Scientist Stories: Silvi Rouskin, Drugging RNA Structure

Hello everyone, welcome to the BUC Sciences webinar session. In case you don't know about us yet, I would like to briefly introduce our product line related to today's topic. BOC Sciences, located in Long Island, New York, is a professional chemical provider and currently the largest supplier of pseudouridine in the world, with the most production capacity of pseudouridine and its derivatives. Our annual output of pseudouridine is currently over 500 kilograms. In the past year, our product N1-methylpseudouridine has been used as a raw material in the production of mRNA vaccine against COVID-19. We are very proud to support RNA-related research and development with our products. In today's webinar session, we have Dr. Silva Ruskin present her works related to alternative RNA structures. We will learn about the novel techniques developed by Ruskin Lab, which can distinguish multiple RNA conformations formed by the same underlying sequence in vivo at single nucleotide resolution at a high throughput scale. During the pandemic, the Ruskin Lab applied the technology to probe the structure of the entire SARS-CoV-2 genome in vivo. Our speaker today, Dr. Ruskin, expertise in investigating how RNA structure and function impacts viral infections and human disease for the design of novel therapeutic approaches based on targeting RNA. Dr. Ruskin immigrated from Bulgaria to the United States alone as a teenager to pursue a career in science. At 16, she became a freshman at the Florida Institute of Technology and obtained a bachelor's degree in physics. Dr. Ruskin did her PhD in Jonathan Westman Lab at UCSF, followed, followed by a traditional postdoc postdoctoral fellowship. In 2015, she started her research independently at the Whitehead Institute for Biomedical Research at MIT. Since last year, September 2021, she became an assistant professor at Harvard Medical School in the Department of Microbiology. So that's all for the introduction. Just keep in mind that you will be able to post your questions related to this webinar content during the presentations and simply just post them using the Q&A panel on the bottom of the screen. And after the presentation, Dr. Ruskin will answer selectively of the questions. So yeah, let's welcome Dr. Ruskin. You can begin. We are ready.

Hi, everyone. Thanks for coming. It's hard to see anyone on the Zoom talks, but I'm very happy to be here. So let me share my screen and make sure you can see my presentation. Can you see it now? I assume yes.

Yes, yes, we can see it well.

Oh, excellent. Thank you. OK, great.

Go ahead.

