kingdom/subkingdom level. Ticks on arrows indicate months. (C-D) Same as A-B except for RNA exposome profiles. (E-F) Heatmap of DNA (E) and RNA (F) chronological exposome profiles at the phylum level. Colored bar denotes the domain of phylum, respectively. (G-H) Correlation plots between DNA and RNA exposome profiles at the kingdom/subkingdom (G) and the phylum (H) level. Correlations with Adj. p < 0.05 are shown. (I) Species richness analyses in DNA and RNA exposome profiles. (J) Functional transcriptomics analyses of the RNA exposome data. Domain-relevant GO terms are denoted by the colored bars next to the heatmap.

Fig. 3.

Decomposing the variation in the human environmental exposome. (A) Partial dbRDA variation partitioning analysis. (B) Ternary plots of variation partitioning analysis of highly prevalent genera (detected in >= 100 samples) in different domains of life. Environmental and spatial/lifestyle variables account for more than 80% of the explained variation in at least 75% genera. Each dot represents a genus and the size of the dot corresponds to the total explained variation. Depending on the genera, either environmental (blue) or spatial/lifestyle (dark yellow) variables may play dominant roles, or neither (grey). Contours denote 0.1 to 0.9 confidence intervals. (C) Samples from consecutive time points of P1 in the same location are more similar than those from different locations. (D) Representative differentially abundant genera between the “Campus” (N = 98) and non-”Campus” locations (N = 103). Boxes are color-coded as in Fig. 2A to denote the kingdom/subkingdom of the respective genus. (E) The location of the four individuals (P1, P3, P5, and P6) in the three-week parallel study. The size of dot corresponds to the self-reported activity level of each individual. Arrows indicate commute. (F) PCA analysis of P1, P3, P5 and P6. The bigger colored dots are geometric centers of respective groups. (G) Bray-Curtis distance profiles between samples from the same individual are more similar. (H) Top contributing genera with respect to the PCA analysis in (F). Color indicates relative contribution of each genus. All ellipses are drawn with axes equal to the standard deviation of the data. The Adj. p values are either directly displayed or denoted using the following notations * <0.05, ** < 0.01, *** < 0.001, and **** < 0.0001.

Fig. 4.

Seasonal influence on the human environmental exposome. (A) Nearest Neighbor (NN) tree constructed on the Bray-Curtis distance matrix calculated between all samples (nodes). If from the same season, nodes are connected by solid edges (pure), otherwise they are connected by dashed lines (mixed). Color denotes season. (B) Graph-based permutation test (N=9999) on the NN tree generated from (A), p = 0.0001. (C) Fuzzy c-means clustering of the genera abundance profiles. The four seasonal clusters are shown. (D) PCA analysis of genera abundance profiles, color-coded by clustering information from (C). (E) The temporal trends of four representative species based either on seasons (left) or months (right). (F) Heatmap of features selected by the regularized multi-class logistic regression model. Colored boxes highlight seasonal abundance profiles. Lollipop charts and the percentage information indicate the relative importance of each genus. (G) Stable internal performance of the season predictive model (resampling 10 times). (H) Macro-average performance metrics from the resampling data (10 times). (I) ROC curves calculated by one-vs-all approach using predictions from the resampling data (10 times). (J) Abundance profiles of representative genera selected for seasonal prediction. The Adj. p values are either directly displayed or denoted using the following notations * <0.05, ** < 0.01, *** < 0.001, and **** < 0.0001.

Fig. 5.

The abiotic exposome and its correlation with the biotic exposome. (A) Line plots of the abundance profiles of chemicals of the location-related cluster Cyan (left, N=84) and cluster Red (Right, N=228) as classified by the fuzzy c-means clustering. Each line represents a chemical feature. (B) Line plots of the abundance profiles of chemicals of the putative season-related clusters Green (N = 456) and Navy (N = 26). Transparency of each line (chemical feature) in (A) and (B) corresponds to the membership (>= 0.65). (C) PCA analysis of the abiotic exposome, colored by season. Note that as time progresses a major shift occurs for samples collected in the spring season. (D) Two anti-correlating chemicals potentially corresponding to different seasons. (E) Phthalate (cluster Cyan) is anti-correlated with geosmin, caprylic acid, and omethoate. (F) Several chemicals of interest show unique location-dependent patterns. (G) A chemical feature is positively correlated with several fungal species. (H) Pyridine, an organic solvent, is anti-correlated with multiple fungal species in a location-dependent manner. Colored boxes around chemicals denote their respective clusters. (I) PCA bi-plot of the sCCA-selected biotic and abiotic features. Samples collected from the “Campus” location are tightly clustered. Colored arrows denote the relative importance of contributing features. All correlations have Adj. p < 0.05.

Fig. 6.

Extensive pan-domain intra-species variation in the exposome. Population genetics analyses on top abundant species (N = 108) from Bacteria, Fungi, and Viruses domains. Archaeon Halococcus thailandensis and Oomycetes Phytophthora lateralis are also included. First row, SNP density, dashed lines denote 100, 10 SNPs/Kbp, respectively; second row, nucleotide diversity (n); third row, reference genome size, dashed line denotes 1e6 bps; and fourth row, average coverage of reference genomes. Dashed lines denote 100- and 500-fold coverage, respectively. All values are in log scale (base 10).

Fig. 7.

The bi-modal exposome interacting cloud. (A) The interaction network includes a plant-centric environmental cloud and a human-centric cloud. The intersection region between the two clouds is labeled by the dark yellow shade. The inset boxplots show that the species (detected in >= 50 samples) belonging to the Human/Animal group (human cloud) have significant less variance when compared to either Plant/Arthropods group (environmental cloud) or the Intersection group. Y-axis, log value of the variances of the aCPM values of analyzed individual species across all samples. (B) The individual exposome cloud of P1, P3, and P5 show diversity/complexity corresponding to their activity level. The number of average interactions (AI) is calculated at per node basis. The average path length (APL) is calculated by averaging the length of paths connecting any two nodes in the network. The Adj. p values are indicated as: N.S. Not Significant, * <0.05, ** < 0.01, *** < 0.001, and **** < 0.0001.