Technical variation is often ignored in human toxicogenomic studies, leading to potential bias in differential expression arising from correlation with technical variation. In the present study, we applied a rigorous study design to assess sources of both potential confounding and experimental variability (nuisance variation) and analyzed the data using statistical techniques that incorporate nonnested sources of variation (i.e., those not eliminated by normalization) and that return estimates of least variability with accurate inference (linear mixed-effects models). This approach increased the power to detect associations between benzene exposure and gene expression, even at low-dose exposure levels.
More genes remained significantly up- or down-regulated compared with controls after multiple test correction in the present study than in an earlier study examining samples from eight pairs of exposed workers and unexposed controls on the Illumina platform (McHale et al. 2009
), likely because of the increased number of individuals and the rigorous approach to study design. Nonetheless, we identified 247 genes in both study populations using the Illumina platform. Of 488 significant genes cross-validated on both Illumina and Affymetrix platforms (McHale et al. 2009
), 147 genes were significant in the present study. ZNF331
(zinc finger protein 331), significant after multiple test correction in individuals occupationally exposed to benzene at levels > 10 ppm compared with controls in two earlier studies (Forrest et al. 2005
; McHale et al. 2009
), was significantly up-regulated at both < 1 ppm and > 10 ppm in the present study.
The finding that genes in the AML pathway were strongly associated with multiple exposure levels of benzene provides support for our approach because epidemiological studies have established that benzene causes AML (Baan et al. 2009
; Smith 2010
). However, such disease associations must be treated cautiously because the KEGG pathway information, on which the pathway analyses were based, is limited for AML, and a KEGG pathway for NHL has not been defined. Information about altered molecular and cellular processes can provide biological plausibility for probable disease associations. Immune response, previously found to be associated with > 10 ppm benzene exposure in our earlier transcriptomic study of eight high-exposed control pairs (McHale et al. 2009
), was one of the major processes significantly altered across multiple exposure levels in the present study, involving both innate (Toll-like receptor signaling) and adaptive (B-cell receptor signaling and T-cell receptor signaling pathway) responses. Additionally, we found central roles for the proinflammatory cytokines IFNG
among genes uniquely altered at low-dose exposure in the present study. A single nucleotide polymorphism in TNF
-α was previously associated with susceptibility to bone marrow dysplasia in chronic benzene poisoning (Lv et al. 2007
). Further, genetic variation in TNF
(Rothman et al. 2006
), Toll-like receptor genes (Purdue et al. 2009
), and IFNG
(Colt et al. 2009
) has previously been associated with NHL risk. Deregulation of pathways involving these genes through sustained alterations in expression provides biological plausibility for the association of benzene with lymphoid neoplasms.
Findings from the present study are consistent with previous reports of adverse effects of benzene on oxidative stress (Kolachana et al. 1993
) and mitochondria (Inayat-Hussain and Ross 2005
). Here, we found highly significant associations with ATP synthesis–coupled proton transport and oxidative phosphorylation at all levels of benzene exposure relative to unexposed controls. Expression of superoxide dismutase (SOD
), a mitochondrial defense against reactive oxygen species, was up-regulated in the present study by 50–100% relative to controls. HMOX1
[heme oxygenase (decycling) 1], an antioxidant and suppressor of TNF-α signaling (Lee et al. 2009
), was down-regulated in the low-dose benzene exposure group. Increased mitochondrial membrane permeability potential induced by benzene metabolites (Inayat-Hussain and Ross 2005
) can lead to the initiation of apoptosis. Indeed, apoptosis was associated with all benzene doses in the present study, consistent with our earlier observation of an association with high-dose benzene exposure (> 10 ppm) (McHale et al. 2009
Previously, we found that chromatin assembly was significantly altered after high-dose benzene exposure (McHale et al. 2009
). The finding that nucleosome assembly (a GO category nested within chromatin assembly) was overrepresented in the highest exposure category in the present study confirms and clarifies this potential mechanism of benzene-associated leukemia.
Although significant involvement of the p53 response pathway was previously found in mice exposed to very high levels of benzene (Faiola et al. 2004
; Yoon et al. 2003
), we did not find such involvement in the present study or in our earlier studies, and the immune and inflammatory effects we found here in humans were not recapitulated in the mouse microarray studies (Faiola et al. 2004
; Yoon et al. 2003
). These differences suggest that human toxicogenomic studies may be more relevant than animal studies, although differences in exposure levels, tissues examined, and uncontrolled confounding in the human study could also be contributing factors.
Our findings suggest two novel hypotheses regarding benzene toxicity. Glycosylphosphatidylinositol (GPI)-anchor biosynthesis was associated with all doses of benzene exposure in the present study. The GPI anchor is a C-terminal posttranslational modification that anchors the modified protein in the outer leaflet of the cell membrane and putatively plays roles in lipid raft partitioning, signal transduction, and cellular communication (Paulick and Bertozzi 2008
). Because epigenetic silencing of genes involved in GPI-anchor biosynthesis may be important in human disease, including lymphomas (Hu et al. 2009
), further investigation of its role in benzene-associated disease is warranted.
ABC transporters were associated highly significantly with only the highest (> 10 ppm) benzene dose. In addition to their capacity to extrude cytotoxic drugs, ABC transporters are known to play important roles in the development, differentiation, and maturation of immune cells and are involved in migration of immune effector cells to sites of inflammation (van de Ven et al. 2009
Our findings also suggest a potential gene expression signature of benzene exposure. In particular, IL1A
played central roles in the interaction network characterizing the gene expression signature associated with benzene in this study. Both molecules are produced by activated macrophages and other cells in inflammatory responses. A single nucleotide polymorphism that increases IL1A
mRNA expression has been inversely associated with granulocyte count in benzene- exposed individuals (Lan et al. 2005
). Overexpression of PTGS2
, which occurs frequently in premalignant and malignant neoplasms, including hematological malignancies (Bernard et al. 2008
), together with overexpression of the prostaglandin cascade, leads to carcinogenesis through a progressive series of highly specific cellular and molecular changes (Harris 2009
The expression pattern of the signature genes suggests a nonlinear response to benzene. Other biomarkers evaluated in populations exposed to benzene have shown similar patterns, including hematotoxicity (Lan et al. 2004
), benzene metabolism (Kim et al. 2006
), and the generation of protein adducts (Rappaport et al. 2002
). Further characterization of the expression levels of these genes across a range of benzene exposures in a larger, independent study is necessary to determine the applicability of the signature genes as biomarkers of early effects and to explore more formally the shape of the dose–response curve.