Just checking. Great. So hi, everyone. I'm Sylvie, and thank you for the kind introduction. My lab generally works on RNA structure, and we're very excited about RNA therapeutics, and we just moved recently to Harvard Medical School in the Department of Microbiology, and I'll be telling you about what we've been up to. So I'd like to start with a recent report from the World Health Organization about the deadliest viruses on Earth. Here I'm showing you the top 9, which include viruses like Zika, Ebola, Dengue, HIV, influenza, and SARS. I hope you can appreciate this is a very diverse set of viruses that infect different cell types and cause different disease. But one thing that's common to all of them is that they're all RNA viruses, meaning that they store their genetic material as RNA. And my lab was previously focused on HIV and looking at the shapes it forms in the cell and, and this, their implication for HIV infection. But with the pandemic, I realized that the methods we developed were directly applicable to SARS-CoV-2. COVID-2. And so we did a lot of work on this virus with the ultimate goal of identifying small molecules or antisense oligos that bind to and inhibit the virus and could represent a novel class of therapeutic strategy. And I'm going to further explain how this works. So first, why is RNA so special? Just like DNA, RNA consists of a string of nucleotides that store genetic information. But DNA always comes in pairs of two molecules that are bound together in this very rigid double helix shape. On the other hand, RNA is one molecule, but instead of being linear or rigid, it is much more flexible and often folds back on itself to form very complex shapes that are very important but really hard to predict. So here I'm showing you one example where the RNA molecule starts with this blue sequence, and this we call stem-loop motifs in the 2D, which is on the left. And on the right, I'm showing you the 3D structure. And what I hope you can appreciate from this example is how complex RNA shapes can be in contrast to DNA. And it turns out that these shapes encode a second layer of regulatory information on top of what is encoded by the linear sequence alone. So RNA can both store genetic information like DNA— and here on the left, I'm showing you the genome of influenza virus— but also RNA can fold into complex shapes like proteins. And This gives RNA some very interesting properties, including the ability to do work. Like, the ribosomal RNA can actually catalyze reactions, and this is typically a function performed by proteins. But in, in a sense, this makes RNA the most versatile of all the biomolecules. So Typically, people think of drugs targeting, let's say, a cellular, or in this case, a viral protein to inhibit some important stage in the viral life cycle, such as viral entry or viral replication, viral assembly, and so forth. And so the idea is that one develops some small molecule that can specifically fit in and bind to and inhibit some viral protein, as shown here on the left. But the issue has been that many proteins, both viral and cellular, are not druggable because they don't have like a nice, good, small molecule binding pocket, like the one illustrated here. And so for a long time, it was thought that RNA is also non-druggable, but the view is now being challenged by multiple scientists and companies. They've been successfully docking small molecules to defined RNA structures, such as the one shown here in the right. And this is from Arrakis Therapeutics. So a big breakthrough came a few years ago where pharmaceutical giants like Novartis and later Roche independently developed similar-looking small molecule drugs that can directly bind to RNA for the treatment of spinal muscular atrophy. And this disease is caused by faulty processing of RNA of the SMN2 gene. And so both of these compounds are performing really well in Phase 2 trials. In addition, there is a different RNA targeting strategy based on antisense oligos or ASOs and this popular drug called Spinraza. And there have been now 7 FDA-approved ASOs for the treatment of SMA. So this has been extremely successful and this great example for targeting RNA. To be able to treat diseases. And so, I hope I've convinced you why we care about RNA structure, and really the two main points is that RNA structure is at the heart of all RNA-based regulation, and is because RNA structure offers new therapeutic avenues. And here I'm just showing you the coronavirus, along with the RNA genome. But despite how important RNA structure is, historically RNA structure has been very difficult to determine because it's limited to pure computational prediction or experimental studies that use short sequences that are arbitrarily cut out of their context and then refolded inside a test tube. So what my lab has been doing is to overcome this, a lot of these limitations, by developing new tools to understand the roles of RNA structures in gene expression and disease, as well as how can we leverage these functional structures to design RNA-based drugs. So what's interesting about RNA in contrast to proteins is on a biophysical level, So in case of proteins, the free energy of folding is often described as a funnel, which means that there is one point of minimal free energy that represents one solution that is optimal for how this protein will fold. This is one very well-defined structure. In contrast, the theoretical RNA folding landscape has multiple wells. That the RNA can get trapped into. And these wells can correspond to very different structures that give an RNA with identical sequence very distinct shapes that could have very distinct properties. And so the question that was very important to us— so again, I'm showing you this is mostly based on theoretical work or experimental work of some very special RNAs, like bacterial riboswitches. But we wanted to— No, for RNA in general and RNA sequence, two identical sequences form different structures as theoretically predicted. And when we started working on this, there were a lot of really great chemical probing approaches developed by us as well as many others to assay RNA structure in high-throughput in cells. However, a big limitation of all these approaches is that at the end, you get some structure that represents a population average. And such average will obscure or blurry signal from alternative structures. So just to illustrate the problem with the average, I'm going to give a real-world example and state a fact that in 1985, the average starting salary of a geography major graduate from UNC was $130,000. So if we step back and think about this information, it seems a little bit off because the year is only 1985 and someone has just graduated and suddenly they're earning $130K. I mean, it seems really nice, but seems a little bit extreme. And if we look at the underlying distribution, actually no one was making $130K. And indeed, the majority of graduates were barely making above $10K. But in this case, there was only one outlier who really shifted the average and made it uninformative and misleading. And so I'm using this example to illustrate that by analogy, for a class of RNA molecules with the same sequence, it's possible that none of them are actually forming the structure that is derived from a population average model. And so these kind of models should always be considered with a grain of salt. So our question was, do the same sequences indeed form multiple different structures. And