2000 Progress Report: Molecular Characterization of a Biological Threshold in Developmental ToxicityEPA Grant Number: R827445
Title: Molecular Characterization of a Biological Threshold in Developmental Toxicity
Investigators: Knudsen, Thomas B. , Charlap, Jeffrey H. , Craig, Robert C.
Current Investigators: Knudsen, Thomas B.
Institution: Jefferson Medical College
EPA Project Officer: Hahn, Intaek
Project Period: October 1, 1999 through September 30, 2002 (Extended to September 30, 2003)
Project Period Covered by this Report: October 1, 1999 through September 30, 2000
Project Amount: $207,170
RFA: Children's Vulnerability to Toxic Substances in the Environment (1999) RFA Text | Recipients Lists
Research Category: Children's Health , Health Effects , Human Health , Health
Objective:The goal of this research is to characterize, in molecular terms, the biological basis of a threshold in developmental toxicity and place this into context of a quantitative dose-response model for risk assessment. An embryo's response to xenobiotic exposure is likely determined by the pattern of genes expressed in precursor target cell populations. Therefore, critical pathway(s) are perturbed as the exposure-disease continuum approximates the threshold for dysmorphogenesis. DNA microarray chips allow investigation of multigenic responses of the embryo at the low end of the dose-response curve. Using functional genomics and computational biology, the studies described here aim to enumerate and classify toxicogenomic responses of the early mouse embryo to teratogen exposure. 2CdA (2-chloro-2'-deoxyadenosine) is the research prototype. This purine analog induces specific eye malformations such as cataract, corneal herniation, and micro-/anophthalmia with susceptibility predetermined by the embryo's genotype with respect to the p53 tumor suppressor gene.
Progress Summary:Studies during the first year of this grant focused on the dose-dependent effects of 2CdA exposure on global gene expression in the whole embryo. Use of DNA microarray chips extended the original research plan, which used reverse transcriptase polymerase chain reaction (RT-PCR) for specific target genes in the embryo. Pregnant outbred (CD-1) mice were dosed with 2CdA at 09.30 hours on day 8 of gestation (headfold stage). Test doses for microarray analysis were 0.625, 1.25, 2.5, and 5.0 mg/kg. Dose selection was guided by previous teratological evaluation: 1.5 mg/kg is the highest dose with no observable adverse effect (NOAEL); 2.5 mg/kg is the estimated dose for an extra 5 percent risk (BMD5) of microphthalmia; and 5.0 mg/kg induces micro-/anophthalmia in 24.4 percent of CD-1 mouse fetuses. Doses 1.25 and 2.5 mg/kg flank the threshold for developmental toxicity whereas 0.625 and 5.0 mg/kg represent exposures well to either side of this threshold. Embryos harvested at 12.30 hours provided RNA for microarray analysis. This post-treatment interval was based on our knowledge of p53 protein induction (270 min) and irrevocable pathogenesis (270-360 min). Reference controls used untreated embryos at the same developmental stage.
Methods: Twenty embryos were pooled in each microarray sample. Quality assurance of RNA was determined by absorbance A260/280 ratio of >1.7 and 1 percent agarose gel electrophoresis to rule out degradation of the isolated RNA and/or DNA contamination, and RT-PCR amplification of positive control genes (beta-actin, 16S ribosomal RNA). We isolated, for each measurement, 15-30 micrograms of total RNA. RNA (1 microgram) was converted to cDNA with incorporation of biotin-dUTP or dinitrophenol-dUTP to generate 800-1200 base labeled nucleotide fragments from the 3'-end of the transcript. Approximately 15-20 ng labeled target RNAs were competitively hybridized (56-59?C) to glass microarray probes printed with 2400 sequence-verified cDNAs (Micromax-1, NEN Life Sciences). The genes represent a cross-section of human diseases, metabolic and regulatory pathways, and control spots. Histochemical detection used peroxidase-conjugated anti-dinitrophenol and cyanine-3-tyramide (Cy3) reaction followed by peroxidase-conjugated streptavidin and cyanine-5-tyramide (Cy5) reaction. Fluorescent images of Cy3 and Cy5 channels were generated with a laser confocal scanner (ScanArray 5000) at 10-micron pixel resolution and analyzed with QuantArray software (GSI Luminomics). Units of intensity for background-corrected signals encompassed three log scales of dynamic response. The results included in this progress report derived from analysis of eight microarray chips.
