Gene Specific Actions of Thyroid Hormone Receptor Subtypes

There are two homologous thyroid hormone (TH) receptors (TRs α and β), which are members of the nuclear hormone receptor (NR) family. While TRs regulate different processes in vivo and other highly related NRs regulate distinct gene sets, initial studies of TR action revealed near complete overlaps in their actions at the level of individual genes. Here, we assessed the extent that TRα and TRβ differ in target gene regulation by comparing effects of equal levels of stably expressed exogenous TRs +/− T3 in two cell backgrounds (HepG2 and HeLa). We find that hundreds of genes respond to T3 or to unliganded TRs in both cell types, but were not able to detect verifiable examples of completely TR subtype-specific gene regulation. TR actions are, however, far from identical and we detect TR subtype-specific effects on global T3 response kinetics in HepG2 cells and many examples of TR subtype specificity at the level of individual genes, including effects on magnitude of response to TR +/− T3, TR regulation patterns and T3 dose response. Cycloheximide (CHX) treatment confirms that at least some differential effects involve verifiable direct TR target genes. TR subtype/gene-specific effects emerge in the context of widespread variation in target gene response and we suggest that gene-selective effects on mechanism of TR action highlight differences in TR subtype function that emerge in the environment of specific genes. We propose that differential TR actions could influence physiologic and pharmacologic responses to THs and selective TR modulators (STRMs).


Introduction
Thyroid hormone (TH) receptors (TRs a and b) are highly homologous transcription factors which transduce signals of active forms of TH (predominantly tri-iodothyronine, T 3 ) [1,2]. Like other nuclear receptors (NRs), TRs bind specific DNA response elements (TREs) comprised of degenerate repeats of the sequence AGGTCA, usually as heterodimers with retinoid X receptors (RXRs). From these locations, the TRs recruit coregulator complexes that influence gene expression and T 3 modulates transcription by inducing conformational changes in the receptor C-terminal ligand binding domain which, in turn, alters the complement of TR associated coregulators [3][4][5]. Despite similarities in structure and function, analysis of TR gene knockout mice and human patients with TRa or TRb mutations has revealed that the two TRs display unique activities in vivo [6][7][8][9][10][11]. TRa plays major roles in regulation of heart rate and muscle whereas TRb modulates serum cholesterol levels via actions in liver and feedback inhibition of TH production through the hypothalamic-pituitary-thyroid axis. TRs also exhibit subtype-specific effects in regulation of basal metabolic rate, bone development and other processes.
Differential effects of TRs have commonly been attributed to variations in TR expression levels in target tissues. Liver, for example, expresses TRb and TRa in a 5:2 ratio and this could explain the predominant role of TRb in cholesterol metabolism [12,13]. Indeed, comparison of T 3 effects in hypothyroid wild type, TRa, TRb and double TR knockout mouse livers failed to identify truly TR subtype selective genes [14]. It is noteworthy, however, that other closely related NRs, including estrogen receptors (ERs) a and b, regulate quite different gene sets [15]. Given that current approaches to selectively modulate TH signaling pathways have often focused on development of TR subtype selective modulators (STRMs) [12], it is important to assess the extent of TR subtype-specific effects on gene regulation.
Several pieces of evidence suggest that TRa and TRb actions may not be absolutely identical at the level of individual target genes. First, stably transfected TRa and TRb regulate the same set of around 40 target genes in HepG2 liver cells, but with significant subtype specificity in magnitude of T 3 response at individual genes [16]. Second, TRa and TRb exhibit similar effects at transfected reporters with the most common form of TRE (direct repeat spaced by 4 nucleotides, DR-4), but TRb is more active at inverted palindromic (IP) TREs [17]. Third, detailed qPCR analysis of expression of liver genes in TR knockout mice has revealed apparently exclusively TRb-dependent T 3 regulated genes, including thrsp and angptl3 [18]. Fourth, TR subtypes exhibit differential effects on ABCD2 gene via indirect differential effects on expression of the transcription factor SREBP1c [19]. Fifth, a constitutively unliganded TRa mutant represses transcription of the liver C/EBPb gene more efficiently than an equivalent TRb mutant, via differential corepressor recruitment [20]. Finally, TRb selectively represses the thyroid stimulating hormone (TSH) gene in cultured pituitary cells even though both TRs are present and TRa is functional when expression of TRb is knocked down [21].
In addition to TR subtype specificity, there is evidence that activity of both TRs is highly gene-context dependent [1,14]. T 3 can either activate or repress transcription of target genes but, within this framework, there are differences in contributions of unliganded and liganded TRs. Commonly, unliganded TRs suppress T 3 -inducible genes and agonists induce transcription by both reversing the inhibitory effects of unliganded TRs and eliciting further transcriptional activation. There are, however, cases in which unliganded TRs fail to suppress T 3 induced genes or where T 3 activates genes solely by relieving unliganded TRdependent basal repression [14,16]. Similar gene-specific variations in the balance of unliganded and liganded TR actions also occur at negatively regulated genes and TRs can even exhibit completely ligand-independent actions [14,16]. T 3 concentration dependence also varies; euthyroid T 3 levels are sufficient for optimal induction of some genes whereas others need higher (hyperthyroid) T 3 levels [14,22].
Here, we set out to define the extent of TR subtype and genespecific variations in transcriptional response by creating cells with similar levels of TRa or TRb. While our data suggests that there are no completely TR subtype specific genes, we observed TRsubtype and cell-specific effects on the kinetics and magnitude of transcriptional response, patterns of TR-dependent gene expression and T 3 concentration dependence that verify and extend conclusions of previous groups. We discuss possible mechanisms of these differential effects and their impacts upon physiological responses to THs and actions of STRMs.

