Scientist Stories: Albert Heck, The Uniqueness of Our Plasma Proteome

So as Bailey said, now it's time for the main talk of the day, and it is my honor to introduce Professor Albert J.R. Heck, Dutch scientist at Utrecht University in the Netherlands. Professor Heck received his PhD from the University in Amsterdam and then moved at Stanford University as a postdoc postdoctoral fellow. And then he was a lecturer in the— in Warwick University, still in the UK. Professor Heck and his group are regarded as pioneers in interactomics and protein-protein interactions, known for developing new techniques to investigate protein complexes and protein foldings, with a particular interest in the immune response and the proteins that are relevant for the for the immune response, especially proteins that are relevant for the antigen-recognizing pathways. Professor Albert Hecht, the stage is yours. Thank you very much for this introduction. And it's a great honor to be speaking here in front of you, a big crowd and all people that are eager to start a career in proteomics. For me, that was quite some time ago. So I wish I was in your place instead of here. Because it's really an exciting time to be in proteomics. It's a real great field. So I'm going to talk today about two stories from our lab, and I hope they are of interest for you. You can also relax a bit because it's not really about MaxQuant. Maybe when— and I'm very happy that Jürgen invites me sometimes to these lectures because I hope then to tell him, oh, we need you to do this for MaxQuant now. So don't Today I will also present some things where I hope that MaxQuant next year will maybe already be able to solve some of the problems we are trying to tackle now. But I also really want to start thanking Jürgen and his team for MaxQuant. It's really been a great tool for many, many years already. Yeah, so in my group we have been working on a lot of alternative things. So of course, when I discuss with Jürgen, he says about normal proteomics, and then he tells that we do abnormal proteomics. And that's partly because I also like to explore, for instance, other proteases, other fragmentation techniques. Very much in the beginning, Jürgen and his team really helped us when we did alternative isotope labeling with dimethyl labeling and even 15N labeling. And indeed, MaxQuant can also do these things. But also with the HLA peptides, first when we started to get bigger and bigger data sets, then MaxQuant also came to help. Now, MaxQuant is not very good yet at glycoproteomics, so write that down, Jürgen. And also top-down proteomics is also a field that is emerging, where I know that Jürgen and his team are interested in, but also their software is still lagging behind quite a bit. So as I showed on the previous slide, I have two stories for you. One about protein histidine phosphorylation. I will start with that. The story is also about how you think everything in your PhD fails and you still can get a nice publication out of it. So, so maybe this is also a good story to tell. And then I want to talk about the immunoglobulin, maybe a new term for you, but we get very much interested in what sort of antibodies you and I are making circulating in our blood. And how we can also use proteomics to study those. So, I will start with the histidine phosphorylation. And as mentioned this morning, when you do phosphoproteomics, I don't know how many of you do phosphoproteomics, in a human system, you typically search for serine, threonine, and tyrosine. And they are also likely the most abundant. But there are maybe some other amino acids that can also be phosphorylated. And One caught our attention already many years ago, and it's called histidine phosphorylation. And it's actually quite interesting because histidine is, of course, a relatively strange amino acid. And you can even have two different isomers of histidine phosphorylation depending on which side in the ring the phosphate group is attached to. But maybe you have never heard about it. And maybe that's good. I'll show you at the end.. But it's also known that you can have histidine phosphorylation, but relatively to serine and threonine, which is also already a label modification, on histidine it's even more labeled. So, it falls off quite easily. And especially under standard condition that you do your phosphoproteomics, the phosphate of the histidine falls off. And this has to do with the temperature you may be using, also with the pH that you are using. So, you haven't heard a lot, I think, yet about phosphohistidine phosphorylation. But a couple of years ago, spearheaded by Simone Lemire here and Clément Potel, we set out to see, can we really measure histidine phosphorylation in samples? Now, the message was that histidine phosphorylation was maybe quite important in bacteria. So we started an approach that you see a bit outlined here. Instead of using very small columns, we used a huge column, an iron IMAC column. We can load milligrams of sample on that. That's not what people do nowadays much anymore. But then in the end, you see this huge peak in the gradient at the front, which is all the boring peptides. And then this small peak that you see at the end, that is supposedly the phosphopeptides. And by doing this with this column, we can really get clean LC-MS chromatograms. And we can get out of 10,000s of phosphorylation or phosphopeptides. Now, but then, of course, most of those are serine-threonine phosphopeptides. So, what can you do to really go for histidine phosphorylated peptides? There are several things that are quite interesting because the modification is quite different. You can use specific parts in the fragmentation spectra that are indicative for histidine phosphorylation. And Clément and Simone worked on that a very long time. So, there is an ammonium ion of histidine that you may know, but there is also a phosphohistidine ammonium ion. So, you can use that. And if you see that in your MS2 spectra, you're pretty sure that the precursor must have had a phosphopeptide in it. Now, you can also use multiple fragmentation steps, and especially with electron-induced fragmentation steps like ETD and ETHCD, you actually keep the labile phosphate on the peptide, and you get better backbone sequence information. So, all this was used also to trigger phosphopeptide identification. And Clément and Simona developed a method that worked pretty well. And the results were a bit astonishing. And I hope you can read the pie chart over there, that when we did this in bacteria, overall, we find not that many phosphorylations because there is less phosphorylation in bacteria. But we found about 1,500, 1,600 phosphosites. And if you read the pie charts, then still half of them are serine, maybe a third of them are threonine.. But actually, the histidines, there are about 250 of them, and even more than tyrosine. So, histidine phosphorylation in bacteria seems to be quite real and seems to be very well detectable and seems to play a role in signaling as well. And of course, that's why it's interesting. And we are now following up in studies where we can see if we can maybe find specific kinases for the