Scientist Stories: Vivian Cheung, Beyond DNA Sequence variation in the RNA

Good afternoon, everybody. I'm Janine Clayton, the Director of the Office of Research on Women's Health and the NIH Associate Director for Research on Women's Health. And it's my pleasure to introduce today's wall speaker, Dr. Vivian Cheung. Dr. Cheung was nominated for her lecture today by the Committee on Women of Color, a part of the NIH Working Group on Women in Biomedical Careers. This committee seeks to increase visibility of prominent women scientists, and it's my pleasure to co-chair that working group with Dr. Francis Collins. Dr. Chung is the Frederick G. L. Hudwell Professor of Pediatrics and Professor of Genetics at the University of Michigan. She has also been a Howard Hughes Medical Institute investigator since 2008. The Chung lab focuses on the genetics of human gene expression Her lab elegantly blends basic and clinical research. Dr. Chung is a strong proponent of precision medicine. In fact, she has stated that it's her dream that everyone will have a genetic test done so we'll know how to best treat each person. Today, Dr. Chung will present Beyond DNA: Sequence Variation in RNA. And I want to make sure that everybody knows we have a reception immediately following this lecture sponsored by FAAS, and it will be in the NIH library. Please join me in welcoming Dr. Vivian Cheung.

Great. Thank you very much, and I feel very honored to be invited. What I'd like to do is to tell you about our recent work. In the last year, I have the honor of spending some time with Kurt Fischbach's group NINDS. And my lab has a longstanding interest in RNA biology, and I'm clinically trained as a pediatric neurologist. And my two sides of my career in basic research and clinical medicine really have kind of advanced in parallel rather than kind of coming together. And the year that I've been spending at the NIH has really allowed me to bring together our interest in RNA biology with several neurogenetic diseases, and so what I'd like to do is to show you how by bringing these things together, we have deepened our understanding of a discovery that we made of RNA sequence variation and also led us to better understand a rare form of ALS. So we know for a long time that this is not a simple one-to-one relationship between DNA, RNA, and protein. And in fact, that one DNA sequence becomes different RNA transcript and even different protein isoforms through the process of RNA splicing, alternative use of UTRs. What I'd like to discuss is how RNA sequences can also be changed to make different RNA transcripts and different protein isoforms to diversify our transcriptomes and genomes. Proteomes. The idea that perhaps I'd like to introduce is that our RNA sequences is not always identical to the underlying DNA. That might seem like a surprising thing because all along we've been thinking of using the DNA as our genetic blueprint. That is for sure still absolutely correct, but there are spots in the genome where the RNA sequences are not identical to the underlying DNA. Perhaps this is not so shocking if we think of what is already known. The term RNA editing was introduced by Bob Benney's group, and now we know quite a bit about RNA editing in trypanosomes, in plants, and also even in human. We know that there are family of deaminases such as ApoBac that deaminate the C in the RNA into a uracil, for example, in apolipoprotein B. In this case, we have one DNA sequence for apolipoprotein B, and when it's completely translated— transcribed into the transcript where the RNA sequence is identical to the underlying DNA, we get 100-kilodalton form of ApoB as expressed in, say, the liver. But when we have this RNA editing event, it converts a C to a U. To make a 48-kilodalton form of apolipoprotein B that is on the chylomicrons and essential for lipid metabolism in the intestine. Other more common form of RNA editing are mediated by the protein ADAR that deaminates the adenosine into an inosine, then it's then read by the translational machinery as a guanosine. So I'll refer to these as A to G editing. The poster child of A to G editing is the glutamate receptor. It is absolutely necessary for this editing to happen to change the glutamine to an arginine, and this allows the glutamate receptors to prevent calcium influx and basically help neurons to guard against excitotoxicity. So besides glutamate receptor, other type of RNA editing events have been found on, say, other neurotransmitters such as serotonin receptor. Receptor, and so on. So what I'd like to do today is perhaps broaden our thoughts about RNA-DNA sequence differences to beyond the known RNA editing categories, and I'll refer to these as RNA-DNA sequence differences, or RDDs. I'll tell you about what we have learned so far about the characteristics of these RDDs, how our beginning understanding of how they are formed and what are their functional of statistical significance. So in some way, our question is fairly simple, right? What we want to know is to compare RNA sequences to the underlying DNA sequences. And of course, with the improvement in sequencing technology, one can do this by sequencing very deeply both the DNA and RNA from the same cells and be able to compare them. So you can say, well, the question is pretty simple. You compare the DNA sequences to the underlying RNA sequences. But like most things, even though the question might be quite simple, the execution can