Characterization of LNPs via Density Matching AUC
文稿
Hello everyone and welcome to today's webinar, Characterization of LNPs via Density Matching AUC.
I'm Cassie Saltman of LabRoots and I'll be your moderator for today's event. Today's educational web seminar is presented by LabRoots and brought to you by Beckman Coulter Life Sciences.
To learn more visit w-w-w-dot-beckman dot com. We encourage you to participate today by submitting any questions … and submit any technical issues here as well, if you have trouble seeing orhearing the presentation. I'd like to now welcome our speaker, Amy Henricksen, market development manager with Beckman Coulter Life Sciences. Amy, you may now begin your presentation. Wonderful. Thank you. Yeah. So my presentation is about the characterization of lipid nanoparticles using the analytical ultracentrifuge, from Beckman Coulter. So first, there's just this brief disclaimer, that the document here is confidential … and that it belongs to Beckman Coulter. So the first part that we're gonna start with is just a brief introduction to AUC. So analytical ultracentrifugation can be used for multiple different purposes and to help you characterize your samples in many different ways. The most …or one of the ways that you can do it is just determine the purity of your sample. So if once you've created your protein or your RNA or DNA or anything like that, you're able to measure it and just make sure that only the analyte of interest is there, or see if there's any contaminants present. You can also use it to get a size distribution of your sample, and to characterize several hydrodynamic or several biophysical properties. Sorry. So you can calculate the molecular weight of your sample, the diffusion coefficient. You can get information on the anisotropy of the sample, so kind of like the shape of your molecule if it's more globular or more elongated. You can use it for aggregation detection and to detect if there was any changes within your system during storage or other factors. So if you wanna see if storing it at four degrees is more stable than storing it at room temperature or at minus eighty, you can use it for studies like that to kind of compare between the different applications. It can also be used to measure equilibrium constants and thermodynamic parameters of the interaction. That one's a little challenging to do. Sometimes your reaction needs to be on their correct time scale, but it's still a possibility. And along those same lines, then you're also able to get information from your samples that are interacting, and you can tell if they're reversible interactions or irreversible written interactions. So it can be used for many things and more things than what's listed here. The way that AUC works is that the rotor seen here is originally put into the centrifuge, which is kind of on the outside here and a little hard to see in this photo. And then the sample …for your the sample to be analyzed is put into one of these AUC cells. So this is a top view of the cell and this is kind of the side view. So each AUC cell can hold up to two samples, in each of the different channels here. And then when you put it inside of the rotor, it's facing straight up like I mentioned in this photo. And the way that it works is, during centrifugation, a light will be flashed onto the sample. And then below the sample, there's a detector that's able to monitor the absorption of the sample during the progress of the run. So when the centrifugal forces are working on the sample, what's gonna happen is that the analytes inside of it are gonna start to sediment, towards the bottom of the AUC cell. The sedimentation of the samples depends on its molecular weight, its density, and as well as, its overall shape. So the more elongated it will be, the more friction there will be on the sample. So those kind of give you an overview of how AUC works. And then throughout this presentation, the software that I did the analysis on was UltraScan. There are several other software packages that can be used to analyze your AUC results. Depending on what you wanna do and what you're looking at, they all have their cons and their benefits. But for this one, I used UltraScan, which is working on helping to move the software towards a more GMP environment, which…can be found in the paper highlighted at the bottom here. So now we'll be moving into the density matching AUC. We've published this work in ACS Nano. We performed it at the University of Lethbridge under my professor there, who is doctor Borries Demeler, and we did it in collaboration with Doctor Pieter Cullis's lab at UBC. So if you want more details on what I'm gonna be discussing here today, feel free to check out the paper or send me any questions that you guys might have. So the LNP samples that we looked at were loaded with siRNA, which I believe was 21 nucleotides in length. They… we were …given three different LNP samples. The first were the empty lipid nanoparticles, seen on the left hand side of the screen. So these didn't contain any RNA introduced into them. They were just made up of the lipid nanoparticle components. The next sample is the NP6 sample, which has some siRNA loaded into it. And then the third sample is the NP. And these samples were overloaded with the RNA and were not clinically relevant samples. So in, like, the NP6, NP1 naming here, the N…just stands for the polyamine group in the lipid, whereas the P stands for the phosphate group and then nucleic acid. So in the NP6 sample, there's a lot more lipid than there is RNA, which makes sense because they have less RNA loaded into them. So our collaborators originally approached us because they were having some trouble differentiating if there was any empty lipid nanoparticles filled into the NP6 sample. When you look at the TEM images seen here, the NP1 sample on the right has these striations kind of inside of the lipid nanoparticles that are imaged. And that allows them to tell these apart from all of… from the other two samples. However when you’re looking at the NP6 sample, it’s really hard to tell if it's different, like it's hard to differentiate it from the empty LNPs. And the issue here then is that you're not sure if your NP6, your clinically relevant sample, contains any empty LNPs in it or if it contains the RNA of interest even. So they asked us to try and help determine if we could tell the difference between the clinically relevant samples, the …empty LNP samples, as well as the NP1. So the first step for us was just to do a typical sedimentation velocity experiment seen here. For this, we just loaded our… in this example, the NP6 sample into an AUC cell, and we centrifuged it, monitored the sedimentation coefficient, and we got the distribution seen here. So this plot might look a little weird to some people even if you're aware of AUC and you've done analysis in it. So I'll kinda walk you through it quickly. So on the side here, we have the boundary fraction. So instead of having the typical AUC peaks that people tend to look at more for AUC, we're just looking at the integral of that, and that allows us to get the line shown here instead of the peaks. So the boundary fraction goes from zero to one… or 0% percent to 100% in other… in another way of