so what we needed was a single-molecule technology, and, and we were interested in the structure of individual molecules instead of a structure of a population average. So we set up to develop that, and our favorite chemical in the lab is called dimethyl sulfate, or DMS. And this is because DMS goes into cells quickly and reacts with open and accessible adenine and cytosine bases. And in short, knowing which bases are open and which are paired allows us to make good models for how the RNA folds in cells. However, to take this to a single molecule level, we first needed to establish a new experimental procedure and then develop an algorithm to deconvolute this data. So just to prepare you for what's coming, I'll first walk you through some technical details because it's important concept for the interpretation of the data. And then I'll show you how we applied these methods to discover new biology and to help us develop therapeutics. So we first developed the ability to probe single molecules. And the way this works is by modifying a single RNA at multiple open bases. And then we detect these modifications or these open bases with a special RT enzyme. And what this RT enzyme does is, as it goes through to make cDNA, every time that it encounters a DMS methylation on the RNA, it incorporates mutation in the cDNA. And so then all we have to do is sequence this single-stranded, this DNA molecules. And we know when we see a DNA molecule, when it has mutations on it, we know that these mutations came from open bases on a parent RNA molecule. And next, we developed an algorithm that uses this chemical probing data to detect RNA folding ensembles with expectation maximization clustering. And I'm going to briefly explain how this works. So imagine you have this RNA sequence, and this sequence is forming two very different alternative structures in this case. There'll be some open bases that are exclusive to one structure or the other, and these open bases will react with DMS, and then they'll get converted to mutations on single molecule reads. And so we end up sequencing these DNA reads that now have co-occurring mutations that are specific for one structure or the other. But what was typically done before this work is to analyze all of these reads together in one bin and get— of course, you get a population average signal, and in this case, it will not correspond to either of the structures that I'm showing here. But what our algorithm did instead is to separate out the sequencing reads based on their mutation pattern into belonging to distinct structures, or what we call clusters, that have distinct DMS signal that now can be used as constraints to generate accurate models for alternative structures. And for the purpose of this seminar, I'm not going to go into all the experimental and computational detail that went into this work. It's been published. You can go and read it, and please don't hesitate to contact me if you have questions. But the key into this whole work here was having some really nice ground truth controls that we could use to test and validate and further develop our both experimental and computational methods. They performed extremely well. Using these tools, we discovered that actually the HIV RNA This is a short— so RNA viruses are typically very short. In case of HIV, it's only 10,000 nucleotides. This is the whole genome. But what's cool about it is that this sequence of 10,000 nucleotides can actually fold in many different ways. And in some ways, kind of what I'm showing you here, I'm showing you a small fraction of it on the left. It folds in a way such as it exposes a given splice site, so now the splice site you see is open and accessible and can be used by cellular host protein factors, can be recognized, and splicing will happen in this HIV molecule that's forming this kind of structure. But the same sequence can fold into alternative structure that now occludes this splice site because it's paired up with something else, some sequence upstream, and now cellular factors can no longer recognize and bind to it, and so that would inhibit splicing. And so our work showed that HIV RNA has this intrinsic ability to fold into multiple structures. Therefore, despite having the same sequence, different molecules can occlude different splice sites by forming RNA structures, and therefore direct which gene is expressed from a given RNA molecule. And in addition, can you imagine an alternative structure that occludes all splice sites,, and that can guarantee that a fraction of molecules will remain full length and will serve as the HIV genome that will get passed on to the next cell. So therefore, there's this role of intrinsic RNA thermodynamics in setting the level of HIV gene expression. And again, this work is published. I encourage you to check it out and ask any questions. But I want to just show you how these new tools enable discovery. And I wanted to go into a little more detail about our work on SARS-CoV-2. And the exciting part of this is because we were so well set up, we actually managed to determine the full structure of the SARS-CoV-2 within a month of starting to work on it. Within a few months, discover new biology such as insights in the mechanism of frameshifting. Within another few months, get this amazing collaborators and start developing therapeutics such as the antisense oligos. I'm going to touch upon this topic. Just as an introduction, the SARS-CoV-2 is another RNA virus is actually one of the largest RNA viruses. It's 3 times larger than HIV. You can see it has about 30,000 nucleotides genome, but it's still very small in terms of genome size. I'm going to describe how we set up to do this. Our initial setup was as follows. We used both Vero cells, which was initially the only model for replicating SARS-CoV-2 in cell culture. Later, we repeated the same experiments in human-derived HEK7 cells. The way it works is basically we get these cells and we infect with virus, and then we ask what the RNA structure looks like. We DMS modify, so we have a snapshot of what the RNA structure looks like of this virus inside the cells. And of course, then we have to do quite a bit of computation and data analysis to figure out how the RNA is folded. So we're actually the first lab to publish a genome-wide population average structure of SARS-2 in infected cells. And this was back in May 2020 because coronaviruses were really understudied and no one had ever done any experiments to see how this, this virus is really folded inside cells. And of course, like every scientific, most scientific endeavors, it took quite a bit longer for this to get finally into press, but it's now, I'm happy to say, in press in Nature Communications. And here I'm just showing you our model over the first 5,000 nucleotides. And again, this is with a grain of salt that here we're assuming that every molecule with the same sequence is folding into the same structure. And, you know, I'm not expecting you to get much out of this besides, okay, there is some very interesting— it's not a linear molecule, you can immediately appreciate that. There's a lot of stem loops, there's a lot of bulges, there's a lot of potentially interesting places that one can develop, one can imagine designing some specific molecule binding to it. And what was important to us, of course, was that, you know, our methods were recapitulating previous work, which was all focused on the SARS coronavirus 5' UTR, because it's extremely