histidine phosphorylation, but maybe also specific phosphatases. And specific inhibitors. So this is a bit an old story, and, and, but of course, when we had this method ready, the big story was, okay, if there is histidine phosphorylation in, in bacteria, is there also histidine phosphorylation in, in mammalian cells? Now, there were some big stories out there also from big names in the field that said, yes, histidine phosphorylation in mammalian systems is quite prevalent is quite important. And even there were groups that had developed specific— and you have tyrosine, phosphotyrosine-specific antibodies— they actually generated phosphohistidine-specific antibodies. And they developed those, and they pulled down with these antibodies supposedly phosphorylated proteins that they analyzed, and they found many, many substrates. And This was quite remarkable. And you see that this was remarkable because there were several Nature and Cell papers about this topic. So, then I have to introduce Niels, poor student, about 4 or 5 years ago, he started his PhD. And we said to him, Simone and I, you have to study mammalian histidine phosphorylation. And we have a method already, which is based on this tyrosine, on this bacterial phosphorylation. So he copied this full method that we used for the bacterial phosphorylation, and the first results were very exciting. This is almost like every PhD project. He came in after a year and he showed this data to us. And here you see the pie charts again, and in black are the serines, in grey are the threonines, then in a color that you almost cannot see, light grey, tyrosines. And then in red there is the histidine. So here we followed the workflow. We had 4 different mammalian cell lines. Our numbers are remarkably high, you know, in total about 25,000 sites per cell line. And so he came and he said, I have thousands, yeah, thousands histidine phosphorylation sites. And we were very excited. But then luckily he and Simone started to look very critical at their data, and this is what you always should do. So one of the things that we realized quite quickly, if we use MaxQuant and if we use localization scoring, we saw something peculiar. If you do stringent localization scoring, you lose quite a bit of your sites. But you see here in B, you see how much sites are still remaining when you do a more stringent Class I filtering of the localization.. And then you see in the black-gray bars, you see serine, threonine, and tyrosine. And with serine, you lose about 20% of your sites. With threonine and tyrosine, you lose about 50% of your sites. But with the histidine, it was already remarkable. We lost 70% of the sites. So, going from all the sites to class I sites, the histidine phosphorylation lost quite a few numbers. So, this triggered us to say, are they really, really histidine phosphorylations? And then we started to do all kinds of experiments, and some of those I want to show you. We did an experiment which is known that if you acidify a sample, then you lose your phosphorylations. But luckily, in serine and threonine and tyrosine, this is relatively slow. In histidine, this is relatively fast. So we did this experiment with bacterial phosphopeptides and with human phosphopeptides. And indeed, you see that on the left here, you see that in the bacterial phosphopeptides, we lost almost all our histidine phosphorylated peptides by doing this acidification. Well, when we did the acidification on the mammalian cells, we almost didn't lose any sites. So again, that triggered us to tell, are they really histidine phosphorylated sites? Now then, in the middle, and this is something I really liked, we said let's look at the localization probability of all the sites. And these curves are also output of MaxQuant or other search algorithms. And you see there the curves again of serine, threonine, and tyrosine in the colors of black to gray. And then you see in red, you see the localization probability score for histidine. So for all these histidine sites, the probability score is way worse than for serine, threonine, and tyrosine. Now then we did an ultimate experiment where we didn't have to do any experiments for. We said maybe let's not look for histidine phosphorylation, let's look for proline phosphorylation. Now of course proline phosphorylation is not possible, but please do it one of these days and tell me how many proline phosphorylated peptides you find., and you will find a lot. And this really clarified to us what is the real problem. Now, the real problem turns out, if you look at serine phosphorylation in human cells, then the adjacent amino acid is 50% of the time a serine. So if you don't find the proper B or Y ion or whatever ion you want to see that tells you that it's on the cleavage between the serine and the proline, Actually, no search algorithm can tell you if the phosphate is on the proline or the serine. Of course, biochemically, you know it's not on the proline, so you're safe, and you know it's a serine phosphorylation. Now, it turns out that if you look at the human genome, that many, many histidines are in regions where serines are adjacent to them. So this turned out to be the big problem. If we now allowed not only the first hit, which said, okay, it's a histidine phosphorylation, but actually looked at the neighboring amino acids, we could clearly see that all the assigned histidine phosphorylations were actually better— no, not better, they didn't come out better, but their second match was very close and was actually a serine phosphorylation. So basically, the take-home message was we cannot detect any histidine phosphorylation in mammalian cells. While we started with thousands of them. So, this is sort of the summary of it. And what I like at the bottom, if you don't find any sites, then people also don't believe you. So, luckily we found 4 or 5 sites that you see here at the bottom where there is histidine phosphorylation in mammalian cells. But this histidine phosphorylation is actually part of an enzymatic reaction. It's an intermediate in an enzyme reaction. And so, it's not done by a kinase. It's just done by the system itself. So, basically, you see here on the left and the right, we started in human cells with way more supposedly histidine phosphorylations. But after doing the class 1 filtering, after doing the acidification, after looking for the imonium ions that I told you about, we were left in this particular experiment with only 2 sites. Well, on the E. coli side, we started with maybe 150 in a single experiment. After the class I site identification, we were left over with about 100. Acidification didn't do there any— also a bit less. But we also found quite a bit of imonium ions. So, with all this, we are pretty clear that there is almost no histidine phosphorylation in human cells. So, I hope that no one of you is really received the task to start histidine phosphorylation studies in mammalian cells, or prove me wrong, that's also good. But I think, but I also want to tell this story because I think this site localization is still something that we not completely understand. So, always stay critical about your phospho datasets, or