Introduction

Control of transcription by RNA polymerase II (Pol II) involves regulated formation of multiprotein complexes at the promoter and enhancer regions of the target genes. These complexes are primarily nucleated by site-specific DNA-binding transcription factors (activators and repressors) that respond to cellular, developmental and environmental signals.1 Transcriptional activators, including members of the large nuclear receptor (NR) superfamily, recruit a series of coactivators that serve both to overcome the chromatin barrier and to directly facilitate the entry of Pol II and its associated general transcription factors (GTFs) to generate the transcriptionally active preinitiation complex (PIC).1 Coactivators acting at the level of chromatin2 include both the ATP-dependent chromatin remodeling factors and enzymes that generate covalent modifications of specific residues in nucleosomal histones. Among the histone-modifying enzymes, p300 is particularly well studied and acetylates histone H3, which both decompacts chromatin and leads to recruitment of acetyl-lysine-binding effector proteins.3 The other major class of coactivators is represented by the TAF components of TFIID4 and by Mediator, a multiprotein complex that directly regulates the formation and function of the Pol II PIC.5,6 Although direct physical interactions of some DNA binding transcriptional activators with selected GTFs were reported earlier,7,8 it has remained unclear whether these interactions contribute to target gene activation.

Estrogen-related receptors (ERRα, ERRβ, and ERRγ), initially identified as factors having homology to the estrogen receptor, are physiologically important members of the NR superfamily that have been implicated in high energy-demanding biological processes as well as development and stem cell maintenance.3d, lanes 7 and 8), consistent with the results of the direct protein–protein interaction assays (Supplementary information, Fig. S2j). Correspondingly, an in vitro chromatin transcription assay showed that ΔC5, but not ΔN1, could facilitate ERRα-dependent transcription activation (Fig. 3e, lane 6 vs. lane 5). These results establish that the N-terminal half (∆C5) of PGC-1α is sufficient not only for ERRα binding, but also for Mediator recruitment to DNA-bound ERRα, and thus for ERRα-, PGC-1α- and Mediator-dependent transcription activation.

In conclusion, the above data suggest that unlike the situation for PPARγ/RXRα, the mechanistic basis for the strict requirement of PGC-1α for ERRα in transcriptional activation of chromatin templates in vitro lies both in ERRα’s inability to directly bind critical coactivators p300 and Mediator and in PGC-1α’s ability to act as an adaptor protein that facilitates these interactions.

PGC-1α- and MED1-dependent recruitment of Mediator to ERRα target genes in mouse embryo fibroblasts (MEFs)

To further verify our in vitro results, we assessed ERRα, PGC-1α, and MED1 requirements for activation of cytochrome c (cycs) and isocitrate dehydrogenase a (idh3a), which are metabolic ERRα target genes,25,37 in MEFs, which lack detectable ERRα and PGC-1α. When ectopic ERRα and PGC-1α were jointly expressed, but not when they were expressed alone, idh3a and cycs were efficiently induced (Supplementary information, Fig. S3a). Importantly, following MED1 knockdown in these cells, the responsiveness to ERRα and PGC-1α was reduced (Supplementary information, Fig. S3a, b).

Next, recruitment of ERRα, PGC-1α and Mediator to the cycs and idh3a enhancers was monitored by chromatin immunoprecipitation (ChIP). As expected, ectopic ERRα alone was recruited to the enhancers (Fig. 4a). Ectopic PGC-1α was similarly recruited to the enhancers, but only when ERRα was co-expressed. Interestingly, whereas PGC-1α recruitment was entirely dependent on ERRα, ERRα binding to the enhancers was enhanced by PGC-1α (Fig. 4a). This PGC-1α-enhanced ERRα binding is consistent with our in vitro ITA results (Fig. 3a, lane 6 vs. lane 7 and lane 9 vs. lane 10). As also observed in the ITA, the PGC-1α binding-deficient ERRα-AF2M6 mutant did not recruit PGC-1α in MEFs (data not shown).

Fig. 4: Requirement of PGC-1α and MED1 for efficient Mediator recruitment and ERRα target gene expression in MEFs.
figure 4

a ERRα-dependent PGC-1α recruitment to ERRα target genes. ChIP assays were performed with anti-ERRα and anti-PGC-1α antibodies in MED1-KO or parental WT MEFs that overexpressed SH-hERRα and/or full-length F-PGC-1α (F) or F-PGC-1α-ΔC5 (C5) mutant as indicated. Error bars indicate standard deviations based on three independent experiments with duplicate qPCR reactions in each case. P-values are also shown. b ERRα-, PGC-1α-, and MED1-dependent Mediator recruitment to ERRα target genes. ChIP assays as in a with anti-MED1 and anti-MED30 antibodies.

