In all 3 studies presented here, Se deficiency did not significantly affect growth, as observed in previous studies [4, 5]. In rat study 2, however, a high Se intake of 5 μg Se/g diet significantly decreased growth as compared to intakes ≤ 2 μg Se/g diet. Growth defects and other symptoms of Se toxicity resulting from intakes of 3 to 5 μg Se/g have been reported in previous studies [1, 31–33]. Although growth was not affected by Se deficiency, liver Se, liver and RBC Gpx1 activity, and plasma Gpx3 activity were all dramatically decreased in Se-deficient mice and rats as compared to Se-adequate animals. Gpx1 mRNA was also significantly down-regulated in Se-deficient tissues, 2.1 to 3.7 fold, and mRNA levels for several other selenoproteins were significantly down-regulated by Se deficiency, as we previously reported [8, 26]. Thus, the Se-deficient animals were Se-deficient at the biochemical and molecular level, but otherwise indistinguishable from Se-adequate animals. Liver Gpx1 activity was not further increased by super-nutritional Se, and was slightly decreased by 5 μg Se/g as compared to the Se-adequate diet (0.24 μg Se/g), most likely due to liver damage.
Microarray analysis of >30,000 transcripts in the liver of rats fed Se deficient to 5 μg Se/g diet found that an intake of 5 μg Se/g is required to alter the expression of a large set of transcripts. This is the first report of a large and distinct transcriptional response profile induced by Se toxicity. In contrast, Se intakes less than 5 μg Se/g diet significantly changed < 10 transcripts as compared to a Se-adequate intake within an experiment. The distinct transcriptional effect of 5 μg Se/g correlates with growth and ALT activity, the other markers of toxicity measured in this study. The vastly expanded number of gene expression changes observed in this study thus is a newly-identified marker of Se toxicity. Importantly, as many as half of the 1193 transcripts altered by the 5 μg Se/g diet treatment may be Se-specific, as 715 transcripts still remained after removing transcripts responding to general toxicity and caloric restriction.
Previous microarray studies have investigated the transcriptional effects of Se supplementation up to 1.0 μg Se/g diet in rodents, of up to 200 μg Se/day in human subjects, and of up to 22 μM Se in cultured cells. A search of the Se-regulated genes in 9 previous Se microarray studies [11–18, 34] found only a few genes that were identified in more than 1 study. The only genes regulated by >3 studies, however, were selenoproteins, suggesting that most of the genes identified were not responding specifically to high Se. Many of these studies are complicated by the fact that high Se treatments were only compared to Se-deficient treatments, so that many of the transcriptional changes are likely to be the result of Se deficiency instead of Se supplementation. One study that did compare a high Se treatment to a Se-adequate treatment (1.0 μg Se/g vs. 0.2 μg Se/g) , found no significantly altered genes. Based on the present study, it is very likely that the lack of Se-specific regulation in these previous studies is due to insufficient Se to cause large, significant transcriptional responses. Additional studies with near toxic Se levels may be necessary to determine whether the transcripts identified in the present study are truly Se-specific.
The levels of high Se in the present studies (8, 20 and 50-times the Se requirement) are relevant to high Se intakes in humans. A study of Chinese subjects living in a high Se area of China found no adverse effects of Se at intakes of 800 μg/day (16X requirement) . A small supplementation trial in men with prostate cancer gave subjects 1600 or 3200 μg Se/d as selenized yeast (32X and 64X requirement) for about 12 months and found no observable adverse effects in the group supplemented with 1600 μg Se/d . The subjects supplemented with 3200 μg Se/d, however, reported symptoms of Se toxicity, and a few men reached plasma Se levels exceeding 1000 ng/ml, which was reported to be the threshold for the onset of Se toxicity in the Chinese study . These reports in humans agree with the observations from the present study of no adverse effects of intakes up to 20-times the requirement and significant signs of Se toxicity at intakes ≥50-times the requirement. Additionally, the microarray data presented here indicates that super-nutritional Se intakes 8 to 20-times the Se requirement are not sufficient to cause a large transcriptional response. This is an important point because levels used in cancer prevention trials are typically 200 μg/day, or 4-times the requirement, which will bring total Se intake to 5 to 8-times the requirement depending on dietary Se intake from foods. This data thus suggest that cancer prevention associated with super-nutritional Se supplementation may not be mediated by transcriptional changes.
Se deficiency highly regulates the expression of some, but not all selenoprotein mRNAs [5, 8, 26, 37]. Most notable of the regulated selenoprotein genes is Gpx1, with its expression dropping to < 10% of Se-adequate levels in the rat model. Our recent studies have shown that in addition to Gpx1, Sepw1 and Selh are also highly regulated by Se deficiency and several other selenoprotein mRNAs are moderately regulated [8, 26]. Interestingly, there were no additional non-selenoprotein genes found to be down-regulated by Se deficiency, reinforcing the Se-specificity of selenoprotein regulation. Importantly, both microarray and qRT-PCR expression of the genes up-regulated in Se deficiency show that like selenoprotein mRNAs, these genes are restored to adequate levels in the mice fed a Se marginal diet (0.05 μg Se/g diet) (Additional file 1, Figure S1).
