Extensive enhancer crosstalk controls PPARG2 activation during adipogenesis

Extensive enhancer crosstalk controls PPARG2 activation during adipogenesis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Extensive enhancer crosstalk controls PPARG2 activation during adipogenesis Anna Cetnarowska, Mette Hyldahl, Marcus Nygård, Hesam Dashti, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6466826/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 15 Mar, 2026 Read the published version in Nature Communications → Version 1 posted You are reading this latest preprint version Abstract Peroxisome proliferator-activated receptor γ (PPARγ) is the master regulator of adipogenesis, but the mechanisms underlying its strong induction in response to adipogenic cues are poorly understood. Using human mesenchymal stem cells as a model system, we show that the PPARG locus is primed for activation in the progenitor state by a highly connected enhancer community, which gets further activated and interconnected during adipogenesis. By systematically deleting individual enhancers in the community and interrogating the effect on enhancer function, connectivity and gene expression, we reveal important, non-redundant and cooperative roles of many enhancers. We show that the promoter-proximal enhancers and the downstream super-enhancer constituents cooperate in cis at early time points of differentiation, whereas at later timepoints PPARγ feedback-activates its own expression. Top-scoring non-coding cardiometabolic variants predicted to affect PPARG2 expression map to key enhancers in the community, indicating that regulation via these enhancers is important for human physiology. Biological sciences/Genetics/Gene regulation Biological sciences/Genetics/Functional genomics Biological sciences/Genetics/Genetic association study/Genome-wide association studies Biological sciences/Molecular biology/Epigenetics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Adipocytes are highly specialized cells that play an important role in metabolic energy storage and homeostasis. Genetic or acquired conditions resulting in decreased lipid storage and endocrine function, so-called partial lipodystrophies, are closely associated with ectopic lipid deposition and cardiometabolic diseases (Fiorenza et al., 2011). Similarly, mounting evidence points to compromised adipocyte function as a key driver of cardiometabolic co-morbidities in obesity (Hagberg & Spalding, 2024). Adipogenesis is the process by which stromal progenitor cells of the mesenchymal lineage develop into lipid-laden adipocytes (Rauch & Mandrup, 2021). The efficiency of this process is of major biomedical importance, as it is required for the development of well-functioning adipose tissue, as well as for healthy expandability of adipose tissue (Ghaben & Scherer, 2019). Moreover, the fact that this process can be efficiently induced in vitro has made it one of the most well-studied differentiation model systems. Studies of many individual genes, as well as genome-wide transcriptional and epigenomic studies, have documented that the transcriptional networks regulating in vitro adipogenesis consist of at least two consecutive waves of transcription factors (TFs), where the first wave is induced directly by the adipogenic inducers (typically a cocktail consisting of insulin, dexamethasone, and a cAMP-elevating agent). This first wave remodels the chromatin template and activates the master regulators peroxisome proliferator-activated receptor γ (PPARγ) and CCAAT/enhancer-binding protein α (C/EBPα), which drive the second wave activating the adipocyte gene program. PPARγ is the most important of the two master regulators, being indispensable for adipocyte differentiation in vitro as well as in vivo and sufficient to drive trans-differentiation of other mesenchymal lineages into adipocytes (Rauch & Mandrup, 2021; Siersbaek et al., 2012). The master regulator attributes of PPARγ rely at least in part on its biochemical properties, as suggested by comparative studies of lipodystrophy mutants of PPARγ (Madsen et al., 2022b) as well as comparative studies with other members of the PPAR family (Bugge et al., 2009; Nielsen et al., 2006). In addition, the ability to dramatically increase PPARG expression in response to adipogenic cues may play a role. Thus, the expression of PPARG must be tightly controlled; initially repressed in the progenitor state, and then rapidly induced in response to the transcription factors of the first wave of adipogenesis. Several transcription factors, most notably C/EBPβ and C/EBPδ, have been shown to be directly involved in the activation of the PPARG locus (Siersbæk et al., 2014). Surprisingly, however, the mechanisms by which the PPARG locus is activated and the transcription of PPARG is switched on during adipogenesis remain poorly understood. Adipogenesis is associated with dramatic chromatin remodeling and de novo enhancer activation (Rauch et al., 2019; Siersbaek et al., 2011; Siersbæk et al., 2014) and dynamic rewiring of promoter-enhancer interactions (Siersbæk et al., 2017). The remodeling of chromatin is a common feature of differentiation pathways (Bonev et al., 2017; Di Giammartino et al., 2019; Freire-Pritchett et al., 2017; Murphy et al., 2024); however, a direct comparison showed that chromatin remodeling in response to adipogenic inducers is much more dramatic than in response to osteogenic inducers (Rauch et al., 2019). Moreover, using enhancer capture Hi-C (ECHi-C), we recently demonstrated extensive 3D interconnectivity of enhancers in multipotent human stromal progenitors (hMSC-TERT4), forming highly interconnected enhancer (HICE) communities that are dynamically regulated during adipogenesis (Madsen et al., 2020). Interestingly, lineage-selective promoters are generally associated with the most highly connected HICE communities, indicating that enhancer cooperativity is required for lineage determination. Intriguingly, the lineage-selective HICE communities associated with the different lineage fates of mesenchymal stem cells (MSCs), including adipogenesis, osteogenesis, myogenesis, and chondrogenesis, are already established at the stem cell state, indicating that these lineage fates are pre-programmed in progenitor cells. In response to the adipogenic cocktail, new enhancers are activated and join the adipogenic HICE communities, thereby increasing the connectivity of these communities (Madsen et al. 2020). These results indicate that HICE communities dynamically integrate signals from multiple enhancers to activate lineage-specific genes. Here, we have investigated the activation of the most important locus in adipogenesis, that of PPARG , during adipocyte lineage determination and differentiation of hMSC-TERT4 cells. Using genome editing, we demonstrate important and non-redundant roles of individual enhancers in the PPARG enhancer community. We report evidence of extensive enhancer crosstalk in cis as well as in through feedback activation . Furthermore, we show that key enhancers in the locus overlap with cardiometabolic disease-associated genetic variants, which we predict to regulate PPARG2 transcription, thereby underscoring the in vivo relevance of the identified enhancers. Results The PPARG locus is primed and connected in undifferentiated human MSC Taking advantage of our previously published Hi-C data from undifferentiated human bone marrow-derived mesenchymal stem cells (hMSC-TERT4) (Madsen et al., 2020), we inspected the overall chromatin topology around the human PPARG locus. This showed that PPARG is isolated in a topologically associating domain (TAD) of 350 kb (chr3: 12,174,868 - 12,526,858) (Fig. 1a) at the stem cell stage prior to chromatin remodeling and induction of PPARG expression (Rauch et al., 2019).Notably, this TAD also contains TIMP4 and the 3’end of SYN2 with no known functions in adipogenesis. Consistent with the TAD structure, ChIP-seq revealed strong CTCF binding at divergent motifs at the boundaries of the PPARG TAD (Fig. 1a). Analysis of publicly available Hi-C data from other stem cells (Dixon et al., 2015) indicated that the locus is also isolated into a partially defined TAD in embryonic (ESC), mesendodermal cells and mesenchymal stem cells (MSC), but not in neuronal progenitor cells, indicating that the locus is organized very early during development but loses its connectivity in some lineages (Extended Data Fig. 1a). Inspection of the PPARG locus connectivity at higher resolution using our previously published ECHi-C data from hMSC-TERT4 cells (Madsen et al., 2020) confirmed that the locus is engaged in an isolated chromatin network in the undifferentiated state (Fig. 1b). Consistent with other adipocyte enhancer communities (Madsen et al., 2020), the locus becomes even more connected and confined during differentiation (Fig. 1b). ChIP-seq analysis revealed that the locus is already primed by H3K4me1 in the undifferentiated state but only gains the active enhancer mark H3K27ac by day 3 of differentiation, coinciding with the activation of PPARG gene expression (Fig. 1a, Fig. 1c). The PPARG locus is transcribed from two major transcription start sites (TSS), PPARG1 and PPARG2 , where PPARG2 is the adipocyte-specific TSS (Fajas 1997). The PPARG2 transcript encodes PPARγ2, which is characterized by an additional 28 amino acids in the N-terminal compared with PPARγ1. PPARγ2 has been shown to be the most adipogenic isoform (Mueller et al., 2002; Ren et al., 2002) and to be required for efficient adipocyte differentiation in vitro and in vivo in mice (Koutnikova et al., 2003; Medina-Gomez et al., 2005; Zhang et al., 2004). Notably, in hMSC-TERT4 cells, the PPARG1 promoter is inactive, and only the adipocyte specific PPARG2 promoter is activated by the adipogenic cocktail (Extended Data Fig. 1b-c), thereby enabling us to specifically study the activation of the PPARG2 promoter without interference from the PPARG1 promoter. Similarly to PPARG2 , the two other genes located in the TAD, SYN2 and TIMP4 , are selectively induced during adipocyte differentiation, whereasthe expression of genes outside the TAD borders, TAMM41 and TSEN2 , are not affected by differentiation (Fig. 1c). Collectively, these results show that the PPARG2 cis -regulatory enhancer network is demarcated prior to cellular commitment and then activated in response to the adipogenic cocktail. Importantly, the regulatory functions of these enhancers appear to be contained within the TAD. The connectivity and activity of the PPARG enhancer community is increased during early adipogenesis Integration of our previously published ECHi-C, MED1 ChIP-seq, and DNase-seq data obtained during adipogenesis and osteogenesis of hMSC-TERT4 cells (Madsen et al., 2020; Rauch et al., 2019) revealed marked activation of several putative enhancers between 1 and 3 days of adipocyte differentiation (Fig. 2a) and showed that these enhancers are part of a highly interconnected enhancer (HICE) community (Fig. 2a top panel). We selected 9 putative enhancers in three regions of interest that are framing the PPARG gene and are already engaged in the enhancer community in the undifferentiated state. Enhancers were named according to their position (in kb) relative to the PPARG2 TSS. Using DNase hypersensitive sites (Fig. 2a) and MED1 occupancy (Fig. 2a) and H3K27 acetylation (Extended Data Fig. 2a) as a proxy for enhancer activity, we showed that the two distal upstream enhancers (E-130, E-123) are already active in undifferentiated cells and remain active throughout both adipocyte and osteoblast differentiation. In contrast, the two promoter-proximal enhancers (E-1, E+2) are activated de novo by day 1 of adipogenesis, and the 5 distal downstream enhancers (E+89, E+102, E+107, E+111, E+125), collectively classified as a super-enhancer by the ROSE algorithm (Lovén et al., 2013; Warren et al., 2013) (Extended Data Fig. 2b), are activated de novo by day 1-3 (Fig. 2a, Extended Data Fig. 2a). During the time course of adipogenesis, none of the nine enhancers bind CTCF, except for E+107, where CTCF binding is observed from day 1 of adipogenesis (Fig. 2a). To specifically investigate the temporal dynamics of chromatin interactions in the PPARG enhancer community during differentiation, we performed Capture-C in the progenitor state and on day 10 of adipogenesis using the nine putative enhancers and the PPARG2 TSS as bait. We focused on pairwise interactions within the specific region of the PPARG locus (chr3: 12.1-12.6 Mb), excluding interactions occurring within ± 2kb of a viewpoint. To compare interactions across samples, we normalized the unique interaction counts within the region ( cis interactions) to 50,000 counts per sample and viewpoint (Supplementary Information Fig. 1). In line with the dynamics of the ECHi-C-based interactions seen in the PPARG2 HICE community (Fig. 2a), Capture-C showed that the main long-distance interaction in the undifferentiated state is between the upstream enhancer E-123 and the super-enhancer constituent E+125, which thereby “pinches” the locus (Fig. 2b-d, Extended Data Fig. 3a). In addition, there are strong interactions between the two upstream enhancers E-130 and E-123 (Fig. 2c). Interestingly, we also observed strong connections between the super-enhancer constituents, which are primed but not yet remodeled or active (Fig. 2c). The most dramatic changes in connectivity during adipogenesis are the strong inductions of interactions between the PPARG2 promoter-proximal regions and the downstream super-enhancer constituents (Fig. 2b, Fig. 2c). Taken together, the connectivity of the PPARG enhancer community is increased during adipogenesis, concomitantly with the sequential activation of putative enhancers near the TSS as well as in the downstream super-enhancer region (Fig. 2d). PPARG2 activation requires cooperativity between many non-redundant enhancers To investigate the role of individual enhancers in chromatin topology and activation of PPARG2 transcription, we generated a hMSC-TERT4-based clonal cell line with inducible Cas9 (iCas9; hMSC-TERT4/iCas9-4) (Fig. 3a, Extended Data Fig. 4a). hMSC-TERT4 cells were transduced with Lenti-iCas9-Neo (Cao et al., 2016) encoding a flag-tagged human codon-optimized S. pyogenes Cas9 coupled to green fluorescent protein (GFP) by a Porcine teschovirus -1 2A (P2A) self-cleavage peptide (Extended Data Fig. 4b-c). In this system, the expression of GFP-coupled Cas9 can be transiently induced by doxycycline (Extended Data Fig. 4b-d), thereby minimizing off-target effects from constitutive Cas9 expression. The differentiation potential of the selected hMSC-TERT4/iCas9-4 clone is comparable to wild-type hMSC-TERT4 cells (Extended Data Fig. 4e). The hMSC-TERT4/iCas9-4 cell line was transfected with pairs of gRNAs designed based on the position and width of the corresponding DNase-seq peak to excise each of the putative enhancers (Fig. 3a, Extended Data Table 3, Supplementary Information Fig. 2). For each deletion, we selected three independent clones. As controls, we used three independent clones not transfected with gRNA (‘No gRNA’). In addition, we generated three clones transfected with control gRNA (‘Ctr gRNA’) targeting intron 1 of the PPP1R12C gene, which is suggested to be a “safe harbor” for incorporation of adeno-associated virus and thus genetic modifications (DeKelver et al., 2010) . For each clone, the zygosity and the size of the deletion were confirmed by Sanger sequencing of regions flanking the deleted sites (Extended Data Table 3, Supplementary Information Table 3). The control and enhancer deletion clones were induced to undergo adipocyte differentiation, RNA was harvested on day 10 and used for quantification of transcripts by qPCR. Interestingly, deletion of several individual enhancers, i.e., the promoter-proximal enhancer E-1 and E+2, as well as the super-enhancer constituents E+102, E+107, and E+111, results in a significantly decreased expression of PPARG2 and SYN2 , with the deletion of E+102 almost completely preventing the induction of PPARG 2 (Fig. 3b). Deletion of the distal upstream E-123 is associated with a slight decrease in PPARG2 expression, whereas deletion of the distal upstream E-130 and distal downstream E+125 enhancers do not affect PPARG2 (Fig. 3b). These results indicate an important and non-redundant role of the promoter-proximal enhancers E-1 and E+2, as well as the super-enhancer constituents E+102, E+107, and E+111. Inspection of epigenetic profiles of these enhancers shows that they are all remodeled and activated (Fig. 2a) and become much more interconnected (Fig. 2c) during adipogenesis. Furthermore, during adipogenesis, all super-enhancer constituents, E+102 in particular, become more connected to the PPARG2 TSS (Fig 2b). Consistent with PPARγ being a master regulator of adipogenesis and a direct activator of most genes involved in lipid handling and metabolism in adipocytes (Lefterova et al., 2014), there is a strong correlation between PPARG2 expression and lipid accumulation in the different enhancer-deleted clones at day 9 of adipocyte differentiation (Pearson R=0.99, Fig. 3c; Extended Data Fig. 4f). In summary, at least six out of nine putative enhancers in the PPARG HICE community are required for the full induction of PPARG2 expression and ultimately adipocyte differentiation. Notably, none of the enhancer deletions affect the expression of TAMM41 and TSEN2 , which are located outside the TAD, or the expression of TIMP4 in the TAD (Fig. 3b). By contrast, the expressions of SYN2 and PPARG2 are similarly affected by the individual enhancer deletions (Fig. 3b). The canonical SYN2 promoter region is outside the TAD and does not seem to interact with the PPARG community enhancers on day 10 of adipogenesis (Fig. 1b), suggesting SYN2 might be transcribed from an alternative TSS. Moreover, both controls, ‘No gRNA’ and ‘Ctr gRNA’, exhibit comparable levels of expression of PPARG 2 and neighboring genes (Fig. 3b). Therefore, for subsequent experiments, we proceeded with ‘No gRNA’ as the control, unless otherwise stated. Deletion of key enhancers affects the connectivity in the PPARG locus To explore the role of individual enhancers in the development of local topology in the PPARG locus, we performed Capture-C in controls and in enhancer-deleted clones before induction of differentiation (day 0) and following exposure to the adipogenic cocktail for 10 days. Results from ‘No gRNA’ and ‘Ctr gRNA’ controls correlated well (Extended Data Fig. 3b) and were therefore considered as combined controls ( n =4) for all statistical tests. Interactions where one of the interacting elements was deleted were disregarded. In the progenitor cell state, where neither the enhancers in the promoter-proximal region nor the super-enhancer constituents are remodeled, as assessed by DNase-seq (Fig. 2a), or active, as assessed by H3K27ac and MED1 ChIP-seq (Fig. 2a, Extended Data Fig. 2a), only a few enhancer interactions are affected by the deletion of other enhancers (Fig. 4a). Interestingly, however, the deletion of E+2, E+107 or E+125 lead to decreased interaction between E-1 and E+89, while that between E+102 and the PPARG2 TSS is increased. This indicates that the E+102-TSS connectivity may be limited by other super-enhancer constituents and E+2, possibly by some competitive mechanisms. Overall, these data indicate that the connectivity of the PPARG HICE community at the progenitor stage is relatively resistant to the deletion of constituents. Moreover, the structural changes that occur upon enhancer deletion at this stage do not correlate well with the ability of the enhancers to induce PPARG2 . At day 10 of adipogenesis, when the community is more connected, many enhancer interactions are affected by the deletion of other enhancers. The most notable changes in enhancer interaction occur in response to the deletion of enhancers that are important for the induction of PPARG2 expression (Fig. 4a), in particular E+102 (Fig. 4b) and E+2 (Fig. 4c). Deletion of E+102 inherently ablates the strongest and most differentiation-dependent interactions between the super-enhancer and the TSS; however, it also leads to a general destabilization of the connectivity between super-enhancer constituents (e.g., the interactions between E+89 and E+107; and between E+107 and E+111) as well as a decrease in the interactions between other super enhancer constituents and the TSS and promoter proximal enhancers (e.g., interactions between E+89 and TSS and E-1; and between E+107 and E+2) (Fig. 4b). Deletion of E+2 leads to a dramatic decrease in the interaction between the super-enhancer constituents and the TSS as well as a decreased connectivity of E+111 with E+107 and with promoter-proximal E-1, respectively (Fig. 4b). Interestingly, however, the deletion of E+2 also increases the interaction frequency between E+102 and E-123 and between E+102 and E-107 (Fig. 4c). This indicates that E+2 may help direct the super-enhancer to interact with the TSS and in doing so may compete with other super-enhancer constituents for interactions with E+102. Taken together, while many genomic interactions in the PPARG locus are unaffected by the deletion of individual enhancers, deletion of E+102 or E+2 interferes with adipogenesis-associated stabilization of interactions in the PPARG HICE community, specifically between the PPARG2 proximal regions and the distal downstream super-enhancer (Fig. 4d). Enhancers in the PPARG locus display wide-spread functional crosstalk To investigate whether the enhancers within the PPARG enhancer community affect the activity of each other, we assessed MED1 occupancy in individual clones at day 10 of adipocyte differentiation, where all enhancers are active in control cells (Fig. 2a). Interestingly, deletion of E+102 leads to decreased MED1 recruitment to other enhancers activated during adipogenesis, i.e. E-1, E+2, E+89, E+107, E+111 and E+125, as assessed by ChIP-PCR (Fig. 5a, Extended Data Fig. 5a) as well as ChIP-seq (Fig S6b). Similar locus-wide effects, but with a smaller magnitude, are observed in response to deletion of other enhancers that are important for PPARG2 expression (Fig. 3b, Fig. 5a). Thus, at day 10 of adipogenesis, MED1 occupancy is not only lost at the deleted enhancer or its closest neighbors but is also lost at other enhancers in the locus. The summed MED1 occupancy at the 9 putative enhancers correlated well (Pearson R=0.83, Fig. 5b) with PPARG2 expression, indicating that transcription of the locus is determined by the sum of enhancer activity. We also assessed MED1 ChIP-seq at day 3 of adipogenesis, the earliest time point with significant MED1 occupancy at the super-enhancer (Fig. 2a). Similar to MED1 ChIP-qPCR data at day 10 of adipogenesis (Fig. 5a, Extended Data Fig. 5b), this showed that the most pronounced effect of E+102 deletion is a lower MED1 signal at the other super-enhancer constituents, E+89, E+107, E+111, and E+125 (Fig. 5c), whereas the most pronounced effect of E+2 deletion is a decreased MED1 occupancy at the neighboring promoter-proximal E-1 and the distal E+125 (Fig. 5c). Importantly, the effects of both E+102 and E+2 deletions are restricted to the PPARG enhancer community (Extended Data Fig. 5c-d). Collectively, these data indicate profound cooperativity between enhancers in the PPARG locus, in particular between those that interact undifferentiated cells (Fig. 5d). PPARG2 enhancer network is activated through extensive enhancer crosstalk We have previously shown that interaction between enhancers occupied by the same transcription factor is correlated with increased occupancy of that factor at both enhancers, a phenomenon which we termed ‘cross-interaction stabilization’ (CIST) (Madsen et al., 2020), and we speculated that such mechanisms could be involved in cis cooperativity between enhancers in the PPARG locus. C/EBPβ is known to be a key driver of early adipogenesis and has been proposed to directly activate Pparg gene expression in mouse preadipocytes (Siersbæk et al., 2014). Consistent with that, C/EBP motif activity and C/EBPβ expression are rapidly increased in hMSC-TERT4 cells in response to the adipogenic cocktail (Fig. 6a). C/EBPβ ChIP-seq analysis showed that the distal upstream enhancers, E-130 and E-123, are bound by C/EBPb in the progenitor state, while the promoter-proximal enhancers, E-1 and E+2 and super-enhancer constituents E+102, E+107, E+111 recruit C/EBPβ at day 1 of differentiation (Fig. 6b). Notably, the binding of C/EBPβ to the super-enhancer at day 1 (Fig. 6b) precedes chromatin remodeling (Fig. 2a), suggesting that C/EBPβ might serve as a pioneering factor in the super-enhancer region. This is consistent with similar analyses in the 3T3-L1 preadipocyte cell line, where C/EBPβ binding also precedes chromatin remodeling (Siersbaek et al., 2011). To explore whether early C/EBPb occupancy in the PPARG HICE community is affected by the deletion of other C/EBPβ bound interacting enhancers, we performed C/EBPb ChIP-seq in control and E+2 and E+102 deletion cell lines at day 1 of adipogenesis (Fig. 6c). Interestingly, and consistent with CIST, deletion of E+102 leads to significantly decreased C/EBPβ occupancy at super-enhancer constituents E+107 and E+125, which are connected to E+102 at the progenitor state (Fig. 6c, Fig. 2c). By contrast, C/EBPβ binding at the distal upstream enhancers, E-130 and E-123, and at the promoter-proximal E-1 and E+2 are not significantly affected (Fig. 6c). Deletion of E+2 decreases C/EBPβ occupancy at distal super-enhancer constituent E+125, the only enhancer that interact with E+2 at the progenitor state (Fig. 6c, Fig. 2c). Importantly, these effects are restricted to the PPARG enhancer community (Extended Data Fig. 6a), as the C/EBPβ ChIP-seq signal in the E+2 and E+102 deletion cell lines is comparable to control lines outside the PPARG locus (Extended Data Fig. 6a) and at a genome-wide level (Extended Data Fig. 6b). Collectively, and in line with the CIST model, these findings indicate that interactions between enhancers bound by C/EBPβ increase the occupancy of C/EBPβ (Fig. 6d). PPARγ has been proposed to positively feedback regulate its own expression. First, PPARγ binds directly to regulatory elements in the PPARG2 promoter in mice and humans (Schmidt et al., 2011). Second, PPARγ is also involved in the induction of other transcription factors, including C/EBPα, which may feedback to regulate the PPARG2 promoter (Salma et al., 2006; Siersbæk et al., 2014). The importance of this feedback regulation is supported by early loss-of-function studies (Lefterova et al., 2008). Indeed, ChIP-seq mapping of PPARγ binding during hMSC-TERT4 adipogenesis showed that PPARγ binds to E-1 and E+102 at day 3 of adipogenesis (Fig. 6e), both of which have moderately conserved PPAR response elements (PPREs) (Extended Data Fig. 6c). At later stages of adipogenesis, PPARγ occupancy is also gained at E+107 and E+2, although no consensus PPRE is found in E+2. In summary, three out of the four most important enhancers for regulating PPARG2 expression, MED1 occupancy, and lipid accumulation exhibit direct binding by PPARγ itself. Moreover, C/EBPα ChIP-seq at day 10 of differentiation showed that C/EBPα binds to the promoter-proximal E-1, E+2, and to super-enhancer constituents E+102, E+107, E+111 by day 10 of adipogenesis, indicating that C/EBPα is also involved in a positive feedback loop (Fig. 6e). Thus, a strong enhancer in the PPARG locus may affect the activity of other enhancers not just in cis but also in trans by contributing to increased expression of PPARγ which then either directly activates other PPARG enhancers or the expression of other adipogenic transcription factors (Fig. 6f). Interestingly, we also noticed a strong binding of PPARγ by day 3 near the TSS of a shorter predicted isoform of SYN2 (Fig. 6e), suggesting that the co-regulation of SYN2 and PPARG2 during adipogenesis might be mediated in part in trans by PPARγ activation of promoter proximal enhancers in the SYN2 locus . Taken together, the enhancers in the PPARG HICE community show clear evidence of cis cooperativity at early stages of differentiation prior to expression of PPARG ; however, at later time points, crosstalk in trans by positive feedback regulation is likely to play a major role as well. Key PPARG2 enhancers overlap with non-coding genetic variants associated with metabolic disease traits Given the importance of well-functioning adipocytes for metabolic health and the obligate role of PPARγ in adipocyte differentiation and function (Hagberg & Spalding, 2024; Lefterova et al., 2014), we rationalized that if the enhancers identified in vitro are also important in vivo , they would overlap with metabolic disease-associated genetic variants. We queried the PPARG TAD (chr3: 12,174,868 - 12,526,858) and extracted 796 haplotypes associated with a series of cardiometabolic traits comprising in total 6,030 variants in high linkage disequilibrium (LD) (r2> 0.8, CEU) from GWAS in the public domain (Fig. 7a). The vast majority of these genetic variants associated with common complex metabolic traits map to non-coding elements in the PPARG locus. Because any given variant in a haplotype could be functional and, therefore, mediate the association with metabolic disease, we used the Enformer model (Žiga Avsec, Vikram Agarwal, et al., 2021), a state-of-the-art sequence-based deep learning approach for the prediction of non-coding variant effects on gene expression and chromatin states. We predicted the variant effects on PPARG2 transcript expression as well as P PARG enhancer regions and ranked the variants based on their positive or negative effects on the target regions (Fig. 7b, Tables S1, S2). The analysis revealed that 1,759 out of 3,596 variants have predicted significant effects on PPARG2 gene expression. Notably, 11 out of the 40 top-scoring variants with predicted (positive and negative) effects on PPARG2 signal and/ or enhancer accessibility, overlap with enhancers identified in the hMSC-TERT4 model system (Fig. 7b, Fig. 7c, Extended Data Table 1-2). In particular, many of these genetic variants are located in the three most important enhancers, E+102, E+2, and E-1 (Fig. 7b, Fig. 7c, Extended Data Table 1-2), indicating that these enhancers are also important for the regulation of PPARG expression in human physiology. One of the top non-coding variants, rs7647481 A/G, located in E-1, is a common variant associated with decreased risk of developing type 2 diabetes. Using label-free quantitative LC–MS/MS proteomics, this variant has previously been shown to enhance the binding of YY1 leading to increased activation of the PPARG2 promoter (Lee et al., 2017). Another interesting, low-frequency variant, rs181025382 G/A (MAF=0.0001, CEU) associated with lipid traits including decreased levels of total cholesterol and non-HDL cholesterol resides in E+102 and is predicted to negatively impact both PPARG2 expression and enhancer accessibility level (Fig. 7c). Interestingly, this variant is located in a peroxisome proliferator response element (PPRE) in E+102, which is bound by PPARγ from day 3 of differentiation (Fig. 6e). In the minor allele variant, the A in the spacer between the two half-sites of the direct repeat is replaced with a less optimal G (Fig. 7d). To estimate the effect of a G in this position on the binding affinity of the PPARγ:RXR heterodimer, we took advantage of the recently published nuclear receptor sequence specific binding preferences for binding to naked DNA in the presence or absence of agonists based on high throughput SELEX data (Bhimsaria et al., 2023). In E+102, the minor variant with a G in the spacer showed an approximately 3-fold lower