Role of transcriptional coregulator GRIP1 in the anti-inflammatory actions of glucocorticoids

Yurii Chinenova,1, Rebecca Gupteb,1, Jana Dobrovolnaa, Jamie R. Flammerb, Bill Liua, Francesco E. Michelassic, and Inez Rogatskya,b,2

aHospital for Special Surgery Research Division, New York, NY 10021; and bWeill Cornell Graduate School of Medical Sciences and cWeill Cornell Medical College, New York, NY 10021

Edited* by Keith R. Yamamoto, University of California, San Francisco, CA, and approved June 11, 2012 (received for review April 10, 2012)

Inhibition of cytokine expression by the hormone-activated tivate the hypothalamic–pituitary–adrenal (HPA) axis stimu- glucocorticoid receptor (GR) is the key component of the anti- lating the adrenal cortex to synthesize and release glucocorti- inflammatory actions of glucocorticoids, yet the underlying mo- coids (GCs), which ultimately suppress proinflammatory gene lecular mechanisms remain obscure. Here we report that gluco- expression (5). A disruption of the HPA axis, either physical or corticoid repression of cytokine in primary macrophages pharmacological, decreases the level of circulating GCs and fl is mediated by GR-interacting (GRIP)1, a transcriptional results in hypersensitivity to in ammatory stimuli (6, 7). Due fl coregulator of the p160 family, which is recruited to the p65-occu- to their uniquely potent ability to counter in ammation, for κ over half a century, GCs have remained among the most cost- pied genomic NF B-binding sites in conjunction with liganded GR. fl We created a mouse strain enabling a conditional hematopoietic effective and widely used anti-in ammatory and immunosup- pressive drugs available. cell-restricted deletion of GRIP1 in adult animals. In this model, GCs act through the cytoplasmic glucocorticoid receptor (GR), GRIP1 depletion in macrophages attenuated in a dose-dependent κ a transcription factor of the nuclear receptor (NR) superfamily manner repression of NF B target genes by GR irrespective of the (8), which, upon ligand binding, translocates into the nucleus and upstream Toll-like receptor pathway responsible for their activa- either binds directly to a palindromic DNA sequence, a gluco- tion. Furthermore, genome-wide transcriptome analysis revealed corticoid response element (GRE), composed of two hexameric a broad derepression of lipopolysaccharide (LPS)-induced glucocor- AGAACA half-sites or “tethers” to DNA through protein:pro- ticoid-sensitive targets in GRIP1-depleted macrophages without tein interactions with other DNA-bound factors such as AP1 or IMMUNOLOGY affecting their activation by LPS. Consistently, conditional GRIP1- NFκB (9). Although GR can either activate or repress tran- deficient mice were sensitized, relative to the wild type, to a sys- scription, typically, binding to “conventional” palindromic GREs temic inflammatory challenge developing characteristic signs of leads to activation, whereas GR tethering to AP1 or NFκB usu- LPS-induced shock. Thus, by serving as a GR corepressor, GRIP1 ally attenuates transcription of target genes. Both activation of facilitates the anti-inflammatory effects of glucocorticoids in vivo. anti-inflammatory genes and repression of proinflammatory genes have been implicated in GC-mediated suppression of in- inflammation | macrophage transcriptome | transcriptional regulation | flammation (2, 9, 10). coactivators | corepressors GR elicits transcriptional changes by recruiting multiple cofac- tors that act upon components of basal machinery or chromatin. From over 100 GR-interacting (GRIP), the members of nflammation is a major defense mechanism against infection the p160 family—SRC1/Ncoa1, GRIP1/TIF2/Ncoa2 (hereafter, and injury. Acute inflammation at the site of invasion is initi- I GRIP1), and SRC3/Ncoa3—are considered primary coactivators ated by resident cells of the innate immune system (neutrophils, by serving as binding platforms for additional cofactors with macrophages, and dendritic cells) that recognize certain mole- chromatin-modifying and remodeling activities, e.g., histone ace- cules shared by pathogenic agents via pattern-recognition tyltransferases CBP/p300, methyltransferases CARM1, G9a, Suv4- receptors such as Toll-like receptors (TLRs), Nod-like receptors, 20h1, and a host of secondary adapter proteins (reviewed in ref. or retinoic acid-inducible gene I family members. TLRs recog- 11). All three p160s are recruited to “conventional” palindromic nize pathogenic molecular patterns, such as lipopolysaccharide GREs via their conserved NR interaction domain (NID) in a (LPS), peptidoglycans, single- and double-stranded nucleic acids, ligand-dependent manner and facilitate transcriptional activation. and common bacterial surface proteins (1). TLR-activated sig- GR-mediated “transrepression,” i.e., the antagonism of AP1- naling cascades ultimately converge on a few transcription fac- and NFκB-dependent transcription by the tethered ligand-acti- tors of the NFκB/Rel, AP1, and IRF families (2), which, acting in vated GR has historically been considered the dominant mech- concert, induce the expression of hundreds of genes encoding anism of the anti-inflammatory actions of GCs (9). Indeed, GR cytokines, chemokines, and other proinflammatory mediators. has long been known to directly bind the p65 subunit of NFκB These, in turn, further activate and stimulate the migration of and the Jun subunit of AP1, which correlates with inhibition monocytes, neutrophils, and leukocytes to injured sites, eventu- of NFκB and AP1 reporters and endogenous targets (12, 13). ally leading to pathogen clearance and tissue repair. Although Furthermore, a recent genome-wide analysis of GR and p65 normally a protective response, an excessive production of binding in HeLa cells costimulated with GC and TNFα revealed cytokines may result in local or systemic tissue damage, multiple a substantial number of shared sites (14); yet, the molecular organ failure, and death. Many pathological and autoimmune consequences of these interactions are unknown. conditions, such as rheumatoid arthritis or lupus, are associated with chronic inflammation and are characterized by unique “signatures” of overproduced cytokines (3). Because unabated inflammation imposes an intrinsic danger to Author contributions: Y.C., R.G., J.R.F., and I.R. designed research; Y.C., R.G., J.D., J.R.F., B.L., and F.E.M. performed research; Y.C. and J.D. contributed new reagents/analytic the host, numerous mechanisms have evolved to control the tools; Y.C., R.G., J.R.F., and I.R. analyzed data; and Y.C., R.G., and I.R. wrote the paper. expression and activities of proinflammatory mediators in a spa- fl tial and temporal manner. Locally, inflammatory stimuli induce The authors declare no con ict of interest. the expression of anti-inflammatory cytokines, such as IL10, fac- *This Direct Submission article had a prearranged editor. tors destabilizing cytokine mRNA, and microRNAs targeting 1Y.C. and R.G. contributed equally to this work. various steps in proinflammatory cytokine production and sig- 2To whom correspondence should be addressed. E-mail: [email protected]. naling (4). When local responses fail to contain an in- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. flammatory reaction, cytokines produced by immune cells ac- 1073/pnas.1206059109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1206059109 PNAS Early Edition | 1of6 Downloaded by guest on October 1, 2021 In principle, GR recruitment to AP1/NFκB sites is likely to alter 1924 α β α A 140 IL1 1400 IL1 70 TNF the composition or function of associated transcriptional regula- 120 1200 60 100 1000 50 tory complexes or the activity of RNA polymerase (Pol)2 itself. 80 800 40 For example, in the context of the IL8 and ICAM1 promoters, GR 60 600 30 40 318 400 20 does not interfere with Pol2 recruitment or initiation complex 20 200 10 assembly, but, instead, blocks Pol2 CTD phosphorylation and Inducon Fold 0 0 0 20 185 CCL2 24 CCL3 20 142 CCL4 productive elongation (13). In an alternate scenario, GR may 20 15 16 fi 16 engage a context-speci c corepressor; however, the identity of 12 10 78 12 such a coregulator remains enigmatic. It has been recently shown 8 56 8 5 that GR bound to atypical “negative” GREs recruits the estab- 4 4 lished NR corepressors NCoR and SMRT (15), yet there is no Inducon Fold 0 0 0 40 Niacr1 8 Plat 18 Arl5c evidence of their function at the GR tethering sites and, consis- 16 fl 32 6 14 LPS tently, GR-mediated transrepression of in ammatory cytokine 24 12 Pam3Cys fi 4 10 CL264 genes was unaffected in NCoR-de cient mice (16). Intriguingly, 16 8 6 the recruitment of the p160 cofactor GRIP1, and not other p160s, 8 2 4

