Insulin action and resistance are dependent on a GSK3β-FBXW7-ERRα transcriptional axis

Insulin resistance, a harbinger of the metabolic syndrome, is a state of compromised hormonal response resulting from the dysregulation of a wide range of insulin-controlled cellular processes. However, how insulin affects cellular energy metabolism via long-term transcriptional regulation and whether boosting mitochondrial function alleviates insulin resistance remains to be elucidated. Herein we reveal that insulin directly enhances the activity of the nuclear receptor ERRα via a GSK3β/FBXW7 signaling axis. Liver-specific deletion of GSK3β or FBXW7 and mice harboring mutations of ERRα phosphosites (ERRα3SA) co-targeted by GSK3β/FBXW7 result in accumulated ERRα proteins that no longer respond to fluctuating insulin levels. ERRα3SA mice display reprogrammed liver and muscle transcriptomes, resulting in compromised energy homeostasis and reduced insulin sensitivity despite improved mitochondrial function. This crossroad of insulin signaling and transcriptional control by a nuclear receptor offers a framework to better understand the complex cellular processes contributing to the development of insulin resistance.

but then a third consensus should be indicated in Fig. 3d.
As a minor point, the coIPs in Ext Data Fig 4e do not necessarily indicate direct interaction, since these were done using transfected cells. Text at top of p. 10 may overstate the interpretation. A similar point is relevant for Fig. 4c; in this case the figure supports the idea that the interaction is physiologic, though not necessarily that it is direct (data in Fig. 4d are more convincing that it is direct, when taken together with previous work). Softening the language would be appropriate. Fig. 4k and Ext Data Fig. 4j present data from FBXW7 floxed mice treated with adenoCre, however no immunoblots of FBXW7 are shown. This is needed to assess the degree of knockout. Fig. 5b and 5c extend the model to muscle, as well as liver, but it is not clear whether FBXW7 is present in muscle. Has this been shown in previous work? Or might some other E3 regulate ERRalpha in muscle?

Data in
In Fig. 6, n is given as a range (e.g. n=6-8). This makes interpretation of the statistical tests impossible. Exact n values need to be provided for each group. This is a concern for other figures as well. In some cases, individual data points are shown, but the reader should not have to count these, and in other cases there is only a bar graph with error bars (and so it is especially important to know n for each bar). Fig. 6 shows that 3SA mice have postprandial insulin resistance, but the impairment in Akt signaling is minimal. Recent data describe the role of TUG cleavage in muscle glucose uptake (PMID: 33686286), and it would be interesting to know if this pathway is impaired. Do the DEGs in 3SA muscle overlap with genes regulated by TUG-C/PPARgamma/PGC-1alpha?
Can a ERRalpha phosphatase be identified? This is certainly beyond the scope of the present work, but this might be an important regulator. This could be pointed out in the discussion.
Reviewer #2 (Remarks to the Author): In this manuscript, Xia et al. discovered an insulin-dependent post-translational modification mechanism for ERRα stability and activity that involves ERRα serine phosphorylation by GSK3β followed by ubiquitination and degradation by FBXW7. This new mechanism is well supported by extensive gain and loss-of-function studies, including in vivo studies using liver GSK3β KO, liver Fbxw7 KO mice and especially the ERRα 3SA mutant mice. Overall, the findings are significant and the manuscript is well written.

REVIEWER COMMENTS
Reviewer #1 (Remarks to the Author): The manuscript by Xia et al reports data from a substantial body of work that, together, supports a model in which insulin inhibits the kinase GSK3beta, which reduces phosphorylation of the transcriptional regulator ERRalpha. As phosphorylation of ERRalpha promotes its FBXW7-mediated ubiquitination and degradation, this pathway can regulate ERRalpha abundance and control oxidative metabolism.
In general the results are convincing and well described. The data will be of high significance for the regulation of metabolism and for insulin action. There are several points that ought to be clarified and/or further supported prior to publication.
Data for genes used to generate the heat maps in Figs. 1g and 1h should be included in a supplementary spreadsheet. The range for the Z score is given in the figure legend, but it would be helpful to have a scale on the figure itself. It would also be helpful to know the fold change, and not just the Z score. This could be on a spreadsheet.
Spreadsheets should similarly be included for genes identified by ChIP-seq in Extended Data Fig. 1. On page 5, the sentence "Consistently, intersection of transcriptome and cistrome datasets revealed…." is confusing. It appears that the cistrome being referred to is the ChIP-seq data, but this experiment has not yet been introduced.