any PTMs that you want to study. So, now I move to the second part of my talk. And since a few years, we are starting to doing serum or plasma proteomics. And when you start to do that, and I don't know, I would be interested to hear how many people do this here, you always see this pie chart. And this pie chart tells you which proteins are there in your serum or blood. And it's actually quite disappointing because about half of the protein mass in your serum comes from albumin.. And also about a quarter of the molecules in your blood come from IgG. So, if you really want to go and do plasma proteomics and you want to go deep to find biomarkers, what people typically do is they first deplete these proteins and quite a few more. Now, we turn actually this around. We said, yeah, but what is this IgG actually? And can we maybe start to look at that in detail. So, that brings me to a couple of questions that you see here on the right that I want to talk about. So, how many proteins are there in serum or plasma? There is sometimes a debate about that. And then people say, oh, I detected more than you. The next question is, how do you define a protein? What is really— if you want to answer the question, how many proteins are there in plasma, you also have to define what is a protein. Now, the question I want to address in this IgG— actually, these are different proteins, they're all called immunoglobulins. So how many of these immunoglobulins are there in our blood? And these other questions I will not come to. So this is the question I want to ask, and therefore I go a little bit first again to numbers. So these amounts are enormous. So we have about 30 grams of IgG molecules in our blood. Which is about 10 to the 20 molecules. And they are also supposedly, you know, you're making new ones every 2 or 3 weeks of them. So, your body does a lot of these things. And of course, these immunoglobulins, I think you know, they are antibodies. We make them to attack or to prevent us from attack from pathogens like viruses and bacteria. Now, here's the difficulty. And also, yeah, what maybe made that no one studied this. So how many immunoglobulins do we have in our body? Now, they are made by B cells, and how they are made is that there is an— and actually here you see a picture of the antibody on the top. The antibody is a big molecule, it's about 150 kilodalton, the complete antibody. And the parts that you see here at the bottom part, the tail at the bottom, is actually what we call the Fc part, the constant part. And this is an amino acid sequence that all the antibodies almost share. So in all the antibodies that our body can make, this is the constant part. All have the same sequence. But there's also a part in the antibody that actually recognizes the antigen, and this is called the variable part. And this variable part you see here at the bottom. In this variable part, there's both an antibody light chain and the heavy chain. You have these regions, and they're called they're called the CDR regions because they recognize the antigens, and this is where all the action takes place. So even in the sequences at the bottom, everything you see in black, which you probably cannot read, is constant, has the same amino acid in whatever antibody you look at. And the things that you also don't see but that are in light gray and white is where all the action happens. This is where all the variation happens. Now, how many antibodies can we make? Now, it depends on how many B cells we have. Now we have a lot of B cells. These are just estimates. So we think that we have about 10 to the 11 different B cells in our body. So this means that we can make 10 to the 11 different sequences. These sequences are made by what is called recoupling of all the genes that make up the antibody. And you see that a little bit depicted over here. So this repertoire of B cells we have, means that we also have this huge repertoire of antibodies that we can make. But which of these then end up in your circulation to these high levels that you see in blood is something that we don't know. But of course, if you think there are 10^13 different sequences, yeah, this is not a plasma proteomics project you want to start on, because how can you ever distinguish all these different sequences by mass spectrometry? Now, just to show it a little bit more, and this is really a problem that I hope Jürgen and his team, and maybe you as well, will solve. This is our problem. We are sequencing, sequencing, sequencing, and again, 90% of the time we are sequencing a peptide that they all have in common, and it doesn't give us any new information. When we are trying to sequence the regions that cover these CDR regions, yeah, then there's always a hit, but likely there are many more hits because they all look almost identical as well. So to distinguish all these antibodies by a bottom-up approach is almost impossible, especially because you're sort of— and that's from my question yesterday— you're, you're trying to sequence antibody protein sequences that are 95% identical. So these mutations, they are only happening in the CDR regions. This is also called the FAB region, the antibody-recognizing part. And this is what we want to sequence. We don't want to sequence this FC part because that's the same for any, any of the proteins. So how do people do it now? Now they try to get as many B cells out of your body as possible, and then you can use expensive and elaborate but very nice tools to sequence all the B cells. And from the B cells, you get B cell receptors. And if you have that information, you get a glimpse of how many antibodies— B cells you have and what their sequences are. Now, there are several problems with this. You have to get the B cells that make the antibodies. Now, that's quite difficult because they are mostly in your lymph nodes, and I think there's no volunteer who wants to give some of their lymph nodes, B cells, to, to study. So mostly people take cells from blood and sequence then the B cells that are circulating in the blood. Now, in the end, if you do an experiment like this and you do it very well, then from a person you can have about 2 million different sequences. And I'm pretty sure that in a couple of years this might be 10 million or maybe even 100 million of sequences. So that could be your database to search your proteomics data against. But again, that's quite big. It's bigger than a human genome database. So in our lab, we, we just said, let's try it. We're just going to do it. Don't look at these numbers. We're just going to do it. So the nice thing is there's a lot of IgG in your blood. So we do an affinity resin to capture all the IgG from our blood, and that you see schematically here.. And then we use a protease that actually cleaves. And, you know, you see the Y-shaped antibody there in the middle that cleaves above the hinge region in the antibody. And then we lose the part that is constant because it remains to the bead, and we're not going to sequence that. I don't want my mass spectrometer to look at things that I know already. So we're only going to sequence the top part, which we call the FAB fragment. So it's