Fold Inducon Fold 2 to an AP1 site in the human MMP13 promoter correlated with 0 0 0 Dex -+ -+ -+ -+ -+ -+ -+ -+ -+ glucocorticoid repression (17), perhaps suggesting that GRIP1 LPS Pam3 CL264 LPS Pam3 CL264 LPS Pam3 CL264 may play a role in GC-mediated repression of proinflammatory Cys Cys Cys 350 IL1α 180 IL1β 45 TNFα genes. GRIP1-null mice display multiple metabolic, endocrine, B 300 160 40 Control 140 35 and reproductive defects (18, 19); however, the role of GRIP1 in 250 120 30 LPS 200 100 25 LPS+ Dex the context of the immune system or inflammation is unknown. In 150 80 20 100 60 15 addition, GRIP1 deficiency disrupts the normal development of mRNA elave 40 10 R levels 50 20 5 the adrenal glands leading to the aberrant regulation of the HPA 0 0 0 - CHX - CHX - CHX axis and GC production (20), making these mice poorly suited for Control assessing immune responses in vivo. To elucidate the function of C p=0.016 D p=0.012 E NS F LPS 5.6 18 45 30 p=0.02 LPS+Dex GRIP1 in a relevant cellular environment in pro- and anti-in- p=0.019 40 4.8 15 p=0.019 35 25 NS flammatory settings, we created a mouse strain enabling a condi- 4.0 12 30 p=0.02 20 3.2 25 9 15 tional GRIP1 depletion in myeloid cells in adult mice. Here we 2.4 20 15 1.6 6 10 describe the genome-wide impact of GRIP1 deficiency in macro- 10 old Enrichment 0.8 3 5 5 F over IgG over phages with respect to the inflammatory gene expression program. 0.0 0 0 0 TSS +1000 TSS +1500 TSS +600 TSS +1000 We further provide evidence for the critical role of GRIP1 in IL1α IL1β TNFα CCL4 controlling systemic inflammation in vivo. Fig. 1. GCs inhibit the transcription of TLR-induced genes. (A)Dexinhibits Results TLR-induced expression of inflammatory mediators. BMMΦ were treated with 10 ng/mL LPS, 100 ng/mL Pam3Cys, or 1 μg/mL CL264 ± 100 nM Dex for 1 h. GR Represses TLR-Induced Cytokine Gene Expression at the Level of mRNA levels of proinflammatory cytokines and chemokines were assayed by Transcription. To evaluate the effect of GCs on cytokine gene RT-quantitative PCR (qPCR), normalized to β-actin or GAPDH, and fold in- expression, we treated bone marrow-derived macrophages = Φ duction was calculated relative to the untreated ( 1). Data are represented as (BMM ) from C57B/6 mice with ligands for TLR2 (Pam3Cys), mean ± SEM. (B) Dex suppresses LPS-induced gene expression in the absence TLR4 (LPS), and TLR7 (CL264) in the absence or presence of of new protein synthesis. BMMΦ were pretreated for 30 min with 20 μg/mL a synthetic GC dexamethasone (Dex). As expected, within 1 h, CHX, followed by treatment with 10 ng/mL LPS ± 100 nM Dex, as indicated. treatment with TLR ligands induced the expression of cytokines mRNA levels of target genes were assessed as in A.(C–F) GCs inhibit cytokine IL1α, IL1β, and TNFα and chemokines CCL2, CCL3, and CCL4 gene expression at the level of transcription. BMMΦ were treated with 10 ng/ and other inflammatory mediators (Fig. 1A). Both synthetic mL LPS ± 100 nM Dex for 30 min, and Pol2 occupancy at the TSS and down- (Dex, Fig. 1A) and natural murine (corticosterone, Fig. S1B) stream regions of the indicated genes was assessed by ChIP with anti-Pol2 GCs markedly decreased the expression of all TLR-induced antibody. For each location, qPCR signals were normalized to those at the genes tested. Although the kinetics and magnitude of inhibition control r28S gene and expressed as fold enrichment over normal IgG (= 1). was gene-specific, it was evident as early as 30 min at Dex con- Data are derived from three or more independent experiments and expressed centration as low as 1 nM (Fig. S1A). These results indicate that as mean ± SD. P values were calculated using two-tailed Student’s t-test. GCs inhibit TLR target genes irrespective of the specific TLR responsible for their activation. TSS (Fig. 1E), consistent with the presence of stalled Pol2 in the GC-activated GR may repress TLR target genes either di- α rectly or by inducing the expression of another protein that in TNF promoter-proximal region (22). LPS treatment resulted in turn down-regulates cytokine genes. To discriminate between a robust recruitment of Pol2 to the TSS as well as intragenic Φ regions of IL1α, IL1β, and CCL4 (Fig. 1 C, D, and F). At the these possibilities, we preincubated BMM with the protein α synthesis inhibitors cycloheximide (CHX) or puromycin (Puro) TNF , Pol2 occupancy was strongly induced intragenically, fl followed by treatment with LPS ± Dex and monitored RNA re ecting the appearance of elongating polymerase in the body levels of GC-sensitive inflammatory cytokines. As seen pre- of the gene (Fig. 1E). This LPS-dependent increase in occupancy viously (21), CHX and Puro elevated the baseline expression of was nearly abrogated by Dex, indicating that liganded GR most analyzed genes (Fig. 1B and Fig. S1C); nonetheless, in- antagonizes transcription initiation (for IL1α, IL1β, and CCL4) cubation with LPS resulted in further cytokine gene induction, or elongation (for TNFα). Further corroborating the transcrip- which was attenuated by Dex, demonstrating that GC-mediated tional effects of GR on IL1α, IL1β, TNFα, and CCL4 expression, cytokine down-regulation is direct and does not rely on the prior the dramatic LPS-dependent accumulation of their nascent un- synthesis of a protein intermediate. processed transcripts was abrogated by Dex (Fig. S1D). Both Pol2 recruitment and productive elongation are rate- limiting steps for signal-regulated transcription. To assess the Dex-Mediated Repression of LPS-Induced Genes Involves GR and effect of GCs on Pol2 dynamics, we evaluated the effect of LPS GRIP1 Recruitment to NFκB-Binding Sites. NFκB is a primary tran- and Dex on Pol2 occupancy at the transcription start sites (TSS) scriptional effector of TLR signaling. To evaluate the effect of of cytokine genes and intragenically by chromatin immunopre- GCs on NFκB recruitment to its targets, we first determined p65 cipitation (ChIP). In uninduced MΦ, little Pol2 was detected at occupancy at the NFκB-binding sites in the regulatory regions of IL1α, IL1β, and CCL4 (Fig. 1 C, D, and F, respectively), and IL1α, IL1β, TNFα, and CCL4 (Fig. 2A). Treatment with LPS comparatively high Pol2 occupancy was detected near the TNFα induced a dramatic recruitment of p65 to several NFκB-binding