saying it. So if you were to, say, draw a line across at, 45% where we have that kinda, like, elbow bend in the sample, you could say that 45% of the sample, is sedimenting or floating faster than the other 65%. The other odd thing about this graph, as I just kind of mentioned, is that the sedimentation coefficient is negative. That means that these samples actually floated when we put them inside of the AUC, and that would be because their density… or… is the partial specific volume is greater than that of water, and that results in this flotation pattern, which is really interesting to see. However, when we also measured the empty lipid nanoparticles, we got fairly similar sedimentation coefficient values. There was some difference, but it was hard to tell exactly what was empty and what was loaded, and if there was any empties in the loaded sample… Therefore, we decided to use the density matching experiment, which is what this is about. To do that, you're gonna take the exact same sample at the exact same concentration, and you're gonna measure it in a buffer that changes only in the density of it. And this can be done by introducing… by changing the ratio of D2Oand H2O in the buffer. And that results in a plot shown here. So when you're adding D2O to your buffer, you're going to be changing its density and viscosity. And as you change the density and viscosity of the buffer, you're going to affect the sedimentation of your sample. In the software, what we do is… we do not account for the change in density that the buffer has on the sample, and this, then allows us to see a shift in the sedimentation pattern. So here you can see in red, we have our NP6 sample measured in light water, and then as we add increasing amounts of D2O, we get faster flotation, seen by increasing negative numbers. The important thing that you want to do when you're doing this experiment is make sure that you're keeping the concentration of your sample the same throughout the entire… throughout the four experiments that you're doing here. So we measured it at four different D2O/H2O ratios. Sorry. I lost my spot there for a second. And you just wanna make sure that you're using the exact same concentration of sample in each… in case there is some sort of issue when you… in case there is some sort of effect that's going to happen on the samples when you're changing their volume. The other thing that you wanna double check once you've lined up all of these different experiments done in the different D2O/H2O concentrations is ensure that the boundary shape is the same for all the different experiments. So here, you can kinda tell that they all have roughly the same kind of, like, shape, and the same, like, kinda elbow bend point in the line. So that shows that the the addition of D2O to our sample doesn't change the sample's inherent sedimentation properties. It does change the sedimentation coefficient, but it doesn't affect the sample. So okay. We'll go back to this. So then once you have all of these different experiments run at the different D2O/H2O buffer densities, what's gonna happen in the software is that it'll go across at each boundary fraction. So say, at 0.4, it'll come across, and it when we'll find out what these sedimentation coefficient is for that boundary fraction at each of the different D2O concentrations. It's then going to take those four sedimentation coefficients and plot them on a graph of sedimentation coefficient versus buffer density. Once these four plots… or once these four spots are plotted for each of the different boundary fractions, what's gonna happen is that the software will create a line of best fit between the different D2O concentrations and sedimentation coefficients and extrapolate that line down to a sedimentation coefficient of zero. At a sedimentation coefficient of zero, this means that your sample is not going to float or sediment in that density of buffer, and therefore, you know the density of your sample. Because at a sedimentation coefficient of zero, the density of the sample will equal the density of the buffer. After you do this for each boundary fraction, you can create an integral or you can create a plot of boundary fraction versus partial specific volume for each sample. So in this plot here, we had a couple of controls that we also did to make sure that the method worked properly. So on the left hand side of the plot, there is a brown or orange line there, which corresponds to the siRNA that was loaded into the LNPs. These have the lowest partial specific volume, which makes sense because nucleic acids have a lower partial specific volume than lipids and proteins. We can also see that the siRNA line as well as the plasma DNA are both very straight and linear, and they only have one value for their partial specific volume, which makes sense. As we move up towards the middle of the graph here, we start to see our protein standard, which we used was BSA, having a partial specific volume of about 6.9. Here you can see that there are two different partial specific volumes, though. There's one that's slightly lower than that at about 6.6. We were kinda surprised to see two different partial specific volumes showing up for the BSA, but we attributed this difference to most likely being that there was monomer and dimer in our BSA sample, and that there was a difference in the… hydrogen deuterium exchange rate happening between the monomer and the dimer. So here we're able to see that the…monomer, is likely the larger portion at about 6.69, partial specific volume, whereas the dimer is probably that smaller portion around .66. This is a value lower than theoretical partial specific volume of BSA, and the reason for that is because of that hydrogen deuterium exchange that's happening. And if that is a concern, you can use H2 18O, which will prevent the hydrogen which doesn't have any deuterium in it, and, therefore, you won't get the hydrogen deuterium exchange, and you're able to get a more accurate partial specific volume. However, from what we were doing, we decided this was not an issue. The lipids most likely would not have too much hydrogen deuterium exchange. And if they did, it would be relatively the same for all three of the lipid nanoparticle samples. And since we're not looking for exact partial specific volumes for this, we're just looking for relative partial specific volumes. We continued forward with this study using deuterium. So if we zoom in on this plot looking at just the lipid nanoparticles on the right hand side of the graph, we can see them showing here. The partial specific volume of water is approximately one. So when you're looking at a graph like this, anything above one should float in water and anything below one should sediment in water. And that's exactly what we saw in our AUC results. If you guys remember, I was just showing you previously the sedimentation plots for the NP6 sample, and this sample had negative sedimentation coefficients, meaning that it had a greater than one partial specific volume. And you can see that in the green trace here, and you can see that it's greater than that and therefore should float in water, which is what we saw. The next thing that you can see here is that as we added in more RNA to get to our NP1 sample, our