formed after digestion with this protease. And we're going to do something that, that you haven't seen so far yet, but I hope you will do more. We're not going to digest this first. We're just going to bring this FAP molecule directly to the mass spectrometer and measure its mass. Now we roughly know what the mass is of such a FAP molecule. It's about 45 to 50 kilodalton. And of course, if it has a different sequence, it has a slightly different mass. If there's a few mutations, the mass of each clone coming from each B cell will have a slightly different mass. Now, the idea was if we did do this, we would maybe get 10^9 to 10^13 different FAB molecules to the mass spectrometer. But being in proteomics, you, you might already guess that this was not the case. So when we brought— and we capture all the IgGs from, let's say, 25 microliters of your blood, we bind them to the beads, then we cleave off the FABs,. And these FAPs we analyze by LC-MS. And this is actually the LC-MS trace. It almost looks like a BSA digest, but it is really a FAP LC-MS trace. And now I have my most complicated animation. Here we go. So we walk through this and we measure the mass of each ion that comes out of the LC-MS, just like you do normal LC-MS. And it's a bit boring, the spectra you see. They are all charge state distributions of molecules of about 45 to 50 kilodalton. And we go through this trace and we see all these different clones with slightly different masses coming through. Now, what tells us directly, it tells that our FAB profile or antibody profile in our blood is highly dominated by a few abundant clones. It's almost the same as the plasma proteome, a few clones make up, let's say, 80-90% of all the antibodies in your blood. But we have now all their masses, and we can say now, and this is an estimate, if something has a specific mass and a specific retention time, we call it a clone. And each antibody clone comes from one B cell. So that's then one antibody identified. Now we can actually convert this to what we call a mass plot.. And here we see all the FAP molecules in a specific donor, and we see their mass, and we see their intensity. And this intensity we can measure because we spike in a recombinant antibody, and we can really say, okay, of this specific molecule of about 4,600, 5,000 Dalton, you have about 40 micrograms per milliliter in your blood at that moment. Now, this profile looks very simple, and so the whole method is now in place. So now the next thing was, can we predict if Jürgen's antibody profile looks like mine or completely different? So that's what we did. And we had 2 donors. And of each of these donors, we were able to get blood at 3 time points. Now here you see what we call now their antibody repertoires. And this is all only IgG1 repertoires, I have to say. And you see that these pictures at the top shows something remarkable. So if you have an antibody profile and you are relatively healthy, then your antibody profile is the same a month later and 2 months later. Now this is already quite interesting because the lifetime of every antibody is only 2-3 weeks. So you actually have made all your antibodies again, but you know, the profile looks stable.. But then the person at the bottom also has a very stable antibody profile with a few clones being very dominant. And these dominant clones, they also don't change over time, but the profile is completely different from the person at the top and at the bottom. So we looked at that and we showed that in many different ways. So the pie chart shows how many clones we identified. So we detect in this particular run— these were early runs— about 500 different clones, so different masses and retention times. And I think in the colored ones you see here how much the top 25 or the top 30 makes up of the total LC-MS chromatogram. So the top 20 or top 30 make up about 70% of the total immunoglobulin. So we make them pictures there on the right. We do a correlation matrix. And you see that the correlation over time within a certain antibody repertoire of a person is about 95%, while the correlation between the antibody profile of Juergen and myself is nonexistent. It's really 0%. I think these numbers here should be zero. So, this is already quite interesting. So, we thought, wow, if we can do this, and again, this is only with 25 microliters of blood, what sort of things we see. So now, we have done this for hundreds of people.— now maybe I exaggerate, let's say for dozens of people. And we see weird things. We see someone having an antibody profile at the top dominated by high-mass antibodies, but only dominated by a few of them, 4 or 5. But we see also people like you see at the bottom. It's not noisy. This person just has a broader range of antibodies in the same intensity. And you see that, you know, on the left, for instance, it's quite surprising if you see the axis. The axis at the left one is up to 3,500 micrograms per milliliter. So, this person is really having some dominating antibodies. While the axis at the bottom is micrograms per milliliter, and it's only 3 or 4 micrograms per milliliter. So, personal antibody profiles are unique for any one of us. But they can also be quite variable in abundance, in diversity, in complexity.. And so that's what we, what we are learning at the moment about these things. So then you can start to think, okay, if you can measure antibody profiles, that's interesting, but what, where do you want to go? What can you do with this? And so one thing we wanted to do is, can we see, okay, maybe the antibody repertoire is stable if you're healthy, but what happens with your antibody profile if you're at the ICU? This is what we're doing here, but we have now also studies what happens with your antibody profile when you get COVID, or what happens with your antibody profile when you get vaccinated. So these are the things that we are studying at the moment. So here's the initial study. We, we had samples from, uh, what is it, 8 different donors. All these donors had surgery, and unfortunately, after the surgery, they got a bacterial infection, and they were admitted to the ICU, which was bad for them. Good for us because then blood was taken at 4 consecutive time points of these patients before surgery, after they were admitted to the ICU, and also when they were released from the ICU. So, we had a sample set of about 32 samples, 8 donors, 4 time points. Yeah, you cannot design an experiment. So, the time points were quite different for some people. Some were released a few days after they were admitted to the ICU. Some of them were only released a few weeks after they were admitted to the ICU. But basically, we did the same trick. We got some blood from them. We did this IgG profiling, as we call it now, FAB profiling. And then our interesting thing that we wanted, are all the clones persistent, stable, like in a healthy person? Or are there clones disappearing? Or are there clones that come up after sepsis? And of course, the ones that come up after sepsis are potentially interesting because sepsis is often induced by bacterial infection. So hopefully the antibodies that you start to make will target the