Quantum-Si incorporated (QSI) Earnings Call Transcript & Summary

Hello, and welcome to this Nature Custom Media webcast titled, Precision Proteomics, Next-Gen Protein Sequencing for Proteoform Detection and Characterization. My name is Alison Halliday, and I will be your moderator. Today's webcast is sponsored by Quantum-SI. Proteins are central to cellular function, yet traditional proteomics methods, including mass spectrometry-based and affinity-based approaches, face challenges in resolving proteoforms, the distinct molecular variations of proteins arising from alternative splicing, genetic mutations and post-translational modifications. A new generation of single molecule sequencing technologies is emerging to meet these challenges. These innovative approaches offer single amino acid revolution and the ability to detect endogenous protein forms with high specificity. In this webcast, we'll introduce next-gen protein sequencing or NGPS on Platinum Pro and illustrate how it complements traditional protein detection methods. You'll learn how NGPS enables the detection of subtle protein variations, integrates with existing proteomics workflows and opens new possibilities in biomarker discovery and disease research. With us in the studio today is Gloria Sheynkman, Associate Professor at the University of Virginia. [Operator Instructions] And now over to our speaker.

Thank you so much for joining today. I am very excited to share with you work that our lab has been doing to implement up-and-coming next-generation protein sequencing methods for studying proteomic diversity. My laboratory is curious about how we can better characterize proteoforms or protein molecular forms in basic biology, but also the clinic. As humans, we only have 20,000 genes, but a gene can produce so [indiscernible] molecular diversity through all sorts of mechanisms such as alternative splicing to generate protein sequences that are different in their amino acid ordering. And this variation is quite massive in scale. Almost every human gene is alternatively spliced. And this is a typical example of a gene locus. This is tropomyosin alpha-1 chain. And you can see that the gene has many different exons as shown in these colors that are spliced in all sorts of different combinations. You can see skipped exons, even retained introns and alternative polyadenylation. And so from this one gene, you're producing upwards of 12 or more different protein isoforms that can have different expression across tissues and disease states as well as different functions. Now to characterize proteins, it is actually very difficult to use genetic -- There is a prevalence and readily read availability of RNA sequencing data. But if you know the expression of an mRNA transcript in your sample, it doesn't necessarily mean that the protein is expressed. There are many post-translational and co-translational regulatory mechanisms. So a transcript may be expressed, but it is -- that particular template is not effectively translated into protein, as you can see here, with a reference protein isoform above and then an alternative protein isoform that has an exon removed below. Furthermore, there are many other post-translational modifications that can occur in the milieu of a cell such as phosphorylation, glycosylation, et cetera. And all of these different molecular changes can lead to functional changes that could be very clinically important in studying disease or indicating disease states through biomarker analysis. And so what techniques are there available today? There are, of course, a variety of different approaches. The proteome is quite complex, physicochemically speaking, and we still don't have a grapple on the full extent of the complexity. There are 3 main methods that are being used at scale. So for example, antibody-based methods, of which there are many variations is if you have an affinity reagent, a probe such as an antibody or an aptamer, you can recognize the [indiscernible] direct binding or specific binding. For distinguishing these different protein forms though, it requires that the antibody is binding to a region of the protein that is different between multiple protein forms. And sometimes it can be difficult to raise antibodies with that proteoform specificity. So specificity towards the epitope that distinguishes 2 highly similar proteins. And other methods that are typically used include mass spectrometry-based methods. A very commonly used method that I will soon be talking about is bottom-up mass spectrometry or so-called shutdown mass spectrometry. And there, you would digest the protein and detect peptides. One of the issues with this method is that not all peptides can be readily detected as I will talk about later. And then lastly, a very exciting approach is something called top-down mass spectrometry in which one can detect, can ionize intact protein molecules and detect them actually measure their master charge at exquisitely high resolution. This technique is quite -- like all methods also has the limitation of throughput and sometimes it is very difficult to analyze really large protein forms. And so I'm going to now talk to you about bottom-up mass spectrometry because it is a very commonly used method and quite mature now. And so with bottom-up mass spectrometry, a typical pipeline is that you would take cells, tissues or whatever specimen of interest, like the cells to liberate the proteins, get them in solution. And then what's typically employed is a proteolytic digest so that the protein now has been digested into shorter peptide fragments. And it is the peptide fragments that we directly analyze because it is much more technically, analytically tractable to measure peptides versus intact proteins. The mixture that's typically produced from any kind of human sample or even eukaryotic sample is very complex. They're typically hundreds of thousands of peptides. And so it is very typical to use chromatographic separations in order to simplify the mixture. And so our lab uses nanospray or nanochromatography. And then the peptides get sprayed into the mass spectrometer, thereby collecting MS1 and MS2 spectra. And really, that's the fundamental unit, this MS2 fragmentation spectra, each representing 1 or at most 2 or 3 peptides. And a typical experiment is such that we collect sometimes millions of experimental spectra that represent a sampling of the peptides in your sample. And so how do we get identifications? The experimental spectra are compared against theoretical spectra that arise from a database. And so matches that are scored very high are more likely to be denoted as true peptide identifications. And so in summary, this is the bottom-up mass spec workflow. Like I said, you start with a sample, generate peptides, sample the peptides and then database search is done in order to get the full list of identified peptides. For the problem of distinguishing proteoforms, there are significant challenges to bottom-up proteomics. For general protein detection, all you need is 1 or 2 peptides that correspond to some portion of the protein from a gene. So what I'm showing you here on the left side is a case of 2 proteoforms and the green peptides below, as you can see, are mapping to both protein isoforms. So this is a really common occurrence with shotgun proteomics that peptides are ambiguously mapped to multiple proteoforms and you don't know or you're not confident in terms of the protein expressed in your sample. What is helpful in distinguishing proteoforms is when one detects peptides that are specific to sequence, such as these 2 cases on the right-hand side in which these pink peptides [indiscernible] to these alternatively spliced exons. And the best case scenario, of course, is where every proteoform has a signature peptide that can be detected. Now mass spectrometry-based proteomics is -- has been very, very I guess, increased or it's advanced quite a bit, especially in the last decade. And so I want to highlight that there are state-of-the-art, really deep coverage proteomic studies in this particular study from Josh Kohn's group at the University of Wisconsin, Madison, they were able to dramatically increase coverage of the proteome. And this was very exciting. However, they still could not detect all alternatively spliced protein isoforms that we believe are in the sample. And to kind of gain a sense of why that would be, if you're pushing a technology and you're still not really analyzing every component that you believe is in the sample is that it really is an issue of a needle in a haystack. So what I'm showing you here is an in silico digest. So these are in silico predictions of the peptides that would be formed from a human proteome. On the left-hand side, you can see that from the 20,000 proteins or 20,000 genes -- in silico digest, you would actually generate almost 3,000 peptides, so tryptic peptides. If you take a look at how -- at this cascade, though, we can start to classify the peptides based on how detectable they should be via mass spec, and we can start classifying peptides based on if they are mapping to shared regions of proteoforms or unique regions that would confirm the presence of a proteoform. So going back to those pink peptides that I showed you in the slide before. And if you trace that, look at these groups, you can see in the far right, that small pink box is the number of isoform-specific or isoform informative peptides. So you can see it really as a needle in the haystack issue. And you can imagine that if you have a really complex mixture, it would be hard to preferentially sample the isoform-specific peptides when it's in a sea of shared peptides that are not as informative. And so why would we not be able to sample every single peptide in a specimen. Well, one thing, first of all, I want to mention is no analytical technique, no protein chemistry approach is going to capture all of the physical chemical diversity of the proteome. The proteomics analysis is quite different compared to nucleotide sequencing, DNA and RNA sequencing. And specifically, there are known properties of peptides such as something called proteotypicity and different characteristics such as really negatively charged peptides or really long peptides can be really difficult to separate and fragment. And so what I'm showing you here is a deep learning-based model to predict how detectable a peptide can be. And so this term is called proteotypicity. And so if a peptide has very high proteotypicity, it is more readily detected by mass spec and will result in identification, whereas a peptide with very low proteotypicity, as you can see in the red here, has a lower probability of being sampled. And this is exactly what we see. I know these are predictions, but then when we are observing our real data, the peptides that we happen to detect do tend to have higher proteotypicity and there really is a large space of peptides that simply allude detection. And so with this issue that my lab but also the community has, I started to get very interested in technologies that could detect proteoform informative peptides using a completely different mode that has the potential for exquisitely high sensitivity, scalability, robustness, ease of use because the more tools we use, especially complementary and orthogonal tools, the more we will have power to cover the whole proteoform space. And so just a few short years ago, there was a publication published in science in 2022 that introduced this brand-new technology called next-generation protein sequencing. And this was actually commercialized a little over a year ago by the company, Quantum-Si. And how it works is that you do the same thing as what was -- what I mentioned with mass spec-based proteomics. You have a sample, you extract the proteins and then you digest the proteins into peptides. We have that same peptide mixture. But instead of introducing those peptides into the mass spectrometer and attempting to ionize those peptides in the gas phase, you would load these on a chip. This is a semiconductor chip, and you can see the wells depicted here. So you would load the peptides in a manner in which one peptide molecule will be immobilized in each of these wells. So we are actually talking about not populations of peptides, but single molecules of peptides. And then what occurs in the sequencing run is that a pool of what are called recognizers are actually in the solution. And over the course of a run, each of these recognizers will bind to the N-terminal amino acid. And that binding event is highly characteristic of the recognizer fluorescence. And it's also -- the binding is very characteristic in terms of its kinetics. And so that's what you're seeing on the far right-hand side, is a peptide that's immobilized in a well. And the first recognizer is recognizing arginine. After that recognition event, the recognizer diffuses away. There's an amino peptides that then cleaves that end terminal amino acid exposing the next amino acid for recognition by the next recognizer and so on. And so you kind of cycle through. And what you can imagine is one by one, you are reading out the identity of the amino acid from the N terminus to the C terminus of the peptide. And so these are kinetic signatures. This is a very sensitive instrument. And overnight, actually, over the course of half a day, what you can see are these profiles. This is a kinetic signature on the x-axis is time, and then you can see colored based on the type of recognizer binding that end to C terminal kinetic pattern that occurs. These signals can then get converted through various computational methods to convert the kinetic signal into the amino acid identity. And so here, you see a peptide DQQ, which was sequenced at the single molecule level. So I was very excited about this technology and excited to see that it has been commercialized. And so early last year, I purchased the Platinum instrument. And it was very interesting because I'm by training mass spectrometrist. And so we have a very complex thermo orbitrap eclipse instrument in the lab. This is a very different modality of detection. We got an instrument that is the size of a thermocycler. It was shipped. And then in just a few days, we got trained and we could run samples. Now one of the things that's really different about running samples on a next-generation sequencer is that you are actually loading all of the reagents. There's a kit, you load the reagents on a chip, and then you're adding the chip to the instrument, push a button and then the sequencing reaction happens and then you get the [indiscernible] software. So we asked the question after implementing or adopting this technology. We asked the question, how can this platform help solve the detection problem that I mentioned, wanting to detect peptides that are specific to proteoforms. There are a couple of notable features of this platform, like I mentioned, the exquisite specificity of that recognition binding, we could resolve single amino acid residues. And so could that be used to distinguish highly similar splice junction peptides. And then the other thing you might have realized is that the way that these peptides are detected is completely different from how we detect peptides via mass spec. And so I also wanted to know