partial specific volume decreased. And that makes sense because the addition of RNA is more dense or is, yeah, more dense, less partial specific volume-y. So you start to get that shift of the overloaded NP1 samples towards a lower partial specific volume. And then the empty LNPs shown in blue… have no RNA in them and therefore have the highest partial specific volume. From this plot, if you look at the green line and the blue line, you can see that there's none to maybe a very little amount of overlap between the green plot and the blue plot. And this indicates to us that our NP6 sample does not contain any empty LNPs. And if it does, it's a very minute portion… approximately 3% of the sample. Whereas when you look at the NP1, you can tell that there is absolutely no empty LNPs present in that plot. If there were empty LNPs present in the NP1 sample, you would see that red line kinda like reach all the way across over to the blue line and touch it. You can also tell that these three plots do not contain any RNA… free RNA. If there was free RNA, then you would see in this graph here that the partial specific volume plot for the lipid nanoparticles would have a tail that kind of reaches over towards where the siRNA on its own sediments. We also then confirmed this doing multiwavelength AUC, which I don't have the results in this presentation just because they take a little bit longer to go over. But by multiwavelength AUC, we were able to see the RNA sedimentation pattern as well as the sedimentation pattern of the lipid nanoparticles, and we could tell that there was no free RNA present in that sample. So once we have the partial specific volume of our samples, what we can do is actually mathematically calculate the hydrodynamic radius. Since this isn't very frequently done with AUC, we compared the results to light scattering and cryo-TEM. You can see that all three of these plots look pretty different. They have different shapes in their lines, and that would be expected just based on the different calculation methods that the three different methods use. The first thing that's kinda different between them is that the light scattering in AUC produced slightly larger, radii than the cryo-TEM did, and that makes perfect sense. Because when you're looking at TEM, you're looking at your particle radius, whereas light scattering and AUC look at the hydrodynamic radius, meaning that it's looking at the particle radius plus the hydration shell around it and therefore should have a larger radius than just the particle radius. The next thing that we see here is that the cryo-TEM tends to kind of have slightly more jaggedy lines, and that's just due to the single particle counting of it. Whereas light scattering tends to average your hydrodynamic radiuses more and tends to overemphasize those larger particles slightly more than the smaller ones. And then AUC is also a bulk counting method, but it doesn't over-emphasize the large particles as much as light scattering does. However, these three techniques all agree fairly well with each other. They all found that the empty LNPs had the lowest, hydrodynamic radius, and empty ones had the largest hydrodynamic radius. And they all cover roughly the same range. For light scattering, the NP6 in green goes from about 10 to 23. In AUC, it goes from about 15 to 24, and cryo-TEM goes from about 15 to 23 as well. So they all agree fairly well across all three of the different methods and for all three of the samples. And this allowed us to determine that AUC could calculate correct hydrodynamic radiuses and still provide high resolution and sensitivity for these heterogeneous samples. Once we were able to calculate the hydrodynamic radius, we could then calculate the molar mass of the samples. So here is our molar mass distribution plot shown. The empty LNPs have the lowest molar mass, whereas the NP1shave the highest molar mass and also have the highest or the largest distribution, all of which makes sense when you think about it. The more RNA you add, the larger the molar mass will be. The NP1s also have the largest hydrodynamic radius. So this plot just helps to confirm our findings as well. So a little over… summary of this section, I guess, then is that AUC can be used to determine the partial specific volume of your lipid nanoparticle samples, the molar mass, and their hydrodynamic radius as well can all be characterized. This allows you to get information on the loading state of your LNPs as well as their purity. So if there's any empty lipid nanoparticles or free RNA present in the sample. And then, following this work, one of our collaborators then furthered the work, Alexander Bepperling, actually then went on and published a paper, where he was able to use this method and then some additional mathematics. And you could calculate the… he could calculate the mRNA copy number loaded inside of the lipid nanoparticles, which then kind of helps to give that, like, stronger identity to…that you're able to get from these lipid nanoparticles is that besides just being able to tell that they're pure, you can tell, the distribution of mRNA copy number loaded into your… LNP distribution. The next section that we're gonna look at is the RNA-RNA interactions with AUC. I thought that this kinda tied into our lipid nanoparticle talk somewhat because of the lipid nanoparticles right now tend to be looking a lot at RNA loading, and you wanna know what you're putting into your LNP samples and what you're putting into your drug products. So therefore, I've included a section on RNA-RNA interaction with AUC because it's pretty interesting, and it shows you that AUC can be used for RNA-RNA work. So the…work that we did here was published in Nucleic Acids Research, and it was done in collaboration with doctor Trushar Patel's lab at the University of Lethbridge. For this work, we were working with a flaviviral virus, and its genome is RNA, single-stranded RNA. It's a linear genome initially, but in order to replicate, the genome needs to cyclize. So the idea here, what they wanted to do was try and find out what the main drivers of this cyclization were. The idea being that if you can find out what drives the cyclization within the sequence, you could potentially try and disrupt that and prevent the cyclization, therefore…preventing replication and ongoing infections. So it's been shown before that the 5 prime TR and the 3 prime TR are what interact to cause this cyclization. Throughout those, we did a computational analysis of many different flaviviral virus genomes, and were able to find several consensus sequences across all of the genomes. With the one showing up the most likely as being identical and not changed being an 11 nucleotide sequence that we've termed the cyclization sequence and is shown in the bottom left hand corner of your screen. So the cyclization sequence then is shown on our linear RNA strand at the top of the screen. Sorry I don’t have my pointer, but showing at the top in the linear strand as those little black lines that's then highlighted in red to kind of help highlight where the cyclization sequence is found. So for this study, what was done is that we took a 225 nucleotide section of the 3 prime TR as well as a 225 nucleotide section of the 