The model in Fig. 2g is confusing. Phosphorylation activates Akt, but inhibits GSK3b. Perhaps the phosphates can be indicated in different colors, to indicate how they modulate the activity of the target enzyme? The arrow from GSK3b to ERRalpha just indicates "stabilization" but this is actually a lack of phosphorylation and, hence, stabilization. The figure legend could be used to help clarify the model. In Ext. Data Fig. 3a it is not clear what SE and LE indicate. Presumably, SE is the slower migrating phosphorylated form and LE is the unphosphorylated protein. However this is not clear from the legend, and there is no description of the Phos-tag gel system in the methods section. Also, the two bands are shown on separate panels, so it is not possible to estimate what fraction of the protein is phosphorylated. As well, the authors conclude from this figure that "insulin-stimulated nuclear ERRα protein was unphosphorylated (Extended Data Fig. 3a), indicating that ERRα translocation between the cytoplasm and nucleus is phosphorylation-dependent." This does not follow from the data, which include only nuclear extracts. The conclusion is supported by Ext. Data Fig.  3b, however this relies on transfected cells. Even so, the language could be softened.
None of the immunoblots shown have molecular weight markers indicated, and these should be added to the figures.
Only two GSK3beta consensus sequences are indicated in Fig. 3d Fig. 4c; in this case the figure supports the idea that the interaction is physiologic, though not necessarily that it is direct (data in Fig. 4d are more convincing that it is direct, when taken together with previous work). Softening the language would be appropriate. Fig. 4k and Ext Data Fig. 4j present data from FBXW7 floxed mice treated with adenoCre, however no immunoblots of FBXW7 are shown. This is needed to assess the degree of knockout. Fig. 5b and 5c extend the model to muscle, as well as liver, but it is not clear whether FBXW7 is present in muscle. Has this been shown in previous work? Or might some other E3 regulate ERRalpha in muscle?

Data in
In Fig. 6, n is given as a range (e.g. n=6-8). This makes interpretation of the statistical tests impossible. Exact n values need to be provided for each group. This is a concern for other figures as well. In some cases, individual data points are shown, but the reader should not have to count these, and in other cases there is only a bar graph with error bars (and so it is especially important to know n for each bar). Fig. 6 shows that 3SA mice have postprandial insulin resistance, but the impairment in Akt signaling is minimal. Recent data describe the role of TUG cleavage in muscle glucose uptake (PMID: 33686286), and it would be interesting to know if this pathway is impaired. Do the DEGs in 3SA muscle overlap with genes regulated by TUG-C/PPARgamma/PGC-1alpha?
Can a ERRalpha phosphatase be identified? This is certainly beyond the scope of the present work, but this might be an important regulator. This could be pointed out in the discussion.
In this manuscript, Xia et al. discovered an insulin-dependent post-translational modification mechanism for ERRα stability and activity that involves ERRα serine phosphorylation by GSK3β followed by ubiquitination and degradation by FBXW7. This new mechanism is well supported by extensive gain and loss-of-function studies, including in vivo studies using liver GSK3β KO, liver Fbxw7 KO mice and especially the ERRα 3SA mutant mice. Overall, the findings are significant and the manuscript is well written. The authors claim that "insulin-stimulated nuclear ERRα protein was unphosphorylated (Extended Data Fig. 3a)", but the short exposure (SE, supposedly more accurate) phospho-ERRα level seems clearly increased. (d) In Fig 3g, why do 3SA mutations affect GSK3β S9 phosphorylation?

Liming Pei
Xia and colleagues showed insulin controls the activity of ERRα via GSK3b/FBXW7 axis in vitro. They demonstrated ERRα is phosphorylated by GSK3b, downstream target of PI3K-AKT pathway, and ubiquitylated by FBXW7 leading to proteasome dependent degradation. In the latter part of the manuscript, they generated mice with ERRa3SA, and found that ERRa3SA mice develop insulin resistance, which showed the biological significance of ERRα in vivo. Overall, the authors have done an excellent analysis, and revealed that ERRα is one of key molecules that control energy homeostasis.
