Genome Architecture from a Different Angle

Abstract

The study of genome architecture has recently been advanced by new techniques combining nuclear proximity ligation and high-throughput sequencing, but independent methods to validate them have been lacking. Reporting in Nature, Beagrie et al. (2017) describe such an orthogonal technique, called genome architecture mapping, to map genomes in 3D space.

Genomes are intricately organized in 3D space in the cell nucleus (Misteli, 2007). The complex higher-order organization of genomes is functionally important, whether by bringing a gene into contact with potent enhancer elements or suppressing a gene by compacting it into dense heterochromatin. Genome organization is developmentally regulated and frequently disrupted in disease. As such, clarifying the way DNA folds within the nucleus by genome mapping is crucial to understanding the regulation of genomes.

Standard biochemical 3D genome mapping methods, generally referred to as C-methods, crosslink chromatin to preserve spatial interactions, fragment the genome with a restriction enzyme, religate the generated fragments, and pull down ligation products for sequencing (Figure 1, right). Regions that are physically adjacent are expected to be more likely to crosslink, ligate, and be pulled down, and they will thus be enriched in the population of sequenced reads (Cullen et al., 1993; Lieberman-Aiden et al., 2009; Fullwood et al., 2009). Reporting in Nature, Beagrie et al. (2017) have now developed a method called genome architecture mapping, or GAM, that is fundamentally different. It uses cryosectioning and microdissection to isolate thin, randomly oriented slices of individual nuclei, each containing about 5% of the genome. These sections are then amplified by PCR and sequenced. Interactions are identified based on the simple rationale that two loci that are frequently near each other in 3D space will be more often found in the same slice, whereas distant loci are less likely to be found in the same section (Figure 1, left). GAM is an important orthogonal method for standard genome mapping techniques.

Figure 1. Schematics of Methodology of Genome Architecture Mapping and C-methods.

In genome architecture mapping: physically proximal regions are selected by taking a thin section of the nuclei; individual cells are selected via microdissection; whole genomes of each nuclear section are sequenced; and interaction frequency is approximated by the number of times two loci were found in the same slice. In C-methods: physically proximal regions are selected by crosslinking, digest, and subsequent ligation; the ligation junctions are then sequenced; and interaction frequency is approximated by the number of times a given ligation junction is observed. While individual cell-level data can be generated after FACS sorting, this is not routinely done in standard C-methods.

Using GAM, Beagrie et al. (2017) confirmed key findings of standard genome mapping methods, including the existence of topologically associated domains (TADs) and of structurally distinct large-scale chromatin compartments. In addition, they identified new features. They find that a full one-third of the most enriched interactions were separated by genomic distances of more than 60 Mb, suggesting that GAM may be better at detecting very long-range associations than C-methods, possibly as a result of improved normalization and the fact that GAM uses 220 nm sections, whereas the reach of crosslinking in C-methods is estimated to be only ~100 nm. Importantly, the interacting regions they identify are highly enriched for previously identified regions containing clusters of enhancers (“superenhancers”) and active genes, suggesting that they may represent functional interactions. They also observed some evidence of superenhancers and active genes interacting in triplets, across tens of millions of base pairs. This analysis would be impossible in ligation-based methods, which, by their nature, rely on the detection of only pairwise interactions. In all, these results not only show that GAM is a confirmatory method, which overall validates general principles of genome structure, but also reveals novel features of genome organization.

GAM provides three key advantages over standard ligation-based methods. First, because ligation by necessity involves a pairwise interaction, C-methods can only determine pairing events and cannot measure clustering of multiple gene loci. GAM, by contrast, can detect clusters of any numbers of sequences, because it probes spatial subportions of nuclei without ligation and simply based on the physical proximity of loci in space. Second, ligation-based techniques are biased toward loci with a high density of restriction enzyme sites, regions where crosslinkers and restriction enzymes can easily bind, and sequences that are more likely to ligate (Simonis et al., 2007). GAM does not have these biases and calculates the expected cosegregation frequency purely based on genomic distance, making data normalization considerably more tractable. Third, while it is technically possible to perform ligation-based methods on single cells, this is currently far from the standard procedure, and most studies using C-methods require millions of cells. GAM, by contrast, generates interactions maps of 30 kb resolution using as few as 400 cells.

As with all methods, GAM has some drawbacks. Chief among these is the amount of time required to individually section and dissect out nuclei. This would be especially limiting if the method were applied either to studying specific interactions between a small set of loci or to comprehensively mapping the higher-order regulatory relationships between sets of interactions in single cells. Further, although observing a ligation product in C-methods ostensibly requires that an interaction take place in at least one cell—or else those two pieces of DNA will not be tethered together to ligate—observing co-segregation in one nuclear section by GAM does not assure interaction, because the regions may be separated by a large distance in the same section. Finally, GAM requires that nuclei being compared have roughly consistent shape, because differences in nuclear shape will lead to different expected values in normalization. A rounder, taller, or overall larger nucleus will have a smaller portion of the genome sorted into any given 220 nm slice than a nucleus that is flatter, shorter, or smaller. Thus, the expected number of co-segregation events for two unlinked loci will be lower in the former case and higher in the latter, and these two populations of cells would best be normalized and analyzed separately.

One of the most significant, and potentially impactful, applications for GAM is in studying chromatin organization in situations in which the sample is limiting and in which sectioning and dissection is already done or provides additional advantages. Clinical samples are an example of this category, as they are often rare and already sectioned, and there is significant advantage to categorizing cells pathologically by tissue type or severity of phenotype. Given the scarcity of information on 3D organization in clinical samples, and its potential importance for diagnostics and understanding of disease mechanisms, this may be a particularly relevant application of GAM. In addition, the ability to interrogate single cells opens the door to probe tissue heterogeneity and differential spatial genome organization in different parts of a tissue or organ. As such, this technique could be applied to examining how genome structure changes in disease (Meaburn et al., 2016). Furthermore, if microdissection and sectioning can be automated, the advantages of this technique’s easier normalization and ability to detect long-range interactions and higher-order clusters might make it as widely used as ligation-based methods. Regardless, this powerful method provides a different angle from which to examine nuclear organization.