conserved in many coronaviruses. And even though it was mostly done in vitro, so almost entirely done in vitro, we agreed really well with the work that was done previously for this particular region of the first about 500 nucleotides of the virus. And then we went ahead to ask the same question that we'd asked for HIV. Does this sequence— we have this nice population average model. How good is this model is it really just one structure or are there multiple structures? And again, I'm going to spare you here all the technical details, but just to show you here, I've plotted the genome sequence, the region, so from 0 nucleotides to 30,000 nucleotides. And in gray, I have highlighted the regions where our methods are saying that this sequence this particular sequence is coming from, it must be coming from at least two alternative structures or more. And so you can see it's actually kind of curiously the 5', the very beginning of this virus was really forming one stable structure. There was indeed one nice solution. We could not see any alternatives in the structures of the 5' UTR. But as soon as you kind of got out of it, there was a lot, there were a lot of sequences, really most, more than half of the genome that were clearly in multiple forms. One sequence that immediately caught our attention at the time was the sequence in the middle of the genome called the frameshift stimulating element, or FSE, that I'm highlighting here in red. We set out to explore this sequence more. And so just to show you the organization of SARS-CoV-2, 30,000 nucleotide genome, FSC is kind of towards the middle. And this sequence is really critical for the virus in order to produce the complete set of proteins that the virus requires. So basically there is one open reading frame, an ORF, and it's called ORF1a or ORF1b. That is kind of defined by the location of this frame shifting element. And here again, I'm showing you the sequence. And what happens is typically the ribosome goes along on a certain open reading frame, and it's reading and it's making this protein, but then it reaches a stop codon, UAA in this case. And when it reaches a stop codon within this sequence, obviously it lets go. The protein, it lets go and it stops translating, so it never makes ORF1b product. But every once in a while, if you have a frameshifting sequence like you do in the coronaviruses, the ribosomes goes, goes, goes, and then there's this slippery site, this UUU, where the ribosomes just tracks, goes backwards, in this case by just one nucleotide, but that's sufficient, of course, to entirely change the reading frame. And now instead of a stop codon, it reads an amino acid, in this case it's valine, and it continues to make this polyprotein and make what's encoded in ORF1b, which is absolutely essential because it includes the viral RNA polymerase. It's the only way that this virus can replicate. You can appreciate how this event is absolutely critical for the virus. It is, okay, so it is thought that maintaining the correct frameshifting rate is actually really important for viral viability. So you don't, you, there's some optimal ratio of this ORF1a and ORF1ab product such that is optimal for viral fitness. And if we were to interfere with this ratio, then this then we can cripple the virus. But what is setting this optimal rate of frameshifting? What causes the ribosome to suddenly decide to backtrack? The model in the field, the most prevailing model based on studying other coronaviruses, it has been that frameshifting is due to kinetic pausing when the ribosome encounters an RNA structure that cannot be unfolded as easily. What happens is it shifts back to relieve mechanical tension. Again, here I'm going to spare a lot of details for the purpose of this talk. But based on all the previous literature had really been done with these short sequences of coronaviruses elements, either whose structure was determined either through computational prediction or looking in vitro. It was thought that this 90-nucleotides sequence of the frame-shifting element is forming a pseudoknot, the structure called pseudoknot. Indeed, we recapitulated all this previous literature and can see that the pseudoknot that forms in this context of this 90 nucleotides gives you a certain frame-shifting rate, about 20%. Initially, that was the rate that was thought that SARS-CoV-2 and other coronaviruses frameshift at 20%. But with our work and then with a lot of newer studies now that were published everywhere in terms of measuring the actual translation rate of SARS-CoV-2 and other coronaviruses in cells, it turns out that the frameshifting event happens a lot more frequently than was initially hypothesized. So 50 to 70% has been depending on the timeframe of infection. And we nicely showed that this is dependent on formation of some really interesting structures that we have never seen before, and actually are very rare in the RNA structure field. They're very long-distance interactions at about 1.2 KB apart. So just to give you an idea, basically, I don't know how well you can see this, but if this was the previous model, was that we have some little like stem loop kind of like local structure forming here and interfering with the ribosome. But what we found is actually there's some sequence that is way far away, and it's coming back and interacting and actually causing a lot more trouble for the ribosome than what was previously thought. So that was one exciting discovery. And again, I'm not going through all the details, but just I hope that I had convinced you that new technology drives discovery. And I told you about the SARS-CoV-2 frameshifting. That paper is coming out hopefully very soon., and also about alternative structures regulating gene expression. And similarly, we have, uh, some very exciting unpublished data for, um, the role of alternative splicing in neurodegenerative disease, including MAPT isoform balance that I won't have a chance to talk about today, but, but please feel free to reach out and chat if you're interested. So, this is great. We see a lot of structures that are interesting, they're alternative, they're doing something for the cell. So, next we thought, okay, if we have a stem loop structure, and, you know, this is in very close collaboration with Anders Snarr lab, who is at Berkeley. And the idea is very simple that, you know, if you have a stem loop structure, you know, or any structure really that is doing something important for the virus in this case. Let's say it's binding some protein and helping translation or transcription or any of this. Then, you know, you can disrupt this structure by adding some antisense oligo. So the idea is the antisense oligo will initially bind, kind of use, land into some open sequence. But then reach out further and unfold some of this stem loop. And now you kind of get rid of this and you unfold this critical structure, and then you can cripple the virus because it no longer is forming the stem loop that's required to bind essential protein factors. So can this work in principle? So here, We're looking at whether an ASO disrupted the structure of SARS-CoV-2 RNA. So on the x-axis is the genome position of the RNA, and on the y-axis is the structure similarity of every nucleotide position. This is between two samples. So if the structure at any particular nucleotide between these two samples is the same, then you have a structure similarity of 1. And here in this graph, we're looking at how similar