Specific points: 1. In Extended Fig 1h, 7079 targets of ERRα ChIP-seq are shown, and it seems that these targets are from 20 kb upstream and downstream of genes. Should the targets from 20 kb downstream of genes be included in this analysis because ERRα is a transcriptional factor? Do the 7079 targets include multiple sites from one gene? 2. Alb-Cre and Gsk3bf/f mice are both controls, but ERRα expressions are so different between them. This figure is confusing ( Figure 2d). Also, it is better to show the data from Alb-Cre/Gsk3b+/f. Figure 2e and 2f, however, mRNA levels should be examined to indicate that these are caused by post-translational modification, not by transcriptional level. 8. In case of analysis of conditional knockout mice, controls should be Alb-Cre or Alb-Cre/Fbxw7+/f (Fig 4k).

The accumulation of ERRα protein is nicely shown in
9. There are three isoforms in FBXW7. FBXW7α is ubiquitously expressed, and FBXW7γ is expressed in muscle. Which isoform is an ubiquitin ligase of ERRα in muscle ( Figure 5b)?
10. The time points with significant difference are varied in the night time, although C29 was injected at the same time ( Figure 7k). Is there any reason for this?

Response to NCOMMS-21-40051-T comments
Reviewer #1 (Remarks to the Author): The manuscript by Xia et al reports data from a substantial body of work that, together, supports a model in which insulin inhibits the kinase GSK3beta, which reduces phosphorylation of the transcriptional regulator ERRalpha. As phosphorylation of ERRalpha promotes its FBXW7-mediated ubiquitination and degradation, this pathway can regulate ERRalpha abundance and control oxidative metabolism.
In general the results are convincing and well described. The data will be of high significance for the regulation of metabolism and for insulin action. There are several points that ought to be clarified and/or further supported prior to publication.
We appreciate that the reviewer recognizes the significance of our study and provides us with helpful comments. We have carefully addressed all points raised in our revised manuscript. Please see our detailed response below.
1. Data for genes used to generate the heat maps in Figs. 1g and 1h should be included in a supplementary spreadsheet. The range for the Z score is given in the figure legend, but it would be helpful to have a scale on the figure itself. It would also be helpful to know the fold change, and not just the Z score. This could be on a spreadsheet.
We thank the reviewer for pointing this out. We presented the specific Z score and fold change of genes used to generate original Figs Supplementary Fig. 6m). Fig. 1. On page 5, the sentence "Consistently, intersection of transcriptome and cistrome datasets revealed…." is confusing. It appears that the cistrome being referred to is the ChIP-seq data, but this experiment has not yet been introduced.

Spreadsheets should similarly be included for genes identified by ChIP-seq in Extended Data
Regarding the cistrome datasets, we previously stated in the Supplementary legend as ERRα ChIP-seq data (GSE43638), which was done in our published study (PMID: 23562079). We apologize for not describing it clearly in the main text and have clarified this point in the revised manuscript.
We also presented the genes identified by ChIP-seq studies in original Extended Data Fig. 1h, i (Now Supplementary Fig. 1k, Fig. 2g is confusing. Phosphorylation activates Akt, but inhibits GSK3b. Perhaps the phosphates can be indicated in different colors, to indicate how they modulate the activity of the target enzyme? The arrow from GSK3b to ERRalpha just indicates "stabilization" but this is actually a lack of phosphorylation and, hence, stabilization. The figure legend could be used to help clarify the model.

The model in
We fully agree with the reviewer on this point and have revised the model in original Fig.  2g (Now Fig. 2j) as suggested. We've colored the activating phosphate for AKT in green and the inhibiting phosphate for GSK3β in black. We've also added an "Activation" beside the glowing p-AKT and added an "Inactivation" beside the p-GSK3β. Please see below.
We have revised the legend as "j Proposed model of insulin-mediated stabilization of ERRα protein through the PI3K/AKT/ GSK3β pathway." We did not state ERRα phosphorylation in this legend, because we only proved that GSK3β kinase activity is required for insulin-mediated stabilization of ERRα protein in Fig 2, and the true effects of GSK3β on ERRα phosphorylation status were not shown until Fig 3. However, the revised model presents this possibility using greyed phosphorylation sites.
4. In Ext. Data Fig. 3a it is not clear what SE and LE indicate. Presumably, SE is the slower migrating phosphorylated form and LE is the unphosphorylated protein. However this is not clear from the legend, and there is no description of the Phos-tag gel system in the methods section. Also, the two bands are shown on separate panels, so it is not possible to estimate what fraction of the protein is phosphorylated. As well, the authors conclude from this figure that "insulin-stimulated nuclear ERRα protein was unphosphorylated (Extended Data Fig. 3a), indicating that ERRα translocation between the cytoplasm and nucleus is phosphorylation-dependent." This does not follow from the data, which include only nuclear extracts. The conclusion is supported by Ext. Data Fig.  3b, however this relies on transfected cells. Even so, the language could be softened.
