Fig 1.
Biomarker construction and screening strategy.
Left, NRF2 biomarker development. Differentially expressed genes (DEGs) from exposures to known NRF2 activators (sulforafan, oltipraz, sulindac, quercetin) and KEAP1 and NFE2L2 siRNA knockdown were assessed in the BSCE environment. Biomarker genes were identified as those consistently up- or down-regulated by the chemical exposures or KEAP1 knockdown and opposing regulation in NFE2L2 knockdown (further details in Results section). Ten biomarkers were initially created from biosets that varied based on chemical, time and dose of exposure, and tissue context.
Table 1.
Biosets used to build the NRF2 biomarker.
Fig 2.
Predictive accuracy of the NRF2 biomarker.
A. Predictions using the NRF2 biomarker were compared to those from the HTS NRF2 activation assays. The dashed line represents the cutoff value of four for significant biomarker activation. Those chemicals that were active (red) or inactive (black) in the HTS assays are shown. Three false positives (blue) and one false negative (yellow) are indicated. Inset, binary classifier calculations based on NRF2 HTS assay results. B. Expression changes of the biomarker genes for the three false positives and one false negative. (Top) The bar graph shows the -Log(p-values) of the comparisons between the biomarker and the indicated chemical. (Bottom) The heatmap shows the expression of the biomarker genes.
Fig 3.
Characterization of the human NRF2 biomarker.
A. Identification of NRF2 biomarker genes was based on consistent directional changes in gene expression resulting from exposure to NRF2 activators sulforaphane (GSE20479 and GSE28813), oltipraz (GSE20479), sulindac (TG-GATES), and quercetin (GSE28878), and knockdown of the NRF2 negative regulator Keap1 (GSE28813). Changes in gene expression were required to be in the opposing direction in cells in which NFE2L2 (NRF2) expression was knocked down (GSE38332). The bars on the left represent the–Log (p-values) for the Running Fisher comparison test between the biomarker and the individual biosets used to construct it. The heatmap on the right depicts gene expression changes for the 143 genes in the biomarker across the individual biosets. B. Canonical pathway analysis of the biomarker genes. Multiple test correction on p-values derived from the right-tailed Fisher’s exact tests was carried out using the Benjamini-Hochberg method. The -Log(q-value)s are shown. C. Potential upstream regulators of NRF2 biomarker genes (activation z-scores > |2|). Only upstream regulators with q-values < 0.05 are shown. Both analyses were conducted using Ingenuity Pathway Analysis software.
Fig 4.
NRF2 activity across biosets from chemically treated human cells.
A. (Top) Biosets derived from microarray comparisons of human cells exposed to chemicals were rank ordered based on their correlation to the biomarker using the -Log(p-value) of the Running Fisher test. Biosets with positive correlation (red) to the biomarker are on the left and biosets with negative correlation (green) to the biomarker are on the right. The dashed lines denote the cutoff p-value = 10−4. (Bottom) The heat map shows the expression of genes in the biomarker across the biosets. B, biomarker genes. The numbers refer to rank-ordered bioset number. B. Top 20 chemical biosets that activate NRF2. Right, heatmap depicting the gene expression changes for the 143 genes in the NRF2 biomarker in each bioset. The biomarker gene expression changes are represented across the bottom of the heatmap. Each bioset is represented by the chemical, time and concentration of exposure, cell line used, and annotated study. One study did not have all information available (GSE13818). Abbreviations: BaP, benzo[a]pyrene; 2CMP, 2-(Chloromethyl) pyridine hydrochloride; NPD, 4-Nitro-o-phenylenediamine; CLEFMA, 4-[3,5-bis(2-chlorobenzylidene-4-oxo-piperidine-1-yl)-4-oxo-2-butenoic acid]; SFN, sulforafan; I3C, indole-3-carbinol. C. Top 20 chemical biosets that suppress NRF2. Abbreviations: 4AAF, 4-acetylaminofluorene; 17DMAG, 17-(dimethylaminoethylamino) 17-demethoxygeldanamycin; CCl4, carbon tetrachloride; Mixture, a mixture of liver toxicants; DMBA, 7,12‑Dimethylbenz[a]anthracene.
Table 2.
Chemicals enriched in NRF2 active biosets.
Fig 5.
Examples of dose-, time-, and mutation-dependent modulation of NRF2.
A. Response to 2 μM benzo[a]pyrene from 12 to 72 hours in HepG2 cells. Red circles indicate NRF2 activation, blue circles indicate inactivity. Data from GSE28878, GSE36242, GSE36243, GSE40117. B. Exposure of HepG2 cells to 0.01 μM and 50 μM quercetin from 12 to 48 hours. Data from E-MEXP-2574, GSE28878. C. Exposure of human primary hepatocytes to 10, 50, and 250 μM diazepam from 2 to 24 hours. Data from the TG-GATES study. D. Exposure of lung fibroblasts to dithiothreitol (2.5 mM) at different time points. Data from GSE4301. E. Examples of chemicals suppressing the activation of NRF2 in cancer cells expressing activated PI3K and EGFR. The activating mutations in PI3K3CA (E545K and H1047R) and overexpression of EGFR in the presence of EGF (caEGFR) lead to activation of NRF2 compared to wild-type cells. HCC827 with EGFR Del15 cells possess an amplified EGFR allele with an activating in frame deletion of 15 nucleotides in exon 19. Treatment of the cells with inhibitors for PI3K (LY294002) or EGFR (erlotinib) in the indicated cells suppresses background NRF2 activation.
Fig 6.
Biomarker prediction agreement with HTS NRF2 assay results.
Maximum biomarker scores for each chemical were extracted from the database of HepG2 and hepatocyte experiments, and activity calls based on these scores were compared with activity determinations from ToxCast and Tox21 NRF2 HTS assays. Chemicals examined at a concentration above the HTS-determined cytotoxicity threshold for that chemical were excluded from the analysis. Lists represent those chemicals classified as NRF2-active by the biomarker only (left) or by HTS only (right). Biomarker score values shaded in red indicate results that are ≥ -Log(p-value) = 4 for activation of NRF2 signaling.
Fig 7.
NRF2 activity of selected chemicals in HepG2 cells.
Thirty chemicals tested for confirmation in the SRXN1-GFP assay. Cells were treated at multiple concentrations and examined for GFP accumulation over a 24-h period (GFP_pos_1–10) for concentrations 1 to 10. The last two columns, PI_pos_10 and cell_count_10, are the Propidium Iodide positive fraction of cells (marker for necrosis/ dead cells) and cell count compared to the negative control, respectively.
Table 3.
Comparison of chemical effects using the biomarker and the SRXN1-GFP assay.