is the structure between two samples that are biological replicates inside the cell, these samples that were not treated with ASO. And so you see there, the structure is the same basically. Now, if we repeat here, if we repeat the experiment, but now we're looking at we add the ASO2 only to one sample, and we compare it to the other sample, we see this very large dip in similarity where the ASO binds. And now on the right is the one that— and as you can see, basically, this is telling us one, where the ASO is binding and how the RNA structure is changing upon the ASO binding. And the questions that we were really interested in here is to answer is how many viral molecules are actually bound by a given ASO in cells? And how specific is this ASO to the virus? So is it binding just the target, or is it binding the target and a bunch of other cellular RNAs and so forth? And what exactly is the ASO doing? Is it just binding? Is it just binding and perturbing the local structure, or is it binding and perturbing some global structure of the virus? And those are the kind of questions that we can now start to answer in a really high-throughput way. And again, I'm not going to show you the 100 ASOs that we tested. I'm just going to show you the one that we ultimately set with and tested in mouse models. But here, first looking at some data, looking at similar experiments but in a different way. So what I'm— so here, basically, we're looking at how similar is the structure of every nucleotide between two samples. And the two samples now, one is on the X-axis, the other sample is on the Y-axis. And again, in this case, the samples are not treated with ASO. And each nucleotide here is— each dot here is a nucleotide. And here are the samples of biological replicates. Again, you see the exact same result, that the two structures are the same. And I specifically color-coded the bases that the ASO is supposed to bind to. Those bases are in pink, and here are some other nucleotides that also you'll see change there in orange. But what we did here is basically first we treated this RNA with antisense oligos, but a small, small amount of antisense oligos. So just for every 10 target RNAs, viral target RNAs, there's only 1 ASO. And you see in those conditions, we actually don't see any structure change. The plot is entirely the same as with replicates. But as we started to add, adding more ASOs, so here is a ratio of 1 to 2, basically. So half of ASO to number of RNA molecules you can see that now this base is exactly the bases that we're expecting are changing, and those three, because those three are not folding, I'm going to show you in a little bit. And then as we add more and more ASO, so this is in a 0.7 ratio of ASO to RNA, you see even greater change. And if we have, you know, 1:1 ratio, this is so 1:1 ratio of ASO to target, you see even greater change. And if you have an excess of ASO to RNA, so presumably under those conditions, every single site is occupied by ASO, you see this huge change in the nucleotides that are binding it, as well as those 3 and some other ones. But based on— I just wanted to show you this really how reproducible this data is. Both in vitro and in cells. And now this is showing adding those ASO in cells and saying we can really tell how much of— this is again showing just one ASO, ASO number 26. Well, we could tell that this ASO is really binding indeed in cells, and we can kind of estimate based on comparing with this in vitro data, we can estimate how many more viral molecules around. To the ASO in cells. So based on this, and again, we could see out of, say, hundreds of ASOs which one is binding the best to its target in cells. And then we have with our collaborators a lot of functional data to see how is that affecting viral replication. And specifically using this data on the DMS reactivities, we can predict how the structure was changing. So this is the 5' UTR of SARS-CoV-2. Just to tell you the terminology, SL1, SL stands for stem loop. So in the RNA structure field, we are very creative. We call all the stem loops stem loop 1, stem loop 2, stem loop 3. You'll see a lot of that if you read the literature for any RNA, so the first stem loop gets number 1. And here I've just color-coded the DMS reactivity per nucleotide, and so again, this is the control, this is the gold standard structure, we know that this is all forming, but when we add the ASO, what happens is, here is now the ASO, we can see this very nicely that the ASO is binding. And those bases become entirely unreactive now in blue by our DMS signal, and the bases next to it become extremely reactive because the stem loop 1 is now being unfolded. And the nice thing about the story is the reason why I wanted to show you this particular example is because our collaborators were able to take it to mouse models, and we have this beautiful mouse data showing how the ASO works in an animal. So in this experiment, we infect the mouse with SARS-CoV-2 and give the mouse this ASO therapeutic through inhalation once per day. And then we harvest the lungs and we measure the viral titer and the presence of M protein and spike protein. And as you can see, there is a very big decrease in all those measurements, including viral load, presence of N, presence of spike. We also stained the lung tissues. Here you're seeing in no treatment, the lungs are chock-full of this N protein. And stained with it. But if you give the therapeutic, then the lungs are nice and clear. And so it's an example of how an ASO targeting and destroying an RNA structure can be used as a therapeutic against SARS-CoV-2 to clear viral infection. And just to zoom out more generally, Because RNA structure is at the heart of RNA-based regulation, and we have developed these new tools, we really want to understand the widespread roles of RNA structure in gene expression and disease, as well as how we can leverage these functional structures for RNA-based therapeutics. And this is the lab focuses and directions, and a very big focus is on gene expression of RNA viruses, that I told you about. A different wing is the RNA-based therapeutics, of course, both with respect to RNA viruses but also with respect to alternative splicing that happen— mistakes in which happen actually in many neurodegenerative diseases. And this is another big focus of the lab. And finally, In terms of— we've become really excited in this intersection of the immune system, even cancer immunotherapy, and activating the immune system through RNA structure as well. So many different lab directions that all have to do with being able to predict and, you know, really determine the RNA structure and target it for functional consequence and therapeutic consequences. Consequence. So with this, I'd like to really thank the people in my lab who've been doing this work, especially Tammy Lan, Matthew Allen, and Jia Wu, who were very heavily involved in the SARS-2 project that's coming out now. Previously, people— these are the people that did all the work on our HIV project, which is Phil Temeschko-Perlmutter Gupta and Harish Srinathan. And specifically, I'd like to thank our fantastic collaborators. I highlighted Anders Snarr, who was critical for, of course, the SARS-CoV-2 work, as well as the antisense oligotherapeutics. We've been very lucky to get multiple fundings, without which this would have been impossible. And yeah, thank you for your attention, and please let me know what kind of questions I can answer.