be quite complicated. Our genome and transcriptomes are very large, and yet each sequence read is very short. So in order for us to be able to compare the DNA and RNA sequences, we have to be able to put all the sequence reads together well in order for comparisons. But like any puzzle, if we can put together the puzzle, usually at the end of the day, we know whether we can put together the puzzle well enough for comparison or not. And I think that's also the case of assembling sequences for comparison between DNA seq and RNA sequencing. Here are some of the inclusion parameters that we've been using through the years and improving in order to get to this perfect assembly of DNA sequencing and RNA sequences. So for sure we need to be— excuse me— to be sure of the DNA sequences and underlying RNA sequences. We have to make sure that each base is covered by multiple reads. We have to be sure of the underlying DNA sequences. For the discussion here, I'm just going to be very conservative and only look at monomorphic sites in the human genome. We're going to exclude regions that are difficult to map, regions that are repetitive regions, exon intron borders, and so on. And we also want to be sure that we know that the RNA base really is different from the underlying DNA by making sure that we see it multiple times. But even so, one of perhaps the easiest, the best way for us to be sure that our assembly, that our putting together this sequencing puzzle is correct, is by looking at the known sites. We already know that ADR mediates a set of RNA editing sites, so we can manipulate those by knocking down ADR, overexpressing ADR, pulling down sequences to know that there are sites that are A to G editing sites that are known. For example, as in this example in the ATM gene, in this individual, we know that person has an A at that site, being covered with 42 reads, and then in that person's RNA, there are 5 reads that are— there are G-bearing form of this ATM gene that is not encoded by the DNA. So after we have kind of sanity check and improve our sensitivity and specificity, of this type of assembly, we went on to look at other type of RDDs. So shown here is an example where the individuals have the G in the DNA, and then the corresponding RNA for this MOB1 gene, there are As that's not encoded by the gene, and in the corresponding WNK1 gene here, this individual has the T in the DNA, and in the RNA, there's also a C-bearing form of WNK1 that's not encoded by the underlying DNA. We don't have to find these by only high-throughput sequencing. We can find also these RNA bases that differ from the underlying DNA by traditional Sanger sequencing as shown here. And perhaps another way to think about this is if we sequence the DNA, sequence the RNA, and see that there are RNA sequences that differ from the underlying DNA, do we see these transcripts translated into proteins? So we take the same cells and do mass spectrometry and ask whether we see these RNA bases translated into proteins. So in this case, in RPL32, we have a G in the DNA, some of the transcripts are made, have a T form of the RPL32, converts an alanine to a serine. And by mass spectrometry, we found both the alanine as well as the serine-bearing form of RPL32. So here are some additional examples of peptides where we found sequences, the peptide sequences corresponding to the RNA sequences at RDD site rather than the underlying DNA. Besides, since our discovery, other groups have also found that there are RNA sequences that differ from the underlying DNA, and I've shown you a list of the recent publications here. So overall, what we have found was that there are all 12 types of RNA-DNA sequence differences. A to G mediated by ADAR being the most common, but we also see the other 11 types as well. What we— the frequency, of course, depends a bit on inclusion threshold, but in general, it's about 1 per 10,000. The patterns are extremely consistent, so if I have a T to A RDD at one site, in the other, the next person will also have a T to A RDD at one site., it will not be to other bases. So far we have found this in all different cell types. I mentioned that one of the reasons I've been spending some time at the NIH is to participate in a natural study of a rare form of ALS. I will tell you more about this ALS study, but now I'd like to show you some data that we have generated using the unaffected individuals from this family, and some of the patients are actually here in the audience today. So using the unaffected individual cells from the individuals, both from the patients and the controls, what we've— they've been generously allowing us to get blood samples and also skin biopsy samples. So from the blood samples, we have made cultured B cells. Of course, we have white blood cells, and from the skin biopsy, we've made them into skin fibroblasts. From the skin fibroblasts, we've been making pluripotent stem cells and driving them into the ectodermal lineage into motor neurons and endodermal lineage into liver cells. So now for each individual, we have a large set, collection of different cell types. For each individual, we carry out genomic DNA sequencing, and for each of the cell types, we also carry out deep RNA sequencing. This allows us to look at some simple questions such as the expression level of genes. Shown here is RB-associated kinase. On the Y-axis, all normalized the same way, so the height of the peak gives you the sense of the