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1206059109 Chinenov et al. Downloaded by guest on October 1, 2021 sites, including those in the IL1α/β intergenic region and up- in response to LPS + Dex, but not Dex alone (Fig. 2D). Con- stream of the TNFα and CCL4 TSS (Fig. 2B). Coadministration versely, treatment with Dex alone was sufficient to facilitate the of Dex had no effect on p65 occupancy at any of the nine sites enrichment of GRIP1 at the GBS of the GILZ promoter analyzed (Fig. 2B), indicating that the loss of Pol2 and repression (Fig. 2D), which correlated with GR loading (Fig. 2C). Com- of transcription are not due to NFκB displacement. bined, these results indicate that GC-mediated repression of Interactions between NFκB and GR in vivo and in vitro (13) cytokine genes correlates with recruitment of GR and GRIP1 to play a critical role in GC-mediated inhibition of NFκB-driven a set of p65-occupied NFκB-binding sites. transcription. To determine whether GR is recruited to NFκB- binding sites, we surveyed GR occupancy by ChIP. In control Generation of Conditionally GRIP1-Deficient Mice. GRIP1 knock-out experiments, GR was recruited to the previously reported GR- (null) mice, in addition to their fertility phenotype, display binding sites (GBS) at the GILZ promoter following a 1-h pleiotropic changes in metabolism and disrupted adrenal archi- treatment with Dex or LPS + Dex (Fig. 2C, Right). Furthermore, tecture and function with compensatory changes in the HPA axis Dex + LPS cotreatment, but not Dex alone, also triggered (20), which complicates the assessment of GC signaling in vivo. a significant GR loading onto a subset of NFκB-binding sites in To overcome these issues, we created a mouse strain enabling an BMMΦ (Fig. 2C), suggesting that NFκB activation and p65 inducible deletion of GRIP1 in adult animals preferentially B in hematopoietic cells including MΦ.TheGRIP1-WTmice binding (Fig. 2 ) are necessary for the recruitment of GR. − − Although the mechanisms of GR-mediated repression at ( / :GRIP1FL/FL) were bred to Mx1Cre mice (Mx1Cre-WT) tethering sites are unknown, the p160 cofactor GRIP1 was shown expressing polyinosinic:polycytidylic acid (pIC)-inducible Cre to be recruited to the AP1 site of the repressed human MMP13 recombinase to ultimately generate GRIP1-KD, GRIP1-HET, gene in osteosarcoma cells (17). To determine whether GRIP1 and GRIP1-WT mice (see SI Materials and Methods and Fig. S2A was a component of GR repression complexes at tethering GBS for details). This protocol resulted in a deletion of a floxed allele of proinflammatory genes in MΦ, we assessed GRIP1 occupancy and a 95% depletion of GRIP1 transcript and protein in BMMΦ at the NFκB-binding sites in our candidate cytokine genes. In from GRIP1-KD mice as determined by RT-quantitative PCR parallel with GR, GRIP1 was recruited to the above NFκB sites (qPCR) and Western blotting, respectively, whereas the expres- sion of GR (Fig. 3A) or the other two p160 family members (Fig. S2B) was unchanged. Notably, pIC injections 2 wk before bone marrow isolation did not trigger nonspecific long-term changes in 129,140,000 129,160,000 129,180,000 129,200,000 the expression or chromatin of TLR-regulated genes (Fig. S2 C A IL1α IL1β Chr2 and D). In addition, “acute” conditional GRIP1 depletion did IMMUNOLOGY -290 -2382 not affect MΦ differentiation based on the expression of the -10K -21K -1728 -10K surface markers CD11b and F4/80 in GRIP1-deficient vs. GRIP1-sufficient BMMΦ (Fig. S2E). 33,335,000 33,337,000 33,339,000 33,341,000 33,343,000 33,345,000 TNFα LTA Chr17 GRIP1 Deletion Attenuates GR-Mediated Repression of LPS-Induced -500 Genes in BMMΦ. To determine the effect of GRIP1 disruption on -200 GC-mediated repression, we treated WT and GRIP1-KD BMMΦ with LPS ± Dex and determined the mRNA levels of ChIP, p65 fl B 15 IL1α 8 IL1β 20 IL1β 14 TNFα 4 CCL4 proin ammatory cytokines. As expected, LPS potently induced 12 Φ 6 16 3 cytokine RNAs in both M populations (Fig. S2F). Strikingly, 10 10 12 8 α β α 4 2 Dex-mediated repression of IL1 , IL1 , TNF , and CCL4 was 8 6 5 significantly attenuated in GRIP1-KD BMMΦ (Fig. 3B). The 2 4 1 4 2 effect of GRIP1 depletion was dose-dependent, as an in- over IgG over Fold Enrichment Fold 0 0 0 0 0 -1728 -10K -20K -290 -2382 -10K -500 -200 TSS termediate level of repression was observed in the GRIP1 haplo- C ChIP, GR * insufficient GRIP1-HET BMMΦ (Fig. 3C). Furthermore, IL1α * IL1β TNFα GILZ 4 4 4 * 20 * GRIP1 deletion similarly reversed GC repression of cytokine 3 3 3 16 * genes induced by ligands to TLR2 and TLR7 (Fig. 3D), sug- 12 2 2 2 er IgG 8 gesting that GR-mediated repression is GRIP1-dependent re- 1 1 1 4 gardless of the TLR pathway responsible for gene activation. ov Fold Enrichment Fold 0 0 0 0 To determine the genome-wide effect of GRIP1 deletion on the D -1728 -10K -20K -290 -2382 -10K -200 GRE fl Φ 2 ChIP, GRIP1 in ammatory gene expression program, we evaluated the M Φ 1.6 transcriptome by RNA-Seq in WT and GRIP1-KD BMM .The 1.2 Control Illumina cDNA libraries were prepared from BMMΦ of both Dex er IgG 0.8 LPS genotypes treated with Dex, LPS, or both for 1 h and untreated 0.4 LPS+Dex control and sequenced. Of 349 genes induced by LPS >2-fold in Fold Enrichment Fold ov 0.0 Φ ≥ -1728 -10K -20K -2382 -10K GRE WT and KD BMM , 152 were repressed 1.4-fold in the WT IL1α IL1β GILZ BMMΦ (Fig. 4A). We ranked Dex-repressed genes in the WT according to their “fold repression” and evaluated the corre- Fig. 2. Transcription complexes at the NFκB-binding sites at activated and sponding levels of repression in the KD. Strikingly, over 60 LPS- repressed cytokine genes. (A)IL1α/β and TNFα loci maps showing the NFκB- induced Dex-repressed genes were derepressed by >15% in GRIP1 binding peaks derived from the p65 ChIP-Seq data set (23). The raw read data KD relative to WT BMMΦ (Fig. 4B; Fig. S3A; Table S1), sug- were mapped to the mouse genome and visualized using CLC Genomics gesting an unexpectedly broad role of GRIP1 in GR-mediated re- κ Workbench 4.8. The center positions of NF B sites that we validated by ChIP pression. term enrichment analysis of derepressed with p65 antibodies are indicated. Binding sites are shown as red, blue, and genes performed with G:Profile (24) revealed a high prevalence of green rectangles under the read distributions for NFκB, GR, and GRIP1, re- fl spectively. (B) Dex does not affect LPS-dependent p65 recruitment to cytokine terms related to regulation of immune and in ammatory responses, genes. BMMΦ were treated with 10 ng/mL LPS ± 100 nM Dex for 1 h, and ChIPs cytokine production, and cell death (Table S2). Interestingly, ± a computational analysis of the 1,000-bp region upstream of the for p65 were performed as in Fig. 1C. Error bars represent mean SD for three fi or more independent experiments. (C and D) GR and GRIP1 are recruited to TSS of these genes showed a signi cant enrichment of the binding p65-binding sites in LPS + Dex-dependent manner. ChIPs for GR (C) and GRIP1 sites for the members of the NFκBandAP1families,butnotofthe (D)wereperformedinBMMΦ treated as indicated. Amplification of the GILZ binding sites for GR or other NRs (Table S3). GRE is used as a positive control for Dex-dependent GR and GRIP1 recruitment. Because we defined fold repression as a ratio of expression in Fold enrichments over normal IgG are calculated as in Fig. 1C (*P < 0.05). the presence of LPS over LPS + Dex, in principle, it could be