about the orthogonality. So the set of peptides that are more amenable to mass spec detection or more prototypic, how is that going to compare to peptides that are sequenceable on using next-generation sequencing. And so we decided to put together a proof of concept. We chose a gene family called tropomyosin or TPM. And this is a highly diverse family. So this gene produces many alternatively spliced protein isoforms, there are many paralogs, there are mutations that are associated with a variety of diseases, and there are post-translational modifications. And so we chose this particular protein. And the goal of the study was could we discern peptides that are uniquely matching to these different TPM proteoforms. Furthermore, we had actually found from a genetic screen. So this is a method that my lab has developed called long-read proteogenomics, where we're not only collecting proteomics data on a sample, but also long-read RNA sequencing data to define the transcript isoforms in a particular sample, including novel transcript isoforms. And then one of the purposes of using proteomics is to validate or determine if the proteins found with genetics or genomics data and transcriptomics data if they are indeed expressed. So for the pipeline, we first started by building a library of possible peptides to target. So what I'm showing you here are 4 different TPM2 splice forms. You can see that there are mutually exclusive exons. So the blue and the orange exons, one is included or the other, but never both. And then you have alternative last exons, so the dark blue and the purple are also mutually exclusive. We take those sequences. We perform an in silico lyse and/or trypsin digestion very similar to a mass spec protocol. And here, you see the peptides aligned to the protein sequences on a browser track. And then we filtered the peptides to choose those with desirable criteria. So a length that is appropriate, so an ideal length of 5 to 30 amino acids that peptides that have at least 3 or more amino acids that are recognizable. As you saw before, there were 6 recognizers that recognized a larger number of amino acids. And then lastly, we found peptides or chose peptides that end in a C-terminal lysine, which is needed for the library preparation. Now I mentioned that the peptide is immobilized on the well and requires a chemical conjugation event that [indiscernible]. So after we did this in silico digest and screening and selection of peptides, these were the peptides that we found. And each of these peptides, as you can see, are paired. So we asked the question, how can we distinguish different types of proteoforms. So you can see that there's a range of different proteoform types such as paralogs, so comparing TPM1 and TPM2 paralogs as well as tissue-specific splicing and phosphorylation. If you're interested in more details about this work, we have a bioarchive preprint that goes into more detail. But for now, I'm going to highlight the key results of the study. So in terms of the experimental design, we took each of these peptides, synthesized them and then subjected them to library preparation. So now the peptides are linked to -- via their C-terminal lysine. We then mixed these peptides together 1:1 and then subjected them to sequencing on the Platinum instrument. And so first, what we found is that we were able to distinguish isoform-informative peptides. And this is -- gives you a sense of what the data actually looks like. So I'll walk you through some of the details here. So for these 2 peptides, these are both -- they both start with TID. So these are 2 TID peptides. And they are almost exactly the same, except for the last few amino acids. And what you can see is that there are these pulse traces. So like I said, these recognizers, we're actually measuring the kinetics of the recognition event over time from the end terminus to the C terminus. And so that kinetic profile that I showed you is actually summarized and aggregated, and that's what you're seeing here is the average pulse duration of the recognition recognizer to amino acid binding. And so what you can see is that for these first few amino acids, isoleucine, leucine and glutamic acid, the pulse durations are very similar, but we were able to distinguish, for example, leucine versus valine here. It's the same recognizer sequence. So you may be thinking how can we distinguish these 2. But actually, the binding kinetics is very characteristic of one amino acid versus the other. So the actual pulsing speed, as you can see in this histogram here, is very different when the identity of the amino acid is leucine versus valine. And so we were able to very confidently distinguish these isoform-informative peptides. I'm going to give you a couple more examples. So this is an example of peptides that are very similar, but they come from 2 different genes, so 2 different paralogs. So there's peptide VIE. And what you can see is that there is a single amino acid difference, the serine to the -- this is the asparagine and glutamine. I always forget with QNN. So we could distinguish these 2. And then what you're seeing here, I want to emphasize that, again, this is coming from single molecule data. So each of these rows, each of these sequencing events is a different peptide. And so you can see there's a little bit of the stochasticity, but we have enough confidence through the analysis through database searching to ascribe the identity of the peptide to each of the kinetic signatures. Now I was talking earlier about some of the limitations with mass spectrometry. One of the limitations is that it is very hard to distinguish peptides that are the same mass. We get information about a peptide's identity through the mass to charge ratios of the intact peptides as well as its fragments. And so we had one case in which in this a peptide called ENA. So we had ENAL leucine and ENAI for isoleucine, it's the exact same mass because leucine and isoleucine are isobaric. And so what you're seeing on the right-hand side is we actually combine these peptides and then ran them via LC-MS. These peptides alluded at the same time. They have the same precursor mass. And when we fragmented this peptide, it generated an MST fragmentation spectrum that was completely indistinguishable. And so we couldn't distinguish these 2 peptides due to that small change. But then when we ran the same mixture on the Platinum, we saw a very clear difference between leucine and isoleucine because when the recognizer binding occurred in the fourth position, we could see that the pulse duration was very different. So even just that small change of the methyl group causes those binding kinetics to differ. And we have a lot of information or a lot of prior information and understanding of how these kinetics change. Lastly, I wanted -- I was very excited to see that we could even push the discrimination power to post-translational modification. So TPM2 had an annotated phosphorylation of a tyrosine, as you can see here. And so we took the TID peptide and synthesized a phosphorylated version of it, combined the 2. And what you can see is the signature, the pulsing signatures were identical for the surrounding amino acids. But when comparing the phosphorylated versus unphosphorylated tyrosine, there was a very -- there's a big difference in the pulsing patterns. And if you think back to how the recognition works, the recognition -- recognizer protein is almost cradling. It really is binding that end terminal amino acid. So if there is a presence of another chemical modi such as a phosphate -- a phospho group, then you could imagine that could ablate binding or really