We apologize for the unexplained abbreviations (SE, LE) and confusing statement. In the original Extended Data Fig.3a (see below), the two bands revealed in the phos-tag gel are the same membrane with a short exposure (SE; middle panel) and longer exposure (LE, lower panel), we could not observe a band shift with the nuclear extract even using longer exposure, thus we previously stated in the main text that "insulinstimulated nuclear ERRα protein was unphosphorylated".
We optimized this experiment by adding whole cell extract (WCE) and cytoplasmic fraction as controls and properly labeled the bands (New Supplementary Fig. 3a). As shown below, in comparison with whole-cell extract, phos-tag gel examination of the nuclear extract from HepG2 cells could not reveal an upper phosphorylated ERRα band shift, indicating that insulin-induced nuclear ERRα protein was dephosphorylated. This was followed by introducing mutations to the insulin-sensitive ERRα residues which showed the nuclear export of ERRα exclusively with the phospho-mimicking ERRα mutant, implying that ERRα translocation between the cytoplasm and nucleus is mediated by phosphorylation ( Supplementary Fig. 3b). We have modified the main text and softened the language as suggested.
Further, methods and mechanism for the phos-tag gel were added in the "Preparation of cell or tissue lysates and immunoblotting" section of Method: SuperSep™ Phos-tag™ (50μmol/L), 7.5% precast gels (FUJIFILM Wako Pure Chemical Corporation, Cat. No. 4548995049858) were used to separate phosphorylated and dephosphorylated ERRα proteins. Gels were agitated gently in a transfer buffer (25 mmol/L Tris, 192 mmol/L Glycine, 10% MeOH) containing 10 mmol/L EDTA for 3*10 minutes, followed by agitation in a transfer buffer that does not contain EDTA for another 10 minutes before transferring to PVDF membranes. The membranes were blocked with 2% Milk/TBST, reacted with ERRα antibody, and detected by chemiluminescence. Given phosphorylated ERRα protein is bound and trapped by Phos-tag™ during SDS-PAGE, it is separated from dephosphorylated ERRα protein and presents as a slower migrating upper band.
Correspondingly, we pinpointed the two ERRα bands revealed by phos-tag gel as phosphorylated (slow-migrating upper band) and dephosphorylated (lower band) in Fig  3b,   9 We thank the reviewer for pointing this out. We have now added the molecular weight markers for all the presented blots.
6. Only two GSK3beta consensus sequences are indicated in Fig. 3d, but Fig. 3k diagrams GSK3B phosphorylating three sites. This can be explained if E30 acts as the priming site, as indicated in Fig. 3k, but then a third consensus should be indicated in Fig. 3d.
We agree with the reviewer and have now included the third putative GSK3β consensus phosphorylation motif in Fig. 3d, see below. We also revised it correspondingly in the main text.
7. As a minor point, the coIPs in Ext Data Fig 4e do not necessarily indicate direct interaction, since these were done using transfected cells. Text at top of p. 10 may overstate the interpretation. A similar point is relevant for Fig. 4c; in this case the figure supports the idea that the interaction is physiologic, though not necessarily that it is direct (data in Fig. 4d are more convincing that it is direct, when taken together with previous work). Softening the language would be appropriate.
As suggested, we revised the description of co-IPs in Ext Data Fig 4e as "Co-IP experiments confirmed the interaction between the 6 E3 candidates and ERRα ( Supplementary Fig. 4e)." Similarly, we revised the description for Fig. 4c as "Using hepatocytes, we confirmed that endogenous FBXW7 and ERRα physiologically interact with each other (Fig. 4c)." 8. Fig. 4k and Ext Data Fig. 4j present data from FBXW7 floxed mice treated with adenoCre, however no immunoblots of FBXW7 are shown. This is needed to assess the degree of knockout.
We thank the reviewer for raising this valid point, we have now confirmed this with immunoblotting as shown in Fig. 4k and original Ext Data Fig. 4j (Now Supplementary  Fig. 4k). We also validated the knockdown via RT-PCR using a primer specifically targeting the floxed FBXW7 exon (New Fig. 4l). See below.
9. Data in Fig. 5b and 5c extend the model to muscle, as well as liver, but it is not clear whether FBXW7 is present in muscle. Has this been shown in previous work? Or might some other E3 regulate ERRalpha in muscle?