bacteria that you're interested in. I don't know if you can see it at the back, but here you see now 4 profiles of 1 donor or 1 patient. And then at the bottom you see some zoom-ins of individual clones, and you see that there's quite a bit of dynamics. So indeed we see antibodies, clones that come up. We see antibodies that disappear. And we see also antibodies that remain relatively stable. Now, of course, unlike that's at number 8, that's an antibody that is quite abundant, but only at T4. So only when the patient is released. And so this is, of course, a clone that you're really interested in, because that might be a clone that is formed after infection. Now, first of all, again, now we had a bigger dataset. So we made this correlation plot again, and you see we all respond differently, but we see that everyone has a repertoire which is completely unique from another person. Is that surprising? If you believe me that we can make 10 to the 11th, why should you and I make the same one? So, if you think about the numbers, it's maybe not so surprising, but I think it was also not that well known or realized so far. Now, you also see that everyone is unique. But here, the boxes, the 4 time points in one box, you see there is more variation because there are really patients that get new antibodies after the sepsis. While some patients like M67, there nothing seems to happen that much. The antibody profile remains sort of the same. But the message is still each donor is unique. Now, we had one donor that was already a bit strange in the beginning. This is a female donor. And already at the beginning, so at T1, her antibody profile was dominated by a few clones, really dominated by a few clones. So you see here the antibody LC-MS profile at the top, the FAB profile, and you only see 4 peaks or so. You know, this is cleaner than a BSA digest. It almost looks like a pure protein.. But then at T2 and T3 and T4, a huge new clone came up that you see here in red. And this is where we thought, okay, we say it's a clone, we say it's an antibody. But of course, now you want to know if this is an antibody in response to the infection. You want to know what this antibody is, and then you first have to know its sequence. So then we said, can we sequence this clone and can we really determine what this antibody is, and can we then maybe start to use it to see if it really responds to the bacterial— that the bacteria that infected this patient. So what we do, we measure these FABs, but you can also measure these FABs after reduction, and then actually the light chain and the Fd part of the FAB dissociate. And you see that at the bottom, it's the same LC-MS trace after reduction, and then of course the FABs fall apart in the light chain and an Fd part. And you see the trace again. So now we have the FAB, we have the LC, we have the FD. But this antibody was so abundant in this patient that we thought, okay, maybe we can also do the bottom-up approach. Maybe we can also analyze it by normal proteomics. So antibody sequencing with bottom-up is possible. It's possible if you have a really pure antibody, and most biopharma can make a real antibody. But it's still very difficult because you have the same problem with this sequence variation. And so what people do to, to, to get antibody sequencing, they don't rely only on trypsin because then often you don't have a peptide in one of the CDR regions. So typically people come in with 4 or 5 proteases and they generate all kinds of peptides. And you see that here on the right, you sort of see with this heavy chain, that's the heavy chain, and then in black you see all the peptides. And you try to get as many peptide reads you can get to get the whole sequence of the antibody. And with trypsin, that trypsin is a very good protease, but you really need multiple proteases to do this. So in our lab, Joe Schneider and Douglas Coulter, they made this Stitch program, which is I think related to genome sequencing. So when you have all these peptides from a pepsin digest, from a lysee digest, from a lysN digest, from a trypsin and chymotrypsin digest, you actually have to stitch them together to get to a full sequence. And you have to see where they agree, where they disagree. And that's why I, for instance, like trypsin is almost too good there that it has no miscleavages. Like pepsin can make a lot of different peptides, and therefore it's more difficult to search.. But it also makes peptides that actually overlay each other. So you get from the same sequence stretch, you get multiple peptides. But if you have all these datasets, you have to reassemble that again into a full sequence. And this we do with this peptide assembler, as we call it, Stitch. So, Jürgen, I probably think you can do better, but this is the start. But the problem is we are still sequencing this antibody in the background of all other antibodies. They are maybe lower abundant.. But they also there, and you know how sensitive bottom-up proteomics is. So it's very difficult if we identify a peptide to say this really comes from this clone. It can also come from a background clone. So that's why we combine it now with what we call bottom-up or middle-down proteomics. So what we also do next to the digestion with trypsin, chymotrypsin, thermolysin, and pepsin, we take these FAP molecules and their reduced LC and FD parts. And then we do directly fragmentation on these molecules, on these ions in the mass spectrometer as well. Now, just to summarize, but I will show a little bit more detail. Of course, with the peptides you get good sequence coverage, but you don't know if these peptides really came from the precursor you were interested in, from the FOP precursor you were interested in. Now, of course, we get less sequence information from doing top-down, compared. But you get really constraints about, okay, if you think you have the full sequence right, you should also be able to predict the mass of the whole FAB molecule or the mass of the FT or LC part. So we use this in combinations to do this, and we did this first on this particular clone. And here you see some ETD spectra of the light chain and ETD of the FAB. Now, relatively, these spectra are poor.. And we don't see that many fragments formed. They are actually all C and Z ions because we do electron capture induced dissociation. But what we get from these spectra is sequence tags. So we can really read off a particular small part of the sequence, maybe 4 or 5 amino acids, and we get the precursor mass of the FD or the LC or the FAB. And taking that together, we can map that on the bottom-up data. And we can see if the bottom-up data says, oh, I maybe have 3 or 4 solutions for this peptide. We can say no, it has to be that one because that only matches with, with the data that we get from the middle-down proteomics. So we do this hand in hand, and but also iteratively. Now you get very complicated outputs that you see over here. We still use a database, it's called the IMG database, and this is a database that is full of germline sequences that once have been sequenced, measured by someone, and put in this enormous database. So this germline database of FAB