expression level of RBAK in these different cell types. And we see here that in, say, the stem cell, say in the fibroblast, the expression level of RBAK is pretty low as seen in the lower size peak, and then in the white blood cells or the leukocytes, the expression level of RBAK is higher. Now we can zoom in into the sequences of RBAK, and what we found was that there are two adjacent A to G editing sites and a nearby T to C RDD. And if we look at the level of these RDD or editing in RBAK, let's say, let's look at the fibroblasts, about 33% of the RBK genes carry the RNA sequences of GGC that's not encoded by the DNA. Whereas in the leukocytes, only about 5%. So even though the expression level of the gene is high in white blood cells, the RDD level is actually relatively low, showing us that RDD adds a whole layer of complexity to gene expression and also suggests that they are regulated differently. You can say, well, this is only samples from one individual. Do we really see this more generally in other individuals? And as we kind of expand this study, we have now quite a few samples from different individuals, and we consistently see this T2C RDD in this RB-associated kinase to be lower in white blood cells than in fibroblasts. So we know that RDDs are found in all human cell types that we have looked at. What about in other organisms? So we decided to go from human all the way down to single-cell organism to yeast. Prior to our study, the only thing that is known about RNA editing in yeast is that there are tRNA editing in yeast. Deep sequencing of DNA and RNA of different wild-type strains of yeast and indeed found again all 12 types of RDDs. Shown here is the reference strain S288C, the distribution of different RDD types, and we see very similar RDD distributions and frequency in all the different wild-type strains of yeast. So now that I've told you a little bit about the— we found RDDs in all human cell types and we know that it's conserved from yeast to human, what I'd like to do is to take a step back and think about how these RDDs come about. When are they made? So we asked two simple questions. Do they occur while like RNA are being synthesized?, and if not, when during the co-transcriptional processes does RDD arise? So to answer this question, we've been collaborating with John Liss' group at Cornell. Professor Liss have invented a method called GroSeq and a similar method called ProSeq that are nuclear run-on assays coupled with high-throughput sequencing. This method allow us to pull out newly made RNA and also know the position of the RNA RNA polymerase on these newly made RNAs. So we can look at these RNAs, know where the RDDs are, and the position relative to the polymerase. So this allows us to start asking when during transcription do RDDs arise. So in the active RNA polymerase bubble, the newly made RNA actually have identical sequences as the underlying DNA. There are no RDDs. Polymerase. So the newly made transcript then begins to exit the polymerase bubble. About 50 to 100 nucleotides outside of the polymerase bubble, we see a sudden emergence of RDDs. So shown here are some raw data from both ProSeq and GrowSeq. Shown on the bubble here is the polymerase. So you can see near the polymerase there is actually no RDD, and then there is a sudden increase of RDD about 50 to 100 nucleotides outside of the bubble. So if we think of this kind of the pathway of an RNA processing, we have the RNA polymerase as the factory where RNAs are made. At that point when the RNA emerge out of the polymerase factory, the RNA sequences are identical to the underlying DNA. But as soon as it comes out, it goes into some processing step where some of the RNA sequences are changed from the underlying DNA where RDD formation occurs. These transcripts then go on to be spliced, UTRs, and so on. At the final product, what we have is one DNA sequence and multiple different forms of RNA transcripts. So what we really want to know is what happened in this processing step that allows Rd deformation to occur. So when we think about what happened right outside of the RNA polymerase bubble, we thought that perhaps there is the R-loop. This is an RNA-DNA hybrid that as the DNA is still slightly unwound and the nascent RNA strand is still close by. Forming this three-stranded structure right outside the bubble. So our question is whether these R-loops could be facilitating RDD formation. Another advantage of spending some time at the NIH recently is that there's a lot of expertise on R-loop biology on campus. So we certainly have been benefiting from all the expertise on campus about this. But in order for us to really ask whether R-loops really play a role in RDD formation, we somehow have to kind of tickle the R-loops. And that's not so easy to do since they are all over the genome following at least the polymerases. But we can take advantage of how cell deals with these R-loops. The cells somehow have to kind of resolve these R-loops in order to kind of facilitate transcription., and the cell does this using different enzymatic steps, including topoisomerase that relaxes the DNA, the RNase H that was identified by Dr. Crouch's group here at the NIH, the RNase H actually digests the RNA within the R-loops, and then there is an RNA helicase called sanataxin that unwinds these structures. So if we can independently kind of play with these R-loops using these different enzymes, perhaps we can get a better understanding