Full size image

As expected, binding of ERRα and PGC-1α to the enhancers was not impacted in Med1 knockout (KO) MEFs (Fig. 4a). To assess endogenous Mediator recruitment by ChIP, we monitored both the NR-interacting, but sub-stoichiometric MED1 subunit38 and MED30, a component of the core complex.39 Mediator recruitment was critically dependent on both ERRα and PGC-1α (Fig. 4b). MED30/Mediator recruitment was also diminished in Med1 KO cells, indicating that Mediator is recruited via the MED1 subunit (Fig. 4b). When WT PGC-1α and mutant PGC-1α-ΔC5, which was competent for Mediator binding and ERRα-dependent transcription in vitro, were overexpressed at comparable levels, induction levels of cycs and idh3a were essentially equivalent (Supplementary information, Fig. S3c). Similarly, the levels of ERRα-dependent recruitment of WT PGC-1α and PGC-1α-ΔC5 to the cycs and idh3a enhancers were also statistically equivalent (Fig. 4a). PGC-1α-ΔC5 also retained the ability to recruit MED1 and MED30 (and hence the Mediator) to the enhancers (Fig. 4b), consistent with the in vitro data indicating that the PGC-1α N-terminus (ΔC5) is sufficient for interaction with Mediator. These results establish that ERRα-mediated recruitment of PGC-1α (via its N-terminus) is essential for recruiting Mediator (via MED1) to ERRα-bound enhancers and consequent target gene expression. Overall, these cell-based results nicely mirror the results of the various biochemical assays above.

Taken together, the preceding results demonstrate that ERRα does not bind p300 and Mediator directly and that PGC-1α is required for efficient recruitment of p300 and Mediator to promoter-bound ERRα and ERRα-dependent transcription activation.

ERRα physically interacts with initiation factors TFIIH and TFIIB

To understand how ERRα activates transcription in the absence of coactivators, as observed in DNA-templated in vitro transcription reactions (Fig. 1), we first examined whether the ERRα AF2 was involved. Surprisingly, we found that an ERRα AF2 deletion mutant (ERRα∆AF2), which was defective in activating cognate ERRα chromatin template (Fig. 2c), activated the naked DNA-templated transcription nearly as well as the full-length protein, both in the absence (Fig. 5a, lane 3 vs. lane 2) and presence of Mediator (Fig. 5a, lane 6 vs. lane 5). The dispensability of the AF2 in this assay system further emphasizes that the observed overall stimulation of transcription by Mediator is through its direct effects on the basal machinery — potentially entailing recruitment via PIC-bound Pol II.40 Moreover, in cell-based assays, an ERRE3 enhancer-driven reporter was also modestly induced by an ectopic ERRα∆AF2 almost as efficiently as by WT ERRα, whereas the AF2 was nonetheless required for the more dramatic PGC-1α-dependent stimulation of reporter gene expression (Fig. 5b). These results point to a potential coactivator-independent mechanism(s) that involves a domain(s) other than the ERRα AF2 (see further below) and that might act in conjunction with the coactivator-dependent mechanism.

Fig. 5: Direct interactions of ERRα with TFIIB and TFIIH.
figure 5

a AF2-independent DNA-templated transcription by ERRα. Assays as in Fig. 1c with the indicated components. b Induction of an ERRα-responsive firefly luciferase reporter by ERRα-ΔAF2. 293T cells were co-transfected with the indicated expression vectors. Average activities from three independent assays are plotted as fold-increases relative to the ERRα(–)/PGC-1α(–) control. c SDS-PAGE analysis (Coomassie brilliant blue (CBB) staining) of recombinant GST and GST-ERRα. d GST pull-down assay for ERRα interactions with TFIID, TFIIH, and RNA polymerase II. Analysis by immunoblot as in all GST pull-down assays. e GST pull-down assay for ERRα interactions with TFIIA, TFIIB, TFIIE, or TFIIF. f Coimmunoprecipitation assay for ERRα interactions with PC4. Indicated proteins (top) were incubated with PC4, immunoprecipitated with M2 antibody, and monitored by immunoblotting (upper panel). SDS-PAGE with CBB staining of inputs (lower panel). g Schematic illustration of ERRα deletion mutants. TFIIB- and TFIIH-interacting regions of ERRα identified in h and i are highlighted. h GST pull-down assays for interactions of TFIIB with hERRα deletion mutants. Assay as in e with ERRα deletion mutants. h GST pull-down assays for interactions of TFIIH with ERRα deletion mutants. Assays as in d with hERRα deletion mutants monitored by SDS-PAGE with CBB staining. TFIIH monitored by immunoblotting of XPB and MO15 subunits.