change the binding properties, which we can measure on the instrument. So I kind of told you a little bit about that. If you want more information, you can check out the bioarchive paper. And then moving forward, some of the limitations of the study is, as you probably noticed, we conducted this study on simple peptide mixtures. Really, this was our first project and proof of concept. And what we're really interested in is demonstrating how this platform can analyze more complex mixtures and to demonstrate the applicability of proteoform detection in more examples. So some of our next steps is that we are endogenously expressed proteins. We are also determining the ability to sequence full proteins. And just to touch a little bit on how we are doing this. Like I said, our lab collects long-read RNA sequencing data. So for our sample, we first have some models, some prior information about what protein isoforms could be expressed in healthy and disease states. We do the in silico peptide digestion and kind of a priority have list of peptides that we would like to detect on our platform. And so current work in peptides that we would like to detect on our platform. And so current work in [indiscernible] based, so immunoprecipitation. And I want to just give you one example of a protein RAB11B. We're optimizing the protocol. And so here's some initial results where we took recombinant RAB11B, digested it with lyse and then we were able to get coverage across the protein, as you can see here. So each of these bars represents a peptide and then you can see the number of independent sequencing events. So each of these hits represents a single peptide well -- a single peptide in a well that was sequenced. And in particular, we're very interested in different proteoforms of RAB11. What you can see here is that there are 4 isoforms of RAB11B. And one of the peptides that we found this NIL peptide is actually isoform specific. So we're very excited to use this ability to measure this peptide to discriminate isoforms of RAB11B in cellular cell-based samples. So like I said, we are using this targeted approach to sequence endogenous proteins. I have 2 more vignettes that I would like to share with you. And so this is actually not work from my lab, but work from Neil Kelleher's lab at Northwestern. And the Kelleher lab is one of the leaders in top-down mass spectrometry. And here, they were interested in determining how next-generation protein sequencing could complement top-down mass spectrometry. Like I said, with top-down mass spectrometry, you are measuring intact protein molecules. And it really is the -- if I would say so, one of the only techniques that can actually give you that proteoform resolution. [indiscernible] was Interleukin-6. This is a very pleiotropic cytokine. It plays widespread roles in the immune system and different inflammatory processes. It is also highly diverse molecularly speaking. There are cleaved signal chains, disulfide bonds and then many phosphorylations and glycosylation. So this is a very interesting target to analyze, and there are also FDA-approved drugs to -- that bind to IL-6. And so when the Kelleher lab digested and sequenced IL-6, this is what they found. So this is from the Platinum instrument. And what you're seeing are all the theoretical lysine digest peptides. And in the colored amino acids show what could be recognized by the recognizer panel. And wherever there is a red star, that peptide was successfully identified. And so there were these 5 peptides that covered IL-6, and they are reshown at the bottom of the slide. So you can see QIR, EAL and these peptides have very, very good coverage across the NTC terminus. When this digest was analyzed via top-down proteomics and actually this new approach that you can read about if you're interested called I2MS, they also detected intact protein molecules as shown in the mass spec trace to the left. Each of these peaks represents protein form ion. And you can see multiple peaks that correspond to proteins that are modified by phosphorylation, glycosylation and such. And then what you are seeing here in B is the full sequence of IL-6. And wherever there is a line, that represents a case in which the fragments were localized or the fragments were sequenced. So you can get a sense of the coverage across IL-6 using top-down mass spec. There are different regions of IL-6 that are more hydrophobic and hydrophilic. I mentioned earlier how physicochemical properties of peptides really influence what can be detected. And so what's highlighted here are the peptides that were detected and the fact that these peptides are associated with variations that are very biologically relevant. And so with this project, overall, there is a demonstration of like the complementarity of using Platinum as well as mass spectrometry. We did show that it can provide single amino acid resolution. And then when combining these tools, you increase sequence coverage. And so again, it's a very complementary approach. So future directions for their lab is to characterize different IL-6 forms in additional systems and really distinguish the various like truncations, amino acid variants and post-translational modifications and really start to study function. Okay. I'm going to end with one last application of next-generation protein sequencing, which is protein barcoding. This is not something that my lab is involved in, but it is something that is another very interesting application. And you can take a look at the bioarchive paper for more details. But essentially, what one can do is fuse protein or actually peptide barcodes to a protein of interest. And that particular peptide that is fused to your protein can be cleaved off and measured. And so you may be wondering why would you want to do this? Well, the particular peptide that is fused to the protein actually acts as a tag. And so the amount of peptide barcode that you measure is proportional to the concentration or the amount of your protein in your sample. And so this is a really nice way of starting to think about multiplexing your protein analysis. And so there are a set of barcodes that are very ideal for very well detected on the Platinum on the Quantum-Si instrument. And so here's an example of a barcode design that you could possibly do in which you fuse your protein to the barcode and [indiscernible] clever library preparation where essentially -- when you input the protein lysate, the protein sample and then the [ sortase ] tag will actually help conjugate the appropriate library linkers and left with that single peptide barcode. And so one of the applications for this is to screen lipid nanoparticles, so you would express the various protein candidates that could render LNP more or less effective. And then instead of having a DNA readout or RNA readout, you would actually be measuring the amount of effective protein that has been successfully delivered and expressed in your target cells of interest or target tissue of interest. So overall, hopefully, I've given you a flavor for how next-generation protein sequencing is a very accessible technology for protein analysis. It really is fundamentally different. It's almost like a paradigm shift compared to maybe what you are familiar with regarding Western blots, antibody-based measurements as well as mass spectrometry. It really is in a category of its own. We're exploring it for proteoform analysis and very excited about the applications -- additional applications that are coming up. I really want to thank you for joining the webinar, and I'm pleased to -- are very excited to answer any questions you may have. Thank you.

