In this talk, Magda Bienko introduces her lab’s recent technological advances for studying 3D genome organization, focusing on a sequencing‑based radial‑mapping method called GP‑Seq (genomic positional/repositioning by sequencing). GP‑Seq uses progressive restriction‑enzyme digestion of fixed chromatin, visualized with a Y‑shaped linker and fluorescent FISH probes, to generate locus‑level radiality (GP6) scores that indicate a region’s distance from the nuclear periphery. The method yields high‑resolution maps (down to 25 kb) that correlate strongly with orthogonal 3D DNA‑FISH measurements. Bienko then shows how GP‑Seq data can be integrated with ATAC‑seq, histone‑mark ChIP‑seq, and Hi‑C subcompartment annotations, revealing distinct radial patterns of DNA accessibility, activating and repressive marks, and transcription‑factor binding across nuclear layers. Functional analyses demonstrate that UV‑responsive genes, transcription‑factor binding sites, and gene‑fusion‑prone loci exhibit characteristic radial distributions. To leverage these radial maps for 3D genome reconstruction, the lab developed ChromFlock, an algorithm that combines GP‑Seq with Hi‑C data to generate thousands of genome‑wide structural models that better recapitulate radial positions than Hi‑C‑only models, as validated by FISH. Using these reconstructions, the team tests the 1975 “bodyguard” hypothesis, finding peripheral enrichment of somatic SNVs in cancers and central enrichment of endogenous double‑strand breaks and fusion‑prone genes, with increased inter‑chromosomal mingling of fusion‑prone loci. The talk concludes with a summary of methodological achievements, ongoing developments, and acknowledgments.
Chapters
Hi everyone, my name is Magda Bienko and I am an assistant professor at Karolinska Institute in Stockholm, where I lead a lab for quantitative biology of the nucleus. And today it is my great pleasure to introduce to you our most recent technological developments with which we are trying to advance knowledge on how our genome is organized in the 3D space of the nucleus. In my lab, we are after this seemingly simple, in reality very complex biological problem, and that is how the 2-meter-long molecule of DNA can fit in the tiny volume of a mammalian nucleus compatibly with all its functions. By now we know that our genome is not entirely randomly organized, there is some structure to it for sure. We know that. For instance, individual chromosomes need to come together for nucleoli to form. We know that the centromeres tend to cluster, telomeres cluster as well, and that certain genomic regions have this strong tendency to be associated with the nuclear lamina to form lamina-associated domains. Individual chromosomes occupy discrete volumes in the nuclear space called chromosomal territories, and at least in some organisms, chromosomal homologues tend to pair or at least be located at certain reciprocal positions. And as I mentioned before, there is a big chunk of DNA that needs to fit in this small volume, so there is the problem of molecular crowding. And so, all in all, all these constraints lead to a certain arrangement of chromatin in 3D. But we are definitely far away from understanding it all. We do not know still, for instance, how individual arrangement of chromosomes, subchromosomal regions, how they differ from one cell to another, from in between different tissue types, let alone what are the driving forces that lead lead to a certain arrangement of chromatin in 3D. And so, in recent years my lab has been developing various methods, various tools and technologies to study chromatin architecture. We have been developing various microscopy approaches, mainly DNA and RNA FISH to look at individual topologies of DNA and how they relate to gene expression. We have been studying using DNA-FISH details of chromosome structures, their arrangement in 3D, and basically looking for various spatial patterns of their localization. And more recently, we have been developing various sequencing-based approaches, and today I am going to present to you one of such approaches, which we called GP-Seq. So, GPC stands for genomic loss repositioning by sequencing. It was developed in my lab by Gabriele Girelli, Tomasz Kallas, and Joaquin Custodio. And I always use this analogy to an onion for this method, because here what we are doing is sort of peeling the nucleus like an onion, trying to understand what DNA sequences tends to sit in which of its concentric layers basically. So, in GPC we have developed a protocol for gradual digestion of fixed chromatin by restriction enzyme. We rely on gradual diffusion of a restriction enzyme through a meshwork of fixed chromatin in cells whereby if we treat our material for a very short duration of digestion, the enzyme should digest only the very outer layer of chromatin. If we let the cells sit for longer, for a longer period of time with such an enzyme, the enzyme will go a little bit deeper and digest a thicker layer of chromatin, always starting from the periphery, and eventually it will go all the way to the center of the nucleus and the whole volume of the nucleus. Will become digested. We obviously wanted to make sure that we can visualize, we can monitor the enzyme digestion to make sure that the enzyme is actually doing what we wanted it to do. So, we developed this new FISH approach where we ligate this Y-shaped linker adapter to the digested sites, and then we are hybridizing short fluorescently labeled oligos to this linker, to this adapter, in order to visualize where which of the cut sites were digested by the enzymes, by the enzyme. And so, what you can see here, this is what we see under a microscope. In green is our whitefish signal, that's where the