5 prime TR, plus a little bit of the structural gene because the cyclization sequence is found inside of the structural gene at the 5 prime end. But both of these were 225 nucleotides long, and they both contain the cyclization sequence. So the way that the replication works is that it cyclizes NS5 then binds to the cyclized region, and it produces the next RNA strand that can then be used to create the replication of the virus. So the first method that was used to characterize this cyclization was SEC MALS. The way that this method works is that we originally separate our sample over a size exclusion column separating any molecules that are present based on their size. The sample then passes through the MALS detector system. While it's passing through, it's shot with or shot, yeah shot with a laser. And when the laser hits the particles inside of the sample, it then causes the laser light to be scattered. It's then detected at multi-angles resulting in it being a multi-angle light scattering method. So we detect the scattered light at multiple angles. And if you know your dn/dc, which is an intrinsic value of your sample and can be found online, as well as the refractive index of your solvent, you're then able to get accurate molecular weights as well as radiuses of gyration. So the results for this sample, are seen here on this slide. So on the left of the slide is the SEC, the SEC results, and the right side shows the MALS results. So if we're looking at the SEC results under the green arrow there, we can see that the peak comes off at about 12 mils. When we're… if we just only look at the brown green line right now for the 3 prime TR as well as the pink line for the 5 prime TR, we see that the majority of the sample comes off at 12 milliliters. When this went through the, the MALS detector, it corresponded to a molecular weight of approximately 75 kilodalton, which is a few… which aligns well with the theoretical molecular weight of the monomer 225 nucleotide RNA. So that's perfect. It shows that when we run our 3 prime TR alone and the 5 prime TR alone, they come off at about 12 mils and have a molecular weight of about 75 kilodaltons corresponding to a monomer RNA strand. When we then mix these together in the SEC plot, we get the purple trace. And we see the addition of this second peak appearing at about 10 milliliters. This 10 milliliter peak corresponded on the… MALS instrument to where the red arrow is there and had a molecular weight of about 150 kilodaltons, which would then be two of our RNAs together. So this helped show that when you mix the 5 prime TR and the 3 prime TR that you get a dimer formation. However, if you look at the SEC plots of the individual RNAs, you see that they both have a slight shoulder peak at approximately 10 to 11 milliliters. And this showed that these RNAs did form homodimers with themselves. The homodimer would also then have the same molecular weight as the heterodimer. There… and we have tried a lot of things to try and purify… purify away this, like, dimer formation within the individual RNAs, and it just didn't happen. They always formed homodimers with themselves. And because of this, we couldn't be certain that the dimer complex we saw when we mix the two of them together was a heterodimer of the 5 prime and 3 prime together and not just somehow, like, forcing the 5 prime to bind with itself. So the next study that we did was turned to AUC. This is just another brief overview of AUC showing the things that you can do with it, but it says the same things that I talked about at the beginning. So we're just gonna keep going. So the first thing that we did with our RNA was we measured the 3 prime terminal region on its own, and we got a sedimentation distribution shown here with the main peak being in about 5.3 s. We then measured the 5 prime terminal region on its own, and we see that the main peak here ends up at about 5.9 s. This was really interesting to us actually, very exciting. These two RNAs have the same nucleotide length, but they produce different sedimentation coefficients, and that allows us to tell these apart. The difference in sedimentation coefficient is most likely due to a difference in folding state. So probably, so one of them is more globular than the other, and this results in a different frictional ratio affecting the sedimentation coefficient of the samples. So we're able to tell them apart. So, of course, then the next, the next experiment we did was mix the 5 prime and 3 prime together and measure them in the AUC. And in that case, we see the addition of that new peak forming at about 7.5 s. The exciting thing here is that we see this formation of that, higher sedimentation coefficient peak showing up, but we also see that the ratio of the 5 prime and 3 prime monomer peaks decreased by relatively the same ratio. And this helps us to, this helps us to identify that this is the formation of a heterodimer and not of a homodimer. Because If it was just the 3 prime binding with itself, you'd only see the 3 prime peak decrease. So it was really exciting, and it helps show that the 5 prime and 3 prime UTRs bound together in a heterodimer formation, resulting in the higher sedimentation coefficient around 7.5 to 8.2 s. We next then, they went on to do some MST. With MST, what you do is you take… we took one of the RNA molecules and fluorescently labeled it and then serially diluted in the second RNA molecule. Then you fill them into the capillaries showing up here on the screen. And as the sample, during the experiment, the capillaries are then, shot with a laser, which results in the sample heating up. This, increased temperature at the area where the laser hits causes the samples to move away from that heated area. And then as the… as the solution cools back down, the samples are gonna diffuse into it, and the diffusion rate is going to be different if the RNA is bound versus unbound. So using this, we're able to calculate a KD of our reaction. So the first thing that was done was MST on the 5 prime and 3 prime UTRs mixed together, and it resulted in a KD of about .6 … .06 micromolar, which is a pretty good KD. However, as I mentioned, there were a few other consensus sequence… sequences found within those 225 nucleotide region. So to see if the 11 nucleotides that we identified was the main driving force for cyclization, what we did was we mutated only those 11 nucleotides that they would not bind with each other, but we left the other 214 nucleotides untouched. So they still had their secondary structure and could still bind if they wanted. And what that resulted in was a significant decrease in KD as seen in the green line here. Here we've got… we went up to…our significant increase in the KD, meaning less tight binding, and it went up to about 7 micromolar. So from these three results, we were able to verify the binding of the RNA using three separate techniques. And then the exciting thing for AUC users is that we were able to measure the RNA in the AUC, and we were able to find that they gave different sedimentation coefficients due to their folding state. And then we also used MST to determine their nanomolar affinity. These RNAs worked really well in our AUC. As you guys saw, there were some…let me just go back to here.
