Human pancreatic islet 3D chromatin architecture provides insights into the genetics of type 2 diabetes

Irene Miguel-Escalada, Silvia Bonàs-Guarch, Inês Cebola, Joan Ponsa-Cobas, Julen Mendieta-Esteban, Delphine M.Y. Rolando, Biola M. Javierre, Goutham Atla, Irene Farabella, Claire C. Morgan, Javier García-Hurtado, Anthony Beucher, Ignasi Morán, Lorenzo Pasquali, Mireia Ramos, Emil V.R. Appel, Allan Linneberg, Anette P. Gjesing, Daniel R. Witte, Oluf Pedersen, Niels Garup, Philippe Ravassard, David Torrents, Josep Maria Mercader, Lorenzo Piemonti, Thierry Berney, Eelco J.P. de Koning, Julie Kerr-Conte, François Pattou, Iryna O. Fedko, Inga Prokopenko, Torben Hansen, Marc A. Marti-Renom, Peter Fraser, Jorge Ferrer

Posted on: 20 September 2018 , updated on: 6 October 2018

Preprint posted on 27 August 2018

Article now published in Nature Genetics at

From 2D to 3D: A chromatin interactome map of human pancreatic islets provides a 3D view into type 2 diabetes and pancreatic islet biology

Selected by Carmen Adriaens


Introduction and context

One of the main two types of diabetes, diabetes mellitus (type 2, T2D) is a polygenic disease that occurs usually in obese subjects and results in systemic peripheral tissue insulin resistance (DeFronzo et al. 2015). A popular approach to study common polygenic diseases is to look at genetic variants that occur in the population and correlate them with the risk for the disease (Torkamani et al. 2018). In the case of T2D, several recent efforts have implicated common Single Nucleotide Polymorphisms (SNPs) in the T2D phenotype and identified a number of susceptibility loci e.g. (K. Sanghera and R. Blackett 2012; Mishra and Hawkins 2017; Mahajan et al. 2018). However, deriving functional and causal information from these variants and their associated genes often proves difficult, especially when they occur in the non-coding portion of the genome, and when information is lacking on what the target genes are and how the variants affect them.

So far, previously identified T2D SNPs and their function in the T2D phenotype have only been studied in the context of a linear DNA map (eg. Which cell-specific transcription factors bind in the enhancer clusters in which the SNP occurs? Or, how do the different variants contribute to gene expression within the Topologically Associated Domain (TAD)?). However, how these SNPs specifically contribute to the disease phenotype on the molecular level in the context of the 3D genome remains to be elucidated.


The preprint

In a recent preprint, members of the group of Prof. Jorge Ferrer and colleagues set out to construct a chromatin interactome map of human pancreatic islets, with the goal to determine beta-cell specific enhancer-promoter interactions and to link variation in the non-coding portion of the genome to the relevant target genes. First, they combined ATAC-sequencing and ChIP-sequencing for Mediator, Cohesin and H3K27Ac with promoter-capture hi-C (pcHi-C). This latter technique increases sensitivity of Hi-C by using fragment baits that cover all known promoter regions (Schoenfelder et al. 2018) to detect lower-frequency, cell type-specific interactions between promoters and their potential regulatory sequences. These experiments demonstrated that islet-specific interactions are strongly enriched for islet-specific enhancers showing ample binding of Mediator, a protein complex known to regulate gene expression by bridging the signals from transcription factors in regulatory regions to RNA Polymerase II. Because these enhancers also showed the most prominent H3K27 acetylation, the authors named these enhancers Class I enhancers.

To identify the target genes of (Class I) pancreatic islet enhancers from Hi-C, the authors first needed to address an important limitation of the Hi-C technique: its data analysis is highly stringent and shows a strong proximity bias and excessive tissue/physiological condition specificity. Therefore, cell type-specific DNA proximity outcomes can easily be underestimated. To overcome this problem, the authors imputed additional potential interactions based on promoter-associated 3D space (PATs). Here, a PAT is defined as the ‘space’ that is constructed from taking together all pcHi-C interactions of a specific bait. Like this, they found that the PATs that show high confidence interactions with islet specific enhancers have a higher chance of also interacting with other enhancers. Next, the PATs were used to identify the target genes of the T2D-relevant enhancers (i.e. those that showed disease-relevant regulatory variation in their sequences). These functional interactions were then validated with several approaches, a.o. using CRISPR-Cas9 modulation of the PAT enhancers, some computational work, and a dynamic-perturbation strategy in which the activity of the enhancers could be modulated by challenging the cells with high or low glucose media.