Reagents
Triiodothyronine (T 3 ) suitable for cell culture was purchased from Sigma Aldrich (T6397).

Construction of HepG2-TR Cells
HepG2 cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 100 U/mL of penicillin, 0.1 g/L of streptomycin and 4 mmol/L glutamine, under 95% air and 5% CO 2 at 37uC. Lentiviral vector pSicoR containing GFP was a gift from the McManus laboratory (UCSF, http://mcmanuslab.ucsf.edu/). Sequences for appropriate tags were added to TR cDNAs derived from previously described eukaryotic expression vectors by standard PCR amplification and resulting hybrid cDNAs were cloned into pSicoR vector at a location C-terminal of GFP. An oligonucleotide encoding a T2A peptide that will mediate cotranslational cleavage of protein was inserted between GFP and TRs coding sequences to facilitate independent expression of GFP and TRs. pSicoR vectors containing tagged TRs were transfected into HEK 293 cells to create packaged virus particles by the Diabetes center core lab at UCSF. The virus was titrated by GFP fluorescence. HepG2 cells grown on a 6-well plate were transduced with lentivirus particles containing tagged TR sequences. After incubation with virus particles for 72 hours, cell culture media were replaced with regular growth media and GFP expression was verified by fluorescent microscopy. Cells were split upon confluence and sorted by fluorescence-activated cell sorting in UCSF core facility after two passages. Clones of GFP positive cells were collected and grown in 10 cm plates. When cells reached confluence, TR expression was analyzed with Western blots using anti-tag antibodies and transient transfection of TRE driven luciferase reporters, described below. Cells were maintained in regular growth medium as above.

Construction of HeLa-TR cells
HeLa cells were grown in similar conditions to HepG2 cells. HeLa cells stably expressing tagged TRa and TRb were generated using the Tet-Off gene expression system (Clontech). Stable clones were selected by hygromycin-resistance (400 mg/ml) and screened for TR protein expression. HeLa cells expressing TRb and TRa were maintained in media supplemented with doxycycline (20 ng/ ml).

Western Analysis
HepG2 or HeLa cells were lysed using Triton X-100 lysis buffer. Cell extracts (10 mg of total protein) were separated by SDS-PAGE and transfer onto a PVDF membrane. Transfer membranes were then incubated with anti-flag M2 antibody (Sigma-Aldrich) or c-Myc antibody (Clontech) at a 1:1000 dilution at 4uC for 16 h followed by goat anti-mouse IgG-horseradish peroxidase antibody (Santa Cruz Biotechnology sc-2004) at a 1:10000 dilution for 45 min at room temperature. Blots were visualized by applying ECL Plus (GE Healthcare).

Hormone Binding
Hormone binding assays were carried out as described in [23]. The K d values were calculated using the Graph-Pad Prism computer program (Graph-Pad Software Inc).

Transfection
Cells were co-transfected with a DR-4 or IP-6 TRE-driven luciferase reporter and constitutive renilla luciferase reporter (Promega) using Transfectin Reagent (BioRad) and plated in 12well plates in growth medium (DMEM with 10% hormonedepleted FBS) [23]. After 16 h of incubation, T 3 (100 nM) or vehicle (DMSO) was added in triplicate. After an additional 24 h of incubation, cells were harvested and assayed for luciferase activity using the Promega Dual Luciferase Reporter Assay (Promega). Data were normalized to the renilla luciferase activity.

mRNA and cDNA Preparation
For HepG2, total RNA was prepared using the Aurum Total RNA kit (Bio-Rad). Reverse transcription reactions in these samples were performed using 1 mg of total RNA with an iScript cDNA Synthesis kit (Bio-Rad). Total RNA concentrations were measured using NanoDrop ND-1000 spectrophometer. For HeLa, total RNA was extracted from cells with Qiazol Lysis Reagent (Invitrogen) and purified with RNeasyH Mini kit (Qiagen) following manufacturer's instructions. mRNA was reverse transcribed into cDNA with a mixture Oligo(dT) 20 and Random Hexamers (1:1 ratio) using SuperScript TM III First-Strand Synthesis System for RT-PCR kit (Invitrogen).