affinity for PPARγ:RXR compared with the major variant, which has an A in that position (Fig. 7d). The synthetic agonist rosiglitazone slightly increased the affinity of PPARγ:RXR for both PPRE variants (Fig. 7d), consistent with previous ChIP-seq data (Haakonsson et al., 2013). To investigate the biological relevance of a 3-fold difference in binding affinity for the PPARγ:RXR heterodimer, we compared PPARγ:RXR affinity (± rosiglitazone) between the 1,000 top PPARγ-bound enhancers and 1000 PPARγ-unbound enhancers from our datasets derived from hMSC-TERT4 (Extended Data Fig. 7a). Importantly, PPARγ generally exhibits about 3-fold higher affinity for bound sites compared to unbound enhancers, with affinity being higher for rosiglitazone-activated PPARγ (Extended Data Fig. 7a). This indicates that a 3-fold difference in naked DNA affinity may have major functional implications for binding of PPARγ to sites in chromatin. To further interrogate PPARγ binding to chromatin in hMSC-TERT4 cells, we applied BPNet (Ž Avsec et al., 2021), a deep learning model to predict base-resolution binding syntax from DNA sequences within PPARγ binding peaks in hMSC-TERT4 (Fig. 7e). We used i n silico mutagenesis to investigate the effect of rs181025382 on PPARγ:RXR binding in E+102 and found that substitution of the major allele A with minor allele G leads to -0.12 log fold change (Fig. 7e). Further, we used TF-MoDIScO for motif discovery within identified seqlets in PPARγ-bound sequences. By inferring the data with previously annotated motifs from JASPAR database, we identified PPARγ:RXR as a top de novo motif matching with PPARγ:RXR (MA0065.2) motif (FDR= 1.3e-10, Supplementary Information Table 6) (Fig. 7f.) To ensure the predicted decrease in binding upon A to G substitution was not specific to E+102, we identified PPARγ binding sites genome-wide that contain the PPARγ motif with an A at the 9 th position. At all these sites, we mutated the A to a G and predicted the difference in binding using the trained BPNet model. This showed -0.09 log fold change in PPARγ:RXR binding affinity genome-wide upon substitution of A with G, similar to results for rs181025282 (Extended Data Fig. 7b-c). This suggests that the G variant of rs181025382, is likely to interfere with the ability of PPARγ to activate this enhancer, thereby reducing the positive feedback activation of PPARγ on PPARG2 expression. Our results show that PPARG enhancers that are critical for in vitro activation of PPARG2 in response to adipogenic inducers overlap with functional non-coding genetic variants associated with metabolic disease. The genetic variants in these enhancers are among the top-scoring variants predicted to affect PPARG2 expression. These results suggest that the enhancers identified in our hMSC-TERT4 model system may be important for human physiology. The finding that one of these notably low-frequency, non-coding genetic variants decreases the binding affinity of PPARγ:RXR for the most important enhancer constituent in the super-enhancer further supports the notion that positive feedback activation of the locus plays an important role in the activation of PPARG2 expression during adipogenesis, a function relevant to cardiometabolic disease in humans. Discussion Here, we show that the activation of the PPARG2 promoter during adipogenesis is driven by a locus-wide HICE community consisting of upstream enhancers, promoter-proximal enhancers as well as a downstream super-enhancer. We show that the enhancers function in a non-redundant and highly cooperative manner involving early cooperativity in cis between the enhancers as well as feedback activation of the enhancers by the product of the locus, PPARγ, and possibly other transcription factors, such as C/EBPα, induced by PPARγ. Disease-associated genetic variants that are predicted to affect PPARG2 expression and enhancer accessibility overlap with the most important enhancers in our hMSC-TERT4 model system, thereby supporting the importance of these enhancers for human physiology and disease. We have previously shown that in the multipotent human mesenchymal stromal cell line, hMSC-TERT4, genes specific for lineage fates, such as the adipocyte, osteoblast, myoblast and chondrocyte lineages, tend to be marked by preestablished HICE communities already at the progenitor state. In response to adipocyte and osteoblast differentiation cocktails, the connectivity of the respective lineage specific HICE communities are strengthened, coinciding with the activation of transcription of the respective target genes (Madsen et al., 2020). Similarly, findings from other studies have indicated a priming role of the chromatin 3D structure at the progenitor state, which shifts to an instructive role during Drosophila (Pollex et al., 2024) and mouse (Chen et al., 2024) embryogenesis. In line with this, we show that the PPARG locus is primed by pre-established enhancer-enhancer interactions in the silent state in progenitors, suggesting a permissive nature of these interactions. In response to the differentiation cocktail, the strength of the existing interactions increases, especially those between the super-enhancer constituents and the promoter-proximal enhancers and PPARG2 TSS. Furthermore, new interactions are formed between the promoter-proximal enhancers and the downstream super-enhancer constituents. In the silent progenitor state, the PPARG locus is also primed by H3K4me1, similar to what has been observed for other developmental genes, such as PDX1 during pancreatic differentiation (Wang et al., 2015), and many early developmental genes in human ESCs (Rada-Iglesias et al., 2011). This epigenetic and 3D chromatin architecture of the PPARG locus is likely to prime rapid lineage-specific activation in response to adipogenic cues. Using inducible CRISPR/Cas9 to individually delete 9 enhancers that are engaged in both pre-established and induced interactions with each other and with the PPARG2 promoter, we show that 6 of the 9 enhancers are required for efficient activation of the PPARG2 promoter. Although, deletion of upstream enhancer E-130 interferes with overall connectivity in the locus in the progenitor state, there is no effect of deletion of this enhancer on PPARG2 expression, indicating that this enhancer may primarily be contributing to overall locus connectivity in the progenitor state. By contrast, individual deletions of most of the enhancers that are activated during differentiation, i.e., the two promoter-proximal enhancers (E+2 and E-1) as well as 3 of the 5 super-enhancer constituents (E+102, E+107, E+111), lead to compromised induction of the PPARG2 promoter. In particular, deletion of the super-enhancer constituent E+102 completely prevents the activation of the promoter. The difference in importance of super-enhancer constituents cannot be explained solely by connectivity, as all constituents interact with the promoter-proximal enhancers and the PPARG2 TSS. However, the importance of the enhancers appears to be closely linked to their effect on the activity of other enhancers in the locus. Previous studies have proposed enhancer cooperativity in cis , particularly in the context of the super-enhancer constituents (Bahr et al., 2018; Choi et al., 2021; Huang et al., 2018; Oudelaar et al., 2018), where cooperativity has been suggested to be related to condensate formation driven by multivalent interactions between intrinsically disordered regions (IDRs) of co-factors and transcription factors (Cho et al., 2018; Du et al., 2024; Lyons et al., 2023; Sabari et al., 2018; Trojanowski et al., 2022). Moreover, QTL analyses have shown that genomic variations within enhancers affect the activity of other enhancers (Grubert et al., 2015). However, documented cooperativity by loss-of-function studies has been limited to enhancers within close genomic proximity, such as enhancer constituents in super-enhancers (Blayney et al., 2023; Huang et al., 2018; Thomas et al., 2021). In this work we demonstrate enhancer cooperativity in cis within a 3D enhancer community that spans the PPARG locus and is sequentially activated during adipogenesis. First, deletion of E+102 or E+2 affects MED1 occupancy at day 3 of adipogenesis at multiple other enhancers in the community, suggesting that enhancers with strong recruitment of Mediator can stimulate Mediator occupancy and activity of nearby enhancers, possibly through condensate formation. Second, the enhancers with the strongest effect on other enhancers and PPARG2 expression are characterized by early recruitment of C/EBPβ at day 1 of differentiation, and deletion of E+2 or E+102 leads to reduced C/EBPβ occupancy at interacting enhancers. Notably, this is consistent with our CIST model, showing strong correlation between binding of homotypic transcription factors at interacting enhancers (Madsen et al., 2020). The underlying mechanism of CIST is unclear but may involve condensate formation and/or more specific protein-protein interactions. C/EBPb has been shown to be an important transcription factor in early adipogenesis (Siersbæk et al., 2014; Tanaka, 1997). Here we show that C/EBPβ appears to act as a pioneer transcription factor at the PPARG super-enhancer by binding to inaccessible enhancer constituents at day 1 of adipogenesis. The structural plasticity of IDRs and post-translational modifications (PTM) of C/EBPb (Dittmar et al., 2019) may facilitate its interaction with DNA on nucleosomes, as well as its interaction with chromatin remodeling complexes and histone modifiers to drive remodeling of the super-enhancer constituents. These features of C/EBPβ may also contribute to enhancer cooperativity in cis . Importantly, since PPARγ binds to several of its own enhancers at later time points of differentiation and induces C/EBPα and other transcription factors that might also stimulate activity of enhancers in the PPARG locus, the enhancers may also cooperate in trans once PPARγ is induced. Thus, at later time points of differentiation (after day 3) it is difficult to distinguish between enhancer crosstalk in cis and trans . Feedback activation has been demonstrated for other core developmental transcription factors (Saint-André et al., 2016), suggesting that feedback activation may also contribute to, and needs to be considered, when evaluating enhancer cooperativity in other loci. The identification of E+102 as an obligate enhancer for PPARG2 induction is intriguing. Deletion of E+102 largely blocks the activation of the super-enhancer, as determined by MED1 occupancy, by day 3 and day 10 of differentiation and interferes with the recruitment of C/EBPβ to the enhancer constituents. Moreover, it is the most prominent target enhancer of PPARγ feedback regulation from day 3, and it remains the most active enhancer constituent (based on MED1 occupancy) in the super-enhancer. Thus, the importance of E+102 may be due to its dual functions, acting as a seed enhancer in the super-enhancer and as a key target enhancer for feedback action by PPARγ (and possibly other adipogenic transcription factors). Such positive feedback regulation through E+102 and other enhancers may contribute to the rapid and efficient activation of PPARG2 during adipogenesis and thereby the ability of PPARγ to act as a master regulator. Similar regulatory principles may be involved in driving the induction other master regulators of different lineages to ensure efficient induction of the master regulator in response to differentiation cues. Moreover, this regulatory principle may also sensitize PPARG2 expression to PPARγ point mutations, such as those observed in lipodystrophies (Broekema et al., 2019; Madsen et al., 2022a), thereby resulting in lower expression of PPARγ variants and further decreasing PPARγ activity. To link these enhancers to human physiology, we applied the sequence-based Enformer machine learning model to infer the effect of non-coding genetic variants associated with cardiometabolic traits on the PPARG locus across the frequency spectrum. Importantly, we showed that 11 out of the 40 variants predicted to have the largest effect on PPARG2 expression overlap with key PPARG2 enhancers identified in hMSC-TERT4 cells, thereby supporting the in vivo importance of these enhancers. Interestingly, we identified a low frequency variant associated with cardiometabolic diseases, which is located in the PPRE in the obligate enhancer, E+102. This variant is predicted to reduce PPARγ binding affinity to both naked and chromatin embedded DNA and may therefore interfere with the ability of PPARγ to feedback activate its own expression. Work from others have shown that single nucleotide polymorphism in the PPRE can affect in vivo PPARγ occupancy, the response to PPARγ agonists as well as human metabolic disease risk (Soccio et al., 2015). The discovery of the PPRE variant in E+102 supports the relevance of cardiometabolic SNPs targeting PPARγ response elements. In relation to the PPARG locus, this finding supports the importance of positive feedback activation for the activation of PPARG2 expression. In conclusion, our study reveals that extensive enhancer cooperativity and feedback activation plays an essential role in the activation of PPARG2 expression during adipogenesis of human stromal progenitor cells. The PPARG locus is primed for activation in the progenitor state by a HICE community and by H3K4me1. Upon induction of differentiation, cooperativity in cis between the promoter-proximal enhancers as well as enhancer constituents in the super-enhancer drive activation of the locus. At later timepoints, the product of the locus, PPARγ, feedback-activates its own expression in trans . We propose that the priming of the locus as well as the elaborate cooperativity in cis and trans play key roles in rapidly switching on the PPARG2 promoterin response to adipogenic cues. This dynamic regulation may also have profound impact on the regulation of PPARG expression in response to metabolic signals. Future detailed investigations of the mechanisms by which the expression of PPARG, the master regulator of adipogenesis and adipocyte function and a key regulator cardiometabolic health, is controlled in adipocytes and other cells in human tissues is likely to offer important insight into personalized risk assessment and treatment of cardiometabolic diseases. Declarations Acknowledgements The work was supported by grants from the Independent Research Fund Denmark (No 12-125524 (Sapere Aude Advanced grant) and No 2034-00335B), the Danish National Research Foundation (DNRF grant No 141) to the Center for Functional Genomics and Tissue Plasticity (ATLAS), grants from the Novo Nordisk Foundation (NNF21OC0071613; the NNF BridgeheadCenter for Basic Mechanisms of Disease; and NNF18OC0033444 to the Challenge Grant Center for Adipocyte Signaling (ADIPOSIGN) the Lundbeck Foundation (grant No R218-2016-1450), the European Union's Horizon 2020 research and innovation programme (grant agreement No 860002) to the ENHPATHY Consortium, the Novo Scholarship Programme, the Fuhrmann Foundation, and the Villum Foundation through support to the Villum Center for Bioanalytical Sciences. We thank Q. Yan from Yale School of Medicine for providing the Lenti-iCas9-neo plasmid and A. M. Oudelaar from the University of Oxford for valuable clarifications on Capture-C protocols and data analysis. We are grateful to our colleagues in the Functional Genomics and Metabolism Research Unit, as well as Eileen Furlong and Tim Pollex, for critical discussions that helped improve the study. Author contributions Conceptualization: A.C., M.H., H.D., A.R., J.G.S.M., M.C., S.M.; Experimental work: A.C., M.H., M.N., L.K.H., K.M., M.S.M., V.S., E.D.M.; Formal analysis, investigation and data curation: A.C, M.H., M.N., B.V.H, H.D., and J.G.S.M.; Visualization: A.C., M.H.; Writing: A.C., M.H., S.M.; Funding acquisition: S.M.; Supervision: S.M., ., A.R., J.G.S.M., M.C. Declaration of interest The authors declare no competing interests. Data availability All generated sequence data will be available at the GEO repository under accession code X. In addition to the data generated for this study, raw sequencing data from RNA-sequencing, DNase-sequencing and MED1 ChIP-sequencing are available from GEO under accession code GSE113253, and raw sequencing data from ECHi-C from GEO under accession code GSE140782. Raw data used for Hi-C analysis are available from GEO under accession code GSE140782 and GSE52457. References Amemiya, H. M., Kundaje, A., & Boyle, A. P. (2019). The ENCODE Blacklist: Identification of Problematic Regions of the Genome. Sci Rep , 9 (1), 9354. https://doi.org/10.1038/s41598-019-45839-z Avsec, Ž., Agarwal, V., Visentin, D., Ledsam, J. R., Grabska-Barwinska, A., Taylor, K. R., Assael, Y., Jumper, J., Kohli, P., & Kelley, D. R. (2021). Effective gene expression prediction from sequence by integrating long-range interactions. Nature Methods , 18 (10), 1196-1203. https://doi.org/10.1038/s41592-021-01252-x Avsec, Ž., Weilert, M., Shrikumar, A., Krueger, S., Alexandari, A., Dalal, K., Fropf, R., McAnany, C., Gagneur, J., Kundaje, A., & Zeitlinger, J. (2021). Base-resolution models of transcription-factor binding reveal soft motif syntax. Nature Genetics , 53 (3), 354-366. https://doi.org/10.1038/s41588-021-00782-6 Avsec, Ž., Weilert, M., Shrikumar, A., Krueger, S., Alexandari, A., Dalal, K., Fropf, R., McAnany, C., Gagneur, J., Kundaje, A., & Zeitlinger, J. (2021). Base-resolution models of transcription-factor binding reveal soft motif syntax. Nature Genetics , 53 (3), 354-366. https://doi.org/10.1038/s41588-021-00782-6 Bahr, C., Von Paleske, L., Uslu, V. V., Remeseiro, S., Takayama, N., Ng, S. W., Murison, A., Langenfeld, K., Petretich, M., Scognamiglio, R., Zeisberger, P., Benk, A. S., Amit, I., Zandstra, P. W., Lupien, M., Dick, J. E., Trumpp, A., & Spitz, F. (2018). A Myc enhancer cluster regulates normal and leukaemic haematopoietic stem cell hierarchies. Nature , 553 (7689), 515-520. https://doi.org/10.1038/nature25193 Bhimsaria, D., Rodríguez-Martínez, J. A., Mendez-Johnson, J. L., Ghoshdastidar, D., Varadarajan, A., Bansal, M., Daniels, D. L., Ramanathan, P., & Ansari, A. Z. (2023). Hidden modes of DNA binding by human nuclear receptors. Nature communications , 14 (1). https://doi.org/10.1038/s41467-023-39577-0 Blayney, J. W., Francis, H., Rampasekova, A., Camellato, B., Mitchell, L., Stolper, R., Cornell, L., Babbs, C., Boeke, J. D., Higgs, D. R., & Kassouf, M. (2023). Super-enhancers include classical enhancers and facilitators to fully activate gene expression. Cell , 186 (26), 5826-5839.e5818. https://doi.org/10.1016/j.cell.2023.11.030 Bonev, B., Mendelson Cohen, N., Szabo, Q., Fritsch, L., Papadopoulos, G. L., Lubling, Y., Xu, X., Lv, X., Hugnot, J.-P., Tanay, A., & Cavalli, G. (2017). Multiscale 3D Genome Rewiring during Mouse Neural Development. Cell , 171 (3), 557-572.e524. https://doi.org/10.1016/j.cell.2017.09.043 Broekema, M. F., Savage, D. B., Monajemi, H., & Kalkhoven, E. (2019). Gene-gene and gene-environment interactions in lipodystrophy: Lessons learned from natural PPARγ mutants. Biochim Biophys Acta Mol Cell Biol Lipids , 1864 (5), 715-732. https://doi.org/10.1016/j.bbalip.2019.02.002 Bugge, A., Grøntved, L., Aagaard, M. M., Borup, R., & Mandrup, S. (2009). The PPARgamma2 A/B-domain plays a gene-specific role in transactivation and cofactor recruitment. Molecular Endocrinology , 23 (6), 794-808. https://doi.org/10.1210/me.2008-0236 Cao, J., Wu, L., Zhang, S.-M., Lu, M., Cheung, W. K. C., Cai, W., Gale, M., Xu, Q., & Yan, Q. (2016). An easy and efficient inducible CRISPR/Cas9 platform with improved specificity for multiple gene targeting. Nucleic Acids Research , gkw660. https://doi.org/10.1093/nar/gkw660 Chen, Z., Snetkova, V., Bower, G., Jacinto, S., Clock, B., Dizehchi, A., Barozzi, I., Mannion, B. J., Alcaina-Caro, A., Lopez-Rios, J., Dickel, D. E., Visel, A., Pennacchio, L. A., & Kvon, E. Z. (2024). Increased enhancer–promoter interactions during developmental enhancer activation in mammals. Nature Genetics . https://doi.org/10.1038/s41588-024-01681-2 Cho, W.-K., Spille, J.-H., Hecht, M., Lee, C., Li, C., Grube, V., & Cisse, I. I. (2018). Mediator and RNA polymerase II clusters associate in transcription-dependent condensates. Science , 361 (6400), 412-415. https://doi.org/10.1126/science.aar4199 Choi, J., Lysakovskaia, K., Stik, G., Demel, C., Söding, J., Tian, T. V., Graf, T., & Cramer, P. (2021). Evidence for additive and synergistic action of mammalian enhancers during cell fate determination. eLife , 10 . https://doi.org/10.7554/eLife.65381 Costanzo, M. C., Von Grotthuss, M., Massung, J., Jang, D., Caulkins, L., Koesterer, R., Gilbert, C., Welch, R. P., Kudtarkar, P., Hoang, Q., Boughton, A. P., Singh, P., Sun, Y., Duby, M., Moriondo, A., Nguyen, T., Smadbeck, P., Alexander, B. R., Brandes, M., . . . Zaghloul, N. A. (2023). The Type 2 Diabetes Knowledge Portal: An open access genetic resource dedicated to type 2 diabetes and related traits. Cell Metabolism , 35 (4), 695-710.e696. https://doi.org/10.1016/j.cmet.2023.03.001 Danecek, P., Bonfield, J. K., Liddle, J., Marshall, J., Ohan, V., Pollard, M. O., Whitwham, A., Keane, T., McCarthy, S. A., Davies, R. M., & Li, H. (2021). Twelve years of SAMtools and BCFtools. GigaScience , 10 (2). https://doi.org/10.1093/gigascience/giab008 Di Giammartino, D. C., Kloetgen, A., Polyzos, A., Liu, Y., Kim, D., Murphy, D., Abuhashem, A., Cavaliere, P., Aronson, B., Shah, V., Dephoure, N., Stadtfeld, M., Tsirigos, A., & Apostolou, E. (2019). KLF4 is involved in the organization and regulation of pluripotency-associated three-dimensional enhancer networks. Nature Cell Biology , 21 (10), 1179-1190. https://doi.org/10.1038/s41556-019-0390-6 Dittmar, G., Hernandez, D. P., Kowenz-Leutz, E., Kirchner, M., Kahlert, G., Wesolowski, R., Baum, K., Knoblich, M., Hofstätter, M., Muller, A., Wolf, J., Reimer, U., & Leutz, A. (2019). PRISMA: Protein Interaction Screen on Peptide Matrix Reveals Interaction Footprints and Modifications- Dependent Interactome of Intrinsically Disordered C/EBPβ. iScience , 13 , 351-370. https://doi.org/10.1016/j.isci.2019.02.026 Dixon, J. R., Jung, I., Selvaraj, S., Shen, Y., Antosiewicz-Bourget, J. E., Lee, A. Y., Ye, Z., Kim, A., Rajagopal, N., Xie, W., Diao, Y., Liang, J., Zhao, H., Lobanenkov, V. V., Ecker, J. R., Thomson, J. A., & Ren, B. (2015). Chromatin architecture reorganization during stem cell differentiation. Nature , 518 (7539), 331-336. https://doi.org/10.1038/nature14222 Dobin, A., Davis, C. A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S., Batut, P., Chaisson, M., & Gingeras, T. R. (2013). STAR: ultrafast universal RNA-seq aligner. Bioinformatics , 29 (1), 15-21. https://doi.org/10.1093/bioinformatics/bts635 Du, M., Stitzinger, S. H., Spille, J.-H., Cho, W.-K., Lee, C., Hijaz, M., Quintana, A., & Cissé, I. I. (2024). Direct observation of a condensate effect on super-enhancer controlled gene bursting. Cell , 187 (2), 331-344.e317. https://doi.org/10.1016/j.cell.2023.12.005 Fiorenza, C. G., Chou, S. H., & Mantzoros, C. S. (2011). Lipodystrophy: pathophysiology and advances in treatment. Nature Reviews Endocrinology , 7 (3), 137-150. https://doi.org/10.1038/nrendo.2010.199 Freire-Pritchett, P., Schoenfelder, S., Várnai, C., Wingett, S. W., Cairns, J., Collier, A. J., García-Vílchez, R., Furlan-Magaril, M., Osborne, C. S., Fraser, P., Rugg-Gunn, P. J., & Spivakov, M. (2017). Global reorganisation of cis-regulatory units upon lineage commitment of human embryonic stem cells. eLife , 6 . https://doi.org/10.7554/elife.21926 Gaulton, K. J., Ferreira, T., Lee, Y., Raimondo, A., Mägi, R., Reschen, M. E., Mahajan, A., Locke, A., William Rayner, N., Robertson, N., Scott, R. A., Prokopenko, I., Scott, L. J., Green, T., Sparso, T., Thuillier, D., Yengo, L., Grallert, H., Wahl, S., . . . Morris, A. P. (2015). Genetic fine mapping and genomic annotation defines causal mechanisms at type 2 diabetes susceptibility loci. Nature Genetics , 47 (12), 1415-1425. https://doi.org/10.1038/ng.3437 Ghaben, A. L., & Scherer, P. E. (2019). Adipogenesis and metabolic health. Nature Reviews Molecular Cell Biology , 20 (4), 242-258. https://doi.org/10.1038/s41580-018-0093-z Grubert, F., Judith, Kasowski, M., Ursu, O., Damek, Alicia, Greenside, P., Srivas, R., Doug, Pekowska, A., Heidari, N., Euskirchen, G., Huber, W., Jonathan, Carlos, Lars, Kundaje, A., & Snyder, M. (2015). Genetic Control of Chromatin States in Humans Involves Local and Distal Chromosomal Interactions. Cell , 162 (5), 1051-1065. https://doi.org/10.1016/j.cell.2015.07.048 Gupta, S., Stamatoyannopoulos, J. A., Bailey, T. L., & Noble, W. S. (2007). Quantifying similarity between motifs. Genome Biol , 8 (2), R24. https://doi.org/10.1186/gb-2007-8-2-r24 Haakonsson, A. K., Stahl Madsen, M., Nielsen, R., Sandelin, A., & Mandrup, S. (2013). Acute Genome-Wide Effects of Rosiglitazone on PPARγ Transcriptional Networks in Adipocytes. Molecular Endocrinology , 27 (9), 1536-1549. https://doi.org/10.1210/me.2013-1080 Hagberg, C. E., & Spalding, K. L. (2024). White adipocyte dysfunction and obesity-associated pathologies in humans. Nat Rev Mol Cell Biol , 25 (4), 270-289. https://doi.org/10.1038/s41580-023-00680-1 Huang, J., Li, K., Cai, W., Liu, X., Zhang, Y., Orkin, S. H., Xu, J., & Yuan, G.-C. (2018). Dissecting super-enhancer hierarchy based on chromatin interactions. Nature communications , 9 (1). https://doi.org/10.1038/s41467-018-03279-9 Lee, H., Qian, K., Christine, Hoerburger, L., Claussnitzer, M., Hoffmann, C., Glunk, V., Wahl, S., Breier, M., Eck, F., Jafari, L., Molnos, S., Grallert, H., Dahlman, I., Arner, P., Brunner, C., Hauner, H., Hauck, S. M., & Laumen, H. (2017). Allele-specific quantitative proteomics unravels molecular mechanisms modulated by cis-regulatory PPARG locus variation. Nucleic Acids Research , 45 (6), 3266-3279. https://doi.org/10.1093/nar/gkx105 Lefterova, M. I., Haakonsson, A. K., Lazar, M. A., & Mandrup, S. (2014). PPARγ and the global map of adipogenesis and beyond. Trends in Endocrinology & Metabolism , 25 (6), 293-302. https://doi.org/10.1016/j.tem.2014.04.001 Lefterova, M. I., Zhang, Y., Steger, D. J., Schupp, M., Schug, J., Cristancho, A., Feng, D., Zhuo, D., Stoeckert, C. J., Liu, X. S., & Lazar, M. A. (2008). PPARγ and C/EBP factors orchestrate adipocyte biology via adjacent binding on a genome-wide scale. Genes & Development , 22 (21), 2941-2952. https://doi.org/10.1101/gad.1709008 Lopez-Delisle, L., Rabbani, L., Wolff, J., Bhardwaj, V., Backofen, R., Grüning, B., Ramírez, F., & Manke, T. (2021). pyGenomeTracks: reproducible plots for multivariate genomic datasets. Bioinformatics , 37 (3), 422-423. https://doi.org/10.1093/bioinformatics/btaa692 Love, M. I., Huber, W., & Anders, S. (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology , 15 (12). https://doi.org/10.1186/s13059-014-0550-8 Lovén, J., Heather, Charles, Lau, A., David, Christopher, James, Tong, & Richard. (2013). Selective Inhibition of Tumor Oncogenes by Disruption of Super-Enhancers. Cell , 153 (2), 320-334. https://doi.org/10.1016/j.cell.2013.03.036 Lyons, H., Veettil, R. T., Pradhan, P., Fornero, C., De La Cruz, N., Ito, K., Eppert, M., Roeder, R. G., & Sabari, B. R. (2023). Functional partitioning of transcriptional regulators by patterned charge blocks. Cell , 186 (2), 327-345.e328. https://doi.org/10.1016/j.cell.2022.12.013 Madsen, J. G. S., Madsen, M. S., Rauch, A., Traynor, S., Van Hauwaert, E. L., Haakonsson, A. K., Javierre, B. M., Hyldahl, M., Fraser, P., & Mandrup, S. (2020). Highly interconnected enhancer communities control lineage-determining genes in human mesenchymal stem cells. Nature Genetics , 52 (11), 1227-1238. https://doi.org/10.1038/s41588-020-0709-z Madsen, J. G. S., Rauch, A., Van Hauwaert, E. L., Schmidt, S. F., Winnefeld, M., & Mandrup, S. (2018). Integrated analysis of motif activity and gene expression changes of transcription factors. Genome research , 28 (2), 243-255. https://doi.org/10.1101/gr.227231.117 Madsen, M. S., Broekema, M. F., Madsen, M. R., Koppen, A., Borgman, A., Gräwe, C., Thomsen, E. G. K., Westland, D., Kranendonk, M. E. G., Koerkamp, M. G., Hamers, N., Bonvin, A. M. J. J., Pittol, J. M. R., Natarajan, K. N., Kersten, S., Holstege, F. C. P., Monajemi, H., van Mil, S. W. C., Vermeulen, M., . . . Kalkhoven, E. (2022a). PPARγ lipodystrophy mutants reveal intermolecular interactions required for enhancer activation. Nature Communications , 13 (1), 7090. https://doi.org/10.1038/s41467-022-34766-9 Madsen, M. S., Broekema, M. F., Madsen, M. R., Koppen, A., Borgman, A., Gräwe, C., Thomsen, E. G. K., Westland, D., Kranendonk, M. E. G., Koerkamp, M. G., Hamers, N., Bonvin, A. M. J. J., Pittol, J. M. R., Natarajan, K. N., Kersten, S., Holstege, F. C. P., Monajemi, H., Van Mil, S. W. C., Vermeulen, M., . . . Kalkhoven, E. (2022b). PPARγ lipodystrophy mutants reveal intermolecular interactions required for enhancer activation. Nature communications , 13 (1). https://doi.org/10.1038/s41467-022-34766-9 Mahajan, A., Go, M. J., Zhang, W., Below, J. E., Gaulton, K. J., Ferreira, T., Horikoshi, M., Johnson, A. D., Ng, M. C. Y., Prokopenko, I., Saleheen, D., Wang, X., Zeggini, E., Abecasis, G. R., Adair, L. S., Almgren, P., Atalay, M., Aung, T., Baldassarre, D., . . . Morris, A. P. (2014). Genome-wide trans-ancestry meta-analysis provides insight into the genetic architecture of type 2 diabetes susceptibility. Nature Genetics , 46 (3), 234-244. https://doi.org/10.1038/ng.2897 Mahajan, A., Spracklen, C. N., Zhang, W., Ng, M. C. Y., Petty, L. E., Kitajima, H., Yu, G. Z., Rüeger, S., Speidel, L., Kim, Y. J., Horikoshi, M., Mercader, J. M., Taliun, D., Moon, S., Kwak, S.-H., Robertson, N. R., Rayner, N. W., Loh, M., Kim, B.-J., . . . Morris, A. P. (2022). Multi-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translation. Nature Genetics , 54 (5), 560-572. https://doi.org/10.1038/s41588-022-01058-3 Mahajan, A., Taliun, D., Thurner, M., Robertson, N. R., Torres, J. M., Rayner, N. W., Payne, A. J., Steinthorsdottir, V., Scott, R. A., Grarup, N., Cook, J. P., Schmidt, E. M., Wuttke, M., Sarnowski, C., Mägi, R., Nano, J., Gieger, C., Trompet, S., Lecoeur, C., . . . McCarthy, M. I. (2018). Fine-mapping type 2 diabetes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps. Nature Genetics , 50 (11), 1505-1513. https://doi.org/10.1038/s41588-018-0241-6 Murphy, D., Salataj, E., Di Giammartino, D. C., Rodriguez-Hernaez, J., Kloetgen, A., Garg, V., Char, E., Uyehara, C. M., Ee, L.-S., Lee, U., Stadtfeld, M., Hadjantonakis, A.-K., Tsirigos, A., Polyzos, A., & Apostolou, E. (2024). 