FBXW7 exists as three protein isoforms (α, β, and γ) that differ in subcellular locations: FBXW7α in the nucleoplasm, FBXW7β in the cytoplasm, and FBXW7γ in the nucleolus. FBXW7α is functionally the most dominant isoform, which is ubiquitously expressed and carries out the most known FBXW7 functions (reviewed in PMID: 25314076).
We first confirmed that ERRα protein was stabilized in C2C12 myoblast cells upon proteasome inhibition without altering its mRNA level (New Supplementary Fig. 5e, f). We then assessed and observed comparable expression levels of FBXW7α mRNA in liver and muscle (New Supplementary Fig. 5g). See below.
Thus, we speculate the accumulated ERRα protein in the muscle of ERRα 3SA mice is caused by the conservation of the insulin-GSK3β-FBXW7-ERRα axis in muscle, which remain to be further validated.
10. In Fig. 6, n is given as a range (e.g. n=6-8). This makes interpretation of the statistical tests impossible. Exact n values need to be provided for each group. This is a concern for other figures as well. In some cases, individual data points are shown, but the reader should not have to count these, and in other cases there is only a bar graph with error bars (and so it is especially important to know n for each bar).
We thank the reviewer for raising these valid points. To addresses these concerns, we have specified the exact sample sizes in related figures, we've also revised the bar graphs to show Individual data points.
11. Fig. 6 shows that 3SA mice have postprandial insulin resistance, but the impairment in Akt signaling is minimal. Recent data describe the role of TUG cleavage in muscle glucose uptake (PMID: 33686286), and it would be interesting to know if this pathway is impaired. Do the DEGs in 3SA muscle overlap with genes regulated by TUG-C/PPARgamma/PGC-1alpha?
Indeed, the attenuation of refeeding-induced AKT phosphorylation is minimal in ERRα 3SA muscle, but the effects of insulin on GSK3β and GS phosphorylation were mostly blunted in ERRα 3SA muscle, accounting for its parallel defect in muscle postprandial glycogenesis (Fig. 6c, d).
We fully agree with the reviewer that other PI3K-AKT independent glucose uptake mechanisms might also contributed to the postprandial insulin resistance in ERRα 3SA muscle.
For example, insulin-stimulated TUG cleavage facilitates GLUT4 translocation to the cell surface and increase muscle glucose uptake, which is independent of PI3K activity.
Here, we examined TUG cleavage by immunoblots using a Tug Antibody (Cell signaling #2049) and observed impaired refeeding-induced TUG cleavage in ERRα 3SA muscle (see below Fig. for Reviewer 1).
We did not include these findings in the main text considering the word limitations, but these data indicate that ERRα is also involved in other pathways of muscle glucose uptake, which will be interesting to explore. We have modified our discussion to acknowledge other potential glucose uptake mechanisms: "It will be interesting to explore whether ERRα is responsible for insulin-stimulated glucose disposal in skeletal muscle that are independent of the PI3K-AKT signaling pathway, as attenuation of refeeding-induced AKT phosphorylation is minimal in ERRα 3SA muscle, and multiple studies revealed additional pathways perturbing muscle glucose tolerance and insulin sensitivity in addition to attenuated AKT signaling." 12. Can a ERRalpha phosphatase be identified? This is certainly beyond the scope of the present work, but this might be an important regulator. This could be pointed out in the discussion.
We agree with the reviewer that this is an important point. The phosphatase(s) involved in ERRα dephosphorylation remains elusive and is a derived project of this manuscript. We speculate that phosphatase calcineurin and PP2A might be involved in the regulation of ERRα phosphorylation and subcellular localization and are currently validating the hypothesis using pharmacological inhibitors. Especially, since PP2A is known to dephosphorylating GSK3β (PMID: 26484916), it might be of interest to investigate whether it is also involved in the insulin/GSK3β/FBXW7/ERRα axis.
As suggested, we have reflected this point in the discussion "Alternatively, the phosphatase(s) involved in ERRα dephosphorylation remains elusive. Identification of ERRα phosphatase is essential and left to be explored. Especially since PP2A is known In this manuscript, Xia et al. discovered an insulin-dependent post-translational modification mechanism for ERRα stability and activity that involves ERRα serine phosphorylation by GSK3β followed by ubiquitination and degradation by FBXW7. This new mechanism is well supported by extensive gain and loss-of-function studies, including in vivo studies using liver GSK3β KO, liver Fbxw7 KO mice and especially the ERRα 3SA mutant mice. Overall, the findings are significant and the manuscript is well written.