Next, the authors assessed the T2D-associated SNPs locating to these 3D regions and asked if they impact on the structure and function of enhancer ‘hubs’ defined by the PATs and their secondary interactors (i.e. their target genes, and other interacting enhancers). Suggestive of a strong tissue-specific role, they observed that the SNPs in the 109 T2D – Fasting Glucose (FG) risk loci were enriched in high-confidence pcHi-C regions with increased Class I (high H3K27Ac – Mediator occupancy) enhancers. Furthermore, they showed also that these SNPs showed higher genetic heritability scores, and thus that they play a prominent role in the heritability of islet function and T2D.

Finally, the authors asked if they could use the islet enhancer hub SNPs to better predict the risk for T2D. They found that, although quantitatively the hub-SNVs were not better at predicting T2D risk than the collection of all common SNPs genome-wide, hub-SNPs could help stratify patients qualitatively. For instance, the Polygenic Risk Scores (PRS) calculated from hub-SNPs in non-obese, younger patients showed higher Odds Ratios than the PRS in obese, older patients, indicating that in these younger individuals the variation occurring in the genomic regions that affect islet transcription impacts more strongly on T2D. Like this, the authors postulated that hub-only SNVs may be useful to further study how islet-specific regulatory variation affects T2D pathogenesis.


What I think about this preprint & open questions

I chose this preprint because it addresses a critical question in cell / systems biology: How does the three-dimensionality of chromatin contribute to cellular identity and disease?

First of all, I like that the 3D interactome map constructed by the authors is browsable with a user-friendly interface; and thus that any individual could go to his/her favorite gene and look at its 3D environment. Furthermore, the authors push the limits of their data by several means, and, although many of these imputed interactions A) may be dynamic/transient and B) may need to be subjected to experimental validation, I love the idea of using ‘hubs’ to represent cell type specific, 3D spatial interactions. Indeed, with this approach, the authors overcome some of the limitations inherent to genome-wide chromatin-capture techniques and identify potential novel, cell type specific or even disease-specific domains, laying the groundwork for further studies.

The fact that many of the ‘hub’ interactions are imputed may also confer a strong limitation to the work: there is no direct evidence that these hubs or all the interactions within them are actually formed (simultaneously), and interactions may be mutually exclusive, or temporally distinct. Therefore, it would be interesting to study if these interactions occur and what their dynamics are using for instance imaging-based methods (e.g. DNA-FISH, or CRISPR/Cas-guided locus identification in live cells, see eg. (Chen et al. 2013)).

With this work, another question emerges: are these 3D interactions functionally relevant to pancreatic islet biology? For instance, would it be interesting to study a time course and see if the enhancer becomes inactive (loses active chromatin marks) before its target gene expression does? Are the 3D connections lost in these and other (experimental/physiological) circumstances?

I am further intrigued by the Class I enhancers defined in this study by high Mediator occupancy and H3K27Ac. Several recent studies (eg. Cho et al. 2018; Sabari et al. 2018) demonstrated that Mediator and RNA Pol II are involved in a phase-separated complex involving super-enhancers. Thinking of how 3D chromatin architecture helps to establish cell identity mechanistically, the question emerges whether the observations in these recent papers can be extended to the islet-specific Class I enhancers observed in this study. More generally, does the liquid demixing through Mediator and RNA Pol II binding provide a chemically favorable environment to establish relevant DNA-DNA contacts in a cell-type specific manner? Thus, it will be interesting to study further what these Class I enhancers are, how they emerge and are maintained, and if they contribute to phase-separation mediated transcriptional output. Furthermore, it remains to be determined what the driving forces are to establish the liquid-demixed environment in this cell type specific manner.

In conclusion, I like this study because it advances the understanding of 1. islet-specific enhancer-promoter interactions and their potential influence on cell-type specific gene expression in the context of human pancreatic beta cell islets; and 2. the functional and mechanistic role of disease-associated variants in establishing relevant interactions in the 3D genome leading to disease through disruption of the normophysiological 3D architecture. Finally, this study also demonstrates the importance for continuous validation studies of predicted interactions in a cell type or physiological condition specific manner, and ultimately may be an important resource to further our understanding of cell identity and how it is influenced by local chromatin architecture.