Microarray hybridization
Human whole genome expression arrays were purchased from Illumina (Human WG-6v2 and Human WG-6v3). cRNA synthesis and labeling were performed using IlluminaH TotalPrep TM -96 RNA Amplification Kit (Ambion). Labeling in vitro transcription reaction was performed at 37uC for 14 h. Biotinylated cRNA samples were hybridized to arrays at 58uC for 18 h according to manufacturer's protocol. Arrays were scanned using BeadArray Reader.

Statistical analysis
Unmodified microarray data obtained from GenomeStudio was background-subtracted and quantile-normalized using the lumi package [24] and analyzed with the limma package [25] within R [26]. To determine the effect of TRa and TRb over-expression in the absence of ligand (''unliganded-effect''), cell lines were analyzed separately by LIMMA (''parental'', with no exogenous TRs, TRa, and TRb), followed by contrast analysis. To better determine TR isoform effects, factorial LIMMA analysis was conducted comparing ligand (T 3 ) with over-expression of the TRa or TRb (''TR over-expression with ligand effect''; interaction between T 3 and over-expression of TRa or TRb), followed by contrast analysis. All analysis was corrected for multiple hypothesis testing [27], and effects determined to be significant when $2-fold with an adjusted p-value # 0.05 when compared to their respective parental cell line. To facilitate comparisons among the various datasets, all data was uploaded into a SQLite3 database [28]. Heatmaps were produced and clustered using multiarray viewer [29]. qPCR Real-time qPCR in HepG2 samples was performed with the Roche LightCycler 480 RT PCR Instrument using SYBR Green Mastermix (Roche). The sequences of the primers are listed in Table S1. The data were collected and analyzed using the comparative threshold cycle method. Experiments were performed at least three times, and the mean 6 SE was calculated using the Prism curve-fitting program (GraphPad Software, version 3.03; GraphPad). For HeLa, qPCR was performed using SYBR Green PCR Master Mix (Applied Biosystems) on ABI 7900HT RT-PCR system (Applied Biosystems) with default two-step QRT-PCR program. Amplification curves were evaluated by the comparative C t analyses.

Calculation for categorizations of expression patterns
Custom Python scripts were used to organize the expression patterns into induced and repressed effects, through the calculation of minimal Euclidean distances between nine hypothetical patterns (2-by-3; unliganded and liganded effect expressed as induced (e.g., 2-fold), repressed (e.g., 0.5-fold) or no effect (e.g., 1)) and that experimentally derived. The hypothetical patterns (unliganded, ligand) were: pattern RR = 0.5, 0.5; pattern RO = 0.5, 1; pattern RI = 0.5, 2; pattern OR = 1, 0.5; pattern OO = 1, 1; pattern OI = 1, 2; pattern IR = 2, 0.5; pattern IO; 2, 1; pattern II = 2, 2. Note, hypothetical pattern OO translates to no substantial effect with TR over-expression (unliganded effect) or T 3 -treatment (ligand effect) and was not included in the final table. All probes with a BH adjusted p-value, = 0.05 within the 3 hr HepG2 and 24 hr Hela treatments were analyzed. To lessen the effect of extremes, fold-change values were first transformed to 2 or 0.5 if their fold-change values were .2 or ,0.5, respectively. Euclidean distances were then calculated, and the probe/gene transcript grouped into the hypothetical pattern that delivered the minimal Euclidean distance between the experimental and/or the transformed vector and the hypothetical pattern vector. The genes were subsequently translated from Probe_ID to official gene name using an SQLite database.

