Scientist Stories: Piero Carninci, Decoding the Human Genome

Thank you very much for the fantastic invitation. It's a big pleasure to be here. Of course, I would like to be there in person, but it's very good to meet and hear many of the friends and collaborators over the years. So, today I would like to talk about my perspective of the effort that has been going on for 20 years after the completion of the human genome. Of course, we know that the completion of the human genome has been given quite a lot of— has caused quite a lot of progress for many fields, of science, biology, medicine, and the achievement of the human genome has truly produced a remarkable progress of science. However, although we have understood quite a, quite a bit of how the genome works, and here I put some of the main facts about the genome and the variability between individuals Actually needs a little bit of more studies, particularly also because it will become obvious that only a minor part of the genome encodes for proteins. Clearly, over the years just after the completion of the Genome Project, various projects started to ask what is needed to characterize, to better understand how the genome is regulated. And immediately after the completion of the genome, the ENCODE project started, was announced in 2004. And in parallel, in here in Japan with the other colleagues, we were starting the FANTOM project. Was quite different ideas, and actually at the end it came— it turned out to be quite complementary to the ENCODE project. I also mentioned various other projects that have been very, very important to complement the human genome. But I'm going to focus essentially on those two projects because I know them better. I've been interacting particularly with ENCODE and I've been also deeply involved in the FANTOM project. In the FANTOM project, actually, we have been focusing over the years to understand what the genome is producing. We've been focusing on identification of RNAs, capped RNAs, in order to capture all the RNAs that are present in tissue as complete copies of them, so the full-length cDNAs. And as everybody already knows, we identify, we and others identify only 20,000 21,000 protein-coding genes. However, what was striking at that time is that there were other transcripts, particularly non-coding RNAs, and they were actually the majority of the transcripts that are present in the genome. At the beginning, we did not know so much about them, but year after year, many groups particularly focusing on the biology of each of those RNAs, start to study them, and it's become clear that a part of them, at least so far at the moment, no more than 4 or 5% of them have been further studied, and they regulate. They regulate chromatin, they regulate transcription, translation, they interact with the proteins. So they do quite different things, although for 95% of them, we still do not know anything about. Here are some numbers. Out of the FANTOM3, at the time it was mouse, telling that 63% of the orthogenome is transcribed and many are long non-coding RNAs, including quite a remarkable number of sense antisense. Actually, we inferred that 73% of genes show some sort of antisense, and for many of them we still do not know the type of regulation. So this is one of the first images of the— of the— of the— how the genes are in the genome. With many exceptions, so we find the genes that start— RNAs that start in one gene and finish in the downstream gene. We often find for the first time at the time quite a complex pattern with genes that are spanning into other genes, antisense region antisense of antisense, so quite complex. And we defined as transcriptional unit everything that clustered together on the exon. So in this case, with so many different transcripts, we have only 3 transcriptional units for this specific example, and 1 gene forest, and where there is nothing, we call it gene desert. So, but this also suggesting us that there are many different transcripts and the transcript variability still has to be fully understood. Of course, there were other groups and this is very inspiring work from Tom Ginger. It's been a real pleasure to work with him over the years, with him and his group over the years. And this was a very inspiring work because this shows that compared to the cytoplasmic poly(A)+ RNA. And actually, there are many more nuclear and nuclear poly(A)- RNAs. So this paper has been inspiring us for a long time with the idea that we should really try to capture all of this difference that usually is not represented in our cDNA libraries that are commonly used to clone our— to identify our preferred genes. So a lot of transcription, and this approach was complementary to what the ENCODE took. And here is one of the famous slides from the ENCODE 1. And the strategy was totally different, was to identify 1% of the human genome, to use some of the cell lines because they can easily be grown and replicated and distributed. And actually, this has been very important work to develop technologies to identify hypersensitive sites, regions bound by chromosomal— by regions of the chromatin bound by transcription factors. And At the beginning, it was all based on microarray