We thank the reviewer for carefully reviewing our manuscript and for the helpful suggestions. We have addressed the points raised. Please see our detailed response below.
1. Gender is known an important factor in metabolism. For mouse metabolic studies presented in Fig 5e-5l, Fig 6, Extended Fig 5f-fl and Fig 6, male and female mice should be grouped and compared separately. Gender information should also be clearly stated in the figure legend and methods.
We fully agree with the reviewer that gender is an important factor in metabolism. In this study all the mouse data were derived from 2-to 3-month-old male littermates except for Supplementary Fig. 6, where we confirmed that the hyperglycemia phenotype is conserved in female mice.
We stated it in the "Mice" section of "Method": "Unless otherwise specified, all experiments used age-matched male littermates (2-to 3-month-old)." We also specified the gender of mice used for Supplementary Fig. 6 in the legend: "Fed blood glucose concentrations of 3-month-old female ERRα 3SA and WT littermates on a ND (WT, n = 7; 3SA, n = 6)." We used 20% change to identify insulin-sensitive DEGs mainly because insulin had moderate effects on a large number of genes, consistent with the widely-accepted notion that transcriptional changes of individual metabolic genes are small. This has been recently revealed in a paper published in Cell (PMID: 30955890): "insulin had moderate effects on a large number of genes ( Figure 4I), consistent with previous reports (Cai et al., 2017), and with insulin's role as a homeostatic factor broadly modulating metabolism and cell growth." As shown below Fig. for Reviewer 2, transcriptional changes of most of the insulinstimulated genes are small.
Also, we can significantly validate a 20% change in gene expression via RT-PCR and a collection of 20% changes in multiple genes enriched in the same metabolic pathway is highly significant. Thus, we used [FC] ≥ 1.2 for the analysis of RNA-seq data in Fig. 1 and Fig. 5.
In our study presented in original Fig 1g (Now Fig. 1m), when we increased the [FC] from 1.2 to 2.0, the insulin-regulated genes reduced from 2455 to 170, supporting the importance of using a less stringent cut-off for insulin-sensitive metabolic genes. Among these 170 genes, 125 are dependent on ERRα, thus the percentage of ERRαdependent insulin-regulated genes increased from 67.4% to 73.5%, indicating the importance of ERRα in insulin-mediated gene expression. See below For Fig 2a-2c and Extended Fig 2a-2c, cells were cultured in optimal medium supplemented with 10% Fetal bovine serum, composing growth factors and hormones to sustain proliferation and maintain normal cell metabolism. Thus, this mimics the physiological fed state, neither starved or insulin stimulated. As we can see in Extended  Fig 2c, the insulin signaling pathway is active in this condition, thus manipulating kinase activity itself would affect ERRα stability if the kinase is involved in the regulation of ERRα stability downstream the insulin signaling. Thus, we performed the initial kinase screen in the physiological state without manipulating the presence or absence of insulin.
After identifying that GSK3β inhibition best replicated the effect of insulin on ERRα in vitro and in vivo, we further validate whether ERRα could still be stabilized by insulin in the absence of GSK3β (Original Fig. 2e, f ;Now Fig. 2e, g) or upon blocking the insulin signaling transduction to GSK3β by AKT inhibitors (Original Extended Fig 2d; Now  Supplementary Fig. 2e).
We fully agree that it will be more convincing to add the with / without insulin conditions for Fig. 2c. To rule out the effect of insulin on endogenous GSK3β, we performed this experiment with cells stably expressing either control or shRNA targeting the 3'UTR of GSK3β. As shown below (New Fig. 2i), re-expression of the constitutively active GSK3β S9A but not the kinase defective K85A mutant degraded the accumulated ERRα protein in the GSK3β knock-down cells (compare lanes 9, 11 to lanes 1, 7), and insulin had no further effects on ERRα protein stability once GSK3β could not be phosphorylated by insulin or lose its kinase activity (compare lanes 9, 10, 11, 12 to lanes 7, 8). Together, our results clearly demonstrate that GSK3β kinase activity is required for insulin control of ERRα protein stability.
4. Additional controls will help the readers. For example, ERRα gene expression (mRNA level) should be included for Fig 1a-1f and Extended Fig 1a-1c to establish whether ERRα transcript is induced in the same experimental conditions.
