Scientist Stories: Matt Bogyo, Chemical Biology to study Human Disease

Welcome everybody. Hi, and welcome again to a new series of the Broad MIT Seminars in Chemical Biology. I'm delighted to introduce today Matt Bojo and welcome him to our series, who's been a professor at Stanford in the departments you see here since 2003. Since then, he's published over 200 papers in a variety of different areas, including chemistry, microbiology, human parasites, life cycle, bacterial pathogens, tumorigenesis. He's also received numerous awards, including election to the American Association of University Pathologists, and he served on numerous editorial boards, as well as being a consultant to several pharmaceutical and biotech companies, including founding on his own Acrotome Imaging and Fasol Therapeutics. And today he's going to tell us about what's been quite a very hot topic lately, and that's covalent ligands for biological applications. So thank you for joining us, and let's all welcome Matt.

All right, thanks very much, and thanks for the invite to come back to MIT. It's exciting for me. This is where I did my PhD work. It's a bit of a homecoming. I just realized I hadn't walked through Building 18 in almost 30 years. It looks kind of the same, but a lot nicer. So yeah, actually, I was asked to give this talk at Tufts, and they wanted me to give this kind of overview a little bit about my background. So I thought it'd be kind of fun to give this sort of history of where this whole interest of mine came with covalent ligands, and it started here at MIT. So I'm going to give you the background a bit, And then I'm going to tell you stories of how we're applying this in some areas of biology, and in particular in the gut microbiome. OK, so I guess I should be using this. So I thought I'd start off and give you kind of overview of the lab interests. Again, it's the unifying theme in my group is the use of these kinds of covalent ligands as a way to study various targets in biology. And so the targets that we've worked on quite a bit are proteases and hydrolases. So enzymes that use a nucleophilic attack mechanism to hydrolyze a bond, an amide, could be an ester, thioesters. So it's pretty broad. And it turns out that proteases and hydrolases tend to make up about 1% to 5% of the genome of most organisms. So they're very prevalent, and they do lots of different things. So we haven't gotten bored yet. And where we've sort of kind of moved into the next generation of technologies, we've started to apply some selection techniques, so phage display and now more recently mRNA display, which I was doing as a sabbatical at Genentech for the first part of this year. And that's to try and get to better structures of ligands that bind with higher specificity to targets so that we can control both the reactivity of the covalent part and then the binding energy of the ligand part to the protein target. In terms of the biological applications, we're very interested in applications in cancer. I'm gonna talk a little bit about our interest in image-guided surgery, so how do we use covalent ligands proteins as a way to image where cancer is present during surgery. And then I'm not going to talk much, but we're also very interested in how immune cells function and how they use proteases in particular in cancer. And then we've done a lot of work on microbes. We started off initially in malaria and human parasites, but we've moved into a number of bacterial pathogens— Staph aureus, C. diff, and MTB— and then more recently into microbes in the microbiome, so commensal bacteria particularly in the gut. Okay, so my training again was here at MIT. I was here from '93 to '97. I came into the Department of Chemistry thinking I wanted to do pure organic chemistry and then eventually realized I wanted to apply what I was making of the molecules, so I ended up finishing in biochemistry. And I actually worked for Hiddeplew here, and he was in biology, so I had to actually have a secondary mentor in chemistry. They were scared if I went over to biology. So I had Larry Stern was my secondary mentor. And so that was kind of a cool setup. And I give MIT a ton of credit and chemistry a ton of credit for letting me do it, which was nice. OK, so I got interested in this protease. This was my first protease that I got interested in was the proteasome. And the reason why was because Hita's lab was studying MHC pathways, so class I and class II pathways. And at the beginning of the MHC Class I pathway is the proteasome. So proteins get degraded by this large protease complex. And they produce peptide fragments that then are presented through MHC Class I molecules. And so at the time, Haida was studying, in particular, MHC Class I molecules that were being degraded when a virus expressed a specific protein. And no one knew what was happening to these MHC molecules. And you needed an inhibitor of the proteasome to be able to block that process. And at the time, there was this company called Myogenics that Fred Goldberg had started. And this company was producing a molecule here called MG132. It was a peptide aldehyde. This was not commercially available. And so Hita came to me and said, hey, we need a proteasome inhibitor. We can't buy it. There's none available. What do you think about this molecule? I looked at it. I said, peptide aldehyde, very simple. Let's make it. So I set up chemistry in Hita's lab, and we made this molecule, and it worked. It was great. It was a good inhibitor. It was actually used by Hita's lab to map out the ERAD pathway at the time. So by basically blocking the proteasome, he was able to see unfolded MHC molecules that were reverse translocated from the ER into the cytosol and being degraded through this complex. And because he had an antibody that could see unfolded MHC molecules, He was able to see this, and they were able to map this pathway, which was totally unknown at the time. So it was an exciting use of a small molecule and a very simple molecule. But then I started looking into the literature, and I found that there was this chemistry that was just reported for protease inhibitors called vinyl sulfones. It was reported by a company out in California called Aris Pharmaceuticals. And what it is is essentially an irreversible Michael acceptor. And so I started thinking about this, like, what if we take this reversibly binding aldehyde and put this new warhead on, namely the vinyl sulfones? Acetylsulfone. It's one step. It's a Horner-Wadsworth-Emmetz chemistry to get you to this molecule. Very simple. And I remember telling my friends about this. They said, well, this isn't going to work because this is a— this likes a thiol nucleophile, not a hydroxyl. And the proteasome uses a threonine in the active site, in its multiple active sites. And I said, well, let's make it anyways. And they said, you're wasting your time. But I did it. So I made the molecule, which we called ZL3VS. Which is basically the exact analog of MG132. And then we actually put a radio label on this back in the day when you put slopped radioactivity around and used I-125. So we put this radio label on it, and for me this was sort of the aha moment of testing this molecule, was putting it into an actual cell in a complex mixture of proteins and asking what happens. And I remember distinctly when this film was coming out of the processor, back when we put it through film processors, And it was blank, dead blank. And I'm like, oh great, another failed experiment. Until this band showed up right here. You can see right here, this one here. And this was actually the proteasome that was being labeled. But what was amazing is it was being labeled in the context of all these thousands of other proteins. So it was amazing to me that I could make this molecule in the hood, put a little radio label on it, drop it onto a live cell, and then be able to see only a single protein getting labeled. And it turns out it wasn't a single protein. 