Figure 1. Competitive fluorescent hybridization of embryonic RNA with Micromax-I microarray probe (59oC). Test RNA was procured from the headfold of day 8 embryos 3h after intrauterine exposure to 5.0 mg/kg 2CdA. Reference was control mouse embryos at the same developmental age. (Note - replacing the dinitrophenol labeling reaction with FITC improved the specificity of the data considerably; however, this report will use data from the original formulation).
Data Analysis: A reasonable microarray signal corresponded to net signal intensity >200 units and signal/background ratio >1.25. For a spot to pass a quality control threshold, both channels must pass the net signal criterion and at least one channel must pass the signal/background criterion. Hybridization to each spot showed systematic variation depending on the specific hapten labeling scheme (e.g., DNP v. biotin). To correct for this technical variable, each experiment was repeated with reversal of the labeling scheme (e.g., DNP-test/biotin-reference v. DNP-test/biotin-reference). Test and reference RNA samples for each replicate in the dye reversal scheme used independent biological samples to improve confidence. Post-hybridized images of the microarray probe were inspected for aberrations in spot morphology that could affect data quality. Genes counted as "present" passed the quality control threshold in at least seven of the eight microarray chips (>88 percent). In total, 1711 genes met this criterion. This represents approximately 5-6 percent of the transcriptome assuming 30,000 genes expressed at the headfold stage of development. Data from each microarray channel was normalized to a net signal = 5000. A correction factor was systematically applied to balance the Cy5 channel for differences due to labeling bias.
1. Does the embryonic transcriptome respond to exposure in a dose-dependent manner that can be correlated with increased risk for disease?
Understanding toxicogenomic responses associated with physiological mechanisms linking environmental exposures to disease requires data reduction standards and practices that have not yet been unified in the field. Because changes in gene expression reflect the primary vulnerability of embryonic systems to toxic insult as well as cellular adjustments that may or may not contribute to the disease process, an immediate goal was to score gene expression profiles as a function of 2CdA dose. The differentiality profile at 3.0 hours post-treatment showed dose dependence in both complexity and magnitude (Figure 2). We saw minimal effects with 0.625 mg/kg 2CdA and a broad response at the teratogenic dose level (5.0 mg/kg). Doses flanking the developmental NOAEL were intermediate in their effect on the embryonic transcriptome. The majority of expression values for the 1711 genes fell within 2.5-fold differential between control and exposed embryos. Outliers (114) selected for 2.5-differential expression with arithmetic averaging numbered as follows: 0.625 mg/kg (8); 1.25 mg/kg (6); 2.5 mg/kg (66); and 5.0 mg/kg (39).
Figure 2. Log-log graph showing the signal values from a comparison of mouse embryos at 3 hours post-exposure (2CdA) and untreated control embryos on day 8 of gestation. Genes were 1711 that passed the quality control and expression criteria. The majority of expression values are within 2-fold, whereas outliers increased with 2CdA doses above the developmental NOAEL).
In a microarray experiment, the interpretation of ratiometric data as upward and downward regulation is consistent with the use of geometric averaging, rather than arithmetic averaging of ratiometric data. Given two experiments, one showing five-fold increase and the other showing one-fifth the expression level, the mean should be 1.0 or no change by geometric averaging as opposed to 2.6 by arithmetic averaging. Use of geometric averaging produced a more robust and quantitative dose response effect than did simple arithmetic averaging. This was shown by transforming the fundamental data to base-2 log scale, averaging the replicate data sets, and standardizing to mean = 0.000 and standard deviation q, where q was extracted from an exponential regression model of the standard deviation v. dose function (r2 = 0.99). Outliers (153) selected for 2.5-fold differential expression with geometric averaging numbered as follows: 0.625 mg/kg (0); 1.25 mg/kg (24); 2.5 mg/kg (57); and 5.0 mg/kg (80). Data for the 1711 genes were subjected to hierarchical modeling using a similarity metric based on standard Pearson correlation and distances computed by average linkage clustering (Gene Cluster software, firstname.lastname@example.org). Setting a high variance filter captured few outliers (e.g., 26 genes at 3.0-fold) whereas setting the variance filter too low produced an unmanageable number of outliers (e.g., 740 genes at 1.0-fold). A reasonable balance was achieved when the variance filter was set at 1.5-fold. Visualization of these 331genes with Tree View software (email@example.com) revealed continuous, threshold, and disease dose-response signature behaviors (Figure 3).