hybridization. And suddenly sequencing started to replace all of this. It's been a very, very long technology marathon. At the same time, we also developed technologies to map promoters that actually turned out to be very complementary to many of the other technologies. This is the CAGE technologies. And to make the story short, essentially CAGE captures those randomly primed cDNAs and sequence only the initial part of the transcript that is very close to the initiation site. By doing this, we can essentially sequence at the initiation site. And with this, we essentially map gene expression, but also each promoter that drives expression in many different tissues. And on the right, there is a simple comparison showing the difference of CAGE that essentially focuses the signal on the initiation site with RNA sequencing that usually covers the whole body of the genes, but is not often very efficient to capture the initiation site. And just to show you a brief example of alternative promoters, just to show that all the CAGE tags here— this is a FANTOM3 many years ago— still shows that genes may have up to 9 or 10 different initiation sites. Many of them are tissue and tissue-specific. So, and actually This is quite important to understand gene regulation because the set of transcription factors in the epigenome differs in each of the cell types. So we did not plan to— we did not know that next-generation sequencing will develop so vigorously and strongly, but clearly the development of the various technologies has been really very welcome not only to sequence genomes, but definitely to understand what the genome is doing by using DNA-seq, ChIP-seq, RNA-seq, CAGE, and many more seq technologies that have really been enabled functional genomics. And of course, this curve shows the decrease of the cost. We would really welcome a further decrease of the cost. This seems to be quite stable over the past 4 or 5 years. And based on the sequencing, the next phase of the ENCODE had been expanding, of course, going genome-wide and including 147 cell types. And actually, there were some findings that were similar to what was the early observation of the phantom, and particularly one that has been discussed very extensively. 80% of the genome participate in some of the biochemical events. So it seems to do something, although we don't understand very well. Of course, other parts of the genome are important for understanding genetics and the role of the GWAS. And there is a complex, better understood quantitative transcription factor-chromatin RNA relationship, both in promoter and splicing. Transcription factors co-associate, and chromatin can be classified in multiple class. This has been quite, quite important project to introduce new concept to analyze the genomes. And here I just put some of the summary that in particularly this 62% here of the genome that produces some sort of RNA has been quite similar to previous observation and actually tend to converge with with observation in the FANTOM project, but of course also in other important projects. And so clearly pervasive transcription of RNA is there. RNA is a very important component, although we know still very little. So again, the FANTOM and the ENCODE have been evolving quite in parallel without any common planning. But often converging and comparing the data around the end of the projects or when the data were prepared. So in 2010, 2011, we started to think that we should prepare a map of promoters using CAGE from as broad as possible collection of primary cells and collection of tissue, because we really need to have a collection of human, of human of human promoter cells. And this was the Phantom 5 where we took 100 primary cells and followed up with 32 of them with the time course to see how the promoter changes during the activation over various time courses. And this is a little bit— a fairly complex slide, so we cannot go into all the details of all the different cell types. There are more than 150 human primary cells and many more tissues here. But just to give you some impression of all the type of sample from the adipocyte immunology and the mesoderm, endoderm, ectoderm, the differentiation of neurons and so on. And this was quite important to identify the diversity of promoter and something else that I'm gonna comment later. In particular, We did for all those samples CAGE. So with CAGE we can identify promoter and infer the type of network, the transcription factors that are important, that are enriched in the promoter, that are present in each cell. For a subset, RNA sequencing, small RNA sequencing, some other assays to identify all those different cells. And to have a first look But how looks this diversity? We've been clustering a variety of promoters that have similar expression features. And how do they map compared to the cells of origin? So we can see the whole diversity of promoter activity here. But also what we can see is that those tend to cluster for different cell types. So immunocells and central nervous system and placenta epithelial mesenchymal cells. And I think that this was a quite important exercise because we also found that the cancer cell lines, and many of them were used during