Chinenov et al. PNAS Early Edition | 3of6 Downloaded by guest on October 1, 2021 8 A 100 WT GRIP1 protein B 11 IL1α 7 IL1β 5 TNFα CCL4 C 4 TNFα D 7 TNFα 46 IL1β P<0.05 6 P<0.01 P<0.05 7 P<0.05 75 KD WT KD 9 37 WT 4 6 5 3 ession 5 KD GRIP1 7 5 28 50 non-spec 4 3 4 % of WT 5 19 25 3 2 3 GR 2 3 3 2 10 Fold Repression Fold Fold Repression Fold 0 tubulin 2 Repr Fold GRIP1 1 1 1 1 1 1 1 mRNA WT KD WT KD WT KD WT KD GRIP1 100 49 9 LPS Pam3 CL264 LPS Pam3 CL264 % WT WTHET KD Cys Cys

Fig. 3. GR represses TLR-induced genes in a GRIP1-dependent manner. (A) GRIP1 expression in the BMMΦ of GRIP1-KD mice. GRIP1 mRNA in WT and KD BMMΦ was measured by RT-qPCR, normalized to β-actin, and expressed relative to WT (= 100%). Error bars show mean ± SD; n = 6. GRIP1 and GR protein levels in WT and KD BMMΦ were assessed by immunoblotting with tubulin as a loading control. (B) GRIP1 depletion attenuates GR-mediated repression of TLR-induced genes. BMMΦ from WT and GRIP1-KD littermate mice were treated as in Fig. 2B. The relative amounts of indicated transcripts were measured and expressed as ratios of LPS/LPS + Dex (“fold repression”) for each genotype. Average fold repression in WT vs. KD was compared using a two-tailed Student’s t-test (nWT = 6, nKD = 7). (C) GRIP1 facilitates repression in a dose-dependent manner. BMMΦ from WT (WT/WT), HET (WT/FL), and GRIP1-KD (FL/FL) mice were treated as in B,andtheTNFα relative RNA amount was measured and expressed as fold repression for each genotype. The corresponding GRIP1 RNA levels are shown as percentage of WT (= 100). (D) GRIP1 mediates GR-dependent repression of cytokine genes irrespective of the TLR pathway responsible for their activation. WT and GRIP1-KD BMMΦ were treated with LPS, Pam3Cys, or CL264 ± Dex,andfoldrepressionofTNFα and IL1β RNA was derived as in B.