2 proteins that you could resolve by 2D gel. And in fact, it was 2 isoforms of the— 2 subunits of the proteasome, the inducible, interferon-inducible, and then the constitutive subunit of the proteasome. And so I realized suddenly that by having covalent molecules, we could actually see the actual subunits that was processing the molecule rather than having to use substrates, which get cleaved and released, and you don't know who cleaved them and where it came from. So this was really, for me, the moment where I realized if you can tune our chemistry, we can get these things to bind in a very selective way. Okay, so I finished up in Hita's lab and then I moved to UCSF as a UCSF fellow. I was there at a really fun time, started at the same time as Joe DeRisi, you may have heard of him, who's now head of the Chan Zuckerberg Institute out in San Francisco, and he was making his spot arrays at the time, and so it was really a fun time, a lot of exciting things happening. They gave me a whopping $75,000 a year to run my lab and a salary of $45,000, which seemed amazing to me as coming from graduate school, but living in San Francisco was not easy on that. So what I did at UCSF was I kind of shifted away from the proteasome for a little bit and started working a bit on lysosomal proteases in particular. We figured, OK, we studied class I proteases. What about MHC class II proteases? And those are the lysosomal cathepsins. So these are interesting proteases. They're a relatively small family. The cysteine cathepsins are only 11 members in the human genome. So they're not that huge of a family. They're predominantly found in the lysosome. And they also get secreted. And they're produced predominantly by macrophages in the tumor microenvironment, so what are called tumor-associated macrophages. So that was interesting to us because it suggested that these proteases could actually be pretty good markers for the tumor. Because what happens is the macrophage has these proteases, they're not, they're not active normally, or they're just mildly active in the lysosome. As soon as they get into the tumor microenvironment where there are cytokines and other signals, they start to produce a lot more of these proteases and they secrete them. And so now we can use that to sort of delineate where the tumor is as opposed to the surrounding normal tissue. So we started targeting these cathepsins. We initially worked back in the early 2000s when I was at UCSF on these kind of epoxide-based compounds. And you can see— let me see— right here, it's based on a natural product called E64. So what happens is the thiol opens up the thiolate from the active site, opens up that ring, and you get a very stable thioether linkage between the molecule and the active site. Site. And now we smartened up and decided to put on fluorophores using these kind of organic BODIPY-based dyes. We could now generate fluorescent images rather than radiolabeled images. And actually, back at the time, there was no scanners to scan for the fluorescence. So we had to use— this was a DNA sequencing rig that we used to do this first gel, literally running our proteins on a DNA gel. And then at the bottom, it was scanning for fluorescence. And we generated an image and we were really excited because we could use different colors and we could resolve all these different cathepsins based on their molecular weight. And then the scanner started coming out. You could buy a flatbed scanner and put your gel literally while it was still in the glass plates. You could just scan it and you'd see all these labeled bands. And unlike the proteasome, which was a single band, this was multiple bands because our probes are not exquisitely selective. They hit that whole family of cysteine cathepsins. So you could actually delineate a number of different active cathepsins that were present in your sample. And then I was right next door to Doug Hanahan's group at the time and a really talented postdoc at the time, Joanna Joyce, who's now at University of Lausanne in Switzerland. And she basically had finished a transcript profiling in a beta cell tumor, a spontaneous beta cell tumor that they had set up in Doug's lab called the RIP tag model. And these beta cell tumors would grow spontaneously in the pancreas, and they could take all those tissues and they did transcript profiling and found that one of the biggest upregulated sets of signals was actually cathepsins. And so she came next door and said, hey, I know you guys work on cathepsins, can we use your probes to look at these things in the mouse? So we said, yeah, let's try. So that was our first attempt to now put this in vivo back in the early 2000s. And essentially what we did was inject these fluorescent probes. They were fluorescent all the time, but they stuck covalently to their targets, and then anything that didn't bind would circulate out. And we got these kinds of images when we did histology of those beta cell tumors. And you could see that essentially you get this line of active cathepsins. Let me see if I can get this. Yeah, right here. And these are the cathepsins getting lit up by our probe. And then blue is just DAPI showing you that this is the tissue here. So this is the tumor. And they tended to be brightest right at this margin between where the tumor and the normal tissue were. So that was really cool. Meant that we could potentially use these to delineate the tumor from the normal tissue. So then I made a little bit of a slight detour. I was at UCSF, ran my group there. At the time, as I said, I was following these companies, this Ares Pharmaceuticals company, because they were the ones that came up with this vinyl sulfone strategy. And in the meantime, they had been converted to this company called Axis Pharmaceuticals, which was literally just down the road from UCSF in South San Francisco. And they had me consulting for them because they were working on proteasome inhibitors and cathepsin inhibitors. So that worked out and I got a sort of flavor for what it was like to be in industry. And then what happened was Axis actually got acquired by Celera in 2001. If you guys remember, Celera was the company that sequenced the human genome and they came to the realization that they weren't going to just be able to sell information on the genome for the rest of their life, so they decided to become a drug discovery company, so they acquired Axis. Lock, stock, and barrel, and it became Celera on the West Coast. They asked me to come in and set up a group there to do chemical proteomics, and this was in November of 2001. And so I decided to go there, and I moved most of my group there, and we started working on probes, covalent probes. The idea was that if we can make a covalent modifier for any of the targets they're interested in from a drug perspective, They could use it to look at target engagement in vivo. So you could ask, is our drug getting to the target? Is it getting to the right place? So there's a lot of really exciting things you can do with it from a drug discovery perspective. So we went there. This is one of the early molecules that my group synthesized, because they were interested in kinase targets, in particular BTK, which, as you guys all know probably, is a B-cell tyrosine kinase that's involved in numerous cancers. This was a big target on the radar in the early 2000s, and everyone was starting to work on it. So we made this probe. It was synthesized basically using a known inhibitor and just attaching to it an electrophile that would be able to target a cysteine in the ATP binding pocket of the kinase. And so this guy, Zhenying Pan, synthesized it in 2003. I actually moved back to academia because I wanted to be back in that environment, and so Right about less than a year after I left, Celera shut down its entire small molecule program, so they basically laid off everyone in South San Francisco in one day. They had everyone come into a big conference room and said, thank you very much for your service, you guys are all, you can leave. And I missed that, but my group was still there, and then what they did is they sold off their assets, including all the BTK assets, to this company called Pharmacyclics. Pharmacyclics, and they sold it for $2 million in cash and $1 million in stock in 2006. And then what happened was Pharmacyclics decided to develop the molecule, which is in fact imbrutinib. So our probe molecule that was not— God forbid we'd call something a drug that's covalent— became imbrutinib, and then they ended up eventually developing it with AbbVie. AbbVie bought Pharmacyclics for $21 billion, and currently the annual sales of imbrutinib are currently over $80 billion. So, this was exciting for us because, and it was really fun for me to see this happen. Obviously, it didn't happen overnight, but see the shift towards covalency as a potential way to actually develop therapeutics. Okay, so then I moved to