ENCODE 1 and ENCODE 2, they occupied quite specific space of the transcriptome. But this suggested that to really broadly identify the human biology, we really need to identify and characterize many different cell types, in particular primary cell types, and from now on, as much as possible, of tissues out of the human body. Some surprises when you use those powerful technologies usually. You always find something new. One was that the promoter architectures can be different in different tissues. So the same promoter can have a different modality. So for instance, in this, in astrocytes, we find only single starting site here. But if you look at the immune cells, CD14 or CD4, the starting sites used migrate more upstream. And this is one of the encoder cell lines and everything moves even more upstream and there is nothing here from promoter that has this sharp end as a TATA box. So the idea is that promoters have submodalities of promoter. Within the very same promoter you have various elements that are differently active. Usually TATA box promotes transcription from a single position in the genome, CpG island, from a more distributed broad regions, and those can coexist in the same promoter. By counting here essentially 1, 2, and 3 groups, we had more than 223,000 promoters in human, less in mouse, not because the mouse is more simple, but because we just did not sample as much. The other things that we did not expect we have already seen a little bit with the ENCODE 2 project that we can also characterize enhancers by their bidirectional transcription. And actually, because of the selection of the samples, they tend to be quite tissue-specific. So the number of the enhancers that the Fantom identifies is much smaller than the number of the enhancers of the ENCODE. However, they tend to be— they have some good future because they are overrepresented in one cell tissue or group. And this may help you if you really are looking for enhancers that are very specific for a given cell type. And importantly, they also are important for genetics, and we have been comparing the GWAS that are mapping in exome, and those are the various diseases which they are related. So exome is a relatively small number compared to the promoters, that is mostly out of the exons, but even more in the enhancers. So this has definitely been helping to map and give some possible function to some of the GWAS over the years following this paper. And again, after a phantom project, there is always some ENCODE project, and this was a really remarkable achievement, and I really would like to congratulate all the authors of those papers. And actually, the number of the cells and the number of the datasets, the number of samples, has been massively expanded 500 biological tissue types. And I would like to say that probably the biggest achievement here is the registry of candidate cis-regulatory elements, so the CCRE. And so because they definitely can help us to understand what is happening here in the genome, what is regulated in the genome, and also and also in different cell types. So the registry is massively large with more than 926,000 candidate cis-regulatory elements, and again less in mouse, again because the sample was lower. And they cover a remarkable part of the human genome, about 8%, And also they cover 80% of elements marked by chromatin marks and a part of the previous GENCODE and FANTOM collections. So essentially, as always, data tend to converge, which is good and is also telling that those complementary projects are always very, very important. Another paper that I like very much is the landscape of chromatin loops in the human genome, which brings for 24 cell types types, everything in 3D, identifying at higher resolution, 125,000 loops, with variation among chromatin loops in different cells, which is also quite important, and also various features about the role of enhancers in genes that are associated with disease and associated with features of the— in the cofactor. And a third important finding, and again, I'm not trying to summarize comprehensively all the findings here, but just some, some of the most important ones, is anyway, is the RNA-binding proteins. This is only in two cell lines, but the extent of the experiment is very remarkable and identifies quite extensively with different technologies what RNA-binding proteins actually bind. And this is very important to understand the regulation at a different level from the promoter and enhancer level. But also, particularly our group is using this very extensively to understand the motifs in many of the long non-coding RNA that we are studying. And particularly we'll have some examples later if there is time to discuss them. So here is for the encode and a little bit what we are working on the long non-coding RNAs because they need to really, they really need some effort. We have been recounting the long non-coding RNA in 2017 with Chung. At the time when we published, we found some divergence with the gene code. We have been discussing after this. I did not update my slide, apologize for this. And actually those phantom cut that for which we found evidence from a CAGE promoter and additional chromatin marks are quite