skewed due to a hypothetical global effect of GRIP1 deletion on GRIP1 Is Protective Against Systemic Inflammatory Responses in Vivo. the induction of various genes by LPS. To rule out this potential The production of endogenous GCs is an important negative variable, we compared fold LPS induction of the 349 genes ac- feedback mechanism that is initiated in response to inflammatory tivated by LPS in WT BMMΦ to corresponding values in the stimuli and inhibits a heightened inflammatory response by KD. Our analysis detected no statistically significant difference limiting the transcription of cytokines and other proin- in the transcriptome-wide LPS induction between WT and KD flammatory mediators. Because GRIP1 depletion in MΦ im- cells (Fig. 4C). Because GRIP1 serves as a GR coactivator in the paired GR-mediated repression of cytokine genes (Fig. 3B; Fig. context of palindromic GREs, we assessed whether GRIP1 de- 4B; Fig. S3), we speculated that GRIP1 may be involved in the pletion also attenuates the induction of the GR transcriptional HPA-dependent anti-inflammatory feedback in vivo, thereby targets with anti-inflammatory activities: DUSP1, IκB, and conferring protection against an exaggerated inflammatory re- GILZ. We found no statistically significant difference in the sponse. LPS-induced toxic shock is commonly used as a model of fl fi expression of these genes between WT vs. KD BMMΦ in two unabridged in ammation. We rst determined that 20 mg/kg of independent RNA-Seq experiments (Fig. S3B), arguing against LPS administered i.p. is the maximum sublethal dose tolerated their contribution to the observed attenuation of glucocorticoid by WT controls. We then injected age-matched WT and GRIP- repression. This finding is consistent with the lack of an effect of KD mice with LPS and monitored them for 96 h for hallmarks of the protein synthesis inhibitors on GR-mediated repression septic shock such as changes in temperature and weight, loco- (Fig. 1B and Fig. S1C). motor activity, and shivering; mice found in a moribund state were euthanized as per the National Institutes of Health and Institutional Animal Care and Use Committee protocols. As shown in Fig. 5A, the mortality rate of GRIP1-KD animals was A B 6 significantly higher (P < 0.05), which correlated with a greater 5.5 WT WT, LPS>2 KD, LPS>2 loss of body weight than that of their WT counterparts (Fig. 5B). 5 IL1α KD IL1β WT=1.885~ These observations suggested that GRIP1-depleted mice may 4.5 ~ 197 MMP13 KD =1.455 suffer from a “cytokine storm” due to failure to control the 29 39 4 n=64, U=3027 Mann-Whitney,p<0.0001 production of proinflammatory cytokines. Indeed, the levels 152 3.5 Niacr1 IFNβ1 α β γ fi 3 of serum TNF ,IL1,andIFN were signi cantly higher CCL5 < 128 Repression Fold 2.5 Cxcl10 (P 0.05)inGRIP1-KDcomparedwiththeWTmice CCL3 CCL2 WT, LPS+Dex >1.4 2 CCL4 (Fig. 5C), suggesting that GRIP1 depletion results in the de- C 1.5 regulation of endogenous responses curbing inflammation. 10 1 9 WT KD Discussion 8 n=349, U=61956.5 7 Mann-Whitney,p= 0.691688 The immune system is under constant pressure from a variety 6 of environmental antigens in a state aptly termed “chronic 5 ”

old Inducon) immunological stress (25). Indeed, from an evolutionary

(F 4 2 perspective, the proper function of the immune system log 3 requires a trade-off between hypo-activation that leads to 2 1 rampant multiplication and spread of pathogens and hyper- activation that may result in immunopathologies and a break- Fig. 4. GRIP1 deletion globally attenuates GR-mediated repression of LPS- down of homeostasis. This delicate equilibrium is maintained induced genes in BMMΦ.(A) Transcriptome analysis of WT and GRIP1-KD by a number of highly redundant mechanisms that gradually BMMΦ treated for 1 h with LPS or LPS + Dex. Venn diagram shows gene escalate immune responses commensurate with the magnitude overlap between treatments and genotypes. (B) A subset of Dex-repressed of injurious signal. The “decision” to proceed to a more vig- Φ genes is derepressed in GRIP1-KD BMM .Geneexpressionlevelswerede- orous response often hinges on a balance between pro- and termined by RNA-Seq, normalized to gene length, and expressed as reads per anti-inflammatory cytokines and chemokines. Tipping this kilobase of exon per million mapped reads (SI Materials and Methods). Fold balance toward proinflammatory mediators triggers a systemic repression by Dex was determined as a ratio of expression levels in LPS− vs. LPS + Dex-treated BMMΦ. LPS-induced (more than twofold) genes repressed response that, among others, increases the level of serum GCs, which, in turn, suppresses the expression of proinflammatory by Dex in WT were ranked by the fold repression, and the corresponding ratios fl for 64 genes derepressed in a KD are plotted. The median fold repression for cytokines, thereby abating in ammation. each population and the results of the Mann–Whitney comparison are shown The effect of GR on gene expression is remarkably broad. in the Inset.(C) GRIP1 depletion does not affect gene induction by LPS. Gene Even a brief exposure of BMMΦ to GCs results in the inhibition expression levels were determined by RNA-Seq, and LPS-induced (more than of over 100 LPS-induced genes. Others have shown that, fol- twofold) genes in WT and GRIP1-KD BMMΦ were compared as in B. lowing a prolonged treatment, up to 30% of total expressed