Cells that Express Comparable Levels of TRa1 or TRb1
We created two sets of TR-expressing cells to compare actions of major TR subtypes at endogenous genes. For HepG2 liver cells, we used a retroviral infection to express epitope-tagged (Flag) TRa1 or TRb1. For HeLa, we used stable transfection to express epitope-tagged (myc) TRs under tetracycline control (tet-off system, Fig. S1). We screened multiple clones of both cell types by western, using antibodies against respective epitope tags to facilitate direct comparisons of protein levels, and identified pairs of cell lines with comparable TRa and TRb expression (Fig. 1A). We confirmed that exogenous TRs were recognized by antibodies against TR primary sequences, that these TRs were expressed at higher levels than endogenous TRs, which were either present at very low levels (TRb in HepG2 [22]) or undetectable (TRa in HepG2 and both TRs in HeLa) and that expressed TRs were of correct molecular weight ( Fig. S2 and not shown). We also verified that TR expression was stable over several passages (not shown).
Exogenously expressed TRs are functional. We verified that the presence of the tag did not affect TR function in transient transfections, in which we compared native and tagged TRs ( Fig.  S3 and not shown). We were also unable to detect any major differences between the function of exogenously expressed TRs versus previous assessments of endogenous TR function. Hormone binding analysis confirmed that TRs exhibit affinities for T 3 that are consistent with previously reported values ( Fig. 1B and not shown). This data also allowed us to estimate numbers of T 3 binding sites per cell (i.e. TRs), which were between 7-10,000 in transfected cells versus essentially undetectable in parental cells [22] and within physiological range (,10,000 receptors per cell) [30]. Exogenous TRs conferred T 3 responses on standard TREdependent reporters in both cell types ( Fig. 1C and not shown). Further, T 3 elicited similar levels of activation with TRa and TRb at a DR-4 reporter but larger levels of activation with TRb at an IP-6 reporter, in accordance with previous results [17]. Finally, TR steady state levels were diminished after T 3 treatment (Fig. 1D); this phenomenon is common to many NRs and a consequence of ubiquitin-dependent turnover of activated receptors [31]. Interestingly, however, TRa levels were rapidly reduced (within 1 hour of T 3 addition) whereas TRb levels only obviously became diminished after extended ligand treatments.

TRa and TRb Regulate Similar Gene Sets in HepG2 with Different Kinetics
We performed transcriptome wide analysis of TR target genes in our HepG2-TR cells and parental HepG2 controls after 3, 6 and 24 hr induction with saturating (100 nM) T 3 . Since HepG2 expresses low levels of endogenous TRb1, we examined interaction between treatment and cell line (i.e. T 3 + TR over-expression) to determine TR-specific effects.
Most T 3 responses required exogenous TRs ( Fig. 2A-D, Fig. 3). As we previously reported [22], a few genes responded to T 3 in parental HepG2 cells, with around 17 meeting cut-offs (.2.0 fold, BH-adjusted P value 0.05) at 24 hrs. This is due to vanishingly low levels of functional TRb present in HepG2 [22]. By contrast, hundreds of genes responded to T 3 in cells that express either of the two TRs ( Fig. 2A-D). Of these, the majority (more than 70%) were induced by T 3 with the remainder repressed. Additionally, most genes that exhibited T 3 responses in parental HepG2 cells exhibited amplified responses in the presence of exogenous TRs. The sole exception was that we discovered expression of the highly T 3 responsive ANGPTL4 gene, a verified direct TRb target in parental HepG2 cells [22], was silenced by exogenous TRs when we performed qPCR analysis (Fig. S4). We confirmed that enhanced T 3 responses seen in the presence of transfected TRs in HepG2 were dependent upon exogenous TR expression using an siRNA specific to the 59 portion of the dual EGFP/TR transcript to inhibit exogenous TR expression (Fig. S5).
Unexpectedly, different numbers of genes met cutoffs for fold induction and statistical significance with TRa and TRb at each of the three times ( Fig. 2A-C). More T 3 responsive genes appeared with TRa at 3 and 6 hrs ( Fig. 2A, B), whereas TRb responses predominated at 24 hrs (Fig. 2C). Overall, similar numbers of genes exhibited T 3 -responses when all three times were considered together (Fig. 2D). Closer analysis revealed no completely TR subtype-specific genes within the datasets, there was a high degree of overlap between TRa and TRb responsive genes and nearly all genes that responded to T 3 with either TRa or TRb at any of the three time points exhibited qualitatively similar responses with the other TR in at least one time point (Fig. 3A and not shown).
To better understand differential kinetics of T 3 response in HepG2-TRa1 cells and HepG2-TRb1 cells, we compared fold T 3 induction/repression of each gene (Probe_ID) in the presence of the two TRs (Fig. 2E-H). Although there were more TRa selective genes at 3 and 6 hrs, there was a strong apparent correlation between fold induction/repression when the two TRs are compared (Fig. 2E, F). Visual inspection (Fig. 3A, not shown) and statistical correlation analysis suggested that many of the apparently TRa-selective genes responded in a similar fashion to TRb, but that T 3 response sometimes failed to meet cutoffs for fold induction and/or statistical significance resulting in the discrepancies between numbers of regulated genes. While there was also apparent correlation between overall TRa and TRb responses at 24 hrs (Figs. 2G, H; Fig. 3A, B), we observed a shift in slope that  Fig. 3B). We used qRT-PCR analysis to confirm that members of this strongly TRb-dependent late responding gene set (G6Pc, GSTA1) retained preferential TRb responses that persisted over multiple T 3 incubation times (Fig. 4).
Together, our data suggests that TRs regulate similar gene sets but with different kinetics in HepG2; T 3 responses emerge earlier with TRa versus TRb. Further, a strongly TRb-dependent subset appears after prolonged T 3 treatment.