remarkable. This is a robust data set if you want to look at the permissive, so the aggressive way to count them there is even more. Importantly, Also, the long non-coding RNAs are mapping— have interesting mapping data compared to the disease. So if you look at the expression of the non-coding RNAs versus the traits, which is traits that have some map, so we find that very often the long non-coding RNAs that are mapping on a trait that is important for some brain disorder, see a list here, are actually expressed in the region and in the cells that are important for that disease. So there is quite a strong correspondence. This is a zoom in of a map for all the tissues that we don't have the time to discuss today, but actually we found this also to be the case for the, for all the other organs, so diseases and the expression of coding RNA tend to correlate. I will skip this and go into a little bit our summary slide on the possible function of the long non-coding RNA inferred by genome conservation, particularly transcriptionalization region conservation is this number. Conservation of exon, lncRNA implicated in GWAS traits, and lncRNA implicated in eQTL, so it's about 3,000, 2,000, and 13,000 each. So in total, we have some suggestion that 19,000 of lncRNA may have some potential function, if you also assume that conservation is associated with the function. With this in mind, we have been over the years running the Phantom6 with the idea to knock down lono-coding RNAs. See here on the left, we knock down and we see that there is something happening after knocking down, let's say transcription starts. And the whole network of all the cells will work. So actually we do this systematically. We analyze the long non-coding RNA to knock down. We do perturbation with a gapmer, so antisense that acts also on the non-coding that are in the nucleus. We extract RNA, prepare library, we sequence, and then we do the bioinformatics analysis. And then we jump here on the right side where we see the pattern. So this is by knocking down about 250 long non-coding RNAs. And looking at which pathways, which GO terms have been changing. At the same time, also looking at the cell morphology proliferation with some robot, with Incocyte measuring cell growth. And we can see interesting is that there is not just a stress response after transfection, but there are various aspects like splicing, translation, cell cycle, metabolism. So, they seem to do quite a bit of things. A caveat, the response is not always strong. You need to replicate the study very carefully. And very often the antisense may differ also because we don't know all the structure of the non-coding RNA. By using two different antisense against what we think is non-coding genes, we may hit two different splicing isoforms. Anyway, the paper was just published in Genome Research and the data is all available. Please feel free to download the data and there are several thousand cage libraries here as well. So, and you can download and check and draw your conclusion. We have a few more studies in the pipeline with the iPS cells on a similar work that I was that have not yet been published. This is a little bit of the interrogation of the function. This is not only genomics, but there's also experiment after scratching cells, how long does it take to regenerate with a different antisense? And it seems that for this gene, one antisense, this antisense of this zinc finger, actually one antisense inhibits quite efficiently the migration of cells. And we've been looking at cell migration and cell proliferation and correlating cell growth and proliferation with the GO terms and KEGG pathways that are coming out from the genomics analysis. And clearly we see that when we see less growth, the knocking down long non-coding RNA causes changes in genes related to apoptosis or apoptosis immune system. Otherwise, genes— if there is an increase of growth or proliferation, there is about cell cycle and DNA replication. So this seems to make sense. We also been working quite a bit to characterize lonocoding RNA that are in the chromatin. Particularly, we know that there is a considerable amount of lonocoding RNA attached to the nucleus in the chromatin. Those are some of the numbers. And with a former postdoc in my lab, Alessandro, and other members have been developing a method to cross-link RNAs bound to the chromatin, bound to the to genomic DNA. And after various steps, the key point is to add a linker that will ligate from one side specifically to the RNA, at the other side specifically to DNA. This is all in the cross-linked nuclei, so molecules don't move too much. So we think we capture quite a lot of the situation as it is cross-linked. After the cross-linking and the transforming of the RNA into cDNA and putting our linkers, we sequence in the Illumina and we do 1 or 2 lanes per each library and we studied this whole interactome. And actually, it is a pattern that somehow has a correlation with the Hi-C pattern, but of course one side is RNA, the correlation is between 0.5 to 0.6. But also we see quite different things as well. We found quite a lot of interaction within the same chromosomes between the RNA on the y-axis and