Full size image

We next analyzed whether ERRα directly binds to the GTFs present in the DNA-templated transcription assay. Pull-down experiments utilized purified GST-ERRα (Fig. 5c) or FLAG (F)-ERRα (Supplementary information, Fig. S1b) and the individual factors (TFIIA, TFIIB, TFIID, TFIIE, TFIIF, TFIIH, RNA polymerase II, and PC4). The results (Fig. 5d–f) showed that TFIIB and TFIIH directly bound to ERRα. Pull-down assays with GST-TFIIB and serial deletion mutants of ERRα (Fig. 5g; Supplementary information, Fig. S2b) showed that TFIIB binds to ERRα-ΔLBD (lacking the ligand-binding domain) but not to ERRα-Δhinge (lacking the hinge region) (Fig. 5h). Additional pull-down assays with GST-ERRα serial deletion mutants and TFIIH showed that TFIIH bound to ERRα-Δhinge but not ERRα-ΔDBD (Fig. 5i). These results indicated that TFIIB and TFIIH bind, respectively, to the ERRα hinge (H) and DNA-binding (DBD) regions (Fig. 5g) and suggested alternative mechanisms for AF2-independent transcription activation by ERRα from DNA templates.

Interaction of ERRα with the TFIIH p52 subunit contributes to DNA-templated transcription activation in vitro by ERRα

Since TFIIB and TFIIH have general roles in PIC formation, our initial studies used DNA-templated assays (Fig. 1a) to test the functional significance of their interactions with ERRα. Assays to date have failed to show a contribution of the ERRα–TFIIB interaction to ERRα-dependent transcription (data not shown), leading us to focus on the ERRα–TFIIH interaction. Further interaction assays with purified proteins (Supplementary information, Fig. S4a) showed that the six-subunit TFIIH core complex, but not the three-subunit CDK-activating complex (CAK), bound specifically to GST-ERRα (Fig. 6a). Subsequent ITA with individually purified components of the core TFIIH complex (Supplementary information, Fig. S4b) showed selective binding of the p52 subunit (TFIIHp52) to DNA-bound ERRα, whereas TFIIHp62 binding to DNA was non-specific (Fig. 6b). As an initial indication of the significance of the ERR–p52/TFIIH interaction, the addition of monomeric p52 to the transcription assay showed only a minimal effect on basal transcription but a major (inhibitory) effect on ERR-dependent transcription, presumably due to the ability of free p52 to inhibit the ERRα–TFIIH interaction through competitive binding but not the p52 interactions within the stable TFIIH complex (Fig. 6c, lanes 4 and 6 vs. lane 2). Further studies of deletion forms of TFIIHp52, analyzed in the context of reconstituted TFIIH, failed to identify mutations that selectively inhibit ERR-dependent transcription relative to basal transcription (data not shown), leading us to focus on an ERR mutagenesis approach to establish the functional significance of the ERRα–TFIIH interaction.

Fig. 6: Interactions with the p52 subunit of TFIIH promote DNA-templated transcription by ERRα.
figure 6

a Interaction of ERRα with holo TFIIH and TFIIH core. GST pull-down assays as in Fig. 5d. b Specific binding of TFIIHp52 to DNA-bound ERRα. ITA as in Fig. 2b using purified TFIIH subunits detected by immunoblotting with anti-FLAG M2 antibody. c Competitive inhibition of ERRα-dependent transcription by free TFIIHp52. DNA-templated assay as in Fig. 1b. d Direct interaction of the ERRα-DBD with TFIIHp52. GST pull-down assays with TFIIHp52 and hERRα serial deletion mutants as in Fig. 5i. e Alignment of DBDs of human and mouse ERRs. Zinc-finger cysteines are in red, and β-sheets and α-helices are in pink and blue boxes, respectively. Residues in direct contact with a base (B) or the phosphate backbone of DNA (P)41 are indicated. Residues mutated to alanine in the DBD region (ERRα-DBDm) are numbered and identified by a green bar underneath. f Screening for DNA binding-competent ERRα-DBD mutants. ITA as in Fig. 2b, with lysates containing bacterially expressed F-hERRα-DBD mutants, under less stringent conditions (no carrier DNA, see Materials and Methods). g Analysis of DNA-binding abilities of purified hERRα-DBD mutants (ERRα-DBDm). ITA as in f. h Identification of ERRα-DBD mutants deficient in TFIIHp52 binding. ITA as in b under more stringent conditions (see Materials and Methods). Bound proteins were detected by immunoblot and relative amounts (lower two panels) were estimated from standard curves (left 5 lanes in each panel). i Elimination of ERRα-dependent transcription from DNA templates by DBDm6/19 mutations. Assays as in c, with the indicated mutants.