T 3 -induced Genes are Direct TR Targets
We determined the proportion of T 3 responsive genes that were direct TR targets (Fig. 5). To do this, we examined effects of pretreatment with the protein synthesis inhibitor cycloheximide (CHX) upon T 3 response at 3 h and defined direct targets by persistence (at least 100%) of gene expression obtained with T 3 after CHX treatment relative to levels obtained with T 3 alone. Most (.80%) of the genes that are positively regulated by T 3 are scored as direct targets by these criteria (Fig. 5A upper panel). Investigation of remaining positively regulated genes revealed that some T 3 activation persisted in the presence of CHX for many of the remaining 20% of genes, suggesting that representation of direct TR targets within this dataset may be even larger than this analysis suggests (Fig. 5B, not shown). Interestingly, a much smaller proportion of negative T 3 responses persisted after CHX treatment versus positive responses (Fig. 5A, lower panel). This suggests that novel protein synthesis is needed for T 3 repression in this cellular context. In total, CHX treatment only completely abolished T 3 response of a small subset of genes (Fig. S6); responses of <11% of all genes that displayed 2-or greater fold responses to T 3 at 3 hrs were completely inhibited by CHX. These are likely to represent secondary responses to T 3 -dependent changes in protein levels.
As seen with the complete dataset, we observed little TR subtype selectivity among direct TR targets (Fig. 5B). Some T 3 responsive genes display stronger TRa responses and qualitatively similar but weaker responses with TRb; this is evident from comparison of columns 1 and 4 in the heat map (some examples of this set of genes marked ''a''). However, T 3 responses mostly persisted with CHX in the presence of both TRs and we even detected cases of amplification of weak T 3 response with TRb in the presence of CHX (examples of this set of genes marked ''b''). There were also gene-specific interactions of TRs and CHX; some genes were selectively de-repressed by CHX treatment alone in the presence of one of the two TRs (TRa selective de-repression is marked ''c'' and TRb selective de-repression marked ''d''). In general, however, most direct target genes appear similarly regulated by both TRs.

TR/Gene-Selectivity in T 3 -response
Within broad TR response patterns outlined above, there was gene-specific variability of T 3 regulation patterns and we verified some of these observations using RT-qPCR (Fig. 6). Many genes displayed similar time courses of T 3 induction with both TRs (PCK1, Fig. 6A), but others exhibited differential responses to the at individual time points (SLC16A6, Fig. 6B) and yet others displayed sustained preferential responses to TRb (HIF2A; Fig. 6C) or to TRa (MYH6 Fig. 6D). Thus, differences in magnitude and kinetics of T 3 response with the two TRs are reflected at the level

Similar TRa and TRb Responses in HeLa cells
To extend comparative analysis of TR subtype effects, we examined T 3 responses in HeLa cells that express exogenous TRs.
Here, TR expression was induced by doxycyclin withdrawal for 24 hrs; this regimen elicited optimal TR mRNA induction (Fig.  S7). We then treated cells +/2 saturating T 3 (100 nM) for a further 24 hrs.
More genes responded to T 3 in the presence of TRs in HeLa cells than HepG2 cells (Fig. 7A). All responses were dependent on exogenous TRs; unlike HepG2 cells, our HeLa cells lack detectable TR protein and transcripts (not shown). Like HepG2 cells, however, the majority of genes were induced by T 3 and there was near complete overlap between TRa and TRb target genes; plots of T 3 responses with TRb versus TRa again revealed apparent correlation between induction/repression for most genes (Fig. 7B). This extends our conclusion that TRa and TRb responses are broadly similar and also implies that late TRbspecific effects observed in HepG2 are a feature of the latter cell type.
Also as seen in HepG2, we detected genes which exhibit preferential responses to TRa or TRb and verified some effects with qRT-PCR. PCK1 was strongly induced by T 3 with TRb but not TRa, although both TR subtypes enhanced transcript abundance without ligand (Fig. 7C). PCK1 was strongly T 3dependent in HepG2-TRa and HepG2-TRb cells (see Fig. 4B) implying that this is a cell-specific effect. More commonly, and similar to HepG2, we observed TR subtype specificity in magnitude of T 3 response; for example, the THRSP gene displayed stronger T 3 induction with TRb versus TRa (Fig. 7D).  Finally, there was only limited overlap between T 3 -regulated gene sets in HepG2 and HeLa (Fig. 8). These differences were not explained by failure to meet cut-offs for fold induction and TRs did regulate different gene sets in the two cell types. This means that strong overlap between TRa and TRb target genes occurs with two largely distinct gene sets of T 3 regulated genes in two cell types.