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1206059109 Chinenov et al. Downloaded by guest on October 1, 2021 The study of GC signaling in the immune system of GRIP1-null A LPS, 20 mg/kg B 90 p<0.05 C 35 p=0.035 8 p=0.02 14 p=0.02 1 fi GRIP1 WT n=5 30 7 12 animals is further complicated by their adrenal insuf ciency and 86 0.8 25 6 10 decreased level of corticosterone in the blood (20). Our condi- e LPS 5 0.6 82 20 8 4 tional GRIP1-KD mice enabled a comprehensive evaluation of p=0.006 15 6 Φ 0.4 n=6 78 3 the GRIP1 function in shaping the M transcriptome and its 10 2 4 74 pg/ml x 100 fl Survival rate 0.2 2 physiological implications for the control of systemic in am- GRIP1 KD 5 1 0 % of pr Weight 70 0 0 0 mation. This genome-wide approach revealed an unexpectedly 0 1224364860728496 WT KD WT KD WT KD WT KD Time, h TNFα IL1β IFNγ broad role of GRIP1 in GR-mediated repression of proin- flammatory genes without appreciably affecting their induction by Fig. 5. GRIP1 plays a protective role in a systemic inflammatory response LPS. Indeed, derepression of many Dex-sensitive TLR target in vivo. (A and B) GRIP1 conditionally depleted mice are sensitized to septic genes in GRIP1-KD was seen not only upon LPS stimulation, but shock. GRIP1-WT and GRIP1-KD mice (n = 6 for each genotype) were injected also at the level of basal expression (not shown). This result is not i.p. with 20 mg/kg LPS and monitored for the indicated time. Survival is unexpected, given that many AP1 sites are constitutively occupied shown via the Kaplan–Meier plot (A). Body weight of WT and GRIP1-KD mice by cJun, which may provide a recruitment site for GR. Consis- 36 h post injection was expressed as a percentage of pre-LPS weight and tently, the regulatory regions of the derepressed genes were compared using the Mann–Whitney test (B). (C) GRIP1 deletion results in devoid of GREs but enriched for AP1- and NFκB-binding sites. increased serum cytokine levels in response to systemic LPS challenge. GRIP1- Taken together, our results identify GRIP1 as a GR corepressor WT and GRIP1-KD mice were injected with LPS, and their serum cytokines fl Φ α β γ = of proin ammatory mediators in BMM ; however, the mecha- were measured at 1.5 h (TNF ) or 5 h (IL1 and IFN ). P values (n 5or6) nistic basis of corepression remains to be determined. The GRIP1 were calculated using two-tailed Student’s t-test. corepressor activity, at least in vitro, was localized to a unique repression domain (RD) encompassing GRIP1 amino acids 767– genes become affected (26, 27), making the mechanistic analysis 1006, which is not shared by other p160 family members (32) and contains no predicted structural motifs or enzymatic activities. of GR action a daunting task. Indeed, although the suppressive Instead, we discovered that RD participates in protein:protein effect of GCs on cytokine transcription has been studied for al- interactions with several transcriptional regulators including most 3 decades, no unifying mechanism has been proposed, and, members of the IRF family and the histone methyltransferase in all likelihood, numerous pathways are engaged to suppress Suv4-20h1 (33–35). Conceivably, the RD-interacting secondary excessive inflammatory signaling. It is conceivable that the GR corepressor mediating GRIP1 actions at tethering elements mechanisms of repression are at least in part determined by the awaits identification. Unexpectedly, much like the effects of GR IMMUNOLOGY events responsible for cytokine gene induction in response to fi itself, the loss of GRIP1 affected both stalled and initiation-con- TLR ligands. Recent studies broadly classi ed TLR-induced trolled genes, perhaps suggesting that, in conjunction with GR, genes into immediate-early (IE) genes, which are activated by GRIP1 has as-yet-unknown functions in elongation control. the resident transcription factors (NFκB, IRFs, and AP1) within “ ” GRIP1 was originally described as an NR coactivator (36, 37); minutes of stimulation and delayed genes, whose activation therefore, in principle it was possible that GRIP1 deletion would requires protein synthesis and amplification of the initial signal fl – also attenuate GR-dependent activation of anti-in ammatory by the IE cytokines (28 30). Furthermore, IE vs. delayed genes genes, such as GILZ, DUSP1, or IκB. Indeed, both GILZ and are often activated at different stages of the transcription cycle DUSP1 contribute to the inhibition of inflammation in various (22, 28). Specifically, IE genes (e.g., TNFα) display features that fi “ ” systems (10). Interestingly, we detected no signi cant difference are characteristic of stalled genes activated by Pol2 release in GILZ, DUSP1, and IκB RNA levels upon a 1-h induction by from the early elongation block: extensive histone acetylation, Dex between the WT and KD BMMΦ in the presence or ab- hypomethylation, and a high level of Pol2 near the TSS in sence of LPS, suggesting either that the expression of these genes uninduced state. Conversely, the delayed genes commonly lack does not rely on GRIP1 or that other members of the p160 detectable Pol2 at the TSS before stimulation and are induced by family expressed in BMMΦ are sufficient. signal-dependent Pol2 recruitment and transcription initiation. Recent analyses of transcriptional regulatory networks revealed Surprisingly, we found that, in both cases, GR activation that a large number of DNA-binding regulators rely on relatively imparted a dramatic, protein synthesis-independent repression, few cofactors to coordinate gene expression in response to a vari- as well as the chromatin marks typical of unstimulated cells. ety of environmental stimuli. Indeed, GRIP1 has a wide-ranging What are the molecular targets of GR for each class of genes? specificity of protein:protein interactions, including those with Physical interactions between NFκB or AP1 family members and MEF2C, MyoD, and IRFs (33, 35, 38, 39), regulating processes as GR have long been implicated in the direct inhibition of proin- diverse as muscle-cell differentiation and innate immune response flammatory gene transcription. In MΦ, we observed GR re- to double-stranded RNA. What promotes the preferential tar- cruitment to several NFκB sites in the regulatory regions of geting of GRIP1 to a specific regulatory complex or its ability to TNFα, IL1α, and IL1β that was contingent upon LPS activation inhibit or activate transcription is unknown, but recently discov- and occurred concomitantly with p65 recruitment. Interestingly, ered GRIP1 posttranslational modifications such as SUMOylation a recent genome-wide analysis of GR and p65 binding in HeLa and phosphorylation (40, 41) may serve as such a mechanism. In cells cotreated with the GC triamcinolone acetonide and TNFα addition, GRIP1 expression itself is potentially amenable to reg- revealed 1,033 GR binding peaks (12% of the total GR cis- ulation: it was found to be overexpressed or fused to other regu- trome), more than half of which were also co-occupied by p65 lators, including MOZ, HEY1, and ETV6 in a number of cancers (14). Consistent with our findings, some of these sites are asso- (refs. 42 and 43 and references therein). Here, we established that ciated with genes encoding proinflammatory mediators whose GRIP1-KD mice are hypersensitive to LPS due to a failure to transcription is inhibited by GCs. repress the IE proinflammatory cytokines (e.g., TNFα,IL1α, The potential involvement of GRIP1 in GR-mediated repres- and IL1β) and a secondary amplification of cytokines such as sion was suggested by an earlier observation of its recruitment to IFNγ, which are not direct targets for GR:GRIP1. Future studies the AP1:GR tethering complex at the MMP13 promoter in the will reveal whether specific changes in GRIP1 modifications or U2OS-GR stable cell line (17). However, the physiological rele- expression occur in pathological conditions characterized by vance of this finding to the inflammatory gene expression program chronic inflammation when the balance of pro- and anti-in- or regulation of inflammation in vivo was never assessed. Among flammatory signaling is disrupted. Intriguingly, a survey of GRIP1 the various phenotypes of GRIP1-deficient mice, those in re- expression in autoimmune diseases using the NextBio database production and energy metabolism were most extensively de- (www.nextbio.com) revealed that in synovial fibroblasts from scribed, but reflect primarily the multifaceted functions of GRIP1 patients with rheumatoid arthritis GRIP1 was down-regulated as a cofactor for other NRs, such as PPARγ and RORα (18, 31). by inflammatory stimulation; GRIP1 mRNA levels were also