of how RDDs are formed. And that's exactly what we did. So we, in yeast, we can look for mutants that are deficient in each of these ways of resolving R-loops. For in each of these mutants, because they have a deficiency in resolving R-loops, we expect to see more R-loops. More R-loops. That's exactly what we see here in this dot plot. It's a dot plot using the antibody S9.6 that's developed by Dr. Leppler's group here at the NIH. We see that in each of these yeast mutants, we have more R-loops relative to wild-type control. And correspondingly, what we also see as in these yeast mutants is that there's more RDDs. So we see more R-loops, more RDDs, suggesting that perhaps these two processes are indeed coupled. So this is in yeast. What happened in human? Before I tell you this, I'd like to take kind of a little introduction to you of this rare form of ALS that we've been studying. This is a form of ALS with juvenile onset. That I will refer to either juvenile ALS or it's also known as ALS-4. It's an autosomal dominant form of amyotrophic lateral sclerosis. It is due to mutation in this gene, senataxin. That's the helicase that I just referred to earlier that resolves RNA-DNA hybrids. The gene discovery was made by David Kornblath with Phil Chance and Kurt Fischbach group in the early 2000s. What distinguishes this form of ALS from classic ALS is that the onset is fairly early, usually in the mid-teens, and is a very indolent disease. It is characterized with progressive muscle wasting, with weakness, and leading to difficulty walking. Another distinguished feature about this form of ALS is the patients do not have any bulbar involvement. So these patients have a mutation in this RNA helicase, in yeast we see that it plays a role in R-loop and perhaps in RDD formation, so what happens in human? One of the reasons that we study this disease here at the NIH is that there's a very large Maryland family with this form of ALS that led to the original identification of the gene. And here's only part of the pedigree, and as I mentioned earlier, that some of the patients are actually here today. So with this family, we've been bringing them into the clinical center, collecting various cells. We've been doing very deep molecular studies of the cells both at the sequencing level. We've been looking at molecular biology such as the R-loops, RDD formations, mitochondrial functions, and so on. So the question right now here relevant to this talk is that what is the effect of the sanataxin to R-loops? We know that it's in amino acid 389. And it's where the protein interaction domain is. Does this mutation actually affect R-loop formation in the patients? And the answer is yes. Again, using the S9.6 antibody and other antibody against R-loop, here's data on the spinal cords from autopsy samples from a patient. What we see here is that in the control samples, there's very clear strong R-loop stain in the motor neuron of the control samples, whereas in the patient, in the motor neuron, there's a clear deficiency of R-loop. Here, just control stains of the nuclei of the motor neurons. So we see that this mutation leads to a decrease of R-loop in the motor neurons of the patients, and if you actually zoom into this slide, you also see that the supporting cells in oligodendrocytes, there's also a decreased R-loop in the patients as well. Now we're going to zoom in. This is just kind of overall in the nuclei. What happened at the gene themselves? We can look at R-loop across all the genome. We do an S9.6 antibody pull-down of the DNA/RNA and sequence the nucleic acid. Rather than show you the whole genome-wide data, I'm going to just focus on kind of Another poster child that people often show for R-loop biology, beta-actin. The reason that people like to use beta-actin to look at R-loop is because in the gene body, there are quite a few R-loops, whereas in the 5' end of this gene, there are very few, so there's an internal control. So again, here it's shown in blue in the bar is where in normal individuals, we see R-loops along the gene body and end of the gene of the beta-actin gene. Whereas in the patients, the ALS4 patient or juvenile ALS patient, we see a significant decrease of R-loop in the beta-actin gene. Beta-actin gene indeed is a perfect gene for this type of study because there's an RDD, a G to A RDD nestled in between of all these R-loops. So we want to know if there's a decrease in R-loops in the patient, do we also see any effect on the RDD? And the answer is yes for this G2A RDD. The patients have a much lower RDD level compared to patients— to controls. So this is of one site in the beta-actin gene, what happened genome-wide. Overall, what we also see is that the frequency of RDD in the patient is lower relative to controls. So what we know so far is that we think that the RNA, when they are made by the polymerase, they do not have any RDDs. RDDs are made soon after the transcript leaves the RNA polymerase bubble, and this RDD formation is coupled to R-loop formation. And we're now working on to find the exact proteins and mechanism that mediates the RDD formations and how that's kind of the R-loops, we believe that forms the shape to facilitate that mechanism. I'd like to spend like the last half of my talk thinking about what is the significance of these RDDs in cells. So at the molecular level, what we know so far is that we have one DNA sequence that made into different RNA transcripts that