Unliganded TRa and TRb Regulate Similar Gene Sets
Since TRs are transcriptionally active without hormone [3,4], we compared effects of unliganded TRa and TRb in both cell types (Fig. 9A-D). To do this, we assessed differences in gene expression in HepG2-TR cells versus HepG2 parental cells and parental HeLa cells versus HeLa-TR cells after 24 hrs doxycyclin withdrawal.
Unliganded TRs influenced many genes in both experimental systems (Fig. 9A, B). Expression of more than two thousand genes was altered by the presence of unliganded TRs relative to parental controls in HepG2 cells; with similar numbers up-and downregulated. Large numbers of genes (<1000) also responded to short term TR induction in HeLa cells and, again, similar numbers of genes were up and down-regulated. More TRdependent genes met fold cutoff and statistical significance with unliganded TRa versus TRb in both cell types. As seen with T 3 regulation, however, these differences were generally qualitative and not absolute and we observed an essentially linear relationship between induction/repression with unliganded TRa and TRb in both cell types even though some probe sets suggested preferential response to one of the two TR subtypes (Fig. 9C, D).

Gene-Specific Variations in Pattern of Response to TRs +/2 T 3
Next, we examined specific patterns of target gene regulation by TRs [14]. To do this, we grouped genes with statistically significant responses to unliganded TR or T 3 into categories according to whether they are repressed (R), unaffected (O) or induced (I) relative to basal gene expression levels in parental cells (Methods and Fig. 10/ Table 1/Table S2). In this way, TR and T 3dependent genes could be placed into one of eight response patterns shown in the heat map at left. Another category, ''OO'', included genes that displayed small changes in response to TR or T 3 that reached statistical significance, but was assigned to the non-responsive pattern and are not shown (see Methods).
We detected examples of all classes of predicted responses to TR +/2 T 3 (Fig. 10/ Table 1/Table S2). As seen in previous study [16], a small percentage of genes were constitutively repressed or induced by TRs (patterns RR and II), and we verified some observations with qRT-PCR (Fig. 11, MST1, hel308 and also see Fig. S4). Other genes were activated or repressed by T 3 , with a majority displaying one of several possible patterns of positive T 3 response (R0, RI, 0I) and a large minority exhibiting one of several patterns of negative regulation in response to T 3 (0R,0I, IR). Distributions of genes between different categories of positive and negative response varied with cell type. In HepG2, large majorities of positively and negatively regulated genes grouped into pattern R0 and pattern I0, respectively. Interestingly, these response patterns were mirror images of each other, with unliganded TR repressing positively regulated genes or activating negatively regulated genes and T 3 reversing these effects. By contrast, the response patterns were more evenly distributed in HeLa.
Although TRs regulated similar numbers of genes in the same way, there was only limited overlap between genes that grouped into the same pattern with TRa and TRb (Fig. 10). This suggests that relatively subtle differences in magnitude of response to TRs +/2 T 3 can translate into different response patterns. We confirmed this impression at a limited set of target genes with qRT-PCR (Fig. 12A-D). For example, Myh6, Furin, ALPI and HIF2A are all induced by T 3 but Myh6 and furin exhibit the same basic regulation pattern (Fig. 12A, B, pattern 0I), but ALPI is induced by unliganded TRb (pattern II) and not TRa (pattern 0I) whereas HIF2A is induced by unliganded TRa (Fig. 12C, pattern II)) and not TRb (Fig. 12D, pattern 0I).

TR Subtype and Gene Selectivity in T 3 Dose Response
Finally, we performed a limited survey of relationships between T 3 dose response, TR subtype and gene ( Table 2 and Fig. S8). To do this, we examined effects of varying doses of T 3 upon selected TR targets in HepG2 cells. Most genes exhibited EC 50 values in the low nM range (<1 nM), but TRa exhibited some gene-specific requirements for lower T 3 concentrations (klf9, pck1) whereas TRb exhibited gene-specific requirements for higher T 3 levels (furin, alpi, myh6, hr). The combination of these effects means that many genes exhibit similar T 3 concentration-dependence in the presence of both TRs whereas others exhibited differential responses to T 3 with TRa and TRb, with EC 50 values varying up to 30-fold when different genes and response to TR subtype are considered. It was noteworthy that higher levels of T 3 were needed for optimal induction with TRb in most cases of differential concentration dependence.