Stanford in 2003. I've been there ever since in the same lab for the last 20 years. And essentially, we started working there. We started thinking about these covalent modifiers and the fact that If we have a fluorophore on our covalent modifier, we still have to wait for it to find its target, circulate out all the unbound probe before we can get any contrast for imaging. So we wanted to ask the question, could we make molecules that are quenched fluorescently? So we would put a fluorophore— we knew we could put a fluorophore on these molecules, but could we put a quencher on them? And if you develop the chemistry just right using these initially acyloxy and then phenoxymethyl ketones, what happens is you get a thiol in the active site. It attacks the carbonyl. It actually migrates over to this carbon, and it kicks out this whole part of the molecule as a leaving group to generate this stable bond, which is a thioether linkage. And so this is perfect. Now we can put a fluorophore on one end and a quencher on the other. When the covalent chemistry happens— this is more accurate what it looks like. Here's the cathepsin. The yellow is the active site cysteine. Covalent binding happens, loss of the leaving group, and now your fluorophore turns on. So it's a smart probe now. You no longer have to wait for that probe to circulate out because it's only on once it's bound to the target. And the other thing is that it's stuck to that target covalently. So now the half-life of your fluorophore is really dependent on how long it takes the cell to turn over that protein. Turns out cavepsins don't turn over very quickly. So you can inject these probes or put them into cells or mice, and they have very long half-lives. And you get signal for 48 hours plus. So that's what we did. We started putting these molecules into mice.. And we had to optimize the chemistry a bit to get them to be stable enough. But the big difference was now you can see that this is a mammary fat pad tumor that you're looking at here. So these are two small tumors you can see here. And then over the course of a few hours, you get really nice contrast as the probe lights up in the tumor tissue because of these high cathepsin activity. And by 8 to 10 hours, you're getting very bright signals, specifically in the tumor. And because the probes are covalent, we can take those tumors out of the animal. They're fluorescent as a piece of tissue, and then you can run them out on a gel here and actually visualize that same series of cathepsin bands that I showed you earlier. So we're actually hitting multiple cathepsins that are related, and we can actually figure out which cathepsins we were hitting. So it's nice, you can go all the way from a whole animal, live living animal, back to biochemistry here, which is why again we like this covalency. Okay, so then we started thinking, well, if we want to image in a clinically relevant system, we can't— you know, no one's gonna— for these mice, we can put them in a nice dark box and open up a camera lens and take a very sensitive picture of what's going on. We can't do that during surgery. But at the same time, there's been a big advance in instrumentation for surgical applications, in particular for endoscopy. I mean, not endoscopy, for endoscopic— what am I trying to say? This is— I'm losing the word here, but this is basically laparoscopic surgery is the word I'm looking for. So you basically have a small— you have these arms on a robot here that would mimic what a normal laparoscopic procedure would be. So rather than opening up the patient, you have just 4 little entry sites for these arms. And then normally under normal lab procedure, the surgeon surgeon would be sitting right over the patient and manually controlling it. But they developed a robotic system here called the da Vinci surgery system, developed by a company called Intuitive Surgical, which is based in Sunnyvale, California, which is luckily for us not very far from Stanford. This was actually developed as a DARPA project initially. So the idea was to make a robot that would go out into the battlefield and allow the surgeon to be sitting safely somewhere else and be able to do the surgery through a robot interface, and obviously it never got to the point where it's used for that purpose, but it works very well and it's now in hospitals pretty much all over the world. Pretty much every hospital you're gonna go to is gonna have these robots. And the way it works is the surgeon sits usually in the same room, maybe in the next room over, but it's hardwired, the robot and the controller. The surgeon looks into a 3D, into an eyepiece which gives them a 3D image down with the camera which is inside the patient., and then they move the arms of the robot by basically exactly doing the same movements themselves. It's very simple to learn because you basically get a very large field of view and you can control it in a very precise way. So it's very good for doing fine procedures. And so this was great for us because they were interested in— can we— they already had a camera that can see fluorescence because you can use a dye called ICG, indocyanine green. And you can inject that into a patient, and wherever there's blood flow, you get a green fluorescence. And so it actually was getting used for fluorescence imaging, not for any kind of targeted imaging. So they wanted targeted contrast, and they came to Stanford, and I was able to link up early on with some of the people at Intuitive. What that allowed us to do is test some of our probes in the surgery system. And you can see here, you can put a mouse under this exact same instrument that's used in humans. It looks a little crazy because the mouse is so small and this is a giant robot tearing it apart, it looks like. But in reality it allows you to do really fine movement and do very precise surgeries on even small animals like mice. And then here's just one of my students doing the surgery. So we get to play around with the robots ourselves. They start you out with these tiny little cones where you move rings onto a cone and they're a few millimeters high. But then after a while you can figure it out. The only challenge about using this system is you get no tactile feedback. It doesn't tell you if you're pressing too hard, but it's quite straightforward and you can do surgeries pretty readily with this. So I'll show you just a quick video. This is a very old video. I just like it because it was one of the earliest videos I did myself doing the surgery. This is sped up 8 times, so it's very fast, and you'll see I'm not a very good surgeon. But basically this is one of the mammary fat pad tumors, and you can see during surgery you can flick a switch with your foot and it turns on the fluorescence, and now you can sort of see where you're getting fluorescence signal. And cut around the fluorescence. In this case, I'm cutting out a very large tumor that's very easy to see with your eye. And I'm cutting in a very crude way, so it's leaving behind virtually no cancer. And you'll see the tumor bed is basically totally non-fluorescent. We did another attempt. And here what you can see is essentially this is the resected tumor from the inside of the patient. Peritoneal cavity here. And you can see here, this is the residual location where the tumor was, and we kind of resected as much as we could and said, okay, let's see how we did, and then turned on the fluorescence. And what you can see is when you flick the fluorescence on, the first thing that happens is there's, there's definitely fluorescence here. This is a piece of tissue that looks totally healthy, but it's in fact tumor. And then you'll see a fine layer of fluorescence left behind over here. Right here, and that fluorescence is residual tumor that we couldn't get. And when you do the histology, you can show that it nicely matches up. So you can see how, you know, in scenarios where you think you got everything, you really didn't. And so this is the real sort of benefit of having contrast on board. So what we did was to take this molecule, which is called BMV-109, developed by one of my postdocs. It had a Cy5 fluorophore and a suitable quencher for that wavelength. And then we just swapped it out for the ICG, which is the one that's— it's a high wavelength dye that's approved by the FDA itself. So we went to that dye. It's what all the instruments are tuned for. And we made this probe, which is now licensed