There are some additional peaks shown here besides just the main one corresponding to most likely the dimer and potentially a small degradant of the RNA. However, this isn't always the case in the RNAs we've measured. If they tend to have lots of different folding states, you will see multiple peaks appearing. So we troubleshot this a little bit to try and figure out if it was…the RNA was aggregating or degrading or what it was. And we found that if you add urea to the sample to your RNA sample and heat it up and then measure it in the AUC inside of that urea, you can get a single peak if your RNA is pure, and that can help you to identify if your RNA is pure or not. It's kinda like doing a urea PAGE inside of an AUC, but it helps to tell you if it's just, folding structure that's causing those different sedimentation coefficients or if it is that you have contaminants in your RNA prep. interacting, and you can tell if they're reversible interactions or irreversible written interactions. So it can be used for many things and more things than what's listed here. The way that AUC works is that the rotor seen here is originally put into the centrifuge, which is kind of on the outside here and a little hard to see in this photo. And then the sample …for your the sample to be analyzed is put into one of these AUC cells. So this is a top view of the celland this is kind of the side view. So each AUC cell can hold up to two samples, in each of the different channels here. And then when you put it inside of the rotor, it's facing straight up like I mentioned in this photo. And the way that it works is, during centrifugation, a light will be flashed onto the sample. And then below the sample, there's a detector that's able to monitor the absorption of the sample during the progress of the run. So when the centrifugal forces are working on the sample, what's gonna happen is that the analytes inside of it are gonna start to sediment, towards the bottom of the AUC cell. The sedimentation of the samples depends on its molecular weight, its density, and as well as, its overall shape. So the more elongated it will be, the more friction there will be on the sample.
So those kind of give you an overview of how AUC works.
And then throughout this presentation, the software that I did the analysis on was UltraScan. There are several other software packages that can be used to analyze your AUC results. Depending on what you wanna do and what you're looking at, they all have their cons and their benefits. But for this one, I used UltraScan, which is working on helping to move the software towards a more GMP environment, which…can be found in the paper highlighted at the bottom here.
So now we'll be moving into the density matching AUC.
We've published this work in ACS Nano.
We performed it at the University of Lethbridge under my professor there, who is doctor Borries Demeler, and we did it in collaboration with Doctor Pieter Cullis's lab at UBC.
So if you want more details on what I'm gonna be discussing here today, feel free to check out the paper or send me any questions that you guys might have.
So the LNP samples that we looked at were loaded with siRNA, which I believe was 21 nucleotides in length.
They… we were …given three different LNP samples. The first were the empty lipid nanoparticles, seen on the left hand side of the screen. So these didn't contain any RNA introduced into them. They were just made up of the lipid nanoparticle components.
The next sample is the NP6 sample, which has some siRNA loaded into it. And then the third sample is the NP. And these samples were overloaded with the RNA and were not clinically relevant samples.
So in, like, the NP6, NP1 naming here, the N…just stands for the polyamine group in the lipid, whereas the P stands for the phosphate group and then nucleic acid. So in the NP6 sample, there's a lot more lipid than there is RNA, which makes sense because they have less RNA loaded into them.
So our collaborators originally approached us because they were having some trouble differentiating if there was any empty lipid nanoparticles filled into the NP6 sample. When you look at the TEM images seen here, the NP1 sample on the right has these striations kind of inside of the lipid nanoparticles that are imaged. And that allows them to tell these apart from all of… from the other two samples. However when you’re looking at the NP6 sample, it’s really hard to tell if it's different, like it's hard to differentiate it from the empty LNPs. And the issue here then is that you're not sure if your NP6, your clinically relevant sample, contains any empty LNPs in it or if it contains the RNA of interest even.
So they asked us to try and help determine if we could tell the difference between the clinically relevant samples, the …empty LNP samples, as well as the NP1.
So the first step for us was just to do a typical sedimentation velocity experiment seen here.
For this, we just loaded our… in this example, the NP6 sample into an AUC cell, and we centrifuged it, monitored the sedimentation coefficient, and we got the distribution seen here. So this plot might look a little weird to some people even if you're aware of AUC and you've done analysis in it. So I'll kinda walk you through it quickly.
So on the side here, we have the boundary fraction. So instead of having the typical AUC peaks that people tend to look at more for AUC, we're just looking at the integral of that, and that allows us to get the line shown here instead of the peaks.
So the boundary fraction goes from zero to one… or 0% percent to 100% in other… in another way of saying it.
So if you were to, say, draw a line across at, 45% where we have that kinda, like, elbow bend in the sample, you could say that 45% of the sample, is sedimenting or floating faster than the other 65%.
The other odd thing about this graph, as I just kind of mentioned, is that the sedimentation coefficient is negative.
That means that these samples actually floated when we put them inside of the AUC, and that would be because their density… or… is the partial specific volume is greater than that of water, and that results in this flotation pattern, which is really interesting to see. However, when we also measured the empty lipid nanoparticles, we got fairly similar sedimentation coefficient values. There was some difference, but it was hard to tell exactly what was empty and what was loaded, and if there was any empties in the loaded sample… Therefore, we decided to use the density matching experiment, which is what this is about.
To do that, you're gonna take the exact same sample at the exact same concentration, and you're gonna measure it in a buffer that changes only in the density of it. And this can be done by introducing… by changing the ratio of D2Oand H2O in the buffer.
And that results in a plot shown here. So when you're adding D2O to your buffer, you're going to be changing its density and viscosity.