and/or antibody sequences is sort of our template. So we try to match our sequences against that template. And here you see the germline sequence, which is the closest match to the sequence that we are observing now. And you see what we call the mature antibody,. And that's the antibody as it is circulating in this patient. Now, between the germline and maturation, actually a lot of things can happen. There are hypermutations taking place while the antibody is made or while the B cell is maturing. And even when it's secreted, you can still have recombinations. So that we don't find a match that is exactly the same as the germline should not come as a surprise. Now, you see here the output of our software now. So we have all the bottom-up peptides in different colors.. And you nicely see how they, you know, actually you also see here that trypsin is poor, and you get very nice peptides, but only a few. But if you also come in with chymotrypsin and thermolysin, you see you get many more peptides. And altogether, that makes the sequence identification more confident. So that's how we use the bottom-up. But we do it again with the middle-down as well. And we have now this iterative search going with this bottom-up and middle-down sequencing at the same time.. Now, just to give you some indication and how it works. So in this IMG/T database, for instance, there are about 4,000 sequences only for a light chain. Now, they have different variants, different gene variants. But if we look at our bottom-up data, we can see that out of the 3,600 sequences, about 60 or 6 and 7 from these different groups still remain after our data analysis. So we filter already out, but maybe the message I want to give, we never get a real hit. We always get, oh, there are 10 solutions that are maybe equally good now. And that's really what we want. And you see that on the right, that's our scoring system. You see we have many potential hits in the IMTT database, that are sort of with the bottom-up data doing equally well. And then we come in with the middle-down data, with the sequencing data that we get, the stretches we get from the middle-down, and then we can apply a scoring and we can really filter out which one of these is really the best one. Now, we think this is really essential to combine this bottom-up data with this middle-down data and the FOP data to find your best hit for the the most likely template. But then you still have only the most likely template. And again, that is not the sequence of your antibody. So what we do then, and then it's a lot of work, and that's why we are just doing this one by one almost, we actually have our germline sequence. And then by looking at the middle down on the bottom of data, we see where we have mismatches. And we do this iteratively. And in the sequence at the top, in red are amino acids in the germline sequence that our data is in, in contrast with. And so, for instance, you see that this Q, according to our data, that has to be an H. But not only this Q, you see actually that there are maybe 15 positions here where the germline sequence, which is our best hit, in the previous slide I showed this is the best hit, we still see that we have like 5, 6 or maybe even more amino acids where this hit is not good. And so we can, by doing these steps, refine our sequence. And in the end, we need to get a sequence that is completely right. So we thought we did, and this is how it goes. So each time you have a prediction and you say, oh, it looks pretty good, but the mass is 58 daltons off. Now, by knowing this, you can say, oh, maybe I look back at the bottom-up data and maybe there is another second hit that actually explains this 58 Dalton better, and then we take that one. So we do this almost manually at the moment, and we hope to, to get this way more automated. But then at the end, we get our final result, and it's, yeah, it's just surprising. We think we have the right sequence now, but there are still 19 residues mutated compared to the closest sequence that is in the IMG database. Now, yeah, This is pretty horrible if you want to do it, but it also shows this is reality. This is how antibodies are in our body. Now, in this case, and maybe also to convince you and the reviewers, we said, okay, if we now predict this de novo sequence from this clone in plasma of this patient, how can we better prove this? Now, so we made the protein recombinantly because we had the sequence, and we Of course, it had exactly the mass that we predicted because it was the sequence that we predicted. And then we also did all the experiments that we did before. So we took bottom-up data, we took middle-down data, we could do FAP of the light chain. And in all these instances, we found that our synthetic recombinant antibody did exactly the same as what we saw in the data that we got from the blood-derived antibody. So, this is where we are now. And this work got published a year ago or so. I hope you got a bit interested in it. So, what I told you in the beginning is what we are doing now. So, I think COVID has been horrible, but it has also been the most beautiful immunology experiment done ever. Because we are now looking at how each of you responds to your vaccine. You make new antibodies against the spike or against another protein of the virus with the vaccine. Not, but if you're having COVID. And we find some very remarkable things, and just a few I want to share, which— because I'm also very much surprised. So you make antibodies when you're vaccinated, but then if you— or most of you do, um, but the next time you get vaccinated, do you think you make the same antibodies? You're going to make quite a few new ones. Some of them you make again, and some of them you don't make anymore. Will any of you make the same antibodies? Answer is no. We all make different antibodies. So, and this is a bit where we want to go to, and I really have to think. So this is a huge project, and some people know that we have a big group, but on the left you see what we call the IgG task force. This is really like 10 people who worked on developing these methods. Both from the purification to the sequencing. But what we hope now is that we can maybe get in the future antibodies out of a donor that has— that is a good responder, maybe not to COVID, there's enough good vaccinations there already, but a good responder to any virus or any bacteria or maybe even in diseases. And by finding the antibodies out of a person's blood, we have already human antibodies, which is a big thing.. And of course, potentially they may then be used to also treat other people who not make the right antibodies against this disease. This is a dream. I don't know if it will be a reality ever, but this is one of the reasons why we do this sort of work. So with that, I like to end. This is the picture of the whole group after COVID, our first group picture. 3 of them I think they're all on the front there. They are also in the room there, so there are 3 of our people here at the group. I hope you enjoyed what I had to present, and I hope in a few years I can present it again and show you that MaxQuant can solve all these problems directly.