by a company called Virgent Biosciences. And it's called VGT309. And then Virgent was able to move this forward. In 2021, they put it into a clinical trial in Australia. So it went first through a phase 1 safety study. The nice thing about these probes is they're used at very low dose relative to a therapeutic, so they're injected somewhere in the 1 to 2 milligram dose for an adult patient. There were no significant adverse events. It was only injected one time. They then did a small phase 2 study in patients with lung cancer, and then it was also advanced to a phase 2 study in the US after we filed IND in the United States. And then it went through a phase 2 study also at UPenn. And so it's now been put into somewhere in the neighborhood of about 30 patients with lung cancer. And the dosing was all sort of worked out. So here's some of the results from human. You can see this is a patient that had a large lung lesion that you can clearly see by eye here. This is the lesion. And you can see the probe is nicely lighting up very specifically in that lesion. This is again in the da Vinci, robot view. Here's a, a different camera system that uses a different coloration system, but it worked very well for other camera systems as well. And then what you can do is assess margins, which is like after the surgeon goes and decides to remove that tissue as they would normally, sort of standard of care, take out the tissue, go to the back table and put it under an even more sensitive camera system. What you can see is that the primary lesion's really nicely fluorescent. And this was a case where the surgeon cut too close to the margin. So this would be called a positive margin, which would mean you'd have to go back and take more tissue because you got too close to where you almost certainly left some cancer behind. This shows another— this is actual video of one of the cancers. And this was a case where they couldn't find the primary lesion that they knew existed by CT scan. And you can see here, this is the lesion that's lighting up. They were able to see it very quickly with contrast. Here you can see they're going to cut, do like a wedge dissection to remove that primary lesion. And as soon as they cut through that first spot, they found a second lesion here below, which they— smaller lesion, did not show up on CT, and they didn't know existed. So there's been a number of cases like this where because of the contrast agent, they were able to identify secondary metastases and other lesions that they didn't know existed. And they were also able to assess their margins. So it's gone very well. There's been now a lot of movement in this space, not just for our probe but for other probes. And I think it's an exciting time. There's one folate receptor targeting agent that's now been approved through the FDA. And it's for, I think, for ovarian cancer. And it hasn't really rolled out yet extensively. But I think there's a number of other broad probes like these that are going to be making it through FDA approval in the next year or two, I would say. So it's kind of an exciting time. OK, so that was the sort of cancer component. I want to shift to talk now more about bacteria and why we're interested in a number of different types of bacteria. The first one I'm going to talk about is Staph aureus. We got interested in Staph because it's a superficial skin infection, but it causes a number of major problems in humans. In particular, we were interested in this condition called endocarditis, and that's a condition where you have staph bacteria that gets into your bloodstream. It finds a nice place to sit down and it forms actually a biofilm where it's very hard to clear, and it tends to be on places like the valve of a heart. And if this is the case, you have to go in and remove valves, and it's very invasive. So the question is, how do you diagnose this? Typically what happens is you get a blood test which tells you you have circulating staph, and that's all good and well, but then as soon as you put antibiotics into that patient, you kill off all the circulating staph very rapidly, and what's left is a biofilm that you don't know where it is, and you don't know if it's responding to the antibiotic. So what you really would like is a molecule that binds to that biofilm and allows you to see it and then ask, okay, how does this patient responding, how big is the biofilm, where is it? And do we really have to go in and start doing invasive procedures? So that was the goal. So we turned to, again, covalent probes. This time we decided rather than using the cathepsins ones, which are very specific for cysteine proteases, we decided to use a broad-spectrum probe developed by Ben Cravatt's group at Scripps, which is called a fluorophosphonate. And this is basically— you can see here it has this fluorophosphate bond, which likes to form bonds with oxygen. So it is extremely good at labeling serine hydrolases.— and not just serine proteases, it's here serine hydrolases, which include lipases and esterases. And because it has this kind of a lipid tail here, it tends to like to bind to lipid esterases, is one of the big families. So I told my postdoc, let's just start with the biofilm. That's what we want to image. Grow it under those conditions. And then put in this probe, which is not super cell-permeant. And let's see if we can tag things that would be accessible on the surface of the biofilm. And so he just ran a gel and came back with this kind of result here. You can see these are a number of labeled proteins, just like with the cathepsins, that were getting labeled by this FP probe. And so we went on to pull out all those targets and identify them using proteomics. And they turned out to be a family of serine hydrolases that all contain these alpha-beta hydrolase domains, and most of which, virtually all of them, had really no annotation or homology to known targets. And specifically almost zero similarity to human serine hydrolases, which was great. So it means good— they're good imaging targets. They potentially could be interesting therapeutic targets. We went on to study a couple of these, in particular FphB, this one right here. We were able to show by transposon mutant that if you knock out FphB, these bacteria grow normally in culture, but when you put them into a mouse, they don't colonize certain organs like the lung, and the liver. So it means that it's probably doing something to help the bacteria decide when and where to form an infection site. So we took this molecule, we had this active site probe, which allows us to see all these guys, and we basically pretreated the biofilm with a library of potential serine protease or serine hydrolase inhibitors, and we looked for competition. So we're looking for molecules like this one, which is one of our best hits. It was a simple chloroiso-coumarin compound originally developed by Jim Powers' group at Georgia Tech. And this molecule, when you titrate it into the biofilm, even at nanomolar concentrations, it completely competes away this one protein called FphB. So that was really cool. We now have a very specific molecule that covalently— this is a covalent modifier. You get— basically, you get chemistry at this chlorine site. But what happens is serine first opens the ring up here, and then there's a secondary attack by histidine at this carbon here. So you get a dual attack of the molecule. So that was neat. We decided, okay, let's take that molecule now and turn it into one of our favorite kinds of fluorescent probes. We just tethered off of this methyl group here, put on a Bodipy fluorophore, and then we could put it into biofilms and Biofilms are kind of messy. They don't look great by microscopy. It looks kind of like a mess. But if you use vancomycin, you can stain for those sort of live cells here that have a viable membrane, and that's in this bluish color. And then the pink color that you're seeing here is the probe. And what we saw was there was a lot of this FphB enzyme on the surface, and it sort of spread throughout the biofilm. So that was really good. It means that it really is a viable target within the biofilm. So fluorescence is great. The problem is you can't really do non-invasive imaging in the heart with fluorescence. It's too deep of a tissue. You can't really— you're not going to be able to image through deep, thick tissue. So we've started to think about other contrasts, and the one that we've