Thank you very much, Dr. Ruski. I believe all the attendees enjoyed this presentation as much as I did. It's very illustrative and the message is very clearly delivered. So we do have some questions from the audience. Like before that, some audiences mentioned that they cannot see the primary screen. It might be because you're the phone is used. So I hope it doesn't apply to a lot of the audiences. And yeah, we can start with the questions from Nick. So he said, you mentioned single-cell work at the beginning. Does the RNA assume one conformation in single cells, or do you deal with a population of different conformations? And if so, how do you deal with this?

Yeah, this is a great question, and I'm going to speak towards what we've seen and what my current theory is. So take it with a grain of salt, but basically from our results, it doesn't— it appears that— and I think that makes more sense— it's not a cell-to-cell difference. 'Cause each cell, depending on the RNA, if it's a viral RNA, the cell will produce thousands or hundreds of thousands of copies of itself, of the virus, right? And most cellular RNAs are expressed at 100 to 1,000 copies per cell. And what we think is happening, we have evidence for happening, is that in a single cell, if you look at 100 RNA molecules, they will actually be folded differently within a single cell. So it's less of a cell-cell— and, you know, we can speculate where this comes from, but clearly part of it at least is the intrinsic thermodynamic ability, and part of it was definitely theoretically predicted based on pure thermodynamics. And the reason is because the rules of RNA folding are actually quite simple. Basically, the biggest energetic stability comes from base pairing, and so you just want to optimize the number of base pairs. The base pairs are so simple, it's just A pairing with U and C with G. If you have a sequence of, let's say, 1,000 nucleotides long, there are many different ways that you can fold up this sequence and have equal amount of number of base pairs folding. Theoretically, you can have, let's say, 100 structures. But with our methods, we don't see 100 structures. That could be because our methods— I think though the reason why that is, is because the rules are a little bit more complicated. We don't know quite enough about the rules, but there's still a lot of different structures, and it's maybe like 2 or 10 at most, depending again on the sequence, the RNA, and so forth. But the— just without trying to veer too much away from your question, is that it's less— we care less about what's happening per cell because we do think that the major changes are on the— are already happening within a cell, such that a cell does often represent the population of cells. Now, of course, there will be exceptions to that because, you know, you can imagine how in some cells there is some different conditions. Maybe the cell is stressed, maybe it has a lot of, you know, different proteins expressed, and that will change the structure in this particular cell. So there are exceptions to that, but I'm saying At the level that we're looking at, this is an assumption we're making. Right now, we don't have a way of doing single-cell RNA structure sequencing. And of course, it's something that we've thought about, but it's technically quite challenging.