digestion happened. So, after 10 minutes of digestion, the enzyme will only cut the very outer layer, 15 minutes a little bit deeper, 30 minutes even deeper, eventually really filling in the entire nucleus with the cuts. Now, if instead of ligating Y-shaped linker, we ligate a linker that supports sequencing, we can generate sequencing libraries from these individual time points. And when we combine the sequencing data all together, we measure, we generate for each locus so-called GP6 score that tells us about its radial location, how distant it is from the nuclear periphery or the nuclear center. And so here I'm showing you a distribution of such a GP6 score from human Hap1 cells. You can see that the profile differs quite a bit all along individual chromosomes as much as in between individual chromosomes. The higher the value, The more central location is, the lower the value, the more peripheral. This map is at 1 megabase resolution, and we get very high reproducibility at 1 mega, which stays still very high even at 100 kb resolution. And now we can even go down to 25 kb resolution. We obviously wanted to validate this approach using orthogonal methods, And our gold standard is the 3D DNA FISH that allows us to measure distances in single cells, directly in cells under a microscope. We used 68 individually sort of randomly selected DNA loci, 8KB loci scattered all throughout the human genome. And we measured their distances, the typical distances from the periphery And we were very pleased to see that these distances correlate very highly with what we get from GP-seq at 1 megabase resolution as well as at 100 kb resolution. And so basically at this point, this gave us a green light to proceed to further analysis, being confident that what GP-seq reveals to us are true radial distances.. And so, what can we use GP6 score for or maps of radiality? We were very keen on adding this radial dimension to the study of various genetic and epigenetic features of chromatin. So, we combined GP6 score with various omics type of data. For instance, we used ATAC-seq data to study radial distribution of DNA accessibility in the nucleus, in the human nucleus. We divided the nucleus into 10 concentric layers with 1 being the very periphery and 10 being the very center. You can see with this dashed line that DNA accessibility as expected goes up towards the nuclear center. This is where most of the transcription is taking place, so this comes as no surprise. What we did on top of that was to stratify this analysis by Hi-C subcompartments. So, different chromatin types, A being the most active so euchromatin B, B being heterochromatin or the B compartment, with B1 being the polycomb-repressed chromatin, B2 pericentromeric mainly with some regions belonging to the lamina-associated domains and nucleoli-associated domains, and B3 being enriched with LADs, lamina-associated domains. And so, what you can see here from such analysis is that for instance when you look at regions belonging to the B1 subcompartment, so polycomb-repressed, they actually show higher DNA accessibility if they are placed in the very center than some of the A1s of the most active chromatin if that is placed more peripherally. So, it really shows you that this radial dimension adds a lot to the picture. Looking at various epigenetic features along the radius in different subcompartments, we really got this impression that these radial profiles change a lot from one sub-compartment to another. So, for instance, looking at H3K27 acetylation, again, as expected genome-wide with this dashed line, you can see it increases towards the center. But when you look, for instance, at the B2 sub-compartment, so pericentromeric, for instance, it actually goes down towards the center, this mark. When we looked at the inactivating mark H3K9 trimethylation, we could see opposite trends. So here genome-wide it seems to be going at least slightly down towards the center. But again, for the B2 subcompartment, it looks very different. There is a sharp increase of this mark towards the center. And that basically made us think of heterochromatin that is present on the small chromosomes like chromosome 17 or 19 that are known to be present typically in the very center of the nucleus. They do carry heterochromatin, they have pericentromeric regions, but these regions are exposed to very highly active sort of transcriptionally active environment. And so perhaps they really need this boost of inactivating mark and a depletion of activating mark. What we also looked at was whether genes that belong to the common pathway show any peculiar radial behavior. Of their distribution. And so what we saw, this was particularly interesting for us, was we saw that basically genes that are responding to UV irradiation, if they get upregulated upon UV, they tend to cluster in the center, while if they are downregulated upon UV, they show the opposite trend, accumulating more towards the periphery of the nucleus, which we find really interesting. We also looked at distribution of transcription factor binding sites, and we could see that for the majority, vast majority of transcription factors, their binding sites show actually strong radial gradients. So either anti-correlating or correlating highly with GP6 score, indicating again some functionality to it. And that is something that we are now trying to explore in greater detail. Now when we developed, or already when we conceptualized GP-Seq, it occurred to us that having these highly accurate radiality maps could be very useful in genome reconstructions, or basically in using data from Hi-C method that reveals inter-intrachromosomal contacts and basically placing these contacts properly in the space of the nucleus. And