differ by one or a few bases because of the RDDs, and then these are translated into proteins that have slightly different protein isoforms. But that's really at the molecular level. Do these proteins with slightly different amino acid sequences really have different functions? Really is the key questions here, right? So I'd like to begin by answering this with an example in yeast with the gene Tup1. So the yeast Tup1 is a very well-characterized transcriptional repressor. Tup1 interacts with the protein CYC8 as a complex to recruit other proteins to then affect the expression levels of many genes. Proteins in the yeast transcriptome. It mostly acts as a repressor and is a highly conserved protein. In the DNA and RNA sequencing of yeast, we found that there's a C to U RDD in the TOP1 gene. It leads to an alanine, a change in alanine in amino acid 459 to a valine. And by mass spectrometry of the yeast, we found both the alanine form as well as the valine form of TUP1. So our question now is, do the alanine form and valine form of TUP1 have different regulatory activities? So the first question we ask is, do these two forms, alanine and valine form of TUP1, have different stability? And by computational prediction is that the valine form of TUP1 is a bit more stable., and with kind of a slightly higher delta G. By cyclohexamide study experimentally, we also showed that the valine form of TUP1 is indeed more stable with a half-life of about 30 minutes longer. This amino acid 459 is in the WD40 domain of TUP1. This is the region of TUP1 that interacts with other protein to influence the expression level of genes. So the crystal structure of the WD40 domain was solved by Cynthia Wolberger's lab, so we know quite a bit about the structure. The question is whether this alanine and valine differences leads to different regulatory activity. So first we use RNA sequencing to identify the target genes of TUP1, and we identified 292 target genes, genes that are regulated by TUP1, and what's nice is that this is a huge amount of overlap with the target genes that were identified by previous microarray studies as target genes of TUP1. Most of the genes, as shown here in the heat maps, are repressed by TUP1, but some genes are indeed activated by TUP1. The question is, if you have a cell that only expressed the alanine form of TUP1 and only expressed the valine form of TUP1, Tup1, whether there are differences in the target genes. So again, we expressed Tup1 in the Tup1 null mutant, we expressed the alanine form of Tup1 and the valine form of Tup1 and we carried out RNA sequencing to look at expression level of the target genes. Of the 292 target genes, 51 of them have differences in expression level based on whether this yeast is expressing the valine or alanine form. In the valine form of the yeast, yeast expressing the valine form of Tup1, 42 genes have higher expression level and 9 genes have lower expression level. All 42 of the genes that have higher expression level are normally repressed by Tup1, and of the 9 genes that have lower expression level, 6 of them are induced by TUP1. So what we show here is that the valine form of TUP1 have a lesser regulatory activity on target genes. So what we know is that that one amino acid differences confer a more stable protein that have lesser regulatory activity. The seeming paradoxical relationship between This protein stability and activity was first introduced in the mid-1990s by Brian Matthews' group for T4 lysozyme and now from several, many other proteins have shown this paradoxical relationship. And here we're showing that RDD also kind of changes the stability as well as regulatory activity. We believe that the RDD is allowing a protein isoform forms that provides a more sustained regulation on the target genes. And this actually has phenotypic effects because the valine form of yeast expressing only the valine form of Tup1 are more sensitive to stress and shown you here is also more sensitive to the antibiotic hygromycin B. So this is in yeast. What I'd like to do is to end by thinking about what does it really mean for the patients, for them to have, in our ALS4 patients, to have fewer R-loops and lower RDD frequency. What have we learned that kind of perhaps allow us to better understand the disease itself? So I already showed you the slide before of the lower R-loop frequency and lower RDD frequency in patients. We also know that synataxin is a gene that is expressed in all cells. And not just motor neurons. So should we begin to think about this as a more broader disease rather than just a pure motor neuron disease? And being at the Clinical Center here, we really can ask this question very deeply. Besides studying the— doing deep molecular phenotyping, we have the possibility of studying the patients clinically at a very deep level and also kind of follow the natural history of the disease. So when the patients, whether it's affected or unaffected individual from this family, comes to the NIH, they get very comprehensive laboratory testing, they get different type of radiological imaging studies, we do neuromuscular testing, EMGs, nerve conduction studies, and we, as I mentioned, we also collect collect different cells and study them at the sequence level. So rather than tell you about kind of the grand synopsis, I'd like to focus on one patient and kind of give you an idea of what we're beginning to take the study— our findings from the lab to help to