And as you change the density and viscosity of the buffer, you're going to affect the sedimentation of your sample.
In the software, what we do is… we do not account for the change in density that the buffer has on the sample, and this, then allows us to see a shift in the sedimentation pattern.
So here you can see in red, we have our NP6 sample measured in light water, and then as we add increasing amounts of D2O, we get faster flotation, seen by increasing negative numbers.
The important thing that you want to do when you're doing this experiment is make sure that you're keeping the concentration of your sample the same throughout the entire… throughout the four experiments that you're doing here. So we measured it at four different D2O/H2O ratios.
Sorry. I lost my spot there for a second. And you just wanna make sure that you're using the exact same concentration of sample in each… in case there is some sort of issue when you… in case there is some sort of effect that's going to happen on the samples when you're changing their volume.
The other thing that you wanna double check once you've lined up all of these different experiments done in the different D2O/H2O concentrations is ensure that the boundary shape is the same for all the different experiments. So here, you can kinda tell that they all have roughly the same kind of, like, shape, and the same, like, kinda elbow bend point in the line.
So that shows that the the addition of D2O to our sample doesn't change the sample's inherent sedimentation properties. It does change the sedimentation coefficient, but it doesn't affect the sample.
So okay. We'll go back to this. So then once you have all of these different experiments run at the different D2O/H2O buffer densities, what's gonna happen in the software is that it'll go across at each boundary fraction. So say, at 0.4, it'll come across, and it when we'll find out what these sedimentation coefficient is for that boundary fraction at each of the different D2O concentrations.
It's then going to take those four sedimentation coefficients and plot them on a graph of sedimentation coefficient versus buffer density.
Once these four plots… or once these four spots are plotted for each of the different boundary fractions, what's gonna happen is that the software will create a line of best fit between the different D2O concentrations and sedimentation coefficients and extrapolate that line down to a sedimentation coefficient of zero.
At a sedimentation coefficient of zero, this means that your sample is not going to float or sediment in that density of buffer, and therefore, you know the density of your sample. Because at a sedimentation coefficient of zero, the density of the sample will equal the density of the buffer.
After you do this for each boundary fraction, you can create an integral or you can create a plot of boundary fraction versus partial specific volume for each sample.
So in this plot here, we had a couple of controls that we also did to make sure that the method worked properly.
So on the left hand side of the plot, there is a brown or orange line there, which corresponds to the siRNA that was loaded into the LNPs.
These have the lowest partial specific volume, which makes sense because nucleic acids have a lower partial specific volume than lipids and proteins.
We can also see that the siRNA line as well as the plasma DNA are both very straight and linear, and they only have one value for their partial specific volume, which makes sense.
As we move up towards the middle of the graph here, we start to see our protein standard, which we used was BSA, having a partial specific volume of about 6.9.
Here you can see that there are two different partial specific volumes, though. There's one that's slightly lower than that at about 6.6.
We were kinda surprised to see two different partial specific volumes showing up for the BSA, but we attributed this difference to most likely being that there was monomer and dimer in our BSA sample, and that there was a difference in the… hydrogen deuterium exchange rate happening between the monomer and the dimer. So here we're able to see that the…monomer, is likely the larger portion at about 6.69, partial specific volume, whereas the dimer is probably that smaller portion around .66.
This is a value lower than theoretical partial specific volume of BSA, and the reason for that is because of that hydrogen deuterium exchange that's happening. And if that is a concern, you can use H2 18O, which will prevent the hydrogen which doesn't have any deuterium in it, and, therefore, you won't get the hydrogen deuterium exchange, and you're able to get a more accurate partial specific volume.
However, from what we were doing, we decided this was not an issue.
The lipids most likely would not have too much hydrogen deuterium exchange. And if they did, it would be relatively the same for all three of the lipid nanoparticle samples. And since we're not looking for exact partial specific volumes for this, we're just looking for relative partial specific volumes. We continued forward with this study using deuterium.
So if we zoom in on this plot looking at just the lipid nanoparticles on the right hand side of the graph, we can see them showing here.
The partial specific volume of water is approximately one. So when you're looking at a graph like this, anything above one should float in water and anything below one should sediment in water. And that's exactly what we saw in our AUC results. If you guys remember, I was just showing you previously the sedimentation plots for the NP6 sample, and this sample had negative sedimentation coefficients, meaning that it had a greater than one partial specific volume. And you can see that in the green trace here, and you can see that it's greater than that and therefore should float in water, which is what we saw.
The next thing that you can see here is that as we added in more RNA to get to our NP1 sample, our partial specific volume decreased. And that makes sense because the addition of RNA is more dense or is, yeah, more dense, less partial specific volume-y.
So you start to get that shift of the overloaded NP1 samples towards a lower partial specific volume. And then the empty LNPs shown in blue… have no RNA in them and therefore have the highest partial specific volume.
From this plot, if you look at the green line and the blue line, you can see that there's none to maybe a very little amount of overlap between the green plot and the blue plot. And this indicates to us that our NP6 sample does not contain any empty LNPs. And if it does, it's a very minute portion… approximately 3% of the sample.
Whereas when you look at the NP1, you can tell that there is absolutely no empty LNPs present in that plot. If there were empty LNPs present in the NP1 sample, you would see that red line kinda like reach all the way across over to the blue line and touch it. You can also tell that these three plots do not contain any RNA… free RNA.
If there was free RNA, then you would see in this graph here that the partial specific volume plot for the lipid nanoparticles would have a tail that kind of reaches over towards where the siRNA on its own sediments.
We also then confirmed this doing multiwavelength AUC, which I don't have the results in this presentation just because they take a little bit longer to go over.