Thanks very much, Gloria, for that presentation, absolutely fascinating. It's now time for our live question-and-answer session. [Operator Instructions]. So we do have a question to get started today that's come in. So Gloria, bear with me, this question is quite long. Have you or anyone attempted to discriminate between different post-translational modifications using this platform? And they've added if the effect of a post-translational modification is strictly steric and/or electrostatic and they can envisage that the ability to discriminate may be low, but if there are kinetic differences between how the recognizes interact with different post-translational modifications, perhaps different chemical moieties could be distinguished. So that makes sense.

That's a really great question. Let's see if I can unpack it. So I think that the person who posed this question is really asking what's the potential and capability of this platform for discovering and distinguishing post-translational modifications. I just showed an example, some preliminary data a couple of slides ago that showed we were able to show like ablation of binding when there was a presence of a phospho group. And so one can think of that as like steric -- steric hindrance, that recognizer is no longer recognizing the amino acid. One of the things that I want to mention is I did kind of oversimplify a little bit, and there's actually much more information. The information is very rich, what you get out of the -- after a run. I don't know how well it was displayed, but these kinetic signatures are really, really precise and they're very context dependent. So in the figure, it actually shows a little bit better than what I was saying that the recognizer not only binds the end terminal amino acid, but is also influenced by the proximal amino acid. So the amino acid in the adjacent positions slightly -- doesn't change completely, but it slightly affects the binding kinetics. And these kinetics are -- we're starting to understand the relationship between recognizer binding and what amino acids and what post-translationally [indiscernible] modified amino acids there are in that vicinity. And we're starting to see characteristic patterns. I think the readers intuition is absolutely correct that for -- because we don't have recognizers that specifically recognize these post-translational modifications, at least not yet, most of how we discriminate or determine the presence of a PTM is by lack of binding or tiny shifts in the context in the -- what I was mentioning, the context-dependent effects. While I'll mention that beyond phosphorylation, there is also work and nice data to show that we can discriminate citrullination. And there's no reason why additional PTMs couldn't also be distinguished.