Thank you very much for your attention. Albert, fantastic, thank you very much. I have, well, question and comment at the same time. I think I'm not very surprised at all, actually, that you have these personalized Ig profiles, right? Because this is actually the nature of the immune system in each of us, right? So, you are responding constantly since birth to different attacks, right, and self-attacks. And then this is creating this repertoire. So, well, first, the question is related to my work. Do you— I mean, maybe you don't have enough data at the moment, but maybe in the future you could look at this. And I will be very interested in the results.

Will you have different repertoires of different IgG clones depending on the time of the day that you have the samples taken? I should have expected that question from you, of course. I'm not sure. You know what I showed first, and we took these repertoires at 3 time points of one person, and they remained very stable. I have no idea when these— when the blood was taken. But so, there might be slight differences.

But to be honest, I don't know the answer to the question. But I don't expect large— Maybe actually the question will be even better because people— and there are some papers, and that's the vaccination. Yeah. So, COVID vaccination. Really produce a different response depending when you got the vaccine. So there, it could be interesting.

You get the vaccination in the morning or the afternoon, do you produce different IgGs or different repertoires, right? Yeah. And so I don't know the answer, but I think you can also respond when are your B cells becoming most proliferative or most active. And I can imagine that there, there is a day-night cycle. Maybe they are more active at night, or maybe they are most active in the morning.

