NEW YORK (GenomeWeb) – An international team led by investigators in Germany has described a ligation-free approach for charting genome architecture mapping and untangling chromatin contacts across the genome.
The method, known as genome architecture mapping (GAM), pairs DNA sequencing with laser microdissection-based cryosectioning, producing sequence reads from series of cell slices that can be used to interpret when and where different chromosomal regions bump up against one another.
"In our first proof of principle of the concept, we decided to use off-the-shelf technologies to see if it worked, so we chose whole-genome amplification, with a very small change to the protocol, and Illumina [sequencing]," co-senior author Ana Pombo, an epigenetic regulation and chromatin architecture researcher affiliated with the Max-Delbrück Centre for Molecular Medicine's Berlin Institute for Medical Systems Biology, said in an interview.
While that did the trick to show that the general method worked, there is now room to come up with amplification and library prep strategies specific to GAM, she noted. "There is a lot of opportunity, in the future, for optimization and we hope some of the companies in the fields of genome amplification and sequencing will be interested in helping us to make it more tailor made, to have a protocol that's specific for these samples."
For an analysis appearing in Nature this week, the researchers use GAM in combination with fairly standard whole-genome amplification and Illumina multiplex sequencing approaches, to map out enhancer and active gene interactions in mouse embryonic stem cells. Their results were consistent with what was known about chromatin interactions in this cell type from past studies, while highlighting complex contacts that were less well described.
Generally speaking, GAM is intended to overcome limitations of chromatin conformation capture (3C) and subsequent ligation-based chromatin interaction detection methods, which can miss complex interactions between multiple chromatin regions.
It also has the potential to help investigators figure out if chromosomal interactions are altered during the various steps needed to do such experiments, University of Massachusetts researcher Job Dekker, a genome organization researcher who was not involved in the study, said in an interview.
"It gives you a completely orthogonal approach to study the 3D structure of the genome and I think this is a really important development for the field," he said. "It's always good to have independent, orthogonal approaches to study the same problems."
Both Dekker and Pombo are part of the National Institutes of Health's 4D Nucleome consortium that is applying varied methods to the same sets of cells to map the genome's 3D structure and organization.
"[GAM] is a really terrific new method for the field of 3D genome studies," Dekker said. "The field was dominated by just a few methods and we need more."
Because the new method typically considers nuclear slices from cells in set locations, Pombo and her colleagues are also considering ways of using GAM datasets to interpret chromatin interactions within each cell, while also pulling back to see interactions between some of the cells containing these nuclei in a given tissue.
"You have your very thin tissue slice and you can collect the cells while recording their geographic position or you can collect the cells based on … any marker you want," Pombo said. "You can choose your single cells with some knowledge of the spatial organization."
Although she and University of Cambridge pathology researcher Paul Edwards came up with the general strategy behind GAM more than a decade ago, the approach only became technically possible with more recent DNA sequencing and bioinformatics advances, Pombo noted.
"We were discussing way to map the physical proximity of every locus in the genome against every other locus," she explained.
The strategy was borne out of such discussions at a time when Pombo and her team were primarily relying on fluorescence in situ hybridization to map chromatin proximity, a lower resolution method that forced them to first select regions of interest and then try to interpret whether the interactions they saw were authentic representations of interactions in cells that had not been FISH labeled.
By bringing so-called Happy linkage mapping together with the type of tissue slicing and dicing used in cryo electron microscopy, the researchers suspected they might be see 3D chromatin interactions in a new way.
After using laser dissection to essentially take a tiny deli slicer to a frozen tissue sample, they are left with impossibly thin cell slabs — some containing bits of nuclei and some that don't. Zoom in further, and each slice of nucleus harbors chromosome bits that were closely packed together before the cryosectioning step. At that point, it's just a matter of extracting, amplifying, and sequencing the DNA to see which sequences turn up together.
But it might not be quite that simple. For a long time, the resulting samples were too technically difficult or expensive to sequence and piece interactions in the nucleus back together again bioinformatically, Pombo explained. That changed with the advent of relatively affordable, multiplexed, high-throughput sequencing methods and other recent advances.
"When Illumina came out, you could sequence, but you could not multiplex — we could show it was possible to make a library from a nuclear slice, but we could not collect all of the datasets because it would have been extremely expensive," she said. When multiplexing with Illumina, for example, became possible "that's when we felt we were ready to go and we implemented the multiplexing to collect our first dataset."
In parallel, co-senior author Mario Nicodemi at the University of Naples spearheaded development of the bioinformatic methods needed to make sense of the resulting datasets, which led to "statistical inference of co-segregation" (SLICE) — a mathematical model designed to discern intermingling enhancers and active genes over long genome distances.
For the mouse embryonic stem cell experiments included in the current study, she and her colleagues did not attempt to tease out specific cellular interactions. Rather, they relied on randomly selected cells in the samples, preparing the nuclear slices from these cells for multiplex sequencing with existing whole-genome amplification and Illumina library preparation methods.
For their initial experiments, Pombo explained, the investigators erred on the side of conditions developed in the electron microscopy field to stabilize the cell's original organization and opted against staining the DNA for their initial experiments. Though some slices did not contain detectable DNA, it found that each of the mouse chromosomes was represented roughly equally after sequencing, arguing against significant biases due to WGA or other steps.
By adding in more and more information on the cells being profiled — whether they express certain transcription factors or long non-coding RNAs, for example — GAM should provide still more regulatory clues and context, Pombo said.
From datasets for 471 mouse embryonic stem cells, the researchers uncovered potential super-enhancers and saw instances of interactions involving multiple genome regions. These results hint that regulatory interactions might be more complex than previously appreciated, though Pombo cautioned that some chromosomal regions may cluster by coincidence, meaning not every interaction can be considered functional just yet.
More broadly, "it's very nice to see that the results they obtain are very consistent with the Hi-C data," Dekker said. He noted that he is keen to take a crack at using GAM, though there are elements of the approach he suspects might be technically challenging, particularly for labs that lack the necessary specialized equipment.
"No doubt this is version one of this method and they will improve on this method," he said. "But I think this is a remarkable study."
Pombo and her colleagues have applied for a patent related to the GAM pipeline, though the corresponding software will be made freely available to other researchers.
At the moment, GAM costs roughly €50 ($53) per sample. In the case of the mouse embryonic stem cell dataset, for example, the team considered about 400 such samples. But the resolution — and price — of GAM notches as additional samples are considered.
The team is now doing GAM experiments on cells taken directly from tissues without fractionation and is particularly interested in using the method to study regulatory networks in pluripotent embryonic cells, as well as terminally differentiated cells, including some neuronal subtypes.
More broadly, the researchers would like to take a crack at combining GAM with other sequencing technologies, including long-read platforms or single-molecule sequencing methods that might eventually make it possible to see longer-range interactions, improve mapping in repeat-rich regions, and/or offer a look at DNA modifications such as cytosine methylation in parallel with the genome organization profiles.