Okay, we have a few more questions. Dr. Ruskin, you can just like read them, maybe just select 2 or 3 more.

Yeah, sure. Like if— Javier is saying, you showed the probing of rRNAs are probably very abundant. Yes. However, is your technique germane sensitive enough to probe for low abundant RNAs? Yes, yes, definitely. So that is the one experimental challenge that was very important to me to overcome. And I didn't— I know I hinted to our splicing work of MAPT, but in that work we're actually probing nascent RNA structure. So in that sense, it's really low abundance because it exists— we're probing intronic sequences that exist for a very short amount of time. And so yes, it does work for low abundant, and the reason is because it only requires specific RT-PCR. So you know how by RT-PCR, theoretically, you can detect a single molecule. The sensitivity is very, very high. And I can talk a lot more about this, but the short answer is yes. It's very well— it works very well in low-abundant RNAs, including super low-abundant ones. So I'm going to move to one more question. How do you differ between the genomic RNA structure from the subgenome? Yeah, of course, that is an issue. And there, depending on how deeply I go into this issue, but the short answer, let me go through a short answer. The short answer is that the way the coronavirus genome works is that the first two-thirds of the genome can only come from the full genome. It's really the last one-third that has all the subgenomic RNAs, and we are looking at that as a separate problem, basically, to see— and I can go into a lot more detail, but yes, to see how much of the subgenomic RNA differs from this full-length genome. And there are some differences, But again, we don't understand yet the functional consequences of that. And I wasn't highlighting that. It's an additional complexity in the last one-third of the genome that is not easy to deal with, but we could deal with a little bit with RT-specific PCR. Again. Okay. If the RNA can fold in many different ways, how can you be sure that the ASO will catch your target sequence? Or is it possible to treat with a pool of many ASOs? Yeah, I mean, great question. Absolutely. I mean, it is possible to treat with pool of many ASOs. And it's just a matter of finding the best ones that will give you the least off-target effects. But yes, this is absolutely. Could you describe the project on MAPT? Understood you well, you mentioned the result of MAPT RNA. Yes, I can. It's basically in very short, kind of like one sentence description, what we found is that the structure that forms co-transcriptionally, so while, you know, when there's an intron present before even splicing has happened, is a huge driver for the splicing of MAPT. So, you know, there's been so much work done on this MAPT, and there are a lot of protein binding sites that have been identified. What we found is that actually, if you stabilize the structure, it doesn't matter, the protein binding, the energy of protein binding is much, basically, it's it controls the splicing much less. The structure is the largest driver of splicing, particularly for MAPT. If you stabilize the structure even a little bit, you can get a very huge difference in how this gene splices. We think that that is probably not specific to MAPT, And so we're kind of exploring that in a more genome-wide scale. But we haven't— this project, as we moved to the microbiology department of Harvard, so we've shifted focus more in microbes, like viruses and bacteria. But yes, it's a result that was extremely impressive to me and something that we're following up on. And I can talk more about it, but— Feel free to email me if there's something particular that you're interested in.