Full size image

As observed with the holo-TFIIH complex, GST pull-down assays with GST-ERRα serial deletion mutants (Fig. 5i) showed that TFIIHp52 bound to the DBD of ERRα (Fig. 6d). We therefore sought to identify ERRα DBD mutants that were selectively deficient in binding to TFIIHp52 and suitable for assessing the requirement of the TFIIHp52–ERRα interaction for DNA-templated transcription activation. Because it was expected that mutations within the DBD might affect the DNA-binding ability of ERRα, serial point mutations were carefully introduced into the DBD (ERRα-DBDm, Fig. 6e) but avoiding cysteine residues that are part of zinc-finger motifs and amino acids that make direct contacts with DNA.41 ITAs showed that several mutants (DBDm1, 6, 9, 14, and 19), when purified, retained DNA-binding abilities at levels comparable to that of WT ERRα (Fig. 6f, g; Supplementary information, Fig. S4c). Further ITAs under very stringent conditions (high detergent concentrations) to more clearly reveal mutant TFIIHp52 binding deficiencies indeed showed variable binding efficiencies, with ERRα-DBDm6 and ERRα-DBDm19 binding comparably, ERRα-DBDm9 and ERRα-DBDm14 binding more strongly, and ERRα-DBDm1 binding more weakly relative to WT ERRα (Fig. 6h, top panel). A parallel analysis of TFIIHp52 recruitment showed significant decreases (relative to WT ERRα) with ERRα-DBDm6 and ERRα-DBDm14 and almost complete loss (near background levels) with ERRα-DBDm1 and ERRα-DBDm19 (Fig. 6h, bottom panel). These results identify ERRα-DBDm6 and ERRα-DBDm19 as TFIIHp52 binding-deficient mutants that retain normal (WT) DNA-binding ability. Notably, the residues mutated in ERRα-DBDm6 and ERRα-DBDm19 are evolutionarily conserved (Supplementary information, Fig. S4d) and closely positioned along the outer surface of the DBD.41 Finally, DNA-templated transcription assays with these mutants showed that ERRα-DBDm1, ERRα-DBDm6, ERRα-DBDm14, and ERRα-DBDm19 did not activate transcription with the GTFs while ERRα-DBDm9 activated as well as WT ERRα (Fig. 6i). Importantly, the transcription activation potential of each mutant corresponded well with the amounts of TFIIHp52 recruited to DNA-bound ERRα in ITA. Additional analyses of ERRα-DBDm6 and ERRα-DBDm19 failed to reveal any effects of the mutations on various ERRα parameters, including dimerization,42 PGC-1α binding, and selected post-translational modifications46 Various other transcription factor-binding motifs were also differentially enriched among the classes; e.g., TEAD4 and BBX were specifically enriched in classes 1 and 8. Thus, although ChIP-seq analyses with individual mutants remain to be performed to dissect the precise mechanism of DNA/TFIIH/AF2-independent ERRβ target gene regulation, various extended ERREs and binding sites for pluripotency-related, as well as other, activators might account for differential functional dependency of ERR-dependent genes on the various pathways in ESCs (see Discussion).

In summary, the diverse interactions of ERRβ with ERREs, TFIIH, and AF2-binding cofactors (PGC-1α and NCOA3) represent physiologically important mechanisms that are critical for ERR functions in maintaining the self-renewal ability of ESCs but are variably required for different target genes.

Discussion

Although the study of transcription activation mechanisms historically began with identification of GTF targets of selected activators, the predominant roles of coactivators was soon realized and emphasis shifted to an understanding of their mechanisms. In this study we used both reconstituted in vitro transcription and cellular and ESC genetic analyses to reveal molecular mechanisms (beyond DNA binding) whereby ERR family members regulate target gene expression. Our data point to two mechanistic pathways for DNA-bound ERRs, one via interaction with GTF TFIIH and the other via interaction with the AF2-binding cofactors PGC-1α and NCOA3, which in turn interact with and facilitate functions of general coactivators (Mediator, p300; Fig. 9). Thus, at least for a subset of transcriptional activators, both GTFs and coactivators are critical mechanistic targets.

