The goal of this study was to compare and contrast the transcriptional response to TCDD in two closely-related rodent species, the rat and the mouse. Because these are the two most commonly used model organisms in pharmacology and toxicology, large divergences in their response to particular compounds is of broad interest. Previous work using cDNA microarrays containing 3,087 orthologous genes found that, while hundreds of genes responded to TCDD, only a small core of 10–20% was common to both species [27
]. We sought to evaluate and extend these findings.
We exploited oligonucleotide arrays that contain 8,125 ortholog pairs – 2.6-fold more than previously studied. We used a single dose, chosen to represent approximately equivalent multiples of the individual species' LD50 values. We assessed expression after 19 hours of TCDD exposure. The selected dose should be sufficient to eventually induce all major overt toxicities of TCDD and the selected time-point should be early enough to allow us to focus on primary responses rather than on the secondary manifestations of toxicity. We further broadened the mouse/rat comparison by using the highly TCDD-sensitive L-E strain, whereas the Boverhof study analyzed Sprague-Dawley rats.
We found that responses to TCDD are moderately correlated between the two species (Figure ), suggesting a broad trend towards similar responses. Further, a novel bi-species clustering approach shows in an unsupervised and unbiased way that the strongest signal within the dataset is the response to TCDD, not the difference between species (Figure ). We also show that hundreds of genes respond to TCDD in each species (Figure ), but only a small core of 10–20% was common to both – corroborating the findings of Boverhof and coworkers (Figure ).
This core set of species-independent TCDD-responsive genes contains many well-characterized genes, including Cyp1a1, Cyp1a2, Cyp1b1, Nqo1, and Tiparp – all of which have been previously detected in high-throughput studies [27
]. However, several species-independent genes have not previously been well-characterized for their response to TCDD but have functions that may be related to the clearance of xenobiotics. For example: indolethylamine N-methyltransferase is an enzyme that methylates indoles such as tryptamine, tagging them for degradation; cysteine conjugate-beta lyase metabolizes cysteine conjugates of alkanes and alkenes. Gene ontology analysis shows that the species-independent genes are enriched for microsomal genes, genes involved in the unfolded protein response, and genes involved in ER-nuclear signalling.
The 30 genes showing concordant responses to TCDD in rat and mouse represent prime candidates for explaining TCDD-induced hepatotoxicities. However, these toxicities might result in at least three other ways. First, toxicities may arise from events outside the liver and thus be independent of changes in hepatic mRNA levels. Indeed, it has been shown that hematopoietic cells contribute to the hepatic lesions caused by TCDD [41
]. Second, if toxicities arise from a cell-type present at a low frequency in the liver rather than from hepatocytes, relevant mRNA changes may be below the threshold of detection of our platform. For example, the Kupffer cells appear to play a central role in the hepatic response to many toxic and carcinogenic agents [42
] and the Ito (stellate) cells are responsible for retinoid storage and metabolism in liver [43
], a target for TCDD action [44
]. Third, toxicities may arise by perturbation of common pathway through different genes in each species.
Testing the first two hypotheses will require the use of tissue-specific transgenic animals and are thus beyond the scope of this study. To test the idea that toxicity results from the dysregulation of a pathway common to mouse and rat we used gene-ontology enrichment analysis. We searched for dysregulated functional groups in the lists of rat-specific and mouse-specific genes and found only one pathway dysregulated in both species – cellular lipid metabolism. Because this minimal overlap could occur by chance alone, and with caveats regarding tissue-specificity and transcriptome-coverage, it appears likely that the genes mediating TCDD-induced toxicity are amongst those reported in Table . While these species-independent genes appear skewed towards up-regulation, it is not clear that highly up-regulated genes are mediators of hepatic toxicities. For example, CYP1A1 and CYP1A2 are both highly induced in the liver of TCDD-resistant Han/Wistar (Kuopio
) rats [25
]. CYP1A1 can modulate some of the hepatotoxic effects of TCDD, as shown by lack of hepatocyte hypertrophy and uroporphyria in Cyp1a1-/-
], but induction of CYP1A enzymes is not, in itself, sufficient to cause hepatotoxicity.
It would be of interest to repeat our analysis using expression profiles collected 96 to 240 hours after TCDD exposure, when overt toxicities are manifest. At these later time-points it would no longer be possible to discern primary from secondary events, but instead there would be an increased ability to contrast the final toxic pathways perturbed in each species.
While about 15% of TCDD-responsive genes are species-independent and thus potentially mediate TCDD toxicities, the roles of the remaining 85% of genes are intriguing to consider. There are two key questions. First, does altered expression of these genes have functional consequences? Second, what mechanisms mediate species-specific responses?