But by multiwavelength AUC, we were able to see the RNA sedimentation pattern as well as the sedimentation pattern of the lipid nanoparticles, and we could tell that there was no free RNA present in that sample.
So once we have the partial specific volume of our samples, what we can do is actually mathematically calculate the hydrodynamic radius.
Since this isn't very frequently done with AUC, we compared the results to light scattering and cryo-TEM.
You can see that all three of these plots look pretty different. They have different shapes in their lines, and that would be expected just based on the different calculation methods that the three different methods use. The first thing that's kinda different between them is that the light scattering in AUC produced slightly larger, radii than the cryo-TEM did, and that makes perfect sense. Because when you're looking at TEM, you're looking at your particle radius, whereas light scattering and AUC look at the hydrodynamic radius, meaning that it's looking at the particle radius plus the hydration shell around it and therefore should have a larger radius than just the particle radius.
The next thing that we see here is that the cryo-TEM tends to kind of have slightly more jaggedy lines, and that's just due to the single particle counting of it. Whereas light scattering tends to average your hydrodynamic radiuses more and tends to overemphasize those larger particles slightly more than the smaller ones. And then AUC is also a bulk counting method, but it doesn't over-emphasize the large particles as much as light scattering does.
However, these three techniques all agree fairly well with each other. They all found that the empty LNPs had the lowest, hydrodynamic radius, and empty ones had the largest hydrodynamic radius. And they all cover roughly the same range. For light scattering, the NP6 in green goes from about 10 to 23. In AUC, it goes from about 15 to 24, and cryo-TEM goes from about 15 to 23 as well. So they all agree fairly well across all three of the different methods and for all three of the samples.
And this allowed us to determine that AUC could calculate correct hydrodynamic radiuses and still provide high resolution and sensitivity for these heterogeneous samples.
Once we were able to calculate the hydrodynamic radius, we could then calculate the molar mass of the samples.
So here is our molar mass distribution plot shown. The empty LNPs have the lowest molar mass, whereas the NP1shave the highest molar mass and also have the highest or the largest distribution, all of which makes sense when you think about it. The more RNA you add, the larger the molar mass will be. The NP1s also have the largest hydrodynamic radius.
So this plot just helps to confirm our findings as well.
So a little over… summary of this section, I guess, then is that AUC can be used to determine the partial specific volume of your lipid nanoparticle samples, the molar mass, and their hydrodynamic radius as well can all be characterized.
This allows you to get information on the loading state of your LNPs as well as their purity. So if there's any empty lipid nanoparticles or free RNA present in the sample.
And then, following this work, one of our collaborators then furthered the work, Alexander Bepperling, actually then went on and published a paper, where he was able to use this method and then some additional mathematics. And you could calculate the… he could calculate the mRNA copy number loaded inside of the lipid nanoparticles, which then kind of helps to give that, like, stronger identity to…that you're able to get from these lipid nanoparticles is that besides just being able to tell that they're pure, you can tell, the distribution of mRNA copy number loaded into your… LNP distribution.
The next section that we're gonna look at is the RNA-RNA interactions with AUC.
I thought that this kinda tied into our lipid nanoparticle talk somewhat because of the lipid nanoparticles right now tend to be looking a lot at RNA loading, and you wanna know what you're putting into your LNP samples and what you're putting into your drug products. So therefore, I've included a section on RNA-RNA interaction with AUC because it's pretty interesting, and it shows you that AUC can be used for RNA-RNA work.
So the…work that we did here was published in Nucleic Acids Research, and it was done in collaboration with doctor Trushar Patel's lab at the University of Lethbridge.
For this work, we were working with a flaviviral virus, and its genome is RNA, single-stranded RNA.
It's a linear genome initially, but in order to replicate, the genome needs to cyclize.
So the idea here, what they wanted to do was try and find out what the main drivers of this cyclization were. The idea being that if you can find out what drives the cyclization within the sequence, you could potentially try and disrupt that and prevent the cyclization, therefore…preventing replication and ongoing infections.
So it's been shown before that the 5 prime TR and the 3 prime TR are what interact to cause this cyclization.
Throughout those, we did a computational analysis of many different flaviviral virus genomes, and were able to find several consensus sequences across all of the genomes. With the one showing up the most likely as being identical and not changed being an 11 nucleotide sequence that we've termed the cyclization sequence and is shown in the bottom left hand corner of your screen.
So the cyclization sequence then is shown on our linear RNA strand at the top of the screen. Sorry I don’t have my pointer, but showing at the top in the linear strand as those little black lines that's then highlighted in red to kind of help highlight where the cyclization sequence is found.
So for this study, what was done is that we took a 225 nucleotide section of the 3 prime TR as well as a 225 nucleotide section of the 5 prime TR, plus a little bit of the structural gene because the cyclization sequence is found inside of the structural gene at the 5 prime end.
But both of these were 225 nucleotides long, and they both contain the cyclization sequence.
So the way that the replication works is that it cyclizes NS5 then binds to the cyclized region, and it produces the next RNA strand that can then be used to create the replication of the virus.
So the first method that was used to characterize this cyclization was SEC MALS.
The way that this method works is that we originally separate our sample over a size exclusion column separating any molecules that are present based on their size.
The sample then passes through the MALS detector system. While it's passing through, it's shot with or shot, yeah shot with a laser. And when the laser hits the particles inside of the sample, it then causes the laser light to be scattered. It's then detected at multi-angles resulting in it being a multi-angle light scattering method. So we detect the scattered light at multiple angles.
And if you know your dn/dc, which is an intrinsic value of your sample and can be found online, as well as the refractive index of your solvent, you're then able to get accurate molecular weights as well as radiuses of gyration.