Fabulous. Thank you, Gloria. That's really thorough. Our next question is, can you please elaborate the major difference between LC-MS and MS-based sequencing with this?

That's a great question. The major difference I would say -- the major difference is that to measure a peptide via mass spectrometry, you must convert or you must allow that peptide or drive that peptide to go from the mobile phase, which is liquid into the gas phase. That's called ionization. And there are many different methods for ionization, Nobel Prize winner, John Fenn with electrospray ionization, but then there are many other modalities, especially like spatial proteomics and [indiscernible] how well you can drive that peptide into the gas phase is like a major determinant of like sensitivity. The second kind of major determinant of how well you can detect a peptide is the patterns of fragmentation, like how it breaks when you're trying to -- when you do -- when you collect an MS2 or fragmentation spectrum, like how well does it break? And how useful is that fragmentation pattern? How characteristic is it of that peptide. So I think those are the 2 things, getting into the gas phase and fragmentation. For Quantum-Si, what is different? It's a completely different way of measuring things. First of all, there's no trying to get the peptide into the gas phase. It's a lot like to me, DNA or RNA sequencing. The peptide stays in a mixture, you basically add this cocktail, you add that cocktail put it in the chip, put that in the instrument. And like I was mentioning, it's all about the sequential binding from the end to the C terminus and what those patterns are on an amino acid by amino acid level.