Figure 3. Toxicogenomic visualization of a teratogenic threshold in early mouse embryos procured 3 hours after 2CdA exposure on day 8 of gestation
Tree View visualization of clustered microarray data.
Clustergram of 331 genes in day 8 mouse embryos 3.0 hours after exposure to
different dose levels of 2CdA and clustered on the slope of their dose response
using a similarity metric based on the standard Pearson correlation. Color
intensity in the expression map measures the relative difference between values
on the same side of zero (black); red genes upwardly expressed in 2CdA-treated
embryos and green signifies downwardly expressed. (A) computational control; (B)
0.625 mg/kg 2CdA; (C) 1.25 mg/kg 2CdA; (D) 2.5 mg/kg 2CdA; and (E) 5.0 mg/kg
2. Does the embryonic transcriptome reveal a quantitative dose response signature in developmental toxicity?
We used computer generated pattern recognition to find the commonality and differentiality of gene expression profiles. Clustering of gene expression data makes sense when associated with a known biological process or replicated to the extent that it can be shown statistically significant. Because the biological significance of primary gene expression profiling data most likely increases with the number of conceptual linkages, one finds inside a suite of genes, we refined the 331 gene set as follows: (1) selecting genes with assigned molecular functions that appeared more than once in the data set; (2) partitioning these genes into distinct cellular processes and grouping them as up-/down-regulated with 2CdA; (3) subjecting these data to two-tailed paired t-test comparison; and (4) eliminating groups not showing a significant effect at any 2CdA dose level. We also removed, arbitrarily, all ribosome subunits from consideration due to shear volume. The result of this complex computational algorithm is shown in Table 1. 2CdA generally stimulated cellular metabolic and regulatory pathways as evidenced by the preponderance of the variance accounted for by genes up-regulated at 3.0 hours post-exposure. For example, of 165 genes identified in this analysis, 124 of them (75 percent) partitioned into cellular processes that showed 2CdA dose-dependent up-regulation of gene expression. Furthermore, this stimulatory effect of 2CdA became significant at the 1.25 mg/kg dose level for gene groups in cellular metabolism, signal transduction pathways, and cell surface-structure, and at the 2.5 mg/kg dose level for gene groups in transcription, translation, and protein sorting. We conclude that disease manifestation (microphthalmia) correlated with a quantitative dose response at dosages encompassing the biological threshold in developmental toxicity.
Table 1. Relational Database Enriched for Responsive Genes Partitioned by Cellular Process
3. What metabolic and regulatory pathways correspond to a "molecular benchmark signature" for 2CdA developmental toxicity?
With the hope of using disease signatures to improve our understanding of how
risks of exposure to environmental contaminants are assessed, important
questions pertain to our own abilities to interpret the commonality and
differentiality of complex multigenic responses within the context of chemical
mode of action. With the help of computer pattern recognition, we selected
suites of genes from the relational database (Table 1) having concerted
molecular function and co-expression with disease potential. Preliminary
analysis identified nine responsive genes associated with ocular defects in
human infants or experimental animals. These included two group IV Pax genes
(Pax6, Oclr) for microphthalmia that are both
down regulated with 2CdA exposure. The drop in Pax6 levels correlated
with independent analysis of Pax6 protein in the murine optic vesicle of
2CdA treated embryos (Figure 4). Other candidate genes for ocular
dysmorphogenesis, all up regulated with 2CdA exposure, encompass phenotypes such
as corneal dystrophy (amyloid, MSS1), congenital cataract
(GalK), and micro-/anophthalmia (TCOF1, RXR-beta,
AF-9). If we define "molecular benchmark signature" as the collective set
of changes that characterize developmental toxicity, then suites of genes
clustering with the candidate disease genes should provide a more complete
picture of 2CdA's mode of action.