settled on now is ultrasound. This is one that's used very commonly for diagnosis in the heart. And there's a contrast agent that's called microbubble. And the way it works is it's basically just a liposome. But instead of having just a normal air pocket or some sort of liquid core, it has a gas core that's an echogenic gas. They use a fluorinated gas that basically when you hit it with ultrasound, it vibrates these little bubbles. And that gives back a very strong signal. And the nice thing about that is now you can start to get to high— both high sensitivity and high resolution imaging using ultrasound, which is where it normally sort of falls down relative to like MRI or or optical. And the nice thing also is these lipid particles can be generated using very straightforward chemistry. You can basically do chemistry on the phosphate head group, and then you can generate your microbubbles with whatever you want coated on the surface. So we've made a number of different types of ones. The control ones are like this one here that has nothing on the surface, but we dope in a little fluorescent-labeled lipids, phospholipids, so that we can actually see these things because they're fluorescent. And then we put a copper-free click ligand on here. So this is these cyclooctyne-type ligands that form reactions with azides. So then we can just use azides on our probe, in this case this chloroiso-coumarin that's a covalent label. We click it onto the surface of the microbubble, and now we have the ability to hopefully get these things to covalently stick. The problem with microbubbles is they are confined to the vasculature. They're big. They're between 1 and 50 microns. So they don't get out of the blood vessels. And they circulate extremely fast. So literally you have to inject these things, and within seconds they're gone. But if you can get these things to stick, and in this case hopefully irreversibly to the biofilm, you'll now be able to image that biofilm at high sensitivity and high spatial resolution. So we wanted to test this out. We didn't want to move in vivo. Quite yet because it's a difficult system to set up in animal models. So we made a synthetic blood vessel system where you just basically take a piece of— it's like a metal tube here, you can see right here. And it's basically, you put it in agar block here. This is just a pipette tip top rack. And all you do is you let that solidify, then you pull out the metal tube in the middle, and now you have this like auger tube that looks kind of like a blood vessel, and you can actually grow bacteria in there. You just put the bacteria into the tube, stick this in the incubator, and the bacteria forms a biofilm inside that little tube. And now you can hook it up to a pump system like this, a peristaltic pump, and you can actually image through this. This is basically like a phantom that looks a lot like skin to the ultrasound. And now you can get a sense of whether your molecules are sticking to the biofilms or not. So we did that, and The results came out pretty interesting. You can see if you take the DBCO-only microbubbles— so these ones don't have our ligand on it, they just have the half of the click reaction— you get basically nothing sticking inside the vessel. If you now put the JCP-251, which is the chloroiso-coumarin, you get the bubbles sticking to the wild-type bacteria that express FphB. And these little individual dots are individual microbubbles that are now really stuck to the biofilm. And then here, is the knockout bacteria where we've knocked out the FphB target. And again, the bubbles don't have anything to stick to, so they're gone and they circulate out. So we're really excited about this. The hard part is now the JCP251 that I just described to you is not a great ligand. It turns out it's not very stable in vivo, and it's certainly not very selective. You can see it's a very small ligand. So that's one of the reasons I've been now doing this sabbatical to do mRNA display and other display technology. We want to get to ligands that are much more complicated, much larger, that bind with much higher specificity, and that are much more stable. So we're working on cyclic peptides and other covalent binding ligands to try and improve this, and then we're gonna move this into the in vivo systems, and that's where sort of we are at the moment. So in the last few minutes, I want to switch gears and describe a project that was started by a really talented postdoc in my lab, Marcus. Lakemeyer, and then a collaboration with Nigel Bounet at NYU. I've known Nigel forever. He was also at UCSF when I first started, so we were collaborating back in the early 2000s. And so it was great to have this project come out. So this involves the gut microbiome, which you probably already know that the human body is made up of, by cell number, more cells, bacterial cells, than human cells. And one of the big places of interest where there's a lot of bacteria is in your gut, and the reason why this is interesting is because it's been shown to have a profound effect on human health and disease depending on the composition of those microbes in your gut. And then the other interesting thing is that it's really a community of microbes that are living in your gut, and so if you really want to— if you want to look closely at most of the disease states, what's happening for the most part is there's a dysbiosis, so you get a change in those populations and it throws off a number of things.. And so the idea is, are there signals that the bacteria use as a way to control that microenvironment or the community of microbes that are around them so that they can maintain a status as being the primary microbe? So we started thinking about this, and Marcus said, well, you know, what's known about this in terms of proteases? So, you know, people have been studying microbes in the gut, and there's been a lot of interest in metabolites in particular. So things produce small molecules produced by the bacteria. That end up in the bloodstream, and then what are they doing. So there's been a lot of work on these kind of like bile, secondary bile acids, for example, amino acid metabolites and sort of fatty acids that get produced. But we started to think, okay, if they're producing these things, what about enzymes? They must secrete enzymes and probably proteases, and what could they be doing? And so we thought candidate receptors would be things like this, G-protein-coupled receptors. That may get processed by proteases that are secreted. So they'd come out of the bacteria and then start cleaving host signals on the inside lining of the gut. And so that's what Marcus started to focus on. In particular, we focused on this class of G-protein-coupled receptors called PARs, which are protease-activated receptors. And they're a very cool family of signaling molecules where they have G-proteins on the inside that signal. And the trigger for the signal is actually a ligand that's on the outside. That has this N-terminal little peptide, and when a protease comes and cuts that, it produces a new little fragment here that's still tethered to the receptor, and it can now bind to itself. And now it turns on signaling, and then eventually the cell has to recycle this because it's basically been cut and turned on, so it turns it off by internalizing the receptor and degrading it. There are actually 4 PARs. The one I'm gonna talk about mainly is PAR2. It's predominantly cleaved on the host side by trypsin and factor Xa, whereas some of the other ones are cleaved by thrombin and are involved in clotting and other things. PAR2 is kind of interesting because it is involved in inflammatory signaling and pain response, which we think would be particularly interesting in the context of inflammatory bowel disease and other conditions in the GI tract. So what is known about PARs and IBDs? Well, they're definitely prevalent in the GI tract, and they've been shown to regulate barrier function as well as involved in signaling. So barrier function is how those tight junctions are formed in epithelial cells in the lining of the gut. And if those get leaky, then you get bacteria into your bloodstream. You get all sorts of issues. And there was definitely at least an example of one bacteria, this E. fecalis, which produces gelatinase that can cleave PARs. So we knew that there's the potential there for there to be proteases. So could we identify bacterial proteases