We agree with the reviewer that ERRα mRNA expression data are important controls and have conducted the examinations for the original Fig 1a-1f and Extended Fig 1a-1c, which showed that these conditions stabilize ERRα proteins without affecting its mRNA levels. Please see the New Fig. 1b, d, f, h, j, l and the New Supplementary Fig. 1b, d, f.
5. Some experimental details/conclusions shall be clarified. (a) Fig1c labeled "nuclear extracts" on the top -are nuclear extracts used for ERRα only or also for other proteins? (b) Legend of Extended Data Fig 1b states that "Hepatocytes were treated with 0, 100 nM or 10 ug/ml insulin…." -please use consistent units of insulin of either nM or ug/ml in this and other figures and throughout the manuscript. (c) The authors claim that "insulin-stimulated nuclear ERRα protein was unphosphorylated (Extended Data Fig. 3a)", but the short exposure (SE, supposedly more accurate) phospho-ERRα level seems clearly increased. (d) In Fig 3g, why do 3SA mutations affect GSK3β S9 phosphorylation?
We thank the reviewer for raising these valid points.
(a) We used nuclear extracts for all the proteins detected for the original Fig. 1c (Now  Fig. 1e, see below). To be consistent with other panels in Fig. 1, we re-examined the whole cell extract of the circadian sample in the revised manuscript, we also changed the light / dark phase sample time points into ZT8 and ZT 16 (4 h before and 4 h after the dark cycle, respectively), when serum insulin levels were significantly upregulated caused by increased food intake at night (See below Fig. for Reviewer 5).
(b) We used insulin solution from bovine pancreas (Sigma, Cat. No. I0516), whose recommended concentration for use in cell culture is 5-10 ug/ml. Alternatively, 100 nM (0.58 ug/ml) is another common concentration used for insulin stimulation in cultured cells. Thus, we used a high (10 ug/ml) or a low (0.58 ug/ml) concentration for insulin stimulation, and found that ERRα protein is induced by insulin in a dose-dependent manner. We've revised the insulin units in ug/ml throughout the manuscript.
(c) We apologize for the unexplained abbreviations (SE, LE) and confusing statement. In the original Extended Data Fig.3a (see below), we could not observe an upper band shift (phosphorylated ERRα) with the nuclear extract even using longer exposure (LE), the only appeared lower band that increased by insulin should be the dephosphorylated ERRα, thus we previously stated in the main text that "insulin-stimulated nuclear ERRα protein was unphosphorylated". We optimized this experiment by adding whole cell extract (WCE) and cytoplasmic fraction as controls and properly labeled the bands (New Supplementary Fig. 3a). As shown below, in comparison with whole-cell extract, phos-tag gel examination of the nuclear extract from HepG2 cells could not reveal an upper phosphorylated ERRα band shift, indicating that insulin-induced nuclear ERRα protein was dephosphorylated. We further explained the methods and mechanism for the phos-tag gel in the Method.
(d) This is likely because ERRα transcriptionally represses the insulin signaling pathway as a feedback mechanism. When we separated the enriched pathway for ERRα-bound up-and down-regulated insulin-sensitive DEGs identified in HepG2 cells presented in original Extended Fig 1h (Now Supplementary Fig. 1k, see below), we observed that ERRα-bound insulin down-regulated DEGs are enriched in the PI3K-AKT signaling pathway. Indeed, 3SA mutations affect GSK3β S9 phosphorylation was also observed in vivo. As shown in Fig. 6c  degradation. In the latter part of the manuscript, they generated mice with ERRa3SA, and found that ERRa3SA mice develop insulin resistance, which showed the biological significance of ERRα in vivo.
Overall, the authors have done an excellent analysis, and revealed that ERRα is one of key molecules that control energy homeostasis.
We thank the reviewer for recognizing the biological significance of our study and the constructive comments. We have carefully addressed the points raised in our revised manuscript. Please see our detailed response below.
Specific points: 1. In Extended Fig 1h, 7079 targets of ERRα ChIP-seq are shown, and it seems that these targets are from 20 kb upstream and downstream of genes. Should the targets from 20 kb downstream of genes be included in this analysis because ERRα is a transcriptional factor? Do the 7079 targets include multiple sites from one gene?
The mechanism of transcription regulation is still not fully understood. The notion of binding "upstream" and "downstream" of a transcriptional unit only makes sense when we read the DNA sequence from left to right, but folding of the genome needs to be considered, which might place regulatory elements close in three-dimensional space.