References/Further reading

Chen B, Gilbert LA, Cimini BA, et al (2013) Dynamic imaging of genomic loci in living human cells by an optimized CRISPR/Cas  system. Cell 155:1479–1491. doi: 10.1016/j.cell.2013.12.001

Cho WK, Spille JH, Hecht M, et al (2018) Mediator and RNA polymerase II clusters associate in transcription-dependent condensates. Science 361:412–415. doi: 10.1126/science.aar4199

DeFronzo RA, Ferrannini E, Groop L, et al (2015) Type 2 diabetes mellitus. Nat Rev Dis Prim 1:15019

Sanghera D, R. Blackett P (2012) Type 2 Diabetes Genetics: Beyond GWAS. J Diabetes Metab. doi: 10.4172/2155-6156.1000198

Mahajan A, Taliun D, Thurner M, et al (2018) Fine-mapping of an expanded set of type 2 diabetes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps. bioRxiv. doi: 10.1101/245506

Mishra A, Hawkins RD (2017) Three-dimensional genome architecture and emerging technologies: Looping in disease. Genome Med 9:87. doi: 10.1186/s13073-017-0477-2

Sabari BR, Dall’agnese A, Boija A, et al (2018) Coactivator condensation at super-enhancers links phase separation and gene control. Science 361:6400. doi: 10.1126/science.aar3958

Schoenfelder S, Javierre B-M, Furlan-Magaril M, et al (2018) Promoter Capture Hi-C: High-resolution, Genome-wide Profiling of Promoter Interactions. J Vis Exp 136:. doi: 10.3791/57320

Torkamani A, Wineinger NE, Topol EJ (2018) The personal and clinical utility of polygenic risk scores. Nat Rev Genet 19:581–590. doi: 10.1038/s41576-018-0018-x




Read preprint (3 votes)

Author's response

Inês Cebola shared

Thanks for selecting our recent preprint to be highlighted in preLights. We are looking forward to see the readers’ comments on it!


Unlike common diseases such as cancer, we are still very far from having therapeutic and preventive interventions that target the molecular basis of type 2 diabetes (T2D). Instead, we currently rely on the management and treatment of its symptoms. Previous work has shown that T2D-associated genetic variants are specifically enriched within pancreatic islet transcriptional enhancers, leaving open the question of which are their target genes. The identification of the true effector genes of T2D risk is key to move forward in the targeted medicine front.

Historically, disease risk haplotypes have been assigned to target genes using linear genome distance, but enhancers and other regulatory elements may act over large distances, even skipping proximal genes. This genome-wide 3D map links T2D-associated variants to their target genes in human pancreatic islets, a tissue that is at the core of the T2D aetiology. More than 70% of enhancers carrying T2D variants interact with unexpected genes, often located hundreds of kilobases away. This observation across many T2D-associated loci brings to light an array of novel genes that may be involved in islet-cell function and whose dysregulation contributes to T2D risk.

The human pancreatic islet interactome also allowed us to identify islet enhancer hubs, which are restricted 3D compartments of the genome where large numbers of enhancers tend to interact with genes related to cell-specific functions. Our data suggests that enhancer hubs are tissue-specific functional units that regulate cell-specific gene expression, although there is still work to be done to understand the dynamic nature of hubs.

The linear genomic space that defines hubs is much smaller than other commonly defined regulatory units such as stretch enhancers, or even super-enhancers, because these tend stitch together enhancers in the linear space without considering the 3D structure of the chromatin. Accounting for how in this case less is more, hubs perform better at finding the SNPs that contribute to T2D heritability. Of course, we did not expect to explain all genetic variation that contributes to T2D within islet hubs, since there are other tissues known to be involved in T2D development. Still, using only genetic variants within islet enhancer hubs we were able to take the first steps in stratifying the population according to their risk of T2D due to islet dysfunction. This is important because T2D is very heterogeneous, and although these are still early days in the polygenic risk score applicability to the clinic, the notion of applying them to stratify populations according to different types of susceptibility will be of great value in efforts to design personalised therapies and preventive measures in the future.


Inês Cebola

1 comment

5 years


Great job!
I would like to know, whether the TAD data of the pancreatic islet is avaliale for public at this moment?
Thank you very much.

Have your say

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sign up to customise the site to your preferences and to receive alerts

Register here