3D Enhancer–promoter networks provide predictive features for gene expression and coregulation in early embryonic lineages. Nature Structural & Molecular Biology , 31 (1), 125-140. https://doi.org/10.1038/s41594-023-01130-4 Nielsen, R., Grøntved, L., Stunnenberg, H. G., & Mandrup, S. (2006). Peroxisome proliferator-activated receptor subtype- and cell-type-specific activation of genomic target genes upon adenoviral transgene delivery. Molecular and Cellular Biology , 26 (15), 5698-5714. https://doi.org/10.1128/mcb.02266-05 Oudelaar, A. M., Davies, J. O. J., Hanssen, L. L. P., Telenius, J. M., Schwessinger, R., Liu, Y., Brown, J. M., Downes, D. J., Chiariello, A. M., Bianco, S., Nicodemi, M., Buckle, V. J., Dekker, J., Higgs, D. R., & Hughes, J. R. (2018). Single-allele chromatin interactions identify regulatory hubs in dynamic compartmentalized domains. Nature Genetics , 50 (12), 1744-1751. https://doi.org/10.1038/s41588-018-0253-2 Pollex, T., Rabinowitz, A., Gambetta, M. C., Marco-Ferreres, R., Viales, R. R., Jankowski, A., Schaub, C., & Furlong, E. E. M. (2024). Enhancer–promoter interactions become more instructive in the transition from cell-fate specification to tissue differentiation. Nature Genetics . https://doi.org/10.1038/s41588-024-01678-x Quinlan, A. R., & Hall, I. M. (2010). BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics , 26 (6), 841-842. https://doi.org/10.1093/bioinformatics/btq033 Rada-Iglesias, A., Bajpai, R., Swigut, T., Brugmann, S. A., Flynn, R. A., & Wysocka, J. (2011). A unique chromatin signature uncovers early developmental enhancers in humans. Nature , 470 (7333), 279-283. https://doi.org/10.1038/nature09692 Ramírez, F., Dündar, F., Diehl, S., Grüning, B. A., & Manke, T. (2014). deepTools: a flexible platform for exploring deep-sequencing data. Nucleic Acids Research , 42 (W1), W187-W191. https://doi.org/10.1093/nar/gku365 Rauch, A., Haakonsson, A. K., Madsen, J. G. S., Larsen, M., Forss, I., Madsen, M. R., Van Hauwaert, E. L., Wiwie, C., Jespersen, N. Z., Tencerova, M., Nielsen, R., Larsen, B. D., Röttger, R., Baumbach, J., Scheele, C., Kassem, M., & Mandrup, S. (2019). Osteogenesis depends on commissioning of a network of stem cell transcription factors that act as repressors of adipogenesis. Nature Genetics , 51 (4), 716-727. https://doi.org/10.1038/s41588-019-0359-1 Rauch, A., & Mandrup, S. (2021). Transcriptional networks controlling stromal cell differentiation. Nature Reviews Molecular Cell Biology , 22 (7), 465-482. https://doi.org/10.1038/s41580-021-00357-7 Rauluseviciute, I., Riudavets-Puig, R., Blanc-Mathieu, R., Castro-Mondragon, J. A., Ferenc, K., Kumar, V., Lemma, R. B., Lucas, J., Chèneby, J., Baranasic, D., Khan, A., Fornes, O., Gundersen, S., Johansen, M., Hovig, E., Lenhard, B., Sandelin, A., Wasserman, W. W., Parcy, F., & Mathelier, A. (2024). JASPAR 2024: 20th anniversary of the open-access database of transcription factor binding profiles. Nucleic Acids Research , 52 (D1), D174-d182. https://doi.org/10.1093/nar/gkad1059 Ross-Innes, C. S., Stark, R., Teschendorff, A. E., Holmes, K. A., Ali, H. R., Dunning, M. J., Brown, G. D., Gojis, O., Ellis, I. O., Green, A. R., Ali, S., Chin, S.-F., Palmieri, C., Caldas, C., & Carroll, J. S. (2012). Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. Nature , 481 (7381), 389-393. https://doi.org/10.1038/nature10730 Sabari, B. R., Dall’Agnese, A., Boija, A., Klein, I. A., Coffey, E. L., Shrinivas, K., Abraham, B. J., Hannett, N. M., Zamudio, A. V., Manteiga, J. C., Li, C. H., Guo, Y. E., Day, D. S., Schuijers, J., Vasile, E., Malik, S., Hnisz, D., Lee, T. I., Cisse, I. I., . . . Young, R. A. (2018). Coactivator condensation at super-enhancers links phase separation and gene control. Science , 361 (6400), eaar3958. https://doi.org/10.1126/science.aar3958 Saint-André, V., Federation, A. J., Lin, C. Y., Abraham, B. J., Reddy, J., Lee, T. I., Bradner, J. E., & Young, R. A. (2016). Models of human core transcriptional regulatory circuitries. Genome Research , 26 (3), 385-396. https://doi.org/10.1101/gr.197590.115 Salma, N., Xiao, H., & Imbalzano, A. N. (2006). Temporal recruitment of CCAAT/enhancer-binding proteins to early and late adipogenic promoters in vivo. J Mol Endocrinol , 36 (1), 139-151. https://doi.org/10.1677/jme.1.01918 Schmidt, S. F., Jørgensen, M., Chen, Y., Nielsen, R., Sandelin, A., & Mandrup, S. (2011). Cross species comparison of C/EBPα and PPARγ profiles in mouse and human adipocytes reveals interdependent retention of binding sites. BMC Genomics , 12 (1), 152. https://doi.org/10.1186/1471-2164-12-152 Shrikumar, A. T., K. Avsec, Ž. Shcherbina, A. Banerjee, A. Sharmin, M. Nair, S. Kundaje, A. (2020). Technical Note on Transcription Factor Motif Discovery from Importance Scores (TF-MoDISco) version 0.5.6.5. Preprint arXiv . Siersbæk, R., Madsen, J. G. S., Javierre, B. M., Nielsen, R., Bagge, E. K., Cairns, J., Wingett, S. W., Traynor, S., Spivakov, M., Fraser, P., & Mandrup, S. (2017). Dynamic Rewiring of Promoter-Anchored Chromatin Loops during Adipocyte Differentiation. Molecular Cell , 66 (3), 420-435.e425. https://doi.org/10.1016/j.molcel.2017.04.010 Siersbaek, R., Nielsen, R., John, S., Sung, M.-H., Baek, S., Loft, A., Hager, G. L., & Mandrup, S. (2011). Extensive chromatin remodelling and establishment of transcription factor ‘hotspots’ during early adipogenesis. The EMBO Journal , 30 (8), 1459-1472. https://doi.org/10.1038/emboj.2011.65 Siersbaek, R., Nielsen, R., & Mandrup, S. (2012). Transcriptional networks and chromatin remodeling controlling adipogenesis. Trends in Endocrinology & Metabolism , 23 (2), 56-64. http://www.sciencedirect.com/science/article/pii/S1043276011001640 (Not in File) Siersbæk, R., Rabiee, A., Nielsen, R., Sidoli, S., Traynor, S., Loft, A., Lars, Rogowska-Wrzesinska, A., Ole, & Mandrup, S. (2014). Transcription Factor Cooperativity in Early Adipogenic Hotspots and Super-Enhancers. Cell Reports , 7 (5), 1443-1455. https://doi.org/10.1016/j.celrep.2014.04.042 Soccio, R. E., Chen, E. R., Rajapurkar, S. R., Safabakhsh, P., Marinis, J. M., Dispirito, J. R., Emmett, M. J., Briggs, E. R., Fang, B., Everett, L. J., Lim, H. W., Won, K. J., Steger, D. J., Wu, Y., Civelek, M., Voight, B. F., & Lazar, M. A. (2015). Genetic Variation Determines PPARγ Function and Anti-diabetic Drug Response In Vivo. Cell , 162 (1), 33-44. https://doi.org/10.1016/j.cell.2015.06.025 Tanaka, T. (1997). Defective adipocyte differentiation in mice lacking the C/EBPbeta and/or C/EBPdelta gene. The EMBO Journal , 16 (24), 7432-7443. https://doi.org/10.1093/emboj/16.24.7432 Thomas, H. F., Kotova, E., Jayaram, S., Pilz, A., Romeike, M., Lackner, A., Penz, T., Bock, C., Leeb, M., Halbritter, F., Wysocka, J., & Buecker, C. (2021). Temporal dissection of an enhancer cluster reveals distinct temporal and functional contributions of individual elements. Molecular Cell , 81 (5), 969-982.e913. https://doi.org/10.1016/j.molcel.2020.12.047 Trojanowski, J., Frank, L., Rademacher, A., Mücke, N., Grigaitis, P., & Rippe, K. (2022). Transcription activation is enhanced by multivalent interactions independent of phase separation. Molecular Cell , 82 (10), 1878-1893.e1810. https://doi.org/10.1016/j.molcel.2022.04.017 Wang, A., Yue, F., Li, Y., Xie, R., Harper, T., Nisha, Muth, K., Palmer, J., Qiu, Y., Wang, J., Dieter, Jeffrey, Doris, Ren, B., & Sander, M. (2015). Epigenetic Priming of Enhancers Predicts Developmental Competence of hESC-Derived Endodermal Lineage Intermediates. Cell Stem Cell , 16 (4), 386-399. https://doi.org/10.1016/j.stem.2015.02.013 Ward, L. D., & Kellis, M. (2012). HaploReg: a resource for exploring chromatin states, conservation, and regulatory motif alterations within sets of genetically linked variants. Nucleic Acids Research , 40 (D1), D930-D934. https://doi.org/10.1093/nar/gkr917 Warren, David, Hnisz, D., Brian, Charles, Michael, Peter, Tong, & Richard. (2013). Master Transcription Factors and Mediator Establish Super-Enhancers at Key Cell Identity Genes. Cell , 153 (2), 307-319. https://doi.org/10.1016/j.cell.2013.03.035 Xiaotao, W. ( 2016). unHiC: a user-friendly Hi-C data processing software based on hiclib. https://doi.org/10.5281/zenodo.55324 . Zhang, Y., Liu, T., Meyer, C. A., Eeckhoute, J., Johnson, D. S., Bernstein, B. E., Nusbaum, C., Myers, R. M., Brown, M., Li, W., & Liu, X. S. (2008). Model-based Analysis of ChIP-Seq (MACS). Genome Biology , 9 (9), R137. https://doi.org/10.1186/gb-2008-9-9-r137 Methods Cell culture and differentiation Telomerase-immortalized human bone marrow-derived stromal progenitor cells (hMSC-TERT4) were kindly provided by Moustapha Kassem (Abdallah et al., 2004; Simonsen et al., 2002)), and maintained in α-MEM with 10 % fetal bovine serum (FBS) and 1% penicillin-streptomycin (P/S) (growth medium) at 37°C, 95% humidity and 5% carbon dioxide. Osteo- or adipogenic differentiation was induced two days post confluence (day 0) by changing the growth medium to the differentiation medium. Osteogenic (α-MEM with 10% FBS, 1% P/S and 10 nM freshly added vitamin D3 in 96% ethanol, 50 μg/mL L-ascorbic acid in MilliQ water, 10 mM β-glycerophosphate in MilliQ water and 10 nM dexamethasone) or adipogenic differentiation medium (DMEM with 10% FBS, 1% P/S and 100nM freshly added dexamethasone in 96% ethanol, 0.5 mM 3-isobutyl-1-methylxanthine in 0.1 M KOH, 2 μM human insulin, and 1 μM rosiglitazone in DMSO) was renewed every second day until cells were harvested. HEK293T cells were maintained in DMEM with 10% FBS and 1% P/S (HEK growth medium) at 37°C, 95% humidity and 5% carbon dioxide. Production of the hMSC-TERT4/iCas9-4 cell line The Lenti-iCas9-Neo plasmid (Addgene #85400, gift from Qin Yan (Cao et al., 2016)) was used to transform competent Escherichia coli DH5α and purified in parallel with psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259). Integrity was evaluated by linearization (using appropriate restriction enzymes for each plasmid) followed by gel electrophoresis. Lenti-iCas9-neo virus was produced in HEK293T cells and used to transduce hMSC-TERT4. Medium was changed to growth medium with 0.5 μg/mL doxycycline 34 hours after transduction to induce GFP and Cas9 expression. After 24 hours, GFP-positive cells were selected by fluorescence-activated cell sorting (FACS) using FACS Aria III (BD Biosciences) and sorted into and maintained in 37°C growth medium. Data analysis was performed with BD FACSDiva software. The initial GFP-positive population was called hMSC-TERT4/iCas9. Single-cell clones of this population were generated and the line hMSC-TERT4/iCas9-4 was selected for further studies. Quantification of incorporated virus copy number Genomic DNA was used for qPCR-based determination of lentiviral transfer sequence copy number in hMSC-TERT4/iCas9-4 cells. Cells were harvested in cold lysis buffer (0.1% SDS, 1% Triton X-100, 0.15 M NaCl, 1 mM EDTA, 20 mM Tris pH 8) and stored at -80 °C prior to sonication. Entire cell lysates were sonicated in 130 μL microTUBE AFA fiber Pre-Slit Snap-Cap 6x16 mm tubes (Covaris) on an ME220 Focused-ultrasonicator (Covaris) for 20 s, peak power 75, duty% factor 26.67, cycles/burst 200. Shreded DNA was purified and used as a template for each qPCR reaction with forward and reverse primer targeting the long terminal repeat (LTR) of the transfer sequence and a genomic region (LTR and NoGene2 primer, respectively, Supplementary Information Table 2) and FastStart Essential DNA Green Master (Roche) on a Lightcycler 480 II (Roche). Incorporated virus copy number was calculated as: Since both primers amplify two target regions per viral transfer sequence and per genome, respectively. Genome editing in hMSC-TERT4/iCas9-4 To design spacer sequences for guide RNAs (gRNAs), enhancer sequences were extracted from the UCSC Genome Browser (Kent et al., 2002, genome.ucsc.edu, hg19 assembly) based on the location and width of peaks from DNase-sequencing of hMSC-TERT4 cells during osteogenic and adipogenic differentiation (Rauch et al., 2019). Spacer sequences were found using the online tool sgRNA Designer (portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design (Doench et al., 2014, 2016)), selected by on-target and off-target rank (higher rank is predicted to be more active and specific) and ordered as default Alt-R CRISPR-Cas9 crRNAs (Integrated DNA Technologies). For spacer sequences see Table S6. tracrRNA for forming tracrRNA:crRNA duplexes, i.e. gRNAs, was also ordered from Integrated DNA Technologies. hMSC-TERT4/iCas9-4 cells were induced with 36 μg doxycycline in MilliQ water per 25 mL medium at 80-90% confluence. After 6 hours, cells were reverse transfected in a 12-well plate with a total of 12 pmol crRNA:tracrRNA duplexes (i.e. gRNA), 2 μL DharmaFECT 1 (Thermo Scientific) and 25,000cells/cm 2 per well. Cells were kept in transfection medium (1/3 Opti-MEM, 2/3 α-MEM with 10 % FBS) for 2 days at 37 °C, 95% humidity, and 5% carbon dioxide, and seeded as single cells in 96-well plates (0.5 cells/100 μL medium). Clonal enhancer-deletion cell lines were then maintained in growth medium. Two controls were included: ‘No gRNA’, which received the same treatment as all other cell lines, except the transfection complexes contained no gRNA; ‘Ctr gRNA’, which received a pair of gRNAs mediating a deletion in intron 1 of PPP1R12C (adeno-associated virus “safe harbor” site, DeKelver et al., 2010) . For screening of enhancer-deleted clones, genomic DNA was harvested in DNeasy Blood and Tissue Kit for DNA isolation (Qiagen), and DNA was extracted according to the manufacturer’s instructions. Target regions in the deleted regions were amplified by end-point PCR using Phusion High-Fidelity DNA polymerase (NEB) (primers in Supplementary Information Table 3), and the presence of PCR product was evaluated on a Fragment Analyzer (Advanced Analytical Technologies). For each enhancer deletion, three independent clones lacking the wild-type PCR product were used to assess the size of the deleted region by PCR analysis of genomic region, followed by purification of PCR products with QIAquick PCR Purification Kit (Qiagen), and Sanger sequencing (primers in Supplementary Information Table 3, deletion size in Extended Data Table 3, and Supplementary Information Fig. 2). Oil-Red-O and Alizarin Red S staining Cells were fixed in 4% formaldehyde in PBS for 15 min. For Oil-red-O staining of lipids, cells were incubated with 0.3% Oil-red-O in 60% isopropanol (filtered) at room temperature in the dark for 30 min. For Alizarin Red S staining of calcified matrix, cells were incubated with 1% Alizarin Red S in MilliQ water, pH adjusted to 4-4.2 with HCl and NH 4 OH. Cells were washed five times with water and subjected to macroscopic and microscopic photography. Western blotting Cell lysates were harvested from hMSC-TERT4 after 0 or 7 days of adipogenic differentiation, or hMSC-TERT4/iCas9-4 cells treated with 0.5 μg/mL doxycycline for 6 hours, in lysis buffer (50 mM Tris-HCl pH 6.8, 10% glycerol, 2.5% SDS, 10 mM β-glycerophosphate, 10 mM NaF, 0.1 mM sodium orthovanadate, 1 mM PMSF (added fresh), 1x Complete (Roche, added fresh), boiled and treated with benzonase. Protein concentration was determined using the Pierce BCA Protein Assay Kit (Thermo Scientific) in accordance with the manufacturer’s instructions. Dithioeythritol (DTE) was then added to 0.01 M. For Western blotting, 10-40 μg protein sample was used per well in a 10% gel (separation: 10% acrylamide/bis-acrylamide solution, 0.4 M Tris-HCl pH 8.8, 0.1% SDS, 0.1% ammonium persulfate in MilliQ (APS), 0.001% tetramethylethylenediamine (TEMED); stacking: 6% acrylamide/bis-acrylamide solution, 0.125 M Tris-HCl pH 6.8, 0.1% SDS, 0.1% APS, 0.001% TEMED). For specific protein detection, the wet-blotted poly-vinyl-difluoride membrane was incubated for 1 hour at room temperature or overnight at 4°C with primary antibody: rabbit-anti-Lamin A (1:10,000, L1293, Sigma-Aldrich), mouse-anti-FLAG (1:500, F1804-1MG, Sigma-Aldrich), rabbit-anti-PPARG (1:1000, 2443S, Cell Signaling) in PBS with 1.5% BSA, and then 1 hour at room temperature with secondary antibody: goat-anti-rabbit (1:1,000, P0448, DAKO) or goat-anti-mouse (1:2,000, P0447, DAKO) in PBS with 1.5% BSA. Enhanced chemiluminescence detection was performed using Immobilon Classico Western HRP substrate (Merck millipore), and either exposing the membrane to an X-ray film, or imaging using the Amersham Imager 680. Quantification of lipid droplet area Lipid accumulation was imaged using Nikon Eclipse Ts2-FL with differential interference contrast and 40x total magnification. Lipid droplet area was quantified using ImageJ (Rueden et al., 2017), by thresholding on the background subtracted greyscale images (using identical thresholds for all images) and quantifying the area in pixels that reached the threshold. Analysis of gene expression using qPCR hMSC-TERT4/iCas9-4 enhancer deletion and control cell lines (three clones for each condition) were harvested using Isol-RNA Lysis Reagent (5PRIME) after 10 days of adipogenic differentiation. RNA was purified using phenol-chloroform extraction and EconoSpin columns (Epoch Life Sciences). For cDNA synthesis, up to 1000 ng RNA (always equal amounts in replicates) was treated with DNase I and reverse-transcribed using Moloney murine leukemia virus reverse transcriptase (Invitrogen). cDNA was used for qPCR with FastStart Essential DNA Green Master (Roche) on a Lightcycler 480 II (Roche) (primers Supplementary Information Table 2). Data for each sample were normalized to the level of transcript of TATA-box protein II B ( TBP ). Gene expression analysis using RNA-seq, and enhancer activity analysis using DNase-seq and MED1 ChIP-seq Our previously published RNA-seq data from hMSC-TERT4 at day 0, 4 hours, day 1, 3, 7, and 14 of adipogenic and osteogenic differentiation (GEO; GSE113253) were analyzed as described by Rauch et al. 2019. Previously published DNase-seq and MED1 ChIP-seq data from hMSC-TERT4 at day 0, and day 1, 3, 7, and 14 of adipogenic and day 0, and day 1, 3, and 7 of osteogenic differentiation (GEO; GSE113253) were analyzed as described (Rauch et al., 2019) and visualized using using the pyGenomeTracks software (Lopez-Delisle et al., 2021). Analysis of putative enhancer-anchored interactions using ECHi-C Our previously published ECHi-C data from hMSC-TERT4 at day 0 and day 10 of adipogenic differentiation (GEO; GSE140782) was analyzed as described (Madsen et al., 2020). Interactions anchored in putative enhancers were visualized using the WashU Epigenome Browser (Zhou et al., 2011). Analysis of chromatin conformation in PPARG locus in different stem cell states using public Hi-C data Hi-C data set for hMSC-TERT4 at progenitor state (GEO; GSE140782) (Madsen et al., 2020) were mapped to human genome (hg19), filtered and processed using runHiC (Xiaotao, 2016). HiC matrixes for ESC, mesodermal cells, MSC, neuronal progenitor cells (GEO; GSE52457) were visualized using Genome Browser from YUE lab (http://3dgenome.fsm.northwestern.edu/view.php). Analysis of super-enhancers in hMSC-TERT4 Super-enhancers were identified and ranked based on MED1 ChIP-seq enrichment at day 14 of adipogenesis in hMSC-TERT4 using ROSE (Lovén et al., 2013; Warren et al., 2013) with default parameters. IMAGE analysis IMAGE analysis was performed according to the instructions (Madsen et al., 2018). Briefly, putative enhancers activity (based on MED1 signal), and RNA-sequencing data from day 0, 1, 3 and 4 hours of adipogenesis in hMSC-TERT4 cells were used as input for a two-step machine-learning algorithm to calculate motif activity depicted as the contribution of given TF motifs to overall enhancer activity and gene expression in given time point of adipogenesis. IMAGE also predicts target enhancers (TE) for all motifs based on an estimated error per enhancer when the modeled enhancer activity is compared with or without a specific motif. TE was used to dissect all motifs targeting TFs at studied PPARG enhancers. ChIP ChIP was performed as described previously (Rauch et al., 2019). Briefly, only for MED1 ChIP cells were crosslinked in 2 mM disuccinimidyl glutarate (DSG) in PBS (100 µL of 0.5 M DSG (Proteochem) in DMSO per 25 ml of PBS) for 20 minutes. For H3K4me1, CTCF, C/EBPβ, C/EBPα and PPARγ ChIP, this step was omitted. Then, cells were crosslinked in 1% formaldehyde in PBS for 10 minutes followed by quenching with 0.125 M glycine for 10 minutes. Cells were harvested in cold lysis buffer (0.1% SDS, 1% Triton X-100, 0.15 M NaCl, 1 mM EDTA, 20 mM Tris pH 8) and stored at -80°C prior to sonication. Samples were sonicated using ME220 Focused-ultrasonicator (Covaris) for 15 min, peak power 75, duty% factor 5, cycles/burst 1000 (for MED1 ChIP-qPCR, μL microTUBE AFA fiber Pre-Slit Snap-Cap 6x16 mm tubes (Covaris)), and for 10 min, peak power 75, duty% factor 26.66, cycles/burst 500 (for ChIP samples followed by sequencing, miliTUBE 1 ml AFA Fiber(100) (Covaris)). Antibodies directed against MED1 (A300-793A, Bethyl Laboratories), H3K4me1 (ab8895, abcam), CTCF (61311, Active Motif), C/EBPβ (sc-7962, Santa Cruz), C/EBPα (8178S, Cell Signaling) and PPARγ (2443S, Cell Signaling) were used for immunoprecipitation. For ChIP-qPCR precipitated DNA was used as a template for qPCR with FastStart Essential DNA Green Master (Roche) on a Lightcycler 480 II (Roche) (primers Supplementary Information Table 2). MED1 occupancy was determined as % recovery of input sample, and the signal in each sample was normalized to signal at the promoter of the house-keeping gene ZRANB3 to account for ChIP efficiency (see Extended Data Fig. 5a for MED1 % recovery in different genomic regions where signal was expected to be similar across samples). For ChIP-seq, precipitated DNA was prepared for sequencing following the manufacturer's protocol (Illumina). ChIP-seq was performed on minimum two independent biological experiments. ChIP libraries were sequenced on the Illumina NovaSeq 6000, and the quality of sequenced reads was estimated using FastQC (https://qubeshub.org/resources/fastqc). Sequencing reads were mapped to the human genome (hg19) using STAR (Dobin et al., 2013). Picard (http://broadinstitute.github.io/picard/) was used to remove duplicates from aligned reads. Peaks were identified using MACS2 (Zhang et al., 2008), and reproducible peaks were estimated using BEDTools Intersect (Quinlan & Hall, 2010). Bigwig files were created using bamCoverage with RPKM normalization using deepTools (Ramírez et al., 2014), and then visualized using pyGenomeTracks software (Lopez-Delisle et al., 2021). For MED1 ChIP-seq at day 10 of adipogenesis in control ‘No gRNA’ and E+102 deletion in hMSC-TERT4-iCas9/4, and for MED1 ChIP-seq at day 3 of adipogenesis in control ‘No gRNA’ and in E+2 deletion and in E+102 deletion in hMSC-TERT4-iCas9/4 , and for C/EBPβ ChIP-seq at d1 of adipogenesis in control ‘No gRNA’ and in E+2 deletion and in E+102 deletion in hMSC-TERT4-iCas9/4, size factors were estimate using DESeqDataSetFromMatrix and estimateSizeFactors functions in DESeq2 v.1.42.0 (Love et al., 2014). Then size factors were used to create normalized bigwig files with BamCoverage from deepTools (Ramírez et al., 2014), and bigwig replicate files were averaged with bigwigAverage from deepTools (Ramírez et al., 2014). For MED1 ChIP-seq at day 3 of adipogenesis, and C/EBPβ ChIP-seq at day 1 of adipogenesis in control ‘No gRNA’ and in E+2 and E+102 deletion in hMSC-TERT4-iCas9/4, differential binding was performed using DiffBind v.3.12.0 with edgeR ( P -value ≤ 0.05) (Ross-Innes et al., 2012). Plot profiles, and heatmaps were created using DeepTools, with commands computeMatrix, plotHeatmap or plotProfile using normalized and averaged bigwig files. Capture-C hMSC-TERT4/iCas9-4 enhancer deletion and control cell lines (one clone for each condition) were harvested at day 0 and 10 of adipogenic differentiation. Capture-C was performed using a customized version of a previously published protocol (Oudelaar et al., 2018). Briefly, cells were cross-linked for 10 minutes in 2% formaldehyde in PBS, quenched with 0.125 M glycine, harvested in fresh cold lysis buffer (10 mM Tris-HCl pH 8, 10 mM NaCl, 0.2% Igepal CA-630 (Sigma), 1x Complete (Roche)), incubated 20 minutes on ice and snap-frozen in liquid nitrogen before storage at -80°C. 3C libraries were generated by in-nucleus double digestion with 3x 150 U of each DdeI and DraI restriction enzymes (New England Biolabs) added over approximately 24 hours (final buffer conditions after last addition of enzyme: 1x enzyme buffer, 0.06% sodium dodecyl sulfate, 1.5% Triton X-100). Restriction enzymes were chosen based on the distribution of restriction sites in the genomic region surrounding the PPARG locus, aiming for the regulatory elements of interest located on restriction fragments (RFs) of approximately 120-250 bp. Digestion overhangs were filled in with dNTP and blunt-end ligated. De-crosslinked chromatin was purified. Approximately 5 μg of purified 3C library was sheared to an average size of 450 bp by sonication on an ME220 Focused-ultrasonicator (Covaris) and cleaned up with 0.85:1 AMPure XP beads (Beckman Coulter) sample ratio to remove fragments <200 bp. For sequencing library preparation, sheared DNA was end-repaired, adenine-tailed and ligated to NEBNext hairpin adaptors (New England Biolabs), followed by U excision by USER Enzyme (New England Biolabs). Libraries were amplified using Q5 high-fidelity DNA polymerase (New England Biolabs), universal primer and 8-bp-index primers (New England Biolabs) in 7 cycles of PCR. For capture, 5’ biotinylated 116-120 bp DNA oligos were designed to target the middle of single RFs spanning, or located as close as possible to, the regulatory elements of interest, avoiding repetitive regions (Sigma standard DNA oligos in tubes (single-stranded), Supplementary Information Table 5). From each sample, 1650 ng of amplified 3C library was used to generate one multiplexed library for capture (biological replicates were kept separate). Per sample in the multiplexed library, 13 fmol in total of pooled biotinylated DNA oligos targeting all 10 regulatory elements of interest was used for capture, i.e. 1.3 fmol of each oligo, and 130 fmol total oligo pool if the multiplexed library consisted of 10 samples. Sequencing adapters and indexes were blocked using a total of 2 nmol custom blockers (Integrated DNA Technologies, PAGE purified DNA oligos) per sample in the multiplexed library, and two rounds of capture were performed as previously described (Oudelaar et al., 2018), using 10 cycles of amplification after each capture. According to the manufacturer's instructions, multiplexed and double-captured libraries were subjected to 150 bp paired-end sequencing (Illumina). The resulting raw sequencing reads were mapped to hg19 and quality-filtered using the CCseqBasicS pipeline (Telenius et al., 2020) in a version customized to handle reads produced with the restriction enzymes DraI and DdeI (see also Supplementary Information Fig. 1). Resulting unique interaction counts from chromosome 3 were normalized to 50,000 counts per sample and viewpoint. As expected, in samples where a captured viewpoint was deleted, the raw interaction count was generally very low compared to samples where the viewpoint was intact (Extended Data Fig. 3c). However, for the viewpoints E-130, E-123 and E+89, the capture oligo and deleted sequence only displayed partial overlap, resulting in some capture of the deleted region. To simplify downstream analysis, counts in all samples, where a captured viewpoint was deleted, were set to 0 for that viewpoint. The analysis was then restricted to the region chr3: 12.1-12.6 Mb (region of interest, ROI) containing the PPARG locus, and interactions to regions ±2 kb from the viewpoint were excluded from further analysis. Normalized interaction counts were averaged over the RFs overlapping bins of 1 kb or 5 kb, giving a count for each 1 kb or 5 kb bin along the ROI seen from each viewpoint in each sample. The expected count for each viewpoint in each sample (at either RF or binned resolution), as a function of distance from the viewpoint, was then predicted by fitting a cubic polynomial model to the observed sample count per 1 kb bin (for element specific interactions, see below) or 5 kb bin (for ROI-wide interactions), given by the geometric mean count per bin across all intact viewpoints in that sample. P -values describing the chance of an observed interaction count at a given distance from a viewpoint randomly being higher than the predicted count for that distance were calculated using Student’s t-test and adjusted for multiple testing using the Benjamini-Hochberg method. Significant interactions (FDR ≤ 0.01, 208 ROI-wide unique interactions out of 1000 possible across all viewpoints) identified in any condition were kept for further analysis. Differential interactions were identified using Student’s t-test but not adjusted for multiple testing to avoid inflating the number of false negatives in this relatively small dataset. Since interaction counts in the ‘No gRNA’ and ‘Ctr gRNA’ controls correlated well, compared to the remaining conditions (Extended Data Fig. 3b), these conditions were considered as four replicates of one control for all statistical tests (instead of two replicates of two controls). To specifically analyze interactions between the selected PPARG community enhancers, element-specific observed and predicted interaction counts were calculated based on un-binned counts as the mean interaction count seen from one viewpoint (enhancer) to all non-captured RFs overlapping another enhancer. This resulted in two interaction counts for each element-specific interaction, one count seen from each viewpoint, which were summed to obtain a single total observed or predicted count for that interaction. Significant interactions (FDR ≤ 0.01, 33 element-specific interactions across all viewpoints out of 45 possible) were identified and subjected to differential analysis as described above. Element-specific interactions were visualized using the circlize package for R (Gu et al., 2014). Analysis of genetic variants within PPARG locus and their predicted effect size using Enfomer To identify variants associated with cardiometabolic diseases, we queried the PPARG topologically associated domain (TAD chr3:12174868-12526858 /hg19) from multiple relevant resources; (a) the Common Metabolic Diseases Knowledge Portal (Costanzo et al., 2023), (b) the CARDIoGRAMplusC4D (Coronary ARtery DIsease Genome wide Replication and Meta-analysis - CARDIoGRAM), plus The Coronary Artery Disease (C4D) Genetics) [The data on coronary artery disease contributed by the CARDIoGRAMplusC4D and UK Biobank CardioMetabolic Consortium CHD working group who used the UK Biobank Resource (application number 9922). Data have been downloaded from www.CARDIOGRAMPLUSC4D.ORG, (c) the DIAGRAM consortium [https://diagram-consortium.org/] (Gaulton et al., 2015; Mahajan et al., 2014; Mahajan et al., 2022; Mahajan et al., 2018) and (d) the variant lists from the GWAS Catalog associated with BMI, T2D, T2DadjBMI, WHR, and WHRadjBMI. From these data resources, the significant variants (p-value 0.8. To further annotate the variants and their corresponding haplotypes, we queried their associated PheWAS from the Common Metabolic Diseases Knowledge Portal API - HuGeAMP (Costanzo et al., 2023), and aggregated traits related to T2D with an association p-value of less than 0.05. These phenotypes or associated traits are reported in the (Extended Data Table 1-2). This table reports the global and race-specific allelic frequencies of the variants which were directly queried from the NIH Allele Frequency Aggregator (ALFA) [https://www.ncbi.nlm.nih.gov/ /docs/gsr/alfa/]. To predict variant effects on the PPARG2 and our enhancer regions we used the Enformer model (Žiga Avsec, Vikram Agarwal, et al., 2021), and we evaluated variant effects on the genomic coordinates of the enhancer regions and the two transcripts of PPARG2 (ENST00000287820 [chr3:12351472-12434356] and ENST00000683699 [chr3:12351500-12434162]). In this analysis, we considered the effects of alternate alleles vs. the reference allele of the variants and their haplotypes to evaluate the effects on PPARG2 enhancer activity (using H3K27ac as proxy) and its expression (CAGE) in adipocyte-related tissues (Supplementary Information Table 1). The maximum absolute effects of the queried variants in these signals were used to represent variant effects on PPARG2 enhancer and CAGE activities. To define a background effect and significance level on the enhancer and CAGE activities, we selected 4000 variants at random from the GWAS catalogue list of significant hits ( p -value < 5 x 10 -8 ). Using Enformer, we calculated the effect of these randomly selected variants on the region of their closest transcripts (the transcript regions were denied using GENCODE v43). The absolute maximum effects of the variants on the same list of adipocyte-related signals were used to define a background effect. The 90 percentiles of the effects from these randomly selected variants were then used as the cutoffs for identifying significant effects of our of variants [Enhancer (90%); negative cutoff: -0.0090, positive cutoff: 0.0088, CAGE (90%): negative cutoff: -0.1147, positive cutoff: 0.1585]. The Enformer results on the GWAS catalogue variants are reported in the (Extended Data Table 1-2). Analysis of binding affinity using HT-SELEX datasets Analysis was performed following the study (Bhimsaria et al., 2023). Briefly, effect of the genetic variant; rs181025382 on DNA binding was estimated as log fold change in MinSeqs using two HT-SELEX data sets for PPARγ: without ligand treatment, and for rosiglitazone-treated PPARγ. These datasets were used to determine the enrichment of sequence flanking ± 20 bp from rs181025382 (chr3:12494623-12494663/hg19) for both the reference allele (A/hg19) and the rs181025382-altered allele (G/hg19). As controls, we selected the top 1000 PPARγ-bound enhancers in hMSC-TERT4 (identified based on a reproducible signal from DNase- and MED1-ChIP seq at day 14 of adipogenesis and PPARγ signal at day 10 of adipogenesis in hMSC-TERT4), and 1000 randomly selected enhancers that were not bound by PPARγ (based on the same signals). MinSeq sequences analysis from two PPARγ datasets mentioned earlier was then performed on 1000 bp sequences these two control enhancers sets. The log fold change in the enrichment was calculated with log 2 , where E (alt.allele) and E (Ref.allele) are enrichment values calculated by scanning MinSeq sequences within the queried reference and altered sequences. is a parameter to facilitate unintended issues caused by division by small numbers. BPNet Model analysis A BPNet (Žiga Avsec, Melanie Weilert, et al., 2021) model was trained on PPARγ ChIP at day 10 of adipogenesis in hMSC-TERT4 data. Replicates were merged using samtools v. 1.13 (Danecek et al., 2021). 