I have no idea, but I can imagine that there is a circadian factor involved in all of this as well. And I think this goes now to my comment, or also question, to this different— so you said, you know, when you get all vaccinated, you produce different antibodies, and then second vaccination gives, even in each of us, a different repertoire, right? So, this also my comment, or well, your dream will be to find an antibody, right? Perfect antibody that you could treat. But would you not think that this will be, or that's the reason why we produce so different ones?

That is not one antibody, but this is a combination of different antibodies that affects us, right? That's a very good question. And I simplify a bit, but our data, and we just got here today, that is accepted. But when you're vaccinated, you generate— or when you get COVID, you generate maybe a few dozen of antibodies directly targeting some of the proteins. And this is in COVID patients, of the COVID proteins. And I think that needs to be— and what we did now, it's a very nice experiment. So, we had— we all worked with very small groups or so. Everything I say, you should take with a pinch of salt, but we had two people that had— and I'm also know that I'm not allowed to call it this, but they had the Chinese COVID or Brazilian COVID or the South African COVID. And all of them made antibodies, like we could detect a dozen of them. And then we see that some of these antibodies actually from a Brazilian patient, they are, you know, they can also neutralize the Chinese variant or they can neutralize the Omicron variant.. But some of the other ones in this dozen of antibodies cannot. So, I think that's also part of our immune system. We don't make one, because if that one is vulnerable, you get into problems. So, we always make an array of antibodies against any antigen, and they target different epitopes. So, they are also responding differently when the virus starts to mutate.

So, and I think, and that this is what we think as well, Maybe in the future we will not use antibodies anymore, recombinant antibodies as treatment, but mixtures, so a polyclonal mixture. And in this sense, maybe more proteomics or limitation of the technique. Do you think, because you saw that you can find, well, in this particular patient it was very clean, right? So you have one Fab and then you produce another one after the sepsis. Do you have a big dynamic range, meaning also What is the functional point of this particular antibody? So the peak is so big.

Do you have also smaller antibodies that you cannot detect maybe because of the dynamic range that are also very important? Yeah, I will be amplified once you have. Yeah, so that's relatively easy to do and we have studies on the way. So what you can also do is if you have a specific antigen, you can also use that to filter out the antibodies that bind to the antigen and only analyze those with the same techniques that I say here. So, we have now antigens against very different diseases where— that we use. And then we sometimes pull down clones that we even don't see in the profiles that I measured today. So, if someone asked me, are there really only 750 clones? No. Or maybe thousands. It's just like in the normal proteomics. There are way lower abundance antibodies also circulating in your blood, and you can filter them out if you have the specific antigen that they bind to.

And so that's one way we go. But I also like this sort of unbiased approach to look at the full.

So, could be also one reason why different people have, or you find different profiles, but this is maybe not that the individual variability is different, but the abundance of each antibody is similar.

Yes, that's true.

But still, I don't know. We all make very abundant antibodies, and no one knows what they are targeting. Yeah, yeah, sure. Which I also still find remarkable. Thanks. I have two small questions. First is, Regarding phosphorylation of histidine that you found in bacteria, do you think it has a role, or are there information available? Yes. So, in bacteria, there are even kinases known that are specifically, you know, doing the histidine phosphorylation. And we also have some studies, or Simone has some studies where we did knockouts of these kinases. And I think in— I have to remember now correctly— like in Bacillus subtilis, I think you only have 3 kinases and 2 phosphatases that supposedly do all the histidine phosphorylation. So, it plays a role in signaling. And by making knockouts of these kinases or phosphatases, you potentially can find the substrates of these kinases and phosphatases. So, in that sense, histidine phosphorylation in bacteria is sort of similar as what serine and threonine phosphorylation is doing in mammalian cells. And my second question is just in continuation with what the former person just talked about. You spoke about the two normal individuals, healthy individuals. One had a higher abundance of a particular set of clones compared to another. So does the first person where there was high abundance represent antibodies that may show that this person has some underlying condition that's not diagnosed? That's a very good question, and the answer is yes and no. The answer, it could be, but we are still learning a lot here, and it's It's funny that not that much is known. And what we actually see, but again, maybe we have hundreds of these profiles now. So, we actually see that if you're young, you— we see more broad profile with equal abundance, plenty of clones. We happen to have also a cohort of people that had COVID and were at the hospital. They are all above 75. And their clonal profile is really dominated by just a few clones. Yeah, I don't know if that's age or COVID. I don't think it's COVID. But so we still have to learn what it all means. There's also people that have this mono-gammopathy, which is that you have a B cell that is almost like a cancer cell, overproductive, produces one antibody to an enormous amount. That is sometimes linked to melanoma.. But, you know, I was also surprised. For me, this is also new. It's also related to MGUS, which means a ridiculously producing B-cell of undefined significance. That is the US. So, so they actually know that some people have this, but they are completely healthy, or sort of. So, so sometimes it's related to melanoma if you have a— a clonal profile that is dominated by one or two clones. And sometimes it doesn't mean anything.

So I wish I could—

maybe in a few years I can answer that better, but at the moment I can't.