There are many regulatory regions of DNA that don't follow the most common model of transcriptional activation: binding of transcriptional factor just upstream of the transcription start site (TSS) to help position RNA polymerase II at the transcriptional start site of the gene. For example, transcriptional regulation may also be achieved by distal transcription factor binding events at genomic regions termed 'enhancers'; some introns could strongly stimulate mRNA accumulation from several hundred nucleotides downstream of the TSS; transcription factor can also form transcription factor complexes with other transcription factors to affect gene transcription.
In the case of ERRα, the receptor binds to promoters but also has a strong preference for binding in introns. Overall, 48.2% of ERRα peaks (11193 of 23226) are found in introns (our previously published study PMID: 23562079, also see below Fig. for Reviewer 6a). Among ERRα-bound introns, there is a stronger preference for binding to the first intron ( Fig. for Reviewer 6b). Actually, we find more peaks +10 to +20 kb compared to -10 to -20 kb of the TSS (Fig. for Reviewer 6b). This explains why ± 20 kb was chosen.
There are 11777 ERRα liver ChIP-seq peaks within ± 20 kb of a TSS, which are attributed to 7079 unique genes.
We also confirmed that STUB1, PARKIN, and other potential ERRα E3 ligases identified in Fig. 4 (FBXO7, FBXO11, WWP1) regulate ERRα stability independent of its phosphorylation at S19, 22, 26, as their interactions with ERRα were not affected by the phospho-defective 3SA mutations (New Supplementary Fig. 4h). The extent and physiological relevance of their associations with ERRα remain to be investigated.
Which E3 mainly regulates ERRα stability in vivo mainly likely depends on tissue type, upstream nutrients and hormone signals. We have also modified our discussion to reflect these points.
7. There is no FBXW7 immunoblot in Figure 4c.
We thank the reviewer for pointing this out. We've included the FBXW7 immunoblot in Fig. 4c, see below.
As presented in our response to point#2 above, we agree with the reviewer that Alb-Cre/Fbxw7+/f littermate is a better control than Alb-Cre and FBXW7 f/f mice. Unfortunately, we did not keep these mice.
Instead, we followed the paper initially generated the liver-specific FBXW7 knockout mice (PMID: 21123947) and a recent paper phenotyping the FBXW7 LKO mice (PMID: 27238018) and used the Floxed littermates as the controls.

24
In the revised manuscript, we've further compared ERRα protein levels in the FBXW7 LKO and age-matched Alb-Cre mice and observed accumulated ERRα protein in the FBXW7-null liver comparing with the Alb-Cre mice (see below New Supplementary  Fig. 4h). Our data convincingly show that FBXW7 negatively regulates ERRα protein stability in vivo.
9. There are three isoforms in FBXW7. FBXW7α is ubiquitously expressed, and FBXW7γ is expressed in muscle. Which isoform is an ubiquitin ligase of ERRα in muscle (Figure 5b)?
FBXW7α is functionally the most dominant isoform, which is ubiquitously expressed and carries out the most known FBXW7 functions (reviewed in PMID: 25314076). We observed comparable expression levels of FBXW7α mRNA in liver and muscle (New Supplementary Fig. 5g). We also assessed the distribution of FBXW7 isoform mRNAs (New Supplementary Fig. 5h) and confirmed that FBXW7γ is the most expressed isoform in muscle, which is consistent with previous observation (PMID: 16989775).
Despite its high expression level in muscle, FBXW7γ remains in the nucleolus and its function is poorly understood. Also, it is unclear whether ERRα displays nucleolar localization in the muscle or whether other components of the SCF complex (Skp, Cullin, F-box containing complex) are localized in the nucleolus. Thus, FBXW7γ may not act as an active E3 ligase for ERRα.
We have searched in commercial sources but failed to access siRNAs and plasmids targeting specific FBXW7 isoforms. Thus, it is difficult to reveal which FBXW7 isoform actively regulates ERRα stability in the muscle, although It will be interesting to define the specific effects of the two isoforms in this tissue. Various factors including fluctuations in physical activity (sleep, locomotion), food and water intake, drug absorption, and equipment noise could contribute to variations in the time points RER achieving significant difference post C29 injection. Similarly, as observed in Fig. 5k, l, RER varies at the same time points on different days. Thus, to achieve trustable measurements, we usually remain the mice in the metabolic cage for multiple days to minimize the effects caused by behavioral and metabolic alterations.