5’ ends of reads were extracted using bedtools v.2.30.0 (Quinlan & Hall, 2010) using subcommand genomecov for each strand, and base pair resolution bigwig files were constructed using bedGraphToBigWig. Low threshold peaks were called with macs2 v.2.2.9.1 using subcommand callpeak using the parameters -p 0.1 --gsize 2701495761 --keep-dup all --shift -75 --extsize 150. Peak regions were blacklist filtered (Amemiya et al., 2019) with bedtools intersect using the ENCODE hg19 blacklist. The model was trained using the BPNet-lite v.0.8.1 implementation of BPNet (https://github.com/jmschrei/bpnet-lite). We used an input sequence length of 2114 bp and an output profile width of 1000 bp. We used 64 convolutional filters and an 8 layers deep tower of dilated convolutions. Augmentation was used on the training sequences, using a max jitter (random lengthwise translation) of 128 bp and reverse complementing with a probability of 0.5. Chromosomes 8 and 10 were used for validation, while the remaining autosomal chromosomes plus chromosome X were used for training. The model with the lowest loss on the validation chromosomes after training for 50 epochs was selected. Motif discovery and attributions To identify sequence motifs driving the ChIP-seq signal we utilized TF-MoDISco (Shrikumar, 2020), running the updated implementation tfmodisco-lite v.2.2.0 (https://github.com/jmschrei/tfmodisco-lite). The identified motifs were matched against the JASPAR motif database (Rauluseviciute et al., 2024) using tomtom (Gupta et al., 2007) from meme v.5.5.5. SNP analysis To analyze the effect of individual SNP at or near the PPARG locus we used tangermeme v.0.2.1 (https://github.com/jmschrei/tangermeme) and performed in-silico saturation mutagenesis (ISM) using the count predictions. Genome-wide PPARG motif mutation analysis We extracted the PPARG:RXR dimer motif (ID: MA0065.2) from the JASPAR database. We then ran FIMO using tangermeme v.0.2.1 (https://github.com/jmschrei/tangermeme) to identify motif matches across all PPARγ peak region PPARγ ChIP-seq from day 10 of adipogenesis in hMSC-TERT4, using a p-value threshold of 0.01. The matches were then narrowed down to unique and high scoring matches, selecting only those with a score above 10. Finally, we selected those matches with an A at position 9 of the motif match, so that we can induce the A > G mutation as for rs181025382. We then calculated the count prediction for sequences centered on the variant both with and without the A > G mutation (Extended Data Fig. 7b-c). Statistics and reproducibility For each enhancer deletion or hMSC-TERT4/iCas9-4 control, mRNA expression and MED1 ChIP-qPCR were assessed based on two independent differentiation experiments of the three independent clones ( n =2 independent biological replicates). For ChIP-seq in enhancer deletion and controls two clonal cell lines per enhancer deletion was used in two independent differentiation experiments ( n =2 independent biological replicates, n =2 clonal cell line per condition). For Capture-C, only one clonal cell line per enhancer deletion was used in two independent differentiation experiments ( n =2 independent biological replicates). For publicly available sequencing data from hMSC-TERT4, data were used from at least two independent differentiation experiments ( n =2 independent biological replicates, n =3 for RNA-sequencing). Lipid accumulation in response to enhancer deletion was assayed based on one differentiation experiment using 3 clonal cell lines per enhancer deletion. For statistics not performed within published packages or methods, normality was evaluated visually before testing using the qqnorm() function in R. Equal variances were either not assumed or tested using var.test() in R. Student’s t-test (for data with equal variances) and Welch’s t-test (for data with non-equal variances) were performed in R using the t.test() function in a two-sided manner. P -values were adjusted for multiple testing in R using the p.adjust() function with the Benjamini-Hochberg method where indicated. Polynomial regression was performed in R using the lm() function. All bar plots show individual measurements and/or mean across a group (defined in Fig. legends), and error bars represent the standard deviation (SD), standard error of the mean (SEM) or range between two experiments (indicated in Fig. legends). Line plots of Capture-C interactions show the mean of two independent experiments with error bars representing a range. Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryInformation.pdf Supplementary material related to computational and experimental procedures. ExtendedData.docx Cite Share Download PDF Status: Published Journal Publication published 15 Mar, 2026 Read the published version in Nature Communications → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6466826","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":449667767,"identity":"48eed30a-ff73-453d-9f04-950e3b8ba13a","order_by":0,"name":"Anna Cetnarowska","email":"","orcid":"https://orcid.org/0000-0002-0539-3152","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Anna","middleName":"","lastName":"Cetnarowska","suffix":""},{"id":449667768,"identity":"a7e6fee6-5ad0-44b5-8671-e5b0708133e8","order_by":1,"name":"Mette Hyldahl","email":"","orcid":"https://orcid.org/0000-0002-7407-6062","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Mette","middleName":"","lastName":"Hyldahl","suffix":""},{"id":449667769,"identity":"0f827c13-9bb2-4d32-aaf6-9c717a577d84","order_by":2,"name":"Marcus Nygård","email":"","orcid":"","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Marcus","middleName":"","lastName":"Nygård","suffix":""},{"id":449667770,"identity":"7d3c7ad3-96ea-4c7a-a436-f774d8814aad","order_by":3,"name":"Hesam Dashti","email":"","orcid":"","institution":"Broad Institute of MIT and Harvard","correspondingAuthor":false,"prefix":"","firstName":"Hesam","middleName":"","lastName":"Dashti","suffix":""},{"id":449667771,"identity":"c2e7072b-fd3c-4ce0-a7b6-5284365963d7","order_by":4,"name":"Bo Vagner Hansen","email":"","orcid":"https://orcid.org/0000-0001-9468-2839","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Bo","middleName":"Vagner","lastName":"Hansen","suffix":""},{"id":449667772,"identity":"e5684e92-301a-42df-82f0-6f6d55d4a655","order_by":5,"name":"Laura Holm","email":"","orcid":"","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Laura","middleName":"","lastName":"Holm","suffix":""},{"id":449667773,"identity":"d9065fa7-9230-45a6-aebe-19b630dd933a","order_by":6,"name":"Kaja Madsen","email":"","orcid":"","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Kaja","middleName":"","lastName":"Madsen","suffix":""},{"id":449667774,"identity":"a28445e0-b502-453f-b7a6-8169731c8b60","order_by":7,"name":"Maria Madsen","email":"","orcid":"https://orcid.org/0000-0002-5039-340X","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"","lastName":"Madsen","suffix":""},{"id":449667775,"identity":"189fd54c-1a8a-4fe8-8bfd-cdb170b6fe1b","order_by":8,"name":"Vallari Shukla","email":"","orcid":"","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Vallari","middleName":"","lastName":"Shukla","suffix":""},{"id":449667776,"identity":"e9034ca6-18c9-4fc9-89f5-7621eb48cbb3","order_by":9,"name":"Esra Durmaz Mitchell","email":"","orcid":"https://orcid.org/0000-0002-4345-2264","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Esra","middleName":"Durmaz","lastName":"Mitchell","suffix":""},{"id":449667777,"identity":"21a216e1-0aed-4021-9bd0-ea2e17add785","order_by":10,"name":"Alexander Rauch","email":"","orcid":"https://orcid.org/0000-0002-9429-7356","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Alexander","middleName":"","lastName":"Rauch","suffix":""},{"id":449667778,"identity":"d4618f12-de48-4af0-bb7b-cf0d74dc5b6a","order_by":11,"name":"Jesper Madsen","email":"","orcid":"https://orcid.org/0000-0002-0518-0800","institution":"University of Southern Denmark","correspondingAuthor":false,"prefix":"","firstName":"Jesper","middleName":"","lastName":"Madsen","suffix":""},{"id":449667779,"identity":"59b7288d-3e8c-4c79-bb17-bd47a91db70b","order_by":12,"name":"Melina Claussnitzer","email":"","orcid":"https://orcid.org/0000-0003-2450-736X","institution":"Massachusetts General Hospital, Harvard Medical School; Type 2 Diabetes Systems Genomics, Broad Institute of MIT and Harvard, Initiative, Cambridge MA","correspondingAuthor":false,"prefix":"","firstName":"Melina","middleName":"","lastName":"Claussnitzer","suffix":""},{"id":449667766,"identity":"7a8553b8-bbe5-4401-8de6-1093c5b2144b","order_by":13,"name":"Susanne Mandrup","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2UlEQVRIiWNgGAWjYNACNiBm72FgBlKMDQwMxkRq4TlDshaJHCK1yDdwJ34uKLuXuF3y7dHNBQx1shsOMG82wKfF4ADvZukZ54oTd87OS7s9g+Gw8YYDbMUJeLUw8G6Q5m1LSNxwO8fsNg/DgcQNB3iMD+B3GO/m32AtN8+AtNQR1sJwgHcbxJYbPCAtzGAt+B12mHebNc+5BOOdPSC/GBw2nnmYrRiv9+Xbezff5ilLkN3OfvbY7YKKOtm+482bJfA6jBlmHZxkxqkW3YXEKhwFo2AUjIKRBwBMlEsAwZZPUwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-0961-5787","institution":"University of Southern Denmark","correspondingAuthor":true,"prefix":"","firstName":"Susanne","middleName":"","lastName":"Mandrup","suffix":""}],"badges":[],"createdAt":"2025-04-17 00:40:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6466826/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6466826/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41467-026-70401-7","type":"published","date":"2026-03-15T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89351909,"identity":"e6da9135-2dd9-4ce4-9afb-d467bf33def9","added_by":"auto","created_at":"2025-08-19 06:25:35","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1465063,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003eenhancer community is highly interconnected already in undifferentiated hMSC.\u003c/strong\u003eGenomic overview of the \u003cem\u003ePPARG\u003c/em\u003e locus and neighboring loci in human stromal progenitors (hMSC-TERT4). \u003cstrong\u003ea\u003c/strong\u003e, Hi-C contact map at 25 kb resolution, \u003cem\u003en\u003c/em\u003e=2 biological replicates. CTCF ChIP-seq track at day 0 of adipogenesis. Forward and reverse-oriented CTCF binding sites represented as blue and green triangles, respectively, \u003cem\u003en\u003c/em\u003e=2 biological replicates. H3K27ac and H3K4me1 ChIP-seq tracks at day 0 and 3 of adipogenesis. \u003cem\u003en\u003c/em\u003e=2 biological replicates. \u003cstrong\u003eb\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eECHi-C interactions at day 0 and 10 of adipogenesis. Interactions with significant signal over background were identified using CHiCAGO (Cairns et al., 2016), see Madsen et al. 2020 for details. All significant interactions between elements within the indicated genomic region are shown (black arcs) and visualized using the WashU Epigenome Browser (Zhou et al., 2011). \u003cem\u003en\u003c/em\u003e=2 biological replicates. \u003cstrong\u003ec\u003c/strong\u003e, RNA-seq-based expression of genes surrounding the most self-interactive part of the genomic region shown in (b) at day 0, 4 hours, and day 1, 3, 7, and 14 of adipogenesis hMSC-TERT4, given in reads per kilobase million (RPKM). Bars indicate mean+1 SD. \u003cem\u003en\u003c/em\u003e=3 biological replicates.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/1a89027b9655d9f1c6d80608.png"},{"id":89351910,"identity":"c9c35523-1f87-4aab-8bf9-e7f313101f19","added_by":"auto","created_at":"2025-08-19 06:25:35","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":324374,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInduction of adipogenesis leads to \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ede novo\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e enhancer activation and increased connectivity of the \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG \u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003eenhancer community. \u003c/strong\u003ehMSC-TERT4 (a, b) or MSC-TERT4/iCas9-4 (see methods for details) (b, c) were induced to undergo adipogenesis\u003cstrong\u003e. a\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eECHi-C interactions within the \u003cem\u003ePPARG\u003c/em\u003eHICE community (Madsen et al., 2020) at day 0 and 10 of adipogenesis. Interactions with significant signal over background were identified using CHiCAGO (Cairns et al., 2016), see (Madsen et al., 2020) for details. CTCF ChIP seq tracks at day 0, day 1 and 10 of adipogenesis. Forward and reverse-oriented CTCF binding sites are represented as blue and green triangles, respectively. \u003cem\u003en\u003c/em\u003e=2 biological replicates. DNase- and MED1 ChIP-seq tracks from day 0, day 1, 3, 7 and 14 of adipogenesis. Putative key enhancers are highlighted by green, red, and blue vertical lines. \u003cem\u003en\u003c/em\u003e=2 biological replicates.\u003cstrong\u003e b\u003c/strong\u003e, Capture-C interactions seen from the \u003cem\u003ePPARG2\u003c/em\u003e TSS (5 kb resolution) at day 0 and day 10 of adipogenesis. Empty circles indicate the mean normalized interaction count of significant non-dynamic interactions (FDR ≤ 0.01, P-value \u0026gt; 0.05), and filled circles indicate the mean normalized interaction count of significant and dynamic interactions (FDR ≤ 0.01, \u003cem\u003eP\u003c/em\u003e-value ≤ 0.05). Error bars indicate the range between two independent experiments. FDR was obtained by two-tailed Student’s t-test and Benjamini-Hochberg correction. \u003cem\u003eP\u003c/em\u003e-values obtained by two-tailed Student’s t-test.\u003cem\u003e n\u003c/em\u003e=2 biological replicates.\u003cstrong\u003e c\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eCircos plot of Capture-C interactions between enhancers of interest from (a) as well as the \u003cem\u003ePPARG2\u003c/em\u003e TSS. Normalized interaction counts detected at either day 0 or day 10 of adipogenesis are represented as lines (FDR ≤ 0.01). Log2 of normalized interaction counts represented as a gray scale gradient (white = the lowest number, black = the highest number). FDR was obtained by two-tailed Student’s t-test and Benjamini-Hochberg correction.\u003cem\u003e n\u003c/em\u003e=2 biological replicates.\u003cstrong\u003e d\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eSchematic depicting the status of enhancer activity, chromatin structure and gene expression in the \u003cem\u003ePPARG\u003c/em\u003ecommunity at day 0 and day 10 of adipogenesis.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/49c7a083c7c4c86affbe0a5d.png"},{"id":89351911,"identity":"d9c2f0c0-2670-42af-9ade-7ea46d09b830","added_by":"auto","created_at":"2025-08-19 06:25:35","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":223172,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePromoter-proximal enhancers and super-enhancer constituents are required for activation of \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG2\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003eand adipogenesis. \u0026nbsp;a\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eWorkflow for the production and characterization of hMSC-TERT4/iCas9-4 clones with deletion of individual enhancers in the \u003cem\u003ePPARG\u003c/em\u003e community. \u003cstrong\u003eb\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eRelative mRNA expression of indicated genes was assayed by cDNA-qPCR at day 10 of adipogenesis in hMSC-TERT4/iCas9-4 with the indicated enhancer deletions. Bars indicate mean+1 SEM. \u003cem\u003eP\u003c/em\u003e-value obtained by two-tailed Welch’s t-test enhancer deletion vs. ‘No gRNA’. \u003cem\u003eP\u003c/em\u003e-values \u0026lt; 0.05 are shown. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=3 clonal cell lines per condition.\u003cstrong\u003e c\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eCorrelation between mean relative \u003cem\u003ePPARG2\u003c/em\u003e expression at day 10 of adipogenesis (\u003cem\u003en\u003c/em\u003e=2 biological replicates) and mean lipid droplet area at day 9 of adipogenesis (\u003cem\u003en\u003c/em\u003e=1 biological replicate) in hMSC-TERT4/iCas9-4 with indicated enhancer deletions.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/ca396a63a33112f25a73d149.png"},{"id":89352174,"identity":"571e544d-0de0-4079-9919-4125102ca6ae","added_by":"auto","created_at":"2025-08-19 06:33:35","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":552121,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eKey enhancers in the \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003ecommunity stabilize local adipogenesis-associated chromatin interactions. \u003c/strong\u003eElement-specific Capture-C in hMSC-TERT4/iCas9-4 with deletion of indicated enhancers. \u003cstrong\u003ea\u003c/strong\u003e, Overview of the most reproducible changes in normalized interaction count for significant (FDR ≤ 0.01 in combined controls or any deletion) interactions at day 0 and 10 of adipogenesis in cells with deletion of the indicated enhancers (top labels). Enhancer deletions were ordered according to rank of importance (Fig. S5f). Interactions that change at either day 0 or 10 of adipogenesis upon any deletion relative to combined controls (\u003cem\u003eP\u003c/em\u003e ≤ 0.05, purple shades) are labeled on the side of each heatmap. log2FC is shown as a number upon interaction change. Crossed-out rectangles indicate the deletion of one of the enhancers in the given interaction. FDR was obtained by two-tailed Student’s t-test and Benjamini-Hochberg correction. \u003cem\u003eP\u003c/em\u003e-values obtained by two-tailed Student’s t-test.\u003cem\u003e n\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=1 clonal cell lines per condition.\u003cstrong\u003e b, \u003c/strong\u003eTop: Capture-C normalized interaction seen from the \u003cem\u003ePPARG2\u003c/em\u003e TSS (5 kb resolution) at day 0 and 10 of adipogenesis in combined controls and in E+102 deletion. Lines represent mean of interactions, shade represents the range between two independent experiments. Circles indicate the mean normalized interaction count of significant and dynamic \u003cem\u003ePPARG2\u003c/em\u003eTSS-baited interactions (FDR ≤ 0.01, \u003cem\u003eP\u003c/em\u003e-value ≤ 0.05). Putative key enhancers are highlighted by green, red, and blue vertical lines. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=1 clonal cell lines per condition. Bottom: Circos plot of reproducible changes (\u003cem\u003eP\u003c/em\u003e ≤ 0.05) in normalized interaction count for significant (FDR ≤ 0.01 in combined controls or any deletion) affected by deletion of E+102 at day 0 and 10 of adipogenesis. Mean log2FC between either day 0 or day 10 of adipogenesis in in combined controls vs. in E+102 deletion shown as line color. FDR was obtained by two-tailed Student’s t-test and Benjamini-Hochberg correction. \u003cem\u003eP\u003c/em\u003e-values obtained by two-tailed Student’s t-test.\u003cem\u003e n\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=1 clonal cell lines per condition.\u003cstrong\u003e c, \u003c/strong\u003eTop: Capture-C normalized interaction seen from the \u003cem\u003ePPARG2\u003c/em\u003e TSS (5 kb resolution) at day 0 and 10 of adipogenesis in combined controls and in E+2 deletion. Lines represent mean of interactions, shade represents the range between two independent experiments. Circles indicate the mean normalized interaction count of significant and dynamic \u003cem\u003ePPARG2\u003c/em\u003e TSS-baited interactions (FDR ≤ 0.01, \u003cem\u003eP\u003c/em\u003e-value ≤ 0.05). Putative key enhancers are highlighted by green, red, and blue vertical lines. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=1 clonal cell lines per condition. Bottom: Circos plot of reproducible changes (\u003cem\u003eP\u003c/em\u003e ≤ 0.05) in normalized interaction count for significant (FDR ≤ 0.01 in controls or any deletion) affected by E+2 deletion at day 0 and 10 of adipogenesis. Mean log2FC between either day 0 or 10 of adipogenesis in in combined controls vs. in E+2 deletion shown as line color. FDR was obtained by two-tailed Student’s t-test and Benjamini-Hochberg correction. \u003cem\u003eP\u003c/em\u003e-values obtained by two-tailed Student’s t-test.\u003cem\u003e n\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=1 clonal cell lines per condition.\u003cstrong\u003e d, \u003c/strong\u003eSchematic depicting the status of enhancer activity, community structure and gene expression in the \u003cem\u003ePPARG\u003c/em\u003e community at day 0 and 10 of adipogenesis and at day 10 of adipogenesis with \u0026nbsp;E+102 deletion.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/74cc1064e2c3382d44cbf47e.png"},{"id":89352902,"identity":"6a06766a-29ee-460a-b3f2-0a2aa29a34e7","added_by":"auto","created_at":"2025-08-19 06:41:35","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":291936,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEnhancers in the\u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e community cooperate to recruit Mediator during adipogenesis of hMSC-TERT4. \u003c/strong\u003ehMSC-TERT4/iCas9-4 and clones with the indicated enhancer deletions were induced to differentiate to adipocytes and MED1 occupancy was assessed by ChIP-qPCR and ChIP-seq. \u003cstrong\u003ea\u003c/strong\u003e, MED1 ChIP-qPCR (recovery of input at indicated enhancer regions relative to the recovery of input at the \u003cem\u003eZRANB3 \u003c/em\u003epromoter, see methods for details) at day 10 of adipogenesis. MED1 occupancy at each of the \u003cem\u003ePPARG\u003c/em\u003e community enhancers in ‘No gRNA’ control and upon deletion of each individual enhancer. The \u003cem\u003eP\u003c/em\u003e-value was obtained by two-tailed Welch’s t-test enhancer deletion vs. ‘No gRNA. *\u003cem\u003eP\u003c/em\u003e-value ≤ 0.05. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=3 clonal cell lines per condition. \u003cstrong\u003eb\u003c/strong\u003e, Correlation between mean relative \u003cem\u003ePPARG2\u003c/em\u003e expression at day 10 of adipogenesis and summed MED1 occupancy at all enhancers at day 10 of adipogenesis. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=3 clonal cell lines per condition. \u003cstrong\u003ec\u003c/strong\u003e, Top: Averaged and normalized MED1 ChIP-seq tracks at day 3 of adipogenesis in ‘No gRNA’ control, and cells with deletion of E+2 or E+102. Enhancers are highlighted by green, red, and blue vertical lines. Dashed squares indicate the deleted enhancer. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=2 clonal cell lines per condition. Bottom: Plot profile of averaged and normalized MED1 ChIP-seq occupancy at studied enhancers in the \u003cem\u003ePPARG\u003c/em\u003e community at day 3 of adipogenesis in ‘No gRNA’ control, E+2 and E+102 deletion clones. Colors represent MED1 occupancy in ‘No gRNA’ control – grey, E+2 deletion – red, E+102 deletion – blue. Respective colored squares indicate enhancers with a significant change in MED1 occupancy (EdgeR,\u003cem\u003e P\u003c/em\u003e-value ≤ 0.05). Visualized using deepTools (Ramírez et al., 2014). \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=2 clonal cell lines per condition. \u003cstrong\u003ed\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eSchematic depicting the status of enhancer activity and gene expression in the \u003cem\u003ePPARG\u003c/em\u003e community at day 3 of adipogenesis in hMSC-TERT4 and with deletion of E+102.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/29e380f93e5c7d4ffa32d7af.png"},{"id":89352175,"identity":"7e59ea2c-f38b-4db7-b3bb-21d335a533e9","added_by":"auto","created_at":"2025-08-19 06:33:35","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1099707,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExtensive enhancer cooperativity is required for activation of \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG \u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003ecommunity. \u003c/strong\u003ehMSC-TERT4, hMSC-TERT4/iCas9-4 control and clones with the indicated enhancer deletions were induced to differentiate to adipocytes.\u003cstrong\u003e a\u003c/strong\u003e, C/EBPβ motif activity (continuous line) based on IMAGE analysis of MED1 ChIP-seq data (see Methods for details) for day 0, 4h, and day 1 and 3 of adipogenesis. The dashed line shows \u003cem\u003eCEBPB \u003c/em\u003eexpression based on RNA-seq. \u003cstrong\u003eb\u003c/strong\u003e, C/EBPβChIP-seq tracks at day 0, 1 and 3 of adipogenesis. Enhancers are highlighted by vertical green, red, and blue lines\u003cem\u003e. n\u003c/em\u003e=2 biological replicates. \u003cstrong\u003ec\u003c/strong\u003e, Top: Averaged and normalized C/EBPβChIP-seq tracks at day 1 of adipogenesis in ‘No gRNA’ control and cells with deletion of E+2 or E+102. Enhancers are highlighted by green, red, and blue vertical lines. Dashed squares indicate enhancer deletions. \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=2 clonal cell lines per condition. Bottom: Plot profile of averaged and normalized C/EBPβ ChIP-seq occupancy at studied enhancers in the \u003cem\u003ePPARG\u003c/em\u003ecommunity at day 1 of adipogenesis in ‘No gRNA’ control, cells with deletion of E+2 or E+102. Colors represent C/EBPβoccupancy as follows: ‘No gRNA’ control – grey, E+2 deletion – red, E+102 deletion – blue. Respective colored squares indicate enhancers with a significant change in C/EBPβoccupancy (EdgeR,\u003cem\u003e P\u003c/em\u003e-value ≤ 0.05). Visualized using deepTools (Ramírez et al., 2014). \u003cem\u003en\u003c/em\u003e=2 biological replicates, \u003cem\u003en\u003c/em\u003e=2 clonal cell lines per condition. \u003cstrong\u003ed\u003c/strong\u003e, Schematic model depicting the principle of enhancer cooperativity in \u003cem\u003ecis \u003c/em\u003ethrough CIST (Madsen et al., 2020). \u003cstrong\u003ee\u003c/strong\u003e, C/EBPα ChIP-seq tracks at day 0 and 10 of adipogenesis, and PPARγChIP-seq tracks at day 0, 1, 3, and 10 of adipogenesis. Putative enhancers are highlighted by green, red, and blue vertical lines\u003cem\u003e. n\u003c/em\u003e=2 biological replicates. \u003cstrong\u003ef\u003c/strong\u003e, Schematic depicting enhancer cooperativity in \u003cem\u003etrans \u003c/em\u003ethrough feedback activation by PPARγ and other PPARγ induced transcription factors on \u003cem\u003ePPARG\u003c/em\u003e enhancers.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/fbca3343805a173411f94529.png"},{"id":89351913,"identity":"12aee717-05f5-409b-bb2c-901843b6a178","added_by":"auto","created_at":"2025-08-19 06:25:35","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":359619,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eKey enhancers overlap with non-coding genetic variants associated with metabolic disease traits and predicted to affect \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePPARG2 \u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003eexpression. a, \u003c/strong\u003eSchematic overview of the computational workflow used to identify genetic variants in the \u003cem\u003ePPARG\u003c/em\u003e locus with predicted Enformer effect (Žiga Avsec, Vikram Agarwal, et al., 2021) on \u003cem\u003ePPARG2\u003c/em\u003esignal and enhancer accessibility (see Methods for details).\u003cstrong\u003e b, \u003c/strong\u003eDNase-seq at day 10 of adipogenesis in hMSC-TERT4 cells, and the top 40 genetic variants associated with metabolic diseases traits and with the biggest predicted effect on \u003cem\u003ePPARG2\u003c/em\u003e signal (top 20 positive, top 20 negative) (see Methods for details). \u003cstrong\u003ec, \u003c/strong\u003eTable depicting the overlap between genetic variants with top Enformer effect (\u003cem\u003ePPARG2\u003c/em\u003e signal, and enhancer accessibility) and enhancers from the \u003cem\u003ePPARG\u003c/em\u003e enhancer community. Green indicates positive effect, red indicates negative effect.\u003cstrong\u003e d, \u003c/strong\u003eTop: MinSeq derived PWM for PPARγ:RXR from HT-SELEX (Bhimsaria et al., 2023). DNA sequences of the PPRE in E+102 for reference (A) and rs181025382-altered allele (G). Bottom: Box plot of PPARγ:RXRbinding affinity to PPRE in E+102 for reference allele and rs181025382-affected allele, based on ± rosiglitazone treated PPARγ HT-SELEX data sets.\u003cstrong\u003e e, \u003c/strong\u003ers181025382 variant location in E+102. Top: DNase-seq peak at day 14, and PPARγ ChIP-seq peak at day 10 of adipogenesis in hMSC-TERT4 at E+102. \u003cem\u003en\u003c/em\u003e=2 biological replicates. Bottom: BPNet-predicted (see Methods for details) profile contribution of the DNA sequences of E+102 region for reference and rs181025382-altered allele.\u003cstrong\u003e f, \u003c/strong\u003eTop: TF-MoDISco-derived PPARγ:RXR motif computed from the profile contribution scores map for PPARγ peaks obtained from PPARγ ChIP-seq at day 10 of adipogenesis in hMSC-TERT4. \u003cem\u003en\u003c/em\u003e=2 biological replicates. DNA sequences of the PPRE in E+102 for reference and rs181025382-altered allele.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/d9442719e257df87b2225c82.png"},{"id":107973397,"identity":"e17cf7e1-5fc1-4923-8d91-39566a506f50","added_by":"auto","created_at":"2026-04-28 07:11:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5112802,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/8f34085b-5de7-4844-b415-a0e55c4f05d7.pdf"},{"id":89351915,"identity":"debe1b1c-7ec3-4cf9-9c68-2f80f7d918ee","added_by":"auto","created_at":"2025-08-19 06:25:35","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3861042,"visible":true,"origin":"","legend":"Supplementary material related to computational and experimental procedures.","description":"","filename":"SupplementaryInformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/4070b603d37cb3faa712e277.pdf"},{"id":89352903,"identity":"89ea365c-e4da-4e02-b65a-3d4f540af8ba","added_by":"auto","created_at":"2025-08-19 06:41:35","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":3081241,"visible":true,"origin":"","legend":"","description":"","filename":"ExtendedData.docx","url":"https://assets-eu.researchsquare.com/files/rs-6466826/v1/a6c1a1503d5e2f0feb66ce90.