So the results for this sample, are seen here on this slide. So on the left of the slide is the SEC, the SEC results, and the right side shows the MALS results.
So if we're looking at the SEC results under the green arrow there, we can see that the peak comes off at about 12 mils. When we're… if we just only look at the brown green line right now for the 3 prime TR as well as the pink line for the 5 prime TR, we see that the majority of the sample comes off at 12 milliliters.
When this went through the, the MALS detector, it corresponded to a molecular weight of approximately 75 kilodalton, which is a few… which aligns well with the theoretical molecular weight of the monomer 225 nucleotide RNA.
So that's perfect. It shows that when we run our 3 prime TR alone and the 5 prime TR alone, they come off at about 12 mils and have a molecular weight of about 75 kilodaltons corresponding to a monomer RNA strand.
When we then mix these together in the SEC plot, we get the purple trace.
And we see the addition of this second peak appearing at about 10 milliliters.
This 10 milliliter peak corresponded on the… MALS instrument to where the red arrow is there and had a molecular weight of about 150 kilodaltons, which would then be two of our RNAs together.
So this helped show that when you mix the 5 prime TR and the 3 prime TR that you get a dimer formation.
However, if you look at the SEC plots of the individual RNAs, you see that they both have a slight shoulder peak at approximately 10 to 11 milliliters.
And this showed that these RNAs did form homodimers with themselves.
The homodimer would also then have the same molecular weight as the heterodimer.
There… and we have tried a lot of things to try and purify… purify away this, like, dimer formation within the individual RNAs, and it just didn't happen. They always formed homodimers with themselves. And because of this, we couldn't be certain that the dimer complex we saw when we mix the two of them together was a heterodimer of the 5 prime and 3 prime together and not just somehow, like, forcing the 5 prime to bind with itself.
So the next study that we did was turned to AUC.
This is just another brief overview of AUC showing the things that you can do with it, but it says the same things that I talked about at the beginning. So we're just gonna keep going.
So the first thing that we did with our RNA was we measured the 3 prime terminal region on its own, and we got a sedimentation distribution shown here with the main peak being in about 5.3 s. We then measured the 5 prime terminal region on its own, and we see that the main peak here ends up at about 5.9 s. This was really interesting to us actually, very exciting.
These two RNAs have the same nucleotide length, but they produce different sedimentation coefficients, and that allows us to tell these apart.
The difference in sedimentation coefficient is most likely due to a difference in folding state.
So probably, so one of them is more globular than the other, and this results in a different frictional ratio affecting the sedimentation coefficient of the samples. So we're able to tell them apart.
So, of course, then the next, the next experiment we did was mix the 5 prime and 3 prime together and measure them in the AUC. And in that case, we see the addition of that new peak forming at about 7.5 s.
The exciting thing here is that we see this formation of that, higher sedimentation coefficient peak showing up, but we also see that the ratio of the 5 prime and 3 prime monomer peaks decreased by relatively the same ratio. And this helps us to, this helps us to identify that this is the formation of a heterodimer and not of a homodimer. Because If it was just the 3 prime binding with itself, you'd only see the 3 prime peak decrease.
So it was really exciting, and it helps show that the 5 prime and 3 prime UTRs bound together in a heterodimer formation, resulting in the higher sedimentation coefficient around 7.5 to 8.2 s.
We next then, they went on to do some MST.
With MST, what you do is you take… we took one of the RNA molecules and fluorescently labeled it and then serially diluted in the second RNA molecule.
Then you fill them into the capillaries showing up here on the screen. And as the sample, during the experiment, the capillaries are then, shot with a laser, which results in the sample heating up. This, increased temperature at the area where the laser hits causes the samples to move away from that heated area. And then as the… as the solution cools back down, the samples are gonna diffuse into it, and the diffusion rate is going to be different if the RNA is bound versus unbound. So using this, we're able to calculate a KD of our reaction.
So the first thing that was done was MST on the 5 prime and 3 prime UTRs mixed together, and it resulted in a KD of about .6 … .06 micromolar, which is a pretty good KD.
However, as I mentioned, there were a few other consensus sequence… sequences found within those 225 nucleotide region. So to see if the 11 nucleotides that we identified was the main driving force for cyclization, what we did was we mutated only those 11 nucleotides that they would not bind with each other, but we left the other 214 nucleotides untouched. So they still had their secondary structure and could still bind if they wanted.
And what that resulted in was a significant decrease in KD as seen in the green line here.
Here we've got… we went up to…our significant increase in the KD, meaning less tight binding, and it went up to about 7 micromolar.
So from these three results, we were able to verify the binding of the RNA using three separate techniques.
And then the exciting thing for AUC users is that we were able to measure the RNA in the AUC, and we were able to find that they gave different sedimentation coefficients due to their folding state.
And then we also used MST to determine their nanomolar affinity.
These RNAs worked really well in our AUC. As you guys saw, there were some let me just go back to here. There are some additional peaks shown here besides just the main one corresponding to most likely the dimer and potentially a small degradant of the RNA. However, this isn't always the case in the RNAs we've measured. If they tend to have lots of different folding states, you will see multiple peaks appearing. So we troubleshot this a little bit to try and figure out if it was…the RNA was aggregating or degrading or what it was. And we found that if you add urea to the sample to your RNA sample and heat it up and then measure it in the AUC inside of that urea, you can get a single peak if your RNA is pure, and that can help you to identify if your RNA is pure or not. It's kinda like doing a urea PAGE inside of an AUC, but it helps to tell you if it's just, folding structure that's causing those different sedimentation coefficients or if it is that you have contaminants in your RNA prep.
