Genome-wide analysis of thyroid hormone receptors shared and specific functions in neural cells

Abstract

TRα1 and TRβ1, the two main thyroid hormone receptors in mammals, are transcription factors that share similar properties. However, their respective functions are very different. This functional divergence might be explained in two ways: it can reflect different expression patterns or result from different intrinsic properties of the receptors. We tested this second hypothesis by comparing the repertoires of 3,3',5-triiodo-L-thyronine (T3)-responsive genes of two neural cell lines, expressing either TRα1 or TRβ1. Using transcriptome analysis, we found that a substantial fraction of the T3 target genes display a marked preference for one of the two receptors. So when placed alone in identical situations, the two receptors have different repertoires of target genes. Chromatin occupancy analysis, performed at a genome-wide scale, revealed that TRα1 and TRβ1 cistromes were also different. However, receptor-selective regulation of T3 target genes did not result from receptor-selective chromatin occupancy of their promoter regions. We conclude that modification of TRα1 and TRβ1 intrinsic properties contributes in a large part to the divergent evolution of the receptors' function, at least during neurodevelopment.

Fig. 1.

Isogenic neural cell lines expressing either TRα1 or TRβ1. (A) Primary sequence of the mouse TRα1 receptor. Italics correspond to the unique N terminus. Red amino acids are different in TRα1 and TRβ1 in both human and mouse receptors. The blue box corresponds to the DNA-binding domain. Within this domain the P box (yellow) and the D box (green) are key elements of the two zinc fingers formed by the DNA-binding domain. Note the divergence in the D box, which has been shown to be important in the in vitro DNA-binding properties of the TR/RXR heterodimers. (B) Expression vector used in C17.2 cells. TRα1 or TRβ1 cDNA is inserted in frame with the GS tag sequence encoding a fragment of protein-G (G) peptide and a streptavidine-binding peptide (S). Transcription is under the control of the CMV promoter. A downstream cassette coding for the enhanced green fluorescent protein (EGFP), translated via an independent internal ribosome entry site (IRES), was used to select TR-expressing cells by fluorescence-assisted cell sorting. (C) C17.2α and C17.2β cells express equal levels of either GS-TRα1 or GS-TRβ1 after two rounds of cell sorting, as judged by Western blotting (based on IgG/protein-G interaction). A faint nonspecific band is observed in both control and transfected cells. (D) Q-RT-PCR confirms that T3 transactivation of Hr expression is restored in both C17.2α and C17.2β cells. Time course of response (10⁻⁷ M T3, Upper) and dose dependence (Lower) are shown. Error bars indicate SD for three independent experiments. Maximum response is achieved at T3 doses superior to 10⁻⁸ M and slowly increases beyond 6 h. Naive C17.2 cells do not display a significant response.

Fig. 2.

Genome-wide analysis of TRα1 and TRβ1 and T3-mediated response. Only the 1,125 genes for which T3 has a significant influence at two consecutive time points are represented. The plots report the log₂ of the induction rate for C17.2α (x axis) and C17.2β cells (y axis) at three different time points. Gray circle corresponds to genes without significant regulation (R < 1; Materials and Methods). Vertical lines and horizontal lines delimit areas with apparent TRβ1-selective and TRα1-selective regulation, respectively.

Fig. 3.

Q-RT-PCR confirmation of receptor-specific regulation. Bars with light shading, C17.2α cells; bars with dark shading, C17.2/β cells. y axis: log₂ of fold change after T3 stimulation. Error bars indicate SD for three independent experiments. (A) Time-course analysis 6,12, and 24 h after T3 addition reveals receptor selective response. (B) Receptor selective response persists within a wide range of T3 concentrations (10⁻¹⁰ M, 10⁻⁹ M, 10⁻⁸ M, 5.10⁻⁸ M, and 10⁻⁷ M for 24 h). TRα1 selective response is observed for Slc26a1, B3galt5, and EphB3. TRβ1 selective response is observed for Adamtsl4, Aoc3, Htra1, Megf6, and Tgm2.

Fig. 4.

Clustering analysis of T3 responsive genes. Heat map (Center) and average behavior of the 10 different clusters of genes obtained after k-means clustering analysis are shown. Blue, low levels of expression; yellow, high levels of expression (normalized and mean-centered values). C17.2α cells are on the left and C17.2β cells on the right. For each cell line, untreated cells (−T3) are on the left and treated cells (+T3) on the right, each category being subdivided into three different columns corresponding to the duration of treatment (6, 12, or 24 h). Each cluster is numbered (K1–K10). Graphs represent the mean value of gene expression ±SD for each cluster at different time points. C17.2α are on the left and C17.2β on the right. Blue line, untreated cells; red line, T3-treated cells.

Fig. 5.

TRα1 and TRβ1 cistromes. (A) representative view of the UCSC mouse genome browser (http://genome.ucsc.edu/cgi-bin/hgGateway) around the Klf9 locus. (Bottom) Klf9 position and exon/intron composition are indicated by boxes (exons) and bars (introns). Mapability for 36-mer tags is indicated as a red line ranging from 0 (not mapable) to 1 (fully mapable) for every position. Each ChAP-Seq experiment is represented as a signal track, providing the number of tags sequenced on 10-bp sliding windows (scale on the right-hand side is the number of counted tags). Two shared TRBS and one binding site detected only for TRα1 are located upstream of Klf9. Note that for Klf9, the TRBS are not at the previously reported position (57). (B) Venn diagrams representing the number of binding sites for each ChAP/ChIP-Seq experiment and the number of shared TRBS. The threshold P value is 10⁻⁷ for the sites identified in a single experiment. Additional sites (in parentheses and italics) are the ones observed in more than one experiment, using a 10⁻⁴ threshold for peak calling, taking into account the data of the RXR ChIP-Seq experiment (Fig. S3). (C) Limited overlap between TRα1 and TRβ1 cistromes. Depending on the P value used for peak calling, the fraction of TRβ1-binding sites overlapping with TRα1-binding site varies. However, unlike what is observed between TRα1 and RXR (Fig. S3), overlap never exceeds 0.8 even for P value <10⁻⁷. The false discovery rate was calculated by assuming that all TRα1-binding sites should be also identified in the RXR ChIP-Seq experiment performed on C17.2α cells. (D) Consensus sequence defined by the CHIP-MEME algorithm using all of the TRBS is close to a direct repeat with a 4-nt spacer (DR4). According to structure analysis, RXR recognizes the 5′ half-site (5′AGGTCA) and TR recognizes the 3′ half-site (5′AGGNCA).

Fig. 6.

Correspondence between chromatin binding by TR and T3 regulation. The 1,125 genes for which T3 has a significant influence at two consecutive time points are represented as in Fig. 2 (gray). (A) Shared binding sites. Genes with a TRBS within 30 kb of the transcription start site are in red. Note the high enrichment (×10) for genes that are positively regulated. Triangles correspond to TRBS with a sequence similar to the consensus DR4 element, as defined in Fig. 5. The distribution of DR4 containing TRBS is not different from the others. (B) Receptor-specific TRBS. Triangles correspond to TRBS with a sequence similar to the consensus DR4 element. Genes with a TRα1-specific binding site are in green, and genes with a TRβ1-specific binding site are in orange. Note that, for technical reasons, we can identify a larger number of TRBS in C17.2α cells than in C17.2β cells. The distribution does not indicate a correlation between the presence of a proximal receptor-specific TRBS and receptor-selective regulation.