Population-based in vitro hazard and concentration-response assessment of chemicals: the 1000 genomes high-throughput screening study

Abstract

Background: Understanding of human variation in toxicity to environmental chemicals remains limited, so human health risk assessments still largely rely on a generic 10-fold factor (10½ each for toxicokinetics and toxicodynamics) to account for sensitive individuals or subpopulations.

Objectives: We tested a hypothesis that population-wide in vitro cytotoxicity screening can rapidly inform both the magnitude of and molecular causes for interindividual toxicodynamic variability.

Methods: We used 1,086 lymphoblastoid cell lines from the 1000 Genomes Project, representing nine populations from five continents, to assess variation in cytotoxic response to 179 chemicals. Analysis included assessments of population variation and heritability, and genome-wide association mapping, with attention to phenotypic relevance to human exposures.

Results: For about half the tested compounds, cytotoxic response in the 1% most "sensitive" individual occurred at concentrations within a factor of 10½ (i.e., approximately 3) of that in the median individual; however, for some compounds, this factor was > 10. Genetic mapping suggested important roles for variation in membrane and transmembrane genes, with a number of chemicals showing association with SNP rs13120371 in the solute carrier SLC7A11, previously implicated in chemoresistance.

Conclusions: This experimental approach fills critical gaps unaddressed by recent large-scale toxicity testing programs, providing quantitative, experimentally based estimates of human toxicodynamic variability, and also testable hypotheses about mechanisms contributing to interindividual variation.

Figure 1

(A) Distribution of the lymphoblastoid cell lines (LCLs) used in this study among the nine populations. Outer boundaries show continental/ancestral origin. (B) Scatter plot for the 1st and 2nd principal components for genotypes across all cell lines; colors represent populations shown in (A).

Figure 2

(A) Comparison of the present study with other comparable lymphoblastoid cell line (LCL) cell line/screening studies, in terms of the number of cell lines and chemicals screened. EC₁₀ values are shown in the heat map (top), and the area of each report is shown in proportion to the present study (bottom); the numbers of cell lines and compounds used in the published studies are listed in Supplemental Material, Table S2. (B) Intraexperimental reproducibility of EC₁₀ values for randomly selected pairs of within-batch replicate plates for all chemicals and cell lines. (C) EC₁₀ values for nine compounds assayed in two independent sets of wells on each plate, shown as side-by-side box plots. Boxes represent interquartile range, lines within boxes are medians, whiskers represent values 1.5*(interquartile range) from the first and third quartiles, and circles indicate outliers. (D) Box plot showing variation of cytotoxicity EC₁₀ values for the 179 chemicals (arranged by mean activity) across the 1,086 cell lines.

Figure 3

(A) Modeling in vitro quantitative high-throughput screening data, using β-nitrostyrene as an example chemical. Logistic dose–response modeling was performed for each individual (plate), as shown by thin gray lines. Bars represent individual EC₁₀ values, and the dashed curve represents the fit of the logistic model to the pooled data. EC₁₀ estimation based on this curve is similar to the average of the individual EC₁₀ values. (B) Histogram (bars) of the toxicodynamic variability factor 10^(q50–q01) for 149 compounds across 1,086 cell lines. The curve shows the same distribution when values are shrunken to account for technical variability; for the 30 compounds not shown, estimated technical variability was too large to calculate a shrunken factor. The inset shows the relationship between range and median estimated EC₁₀ for each chemical. (C) Cumulative distribution functions for the in vitro toxicodynamic variability factor shrunken to account for technical variability 10^{(q50–q01)/VIF1/2}, across 149 compounds (present study) and the human in vivo toxicodynamic variability factors across 34 compounds (IPCS 2014). (D) Hierarchical clustering for the 179-length profiles of mean EC₁₀, computed within each population, and shown by continental ancestral origin of the population. AM, Americas. (E) Box plot of EC₁₀ values by population for two example chemicals with different potency levels, showing significant population differences by analysis of variance (top; cycloheximide, p = 6.0 × 10^–6; bottom; triamterene, p = 3.6 × 10^–4). Boxes represent interquartile range, lines within boxes are medians, whiskers represent values 1.5*(interquartile range) from the first and third quartiles, and circles indicate outliers. (F) Trio-based heritability estimates (h²) for compounds with evidence of additive heritability (22 chemicals with p < 0.05 are shown, the top 17 having q < 0.2).

Figure 4

(A) Manhattan plot of MAGWAS –log₁₀(p) versus genomic position for association of genotype and cytotoxicity to 2-amino-4-methylphenol. The green dashed line indicates the significance threshold for suggestive association (expected once per genome scan), and the black dotted line represents the Bonferroni-corrected significance for a single chemical. (B) LocusZoom plot of the most significant region. Abbreviations: cM, centimorgans; Mb, megabase pair. SNP rs13120371 was the most significant (p = 8.4 × 10^–10), and the nearby rs7674870 was used for comparison of linkage disequilibrium patterns in the region. See Supplemental Material, Figure S5, for color heat maps of the significance association of the individual SNPs. (C) Average concentration–response profiles of cytotoxicity of 2-amino-4-methylphenol plotted separately for each rs13120371 genotype (AA, AT, and TT); genotype effects were observed only for the highest concentrations. (D) Histogram of EC₁₀-based p-values for all 179 chemicals for rs13120371, showing an excess of small p-values.