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Extensive enhancer crosstalk controls PPARG2 activation during adipogenesis","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAdipocytes are highly specialized cells that play an important role in metabolic energy storage and homeostasis. Genetic or acquired conditions resulting in decreased lipid storage and endocrine function, so-called partial lipodystrophies, are closely associated with ectopic lipid deposition and cardiometabolic diseases (Fiorenza et al., 2011). Similarly, mounting evidence points to compromised adipocyte function as a key driver of cardiometabolic co-morbidities in obesity (Hagberg \u0026amp; Spalding, 2024).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdipogenesis is the process by which stromal progenitor cells of the mesenchymal lineage develop into lipid-laden adipocytes\u0026nbsp;(Rauch \u0026amp; Mandrup, 2021). The efficiency of this process is of major biomedical importance, as it is required for the development of well-functioning adipose tissue, as well as for healthy expandability of adipose tissue (Ghaben \u0026amp; Scherer, 2019). Moreover, the fact that this process can be efficiently induced \u003cem\u003ein vitro\u003c/em\u003e has made it one of the most well-studied differentiation model systems. Studies of many individual genes, as well as genome-wide transcriptional and epigenomic studies, have documented that the transcriptional networks regulating \u003cem\u003ein vitro\u003c/em\u003e adipogenesis consist of at least two consecutive waves of transcription factors (TFs), where the first wave is induced directly by the adipogenic inducers (typically a cocktail consisting of insulin, dexamethasone, and a cAMP-elevating agent). This first wave remodels the chromatin template and activates the master regulators peroxisome proliferator-activated receptor γ (PPARγ) and CCAAT/enhancer-binding protein α (C/EBPα), which drive the second wave activating the adipocyte gene program. PPARγ is the most important of the two master regulators, being indispensable for adipocyte differentiation \u003cem\u003ein vitro\u003c/em\u003e as well as \u003cem\u003ein vivo\u003c/em\u003e and sufficient to drive trans-differentiation of other mesenchymal lineages into adipocytes\u0026nbsp;(Rauch \u0026amp; Mandrup, 2021; Siersbaek et al., 2012). The master regulator attributes of PPARγ rely at least in part on its biochemical properties, as suggested by comparative studies of lipodystrophy mutants of PPARγ\u0026nbsp;(Madsen et al., 2022b)\u0026nbsp;as well as comparative studies with other members of the PPAR family\u0026nbsp;(Bugge et al., 2009; Nielsen et al., 2006). In addition, the ability to dramatically increase \u003cem\u003ePPARG\u003c/em\u003e expression in response to adipogenic cues may play a role. Thus, the expression of \u003cem\u003ePPARG\u003c/em\u003e must be tightly controlled; initially repressed in the progenitor state, and then rapidly induced in response to the transcription factors of the first wave of adipogenesis. Several transcription factors, most notably C/EBPβ and C/EBPδ, have been shown to be directly involved in the activation of the \u003cem\u003ePPARG\u003c/em\u003e locus\u0026nbsp;(Siersbæk et al., 2014). Surprisingly, however, the mechanisms by which the \u003cem\u003ePPARG\u003c/em\u003e locus is activated and the transcription of \u003cem\u003ePPARG\u003c/em\u003e is switched on during adipogenesis remain poorly understood.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdipogenesis is associated with dramatic chromatin remodeling and \u003cem\u003ede novo\u003c/em\u003e enhancer activation (Rauch et al., 2019; Siersbaek et al., 2011; Siersbæk et al., 2014) and dynamic rewiring of promoter-enhancer interactions (Siersbæk et al., 2017). The remodeling of chromatin is a common feature of differentiation pathways\u0026nbsp;(Bonev et al., 2017; Di Giammartino et al., 2019; Freire-Pritchett et al., 2017; Murphy et al., 2024); however, a direct comparison showed that chromatin remodeling in response to adipogenic inducers is much more dramatic than in response to osteogenic inducers\u0026nbsp;(Rauch et al., 2019).\u0026nbsp;Moreover, using enhancer capture Hi-C (ECHi-C), we recently demonstrated extensive 3D interconnectivity of enhancers in multipotent human stromal progenitors (hMSC-TERT4), forming highly interconnected enhancer (HICE) communities that are dynamically regulated during adipogenesis\u0026nbsp;(Madsen et al., 2020). Interestingly, lineage-selective promoters are generally associated with the most highly connected HICE communities, indicating that enhancer cooperativity is required for lineage determination. Intriguingly, the lineage-selective HICE communities associated with the different lineage fates of mesenchymal stem cells (MSCs), including adipogenesis, osteogenesis, myogenesis, and chondrogenesis, are already established at the stem cell state, indicating that these lineage fates are pre-programmed in progenitor cells. In response to the adipogenic cocktail, new enhancers are activated and join the adipogenic HICE communities, thereby increasing the connectivity of these communities (Madsen et al. 2020). These results indicate that HICE communities dynamically integrate signals from multiple enhancers to activate lineage-specific genes.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHere, we have investigated the activation of the most important locus in adipogenesis, that of \u003cem\u003ePPARG\u003c/em\u003e, during adipocyte lineage determination and differentiation of hMSC-TERT4 cells. Using genome editing, we demonstrate important and non-redundant roles of individual enhancers in the \u003cem\u003ePPARG\u0026nbsp;\u003c/em\u003eenhancer community. We report evidence of extensive enhancer crosstalk in \u003cem\u003ecis\u003c/em\u003e as well as in through feedback activation\u003cem\u003e.\u0026nbsp;\u003c/em\u003eFurthermore, we show that key enhancers in the locus overlap with cardiometabolic disease-associated genetic variants, which we predict to regulate \u003cem\u003ePPARG2\u003c/em\u003e transcription, thereby underscoring the\u003cem\u003e\u0026nbsp;in vivo\u0026nbsp;\u003c/em\u003erelevance of the identified enhancers.\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eThe PPARG\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;locus is primed and connected in undifferentiated human MSC\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTaking advantage of our previously published Hi-C data from undifferentiated human bone marrow-derived mesenchymal stem cells (hMSC-TERT4) (Madsen et al., 2020), we inspected the overall chromatin topology around the human \u003cem\u003ePPARG\u003c/em\u003e locus. This showed that \u003cem\u003ePPARG\u003c/em\u003e is isolated in a topologically associating domain (TAD) of \u0026nbsp;350 kb\u0026nbsp;(chr3: 12,174,868 - 12,526,858) (Fig. 1a) at the stem cell stage prior to chromatin remodeling and induction of \u003cem\u003ePPARG\u003c/em\u003e expression (Rauch et al., 2019).Notably, this TAD also contains \u003cem\u003eTIMP4\u003c/em\u003e and the 3\u0026rsquo;end of \u003cem\u003eSYN2\u0026nbsp;\u003c/em\u003ewith no known functions in adipogenesis. Consistent with the TAD structure, ChIP-seq revealed strong CTCF binding at divergent motifs at the boundaries of the \u003cem\u003ePPARG\u003c/em\u003e TAD (Fig. 1a). Analysis of publicly available Hi-C data from other stem cells\u0026nbsp;(Dixon et al., 2015)\u0026nbsp;indicated that the locus is also isolated into a partially defined TAD in embryonic (ESC), mesendodermal cells and mesenchymal stem cells (MSC), but not in neuronal progenitor cells, indicating that the locus is organized very early during development but loses its connectivity in some lineages \u0026nbsp;(Extended Data Fig. 1a).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eInspection of the \u003cem\u003ePPARG\u003c/em\u003e locus connectivity at higher resolution using our previously published ECHi-C data from hMSC-TERT4 cells \u0026nbsp;(Madsen et al., 2020) confirmed that the locus is engaged in an isolated chromatin network in the undifferentiated state (Fig. 1b). Consistent with other adipocyte enhancer communities (Madsen et al., 2020), the locus becomes even more connected and confined during differentiation (Fig. 1b). ChIP-seq analysis revealed that the locus is already primed by H3K4me1 in the undifferentiated state but only gains the active enhancer mark H3K27ac by day 3 of differentiation, coinciding with the activation of \u003cem\u003ePPARG\u003c/em\u003e gene expression (Fig. 1a, Fig. 1c).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe \u003cem\u003ePPARG\u0026nbsp;\u003c/em\u003elocus is transcribed from two major transcription start sites (TSS), \u003cem\u003ePPARG1\u003c/em\u003e and\u003cem\u003e\u0026nbsp;PPARG2\u003c/em\u003e, where \u003cem\u003ePPARG2\u003c/em\u003e is the adipocyte-specific TSS\u0026nbsp;(Fajas 1997). The \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003etranscript encodes PPAR\u0026gamma;2, which is characterized by an additional 28 amino acids in the N-terminal compared with PPAR\u0026gamma;1. PPAR\u0026gamma;2 has been shown to be the most adipogenic isoform (Mueller et al., 2002; Ren et al., 2002) and to be required for efficient adipocyte differentiation \u003cem\u003ein vitro\u003c/em\u003e and \u003cem\u003ein vivo\u003c/em\u003e in mice (Koutnikova et al., 2003; Medina-Gomez et al., 2005; Zhang et al., 2004). Notably, in hMSC-TERT4 cells, the \u003cem\u003ePPARG1\u003c/em\u003e promoter is inactive, and only the adipocyte specific \u003cem\u003ePPARG2\u003c/em\u003e promoter is activated by the adipogenic cocktail (Extended Data Fig. 1b-c), thereby enabling us to specifically study the activation of the \u003cem\u003ePPARG2\u003c/em\u003e promoter without interference from the \u003cem\u003ePPARG1\u003c/em\u003e promoter. Similarly to \u003cem\u003ePPARG2\u003c/em\u003e, the two other genes located in the TAD, \u003cem\u003eSYN2\u003c/em\u003e and \u003cem\u003eTIMP4\u003c/em\u003e, are selectively induced during adipocyte differentiation, whereasthe expression of genes outside the TAD borders, \u003cem\u003eTAMM41\u003c/em\u003e and \u003cem\u003eTSEN2\u003c/em\u003e, are not affected by differentiation (Fig. 1c).\u003c/p\u003e\n\u003cp\u003eCollectively, these results show that the \u003cem\u003ePPARG2\u003c/em\u003e \u003cem\u003ecis\u003c/em\u003e-regulatory enhancer network is demarcated prior to cellular commitment and then activated in response to the adipogenic cocktail. Importantly, the regulatory functions of these enhancers appear to be contained within the TAD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe connectivity and activity of the \u003cem\u003ePPARG\u003c/em\u003e enhancer community is increased during early adipogenesis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIntegration of our previously published ECHi-C, MED1 ChIP-seq, and DNase-seq data obtained during adipogenesis and osteogenesis of hMSC-TERT4 cells (Madsen et al., 2020; Rauch et al., 2019) revealed marked activation of several putative enhancers between 1 and 3 days of adipocyte differentiation (Fig. 2a) and showed that these enhancers are part of a highly interconnected enhancer (HICE) community (Fig. 2a top panel). We selected 9 putative enhancers in three regions of interest that are framing the \u003cem\u003ePPARG\u003c/em\u003e gene and are already engaged in the enhancer community in the undifferentiated state. Enhancers were named according to their position (in kb) relative to the \u003cem\u003ePPARG2\u003c/em\u003e TSS. Using DNase hypersensitive sites (Fig. 2a) and MED1 occupancy (Fig. 2a) and H3K27 acetylation (Extended Data Fig. 2a) as a proxy for enhancer activity, we showed that the two distal upstream enhancers (E-130, E-123) are already active in undifferentiated cells and remain active throughout both adipocyte and osteoblast differentiation. In contrast, the two promoter-proximal enhancers (E-1, E+2) are activated \u003cem\u003ede novo\u003c/em\u003e by day 1 of adipogenesis, and the 5 distal downstream enhancers (E+89, E+102, E+107, E+111, E+125), collectively classified as a super-enhancer by the ROSE algorithm (Lov\u0026eacute;n et al., 2013; Warren et al., 2013) (Extended Data Fig. 2b), are activated \u003cem\u003ede novo\u003c/em\u003e by day 1-3 (Fig. 2a, Extended Data Fig. 2a). During the time course of adipogenesis, none of the nine enhancers bind CTCF, except for E+107, where CTCF binding is observed from day 1 of adipogenesis (Fig. 2a).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo specifically investigate the temporal dynamics of chromatin interactions in the \u003cem\u003ePPARG\u003c/em\u003e enhancer community during differentiation, we performed Capture-C in the progenitor state and on day 10 of adipogenesis using the nine putative enhancers and the \u003cem\u003ePPARG2\u003c/em\u003e TSS as bait. We focused on pairwise interactions within the specific region of the \u003cem\u003ePPARG\u003c/em\u003e locus (chr3: 12.1-12.6 Mb), excluding interactions occurring within \u0026plusmn; 2kb of a viewpoint. To compare interactions across samples, we normalized the unique interaction counts within the region (\u003cem\u003ecis\u003c/em\u003e interactions) to 50,000 counts per sample and viewpoint (Supplementary Information Fig. 1). In line with the dynamics of the ECHi-C-based interactions seen in the \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003eHICE community (Fig. 2a), Capture-C showed that the main long-distance interaction in the undifferentiated state is between the upstream enhancer E-123 and the super-enhancer constituent E+125, which thereby \u0026ldquo;pinches\u0026rdquo; the locus (Fig. 2b-d, Extended Data Fig. 3a). In addition, there are strong interactions between the two upstream enhancers E-130 and E-123 (Fig. 2c). Interestingly, we also observed strong connections between the super-enhancer constituents, which are primed but not yet remodeled or active (Fig. 2c). The most dramatic changes in connectivity during adipogenesis are the strong inductions of interactions between the \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003epromoter-proximal regions and the downstream super-enhancer constituents (Fig. 2b, Fig. 2c).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTaken together, the connectivity of the \u003cem\u003ePPARG\u003c/em\u003e enhancer community is increased during adipogenesis, concomitantly with the sequential activation of putative enhancers near the TSS as well as in the downstream super-enhancer region (Fig. 2d).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePPARG2\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;activation requires cooperativity between many non-redundant enhancers\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo investigate the role of individual enhancers in chromatin topology and activation of \u003cem\u003ePPARG2\u003c/em\u003e transcription, we generated a hMSC-TERT4-based clonal cell line with inducible Cas9 (iCas9; hMSC-TERT4/iCas9-4) (Fig. 3a, Extended Data Fig. 4a). hMSC-TERT4 cells were transduced with Lenti-iCas9-Neo (Cao et al., 2016) encoding \u0026nbsp;a flag-tagged human codon-optimized \u003cem\u003eS. pyogenes\u0026nbsp;\u003c/em\u003eCas9 coupled to green fluorescent protein (GFP) by a \u003cem\u003ePorcine teschovirus\u003c/em\u003e-1 2A (P2A) self-cleavage peptide (Extended Data Fig. 4b-c). In this system, the expression of GFP-coupled Cas9 can be transiently induced by doxycycline (Extended Data Fig. 4b-d), thereby minimizing off-target effects from constitutive Cas9 expression. The differentiation potential of the selected hMSC-TERT4/iCas9-4 clone is comparable to wild-type hMSC-TERT4 cells (Extended Data Fig. 4e). The hMSC-TERT4/iCas9-4 cell line was transfected with pairs of gRNAs designed based on the position and width of the corresponding DNase-seq peak to excise each of the putative enhancers (Fig. 3a, Extended Data Table 3, Supplementary Information Fig. 2). For each deletion, we selected three independent clones. As controls, we used three independent clones not transfected with gRNA (\u0026lsquo;No gRNA\u0026rsquo;). In addition, we generated three clones transfected with control gRNA (\u0026lsquo;Ctr gRNA\u0026rsquo;) targeting intron 1 of the \u003cem\u003ePPP1R12C\u0026nbsp;\u003c/em\u003egene, which is suggested to be a \u0026ldquo;safe harbor\u0026rdquo; for incorporation of adeno-associated virus and thus genetic modifications (DeKelver et al., 2010)\u003cem\u003e.\u003c/em\u003e For each clone, the zygosity and the size of the deletion were confirmed by Sanger sequencing of regions flanking the deleted sites (Extended Data Table 3, Supplementary Information Table 3).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe control and enhancer deletion clones were induced to undergo adipocyte differentiation, RNA was harvested on day 10 and used for quantification of transcripts by qPCR. Interestingly, deletion of several individual enhancers, i.e., the promoter-proximal enhancer E-1 and E+2, as well as the super-enhancer constituents E+102, E+107, and E+111, results in a significantly decreased expression of \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003eand\u003cem\u003e\u0026nbsp;SYN2\u003c/em\u003e, with the deletion of E+102 almost completely preventing the induction of \u003cem\u003ePPARG\u003c/em\u003e2 (Fig. 3b). Deletion of the distal upstream E-123 is associated with a slight decrease in \u003cem\u003ePPARG2\u003c/em\u003e expression, whereas deletion of the distal upstream E-130 and distal downstream E+125 enhancers do not affect \u003cem\u003ePPARG2\u003c/em\u003e (Fig. 3b). These results indicate an important and non-redundant role of the promoter-proximal enhancers E-1 and E+2, as well as the super-enhancer constituents E+102, E+107, and E+111. Inspection of epigenetic profiles of these enhancers shows that they are all remodeled and activated (Fig. 2a) and become much more interconnected (Fig. 2c) during adipogenesis. Furthermore, during adipogenesis, all super-enhancer constituents, E+102 in particular, become more connected to the \u003cem\u003ePPARG2\u003c/em\u003e TSS (Fig 2b). Consistent with PPAR\u0026gamma; being a master regulator of adipogenesis and a direct activator of most genes involved in lipid handling and metabolism in adipocytes (Lefterova et al., 2014), there is a strong correlation between \u003cem\u003ePPARG2\u003c/em\u003e expression and lipid accumulation in the different enhancer-deleted clones at day 9 of adipocyte differentiation (Pearson R=0.99, Fig. 3c; Extended Data Fig. 4f). In summary, at least six out of nine putative enhancers in the \u003cem\u003ePPARG\u003c/em\u003e HICE community are required for the full induction of \u003cem\u003ePPARG2\u003c/em\u003e expression and ultimately adipocyte differentiation. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNotably, none of the enhancer deletions affect the expression of \u003cem\u003eTAMM41\u003c/em\u003e and \u003cem\u003eTSEN2\u003c/em\u003e, which are located outside the TAD, or the\u003cem\u003e\u0026nbsp;\u003c/em\u003eexpression of\u003cem\u003e\u0026nbsp;TIMP4\u0026nbsp;\u003c/em\u003ein\u003cem\u003e\u0026nbsp;\u003c/em\u003ethe TAD (Fig. 3b). By contrast, the expressions of \u003cem\u003eSYN2\u003c/em\u003e and \u003cem\u003ePPARG2\u003c/em\u003e are similarly affected by the individual enhancer deletions (Fig. 3b). The canonical \u003cem\u003eSYN2\u003c/em\u003e promoter region is outside the TAD and does not seem to interact with the \u003cem\u003ePPARG\u003c/em\u003e community enhancers on day 10 of adipogenesis (Fig. 1b), suggesting \u003cem\u003eSYN2\u003c/em\u003e might be transcribed from an alternative TSS. Moreover, both controls, \u0026lsquo;No gRNA\u0026rsquo; and \u0026lsquo;Ctr gRNA\u0026rsquo;, exhibit comparable levels of expression of \u003cem\u003ePPARG\u003c/em\u003e2 and neighboring genes (Fig. 3b). Therefore, for subsequent experiments, we proceeded with \u0026lsquo;No gRNA\u0026rsquo; as the control, unless otherwise stated.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eDeletion of key enhancers affects the connectivity in the \u003cem\u003ePPARG\u003c/em\u003e locus\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo explore the role of individual enhancers in the development of local topology in the \u003cem\u003ePPARG\u003c/em\u003e locus, we performed Capture-C in controls and in enhancer-deleted clones before induction of differentiation (day 0) and following exposure to the adipogenic cocktail for 10 days. Results from \u0026lsquo;No gRNA\u0026rsquo; and \u0026lsquo;Ctr gRNA\u0026rsquo; controls correlated well (Extended Data Fig. 3b) and were therefore considered as combined controls (\u003cem\u003en\u003c/em\u003e=4) for all statistical tests. Interactions where one of the interacting elements was deleted were disregarded. In the progenitor cell state, where neither the enhancers in the promoter-proximal region nor the super-enhancer constituents are remodeled, as assessed by DNase-seq (Fig. 2a), or active, as assessed by H3K27ac and MED1 ChIP-seq (Fig. 2a, Extended Data Fig. 2a), only a few enhancer interactions are affected by the deletion of other enhancers (Fig. 4a). Interestingly, however, the deletion of E+2, E+107 or E+125 lead to decreased interaction between E-1 and E+89, while that between E+102 and the \u003cem\u003ePPARG2\u003c/em\u003e TSS is increased. This indicates that the E+102-TSS connectivity may be limited by other super-enhancer constituents and E+2, possibly by some competitive mechanisms. Overall, these data indicate that the connectivity of the \u003cem\u003ePPARG\u003c/em\u003e HICE community at the progenitor stage is relatively resistant to the deletion of constituents. Moreover, the structural changes that occur upon enhancer deletion at this stage do not correlate well with the ability of the enhancers to induce \u003cem\u003ePPARG2\u003c/em\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAt day 10 of adipogenesis, when the community is more connected, many enhancer interactions are affected by the deletion of other enhancers. The most notable changes in enhancer interaction occur in response to the deletion of enhancers that are important for the induction of \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003eexpression (Fig. 4a), in particular E+102 (Fig. 4b) and E+2 (Fig. 4c). Deletion of E+102 inherently ablates the strongest and most differentiation-dependent interactions between the super-enhancer and the TSS; however, it also leads to a general destabilization of the connectivity between super-enhancer constituents (e.g., the interactions between E+89 and E+107; and between E+107 and E+111) as well as a decrease in the interactions between other super enhancer constituents and the TSS and promoter proximal enhancers (e.g., interactions between E+89 and TSS and E-1; and between E+107 and E+2) (Fig. 4b). Deletion of E+2 leads to a dramatic decrease in the interaction between the super-enhancer constituents and the TSS as well as a decreased connectivity of E+111 with E+107 and with promoter-proximal E-1, respectively (Fig. 4b). Interestingly, however, the deletion of E+2 also increases the interaction frequency between E+102 and E-123 and between E+102 and E-107 (Fig. 4c). This indicates that E+2 may help direct the super-enhancer to interact with the TSS and in doing so may compete with other super-enhancer constituents for interactions with E+102.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTaken together, while many genomic interactions in the \u003cem\u003ePPARG\u003c/em\u003e locus are unaffected by the deletion of individual enhancers, deletion of E+102 or E+2 interferes with adipogenesis-associated stabilization of interactions in the \u003cem\u003ePPARG\u003c/em\u003e HICE community, specifically between the \u003cem\u003ePPARG2\u003c/em\u003e proximal regions and the distal downstream super-enhancer (Fig. 4d).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEnhancers in the\u003cem\u003e\u0026nbsp;PPARG\u003c/em\u003e locus display wide-spread functional crosstalk\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo investigate whether the enhancers within the \u003cem\u003ePPARG\u003c/em\u003e enhancer community affect the activity of each other, we assessed MED1 occupancy in individual clones at day 10 of adipocyte differentiation, where all enhancers are active in control cells (Fig. 2a). Interestingly, deletion of E+102 leads to decreased MED1 recruitment to other enhancers activated during adipogenesis, i.e. \u0026nbsp;E-1, E+2, E+89, E+107, E+111 and E+125, as assessed by ChIP-PCR (Fig. 5a, Extended Data Fig. 5a) as well as ChIP-seq (Fig S6b). Similar locus-wide effects, but with a smaller magnitude, are observed in response to deletion of other enhancers that are important for \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003eexpression (Fig. 3b, Fig. 5a). Thus, at day 10 of adipogenesis, MED1 occupancy is not only lost at the deleted enhancer or its closest neighbors but is also lost at other enhancers in the locus. The summed MED1 occupancy at the 9 putative enhancers correlated well (Pearson R=0.83, Fig. 5b) with \u003cem\u003ePPARG2\u003c/em\u003e expression, indicating that transcription of the locus is determined by the sum of enhancer activity.\u003c/p\u003e\n\u003cp\u003eWe also assessed MED1 ChIP-seq at day 3 of adipogenesis, the earliest time point with significant MED1 occupancy at the super-enhancer (Fig. 2a). Similar to MED1 ChIP-qPCR data at day 10 of adipogenesis (Fig. 5a, Extended Data Fig. 5b), this showed that the most pronounced effect of E+102 deletion is a lower MED1 signal at the other super-enhancer constituents, E+89, E+107, E+111, and E+125 (Fig. 5c), whereas the most pronounced effect of E+2 deletion is a decreased MED1 occupancy at the neighboring promoter-proximal E-1 and the distal E+125 (Fig. 5c). Importantly, the effects of both E+102 and E+2 deletions are restricted to the \u003cem\u003ePPARG\u003c/em\u003e enhancer community (Extended Data Fig. 5c-d).\u003c/p\u003e\n\u003cp\u003eCollectively, these data indicate profound cooperativity between enhancers in the \u003cem\u003ePPARG\u003c/em\u003e locus, in particular between those that interact undifferentiated cells (Fig. 5d).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eenhancer network is activated through extensive enhancer crosstalk\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe have previously shown that interaction between enhancers occupied by the same transcription factor is correlated with increased occupancy of that factor at both enhancers, a phenomenon which we termed \u0026lsquo;cross-interaction stabilization\u0026rsquo; (CIST) (Madsen et al., 2020), and we speculated that such mechanisms could be involved in \u003cem\u003ecis\u003c/em\u003e cooperativity between enhancers in the \u003cem\u003ePPARG\u003c/em\u003e locus. C/EBP\u0026beta; is known to be a key driver of early adipogenesis and has been proposed to directly activate \u003cem\u003ePparg\u003c/em\u003e gene expression in mouse preadipocytes (Siersb\u0026aelig;k et al., 2014). Consistent with that, C/EBP motif activity and C/EBP\u0026beta; expression are rapidly increased in hMSC-TERT4 cells in response to the adipogenic cocktail (Fig. 6a). C/EBP\u0026beta; ChIP-seq analysis showed that the distal upstream enhancers, E-130 and E-123, are bound by C/EBPb in the progenitor state, while the promoter-proximal enhancers, E-1 and E+2 and super-enhancer constituents E+102, E+107, E+111 recruit C/EBP\u0026beta; at day 1 of differentiation (Fig. 6b). Notably, the binding of C/EBP\u0026beta; to the super-enhancer at day 1 (Fig. 6b) precedes chromatin remodeling (Fig. 2a), suggesting that C/EBP\u0026beta; might serve as a pioneering factor in the super-enhancer region. This is consistent with similar analyses in the 3T3-L1 preadipocyte cell line, where C/EBP\u0026beta; binding also precedes chromatin remodeling (Siersbaek et al., 2011).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo explore whether early C/EBPb occupancy in the \u003cem\u003ePPARG\u003c/em\u003e HICE community is affected by the deletion of other C/EBP\u0026beta; bound interacting enhancers, we performed C/EBPb ChIP-seq in control and E+2 and E+102 deletion cell lines at day 1 of adipogenesis (Fig. 6c). Interestingly, and consistent with CIST, deletion of E+102 leads to significantly decreased C/EBP\u0026beta; occupancy at super-enhancer constituents E+107 and E+125, which are connected to E+102 at the progenitor state (Fig. 6c, Fig. 2c). \u0026nbsp;By contrast, C/EBP\u0026beta; binding at the distal upstream enhancers, E-130 and E-123, and at the promoter-proximal E-1 and E+2 are not significantly affected (Fig. 6c). Deletion of E+2 decreases C/EBP\u0026beta; occupancy at distal super-enhancer constituent E+125, the only enhancer that interact with E+2 at the progenitor state (Fig. 6c, Fig. 2c). Importantly, these effects are restricted to the \u003cem\u003ePPARG\u003c/em\u003e enhancer community (Extended Data Fig. 6a), as the C/EBP\u0026beta; ChIP-seq signal in the E+2 and E+102 deletion cell lines is comparable to control lines outside the \u003cem\u003ePPARG\u003c/em\u003e locus (Extended Data Fig. 6a) and at a genome-wide level (Extended Data Fig. 6b). Collectively, and in line with the CIST model, these findings indicate that interactions between enhancers bound by C/EBP\u0026beta; increase the occupancy of C/EBP\u0026beta; (Fig. 6d).\u003c/p\u003e\n\u003cp\u003ePPAR\u0026gamma; has been proposed to positively feedback regulate its own expression. First, PPAR\u0026gamma; binds directly to regulatory elements in the \u003cem\u003ePPARG2\u003c/em\u003e promoter in mice and humans (Schmidt et al., 2011). Second, PPAR\u0026gamma; is also involved in the induction of other transcription factors, including C/EBP\u0026alpha;, which may feedback to regulate the \u003cem\u003ePPARG2\u003c/em\u003e promoter (Salma et al., 2006; Siersb\u0026aelig;k et al., 2014). The importance of this feedback regulation is supported by early loss-of-function studies (Lefterova et al., 2008). Indeed, ChIP-seq mapping of PPAR\u0026gamma; binding during hMSC-TERT4 adipogenesis showed that PPAR\u0026gamma; binds to E-1 and E+102 at day 3 of adipogenesis (Fig. 6e), both of which have moderately conserved PPAR response elements (PPREs) (Extended Data Fig. 6c). At later stages of adipogenesis, PPAR\u0026gamma; occupancy is also gained at E+107 and E+2, although no consensus PPRE is found in E+2. In summary, three out of the four most important enhancers for regulating \u003cem\u003ePPARG2\u003c/em\u003e expression, MED1 occupancy, and lipid accumulation exhibit direct binding by PPAR\u0026gamma; itself. Moreover, C/EBP\u0026alpha; ChIP-seq at day 10 of differentiation showed that C/EBP\u0026alpha; binds to the promoter-proximal E-1, E+2, and to super-enhancer constituents E+102, E+107, E+111 by day 10 of adipogenesis, indicating that C/EBP\u0026alpha; is also involved in a positive feedback loop (Fig. 6e). Thus, a strong enhancer in the \u003cem\u003ePPARG\u003c/em\u003e locus may affect the activity of other enhancers not just in \u003cem\u003ecis\u003c/em\u003e but also in \u003cem\u003etrans\u003c/em\u003e by contributing to increased expression of PPAR\u0026gamma; which then either directly activates other \u003cem\u003ePPARG\u003c/em\u003e enhancers or the expression of other adipogenic transcription factors (Fig. 6f). Interestingly, we also noticed a strong binding of PPAR\u0026gamma; by day 3 near the TSS of a shorter predicted isoform of \u003cem\u003eSYN2\u0026nbsp;\u003c/em\u003e(Fig. 6e), suggesting that the co-regulation of \u003cem\u003eSYN2\u003c/em\u003e and \u003cem\u003ePPARG2\u003c/em\u003e during adipogenesis might be mediated in part in \u003cem\u003etrans\u003c/em\u003e by PPAR\u0026gamma; activation of promoter proximal enhancers in the \u003cem\u003eSYN2\u003c/em\u003e locus\u003cem\u003e.