Fig. 9: ERR activates transcription through two different pathways.
figure 9

Pathway 1 (blue arrow): in the case of cells expressing PGC-1α (e.g., MEFs, upper panel), the AF2 region of a DNA-bound ERR binds to LxxLL motifs 2 and 3 of PGC-1α, whose N-terminal half in turn facilitates the recruitment of p300 and MED1-containing Mediator. Alternatively, in the case of cells lacking PGC-1α (e.g., ESCs, lower panel), the AF2 region of a DNA-bound ERR binds to NCOA3, which facilitates the recruitment of Mediator and likely p300 as well.13 Histone acetylase p300 facilitates chromatin remodeling and Mediator directly promotes PIC formation (blue dashed line). Pathway 2 (red arrow): the DBD of DNA-bound ERR interacts with the p52 subunit of GTF TFIIH, thereby impacting its transcription initiation functions (red dashed arrow). Both pathways can operate simultaneously to activate transcription on individual target genes.

Full size image

ERR- and PGC-1α-dependent transcription activation

Consistent with results of cell-based studies,25,26,27,28 our in vitro chromatin transcription assays showed that activation by ERRα is strongly dependent on PGC-1α. Whereas we found no direct physical interactions between ERRα and the coactivators p300 and Mediator, unlike the case for many other NRs, biochemical analyses showed that the PGC-1α dependency results from its ability to directly facilitate recruitment of Mediator and p300 to template-bound ERRα. PGC-1α has been suggested to act like a “protein ligand” for orphan ERRs because of specific binding between ERR AF2 and the PGC-1α LxxLL motif and a corresponding strong requirement for ERR target gene activation.47,48 We here provide further evidence, as well as a potential mechanism, to support the notion of PGC-1α as a protein ligand. Thus, analogous to what ensues upon binding of small-molecule ligands to cognate NRs, p300 and Mediator sequentially interact with ERRs only after PGC-1α binds to ERRs. Whether PGC-1α fulfills this role by inducing ERR conformational changes, similar to authentic ligands, or by serving as a scaffold for the recruitment of the cofactors (as seems likely from the observed interactions) remains to be investigated.

Overall, we envision a general multi-step pathway for ERR target gene activation involving (i) ERR binding to ERREs, (ii) ERR recruitment of PGC-1α, (iii) PGC-1α-mediated recruitment of p300 to effect chromatin opening, likely in conjunction with other chromatin cofactors, and (iv) PGC-1α-mediated recruitment of Mediator to facilitate PIC formation and function. In this regard, our finding that Mediator recruited to ERR target genes contains (and requires) the MED1 subunit, which anchors many other direct NR–Mediator interactions, has interesting mechanistic implications. With a wide range of newly described effector functions, Mediator’s role is not simply to facilitate formation of the PIC but also to fine-tune its function at post-recruitment stages.5,6 Although our data do not allow us to conclude whether any direct contacts between ERRs and MED1 are ultimately established, the selection of otherwise sub-stoichiometric MED1-containing Mediator complexes38 implies that Mediator effector functions that might preferentially be induced via this subunit are relevant for ERR activity.

We previously suggested that, through dynamic interactions between the C-terminal RNA-binding and RS domains of PGC-1α and its MED1 subunit, Mediator might coordinate the chromatin remodeling and PIC formation steps.19,20 The C-terminal region of PGC-1α has also been reported to regulate RNA processing,49 as well as to modulate PGC-1α coactivator activity through post-translational modifications.50,51 In the present study, focusing on interactions of template-bound ERR, we identified additional critical roles for the N-terminal region of PGC-1α in Mediator-dependent gene activation by ERR. Indeed, a PGC-1α deletion mutant (ΔC5) that lacks the RNA-binding region and the RS domain, retained full capability of binding to ERRα and Mediator and activating genes. Although these results might be a reflection of the dynamic nature of PGC-1α interactions, or of additional PGC-1α interactions with other Mediator subunits, they also highlight a potential dispensability of the C-terminal half of PGC-1 in certain scenarios, as also suggested by the existence of functional isoforms lacking these domains.52