Does the drug— I think we only have time for 2 or 3 more questions.

So yeah, we can— Yeah, I can go to a couple more. Does the drugable RNA target and deflection strategy you mentioned for SARS-CoV-2 also apply for human? Yeah, absolutely. Yes. That's exactly— so the drugable RNA targeted deflection strategy does apply for human mRNA UTRs. The key, of course, is knowing that there is— like, first is identifying that there is a structure in the mammalian UTR that is functional, that's helping translation or, you know, doing something— splicing, translation, transcription, things like that. But it does. I had two questions. While I appreciate how important the view structures are, do you think the Beat Research Tool will be sufficient to identify lead structures for small molecule inhibitors, as interference with structure in the small scale would likely also influence long-range interactions. Sorry if I didn't read this question too loud, but yes, I think the in vitro work is definitely— I mean, whether it's sufficient, I think the visual work is very important, and I think it's just that when you know about the long-range interactions, you could— we actually, I mean, I didn't have time to go over this data, but we can entirely recapitulate the structure in vitro also if we include the sequence that permits the long-range attraction. So our whole point was, yes, you could probably interfere with both the small and the long range if you're just looking at the, the small thing, but you might as well look at the whole thing in vitro as well and give yourself better chances for drugs that will interfere for sure the long and the small. Have you performed studies with +1 ribosome? Is the mechanism quite different? Can you transfer your technique to this particular phenomenon? Oh, we have not performed studies with +1, Is it mechanistically different? There's not too, to my knowledge, there's not. I'm not sure it is mechanistically different and we haven't really looked at it, but yeah, I'd love to talk more. Again, I'm sorry if I didn't get into everybody's questions, but please, if you're interested, follow up with me. Yeah, I can, I can post my email or if you Google me, I think my email should be available.

But yeah, we can definitely Google Dr. Skin Lab and can see more research work and she published and follow up and see what she will publish. Like if you have a Twitter account, I think the audience can follow as well.

Twitter account too, Skin Lab. But contact me.

I'm very responsive over email usually. Yeah, Twitter is like a new science discussion field for professors right now. Yeah, so well, thank you all for the questions. Sorry that we cannot go through everything. And thank Dr. Ruskin again for this fantastic presentation. So after the seminar, we will post this video, the record, to our official website. And if you want to recall, you can just check our website if you're interested. So yeah, that will— that's all for today. And thank you one more time for attendance. Stay safe, and we look forward to seeing you all next time. Bye, everyone. Thanks. Bye.