\u003c/em\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTaken together, the enhancers in the \u003cem\u003ePPARG\u003c/em\u003e HICE community show clear evidence of cis cooperativity at early stages of differentiation prior to expression of \u003cem\u003ePPARG\u003c/em\u003e; however, at later time points, crosstalk in trans by positive feedback regulation is likely to play a major role as well.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eKey \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003eenhancers overlap with non-coding genetic variants associated with metabolic disease traits\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGiven the importance of well-functioning adipocytes for metabolic health and the obligate role of PPAR\u0026gamma; in adipocyte differentiation and function (Hagberg \u0026amp; Spalding, 2024; Lefterova et al., 2014), we rationalized that if the enhancers identified \u003cem\u003ein vitro\u003c/em\u003e are also important \u003cem\u003ein vivo\u003c/em\u003e, they would overlap with metabolic disease-associated genetic variants. We queried the \u003cem\u003ePPARG\u003c/em\u003e TAD (chr3: 12,174,868 - 12,526,858) and extracted 796 haplotypes associated with a series of cardiometabolic traits comprising in total 6,030 variants in high linkage disequilibrium (LD) (r2\u0026gt; 0.8, CEU) from GWAS in the public domain (Fig. 7a). The vast majority of these genetic variants associated with common complex metabolic traits map to non-coding elements in the \u003cem\u003ePPARG\u003c/em\u003e locus. Because any given variant in a haplotype could be functional and, therefore, mediate the association with metabolic disease, we used the Enformer model (Žiga Avsec, Vikram Agarwal, et al., 2021), a state-of-the-art sequence-based deep learning approach for the prediction of non-coding variant effects on gene expression and chromatin states. We predicted the variant effects on \u003cem\u003ePPARG2\u003c/em\u003e transcript expression as well as P\u003cem\u003ePARG\u003c/em\u003e enhancer regions and ranked the variants based on their positive or negative effects on the target regions (Fig. 7b, Tables S1, S2). The analysis revealed that 1,759 out of 3,596 variants have predicted significant effects on \u003cem\u003ePPARG2\u003c/em\u003e gene expression.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNotably, 11 out of the 40 top-scoring variants with predicted (positive and negative) effects on \u003cem\u003ePPARG2\u003c/em\u003e signal and/ or enhancer accessibility, overlap with enhancers identified in the hMSC-TERT4 model system (Fig. 7b, Fig. 7c, Extended Data Table 1-2). In particular, many of these genetic variants are located in the three most important enhancers, E+102, E+2, and E-1 (Fig. 7b, Fig. 7c, Extended Data Table 1-2), indicating that these enhancers are also important for the regulation of \u003cem\u003ePPARG\u003c/em\u003e expression in human physiology. One of the top non-coding variants, rs7647481 A/G, located in E-1, is a common variant associated with decreased risk of developing type 2 diabetes. Using label-free quantitative LC\u0026ndash;MS/MS proteomics, this variant has previously been shown to enhance the binding of YY1 leading to increased activation of the \u003cem\u003ePPARG2\u003c/em\u003e promoter (Lee et al., 2017).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAnother interesting, low-frequency variant, rs181025382 G/A (MAF=0.0001, CEU) associated with lipid traits including decreased levels of total cholesterol and non-HDL cholesterol resides in E+102 and is predicted to negatively impact both \u003cem\u003ePPARG2\u003c/em\u003e expression and enhancer accessibility level (Fig. 7c). Interestingly, this variant is located in a peroxisome proliferator response element (PPRE) in E+102, which is bound by PPAR\u0026gamma; from day 3 of differentiation (Fig. 6e). In the minor allele variant, the A in the spacer between the two half-sites of the direct repeat is replaced with a less optimal G (Fig. 7d). To estimate the effect of a G in this position on the binding affinity of the PPAR\u0026gamma;:RXR heterodimer, we took advantage of the recently published nuclear receptor sequence specific binding preferences for binding to naked DNA in the presence or absence of agonists based on high throughput SELEX data (Bhimsaria et al., 2023). In E+102, the minor variant with a G in the spacer showed an approximately\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e3-fold lower affinity for PPAR\u0026gamma;:RXR compared with the major variant, which has an A in that position (Fig. 7d). The synthetic agonist rosiglitazone slightly increased the affinity of PPAR\u0026gamma;:RXR for both PPRE\u0026nbsp;\u003c/p\u003e\n\u003cp\u003evariants (Fig. 7d), consistent with previous ChIP-seq data (Haakonsson et al., 2013). To investigate the biological relevance of a 3-fold difference in binding affinity for the PPAR\u0026gamma;:RXR heterodimer, we compared PPAR\u0026gamma;:RXR affinity (\u0026plusmn; rosiglitazone) between the 1,000 top PPAR\u0026gamma;-bound enhancers and 1000 PPAR\u0026gamma;-unbound enhancers from our datasets derived from hMSC-TERT4 (Extended Data Fig. 7a). Importantly, PPAR\u0026gamma; generally exhibits about 3-fold higher affinity for bound sites compared to unbound enhancers, with affinity being higher for rosiglitazone-activated PPAR\u0026gamma; (Extended Data Fig. 7a). This indicates that a 3-fold difference in naked DNA affinity may have major functional implications for binding of PPAR\u0026gamma; to sites in chromatin.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo further interrogate PPAR\u0026gamma; binding to chromatin in hMSC-TERT4 cells, we applied BPNet (Ž Avsec et al., 2021), a deep learning model to predict base-resolution binding syntax from DNA sequences within PPAR\u0026gamma; binding peaks in hMSC-TERT4 (Fig. 7e). We used \u003cem\u003ei\u003c/em\u003e\u003cem\u003en silico\u0026nbsp;\u003c/em\u003emutagenesis to investigate the effect of rs181025382 on PPAR\u0026gamma;:RXR binding in E+102 and found that substitution of the major allele A with minor allele G leads to -0.12 log fold change (Fig. 7e). Further, we used TF-MoDIScO for motif discovery within identified seqlets in PPAR\u0026gamma;-bound sequences. By inferring the data with previously annotated motifs from JASPAR database, we identified PPAR\u0026gamma;:RXR as a top \u003cem\u003ede novo\u003c/em\u003e motif matching with PPAR\u0026gamma;:RXR (MA0065.2) motif (FDR= 1.3e-10, Supplementary Information Table 6) (Fig. 7f.) \u0026nbsp;To ensure the predicted decrease in binding upon A to G substitution was not specific to E+102, we identified PPAR\u0026gamma; binding sites genome-wide that contain the PPAR\u0026gamma; motif with an A at the 9\u003csup\u003eth\u003c/sup\u003e position. At all these sites, we mutated the A to a G and predicted the difference in binding using the trained BPNet model. This showed -0.09 log fold change in PPAR\u0026gamma;:RXR binding affinity genome-wide upon substitution of A with G, similar to results for rs181025282 (Extended Data Fig. 7b-c). This suggests that the G variant of rs181025382, is likely to interfere with the ability of PPAR\u0026gamma; to activate this enhancer, thereby reducing the positive feedback activation of PPAR\u0026gamma; on \u003cem\u003ePPARG2\u003c/em\u003e expression.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur results show that \u003cem\u003ePPARG\u003c/em\u003e enhancers that are critical for \u003cem\u003ein vitro\u003c/em\u003e activation of \u003cem\u003ePPARG2\u003c/em\u003e in response to adipogenic inducers overlap with functional non-coding genetic variants associated with metabolic disease. The genetic variants in these enhancers are among the top-scoring variants predicted to affect \u003cem\u003ePPARG2\u003c/em\u003e expression. These results suggest that the enhancers identified in our hMSC-TERT4 model system may be important for human physiology. The finding that one of these notably low-frequency, non-coding genetic variants decreases the binding affinity of PPAR\u0026gamma;:RXR for the most important enhancer constituent in the super-enhancer further supports the notion that positive feedback activation of the locus plays an important role in the activation of \u003cem\u003ePPARG2\u003c/em\u003e expression during adipogenesis, a function relevant to cardiometabolic disease in humans.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eHere, we show that the activation of the \u003cem\u003ePPARG2\u003c/em\u003e promoter during adipogenesis is driven by a locus-wide HICE community consisting of upstream enhancers, promoter-proximal enhancers as well as a downstream super-enhancer. We show that the enhancers function in a non-redundant and highly cooperative manner involving early cooperativity in \u003cem\u003ecis\u003c/em\u003e between the enhancers as well as feedback activation of the enhancers by the product of the locus, PPARγ, and possibly other transcription factors, such as C/EBPα, induced by PPARγ. Disease-associated genetic variants that are predicted to affect \u003cem\u003ePPARG2\u003c/em\u003e expression and enhancer accessibility overlap with the most important enhancers in our hMSC-TERT4 model system, thereby supporting the importance of these enhancers for human physiology and disease.\u003c/p\u003e\n\u003cp\u003eWe have previously shown that in the multipotent human mesenchymal stromal cell line, hMSC-TERT4, genes specific for lineage fates, such as the adipocyte, osteoblast, myoblast and chondrocyte lineages, tend to be marked by preestablished HICE communities already at the progenitor state. In response to adipocyte and osteoblast differentiation cocktails, the connectivity of the respective lineage specific HICE communities are strengthened, coinciding with the activation of transcription of the respective target genes\u0026nbsp;(Madsen et al., 2020). Similarly, findings from other studies have indicated a priming role of the chromatin 3D structure at the progenitor state, which shifts to an instructive role during \u003cem\u003eDrosophila\u0026nbsp;\u003c/em\u003e(Pollex et al., 2024) and mouse (Chen et al., 2024) embryogenesis. In line with this, we show that the \u003cem\u003ePPARG\u003c/em\u003e locus is primed by pre-established enhancer-enhancer interactions in the silent state in progenitors, suggesting a permissive nature of these interactions. In response to the differentiation cocktail, the strength of the existing interactions increases, especially those between the super-enhancer constituents and the promoter-proximal enhancers and \u003cem\u003ePPARG2\u003c/em\u003e TSS. Furthermore, new interactions are formed between the promoter-proximal enhancers and the downstream super-enhancer constituents. In the silent progenitor state, the \u003cem\u003ePPARG\u003c/em\u003e locus is also primed by H3K4me1, similar to what has been observed for other developmental genes, such as \u003cem\u003ePDX1\u003c/em\u003e during pancreatic differentiation (Wang et al., 2015), and many early developmental genes in human ESCs\u0026nbsp;(Rada-Iglesias et al., 2011). This epigenetic and 3D chromatin architecture of the\u0026nbsp;\u003cem\u003ePPARG\u003c/em\u003e locus is likely to prime rapid lineage-specific activation in response to adipogenic cues.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUsing inducible CRISPR/Cas9 to individually delete 9 enhancers that are engaged in both pre-established and induced interactions with each other and with the \u003cem\u003ePPARG2\u003c/em\u003e promoter, we show that 6 of the 9 enhancers are required for efficient activation of the \u003cem\u003ePPARG2\u003c/em\u003e promoter. Although, deletion of\u0026nbsp;upstream enhancer E-130 interferes with overall connectivity in the locus in the progenitor state, there is no effect of deletion of this enhancer on \u003cem\u003ePPARG2\u003c/em\u003e expression, indicating that this enhancer may primarily be contributing to overall locus connectivity in the progenitor state. By contrast, individual deletions of most of the enhancers that are activated during differentiation, i.e., the two promoter-proximal enhancers (E+2 and E-1) as well as 3 of the 5 super-enhancer constituents (E+102, E+107, E+111), lead to compromised induction of the \u003cem\u003ePPARG2\u003c/em\u003e promoter. In particular, deletion of the super-enhancer constituent E+102 completely prevents the activation of the promoter. The difference in importance of super-enhancer constituents cannot be explained solely by connectivity, as all constituents interact with the promoter-proximal enhancers and the \u003cem\u003ePPARG2\u003c/em\u003e TSS. However, the importance of the enhancers appears to be closely linked to their effect on the activity of other enhancers in the locus.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePrevious studies have proposed enhancer cooperativity in \u003cem\u003ecis\u003c/em\u003e, particularly in the context of the super-enhancer constituents (Bahr et al., 2018; Choi et al., 2021; Huang et al., 2018; Oudelaar et al., 2018), where cooperativity has been suggested to be related to condensate formation driven by multivalent interactions between intrinsically disordered regions (IDRs) of co-factors and transcription factors (Cho et al., 2018; Du et al., 2024; Lyons et al., 2023; Sabari et al., 2018; Trojanowski et al., 2022). Moreover, QTL analyses have shown that genomic variations within enhancers affect the activity of other enhancers (Grubert et al., 2015). However, documented cooperativity by loss-of-function studies has been limited to enhancers within close genomic proximity, such as enhancer constituents in super-enhancers (Blayney et al., 2023; Huang et al., 2018; Thomas et al., 2021).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn this work we demonstrate enhancer cooperativity in \u003cem\u003ecis\u0026nbsp;\u003c/em\u003ewithin a 3D enhancer community that spans the \u003cem\u003ePPARG\u003c/em\u003e locus and is sequentially activated during adipogenesis.\u0026nbsp;First, deletion of E+102 or E+2 affects MED1 occupancy at day 3 of adipogenesis at multiple other enhancers in the community, suggesting that enhancers with strong recruitment of Mediator can stimulate Mediator occupancy and activity of nearby enhancers, possibly through condensate formation. Second, the enhancers with the strongest effect on other enhancers and \u003cem\u003ePPARG2\u003c/em\u003e expression are characterized by early recruitment of C/EBPβ at day 1 of differentiation, and deletion of E+2 or E+102 leads to reduced C/EBPβ occupancy at interacting enhancers. Notably, this is consistent with our CIST model, showing strong correlation between binding of homotypic transcription factors at interacting enhancers (Madsen et al., 2020). The underlying mechanism of CIST is unclear but may involve condensate formation and/or more specific protein-protein interactions. C/EBPb\u0026nbsp;has been shown to be an important transcription factor in early adipogenesis\u0026nbsp;(Siersbæk et al., 2014; Tanaka, 1997). Here we show that C/EBPβ appears to act as a pioneer transcription factor at the \u003cem\u003ePPARG\u003c/em\u003e super-enhancer by binding to inaccessible enhancer constituents at day 1 of adipogenesis. The structural plasticity of IDRs and post-translational modifications (PTM) of C/EBPb\u0026nbsp;(Dittmar et al., 2019)\u0026nbsp;may facilitate its interaction with DNA on nucleosomes, as well as its interaction with chromatin remodeling complexes and histone modifiers to drive remodeling of the super-enhancer constituents. These features of C/EBPβ may also contribute to enhancer cooperativity in \u003cem\u003ecis\u003c/em\u003e.\u003c/p\u003e\n\u003cp\u003eImportantly, since PPARγ binds to several of its own enhancers at later time points of differentiation and induces C/EBPα and other transcription factors that might also stimulate activity of enhancers in the \u003cem\u003ePPARG\u003c/em\u003e locus, the enhancers may also cooperate in \u003cem\u003etrans\u003c/em\u003e once PPARγ is induced. Thus, at later time points of differentiation (after day 3) it is difficult to distinguish between enhancer crosstalk in \u003cem\u003ecis\u003c/em\u003e and \u003cem\u003etrans\u003c/em\u003e. Feedback activation has been demonstrated for other core developmental transcription factors (Saint-André et al., 2016), suggesting that feedback activation may also contribute to, and needs to be considered, when evaluating \u0026nbsp;enhancer cooperativity in other loci.\u003c/p\u003e\n\u003cp\u003eThe identification of E+102 as an obligate enhancer for \u003cem\u003ePPARG2\u003c/em\u003e induction is intriguing. Deletion of E+102 largely blocks the activation of the super-enhancer, as determined by MED1 occupancy, by day 3 and day 10 of differentiation and interferes with the recruitment of C/EBPβ to the enhancer constituents. Moreover, it is the most prominent target enhancer of PPARγ feedback regulation from day 3, and it remains the most active enhancer constituent (based on MED1 occupancy) in the super-enhancer. Thus, the importance of E+102 may be due to its dual functions, acting as a seed enhancer in the super-enhancer and as a key target enhancer for feedback action by PPARγ (and possibly other adipogenic transcription factors).\u0026nbsp;Such positive feedback regulation through E+102 and other enhancers may contribute to the rapid and efficient activation of \u003cem\u003ePPARG2\u003c/em\u003e during adipogenesis and thereby the ability of PPARγ to act as a master regulator. Similar regulatory principles may be involved in driving the induction other master regulators of different lineages to ensure efficient induction of the master regulator in response to differentiation cues. Moreover, this regulatory principle may also sensitize \u003cem\u003ePPARG2\u003c/em\u003e expression to PPARγ point mutations, such as those observed in lipodystrophies (Broekema et al., 2019; Madsen et al., 2022a), thereby resulting in lower expression of PPARγ variants and further decreasing PPARγ activity.\u003c/p\u003e\n\u003cp\u003eTo link these enhancers to human physiology, we applied the sequence-based Enformer machine learning model to infer the effect of non-coding genetic variants associated with cardiometabolic traits on the \u003cem\u003ePPARG\u003c/em\u003e locus across the frequency spectrum. Importantly, we showed that 11 out of the 40 variants predicted to have the largest effect on \u003cem\u003ePPARG2\u003c/em\u003e expression overlap with key \u003cem\u003ePPARG2\u003c/em\u003e enhancers identified in hMSC-TERT4 cells, thereby supporting the \u003cem\u003ein vivo\u003c/em\u003e importance of these enhancers. Interestingly, we identified a low frequency variant associated with cardiometabolic diseases, which is located in the PPRE in the obligate enhancer, E+102. This variant is predicted to reduce PPARγ binding affinity to both naked and chromatin embedded DNA and may therefore interfere with the ability of PPARγ to feedback activate its own expression. Work from others have shown that single nucleotide polymorphism in the PPRE can affect \u003cem\u003ein vivo\u003c/em\u003e PPARγ occupancy, the response to PPARγ agonists as well as human metabolic disease risk (Soccio et al., 2015). The discovery of the PPRE variant in E+102 supports the relevance of cardiometabolic SNPs targeting PPARγ response elements. In relation to the\u003cem\u003e\u0026nbsp;PPARG\u003c/em\u003e locus, this finding supports the importance of positive feedback activation for the activation of \u003cem\u003ePPARG2\u003c/em\u003e expression.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn conclusion, our study reveals that extensive enhancer cooperativity and feedback activation plays an essential role in the activation of \u003cem\u003ePPARG2\u003c/em\u003e expression during adipogenesis of human stromal progenitor cells. The \u003cem\u003ePPARG\u003c/em\u003e locus is primed for activation in the progenitor state by a HICE community and by H3K4me1. Upon induction of differentiation, cooperativity in \u003cem\u003ecis\u003c/em\u003e between the promoter-proximal enhancers as well as enhancer constituents in the super-enhancer drive activation of the locus. At later timepoints, the product of the locus, PPARγ, feedback-activates its own expression in \u003cem\u003etrans\u003c/em\u003e. We propose that the priming of the locus as well as the elaborate cooperativity in \u003cem\u003ecis\u003c/em\u003e and \u003cem\u003etrans\u003c/em\u003e play key roles in rapidly switching on the \u003cem\u003ePPARG2\u0026nbsp;\u003c/em\u003epromoterin response to adipogenic cues. This dynamic regulation may also have profound impact on the regulation of \u003cem\u003ePPARG\u003c/em\u003e expression in response to metabolic signals. Future detailed investigations of the mechanisms by which the expression of \u003cem\u003ePPARG,\u0026nbsp;\u003c/em\u003ethe master regulator of adipogenesis and adipocyte function and a key regulator cardiometabolic health, is controlled in adipocytes and other cells in human tissues is likely to offer important insight into personalized risk assessment and treatment of cardiometabolic diseases.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe work was supported by grants from the Independent Research Fund Denmark (No 12-125524 (Sapere Aude Advanced grant) and No 2034-00335B), the Danish National Research Foundation (DNRF grant No 141) to the Center for Functional Genomics and Tissue Plasticity (ATLAS), grants from the Novo Nordisk Foundation (NNF21OC0071613; the NNF\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;BridgeheadCenter for Basic Mechanisms of Disease; and NNF18OC0033444 to the Challenge Grant Center for Adipocyte Signaling (ADIPOSIGN) the Lundbeck Foundation (grant \u0026nbsp;No R218-2016-1450), the European Union\u0026apos;s Horizon 2020 research and innovation programme (grant agreement No 860002) to the ENHPATHY Consortium, the Novo Scholarship Programme, the Fuhrmann Foundation, and the Villum Foundation through support to the Villum Center for Bioanalytical Sciences. We thank Q. Yan from Yale School of Medicine for providing the Lenti-iCas9-neo plasmid and A. M. Oudelaar from the University of Oxford for valuable clarifications on Capture-C protocols and data analysis. We are grateful to our colleagues in the Functional Genomics and Metabolism Research Unit, as well as Eileen Furlong and Tim Pollex, for critical discussions that helped improve the study. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization: A.C., M.H., H.D., A.R., J.G.S.M., M.C., S.M.; Experimental work: A.C., M.H., M.N., L.K.H., K.M., M.S.M., V.S., E.D.M.; Formal analysis, investigation and data curation: A.C, M.H., M.N., B.V.H, H.D., and J.G.S.M.; Visualization: A.C., M.H.; Writing: A.C., M.H., S.M.; Funding acquisition: S.M.; Supervision: S.M., ., A.R., J.G.S.M., M.C.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll generated sequence data will be available at the GEO repository under accession code X. In addition to the data generated for this study, raw sequencing data from RNA-sequencing, DNase-sequencing and MED1 ChIP-sequencing are available from GEO under accession code GSE113253, and raw sequencing data from ECHi-C from GEO under accession code GSE140782. Raw data used for Hi-C analysis are available from GEO under accession code GSE140782 and GSE52457.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAmemiya, H. M., Kundaje, A., \u0026amp; Boyle, A. P. (2019). The ENCODE Blacklist: Identification of Problematic Regions of the Genome. \u003cem\u003eSci Rep\u003c/em\u003e,\u003cem\u003e\u0026nbsp;9\u003c/em\u003e(1), 9354. https://doi.org/10.1038/s41598-019-45839-z\u003c/li\u003e\n \u003cli\u003eAvsec, Ž., Agarwal, V., Visentin, D., Ledsam, J. R., Grabska-Barwinska, A., Taylor, K. R., Assael, Y., Jumper, J., Kohli, P., \u0026amp; Kelley, D. R. (2021). Effective gene expression prediction from sequence by integrating long-range interactions. \u003cem\u003eNature Methods\u003c/em\u003e,\u003cem\u003e\u0026nbsp;18\u003c/em\u003e(10), 1196-1203. https://doi.org/10.1038/s41592-021-01252-x\u003c/li\u003e\n \u003cli\u003eAvsec, Ž., Weilert, M., Shrikumar, A., Krueger, S., Alexandari, A., Dalal, K., Fropf, R., McAnany, C., Gagneur, J., Kundaje, A., \u0026amp; Zeitlinger, J. (2021). Base-resolution models of transcription-factor binding reveal soft motif syntax. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;53\u003c/em\u003e(3), 354-366. https://doi.org/10.1038/s41588-021-00782-6\u003c/li\u003e\n \u003cli\u003eAvsec, Ž., Weilert, M., Shrikumar, A., Krueger, S., Alexandari, A., Dalal, K., Fropf, R., McAnany, C., Gagneur, J., Kundaje, A., \u0026amp; Zeitlinger, J. (2021). Base-resolution models of transcription-factor binding reveal soft motif syntax. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;53\u003c/em\u003e(3), 354-366. https://doi.org/10.1038/s41588-021-00782-6\u003c/li\u003e\n \u003cli\u003eBahr, C., Von Paleske, L., Uslu, V. V., Remeseiro, S., Takayama, N., Ng, S. W., Murison, A., Langenfeld, K., Petretich, M., Scognamiglio, R., Zeisberger, P., Benk, A. S., Amit, I., Zandstra, P. W., Lupien, M., Dick, J. E., Trumpp, A., \u0026amp; Spitz, F. (2018). A Myc enhancer cluster regulates normal and leukaemic haematopoietic stem cell hierarchies. \u003cem\u003eNature\u003c/em\u003e,\u003cem\u003e\u0026nbsp;553\u003c/em\u003e(7689), 515-520. https://doi.org/10.1038/nature25193\u003c/li\u003e\n \u003cli\u003eBhimsaria, D., Rodr\u0026iacute;guez-Mart\u0026iacute;nez, J. A., Mendez-Johnson, J. L., Ghoshdastidar, D., Varadarajan, A., Bansal, M., Daniels, D. L., Ramanathan, P., \u0026amp; Ansari, A. Z. (2023). Hidden modes of DNA binding by human nuclear receptors. \u003cem\u003eNature communications\u003c/em\u003e,\u003cem\u003e\u0026nbsp;14\u003c/em\u003e(1). https://doi.org/10.1038/s41467-023-39577-0\u003c/li\u003e\n \u003cli\u003eBlayney, J. W., Francis, H., Rampasekova, A., Camellato, B., Mitchell, L., Stolper, R., Cornell, L., Babbs, C., Boeke, J. D., Higgs, D. R., \u0026amp; Kassouf, M. (2023). Super-enhancers include classical enhancers and facilitators to fully activate gene expression. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;186\u003c/em\u003e(26), 5826-5839.e5818. https://doi.org/10.1016/j.cell.2023.11.030\u003c/li\u003e\n \u003cli\u003eBonev, B., Mendelson Cohen, N., Szabo, Q., Fritsch, L., Papadopoulos, G. L., Lubling, Y., Xu, X., Lv, X., Hugnot, J.-P., Tanay, A., \u0026amp; Cavalli, G. (2017). Multiscale 3D Genome Rewiring during Mouse Neural Development. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;171\u003c/em\u003e(3), 557-572.e524. https://doi.org/10.1016/j.cell.2017.09.043\u003c/li\u003e\n \u003cli\u003eBroekema, M. F., Savage, D. B., Monajemi, H., \u0026amp; Kalkhoven, E. (2019). Gene-gene and gene-environment interactions in lipodystrophy: Lessons learned from natural PPAR\u0026gamma; mutants. \u003cem\u003eBiochim Biophys Acta Mol Cell Biol Lipids\u003c/em\u003e,\u003cem\u003e\u0026nbsp;1864\u003c/em\u003e(5), 715-732. https://doi.org/10.1016/j.bbalip.2019.02.002\u003c/li\u003e\n \u003cli\u003eBugge, A., Gr\u0026oslash;ntved, L., Aagaard, M. M., Borup, R., \u0026amp; Mandrup, S. (2009). The PPARgamma2 A/B-domain plays a gene-specific role in transactivation and cofactor recruitment. \u003cem\u003eMolecular Endocrinology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;23\u003c/em\u003e(6), 794-808. https://doi.org/10.1210/me.2008-0236\u003c/li\u003e\n \u003cli\u003eCao, J., Wu, L., Zhang, S.-M., Lu, M., Cheung, W. K. C., Cai, W., Gale, M., Xu, Q., \u0026amp; Yan, Q. (2016). An easy and efficient inducible CRISPR/Cas9 platform with improved specificity for multiple gene targeting. \u003cem\u003eNucleic Acids Research\u003c/em\u003e, gkw660. https://doi.org/10.1093/nar/gkw660\u003c/li\u003e\n \u003cli\u003eChen, Z., Snetkova, V., Bower, G., Jacinto, S., Clock, B., Dizehchi, A., Barozzi, I., Mannion, B. J., Alcaina-Caro, A., Lopez-Rios, J., Dickel, D. E., Visel, A., Pennacchio, L. A., \u0026amp; Kvon, E. Z. (2024). Increased enhancer\u0026ndash;promoter interactions during developmental enhancer activation in mammals. \u003cem\u003eNature Genetics\u003c/em\u003e. https://doi.org/10.1038/s41588-024-01681-2\u003c/li\u003e\n \u003cli\u003eCho, W.-K., Spille, J.-H., Hecht, M., Lee, C., Li, C., Grube, V., \u0026amp; Cisse, I. I. (2018). Mediator and RNA polymerase II clusters associate in transcription-dependent condensates. \u003cem\u003eScience\u003c/em\u003e,\u003cem\u003e\u0026nbsp;361\u003c/em\u003e(6400), 412-415. https://doi.org/10.1126/science.aar4199\u003c/li\u003e\n \u003cli\u003eChoi, J., Lysakovskaia, K., Stik, G., Demel, C., S\u0026ouml;ding, J., Tian, T. V., Graf, T., \u0026amp; Cramer, P. (2021). Evidence for additive and synergistic action of mammalian enhancers during cell fate determination. \u003cem\u003eeLife\u003c/em\u003e,\u003cem\u003e\u0026nbsp;10\u003c/em\u003e. https://doi.org/10.7554/eLife.65381\u003c/li\u003e\n \u003cli\u003eCostanzo, M. C., Von Grotthuss, M., Massung, J., Jang, D., Caulkins, L., Koesterer, R., Gilbert, C., Welch, R. P., Kudtarkar, P., Hoang, Q., Boughton, A. P., Singh, P., Sun, Y., Duby, M., Moriondo, A., Nguyen, T., Smadbeck, P., Alexander, B. R., Brandes, M., . . . Zaghloul, N. A. (2023). The Type 2 Diabetes Knowledge Portal: An open access genetic resource dedicated to type 2 diabetes and related traits. \u003cem\u003eCell Metabolism\u003c/em\u003e,\u003cem\u003e\u0026nbsp;35\u003c/em\u003e(4), 695-710.e696. https://doi.org/10.1016/j.cmet.2023.03.001\u003c/li\u003e\n \u003cli\u003eDanecek, P., Bonfield, J. K., Liddle, J., Marshall, J., Ohan, V., Pollard, M. O., Whitwham, A., Keane, T., McCarthy, S. A., Davies, R. M., \u0026amp; Li, H. (2021). Twelve years of SAMtools and BCFtools. \u003cem\u003eGigaScience\u003c/em\u003e,\u003cem\u003e\u0026nbsp;10\u003c/em\u003e(2). https://doi.org/10.1093/gigascience/giab008\u003c/li\u003e\n \u003cli\u003eDi Giammartino, D. C., Kloetgen, A., Polyzos, A., Liu, Y., Kim, D., Murphy, D., Abuhashem, A., Cavaliere, P., Aronson, B., Shah, V., Dephoure, N., Stadtfeld, M., Tsirigos, A., \u0026amp; Apostolou, E. (2019). KLF4 is involved in the organization and regulation of pluripotency-associated three-dimensional enhancer networks. \u003cem\u003eNature Cell Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;21\u003c/em\u003e(10), 1179-1190. https://doi.org/10.1038/s41556-019-0390-6\u003c/li\u003e\n \u003cli\u003eDittmar, G., Hernandez, D. P., Kowenz-Leutz, E., Kirchner, M., Kahlert, G., Wesolowski, R., Baum, K., Knoblich, M., Hofst\u0026auml;tter, M., Muller, A., Wolf, J., Reimer, U., \u0026amp; Leutz, A. (2019). PRISMA: Protein Interaction Screen on Peptide Matrix Reveals Interaction Footprints and Modifications- Dependent Interactome of Intrinsically Disordered C/EBP\u0026beta;. \u003cem\u003eiScience\u003c/em\u003e,\u003cem\u003e\u0026nbsp;13\u003c/em\u003e, 351-370. https://doi.org/10.1016/j.isci.2019.02.026\u003c/li\u003e\n \u003cli\u003eDixon, J. R., Jung, I., Selvaraj, S., Shen, Y., Antosiewicz-Bourget, J. E., Lee, A. Y., Ye, Z., Kim, A., Rajagopal, N., Xie, W., Diao, Y., Liang, J., Zhao, H., Lobanenkov, V. V., Ecker, J. R., Thomson, J. A., \u0026amp; Ren, B. (2015). Chromatin architecture reorganization during stem cell differentiation. \u003cem\u003eNature\u003c/em\u003e,\u003cem\u003e\u0026nbsp;518\u003c/em\u003e(7539), 331-336. https://doi.org/10.1038/nature14222\u003c/li\u003e\n \u003cli\u003eDobin, A., Davis, C. A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S., Batut, P., Chaisson, M., \u0026amp; Gingeras, T. R. (2013). STAR: ultrafast universal RNA-seq aligner. \u003cem\u003eBioinformatics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;29\u003c/em\u003e(1), 15-21. https://doi.org/10.1093/bioinformatics/bts635\u003c/li\u003e\n \u003cli\u003eDu, M., Stitzinger, S. H., Spille, J.