guide us in management of the patients. So JL is a 35-year-old male, also from this Maryland family with ALS. He developed symptoms of juvenile ALS when he was approximately age 13. When we met him during the review of system, what's kind of a bit unusual was he reported shortness of breath with mild exertions, and that's really disproportional to his weakness. And at the beginning I mentioned that this form of ALS have no bulbar involvement, so this is a bit unusual. Unusual. And he also reported a sense of fullness after eating. On physical exam, his neurological exam was very consistent with someone with a motor neuron disease. He has distal weakness, he had hyperreflexia, and he has an extensor Babinski sign. On laboratory findings, he had an iron deficiency anemia. On his CBC, his hemoglobin was 9. With an iron level of 14. And even though we're neurologists, kind of, we had, by that time, we already starting to have some data from our RDD studies, and part of those RDD studies kind of whispered to us that we really need to look at this iron deficiency and some of these, the shortness of breath and the fullness that the patient was experiencing more carefully. So we proceeded to do a colonoscopy on the— suggested a colonoscopy on the patient which revealed a greater than 4-centimeter mass at the hepatic flexure. The patient then underwent surgical resection of that mass that showed out to be adenocarcinoma with microsatellite instability. So what was the data in the lab that was whispering to us us that to look to go ahead and suggest a colonoscopy. I mentioned that in the patients we see lower RDD level genome-wide, but a set of genes that have particularly low RDD levels are genes that play a role in growth. Here I show you the RDD level of the patients relative to control for a set of these genes that participate in growth. They all have lower RDD levels. In the patients, the transcript that have the RDD-bearing form of these transcripts are lower abundance in the patient relative to the controls. And when we look at these genes, they are not just the ones that participate in cell growth, but many of them have been associated with adenocarcinoma, particularly in colorectal adenocarcinoma. So from this result and the clinical finding kind of prompted us to to suggest a colonoscopy. So what we think that the connection between RDD and the disease is right now is that we believe that in the juvenile ALS for patients, perhaps the RDD, lower RDD frequency is leading to a deficiency in some transcript in their cells, and some of these transcripts play a role either in necessary for proper motor neuron function and perhaps in other cellular functions such as growth. And this is perhaps not so surprising. Previous study by Professor Kwak's group in Japan have also suggested that the GluR2 editing level is lower in classic ALS patients. So perhaps this is kind of all along the similar spectrum of either lower editing transcript or lower RDD levels leads to some vulnerability in motor neurons. So in summary, what I've told you is that these RNA-DNA sequence differences are found in yeast to humans, so it is conserved. We believe that the formation of RDD occurs soon after the nascent RNA exits the polymerase complex and it is coupled to R-loops. And we believe that RDD, like say splicing and other type of RNA processing steps, plays a role to diversify the transcriptome as well as the proteome and kind of provide us forms of proteins with either different stability or even different activity. And so what I hope that I have shown you today is that by intertwining very deep basics on the study of RNA biology together with careful clinical studies of, in this case, juvenile ALS, that we are able to kind of bring together knowledge that is important for patient care. And I hope that this is the kind of the direction that genetics is allowing us to do today that the precision medicine is really allowing us to more precisely take care of our patients based on not just the mutation but better understanding of the cell, the functions of the mutations, and the underlying genes. So I'd like to thank all the people who have allowed us to do— me today to tell you about this work. In particular, my junior colleague, Isabel Wang, in my group. And Christopher Grunzike in NINDS, who the two of them together really have been responsible for most of the work that I'm showing you today. So thank you very much. An amazing talk. Do the ALS4 patients have TDP-43 abnormalities? Do they lose it in the nuclei? Yes. So of course, as a sequence, there's no mutations in TDP-43, but the expression level of TDP-43 are different in the ALS4 patients. That fits very nicely with the story you're putting together. You spoke of lowered RDDs in adenocarcinoma of the colon and rectum with microsatellite instability. Can that be generalized to any other types of adenocarcinomas and perhaps be used as a marker or maybe even a screening tool? Right. So, yes. Our general thinking, of course, this is our first year of studying the patients. Our general thinking is that perhaps there is a slight increased risk of overgrowth and perhaps adenocarcinoma. It's maybe too early days for us to comment on that right now, but we are interested into looking into cancer cells and see whether that deficiency in certain transcripts mediated by RDD is leading to cell growth in particular into that type of cancers. Okay, thank you. Great, thank you very much.