Chinenov et al. PNAS Early Edition | 5of6 Downloaded by guest on October 1, 2021 reduced in peripheral monocytes of patients with lupus and in Survival Analysis. GRIP1-WT and KD mice were injected i.p. with 20 mg/kg of lupus models. Thus, a dysregulation of GRIP1 may provide the LPS. Mice were weighed before and at 36 h post injection, and the signifi- cance of difference between WT and KD groups was determined using the molecular basis for the instances of clinical GC resistance when ’ — two-tailed Student s t-test. The recovery of mice following LPS injection was the function of GR itself appears grossly normal an attractive monitored for up to 96 h. The survival curves of the two groups were hypothesis that remains to be tested. evaluated by Kaplan–Meyer analysis, and the P value was determined using a log-rank test. Materials and Methods Cell Culture and Reagents. BMMΦ were prepared from 8- to 12-wk-old mice Cytokine Measurement. Mouse inflammation and IL1β Flex Cytometric Bead as described (36). Dex, corticosterone, LPS, pIC, and CHX were purchased Array kits (BD Biosciences) were used for detecting TNFα,IFNγ, and IL1β,and from Sigma. Pam3Cys, CL264, and Puro were purchased from Invivogen. the data were analyzed with FACScan.

RNA Isolation, Real-Time PCR, and RNA-Seq. Total RNA isolation, random- ACKNOWLEDGMENTS. We thank P. Chambon (Institut de Génétique et de FL/FL primed cDNA synthesis, qPCR with ROX containing SYBR-green master mix Biologie Moléculaire et Cellulaire) for the kind gift of TIF2/GRIP1 mice and the staff of the Hospital for Special Surgery Transgenic Animal (Fermentas), and δδCt analysis were performed as described (34). Primers are Facility, Flow Cytometry Core, and the Weill Cornell Genomic Resources Core listed in Table S4. For RNA-Seq, total RNA was isolated from BMMΦ from Facility. We also thank P. D. A. Issuree for help with FACS analyses; two independent pairs of GRIP1-WT and KD littermates and subjected to M. Kennedy for technical assistance; J. Zavadil (New York University) and RNA-Seq as described in SI Materials and Methods. O. Elemento (Weill Cornell) for help with RNA-Seq data analysis; and S. Durum (National Institutes of Health), L. Ivashkiv (Hospital for Special Surgery), and M. Shtutman (South Carolina College of Pharmacy) for insight- Immunoblotting and ChIP. See SI Materials and Methods. ful discussion. This work was supported by grants from the National Insti- tutes of Health and the Kirkland Center (to I.R.) and from American Heart Transgenic Mice. See SI Materials and Methods. Association Grant 11SDG5160006 (to Y.C.).