TFIIH as an ERR activation target

Our study also revealed direct physical interactions between ERRs and TFIIH that were functionally critical both in vitro and in ESCs. Although several transcription factor–TFIIH interactions have been reported (Supplementary information, Table S3), possible functional effects, including transcription factor phosphorylation by TFIIH (see below), are largely unexplored. Even in the exceptional cases of xeroderma pigmentosum and trichothiodystrophy,53,54,55,56 where TFIIH subunits (XPB, XPD, p62, CAK) have been implicated in these pathologies (Supplementary information, Table S3), there has been no clear-cut identification of specific transcription factor-interacting TFIIH subunits. This likely reflects the inherent difficulty in maintaining the essential function of a GTF such as TFIIH in global transcription while assessing (by mutation) its gene-specific functions through interactions with distinct activators. Here, by uniquely combining biochemical and genetic analyses and taking advantage of identified ERR DBD mutants that are selectively defective in GTF (TFIIH) interactions relative to other activator functions, we have clearly established: (i) direct binding of ERRs to TFIIHp52; (ii) ERR domains and residues selectively required for TFIIHp52 binding; (iii) the importance of the TFIIH–ERRβ interaction for activation of select ERRβ target genes in ESCs; and (iv) a corresponding requirement of the TFIIH–ERRβ/γ interaction for ESC self-renewal. We also report, for the first time, that by targeting a general initiation factor, an NR DBD can execute an activating function. Notably, our approach of identifying interactions of immobilized DNA-bound ERR that may have undergone significant conformational changes,41,57,58 and eliminates non-specific cofactor interactions through the DNA-binding surface of the DBD, was instrumental in revealing this unusual interaction.

How TFIIH binding results in activation of ERR-dependent genes remains unclear. Reports that various activators, including NRs, are phosphorylated by TFIIH CAK56,59,60 suggested that CAK-dependent phosphorylation of ERR might modulate its activity. However, we found that ERRα phosphorylation efficiency (by TFIIH) was very low (< 1%) and did not affect ERRα-dependent transcription in vitro (Supplementary information, Fig. S4i–m). Because p52 modulates TFIIH/XPB ATPase activity required for promoter opening and promoter escape,61,62,63 the ERRα–TFIIH interaction might yet regulate these TFIIH activities, potentially in concert with the Mediator.64

Variable requirements for the functions of ERRβ interactions with ERREs, TFIIH and AF2-binding factors in ESCs

It has been reported that ESCs are deficient in PGC-1α and PGC-1β (although they do contain the family member PRC) and are dependent on another cofactor (NCOA3/SRC3) that also interacts with the ERRβ AF2 in ESCs.45 We now show that NCOA3, like PGC-1α, can facilitate Mediator recruitment to DNA-bound ERRβ/γS through a direct interaction with Mediator (Supplementary information, Fig. S6i, j). Moreover, the ERRβ/γS AF2M6 mutant that is defective in binding PGC-1α is also defective in binding NCOA3 (Supplementary information, Fig. S6h). Thus, NCOA3 is a functional analog of PGC-1α (see above) and, given further its predominance in ESCs, the ERRβ/γS AF2M6 mutant thus served as a good proxy for assessing the roles of ERR–NCOA3 interactions in our genomic analyses of ERR-dependent gene regulation in ESCs.

In particular, our detailed analyses of ERRβ function in ESCs helped determine whether the activation pathways dependent, respectively, on interactions with TFIIH and the AF2-binding cofactor (probably NCOA3) operate independently or cooperatively. A large number of ERRβ target genes were identified by RNA-seq analysis and, based on complementation experiments, were grouped into 8 classes based on dependency on a given interaction (ERRE, TFIIH, or NCOA3). Interestingly, by GO analyses, each class was enriched with different biological gene sets (Fig. 8; Supplementary information, Fig. S6k, n, and Table S1). It should be emphasized that the largest class contained genes that were not rescued by ERRβ derivatives with mutations that individually compromised interactions with ERREs, TFIIH, and NCOA3 (Class 1 in Fig. 8d). These results, together with CFAs, which demonstrated that individual ERRβ mutants do not restore the self-renewal ability of ERRβ-KO ESCs (Fig. 7b), underscore a requirement for all interactions for the ERR transactivation function and support the general model from the biochemical studies. On the other hand, a subset of genes was not reliant on one or two of the analyzed interactions (Classes 2–7 in Fig. 8d), indicating that all three kinds of interactions are not always required and are redundant for target gene activation. Moreover, other subsets of target genes were successfully rescued by individual DNA-, TFIIH-, and NCOA3-binding mutants of ERRβ (Class 8 in Fig. 8d).