that modulate PAR signaling? You know, there's potential different ways they can trigger signals. They can turn on the receptors, so they could cleave at the native, like, trypsin site, for example, and turn it on. Or they could cut at a new site and deactivate it, right? If you cut the ligand right off, now it won't signal. And there's the potential for production of actual ligands that don't care about this receptor piece here, and they can actually just trigger directly signaling. So you might be able to cleave a peptide and produce a fragment that signals. The other thing is that there's what's called biased signaling, which is particularly interesting about PARs. If you cut that tail piece at different locations, you can actually turn on different signals inside the cell. So it has the potential for the bacteria to produce a protease that's very specific for a certain cut site, and that gives a certain outcome that the it's beneficial for the bacteria. Okay, so how did Marcus go about doing this? Well, he kind of did it the hard way. We thought, all right, well, let's make a substrate that looks like the PAR N-terminus, and then we can just take extracts from bacteria grown in culture and look for any activity that cleaves that substrate. Well, that would be easy to just synthesize a short peptide, but the problem is we didn't know where the cut site is, so we had to make this whole protein. And if you really want to be careful, you're going to express that whole N-terminal domain, which is between 4 and 10 kilodaltons depending on the PAR that you're looking at. And that's obviously too big. Well, Brad, if he's here, would say that it's not too big to make synthetically. But we decided that the easiest thing to do, which turned out to be not that easy, would be to express it. And so Marcus spent a lot of time trying to express this, which turned out to be very insoluble as a protein, but managed to get it to work finally. The trick was to put a His tag on one end, And then a fluorophore on the other end that he used native chemical ligation to chemically modify the C-terminus with a dye, a small dye, in this case Cy5. And then he created this assay, which works quite nicely, is you take conditioned media. So you have to grow up all these commensal strains. And then you take the conditioned media. Let's see. And let me see if I can get this. Maybe I need to go here. Anyway, the pointer is not working. You grow conditioned media and you collect these from individual strains that you're growing under anaerobic conditions. If they have a protease present, you put them in with the probe, and what happens is the probe will cleave— will get cleaved if there's a protease. If there's no protease, the probe stays intact. And because it has this His tag on it, you do a pulldown step, and the pulldown gets you basically all of your molecules stuck to the beads, and now your supernatant is totally clear. So essentially, the assay allows you to then determine the difference between an extract that contains a protease and one that doesn't. And when it does contain the protease, you're going to basically get a signal in your solution that you can monitor just by measuring fluorescence. So to do this, we originally thought, OK, how are we going to screen a bunch of microbes? Luckily, we had collaborators, in particular Justin Sonnenburg at Stanford, who curated a library of commensal bacteria. So these are anaerobic bacteria that they grow in parallel using deep well, 96-well plates. And so we could actually grow up all his whole select collection, which is in this case was close to 200 strains. We take those conditioned media and then we use those to measure that activity using the assay I just showed you. And this is kind of how the data looks. You can see MM is mega media, then there's the trypsin signal next, which is a very high signal because it cleaves the PAR receptor very nicely. And then you get a bunch of hits where you're getting signal cleavage. So if you map that activity back onto the sort of family tree here, the dendrogram here, what you see— the blue bars are the amount of cleavage of the PAR2 substrate. And what you can see at the bottom is that there's a huge family of bacteria, and in particular, Bacteroidetes, that have a lot of activity that cleave the PAR2 receptor— substrate quite nicely. And so we ended up with a total of about 50 strains, about 52 strains with greater than 50% cleavage of the PAR substrate. So this is very interesting. It's clear that, that a number of different bacteria can produce PAR processing proteases. So that was exciting. The question is, what are these proteases? Now you have to find them. You have just a functional assay here. So that's where it comes back to our covalent molecules. So it brings back the theme that I've been hammering since the beginning here, is that if you have covalent modifiers, you can now use them to find targets. So we started with these two molecules, again, the FP molecule, which is a fluorophosphonate. So if there's a serine protease, it'll probably get inhibited by this molecule, FP-Alkine. If there's a cysteine protease, it can almost certainly get inhibited by a class of molecules here called E64, which again is very similar to the molecules we were playing around with back in the early 2000s at UCSF. So, this still is paying out dividends. So, here is the FP treatment. So, we took the 52 hits, we pretreated those extracts with the FP molecule, and you're looking for the bars to go down. And the ones where the bars are low means that they probably are producing a serine protease that was inhibited by our molecule, and those are the ones we decided to focus on initially. We did the same thing with E64. We got a couple hits there that are going down specifically with E64. So if you focus on the serine proteases, we decided to look at Bacteroidetes fragilis, so B. fragilis. That one was particularly interesting to us because it's known as what's called a pathobiont, which means that it can survive inside the gut as a sort of normal, healthy part of the flora. But if it takes over, it can become pathogenic. So that would be interesting if it uses signaling through PARs as a way to trigger pathogenesis. And then we also started looking at these Bacteroidetes cellulocilidicus, which produce what looks like a cysteine protease. So this just shows you we can take our initial probe again. This is the same probe that now is in clinical trials for cancer imaging. We just use that exact same probe with the Cy5 on there, and we can put it into the conditioned media and look for labeling. And what we see is a number of bands. Of activity. These are likely cysteine proteases that are competed away by pretreatment with E64. And so that is— let me see if I can use this. No, it's not going to work. Yeah, so we now have a number of candidates here. We've now undergone the process of doing proteomics, much like what we did for the original FP proteins in Staph. We do the same thing here with this probe. And we now have some candidates which we literally identified a couple weeks ago and we're now trying to validate. So I'm not gonna talk about those, the cysteine ones. We've done a lot more looking at the serine protease targets in B. fragilis, so I'm gonna talk about those. Marcus had an inhibitor that he had synthesized in the lab that he calls MLP-7. It's a chloromethylketone. This compound nicely inhibits the activity we saw in B. fragilis. And so you can see again, the blue bar is just DMSO-treated. You get a nice cleavage of the PAR substrate. And in red, if you put in FPL kinase, you can block it. And if you put in the MLP7 in green, you can completely shut down that PAR2 processing activity. So this is a great inhibitor. We can use it as a competitor for our proteomics analysis. And so the first thing we did was just try labeling extracts. You can see if we just take extracts that had not been pretreated and we put the FP probe in, there's a number of bands that label here. If you now pretreat with MLP7, it's a little hard to see, several of these bands disappear. Here you can see it a little bit with the arrows and higher contrast. There's 3 bands in particular at different molecular weights that all competed away when we added that MLP7 probe. And those are the ones that we think are the critical ones involved in PAR processing. So we did the proteomics. Marcus was able to prep these samples. And