-H., Cho, W.-K., Lee, C., Hijaz, M., Quintana, A., \u0026amp; Ciss\u0026eacute;, I. I. (2024). Direct observation of a condensate effect on super-enhancer controlled gene bursting. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;187\u003c/em\u003e(2), 331-344.e317. https://doi.org/10.1016/j.cell.2023.12.005\u003c/li\u003e\n \u003cli\u003eFiorenza, C. G., Chou, S. H., \u0026amp; Mantzoros, C. S. (2011). Lipodystrophy: pathophysiology and advances in treatment. \u003cem\u003eNature Reviews Endocrinology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;7\u003c/em\u003e(3), 137-150. https://doi.org/10.1038/nrendo.2010.199\u003c/li\u003e\n \u003cli\u003eFreire-Pritchett, P., Schoenfelder, S., V\u0026aacute;rnai, C., Wingett, S. W., Cairns, J., Collier, A. J., Garc\u0026iacute;a-V\u0026iacute;lchez, R., Furlan-Magaril, M., Osborne, C. S., Fraser, P., Rugg-Gunn, P. J., \u0026amp; Spivakov, M. (2017). Global reorganisation of cis-regulatory units upon lineage commitment of human embryonic stem cells. \u003cem\u003eeLife\u003c/em\u003e,\u003cem\u003e\u0026nbsp;6\u003c/em\u003e. https://doi.org/10.7554/elife.21926\u003c/li\u003e\n \u003cli\u003eGaulton, K. J., Ferreira, T., Lee, Y., Raimondo, A., M\u0026auml;gi, R., Reschen, M. E., Mahajan, A., Locke, A., William Rayner, N., Robertson, N., Scott, R. A., Prokopenko, I., Scott, L. J., Green, T., Sparso, T., Thuillier, D., Yengo, L., Grallert, H., Wahl, S., . . . Morris, A. P. (2015). Genetic fine mapping and genomic annotation defines causal mechanisms at type 2 diabetes susceptibility loci. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;47\u003c/em\u003e(12), 1415-1425. https://doi.org/10.1038/ng.3437\u003c/li\u003e\n \u003cli\u003eGhaben, A. L., \u0026amp; Scherer, P. E. (2019). Adipogenesis and metabolic health. \u003cem\u003eNature Reviews Molecular Cell Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;20\u003c/em\u003e(4), 242-258. https://doi.org/10.1038/s41580-018-0093-z\u003c/li\u003e\n \u003cli\u003eGrubert, F., Judith, Kasowski, M., Ursu, O., Damek, Alicia, Greenside, P., Srivas, R., Doug, Pekowska, A., Heidari, N., Euskirchen, G., Huber, W., Jonathan, Carlos, Lars, Kundaje, A., \u0026amp; Snyder, M. (2015). Genetic Control of Chromatin States in Humans Involves Local and Distal Chromosomal Interactions. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;162\u003c/em\u003e(5), 1051-1065. https://doi.org/10.1016/j.cell.2015.07.048\u003c/li\u003e\n \u003cli\u003eGupta, S., Stamatoyannopoulos, J. A., Bailey, T. L., \u0026amp; Noble, W. S. (2007). Quantifying similarity between motifs. \u003cem\u003eGenome Biol\u003c/em\u003e,\u003cem\u003e\u0026nbsp;8\u003c/em\u003e(2), R24. https://doi.org/10.1186/gb-2007-8-2-r24\u003c/li\u003e\n \u003cli\u003eHaakonsson, A. K., Stahl Madsen, M., Nielsen, R., Sandelin, A., \u0026amp; Mandrup, S. (2013). Acute Genome-Wide Effects of Rosiglitazone on PPAR\u0026gamma; Transcriptional Networks in Adipocytes. \u003cem\u003eMolecular Endocrinology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;27\u003c/em\u003e(9), 1536-1549. https://doi.org/10.1210/me.2013-1080\u003c/li\u003e\n \u003cli\u003eHagberg, C. E., \u0026amp; Spalding, K. L. (2024). White adipocyte dysfunction and obesity-associated pathologies in humans. \u003cem\u003eNat Rev Mol Cell Biol\u003c/em\u003e,\u003cem\u003e\u0026nbsp;25\u003c/em\u003e(4), 270-289. https://doi.org/10.1038/s41580-023-00680-1\u003c/li\u003e\n \u003cli\u003eHuang, J., Li, K., Cai, W., Liu, X., Zhang, Y., Orkin, S. H., Xu, J., \u0026amp; Yuan, G.-C. (2018). Dissecting super-enhancer hierarchy based on chromatin interactions. \u003cem\u003eNature communications\u003c/em\u003e,\u003cem\u003e\u0026nbsp;9\u003c/em\u003e(1). https://doi.org/10.1038/s41467-018-03279-9\u003c/li\u003e\n \u003cli\u003eLee, H., Qian, K., Christine, Hoerburger, L., Claussnitzer, M., Hoffmann, C., Glunk, V., Wahl, S., Breier, M., Eck, F., Jafari, L., Molnos, S., Grallert, H., Dahlman, I., Arner, P., Brunner, C., Hauner, H., Hauck, S. M., \u0026amp; Laumen, H. (2017). Allele-specific quantitative proteomics unravels molecular mechanisms modulated by cis-regulatory PPARG locus variation. \u003cem\u003eNucleic Acids Research\u003c/em\u003e,\u003cem\u003e\u0026nbsp;45\u003c/em\u003e(6), 3266-3279. https://doi.org/10.1093/nar/gkx105\u003c/li\u003e\n \u003cli\u003eLefterova, M. I., Haakonsson, A. K., Lazar, M. A., \u0026amp; Mandrup, S. (2014). PPAR\u0026gamma; and the global map of adipogenesis and beyond. \u003cem\u003eTrends in Endocrinology \u0026amp; Metabolism\u003c/em\u003e,\u003cem\u003e\u0026nbsp;25\u003c/em\u003e(6), 293-302. https://doi.org/10.1016/j.tem.2014.04.001\u003c/li\u003e\n \u003cli\u003eLefterova, M. I., Zhang, Y., Steger, D. J., Schupp, M., Schug, J., Cristancho, A., Feng, D., Zhuo, D., Stoeckert, C. J., Liu, X. S., \u0026amp; Lazar, M. A. (2008). PPAR\u0026gamma; and C/EBP factors orchestrate adipocyte biology via adjacent binding on a genome-wide scale. \u003cem\u003eGenes \u0026amp; Development\u003c/em\u003e,\u003cem\u003e\u0026nbsp;22\u003c/em\u003e(21), 2941-2952. https://doi.org/10.1101/gad.1709008\u003c/li\u003e\n \u003cli\u003eLopez-Delisle, L., Rabbani, L., Wolff, J., Bhardwaj, V., Backofen, R., Gr\u0026uuml;ning, B., Ram\u0026iacute;rez, F., \u0026amp; Manke, T. (2021). pyGenomeTracks: reproducible plots for multivariate genomic datasets. \u003cem\u003eBioinformatics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;37\u003c/em\u003e(3), 422-423. https://doi.org/10.1093/bioinformatics/btaa692\u003c/li\u003e\n \u003cli\u003eLove, M. I., Huber, W., \u0026amp; Anders, S. (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. \u003cem\u003eGenome Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;15\u003c/em\u003e(12). https://doi.org/10.1186/s13059-014-0550-8\u003c/li\u003e\n \u003cli\u003eLov\u0026eacute;n, J., Heather, Charles, Lau, A., David, Christopher, James, Tong, \u0026amp; Richard. (2013). Selective Inhibition of Tumor Oncogenes by Disruption of Super-Enhancers. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;153\u003c/em\u003e(2), 320-334. https://doi.org/10.1016/j.cell.2013.03.036\u003c/li\u003e\n \u003cli\u003eLyons, H., Veettil, R. T., Pradhan, P., Fornero, C., De La Cruz, N., Ito, K., Eppert, M., Roeder, R. G., \u0026amp; Sabari, B. R. (2023). Functional partitioning of transcriptional regulators by patterned charge blocks. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;186\u003c/em\u003e(2), 327-345.e328. https://doi.org/10.1016/j.cell.2022.12.013\u003c/li\u003e\n \u003cli\u003eMadsen, J. G. S., Madsen, M. S., Rauch, A., Traynor, S., Van Hauwaert, E. L., Haakonsson, A. K., Javierre, B. M., Hyldahl, M., Fraser, P., \u0026amp; Mandrup, S. (2020). Highly interconnected enhancer communities control lineage-determining genes in human mesenchymal stem cells. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;52\u003c/em\u003e(11), 1227-1238. https://doi.org/10.1038/s41588-020-0709-z\u003c/li\u003e\n \u003cli\u003eMadsen, J. G. S., Rauch, A., Van Hauwaert, E. L., Schmidt, S. F., Winnefeld, M., \u0026amp; Mandrup, S. (2018). Integrated analysis of motif activity and gene expression changes of transcription factors. \u003cem\u003eGenome research\u003c/em\u003e,\u003cem\u003e\u0026nbsp;28\u003c/em\u003e(2), 243-255. https://doi.org/10.1101/gr.227231.117\u003c/li\u003e\n \u003cli\u003eMadsen, M. S., Broekema, M. F., Madsen, M. R., Koppen, A., Borgman, A., Gr\u0026auml;we, C., Thomsen, E. G. K., Westland, D., Kranendonk, M. E. G., Koerkamp, M. G., Hamers, N., Bonvin, A. M. J. J., Pittol, J. M. R., Natarajan, K. N., Kersten, S., Holstege, F. C. P., Monajemi, H., van Mil, S. W. C., Vermeulen, M., . . . Kalkhoven, E. (2022a). PPAR\u0026gamma; lipodystrophy mutants reveal intermolecular interactions required for enhancer activation. \u003cem\u003eNature Communications\u003c/em\u003e,\u003cem\u003e\u0026nbsp;13\u003c/em\u003e(1), 7090. https://doi.org/10.1038/s41467-022-34766-9\u003c/li\u003e\n \u003cli\u003eMadsen, M. S., Broekema, M. F., Madsen, M. R., Koppen, A., Borgman, A., Gr\u0026auml;we, C., Thomsen, E. G. K., Westland, D., Kranendonk, M. E. G., Koerkamp, M. G., Hamers, N., Bonvin, A. M. J. J., Pittol, J. M. R., Natarajan, K. N., Kersten, S., Holstege, F. C. P., Monajemi, H., Van Mil, S. W. C., Vermeulen, M., . . . Kalkhoven, E. (2022b). PPAR\u0026gamma; lipodystrophy mutants reveal intermolecular interactions required for enhancer activation. \u003cem\u003eNature communications\u003c/em\u003e,\u003cem\u003e\u0026nbsp;13\u003c/em\u003e(1). https://doi.org/10.1038/s41467-022-34766-9\u003c/li\u003e\n \u003cli\u003eMahajan, A., Go, M. J., Zhang, W., Below, J. E., Gaulton, K. J., Ferreira, T., Horikoshi, M., Johnson, A. D., Ng, M. C. Y., Prokopenko, I., Saleheen, D., Wang, X., Zeggini, E., Abecasis, G. R., Adair, L. S., Almgren, P., Atalay, M., Aung, T., Baldassarre, D., . . . Morris, A. P. (2014). Genome-wide trans-ancestry meta-analysis provides insight into the genetic architecture of type 2 diabetes susceptibility. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;46\u003c/em\u003e(3), 234-244. https://doi.org/10.1038/ng.2897\u003c/li\u003e\n \u003cli\u003eMahajan, A., Spracklen, C. N., Zhang, W., Ng, M. C. Y., Petty, L. E., Kitajima, H., Yu, G. Z., R\u0026uuml;eger, S., Speidel, L., Kim, Y. J., Horikoshi, M., Mercader, J. M., Taliun, D., Moon, S., Kwak, S.-H., Robertson, N. R., Rayner, N. W., Loh, M., Kim, B.-J., . . . Morris, A. P. (2022). Multi-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translation. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;54\u003c/em\u003e(5), 560-572. https://doi.org/10.1038/s41588-022-01058-3\u003c/li\u003e\n \u003cli\u003eMahajan, A., Taliun, D., Thurner, M., Robertson, N. R., Torres, J. M., Rayner, N. W., Payne, A. J., Steinthorsdottir, V., Scott, R. A., Grarup, N., Cook, J. P., Schmidt, E. M., Wuttke, M., Sarnowski, C., M\u0026auml;gi, R., Nano, J., Gieger, C., Trompet, S., Lecoeur, C., . . . McCarthy, M. I. (2018). Fine-mapping type 2 diabetes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;50\u003c/em\u003e(11), 1505-1513. https://doi.org/10.1038/s41588-018-0241-6\u003c/li\u003e\n \u003cli\u003eMurphy, D., Salataj, E., Di Giammartino, D. C., Rodriguez-Hernaez, J., Kloetgen, A., Garg, V., Char, E., Uyehara, C. M., Ee, L.-S., Lee, U., Stadtfeld, M., Hadjantonakis, A.-K., Tsirigos, A., Polyzos, A., \u0026amp; Apostolou, E. (2024). 3D Enhancer\u0026ndash;promoter networks provide predictive features for gene expression and coregulation in early embryonic lineages. \u003cem\u003eNature Structural \u0026amp; Molecular Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;31\u003c/em\u003e(1), 125-140. https://doi.org/10.1038/s41594-023-01130-4\u003c/li\u003e\n \u003cli\u003eNielsen, R., Gr\u0026oslash;ntved, L., Stunnenberg, H. G., \u0026amp; Mandrup, S. (2006). Peroxisome proliferator-activated receptor subtype- and cell-type-specific activation of genomic target genes upon adenoviral transgene delivery. \u003cem\u003eMolecular and Cellular Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;26\u003c/em\u003e(15), 5698-5714. https://doi.org/10.1128/mcb.02266-05\u003c/li\u003e\n \u003cli\u003eOudelaar, A. M., Davies, J. O. J., Hanssen, L. L. P., Telenius, J. M., Schwessinger, R., Liu, Y., Brown, J. M., Downes, D. J., Chiariello, A. M., Bianco, S., Nicodemi, M., Buckle, V. J., Dekker, J., Higgs, D. R., \u0026amp; Hughes, J. R. (2018). Single-allele chromatin interactions identify regulatory hubs in dynamic compartmentalized domains. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;50\u003c/em\u003e(12), 1744-1751. https://doi.org/10.1038/s41588-018-0253-2\u003c/li\u003e\n \u003cli\u003ePollex, T., Rabinowitz, A., Gambetta, M. C., Marco-Ferreres, R., Viales, R. R., Jankowski, A., Schaub, C., \u0026amp; Furlong, E. E. M. (2024). Enhancer\u0026ndash;promoter interactions become more instructive in the transition from cell-fate specification to tissue differentiation. \u003cem\u003eNature Genetics\u003c/em\u003e. https://doi.org/10.1038/s41588-024-01678-x\u003c/li\u003e\n \u003cli\u003eQuinlan, A. R., \u0026amp; Hall, I. M. (2010). BEDTools: a flexible suite of utilities for comparing genomic features. \u003cem\u003eBioinformatics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;26\u003c/em\u003e(6), 841-842. https://doi.org/10.1093/bioinformatics/btq033\u003c/li\u003e\n \u003cli\u003eRada-Iglesias, A., Bajpai, R., Swigut, T., Brugmann, S. A., Flynn, R. A., \u0026amp; Wysocka, J. (2011). A unique chromatin signature uncovers early developmental enhancers in humans. \u003cem\u003eNature\u003c/em\u003e,\u003cem\u003e\u0026nbsp;470\u003c/em\u003e(7333), 279-283. https://doi.org/10.1038/nature09692\u003c/li\u003e\n \u003cli\u003eRam\u0026iacute;rez, F., D\u0026uuml;ndar, F., Diehl, S., Gr\u0026uuml;ning, B. A., \u0026amp; Manke, T. (2014). deepTools: a flexible platform for exploring deep-sequencing data. \u003cem\u003eNucleic Acids Research\u003c/em\u003e,\u003cem\u003e\u0026nbsp;42\u003c/em\u003e(W1), W187-W191. https://doi.org/10.1093/nar/gku365\u003c/li\u003e\n \u003cli\u003eRauch, A., Haakonsson, A. K., Madsen, J. G. S., Larsen, M., Forss, I., Madsen, M. R., Van Hauwaert, E. L., Wiwie, C., Jespersen, N. Z., Tencerova, M., Nielsen, R., Larsen, B. D., R\u0026ouml;ttger, R., Baumbach, J., Scheele, C., Kassem, M., \u0026amp; Mandrup, S. (2019). Osteogenesis depends on commissioning of a network of stem cell transcription factors that act as repressors of adipogenesis. \u003cem\u003eNature Genetics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;51\u003c/em\u003e(4), 716-727. https://doi.org/10.1038/s41588-019-0359-1\u003c/li\u003e\n \u003cli\u003eRauch, A., \u0026amp; Mandrup, S. (2021). Transcriptional networks controlling stromal cell differentiation. \u003cem\u003eNature Reviews Molecular Cell Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;22\u003c/em\u003e(7), 465-482. https://doi.org/10.1038/s41580-021-00357-7\u003c/li\u003e\n \u003cli\u003eRauluseviciute, I., Riudavets-Puig, R., Blanc-Mathieu, R., Castro-Mondragon, J. A., Ferenc, K., Kumar, V., Lemma, R. B., Lucas, J., Ch\u0026egrave;neby, J., Baranasic, D., Khan, A., Fornes, O., Gundersen, S., Johansen, M., Hovig, E., Lenhard, B., Sandelin, A., Wasserman, W. W., Parcy, F., \u0026amp; Mathelier, A. (2024). JASPAR 2024: 20th anniversary of the open-access database of transcription factor binding profiles. \u003cem\u003eNucleic Acids Research\u003c/em\u003e,\u003cem\u003e\u0026nbsp;52\u003c/em\u003e(D1), D174-d182. https://doi.org/10.1093/nar/gkad1059\u003c/li\u003e\n \u003cli\u003eRoss-Innes, C. S., Stark, R., Teschendorff, A. E., Holmes, K. A., Ali, H. R., Dunning, M. J., Brown, G. D., Gojis, O., Ellis, I. O., Green, A. R., Ali, S., Chin, S.-F., Palmieri, C., Caldas, C., \u0026amp; Carroll, J. S. (2012). Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. \u003cem\u003eNature\u003c/em\u003e,\u003cem\u003e\u0026nbsp;481\u003c/em\u003e(7381), 389-393. https://doi.org/10.1038/nature10730\u003c/li\u003e\n \u003cli\u003eSabari, B. R., Dall\u0026rsquo;Agnese, A., Boija, A., Klein, I. A., Coffey, E. L., Shrinivas, K., Abraham, B. J., Hannett, N. M., Zamudio, A. V., Manteiga, J. C., Li, C. H., Guo, Y. E., Day, D. S., Schuijers, J., Vasile, E., Malik, S., Hnisz, D., Lee, T. I., Cisse, I. I., . . . Young, R. A. (2018). Coactivator condensation at super-enhancers links phase separation and gene control. \u003cem\u003eScience\u003c/em\u003e,\u003cem\u003e\u0026nbsp;361\u003c/em\u003e(6400), eaar3958. https://doi.org/10.1126/science.aar3958\u003c/li\u003e\n \u003cli\u003eSaint-Andr\u0026eacute;, V., Federation, A. J., Lin, C. Y., Abraham, B. J., Reddy, J., Lee, T. I., Bradner, J. E., \u0026amp; Young, R. A. (2016). Models of human core transcriptional regulatory circuitries. \u003cem\u003eGenome Research\u003c/em\u003e,\u003cem\u003e\u0026nbsp;26\u003c/em\u003e(3), 385-396. https://doi.org/10.1101/gr.197590.115\u003c/li\u003e\n \u003cli\u003eSalma, N., Xiao, H., \u0026amp; Imbalzano, A. N. (2006). Temporal recruitment of CCAAT/enhancer-binding proteins to early and late adipogenic promoters in vivo. \u003cem\u003eJ Mol Endocrinol\u003c/em\u003e,\u003cem\u003e\u0026nbsp;36\u003c/em\u003e(1), 139-151. https://doi.org/10.1677/jme.1.01918\u003c/li\u003e\n \u003cli\u003eSchmidt, S. F., J\u0026oslash;rgensen, M., Chen, Y., Nielsen, R., Sandelin, A., \u0026amp; Mandrup, S. (2011). Cross species comparison of C/EBP\u0026alpha; and PPAR\u0026gamma; profiles in mouse and human adipocytes reveals interdependent retention of binding sites. \u003cem\u003eBMC Genomics\u003c/em\u003e,\u003cem\u003e\u0026nbsp;12\u003c/em\u003e(1), 152. https://doi.org/10.1186/1471-2164-12-152\u003c/li\u003e\n \u003cli\u003eShrikumar, A. T., K. Avsec, Ž. Shcherbina, A. Banerjee, A. Sharmin, M. Nair, S. Kundaje, A. (2020). Technical Note on Transcription Factor Motif Discovery from Importance Scores (TF-MoDISco) version 0.5.6.5. \u003cem\u003ePreprint arXiv\u003c/em\u003e.\u003c/li\u003e\n \u003cli\u003eSiersb\u0026aelig;k, R., Madsen, J. G. S., Javierre, B. M., Nielsen, R., Bagge, E. K., Cairns, J., Wingett, S. W., Traynor, S., Spivakov, M., Fraser, P., \u0026amp; Mandrup, S. (2017). Dynamic Rewiring of Promoter-Anchored Chromatin Loops during Adipocyte Differentiation. \u003cem\u003eMolecular Cell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;66\u003c/em\u003e(3), 420-435.e425. https://doi.org/10.1016/j.molcel.2017.04.010\u003c/li\u003e\n \u003cli\u003eSiersbaek, R., Nielsen, R., John, S., Sung, M.-H., Baek, S., Loft, A., Hager, G. L., \u0026amp; Mandrup, S. (2011). Extensive chromatin remodelling and establishment of transcription factor \u0026lsquo;hotspots\u0026rsquo; during early adipogenesis. \u003cem\u003eThe EMBO Journal\u003c/em\u003e,\u003cem\u003e\u0026nbsp;30\u003c/em\u003e(8), 1459-1472. https://doi.org/10.1038/emboj.2011.65\u003c/li\u003e\n \u003cli\u003eSiersbaek, R., Nielsen, R., \u0026amp; Mandrup, S. (2012). Transcriptional networks and chromatin remodeling controlling adipogenesis. \u003cem\u003eTrends in Endocrinology \u0026amp; Metabolism\u003c/em\u003e,\u003cem\u003e\u0026nbsp;23\u003c/em\u003e(2), 56-64. http://www.sciencedirect.com/science/article/pii/S1043276011001640 (Not in File)\u003c/li\u003e\n \u003cli\u003eSiersb\u0026aelig;k, R., Rabiee, A., Nielsen, R., Sidoli, S., Traynor, S., Loft, A., Lars, Rogowska-Wrzesinska, A., Ole, \u0026amp; Mandrup, S. (2014). Transcription Factor Cooperativity in Early Adipogenic Hotspots and Super-Enhancers. \u003cem\u003eCell Reports\u003c/em\u003e,\u003cem\u003e\u0026nbsp;7\u003c/em\u003e(5), 1443-1455. https://doi.org/10.1016/j.celrep.2014.04.042\u003c/li\u003e\n \u003cli\u003eSoccio, R. E., Chen, E. R., Rajapurkar, S. R., Safabakhsh, P., Marinis, J. M., Dispirito, J. R., Emmett, M. J., Briggs, E. R., Fang, B., Everett, L. J., Lim, H. W., Won, K. J., Steger, D. J., Wu, Y., Civelek, M., Voight, B. F., \u0026amp; Lazar, M. A. (2015). Genetic Variation Determines PPAR\u0026gamma; Function and Anti-diabetic Drug Response In Vivo. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;162\u003c/em\u003e(1), 33-44. https://doi.org/10.1016/j.cell.2015.06.025\u003c/li\u003e\n \u003cli\u003eTanaka, T. (1997). Defective adipocyte differentiation in mice lacking the C/EBPbeta and/or C/EBPdelta gene. \u003cem\u003eThe EMBO Journal\u003c/em\u003e,\u003cem\u003e\u0026nbsp;16\u003c/em\u003e(24), 7432-7443. https://doi.org/10.1093/emboj/16.24.7432\u003c/li\u003e\n \u003cli\u003eThomas, H. F., Kotova, E., Jayaram, S., Pilz, A., Romeike, M., Lackner, A., Penz, T., Bock, C., Leeb, M., Halbritter, F., Wysocka, J., \u0026amp; Buecker, C. (2021). Temporal dissection of an enhancer cluster reveals distinct temporal and functional contributions of individual elements. \u003cem\u003eMolecular Cell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;81\u003c/em\u003e(5), 969-982.e913. https://doi.org/10.1016/j.molcel.2020.12.047\u003c/li\u003e\n \u003cli\u003eTrojanowski, J., Frank, L., Rademacher, A., M\u0026uuml;cke, N., Grigaitis, P., \u0026amp; Rippe, K. (2022). Transcription activation is enhanced by multivalent interactions independent of phase separation. \u003cem\u003eMolecular Cell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;82\u003c/em\u003e(10), 1878-1893.e1810. https://doi.org/10.1016/j.molcel.2022.04.017\u003c/li\u003e\n \u003cli\u003eWang, A., Yue, F., Li, Y., Xie, R., Harper, T., Nisha, Muth, K., Palmer, J., Qiu, Y., Wang, J., Dieter, Jeffrey, Doris, Ren, B., \u0026amp; Sander, M. (2015). Epigenetic Priming of Enhancers Predicts Developmental Competence of hESC-Derived Endodermal Lineage Intermediates. \u003cem\u003eCell Stem Cell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;16\u003c/em\u003e(4), 386-399. https://doi.org/10.1016/j.stem.2015.02.013\u003c/li\u003e\n \u003cli\u003eWard, L. D., \u0026amp; Kellis, M. (2012). HaploReg: a resource for exploring chromatin states, conservation, and regulatory motif alterations within sets of genetically linked variants. \u003cem\u003eNucleic Acids Research\u003c/em\u003e,\u003cem\u003e\u0026nbsp;40\u003c/em\u003e(D1), D930-D934. https://doi.org/10.1093/nar/gkr917\u003c/li\u003e\n \u003cli\u003eWarren, David, Hnisz, D., Brian, Charles, Michael, Peter, Tong, \u0026amp; Richard. (2013). Master Transcription Factors and Mediator Establish Super-Enhancers at Key Cell Identity Genes. \u003cem\u003eCell\u003c/em\u003e,\u003cem\u003e\u0026nbsp;153\u003c/em\u003e(2), 307-319. https://doi.org/10.1016/j.cell.2013.03.035\u003c/li\u003e\n \u003cli\u003eXiaotao, W. ( 2016). unHiC: a user-friendly Hi-C data processing software based on\u003c/li\u003e\n \u003cli\u003ehiclib.\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003cem\u003ehttps://doi.org/10.5281/zenodo.55324\u003c/em\u003e.\u003c/li\u003e\n \u003cli\u003eZhang, Y., Liu, T., Meyer, C. A., Eeckhoute, J., Johnson, D. S., Bernstein, B. E., Nusbaum, C., Myers, R. M., Brown, M., Li, W., \u0026amp; Liu, X. S. (2008). Model-based Analysis of ChIP-Seq (MACS). \u003cem\u003eGenome Biology\u003c/em\u003e,\u003cem\u003e\u0026nbsp;9\u003c/em\u003e(9), R137. https://doi.org/10.1186/gb-2008-9-9-r137\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Methods","content":"\u003ch2\u003e\u003cstrong\u003eCell culture and differentiation\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTelomerase-immortalized human bone marrow-derived stromal progenitor cells (hMSC-TERT4) were kindly provided by Moustapha Kassem (Abdallah et al., 2004; Simonsen et al., 2002)), and maintained in \u0026alpha;-MEM with 10 % fetal bovine serum (FBS) and 1% penicillin-streptomycin (P/S) (growth medium) at 37\u0026deg;C, 95% humidity and 5% carbon dioxide. Osteo- or adipogenic differentiation was induced two days post confluence (day 0) by changing the growth medium to the differentiation medium. Osteogenic (\u0026alpha;-MEM with 10% FBS, 1% P/S and 10 nM freshly added vitamin D3 in 96% ethanol, 50 \u0026mu;g/mL L-ascorbic acid in MilliQ water, 10 mM \u0026beta;-glycerophosphate in MilliQ water and 10 nM dexamethasone) or adipogenic differentiation medium (DMEM with 10% FBS, 1% P/S and 100nM freshly added dexamethasone in 96% ethanol, 0.5 mM 3-isobutyl-1-methylxanthine in 0.1 M KOH, 2 \u0026mu;M human insulin, and 1 \u0026mu;M rosiglitazone in DMSO) was renewed every second day until cells were harvested. HEK293T cells were maintained in DMEM with 10% FBS and 1% P/S (HEK growth medium) at 37\u0026deg;C, 95% humidity and 5% carbon dioxide.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eProduction of the hMSC-TERT4/iCas9-4 cell line\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe Lenti-iCas9-Neo plasmid (Addgene #85400, gift from Qin Yan (Cao et al., 2016)) was used to transform competent \u003cem\u003eEscherichia coli\u003c/em\u003e DH5\u0026alpha; and purified in parallel with psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259). Integrity was evaluated by linearization (using appropriate restriction enzymes for each plasmid) followed by gel electrophoresis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLenti-iCas9-neo virus was produced in HEK293T cells and used to transduce hMSC-TERT4. Medium was changed to growth medium with 0.5 \u0026mu;g/mL doxycycline 34 hours after transduction to induce GFP and Cas9 expression. After 24 hours, GFP-positive cells were selected by fluorescence-activated cell sorting (FACS) using FACS Aria III (BD Biosciences) and sorted into and maintained in 37\u0026deg;C growth medium. Data analysis was performed with BD FACSDiva software. The initial GFP-positive population was called hMSC-TERT4/iCas9. Single-cell clones of this population were generated and the line hMSC-TERT4/iCas9-4 was selected for further studies.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eQuantification of incorporated virus copy number\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eGenomic DNA was used for qPCR-based determination of lentiviral transfer sequence copy number in hMSC-TERT4/iCas9-4 cells. Cells were harvested in cold lysis buffer (0.1% SDS, 1% Triton X-100, 0.15 M NaCl, 1 mM EDTA, 20 mM Tris pH 8) and stored at -80 \u0026deg;C prior to sonication. Entire cell lysates were sonicated in 130 \u0026mu;L microTUBE AFA fiber Pre-Slit Snap-Cap 6x16 mm tubes (Covaris) on an ME220 Focused-ultrasonicator (Covaris) for 20 s, peak power 75, duty% factor 26.67, cycles/burst 200. Shreded DNA was purified and used as a template for each qPCR reaction with forward and reverse primer targeting the long terminal repeat (LTR) of the transfer sequence and a genomic region (LTR and NoGene2 primer, respectively, Supplementary Information Table 2) and FastStart Essential DNA Green Master (Roche) on a Lightcycler 480 II (Roche). Incorporated virus copy number was calculated as:\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003eSince both primers amplify two target regions per viral transfer sequence and per genome, respectively.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eGenome editing in hMSC-TERT4/iCas9-4\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo design spacer sequences for guide RNAs (gRNAs), enhancer sequences were extracted from the UCSC Genome Browser (Kent et al., 2002, genome.ucsc.edu, hg19 assembly) based on the location and width of peaks from DNase-sequencing of hMSC-TERT4 cells during osteogenic and adipogenic differentiation (Rauch et al., 2019). Spacer sequences were found using the online tool sgRNA Designer (portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design (Doench et al., 2014, 2016)), selected by on-target and off-target rank (higher rank is predicted to be more active and specific) and ordered as default Alt-R CRISPR-Cas9 crRNAs (Integrated DNA Technologies). For spacer sequences see Table S6. tracrRNA for forming tracrRNA:crRNA duplexes, i.e. gRNAs, was also ordered from Integrated DNA Technologies.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ehMSC-TERT4/iCas9-4 cells were induced with 36 \u0026mu;g doxycycline in MilliQ water per 25 mL medium at 80-90% confluence. After 6 hours, cells were reverse transfected in a 12-well plate with a total of 12 pmol crRNA:tracrRNA duplexes (i.e. gRNA), 2 \u0026mu;L DharmaFECT 1 (Thermo Scientific) and 25,000cells/cm\u003csup\u003e2\u003c/sup\u003e per well. Cells were kept in transfection medium (1/3 Opti-MEM, 2/3 \u0026alpha;-MEM with 10 % FBS) for 2 days at 37 \u0026deg;C, 95% humidity, and 5% carbon dioxide, and seeded as single cells in 96-well plates (0.5 cells/100 \u0026mu;L medium). Clonal enhancer-deletion cell lines were then maintained in growth medium. Two controls were included: \u0026lsquo;No gRNA\u0026rsquo;, which received the same treatment as all other cell lines, except the transfection complexes contained no gRNA; \u0026lsquo;Ctr gRNA\u0026rsquo;, which received a pair of gRNAs mediating a deletion in intron 1 of \u003cem\u003ePPP1R12C\u0026nbsp;\u003c/em\u003e(adeno-associated virus \u0026ldquo;safe harbor\u0026rdquo; site, DeKelver et al., 2010)\u003cem\u003e.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eFor screening of enhancer-deleted clones, genomic DNA was harvested in DNeasy Blood and Tissue Kit for DNA isolation (Qiagen), and DNA was extracted according to the manufacturer\u0026rsquo;s instructions. Target regions in the deleted regions were amplified by end-point PCR using Phusion High-Fidelity DNA polymerase (NEB) (primers in Supplementary Information Table 3), and the presence of PCR product was evaluated on a Fragment Analyzer (Advanced Analytical Technologies). For each enhancer deletion, three independent clones lacking the wild-type PCR product were used to assess the size of the deleted region by PCR analysis of genomic region, followed by purification of PCR products with QIAquick PCR Purification Kit (Qiagen), and Sanger sequencing (primers in Supplementary Information Table 3, deletion size in Extended Data Table 3, and Supplementary Information Fig. 2).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eOil-Red-O and Alizarin Red S staining\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eCells were fixed in 4% formaldehyde in PBS for 15 min. For Oil-red-O staining of lipids, cells were incubated with 0.3% Oil-red-O in 60% isopropanol (filtered) at room temperature in the dark for 30 min. For Alizarin Red S staining of calcified matrix, cells were incubated with 1% Alizarin Red S in MilliQ water, pH adjusted to 4-4.2 with HCl and NH\u003csub\u003e4\u003c/sub\u003eOH. Cells were washed five times with water and subjected to macroscopic and microscopic photography.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eWestern blotting\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eCell lysates were harvested from hMSC-TERT4 after 0 or 7 days of adipogenic differentiation, or hMSC-TERT4/iCas9-4 cells treated with 0.5 \u0026mu;g/mL doxycycline for 6 hours, in lysis buffer (50 mM Tris-HCl pH 6.8, 10% glycerol, 2.5% SDS, 10 mM \u0026beta;-glycerophosphate, 10 mM NaF, 0.1 mM sodium orthovanadate, 1 mM PMSF (added fresh), 1x Complete (Roche, added fresh), boiled and treated with benzonase. Protein concentration was determined using the Pierce BCA Protein Assay Kit (Thermo Scientific) in accordance with the manufacturer\u0026rsquo;s instructions. Dithioeythritol (DTE) was then added to 0.01 M. For Western blotting, 10-40 \u0026mu;g protein sample was used per well in a 10% gel (separation: 10% acrylamide/bis-acrylamide solution, 0.4 M Tris-HCl pH 8.8, 0.1% SDS, 0.1% ammonium persulfate in MilliQ (APS), 0.001% tetramethylethylenediamine (TEMED); stacking: 6% acrylamide/bis-acrylamide solution, 0.125 M Tris-HCl pH 6.8, 0.1% SDS, 0.1% APS, 0.001% TEMED). For specific protein detection, the wet-blotted poly-vinyl-difluoride membrane was incubated for 1 hour at room temperature or overnight at 4\u0026deg;C with primary antibody: \u0026nbsp; rabbit-anti-Lamin A (1:10,000, L1293, Sigma-Aldrich), mouse-anti-FLAG (1:500, F1804-1MG, Sigma-Aldrich), rabbit-anti-PPARG (1:1000, 2443S, Cell Signaling) in PBS with 1.5% BSA, and then 1 hour at room temperature with secondary antibody: goat-anti-rabbit (1:1,000, P0448, DAKO) or goat-anti-mouse (1:2,000, P0447, DAKO) in PBS with 1.5% BSA. Enhanced chemiluminescence detection was performed using Immobilon Classico Western HRP substrate (Merck millipore), and either exposing the membrane to an X-ray film, or imaging using the Amersham Imager 680. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eQuantification of lipid droplet area\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eLipid accumulation was imaged using Nikon Eclipse Ts2-FL with differential interference contrast and 40x total magnification. Lipid droplet area was quantified using ImageJ (Rueden et al., 2017), by thresholding on the background subtracted greyscale images (using identical thresholds for all images) and quantifying the area in pixels that reached the threshold.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eAnalysis of gene expression using qPCR\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003ehMSC-TERT4/iCas9-4 enhancer deletion and control cell lines (three clones for each condition) were harvested using Isol-RNA Lysis Reagent (5PRIME) after 10 days of adipogenic differentiation. RNA was purified using phenol-chloroform extraction and EconoSpin columns (Epoch Life Sciences). For cDNA synthesis, up to 1000 ng RNA (always equal amounts in replicates) was treated with DNase I and reverse-transcribed using Moloney murine leukemia virus reverse transcriptase (Invitrogen). cDNA was used for qPCR with FastStart Essential DNA Green Master (Roche) on a Lightcycler 480 II (Roche) (primers Supplementary Information Table 2). Data for each sample were normalized to the level of transcript of TATA-box protein II B (\u003cem\u003eTBP\u003c/em\u003e).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eGene expression analysis using RNA-seq, and enhancer activity analysis using DNase-seq and MED1 ChIP-seq\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eOur previously published RNA-seq data from hMSC-TERT4 at day 0, 4 hours, day 1, 3, 7, and 14 of adipogenic and osteogenic differentiation (GEO; GSE113253) were analyzed as described by Rauch et al. 2019. Previously published DNase-seq and MED1 ChIP-seq data from hMSC-TERT4 at day 0, and day 1, 3, 7, and 14 of adipogenic and day 0, and day 1, 3, and 7 of osteogenic differentiation (GEO; GSE113253) were analyzed as described (Rauch et al., 2019) and visualized using using the pyGenomeTracks software (Lopez-Delisle et al., 2021).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eAnalysis of putative enhancer-anchored interactions using ECHi-C\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eOur previously published ECHi-C data from hMSC-TERT4 at day 0 and day 10 of adipogenic differentiation (GEO; GSE140782) was analyzed as described (Madsen et al., 2020). Interactions anchored in putative enhancers were visualized using the WashU Epigenome Browser (Zhou et al., 2011).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of chromatin conformation in\u003cem\u003e\u0026nbsp;PPARG\u003c/em\u003e locus in different stem cell states using public Hi-C data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHi-C data set for hMSC-TERT4 at progenitor state (GEO; GSE140782) (Madsen et al., 2020) were mapped to human genome (hg19), filtered and processed using runHiC (Xiaotao, 2016). HiC matrixes for ESC, mesodermal cells, MSC, neuronal progenitor cells (GEO; GSE52457) were visualized using Genome Browser from YUE lab (http://3dgenome.fsm.northwestern.edu/view.php). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of super-enhancers in hMSC-TERT4\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSuper-enhancers were identified and ranked based on MED1 ChIP-seq enrichment at day 14 of adipogenesis in hMSC-TERT4 using ROSE (Lov\u0026eacute;n et al., 2013; Warren et al., 2013) with default parameters.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIMAGE analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIMAGE analysis was performed according to the instructions (Madsen et al., 2018). Briefly, putative enhancers activity (based on MED1 signal), and RNA-sequencing data from day 0, 1, 3 and 4 hours of adipogenesis in hMSC-TERT4 cells were used as input for a two-step machine-learning algorithm to calculate \u003cem\u003emotif activity\u0026nbsp;\u003c/em\u003edepicted as the contribution of given TF motifs to overall enhancer activity and gene expression in given time point of adipogenesis. IMAGE also predicts target enhancers (TE) for all motifs based on an estimated error per enhancer when the modeled enhancer activity is compared with or without a specific motif. TE was used to dissect all motifs targeting TFs at studied \u003cem\u003ePPARG\u003c/em\u003e enhancers.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eChIP\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eChIP was performed as described previously (Rauch et al., 2019). Briefly, only for MED1 ChIP cells were crosslinked in 2 mM disuccinimidyl glutarate (DSG) in PBS (100 \u0026micro;L of 0.5 M DSG (Proteochem) in DMSO per 25 ml of PBS) for 20 minutes. For H3K4me1, CTCF, C/EBP\u0026beta;, C/EBP\u0026alpha; and PPAR\u0026gamma; ChIP, this step was omitted. Then, cells were crosslinked in 1% formaldehyde in PBS for 10 minutes followed by quenching with 0.125 M glycine for 10 minutes. Cells were harvested in cold lysis buffer (0.1% SDS, 1% Triton X-100, 0.15 M NaCl, 1 mM EDTA, 20 mM Tris pH 8) and stored at -80\u0026deg;C prior to sonication. Samples were sonicated using ME220 Focused-ultrasonicator (Covaris) for 15 min, peak power 75, duty% factor 5, cycles/burst 1000 (for MED1 ChIP-qPCR, \u0026mu;L microTUBE AFA fiber Pre-Slit Snap-Cap 6x16 mm tubes (Covaris)), and for 10 min, peak power 75, duty% factor 26.66, cycles/burst 500 (for ChIP samples followed by sequencing, miliTUBE 1 ml AFA Fiber(100) (Covaris)). Antibodies directed against MED1 (A300-793A, Bethyl Laboratories), H3K4me1 (ab8895, abcam), CTCF (61311, Active Motif), C/EBP\u0026beta; (sc-7962, Santa Cruz), C/EBP\u0026alpha; (8178S, Cell Signaling) and PPAR\u0026gamma; (2443S, Cell Signaling) were used for immunoprecipitation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor ChIP-qPCR precipitated DNA was used as a template for qPCR with FastStart Essential DNA Green Master (Roche) on a Lightcycler 480 II (Roche) (primers Supplementary Information Table 2). MED1 occupancy was determined as % recovery of input sample, and the signal in each sample was normalized to signal at the promoter of the house-keeping gene \u003cem\u003eZRANB3\u003c/em\u003e to account for ChIP efficiency (see Extended Data Fig. 5a for MED1 % recovery in different genomic regions where signal was expected to be similar across samples).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor ChIP-seq, precipitated DNA was prepared for sequencing following the manufacturer\u0026apos;s protocol (Illumina). ChIP-seq was performed on minimum two independent biological experiments. ChIP libraries were sequenced on the Illumina NovaSeq 6000, and the quality of sequenced reads was estimated using FastQC (https://qubeshub.org/resources/fastqc). Sequencing reads were mapped to the human genome (hg19) using STAR (Dobin et al., 2013). Picard (http://broadinstitute.github.io/picard/) was used to remove duplicates from aligned reads. \u0026nbsp; Peaks were identified using MACS2 (Zhang et al., 2008), and reproducible peaks were estimated using BEDTools Intersect (Quinlan \u0026amp; Hall, 2010). Bigwig files were created using bamCoverage with RPKM normalization using deepTools (Ram\u0026iacute;rez et al., 2014), and then visualized using pyGenomeTracks software (Lopez-Delisle et al., 2021). For MED1 \u0026nbsp; ChIP-seq at day 10 of adipogenesis in control \u0026lsquo;No gRNA\u0026rsquo; and E+102 deletion in hMSC-TERT4-iCas9/4, and for MED1 ChIP-seq at day 3 of adipogenesis in control \u0026lsquo;No gRNA\u0026rsquo; and in E+2 deletion and in E+102 deletion in hMSC-TERT4-iCas9/4 , and for C/EBP\u0026beta; ChIP-seq at d1 of adipogenesis in control \u0026lsquo;No gRNA\u0026rsquo; and in E+2 deletion and in E+102 deletion in hMSC-TERT4-iCas9/4, size factors were estimate using DESeqDataSetFromMatrix and estimateSizeFactors functions in DESeq2 v.1.42.0 (Love et al., 2014). Then size factors were used to create normalized bigwig files with BamCoverage from deepTools (Ram\u0026iacute;rez et al., 2014), and bigwig replicate files were averaged with bigwigAverage from deepTools (Ram\u0026iacute;rez et al., 2014). For MED1 ChIP-seq at day 3 of adipogenesis, and C/EBP\u0026beta; ChIP-seq at day 1 of adipogenesis in control \u0026lsquo;No gRNA\u0026rsquo; and in \u0026nbsp;E+2 and E+102 deletion in hMSC-TERT4-iCas9/4, differential binding was performed using DiffBind v.3.12.0 with edgeR (\u003cem\u003eP\u003c/em\u003e-value \u0026le; 0.05) (Ross-Innes et al., 2012). Plot profiles, and heatmaps were created using DeepTools, with commands computeMatrix, plotHeatmap or plotProfile using normalized and averaged bigwig files.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCapture-C\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ehMSC-TERT4/iCas9-4 enhancer deletion and control cell lines (one clone for each condition) were harvested at day 0 and 10 of adipogenic differentiation. Capture-C was performed using a customized version of a previously published protocol (Oudelaar et al., 2018). Briefly, cells were cross-linked for 10 minutes in 2% formaldehyde in PBS, \u0026nbsp;quenched with 0.125 M glycine, harvested in fresh cold lysis buffer (10 mM Tris-HCl pH 8, 10 mM NaCl, 0.2% Igepal CA-630 (Sigma), 1x Complete (Roche)), incubated 20 minutes on ice and snap-frozen in liquid nitrogen before storage at -80\u0026deg;C. 3C libraries were generated by in-nucleus double digestion with 3x 150 U of each DdeI and DraI restriction enzymes (New England Biolabs) added over approximately 24 hours (final buffer conditions after last addition of enzyme: 1x enzyme buffer, 0.06% sodium dodecyl sulfate, 1.5% Triton X-100). Restriction enzymes were chosen based on the distribution of restriction sites in the genomic region surrounding the \u003cem\u003ePPARG\u003c/em\u003e locus, aiming for the regulatory elements of interest located on restriction fragments (RFs) of approximately 120-250 bp. Digestion overhangs were filled in with dNTP and blunt-end ligated. De-crosslinked chromatin was purified. Approximately 5 \u0026mu;g of purified 3C library was sheared to an average size of 450 bp by sonication on an ME220 Focused-ultrasonicator (Covaris) and cleaned up with 0.85:1 AMPure XP beads (Beckman Coulter) sample ratio to remove fragments \u0026lt;200 bp. For sequencing library preparation, sheared DNA was end-repaired, adenine-tailed and ligated to NEBNext hairpin adaptors (New England Biolabs), followed by U excision by USER Enzyme (New England Biolabs). Libraries were amplified using Q5 high-fidelity DNA polymerase (New England Biolabs), universal primer and 8-bp-index primers (New England Biolabs) in 7 cycles of PCR.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor capture, 5\u0026rsquo; biotinylated 116-120 bp DNA oligos were designed to target the middle of single RFs spanning, or located as close as possible to, the regulatory elements of interest, avoiding repetitive regions (Sigma standard DNA oligos in tubes (single-stranded), Supplementary Information Table 5). From each sample, 1650 ng of amplified 3C library was used to generate one multiplexed library for capture (biological replicates were kept separate). Per sample in the multiplexed library, 13 fmol in total of pooled biotinylated DNA oligos targeting all 10 regulatory elements of interest was used for capture, i.e. 1.3 fmol of each oligo, and 130 fmol total oligo pool if the multiplexed library consisted of 10 samples. Sequencing adapters and indexes were blocked using a total of 2 nmol custom blockers (Integrated DNA Technologies, PAGE purified DNA oligos) per sample in the multiplexed library, and two rounds of capture were performed as previously described (Oudelaar et al., 2018), using 10 cycles of amplification after each capture. According to the manufacturer\u0026apos;s instructions, multiplexed and double-captured libraries were subjected to 150 bp paired-end sequencing (Illumina).\u003c/p\u003e\n\u003cp\u003eThe resulting raw sequencing reads were mapped to hg19 and quality-filtered using the CCseqBasicS pipeline (Telenius et al., 2020) in a version customized to handle reads produced with the restriction enzymes DraI and DdeI (see also Supplementary Information Fig. 1). Resulting unique interaction counts from chromosome 3 were normalized to 50,000 counts per sample and viewpoint. As expected, in samples where a captured viewpoint was deleted, the raw interaction count was generally very low compared to samples where the viewpoint was intact (Extended Data Fig. 3c). However, for the viewpoints E-130, E-123 and E+89, the capture oligo and deleted sequence only displayed partial overlap, resulting in some capture of the deleted region. To simplify downstream analysis, counts in all samples, where a captured viewpoint was deleted, were set to 0 for that viewpoint. The analysis was then restricted to the region chr3: 12.1-12.6 Mb (region of interest, ROI) containing the \u003cem\u003ePPARG\u003c/em\u003e locus, and interactions to regions \u0026plusmn;2 kb from the viewpoint were excluded from further analysis. Normalized interaction counts were averaged over the RFs overlapping bins of 1 kb or 5 kb, giving a count for each 1 kb or 5 kb bin along the ROI seen from each viewpoint in each sample. The expected count for each viewpoint in each sample (at either RF or binned resolution), as a function of distance from the viewpoint, was then predicted by fitting a cubic polynomial model to the observed sample count per 1 kb bin (for element specific interactions, see below) or 5 kb bin (for ROI-wide interactions), given by the geometric mean count per bin across all intact viewpoints in that sample. \u003cem\u003eP\u003c/em\u003e-values describing the chance of an observed interaction count at a given distance from a viewpoint randomly being higher than the predicted count for that distance were calculated using Student\u0026rsquo;s t-test and adjusted for multiple testing using the Benjamini-Hochberg method. Significant interactions (FDR \u0026le; 0.01, 208 ROI-wide unique interactions out of 1000 possible across all viewpoints) identified in any condition were kept for further analysis. Differential interactions were identified using Student\u0026rsquo;s t-test but not adjusted for multiple testing to avoid inflating the number of false negatives in this relatively small dataset. Since interaction counts in the \u0026lsquo;No gRNA\u0026rsquo; and \u0026lsquo;Ctr gRNA\u0026rsquo; controls correlated well, compared to the remaining conditions (Extended Data Fig. 3b), these conditions were considered as four replicates of one control for all statistical tests (instead of two replicates of two controls). To specifically analyze interactions between the selected \u003cem\u003ePPARG\u003c/em\u003e community enhancers, element-specific observed and predicted interaction counts were calculated based on un-binned counts as the mean interaction count seen from one viewpoint (enhancer) to all non-captured RFs overlapping another enhancer. This resulted in two interaction counts for each element-specific interaction, one count seen from each viewpoint, which were summed to obtain a single total observed or predicted count for that interaction. Significant interactions (FDR \u0026le; 0.01, 33 element-specific interactions across all viewpoints out of 45 possible) were identified and subjected to differential analysis as described above. Element-specific interactions were visualized using the circlize package for R (Gu et al., 2014).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of genetic variants within PPARG locus and their predicted effect size using Enfomer\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo identify variants associated with cardiometabolic diseases, we queried the \u003cem\u003ePPARG\u003c/em\u003e topologically associated domain (TAD chr3:12174868-12526858 /hg19) from multiple relevant resources; (a) the Common Metabolic Diseases Knowledge Portal (Costanzo et al., 2023), (b) the CARDIoGRAMplusC4D (Coronary ARtery DIsease Genome wide Replication and Meta-analysis - CARDIoGRAM), plus The Coronary Artery Disease (C4D) Genetics) [The data on coronary artery disease contributed by the CARDIoGRAMplusC4D and UK Biobank CardioMetabolic Consortium CHD working group who used the UK Biobank Resource (application number 9922). Data have been downloaded from www.CARDIOGRAMPLUSC4D.ORG, (c) the DIAGRAM consortium [https://diagram-consortium.org/] (Gaulton et al., 2015; Mahajan et al., 2014; Mahajan et al., 2022; Mahajan et al., 2018) and (d) the variant lists from the GWAS Catalog associated with BMI, T2D, T2DadjBMI, WHR, and WHRadjBMI. From these data resources, the significant variants (p-value \u0026lt; 5 x 10\u003csup\u003e-8\u003c/sup\u003e) were extracted and submitted to the HaploReg server (Ward \u0026amp; Kellis, 2012)\u0026nbsp;for identifying their haplotypes in LD \u0026gt; 0.8.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo further annotate the variants and their corresponding haplotypes, we queried their associated PheWAS from the Common Metabolic Diseases Knowledge Portal API - HuGeAMP (Costanzo et al., 2023), and aggregated traits related to T2D with an association p-value of less than 0.05. These phenotypes or associated traits are reported in the (Extended Data Table 1-2). This table reports the global and race-specific allelic frequencies of the variants which were directly queried from the NIH Allele Frequency Aggregator (ALFA) [https://www.ncbi.nlm.nih.gov/\u003c/p\u003e\n\u003cp\u003e/docs/gsr/alfa/].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo predict variant effects on the \u003cem\u003ePPARG2\u003c/em\u003e and our enhancer regions we used the Enformer model (Žiga Avsec, Vikram Agarwal, et al., 2021), and we evaluated variant effects on the genomic coordinates of the enhancer regions and the two transcripts of \u003cem\u003ePPARG2\u003c/em\u003e (ENST00000287820 [chr3:12351472-12434356] and ENST00000683699 [chr3:12351500-12434162]). In this analysis, we considered the effects of alternate alleles vs. the reference allele of the variants and their haplotypes to evaluate the effects on \u003cem\u003ePPARG2\u003c/em\u003e enhancer activity (using H3K27ac as proxy) and its expression (CAGE) in adipocyte-related tissues (Supplementary Information Table 1). The maximum absolute effects of the queried variants in these signals were used to represent variant effects on \u003cem\u003ePPARG2\u003c/em\u003e enhancer and CAGE activities. To define a background effect and significance level on the enhancer and CAGE activities, we selected 4000 variants at random from the GWAS catalogue list of significant hits (\u003cem\u003ep\u003c/em\u003e-value \u0026lt; 5 x 10\u003csup\u003e-8\u003c/sup\u003e). Using Enformer, we calculated the effect of these randomly selected variants on the region of their closest transcripts (the transcript regions were denied using GENCODE v43). The absolute maximum effects of the variants on the same list of adipocyte-related signals were used to define a background effect. The 90 percentiles of the effects from these randomly selected variants were then used as the cutoffs for identifying significant effects of our of variants [Enhancer (90%); negative cutoff: -0.0090, positive cutoff: 0.0088, CAGE (90%): negative cutoff: -0.1147, positive cutoff: 0.1585]. The Enformer results on the GWAS catalogue variants are reported in the (Extended Data Table 1-2).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of binding affinity using HT-SELEX datasets\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnalysis was performed following the study (Bhimsaria et al., 2023). Briefly, effect of the genetic variant; rs181025382 on DNA binding was estimated as log fold change in MinSeqs using two HT-SELEX data sets for PPAR\u0026gamma;: without ligand treatment, and for rosiglitazone-treated PPAR\u0026gamma;. These datasets were used to determine the enrichment of sequence flanking\u0026nbsp;\u0026plusmn;\u0026nbsp;20 bp from rs181025382 (chr3:12494623-12494663/hg19)\u0026nbsp;for both the reference allele (A/hg19) and the rs181025382-altered allele (G/hg19).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs controls, we selected the top 1000 PPAR\u0026gamma;-bound enhancers in hMSC-TERT4 (identified based on a reproducible signal from DNase- and MED1-ChIP seq at day 14 of adipogenesis and PPAR\u0026gamma; signal at day 10 of adipogenesis in hMSC-TERT4), and 1000 randomly selected enhancers that were not bound by PPAR\u0026gamma; (based on the same signals). MinSeq sequences analysis from two PPAR\u0026gamma; datasets mentioned earlier was then performed on 1000 bp sequences these two control enhancers sets.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe log fold change in the enrichment was calculated with log\u003csub\u003e2\u003c/sub\u003e \u003cimg width=\"97\" height=\"47\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAGEAAAAvCAMAAADEilSNAAAAAXNSR0IArs4c6QAAAKhQTFRFAAAAAAAAAAA6ADpmADqQAGa2OgAAOgA6OgBmOjpmOjqQOma2OpC2OpDbZgAAZgA6ZgBmZjoAZjpmZjqQZpDbZrbbZrb/kDoAkDo6kDpmkGY6kGZmkGaQkLbbkNv/tmYAtmY6tmZmtpBmtrbbtrb/ttu2ttvbttv/tv//25A625Bm25C227Zm27aQ29u229vb2/+22//b2////7Zm/9uQ/9u2//+2///b6nmGQAAAAAF0Uk5TAEDm2GYAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAAZdEVYdFNvZnR3YXJlAE1pY3Jvc29mdCBPZmZpY2V/7TVxAAACMUlEQVRYR+1WW1uCQBDdRSVLLUsFKwu7KmU3QPb//7POzAouaimIfT0wD33bMDuzc+ZyFOLvZX49yRt01s9zI+4Fecy1bdTa5U4oIfXAd5bGnnW7cpM0qdb4PGZDv/NrIDW8FVFfXU6XVpRPfGEoBGuSLI10te8tEaKTIHRE3CWH6kpa9UCELRHZUjYyGtIKNZaWkx6ECMlmESGyGy82bq8JUAIEHEG57cDHHUJMvyujcYQaDj7GDj7rA0egt7CL97ejAZ67HsHriLupYJTou0fXga5G2NAwmra02l/JAc4yKOEfnVMq6r4FdM9wEfnTu8N6MGs+P8S9p0fk9Ip8TA20cDAffSYHRDArrVxHR/ThlgWZ4q9HOULPtTy3Bm5tGp/Wg9i1YN8wNdAqV9Ym9JkPabfqTPAozh7qY90lKx0wH+0/cVFTu0bqFBLwlC1JGQiwJE7ZMRb+0D4oQLbs5YbyCSYOQ2UuW9JCcIRDCZf68CjpjjqQ8POz3VqAf+hxP3CQ7tbMxBXiHx7fjRy0wIe3hsk/dK4txzC05Qor7cpBydYQtPlM/qHzOJ2SuJvZIjk4KN18XGSTf3AmAJlleO1zCktWKsZBBv8QYrWblFyIiiiAyUpFOMjgH0w5pcAsQ/Oii7AnB2X4h9oXfcwso5kJ/VaMg4wpM/nHA/WH0lmQi264ohy0YZDX+Ge3pbI2cQkHbd0VCmS61WiTwcpPgUI+qksVAhUCFQIVAhUC/xuBb2YVXZTvBPGVAAAAAElFTkSuQmCC\" alt=\"image\"\u003e, where \u003cem\u003eE\u003c/em\u003e(alt.allele) and \u003cem\u003eE\u003c/em\u003e(Ref.allele) are enrichment values calculated by scanning MinSeq sequences within the queried reference and altered sequences.\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003cimg width=\"8\" height=\"36\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAgAAAAkCAMAAACDv4v0AAAAAXNSR0IArs4c6QAAAE5QTFRFAAAAAAAAADqQAGa2OgAAOgA6OgBmOpDbZgAAZjoAZpBmZrb/kDoAkNv/tmYAtmY6tpCQtv/btv//25A629uQ2////7Zm/9uQ//+2///bKRlF6QAAAAF0Uk5TAEDm2GYAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAAZdEVYdFNvZnR3YXJlAE1pY3Jvc29mdCBPZmZpY2V/7TVxAAAAS0lEQVQoU2NgoAxIcDIJ8jMyiTIIiLBx8Yqz8jIwiLNwM4izCTEwCAOFxYCYgY8dgiU4uBkkebiBSoDSYLUgAFYCAsLMEBqsZAgBAGf/AsIWnkOMAAAAAElFTkSuQmCC\" alt=\"image\"\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003eis a parameter to facilitate unintended issues caused by division by small numbers.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBPNet Model analysis\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA BPNet (Žiga Avsec, Melanie Weilert, et al., 2021) \u0026nbsp;model was trained on PPAR\u0026gamma; ChIP at day 10 of adipogenesis in hMSC-TERT4 data. Replicates were merged using samtools v. 1.13 (Danecek et al., 2021). 5\u0026rsquo; ends of reads were extracted using bedtools v.2.30.0 (Quinlan \u0026amp; Hall, 2010) using subcommand genomecov for each strand, and base pair resolution bigwig files were constructed using bedGraphToBigWig. Low threshold peaks were called with macs2 v.2.2.9.1 using subcommand callpeak using the parameters -p 0.1 --gsize 2701495761 --keep-dup all --shift -75 --extsize 150. Peak regions were blacklist filtered (Amemiya et al., 2019) with bedtools intersect using the ENCODE hg19 blacklist. The model was trained using the BPNet-lite v.0.8.1 implementation of BPNet (https://github.com/jmschrei/bpnet-lite). We used an input sequence length of 2114 bp and an output profile width of 1000 bp. We used 64 convolutional filters and an 8 layers deep tower of dilated convolutions. Augmentation was used on the training sequences, using a max jitter (random lengthwise translation) of 128 bp and reverse complementing with a probability of 0.5. Chromosomes 8 and 10 were used for validation, while the remaining autosomal chromosomes plus chromosome X were used for training. The model with the lowest loss on the validation chromosomes after training for 50 epochs was selected.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eMotif discovery and attributions\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo identify sequence motifs driving the ChIP-seq signal we utilized TF-MoDISco (Shrikumar, 2020), running the updated implementation tfmodisco-lite v.2.2.0 (https://github.com/jmschrei/tfmodisco-lite). The identified motifs were matched against the JASPAR motif database (Rauluseviciute et al., 2024) using tomtom (Gupta et al., 2007) from meme v.5.5.5.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eSNP analysis\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eTo analyze the effect of individual SNP at or near the PPARG locus we used tangermeme v.0.2.1 (https://github.com/jmschrei/tangermeme) and performed in-silico saturation mutagenesis (ISM) using the count predictions.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eGenome-wide PPARG motif mutation analysis\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eWe extracted the PPARG:RXR dimer motif (ID: MA0065.2) from the JASPAR database. We then ran FIMO using tangermeme v.0.2.1 (https://github.com/jmschrei/tangermeme) to identify motif matches across all PPAR\u0026gamma; peak region PPAR\u0026gamma; ChIP-seq \u0026nbsp;from day 10 of adipogenesis in hMSC-TERT4, \u0026nbsp;using a p-value threshold of 0.01. The matches were then narrowed down to unique and high scoring matches, selecting only those with a score above 10. Finally, we selected those matches with an A at position 9 of the motif match, so that we can induce the A \u0026gt; G mutation as for rs181025382. We then calculated the count prediction for sequences centered on the variant both with and without the A \u0026gt; G mutation (Extended Data Fig. 7b-c).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eStatistics and reproducibility\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eFor each enhancer deletion or hMSC-TERT4/iCas9-4 control, mRNA expression and MED1 ChIP-qPCR were assessed based on two independent differentiation experiments of the three independent clones (\u003cem\u003en\u003c/em\u003e=2 independent biological replicates). For ChIP-seq in enhancer deletion and controls two clonal cell lines per enhancer deletion was used in two independent differentiation experiments (\u003cem\u003en\u003c/em\u003e=2 independent biological replicates, \u003cem\u003en\u003c/em\u003e=2 clonal cell line per condition). For Capture-C, only one clonal cell line per enhancer deletion was used in two independent differentiation experiments (\u003cem\u003en\u003c/em\u003e=2 independent biological replicates). For publicly available sequencing data from hMSC-TERT4, data were used from at least two independent differentiation experiments (\u003cem\u003en\u003c/em\u003e=2 independent biological replicates, \u003cem\u003en\u003c/em\u003e=3 for RNA-sequencing). Lipid accumulation in response to enhancer deletion was assayed based on one differentiation experiment using 3 clonal cell lines per enhancer deletion. For statistics not performed within published packages or methods, normality was evaluated visually before testing using the qqnorm() function in R. Equal variances were either not assumed or tested using var.test() in R. Student\u0026rsquo;s t-test (for data with equal variances) and Welch\u0026rsquo;s t-test (for data with non-equal variances) were performed in R using the t.test() function in a two-sided manner. \u003cem\u003eP\u003c/em\u003e-values were adjusted for multiple testing in R using the p.adjust() function with the Benjamini-Hochberg method where indicated. Polynomial regression was performed in R using the lm() function. All bar plots show individual measurements and/or mean across a group (defined in Fig. legends), and error bars represent the standard deviation (SD), standard error of the mean (SEM) or range between two experiments (indicated in Fig. legends). Line plots of Capture-C interactions show the mean of two independent experiments with error bars representing a range. \u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6466826/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6466826/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Peroxisome proliferator-activated receptor γ (PPARγ) is the master regulator of adipogenesis, but the mechanisms underlying its strong induction in response to adipogenic cues are poorly understood. Using human mesenchymal stem cells as a model system, we show that the PPARG locus is primed for activation in the progenitor state by a highly connected enhancer community, which gets further activated and interconnected during adipogenesis. By systematically deleting individual enhancers in the community and interrogating the effect on enhancer function, connectivity and gene expression, we reveal important, non-redundant and cooperative roles of many enhancers. We show that the promoter-proximal enhancers and the downstream super-enhancer constituents cooperate in cis at early time points of differentiation, whereas at later timepoints PPARγ feedback-activates its own expression. Top-scoring non-coding cardiometabolic variants predicted to affect PPARG2 expression map to key enhancers in the community, indicating that regulation via these enhancers is important for human physiology.","manuscriptTitle":"Extensive enhancer crosstalk controls PPARG2 activation during adipogenesis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-19 06:25:30","doi":"10.21203/rs.3.rs-6466826/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"e06e96e9-c937-481f-b06f-33bc1e0bded8","owner":[],"postedDate":"August 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":47849280,"name":"Biological sciences/Genetics/Gene regulation"},{"id":47849281,"name":"Biological sciences/Genetics/Functional genomics"},{"id":47849282,"name":"Biological sciences/Genetics/Genetic association study/Genome-wide association studies"},{"id":47849283,"name":"Biological sciences/Molecular biology/Epigenetics"}],"tags":[],"updatedAt":"2026-04-28T07:10:58+00:00","versionOfRecord":{"articleIdentity":"rs-6466826","link":"https://doi.org/10.1038/s41467-026-70401-7","journal":{"identity":"nature-communications","isVorOnly":false,"title":"Nature Communications"},"publishedOn":"2026-03-15 04:00:00","publishedOnDateReadable":"March 15th, 2026"},"versionCreatedAt":"2025-08-19 06:25:30","video":"","vorDoi":"10.1038/s41467-026-70401-7","vorDoiUrl":"https://doi.org/10.1038/s41467-026-70401-7","workflowStages":[]},"version":"v1","identity":"rs-6466826","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6466826","identity":"rs-6466826","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}