Although the underlying basis for what determines pathway choice for each class remains unclear, enriched motif analysis at ERRβ-binding sites proximal to target genes indicated several potential mechanisms. First, 1–3 bp sequences adjacent to core ERREs were found to vary between classes (Supplementary information, Fig. S6n). In light of NR structural studies that show how different recognition sequences confer different NR conformations and cofactor dependencies,65,66,67 one possibility is that the ERRβ interactions with ERREs containing class-specific adjacent sequences might affect ERRβ conformation to facilitate preferential binding with one or the other factor (e.g., TFIIH vs. NCOA3). Alternatively, because binding motifs for various transcription factors, including pluripotency-related activators, were differentially enriched in each class (Supplementary information, Fig. S6n), these colocalizing activators might substitute for ERRβ and interact with their preferred factor. Indeed, specific genomic sites in ESCs are reported to be targeted by multiple transcription factors, including ERRβ,68,69,70 further raising the possibility that at some loci ERRβ might function just as an architectural factor to facilitate formation of multi-protein regulatory complexes on target genes even without interacting with cognate ERRE, TFIIH, or NCOA3 (Class 8 in Fig. 8d). Future studies will be directed at understanding how these variables control ERR–cofactor interactions. Furthermore, it remains unclear whether the observed ERRβ-mediated repression of genes regulating RNA-related cellular function reflects direct effects — our in vitro studies revealed evidence only of transcription activation — this possibility is consistent with previously reported interactions of ERRβ with transcriptional repressors and corepressors.71,72

Distinct and overlap** functions of ERR family members in ESCs

This study also highlights both redundancies and functional differences among ERR family members in ESCs, even as they behave indistinguishably in our current in vitro transcription assays that are mainly designed to reveal critical cofactor dependencies under specified constraints. It was previously shown that ERRα and ERRγ targeted a common set of promoters and interacted with a similar set of proteins.73,74 Nonetheless, we found that ERRγ, but not ERRα, could replace ERRβ for ESC self-renewal; indeed, ERRβ and ERRγ regulated almost the same set of genes (Fig. 8a). Moreover, and most importantly, our biochemical analyses showed AF2-dependent interactions of ERRβ and ERRγ, but not ERRα, with NCOA3. Thus, because it mimics an ERR mutant defective in interaction with NCOA3, ERRα is unable to substitute for ERRβ in ESCs. While not ruling out interactions with additional AF2-binding cofactors in ESCs, these results simultaneously confirm the reported requirement of NCOA3 and establish a basis for the selective function of ERRβ (and ERRγ) in these cells. These results also indicate that in ESCs, ERRβ and ERRγ may selectively interact with common coactivators besides TFIIH and NCOA3, possibly through their N-terminal AF1 regions that have relatively high homology (~60%) to each other but low homology (~20%) to ERRα.S6c. These cells were designated KO/ERRβ-wt cells whereas ERRβ-KO ESCs that were transduced with a control lentivirus vector that does not express ERRβ were designated KO/mock cells.

CFA of ESCs and qPCR

For CFA, a 24-well plate was first coated with 400 µL of 0.01% poly-L-ornithine (Sigma; dissolved in water) at RT overnight. After washing with PBS (–) twice, it was coated with 310 µg/mL laminin (Invitrogen; dissolved in PBS (–)) at 37 °C overnight. After two washes with PBS (–), N2B27 medium was added to the plate. ESCs maintained in N2B27-LIF/PD/CH medium, with or without 2 µg/mL Dox, were seeded at the appropriate clonal density (200 cells/24-well plate). Medium contained either LIF/PD/CH or only PD/CH, with or without 2 µg/mL Dox. For RNA preparation, 1600 cells were seeded in a poly-L-ornithine- and laminin-coated 12-well plate in N2B27-LIF/PD/CH medium. After 4 days in culture, the cells were fixed in 4% paraformaldehyde and stained with alkaline phosphatase as described.94 The stained colonies were scanned and quantified by using Photoshop (Adobe) software as follows. The colonies were first automatically detected by the “color range” function against a fixed level of red color that covers all red colonies. This was followed by a manual correction to remove false positives, i.e., diffuse reflection from solution surface. Then paths against the selected pixels were made and the paths were filled with flat blue color, i.e., RGB code (0, 0, 255). The blue pixel was then automatically counted using the “counting tool” function against RGB (0, 0, 255) color and the counting log was recorded via the “record measurement” function. The colony number was finally calculated from total area value of the log. Note that we always used the same color extraction file for the “color range” function to detect red colonies so that the definition of red-colored colonies was statistically the same across the various assays.