you can see here we're looking at a volcano plot. So as you move to the right here, that's fold enrichment. So the more to the right shifted things are, the more abundantly labeled they are with the probe. So, there was this one, Q5LDF9, which is basically just a notation of the location of this gene. It is— it's highly enriched and it has a good p-value, so that's probably a target. We then did competition, and both two targets showed out. So, now to the right, you're looking at how much it's competed away by the MLP7 inhibitor. And again, Q5LDF9 was the most competed of all of them. The second one Q5LI A5 was also a pretty good candidate. So we focused on those. The other ones didn't make a lot of sense, and they weren't getting as much signal. So we decided to focus on these two as a starting point. So what's known about these two? Well, the one that I'm just going to call A5 is, upon BLAST analysis, it really had only one homolog that we could find in P. meridae. And it basically has some homologs in other Bacteroidetes, but it's very low homology. It has what looks like a very uncommon S41 protease domain fold. And what was also really interesting is it has a clear lipidation sequence, which looks like it's secreted and then lipidated on the N-terminus, which would make sense if these are actually membrane-associated proteases that are cleaving a membrane-associated associated receptor. When we looked at the larger one, the F9, it's a higher molecular weight protein, but it also had the S41 peptidase domain and the N-terminal lipidation sequence. It had even less homologs. We couldn't find anything outside of B. fragilis that's homologous to this thing, which is quite interesting. We did the AlphaFold analysis, as can be expected, and they have similar types of fold. You can see here. The F9 and the A5. And the AlphaFold was a pretty high confidence score based on the Pmeridae structure that had been solved. So we were pretty confident these are actual pretty good models of what's going on. You can see in green, the F9 has this extra domain that is— we don't know what it would be doing. So we overlaid the two, and they seem to look very similar in their fold. And in fact, they have two residues that would be consistent with the same serine protease. They have the serine and then a histidine, which would be a base. We couldn't find a third residue. Usually serine proteases have an aspartate as well as part of their catalytic triad, but we couldn't find one, so we don't know if this is just a dyad. We tried expressing these things to do some biochemistry, and when, when we try to express the full-length with the lipid domain, you get nothing. They just don't express at all. And then when you cut off the lipid domain, they express really nicely. We were excited. These proteins were coming out in like milligram quantities from a liter of culture. But unfortunately, when we tried to look at these things for activity, they showed no activity, even though by CD spectra they appear to be folded. So we think that the lipid sequence is really critical for these things. They might form multimers, we don't know. But simply cutting off the, the N-terminal lipidation and expressing them wasn't good enough to get activity back. So we had to turn to genetics, and Marcus had to learn how to do genetics in B. fragilis, which isn't all that easy. In fact, they don't transfect well, so you have to basically do a conjugation where you put your construct to knock out your gene of interest into E. coli, and then you grow the E. coli with the fragilis under anaerobic conditions, and they basically exchange plasmids and you get knockouts. So he was able to get that working and was able to knock out both the F9 and the A5 genes. And the first thing we did was just grow those strains after we confirmed they were in fact knockouts, and they grow totally normally, no defects in growth. Then the question is, what happens in our PAR processing activity assay? So again, we took our PAR substrate. This is just wild-type B. fragilis to remind you that its activity is in blue. Inhibited activity with the FP probe is in red. If you do that same assay now in the A5 knockout, you just see virtually no difference. You still have activity. It's inhibitable by the FP. If you do the same thing with the F9 knockout, you see you totally lose the PAR2 processing. Activity. So this F9, which was the best hit of the proteomics, is in fact the one that's processing our substrate. It's just telling us our substrate is being cleaved by this protease, which is good. But the question is, is that relevant? So that's when we turned to Nigel for some help, because he had really nice assays already set up in his lab for looking at PAR signaling. And one is a primary neuron assay where you look using a patch clamp measurement, and you're measuring basically stimulation to fire action potential in the neurons.. And the first thing you wanted to do is make sure that the wild-type B. fragilis media has some effect. And so you're looking at as the bars get lower, they get a strong— that's a stronger signaling or nociception in the neuron. So you can see in medium, they have very little signal. If you throw in vehicle-treated B. fragilis supernatant, that's the blue bar, you get a very strong response. The neurons fire more often. If you do a PAR2 knockout at the far end in pink, you can see you can go back. You lose that increased nociception, which tells us it's PAR2 dependent. So Fragilis is turning on signaling in neurons, and it's PAR2 dependent. The AZ molecule is just an inhibitor, AstraZeneca inhibitor. And then FP was our inhibitor of the serine protease. So if you treat the extracts, you get loss of the nociception. So that means that at least in primary neurons, we're seeing it in wild type B. fragilis. So what about the knockout? The good news is the knockout nicely phenocopies what you're seeing with inhibition or PAR2 knockout. You can see the wild type again is in blue, and in red is the knockout. You go back to essentially normal levels of signaling in the neuron in the knockout. So it's telling us really that one protease is definitely what's responsible for giving us this increased signaling in the neuron. So then we moved it in vivo. This is sort of the simplest mouse model we could start with, and it's basically you intracolonically inject the supernatants from these bacteria. And then you use this thing called a von Frey filament to kind of poke the belly of the mouse, and they get a response. And the sort of the graphs look like this. You can see again, here's the wild-type mice in blue and the knockout, PAR2 knockout mice in red. And what you can see is you get a strong response from just wild-type fragilis extracts, and that signal goes away when you knock out the PAR2. So it's a PAR2-dependent signal. And then when we do this thing again, now using the the wild type or the knockout for our protease, you can see the wild type gives the blue signal. The knockout looks just like sort of the background that you see in the PAR2 knockout. And so basically we think that this thing is really inducing a response. We can also do histology on these mice because they're GFP tagged on their PAR2 receptor. And what you can see by GFP signaling is if you put wild type Fragilis supernatant into these mice, they get internalization. So this signal you see here is kind of diffuse. It's because it's in— these little vesicles. So you've turned on PAR2 and it's been recycled and being endocytosed. If you now do this in the knockout, you see that there's very clear retention of PAR2 at the surface. This is how it should look prior to being stimulated. So we really think that these things are turning on PAR signaling in vivo, which is great, and it's a single protease. So that's pretty much what I wanted to share with you guys, and I'll just summarize briefly. That there's a number of gut commensals that produce PAR cleaving proteases. And we think that there's probably more. And we're trying to understand the biology of this. But we do think they could be interesting targets from a therapeutic perspective. OK, so I just want to thank everyone, collaborators in particular, Justin Sonnenberg's group, and Casey Huang, and Ami Bhatt, and Michael Howitt, and then Virgent for their help with that, and Intuitive Surgical. And since we're getting short on time here, I'll stop and take questions if you guys have them.