1. Moresco EM, LaVine D, Beutler B (2011) Toll-like receptors. Curr Biol 21:R488--R493. 24. Reimand J, Arak T, Vilo J (2011) g:Profiler–a web server for functional interpretation 2. Chinenov Y, Rogatsky I (2007) Glucocorticoids and the innate immune system: of gene lists (2011 update). Nucleic Acids Res 39:W307–W315. Crosstalk with the toll-like receptor signaling network. Mol Cell Endocrinol 275(1-2): 25. Lochmille RL, Deerenber C (2000) Trade-offs in evolutionary immunology: Just what is 30–42. the cost of immunity? Oikos 88:87–98. 3. Palucka AK, Blanck JP, Bennett L, Pascual V, Banchereau J (2005) Cross-regulation of 26. Chrousos GP (2010) Stress and sex versus immunity and inflammation. Sci Signal TNF and IFN-alpha in autoimmune diseases. Proc Natl Acad Sci USA 102:3372–3377. 3:pe36. 4. Medzhitov R, Horng T (2009) Transcriptional control of the inflammatory response. 27. Maranville JC, et al. (2011) Interactions between glucocorticoid treatment and cis- – Nat Rev Immunol 9:692 703. regulatory polymorphisms contribute to cellular response phenotypes. PLoS Genet 5. Sternberg EM (2006) Neural regulation of innate immunity: A coordinated non- 7:e1002162. fi – speci c host response to pathogens. Nat Rev Immunol 6:318 328. 28. Escoubet-Lozach L, et al. (2011) Mechanisms establishing TLR4-responsive activation 6. Sternberg EM, et al. (1989) Inflammatory mediator-induced hypothalamic-pituitary- states of inflammatory response genes. PLoS Genet 7:e1002401. adrenal axis activation is defective in streptococcal cell wall arthritis-susceptible Lewis 29. Ramirez-Carrozzi VR, et al. (2006) Selective and antagonistic functions of SWI/SNF and rats. Proc Natl Acad Sci USA 86:2374–2378. Mi-2beta nucleosome remodeling complexes during an inflammatory response. 7. Gómez SA, et al. (2003) Endogenous glucocorticoids attenuate Shiga toxin-2-induced Genes Dev 20:282–296. toxicity in a mouse model of haemolytic uraemic syndrome. Clin Exp Immunol 131: 30. Takeuchi O, Akira S (2010) Pattern recognition receptors and inflammation. Cell 217–224. 140:805–820. 8. Gronemeyer H, Gustafsson JA, Laudet V (2004) Principles for modulation of the nu- 31. Picard F, et al. (2002) SRC-1 and TIF2 control energy balance between white and clear receptor superfamily. Nat Rev Drug Discov 3:950–964. – 9. Glass CK, Saijo K (2010) Nuclear receptor transrepression pathways that regulate in- brown adipose tissues. Cell 111:931 941. flammation in macrophages and T cells. Nat Rev Immunol 10:365–376. 32. Rogatsky I, Luecke HF, Leitman DC, Yamamoto KR (2002) Alternate surfaces of 10. Clark AR, Belvisi MG (2012) Maps and legends: The quest for dissociated ligands of the transcriptional coregulator GRIP1 function in different glucocorticoid receptor acti- – glucocorticoid receptor. Pharmacol Ther 134:54–67. vation and repression contexts. Proc Natl Acad Sci USA 99:16701 16706. 11. York B, O’Malley BW (2010) Steroid receptor coactivator (SRC) family: Masters of 33. Reily MM, Pantoja C, Hu X, Chinenov Y, Rogatsky I (2006) The GRIP1:IRF3 interaction as systems biology. J Biol Chem 285:38743–38750. a target for glucocorticoid receptor-mediated immunosuppression. EMBO J 25:108–117. 12. Yang-Yen H-F, et al. (1990) Transcriptional interference between c-Jun and the glu- 34. Chinenov Y, Sacta MA, Cruz AR, Rogatsky I (2008) GRIP1-associated SET-domain cocorticoid receptor: Mutual inhibition of DNA binding due to direct protein-protein methyltransferase in glucocorticoid receptor target gene expression. Proc Natl Acad interaction. Cell 62:1205–1215. Sci USA 105:20185–20190. 13. Nissen RM, Yamamoto KR (2000) The glucocorticoid receptor inhibits NFkappaB by 35. Flammer JR, et al. (2010) The type I interferon signaling pathway is a target for interfering with serine-2 phosphorylation of the RNA polymerase II carboxy-terminal glucocorticoid inhibition. Mol Cell Biol 30:4564–4574. domain. Genes Dev 14:2314–2329. 36. Voegel JJ, Heine MJ, Zechel C, Chambon P, Gronemeyer H (1996) TIF2, a 160 kDa 14. Rao NA, et al. (2011) Coactivation of GR and NFKB alters the repertoire of their transcriptional mediator for the ligand-dependent activation function AF-2 of nuclear binding sites and target genes. Genome Res 21:1404–1416. receptors. EMBO J 15:3667–3675. 15. Surjit M, et al. (2011) Widespread negative response elements mediate direct re- 37. Hong H, Kohli K, Garabedian MJ, Stallcup MR (1997) GRIP1, a transcriptional co- – pression by agonist-liganded glucocorticoid receptor. Cell 145:224 241. activator for the AF-2 transactivation domain of steroid, thyroid, retinoid, and vita- 16. Ogawa S, et al. (2005) Molecular determinants of crosstalk between nuclear receptors min D receptors. Mol Cell Biol 17:2735–2744. – and toll-like receptors. Cell 122:707 721. 38. Chen SL, Dowhan DH, Hosking BM, Muscat GE (2000) The steroid receptor coactivator, 17. Rogatsky I, Zarember KA, Yamamoto KR (2001) Factor recruitment and TIF2/GRIP1 GRIP-1, is necessary for MEF-2C-dependent gene expression and skeletal muscle dif- corepressor activity at a collagenase-3 response element that mediates regulation by ferentiation. Genes Dev 14:1209–1228. phorbol esters and hormones. EMBO J 20:6071–6083. 39. Wu HY, et al. (2005) Nuclear hormone receptor coregulator GRIP1 suppresses, 18. Chopra AR, et al. (2008) Absence of the SRC-2 coactivator results in a glycogenopathy whereas SRC1A and p/CIP coactivate, by domain-specific binding of MyoD. J Biol resembling Von Gierke’s disease. Science 322:1395–1399. Chem 280:3129–3137. 19. Gehin M, et al. (2002) The function of TIF2/GRIP1 in mouse reproduction is distinct 40. Kotaja N, Karvonen U, Jänne OA, Palvimo JJ (2002) The nuclear receptor interaction from those of SRC-1 and p/CIP. Mol Cell Biol 22:5923–5937. domain of GRIP1 is modulated by covalent attachment of SUMO-1. J Biol Chem 20. Patchev AV, et al. (2007) Insidious adrenocortical insufficiency underlies neuroendo- – crine dysregulation in TIF-2 deficient mice. FASEB J 21:231–238. 277:30283 30288. 21. Croons V, Martinet W, Herman AG, De Meyer GR (2008) Differential effect of the 41. Dobrovolna J, Chinenov Y, Kennedy MA, Liu B, Rogatsky I (2012) Glucocorticoid- protein synthesis inhibitors puromycin and cycloheximide on vascular smooth muscle dependent phosphorylation of the transcriptional coregulator GRIP1. Mol Cell Biol cell viability. J Pharmacol Exp Ther 325:824–832. 32:730–739. 22. Adelman K, et al. (2009) Immediate mediators of the inflammatory response are 42. Taylor BS, et al. (2010) Integrative genomic profiling of human prostate cancer. poised for gene activation through RNA polymerase II stalling. Proc Natl Acad Sci USA Cancer Cell 18:11–22. 106:18207–18212. 43. Wang L, et al. (2012) Identification of a novel, recurrent HEY1-NCOA2 fusion in 23. Barish GD, et al. (2010) Bcl-6 and NF-kappaB cistromes mediate opposing regulation mesenchymal chondrosarcoma based on a genome-wide screen of exon-level ex- of the innate immune response. Genes Dev 24:2760–2765. pression data. Genes Cancer 51:127–139.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1206059109 Chinenov et al. Downloaded by guest on October 1, 2021