That's a great question. That puts you on the spot. So the next question is, if I understand it correctly, this is what the person is asking, it has not been tested in the lysates from cells or any organisms. So if that's correct, can it be used for that?

What was the last part of what you said?

So they're just asking whether or not it's been tested in lysates from cells or any organisms. They're saying, they're assuming it hasn't. And if they're correct in that assumption, can it be used for that?

There -- it has been. So probably it's apparent that right now, we're using a targeted approach where you have a simple or moderately complex mixture in which there -- the protein of interest is either highly abundant, moderately to highly abundant or you have enriched that particular protein. And if I remember correctly, I think there are some white papers to show that -- I'm not going to remember, but there was some serum studies done in which a high abundant protein was detected from blood. And there are -- this is unpublished, but we are enriching for the proteins, and we have successfully -- we have some preliminary data to show that we've detected that. So really, it's a situation where it's really biochemistry. If you can get the protein in sufficient concentrations and remove and not have too much background, like we're there. Many groups, including ours, are -- that's what we're working on today.

Thank you, Gloria. So we've had a question for someone who's also given you thank you. They said thanks for the informative talk. The question is, can you elaborate on the throughput of the instrument and what are the limits of detection in upper and lower bounds in terms of peptide number/concentration that can be captured in a single run.

So each chip, so you can run 2 samples per run. I didn't talk about the numbers, but each chip will have, like I said, 2 flow cells. Each flow cell will have 1 million apertures or wells, as I was calling them. And then you could see that we get anywhere from -- not every well will be used because, again, it's like diffusion-based loading, a lot like these other approaches like PacBio sequencing. We get anywhere from 10,000 to 100,000 different sequencing events, like so I kind of think of it as a peptide like sampling event. And then in terms of the [indiscernible] one of those chips overnight. So if you really want to max it out, you could probably do 4 samples per 24-hour cycle. And then I think the other part of the question is what is the complexity of the sample that could be analyzed. And so I would say where we're at is maybe up to a dozen, a dozen different proteins is what has been tested. And so we're kind of in that like moderate throughput. One thing I might want to mention also is that it's changing quite a bit, like the Quantum-Si has been putting a lot of effort into R&D... So when we first started the project, we could only distinguish maybe like 1 or 2 peptides, but then we got new kits with lower input requirements. And so we're kind of in that -- I think that positive slope. In terms of the sensitivity, I will need to ask my student because we do a lot of back of envelope calculations, and I don't think I'm going to get it quite correct. So I will need to pass on that. But perhaps afterwards, we can provide material because you can think of sensitivity in so many different ways like copies per cell of the protein, micromolar concentration, et cetera. I don't know if anybody can jump in. I guess I might be the sole person.