If the species-independent genes had no functional consequence we might expect them to be randomly distributed across different functional classes. Instead, gene-ontology enrichment analysis (Table ) shows that they represent specific responses. In particular, the formation and function of ribonucleoprotein complexes involved in translation is significantly dysregulated in mice but not in rats. In our study of AHR-knockout mice we showed that this effect is AHR-dependent [28
] and suggested that it may mediate some TCDD-toxicities. Because this effect is absent in the rat, this pathway may contribute to degenerative changes induced by TCDD that are more prominent in mouse liver relative to rat liver.
The vast majority of TCDD toxicities are common to both species, but some species-specific differences do exist. For example, multinucleated giant hepatocytes occur only in rats, while hydropic degeneration, fat accumulation, inflammatory cell infiltration as well as apoptosis and necrosis are more pronounced in mice [27
]. We selected our TCDD dose to reflect approximately equal multiples of the LD50
values for each species to ensure that a similar ultimate phenotype was being considered. Nevertheless it is possible that the type of hepatic injury is species-specific and that this leads to the observed differences in transcript profiles. Several lines of evidence suggest against this possibility, including the evolutionary proximity of rat and mouse, the similar responses of known TCDD-responsive genes, and the inter-species similarity of the vast majority of morphological changes. Further work will be needed to clarify between these two situations.
Many different mechanisms can be proposed to explain the large species-specific responses observed, including copy-number variation, differences in transcription-factor binding affinities, and differential transcription-factor activities. We did not find evidence that copy-number variation plays a role (Figure ), so we focused on the potential role of transcription-factors in species-specific responses.
It has been established that transcription-factor binding is poorly correlated with changes in mRNA expression [49
]. Further, when the transcriptional regulatory networks were compared between human and mouse liver, they were found to differ dramatically [50
]. To determine whether specific transcription-factors are associated with the species-specific responses we performed two separate analyses of transcription-factor binding-site motifs. First we used a library-based enrichment analysis [51
] with a database of 130 known motifs that we supplemented with 5 AHR-associated motifs [37
]. Because the majority of sequence-specific DNA-binding proteins do not have known binding-site motifs, we complemented this analysis with an unsupervised analysis using the MotifSampler pattern-discovery algorithm [39
Our library-based analysis identified numerous motifs enriched in both species-dependent and species-independent manners, some clearly associated to the biology of dioxin-toxicity. For example, the enrichment of p53 motifs in mouse promoters suggests that a p53-dependent apoptotic response was initiated in mouse, but not in rat. Apoptosis is a common sequela to TCDD exposure in mouse liver [41
] and, while it also occurs in rats, it appears less frequent [52
]. One motif, Nr2f1 (COUP-TF1) showed divergent enrichment between the promoters of mouse and rat that responded similarly to TCDD. The AHR has been shown to directly bind COUP-TF1 in human cells [53
]. When we consider the overall patterns of TFBS-enrichment in an unbiased manner using unsupervised machine-learning, however, we found that species-specificity was the dominant trend. This suggests that the transcriptional regulatory networks are sufficiently diverged in mouse and rat as to make non-phylogenetic motif analysis highly challenging [50
MotifSampler is a variant of Gibbs sampling that introduces a higher-order Markov background model and incorporates a Bayesian mechanism to estimate the missing value: the number of motifs occurring in each sequence [39
]. MotifSampler was applied here to six different promoter datasets to successfully identify novel motifs. Biochemical analyses are required to demonstrate the functionality of these motifs, especially in the absence of an appropriate validation dataset, but the success of this analysis suggests that significant components of the AHR transcriptional network remain to be elucidated.
We directly compared our results to those of Boverhof and coworkers at the gene- level (Figure ). This is important because the two studies differed in animal handling, time-points, strain of rat used, gender and age of animals, array platforms employed, and statistical methodologies. Despite all these differences Spearman's correlations above 0.8 were observed between the two studies for both mouse and rat – an outstanding concordance. However, despite this correlation it is notable that the majority of genes identified as TCDD-responsive were found in only one of the two studies. This provides a cautionary note to the use of toxicogenomics for high-throughput testing, and stands in contrast to the optimistic findings of the MAQC study [54
Thus we have shown that most TCDD-responsive genes are species-specific and that the small core of species-independent genes shares both common functions and common transcriptional-regulatory elements. Taken together these data suggest that the vast majority of dioxin-responsive genes play no direct role in the toxic responses. Rather, it may be that a small number of regulatory genes under direct AHR control lead to dramatic inter-species differences. It would be highly advantageous to further refine the list of genes associated with dioxin-toxicities by profiling the transcriptional response of additional strains of mouse and rat, or by exploiting additional rodent species.
This study has significant implications for the use of rodent models to understand mechanisms of pharmacological or toxicological compounds. If transcriptional responses have been extensively remodelled in rat and mouse since their last common ancestor then hepatic expression studies may yield contradictory and species-specific results. It may be necessary to test both species and focus on the subset of overlapping genes. An optimal study design might include several strains of mice and several strains of rat, thereby focusing the list of transcriptional responses to a small number that exhibit both intra- and inter-species homogeneity and are thus highly likely to be directly involved in the phenotypic effects of the compound under investigation.