Discussion
In this study, we compared effects of equal levels of exogenous TRs upon endogenous genes, +/2 T 3 , in different cell backgrounds (HepG2 and HeLa) to determine the prevalence of TR subtype specific genes. While hundreds of genes respond to T 3 or to unliganded TRs in both cell types, and there are differences in degree of response +/2 T 3 , we were unable to identify truly TR subtype-specific genes in either cell background; i.e. genes which respond only to one TR subtype and not the other. We conclude that TRa and TRb regulate the same genes, different from some homologous NR pairs such as the ERs and classic steroid receptors. Our results agree with previous studies of other groups who found that TRs regulate the same genes in mouse liver [14] and in a prior study conducted in HepG2 cells [16].
While TRa and TRb may regulate the same genes, their actions are far from identical. Part of our analysis overlaps a previous study of similar design by Chan and Privalsky [16], who also stably expressed exogenous TRs in HepG2 cells and found strongly overlapping responses to TRa and TRb after 6 hrs T 3 induction but with gene-specific differences in magnitude of response to the two TR subtypes. While our results do not precisely reproduce previous findings, we regard differences as minor and emphasize that there is remarkable concordance between the key points of the two studies. Chan and Privalsky observed fewer genes that responded to T 3 (<40 versus 150-300) or unliganded TRs (<100 versus 1500-2000) and detected a stronger bias towards TRa versus TRb responses. Both discrepancies are probably explained by differences in TR expression levels. Our comparisons of levels of TR transcripts relative to parental cells in both datasets (not shown) suggest that our cells express more TRs, explaining detection of more TR target genes in our stably transfected  Table 1 HepG2 cells. Moreover, evidence described in the Chan and Privalsky study suggests that their cells may express more TRa than TRb, explaining observed TRa bias. More importantly, however, both groups conclude that TRs regulate the same genes, all genes flagged as T 3 regulated in the Chan study appear in our dataset at the 6 hr time point and regulation patterns appear very similar in both analyses (not shown). Our analysis therefore confirms conclusions of this study and, because it also includes extra time points and two cell types, confirms and extends the conclusion that TRs regulate the same genes with differences in precise magnitude of response. Additionally, the fact that we have confirmed that many early responding genes are direct TR targets with CHX treatment indicates that many differential effects must be related to primary differences in direct TR actions.
Our study has also uncovered other ways that TRa and TRb differ. First, most T 3 responses tend to be stronger with TRa at earlier times and with TRb at 24 hrs, although there are numerous counter-examples of individual genes which deviate from this pattern. Second, a large set of late responding highly TRb-specific genes appears in HepG2 cells. Third, we find that TR subtype preferences in magnitude of T 3 response may appear at selected time points or persist across several time points. Fourth, differences in magnitude of response +/2 T 3 means that the two TR subtypes exhibit different regulation patterns at individual genes. Finally, we detect variations in T 3 dose response in a limited survey of target genes and this effect displays a TR subtypeselective component.
Our experiments do not address mechanisms of differential effects, but do permit speculation about possible causes: 1. We suspect that there are fundamental differences in kinetics of T 3 activation processes in the HepG2 experimental system. We note that: i) faster transcriptional responses to TRa are paralleled by faster T 3 -dependent reductions in steady state TRa levels versus TRb; this often reflects ubiquitin-dependent turnover of transcriptionally active complexes [31] and ii) some verified early direct TRa targets display similar but slower T 3 responses with TRb (see heat map in Figure 5). 2. Early T 3 -dependent changes in gene expression seem to foster an intracellular environment that enhances TRb actions at some genes; the late HepG2 TRb-dependent gene set also responds weakly to TRa suggesting that prolonged T 3 treatment selectively augments TRb action in these contexts. One possible explanation for this effect is that TRs may induce differential expression or activity of transcription factors that regulate downstream genes and possibly cooperate with TRs in some contexts. 3. We note that TRa and TRb subtype-and gene-selective actions emerge within the context of wide gene-specific variations in TR action that have also been observed by other groups [14,16]. We observed that: a) some genes respond to low levels of endogenous TRs in HepG2 parental cells whereas others require exogenous TR expression to mount a detectable response and one gene (ANGPTL4) that is a verified direct TRb target in parental HepG2 cells [22] is even silenced by TR overexpression, b) magnitude and direction of response to T 3 and unliganded TRs varies widely, c) there are gene-specific interactions between CHX and T 3 , see Figure 5 d) there are variations in response patterns of both activated and repressed genes +/2T 3 and this effect displays a cell-specific component and e) T 3 dose response is gene-specific. We suggest that these gene-specific variations in response reflect gene-context specific variations in mechanisms of TR action and that some of these, in turn, highlight differences in TRa and TRb function that are not always apparent from standard reporter assays alone. Elucidation of mechanisms of these effects will require better understanding of gene architecture and TR influences upon transcription factor and cofactor recruitment and we propose that systems described within this paper will help us to dissect influences of gene context upon precise mechanisms of TR action.  Finally, it is important to consider whether gene-specific differential actions of TRa and TRb also occur in vivo and possible physiologic impacts of such effects. We know from previously published studies of wild type and TR gene knockout mice that some gene-specific TR behaviors that we have been able to document in culture have direct parallels in vivo; these include variations in patterns of TR regulation +/2 T 3 and in T 3 concentration dependence [14]. This implies that at least some of the gene-specific differential effects seen in our study will also be observed in vivo. Indeed, our initial survey of effects of TRb knockout upon endogenous liver genes in mice revealed that T 3 response of all genes is diminished but that there are more severe effects at some genes than others (not shown), implying different contributions of TRa and TRb to T 3 response.
We can imagine two situations in which gene-specific differential TR actions would be physiologically important. First, TRa and TRb expression is highly influenced by diurnal rhythm [32]. Thus, variations in TR protein levels could affect the TR target repertoire based upon whether a particular target gene responds preferentially to TRa or TRb; such differential effects would be observed even without changes in T 3 levels. Second, TRb and liver selective thyromimetics (STRMs) such as GC-1 (sobetirome) and KB2115 (eproterome) have been developed to selectively lower serum cholesterol without deleterious effects on heart and combat other aspects of metabolic disease [12]. Some natural TR ligands such as TRIAC are also TRb selective [23]. Doses of such ligands that activate TRb but not TRa would be expected to alter gene expression in a TRb biased manner that would differ from T 3 , which binds the two TRs with similar affinity. It will be interesting to determine whether any such TR subtype and genespecific effects occur in vivo and whether their impact is physiologically or pharmacologically important.  Figure S2 Exogenous TRs are recognized by antibodies against TR primary sequences. We compared TRb and TRa levels in HepG2-TR cells versus HepG2 cells +/21 hr T 3 treatment. The panel shows representative western blots of HepG2-TR and HepG2 cell extracts probed with TR antibodies. Antibodies were TRb (TRb72-93; BabCO, Berkeley Antibody Co., now Covance, Richmond, CA, see reference [22]) and TRa (abcam: ab53729). Secondary antibody was goat anti-mouse IgG-HRP conjugate from Santa Cruz Antibodies. (PPTX) Figure S3 The TR expression tag does not affect T 3 response in transient transfection assays. The panel shows luciferase activities at a standard DR-4 reporter with equivalent (optimal) levels of transfected wild type TRb expression vector or similar flag-TRb vector. (PPTX) Figure S4 Silencing of ANGPTL4 expression by TR overexpression. Results of qPCR analysis to show ANGPTL4 expression levels +/_ T 3 in parental HepG2 cells versus HepG2-TR cells.

