Globally, millions of people are at risk for the detrimental effects of chronic arsenic exposure with drinking water levels far exceeding the WHO guideline [1
]. Prenatal arsenic exposure in human populations has been associated with pronounced long-term health consequences [4
]. Here, we address the impact of maternal arsenic exposure on fetal gene expression in a human population. Our goals were 2-fold: first, to establish the extent to which chronic arsenic exposure in mothers impacts newborn gene expression, and second, to identify genes that could be used as potential biomarkers of prenatal arsenic exposure and targets for remedial therapy.
Differential expression testing of training populations of newborns whose mothers had varied exposures to arsenic identified three arsenic-associated gene expression signatures comprised of 170, 38, and 11 genes. Analysis of the predictive capacity of each of these gene sets using the Support Vector Machine two-class prediction algorithm showed that each of these gene sets is highly predictive of arsenic exposure in a test population. Notably, even the smallest gene set comprised of 11 genes was powerful, with 83% accuracy in predicting prenatal arsenic exposure in the test population. The 11 potential biomarkers of prenatal arsenic exposure include CXL1
, and SOC3.
The set of 11 genes show a striking dose response to prenatal arsenic exposure. Stress response and cell cycle regulation are associated molecular functions of the potential biomarker set. Arsenic exposure is known to activate stress-related transcripts in yeast, animal models and human subjects [24
]. Here, we find that stress-response genes are differentially expressed among a population of newborns whose mothers were exposed to varying levels of arsenic.
To assess the genome-wide impact of prenatal arsenic exposure on newborn gene expression, we identified all transcripts that showed differential expression between two populations; the 21 newborns whose mothers had been exposed to arsenic versus the 11 newborns whose mothers were unexposed. These analyses identified ~450 genes differentially expressed between the two populations, of which 90% had expression levels that were increased (rather than decreased) by arsenic exposure. Clearly, there is a robust genome-wide response to prenatal arsenic exposure with ~3% of the expressed genes significantly altered in the newborn. Gene ontology and GSEA highlight the activation of stress-related transcripts in the cord blood of infants exposed prenatally to arsenic.
Furthermore, integration of the gene products of the ~450 transcripts with known molecular interactions identified the existence of a large arsenic-modulated interacting network of 105 proteins. Embedded within this large interacting network are three sub-networks that highlight that prenatal arsenic exposure activates inflammation-related molecules. Specifically, the first of the sub-networks centers around NF-κB and IL1-β. NF-κB regulates a large number of genes critical for apoptosis, as well as inflammation-related molecules such as cytokines (interleukins). IL1-β belongs to the class of acute phase proteins known to be increased in response to inflammation. Links between prenatal arsenic exposure and the activation of a stress response are also evident in the second and third sub-networks. Prenatal arsenic exposure resulted in the induction of the stress-related transcription factors STAT1 and HIF-1α, both of which are known to be activated by arsenic in model systems [19
]. Here, we identify STAT1 and HIF-1α activation in newborn cord blood upon prenatal arsenic exposure. The activation of stress-response proteins such as interleukin-8 (IL-8) in response to prenatal arsenic exposure is also evident in sub-network three. The gene expression signatures identified here as modulated by prenatal arsenic exposure were compared to arsenic-induced gene expression changes in the mouse model and also with a separate human population. These comparisons highlight the common pattern of activation of stress-related transcripts in response to arsenic exposure.
Additionally, eight of the 11 biomarker gene products were found to have significant interactions with the proinflammatory cytokine TNF-α. Several studies in animal models have shown that arsenic exposure results in TNF-α stimulation [27
]. In this study, TNF-α activation is identified in newborn cord blood upon prenatal arsenic exposure. Taken together, the network findings underscore that a mother's arsenic exposure results in a robust response in the fetus, indicative of a systemic inflammatory response along with the modulation of numerous other biological processes including apoptosis, signal transduction, cell adhesion, and transcription.
We further show that the extensive genome-wide newborn response to prenatal arsenic exposure may be regulated by at least three transcription factors. Analysis of the promoter regions of the arsenic-modulated genes showed enrichment for NF-κB and SRF in all four arsenic-modulated gene sets. SRF transcriptionally activates the expression of immediate early response genes, including C-FOS
], two members of the potential arsenic biomarker set. Moreover, binding sites for the metal-responsive transcription factor-1 (MTF1) are enriched in three of the four gene sets (sets 1, 3, and 4). MTF1 was shown to be activated upon arsenic exposure in animal models [23
]. That gene targets for a known arsenic-inducible transcription factor are found among the transcripts modulated in the cord blood of infants born to arsenic exposed mothers supports our conclusions that the transcriptional changes reported here are likely due to prenatal arsenic exposure.
Our findings clearly demonstrate the robust impact of a mother's arsenic consumption on gene expression in utero as evidenced by transcript levels in the newborn's cord blood. More specifically, our data suggest that prenatal arsenic exposure acts as an inflammatory stimulus that activates the NF-κB signaling cascade. NF-κB activation plays a critical role in inflammation-driven tumor progression [31
], and thus key players in tumor progression are modulated in the blood of newborns exposed to arsenic. To determine the extent to which these exposures and the resultant expression changes are associated with susceptibility to disease in later life, the health status of these children is currently being followed.
In summary, class prediction algorithms identified gene expression signatures that predict arsenic exposure in a test population with about 80% accuracy. Notably, by integrating training populations with varied exposures, a highly predictive potential biomarker gene set composed of just 11 genes was identified. These genes are promising as genetic biomarkers for prenatal arsenic exposure. Currently, we cannot eliminate the possibility that the gene expression signatures identified here are not absolutely specific for arsenic; they may also be predictive of other environmental exposures, e.g., exposure to other heavy metals. Nevertheless, this study underscores that there is a robust prenatal response that correlates with arsenic-exposure levels that could modulate numerous biological pathways including apoptosis, cell signaling, the inflammatory response, and other stress responses, and ultimately affect health status. Arsenic contamination of the drinking water in the Ron Pibul area of Thailand is representative of that seen in many other areas of South East Asia, most notably Bangladesh [9
], suggesting that prenatal exposures are likely to be endemic in these areas. Moreover, arsenic contamination of the Ron Pibul drinking water is roughly the same as that known to be present in many of the western United States [2
], suggesting that prenatal arsenic exposure may also be a problem in the United States. These data contribute to our understanding of biological responses upon arsenic exposure, and show that prenatal exposure in humans results in measurable phenotypic responses in the newborn.