Figure 4. Candidate genes for ocular disease perturbed by 2CdA exposure on day 8 of gestation.
LEFT: Immunoperoxidase staining of the optic vesicle (OV) and preplacodal lens ectoderm (PLP) of control day 9 mouse embryo (left section) and embryo exposed to 10 mg/kg 2CdA on day 8 of gestation (right section). Diminuitive Pax6 staining phenocopies the pathogenesis of micro-/anophthalmia in the small eye mouse mutant and the blind cavefish.
RIGHT: Tree View clustergram of 51 genes with 9 ocular disease genes expressed in early mouse embryos exposed to different dose levels of 2CdA and clustered using a similarity metric based on the uncentered Pearson correlation. Columns were: (A) computational control; (B) 0.625 mg/kg 2CdA; (C) 1.25 mg/kg 2CdA; (D) 2.5 mg/kg 2CdA; and (E) 5.0 mg/kg 2CdA.
At present, there is no easy way to reconstruct cellular pathways from gene
expression data. Bioinformatics tools used for pathway reconstruction include
the Kyoto Encyclopedia for Genes and Genomes (KEGG) and related database links
including PubMed and MedLine. Such an analysis of the relational database
revealed for nucleotide metabolism, cellular redox control, beta-oxidation, and
steroid biosynthesis among the clusters of candidate genes for ocular
dysmorphogenesis. For example, 2CdA exposure stimulated expression of the de
novo pathway of purine and pyrimidine metabolism as evidenced in the
increase of GARS, AICAR, CAD, and ADK2. Although
this may partly reflect biochemical imbalances associated with 2CdA entry into
cellular nucleotide pools, the involvement of purine and pyrimidine de novo
genes suggests an adaptive response to generalized nucleotide imbalances. This
is supported by increased expression of thiopurine methyltransferase
(TPMT), an enzyme known to directly influence cellular susceptibility to
cytotoxic nucleoside analogs. Low doses of 2CdA up regulated several redox
cycling genes (TRR, Q6, Gpx, cyt-c1) and redox
response genes (ref-1, CaMKII, RC3). This was supported by
increased expression of lactoyl glutathione lyase (GLO1), a phenomenon
seen with phenytoin, and XRCC1 to repair oxidative damage of DNA. It also
is worth noting up regulation of several genes in the Alzheimer's disease
pathway (Fe65, KGDHC, MSS1, APP, alpha2-
macroGR). We speculate that oxidative stress is a penultimate signal for p53
protein induction. Yet, the largest assembly of responsive genes included
genetic regulators (SREBP-1, TIF2, TAFII68), enzymes
(Epxh-1, cyt-b5, GLO1, ACL, ICDH, LS,
KGDHC), and subcellular transport (cav-1, MLN64,
RXR-beta, TR3 orphan receptor, TRAC) for pathways in
beta oxidation and steroid biosynthesis. Nearly all of these genes increased
with 2CdA exposure to imply a steroidogenic effect, an effect novel to our
present understanding of 2CdA. Finally, several cell signaling pathways could be
partly reconstructed from genes in the relational database. Down regulated cell
signaling systems included the receptor serine/threonine kinase TGF-beta
receptor pathway, whereas the receptor tyrosine kinase pathway generally
increased. Genes affiliated with fundamental morphogenetic processes (cell
growth, apoptosis, shape, adhesion, and motility) had adequate representation in
the relational database although without clear linkage to signaling
In conclusion, gene expression profiling data derived a preliminary "molecular benchmark signature" for a biological threshold in 2CdA-induced microphthalmia. Risk assessment methods commonly used to evaluate the developmental toxicity of environmental exposures have traditionally focused on identification of the NOAEL from disease endpoints. Gene expression profiling showed a surprisingly conservative response of the embryonic transcriptome as the dosage approaches the NOAEL. The robust response invoked by teratogenic exposure is presumed to have reflected primary vulnerability to toxic insult, cellular adjustments at the limit of tolerance to intermediate dosages, and the disease process itself. Complex relationships such as these must still be sorted out in terms of distinct molecular functions and biological processes in order to simulate and model the disease pathways.