Supporting Information
Note the silencing of T 3 response and strong suppression of basal expression levels by unliganded TRs. (PPTX) Figure S5 SiRNA directed against the upstream EGFP coding sequences selectively inhibits responses to exogenous TRs. HepG2-TRb cells were transfected with 5 nM Qiagen negative control siRNA (siRNA_Qiagen-NC) ''UUCUCCGAACGUGU-CACGU'' or and siRNA-EGFP ''GCCACAACGUCUAUAU-CAUGG'', treated with DMSO (vehicle) or 100 nM T3 24 hr later post siRNA transfection, and RNA was isolated 24 hr later. The left panel shows relative expression of TRb +/2 T3 in the presence of control or EGFP siRNA confirming efficient knockdown of TRb transcripts in HepG2-TRb cells (similar knockdown was observed for EGFP). The right panel shows inhibition of T 3 response at the highly induced C10orf10 gene with the EGFP siRNA treatment and not Qiagen negative control (QNC) siRNA. Similar results were obtained with other T 3 induced genes in the HepG2-TRb cells, including anxa1, pck1, slc16a6, and scnn1a. Data presented represents 3 biological replicates, and bars with the same letters are not statistically different (adjusted p-value.0.05; ANOVA, Tukey-HSD on (C10orf10_Cp) -(RPS27A_Cp)).