Okay. We'll park that one. And hopefully, we'll be able to get back to that [ acquirer ] later on after the seminar. And we have another question. It says, is this Platinum instrument truly revolutionary in proteomics from your current search perspective?

Is it truly revolutionary?

That's what they said. And they said, is it the next step evolution in your research?

To me, I'm a technology person. So I think like when I see new technologies that are completely different and not incremental, although incremental is important because you need to, I think there are different stages of technology. So I trained with Lloyd Smith, and he took Sangers DNA sequencing, those were slab gels and he thought of the idea of how to automate that process. So my thought on the field is we [indiscernible] it has started. We're right at the -- I mean, if you look at these papers, look at where we're at, it's only been 1 or 2 years. So I do -- my belief is that seeing where mass spec being an analytical chemist, seeing how technologies have developed over the last few decades, it was unimaginable what we could do with mass spec-based proteomics 10, 20 years ago. And so I definitely -- I think that it will be revolutionary and that these type of really creative like different approaches and having more people be involved in the process. And because there also is the question of what other applications could there be. So there are some applications that we were coming up with that we never thought about because we were in the mass spectrometry mindset. And I think that each of these platforms, it's going to be so different. This is a single molecule sequencing technique. So it is theoretically possible to get down to single cell resolution. Is it happening right now? No. But I'm a very -- I'm a big technological optimist, and I think we're going to see some really exciting things to come.

Super. And you probably just answered this question just there, but it's a nice one to segue on to. What do you think the impact of NGPS will be on the field of proteomics and they put a time line on that in the next 5 years. But you can go beyond that if you want.

In the next 5 years?

Yes.

I think that -- if I were to say one thing, -- we haven't had the ability -- if you really think of what we've had, we've had Western blots, antibody-based detection, and we're still -- it's a very robust process, but we're still doing a lot of Western blots, right? And anybody doing proteomics need -- I mean, I've been trained in mass spec-based proteomics for over 15 years. But now we have something that is completely different, and it's almost like a thermocycler. I'd like to use that term because [indiscernible] people who are not like me. I'm a technology geek, really, really analytical, really into all of these details. But I think what the impact will be is that it will maybe more democratize protein measurements. And the way that you measure proteins will be different. And it won't be like running a gel and looking at a band but it will be single amino acid resolution. So we're going to get to the point where you might not know the identity of the protein in your sample, but you can just go ahead and sequence it. You don't have to send it to a core, you have an instrument on your bench. So I think it's going to maybe change the behavior of like how instrument -- how experiments might be designed.

Thank you. And we've got one last question here. You've already touched upon this sort of going after that moderate throughput, but someone is asking whether it can be used for high throughput protein-wide studies? Or is it best suited to those targeted applications? And I guess it would be good to sort of see [indiscernible].

Yes, it's best suited for the targeted application. So like I said, going back to the complexity. So I would say it's moderate throughput. So you need semi simplified mixtures. And so you need -- for your application, you want to think about what the question is. If you boil it down to identify, discriminate or quantify a certain amount of protein, I guess, in our case, proteoform. And actually, my student came through for me in terms of the sensitivity for sequencing a peptide, it's as low as 500 fentanyls. And then for the barcoding because they really optimize those peptides, they're really ideal for sequencing, you can go as low as about 400 [ fentanyls ] in an 8-plex mix. So hopefully, that answers the earlier individual's question.

Fabulous. Well, thank you so much today for your presentation and for answering all of our questions. That's all we do have time for today. So I'd -- obviously like to thank you again. And I would also like to thank our sponsor, which was Quantum-Si. And of course, you, the audience for tuning in today. Remember, you can watch this again any time on demand at www.nature.com/natportwebcast. Thank you for watching, and I hope you can join us again soon.