Thanks, Matt, for a fantastic talk. And it's amazing to see all the applications that you have been able to, to use with your compounds. We have time for a couple of questions. If you have them, please wait for me to bring you the microphone for our friends online.

Hi there. I just had a question about the B. fragilis. Do you have a sense for this— if this protease that they're expressing is expressed in the B. fragilis that is not pathogenic and just sort of existing in the gut, or is it sort of correlated with pathogenicity? Yeah, that's an interesting question. I mean, the fragilis, because it's a pathobiont, it means that it has the potential to become pathogenic at any point. So it's probably some triggers within the gut environment that make it go from being healthy to pathogenic. So it's going to express this. Thing all the time, whether it's secreted or— we don't know. I mean, there's probably other ways that it's regulated because it's a secreted protein. So yeah, we don't know what would be the triggers there. But we think that what probably is going on is that if it secretes this protease and it triggers PAR signaling, it can induce inflammation. And that high inflammation can actually be beneficial because B. fragilis can grow under fairly high oxygen tension, whereas a lot of other bacteria can't. So it may be a mechanism to keep itself sort of dominant in the niche is what we think.

Thank you.

Yeah.

Anyone else? Oh, Dr.

Keesling. Excuse me. Thank you. Thanks. Thanks, Matt, for that interesting talk. I'm curious about— so you presented two potential roles for these peptides, right? One role could be that they're basically agonists of the GPCR, and the other is their potential antagonist.

Do you have any sense of whether there are antagonists and whether there's some competition Yeah. I mean, that's what we think with the Cellulosilidicus. Actually, I don't have enough data to show it yet, but we've started to try and map the cut sites for these proteases on our little receptor— I mean, the PAR N-terminus. And it looks like the one from Fragilis is cleaving right by where the trypsin cleavage happens, so we think that's an activator. And that fits with all the VO data we have now. But Cellulosilidicus one looks like it's cleaving inside the normal peptide. So I think it's a deactivator, but we don't know 100%. But it would be really cool because then it would be sort of a battle of, you know, do certain bacteria cleave those receptors to prevent them from getting triggered? Yeah, I think that that probably is what's happening.

I mean, it's got to be kind of an arms race among bacteria in the gut, right? Yeah, I think that's a really interesting model too, because as you mentioned, you know, you don't want to flip the balance in the gut, and that could be really what's going on.

It's really interesting.

Yeah.

OK.

Oh, is there one? Oh, Emily. Sorry. Hi, Matt. That was a great talk. Hey, thanks. So one of the most abundant or frequently distributed classes of biosynthetic gene clusters in human gut microbes is actually a group of gene clusters that's predicted to make peptide aldehyde natural products. So I was curious if you had looked at those clusters or thought about seeing whether there might be natural products that interfere with the activity of these enzymes.

No, that's a great suggestion. I mean, it would be very cool if these things are inhibitors of serine protease. Yeah, yeah, I mean, it's definitely something we should look at. I mean, that's one of the other things that we could start to look at is in community setup. Casey's set up really nicely to do these artificial communities to ask the question if we co-culture one with another, does that shut down the PAR processing activity? That would probably be a good start, but looking at gene clusters I think is also a great idea. Thanks.

Thanks. Well, please join me in thanking Matt for a fantastic talk.