A small set of genes were found to be significantly and consistently up-regulated by Se deficiency. Comparison of these genes with a dataset of 21 well-characterized Nrf2-targeted genes containing Nrf2-binding sites  plus a dataset of 1055 Nrf2-targeted genes identified recently by ChiP analysis  revealed that the majority of the genes up-regulated by Se deficiency were Nrf2 targets (Table 1). NADPH dehydrogenase quinone 1 (Nqo1), a classic target of Nrf2 [40, 41], did not reach significance in this study, but was up-regulated 1.5 to 2 fold by Se-deficient treatments. Similarly, ABC transporters (also known as multidrug resistance proteins) and glutathione S-transferases, are known targets of Nrf2 regulation [42, 43], and were consistently up-regulated by Se deficiency. Two of the up-regulated ABC transporters, Abcc4 and Abcc3, are reported to be involved in detoxification of a variety of drugs including acetaminophen . In comparison, the only gene down-regulated by Se-deficiency and overlapping with these Nrf2 datasets was thioredoxin reductase-1 in mouse liver (Table 2). Two additional Nrf2 target genes, Rgs4 and Tsku, were also found to be up-regulated by 2 μg Se/g diet (Table 3). Overlap analysis of the 1193 transcripts that were significantly altered by 5 μg Se/g diet and the Nrf2-regulated gene datasets identified 99 Nrf2 targets that were differentially-regulated in Se toxicity (Additional file 3, Table S1), including 49 genes retained in the Se-specific dataset. The prevalence of Nrf2-regulated genes in genes significantly altered by Se toxicity as well as up-regulated by Se deficiency indicates that Se excess as well as Se deficiency increases oxidative stress.
Functional analysis of the genes in the Se-specific dataset indicates their involvement in processes such as cell movement/morphogenesis and development/angiogenesis (Table 4). ECM-related genes were particularly affected by Se toxicity. For example, collagen fibril organization was enriched by a factor of 6.72 (6 of 23 total genes present in the Se-specific set, FDR = 0.006). The genes in this category included four collagen genes (Col12a1, Col5a2, Col1a2, Col11a1), one serine proteinase inhibitor (Serpinh1) and one annexin (Anxa2). Further searching of the Se-specific dataset found a total of seven collagen genes. In addition, many other ECM-related genes were present in the 5 μg Se/g dataset. Collagen genes were also reported to be regulated by Se-methylselenocysteine in a prostate cancer cell line (LNCaP), but only one of these (Col4a5) was regulated in the same direction in the present study . Additional evidence for an association between Se and collagen metabolism comes from a study which found that Se supplementation of rats with 0.3 μg Se/g diet for 10 weeks increased collagen content of the skin, but decreased it in other tissues including liver . The regulation of ECM components may also be relevant to the anti-carcinogenic effect of Se as the ECM plays an important role in cell migration and tumor progression. In addition, collagen-regulation by Se may underlie some physiologic effects of Se toxicity, as the symptoms of Se toxicity include changes or malformations in tissues comprised primarily of collagen, nails and hair in humans and hooves in grazing animals [9, 46].
Genes related to glucose metabolism were enriched in the clusters that were up-regulated to some extent by 2 as well as 5 μg Se/g diet. There is some evidence that high Se status is related to diabetes and it is known that a selenosugar is one of the major excretory metabolites for Se [24, 25, 47]. Se-regulation of glucose-related genes may be another piece of the puzzle linking Se and glucose metabolism.
We also conducted GOMiner analysis of the genes that were removed from the original 1193-transcript toxic Se dataset due to overlap with genes altered by general toxicity and/or calorie restriction. This analysis identified 90 significantly enriched (False Discovery Rate = FDR < 0.05) biological processes that were not specifically associated with Se toxicity; these processes appeared to be more varied than those found in the Se-specific dataset (Additional file 4, Table S2).
Very few transcripts were significantly regulated by the sub-toxic intake of 2 μg Se/g diet, but there were several clusters of transcripts altered significantly by 5 μg Se/g which were also variably affected by 2 μg Se/g diet. Two of the conventional biomarkers, growth and ALT activity, suggest 5 but not 2 μg Se/g diet is toxic. AST activity, however, suggests that high Se intake at 2 μg Se/g as well as in Se deficiency is also just beginning to cause some liver damage. Previous studies have reported adverse effects, such as reduced body weight and liver damage, from Se intakes as low as 3 μg Se/g . Further studies are necessary to determine whether the transcripts within the clusters similarly regulated by 2 and 5 μg Se/g diet would gain significance and correlate with other markers of Se toxicity if studied at intakes between 2 and 5 μg Se/g diet. As 2 μg Se/g diet borders on a toxic Se intake, the 6 transcripts that were significantly regulated by this treatment may prove to be useful biomarkers of high Se intake before adverse effects are observed.
To illustrate the potential utility of a small set of high Se regulated transcripts as biomarkers of Se status, these transcripts and those regulated by Se deficiency were used successfully to predict liver Se concentration. Liver Se concentration was chosen as a marker of Se status for this example because it represents a pool of Se that continues to increase above the Se requirement, similar to hypothetical pools of Se associated with the anticarcinogenic activity of Se. The resulting 11-gene biomarker panel accounted for 99% of the variability in liver Se concentration across the range from Se-deficient to toxic Se, illustrating the potential of molecular biomarkers to predict Se status. Comparison of the molecular biomarker panel curve in Figure 7 with the conventional biomarker curves shown in Figure 2B-D clearly illustrates why these conventional biomarkers are ineffective for assessment of high Se status.
The high Se regulated transcripts identified in this study were associated with high Se intake in rats, but these studies did not show that these changes are specific for super-nutritional or toxic Se status. The use of transcript datasets associated with general toxicity or calorie restriction reduced the dataset to 667 unique transcripts which were more likely to be Se-specific, but, as shown by the variety of biological processes enriched in this dataset (Table 4), it is clear that most of these transcripts are likely to be altered by a variety of conditions, not just high Se. The molecular biomarker panel, however, illustrates that it may be possible to identify panels of these transcripts that collectively can be specific for a condition such as high Se status. Additional studies will be needed to further refine the sets of Se-specific transcripts that can be used as biomarkers and to uncover genes and processes associated with Se homeostasis and the effects of high Se supplementation.