so Erik Wernersson, a senior scientist in my lab, developed an algorithm which he called ChromFlock that combines GP-seq data with Hi-C data. It's based on the PGS software from Frank Albers' lab that uses Hi-C only. And basically he the algorithm generates thousands of single instantiations of genome structure, either at 1 megabase resolution or at 100 kb resolution. Still, even at 100 kb, recapitulating basically the expected pattern of arrangement of various subcompartments. When we looked at our structures from ChromFlock, we could see that they correlate very highly with our GP-seq maps, as expected since we are using them to generate the structures, and they correlate very highly with the original Hi-C maps, again as expected, sort of a proof that we are doing something right here. But when we generate such structures using Hi-C only, they actually do not correlate very well with our radiality map. So we believe that they do not preserve, or Hi-C might not be enough to position properly contacts in the space of the nucleus. And even when we look at correlation with the Hi-C data, the original input data, the correlation looks worse than when we use also GP-seq to generate these structures. Most importantly though, When we look at correlation of these structures with the 3D DNA fish, we see also a very high correlation for our structures when we use Hi-C and GPC, and no correlation really when we use Hi-C only. We went a step ahead further and basically wondered what would happen if we entirely omitted trans contacts from Hi-C, so really focusing only on intrachromosomal contacts, taking only those with GPC. What would happen if we generate created 10,000 structures there. And so, as expected again, they correlate very well with the GP-seq, they actually correlate equally well with the original Hi-C data that has both trans and intra contacts, and to our big surprise but also big satisfaction of course, they correlated very highly with our FISH data, again suggesting that that GP-seq does a really good job in positioning contacts properly in the space of the nucleus and doing a good job in aiding in the reconstructions of the genome. When we generated, obtained our radiality maps, given the fact that here we have sort of data all along the radius. gPCIC allows us to study radiality all across the radius, not only looking at what is at the very periphery versus everything else, or what is in the very center versus everything else. We really profile everything all throughout the radius. And so we were very keen on using our data to test the longstanding so-called bodyguard hypothesis proposed back in 1975 that posits that that says that the very outer layer of chromatin, heterochromatin sitting at the very periphery, acts as a shield protecting euchromatin from insults by mutagens that would enter the nucleus from the outside. And so to address this, we looked at the distribution, radial distribution of single or SNVs, single nucleotide variants found in different cancer types. And basically looked for their density in different radial layers. And we could indeed see that they do accumulate, they are enriched at these peripheral layers in comparison to the more central ones, especially pronounced for cancers such as melanoma or lung cancer. On the other hand, when we looked at radial distribution of endogenous DNA double-stranded breaks, we could see that those tend to accumulate more in the very center of the nucleus. And this made us wonder or made us think about a related phenomenon, basically gene fusion found in cancer. And we want— because we know that double-stranded breaks are required for gene fusions to occur. And so we wondered about where genes which are found frequently being engaged in gene fusions and chromosomal translocations, where would those be located radially speaking? And indeed, we could see that also in this case, they do accumulate or they are enriched in the very centre of the nucleus as compared to regions that do not carry SARG genes. And we moved again a step further, digged a little bit deeper Thanks to the fact that we have chromatin structures, we could actually ask more questions sort of going into further mechanistic details of gene fusions and ask a question of whether beads in our structures that carry genes engaged in gene fusions would mingle more. We hypothesized that such regions would be engaged much more in interchromosomal interchromosomal mingling. And so, we asked exactly that and we could indeed see from our chromatin structures that beads which carry genes engaged in gene fusions, they are much more frequently engaged in interchromosomal mingling events. All of this basically pointing to certain mechanistic details behind chromosomal translocation formation, for instance. So, sort of we like to think that this approach is really good for studying the interplay between genome structure and genome fragility. And with this, I basically summarize my talk giving you some take-home messages. We have established a protocol for gradual penetration of proteins through chromatin in morphologically preserved cells. I showed you this for restriction enzymes, but we tested also for antibodies for instance, and that works really well. GBSeq allows us to generate highly resolved radiality maps. ChromFlock generates state-of-the-art genome reconstructions. Our data is in line with the bodyguard hypothesis. And the GBSeq serves the basis for new methods developments, and that is sort of ongoing right now in the lab, so please stay tuned. And with that, I would like to thank all the members of my lab, as well as all our collaborators, as well as all our funding bodies. And I thank you for your attention, and I'm happy to take questions. Thank you.