To analyze pluripotency-related genes under 2i conditions, each ESC line was first seeded in N2B27-LIF/PD/CH medium and cultured for 12 h; after two rinses with a 1:1 mixture of DMEM/F12 and Neurobasal media, N2B27-PD/CH was finally added. After culturing for 3 days, the cells were harvested for RNA preparation using RNeasy (Qiagen) with inclusion of DNase treatment step. RT-qPCR was performed with Quantitect SYBR Green PCR master mix (Qiagen), appropriate primer sets (Supplementary information, Table S4), and cDNAs prepared from RNA purified using a SuperScript III kit (Invitrogen). Fold-changes of each gene expression were calculated by the 2ΔCt method. Relative expression levels of ERRα/β/γ were determined by RT-qPCR with appropriate primer sets (Supplementary information, Table S4) and copy number-known plasmid vectors containing ERRα/β/γ cDNAs as standards.

RNA-seq and data analysis

RNA samples were obtained from two clones of each of WT/mock, KO/mock, KO/ERRβ-wt, KO/ERRβ-DBDm3, KO/ERRβ-DBDm6, KO/ERRβ-AF2M6, and KO/ERRβ (described in detail in the main text) ESC lines that had been cultured in N2B27-PD/CH medium and Dox for 3 days. The library for RNA-seq was prepared from 400–500 ng of isolated RNA by using TruSeq Stranded Total RNA with Ribo-Zero kit (Illumina) according to the manufacturer’s instructions. The pooled libraries with unique index sequences were sequenced using NovaSeq SP (Illumina) with single-end 100 bp reads at the Rockefeller University Genomics Resource Center. The raw data and a count matrix (see below) are posted at Gene Expression Omnibus with accession number GSE196202.

To generate expression count matrix, raw reads were trimmed to remove adaptor sequences by Skewer (v.0.2.2) and mapped to mm10 genome by STAR (v.2.7.8a), and then mapped reads were counted by featureCounts (v.2.0.10).

The package of edgeR glmQLFTest (v3.32.1) was used to identify differentially expressed genes (DEGs) using the count matrix. Briefly, the library size normalization, the dispersion estimation, and then the generalized linear model fitting were sequentially performed with ‘calcNormFactors(y, method = TMM), ‘estimateDisp(y, design = design, robust = TRUE)’, and ‘glmQLFit(y, design = design)’, respectively. Finally, log2 fold change (FC) and false discovery rate (FDR) of each gene between two groups were calculated by glmQLFTest. Genes with FDR < 0.05 and abs(log2FC) > 0 were identified as DEGs. To identify ERRβ target genes, WT/Mock and KO/ERRβ-wt groups were compared with KO/Mock group to select DEGs. To classify the ERRβ target genes into classes 1–8, KO/Mock group was compared with groups of all combinations of each mutant expressing clones to obtain temporal DEGs. For example, the comparison of KO/Mock vs. KO/ERRβ-DBDm3 + KO/ERRβ-DBDm6 + KO/ERRβ-AF2M6 was used for class 8, and the comparison of KO/Mock vs. KO/ERRβ-DBDm3 + KO/ERRβ-AF2M6 was used for class 2. Then, if the temporal DEGs were found in the original ERRβ target genes, they were finally identified as the classified genes. If genes existed across multiple classes, those were allocated to one class according to the following priority: classes 1, 8 > 2, 3, 5 > 4, 6, 7. To identify ERRγ target genes, the ERRβ target genes were compared between KO/ERRγ and KO/ERRβ + WT/Mock groups, and DEGs were defined as non-ERRγ target genes.

Heatmaps were generated from z-score log2(transcripts per million) with complexHeatmap package. GO enrichment was computed by ‘clusterProfiler’s enrichGO’ and compareCluster (v3.18.1).

To identify ERRβ binding region within ±50 kb from TSSs of the classified genes, the called peaks (q < 1E−5) of public ChIP-seq data performed with anti-ERRβ antibody and ESCs (SRX093166, SRX4004790, SRX4004792, SRX4158594, SRX4158595, SRX4167129, SRX4167130, SRX5023707, SRX5023708, SRX5023709, SRX5023710, ChIP-Atlas (https://chip-atlas.org)) were merged by Homer (v.4.10) to obtain ERRβ binding regions in ESCs (Supplementary information, Table S2). Next, the classified gene coordinates were obtained by ‘TxDB.Mmusculus.UCSC.mm10.knownGene’. Then, ±50 kb window from each TSS was created by ‘resize (gr, width = 100000, fix = “start”)’, and ‘subsetByOverlaps’ was run to select the regions that overlap the merged ERRβ binding regions and the ±50 kb windows. Then, we used Homer (v.4.10) for de novo motif enrichment analysis with ‘findMotifGenome.pl -size 200 -mask’ at the selected regions.