DNA on the x-axis. And this interaction is usually caused by introns of protein coding genes. So this is quite surprising. So introns are often not degraded and they interact along the same chromosome. The other are those distributed spots, often in green, and then often are long non-coding RNAs or other non-coding RNAs that show some various interactions. Those are then the number of the events, huge number of interactions, and including also non-coding RNAs.. And this is one example, NIT1, that is not so active in mouse embryonic stem cells. So it interacts here with its own locus on the chromosome 19 and a little bit outside. But if you look on oligodendrocyte progenitor cells, this is all mouse, it interacts not only with its own locus but with different regions of the same chromosome and also in trans with different chromosomes. So we have hundreds and hundreds of those graphs. If you look at the paper also just recently out in Nature Communications, you can find all the data there as well. There is some specific direction of those interactions. RNA-chromatin interactions are enriched at TAD boundaries, and also there is an effect on the distribution, so the the interaction of RNA within the same TAD is larger than the interaction outside the same TAD. So it's telling us that those interactions probably have some biological meaning which we still do not understand. And importantly, also what we see is important role of repeat elements. We don't know how important it is, but repeat elements, if compared to non-repeat transcript, they tend to distribute over long range distance along the chromosome, and this is all statistically relevant for SINE, LINE, and LTR elements. Repeat elements, and I just bring here in the last few minutes one example of repeat elements that are embedded in in the transcript, and this is one repeat element embedded in this antisense in red. And actually, once we overexpress the antisense, it does not affect the level of the sensor RNA, but it affects the level of protein. So it overexpresses— it causes overexpression of the protein. This was published before, but actually what Andy Actually, this is caused by a sine element, sine B2 element. If you take this out, we don't see this effect anymore. So what we've been studying with Harshita, still unpublished work, is that this is not only one exceptional RNA, but we have a whole class of sine B2 RNAs, but that also have a human repeat element embedded in the transcripts. That are effective to enhance protein. So this is a class in the mouse, is a class in the human, but we've identified also in Arabidopsis, in horse, in the fishes, in human, and actually they work in different species. So repeat elements embedded in antisense enhance protein expression of the targeted RNA in different in speech. And is this caused by primary sequence? And by looking at the sequence, we believe it's not. It's mostly a combination of motifs. And I'll show you just very briefly using a model of COX7B, that is a helper insufficiency, that in medaka fish, the mouse, element can rescue a phenotype induced by morpholino that I need with splicing here, and a mouse SMB2 can actually enhance translation of the fish protein and rescue the phenotype. Of course, this is quite interesting for future application in the human, and again, Please do not underestimate the importance of repeat element. Now, perspective. Everything is going to move at single cell. And we just published last year the cage that works in single cell. First with the C1, we see similar things in the 10X. We identify enhancers. Bidirectional enhancers, but actually in each cell enhancers are essentially monodirectional. They are not bidirectional in each cell. On average they are bidirectional if you look at cell population. There's a good correlation with the DNA hypersensitive sites and chromatin marks, so essentially we can map in a single cell trace, identify promoters, and go into genetics working with single cells. Particularly, the concept here is that we will map those regulatory elements correlated to the SNP and try to understand the expression of QTL in each of the specific cell types. I've been going very, very quickly here. I'd just like to mention that this is also part of the Human Cell Atlas, And there is no need probably to discuss very extensively the Human Cell Atlas here, but we are also quite strongly engaged with this project. There is a Human Cell Atlas activity in Japan with a collection of data starting from various hospitals, and we're looking forward to do this. And some general details of how the project will will go. And there's also the Human Lateral Initiative in Asia. We're going to have the next meeting later this month in China, so please stay tuned. If you are in Asia, please contact us if you want to be a part of this. And finally, I would like to introduce the Human Technopole, one institute in Milan where I'm progressively moving a good part of my activity in in this year and more in the coming years. Thank you very much for your attention. This was done in collaboration with many, and there is no time, but I just leave this here for any questions that you may have. Thank you very much for your attention.