Scientist Stories: Thomas Lanyon-Hogg, Chemical Biology in Drug Discovery

Hi everyone, I think we're going to start now. Um, thank you for coming for today's talk. Today, um, we welcome Dr. Thomas Leung Hong from the Department of Pharmacology at the University of Oxford. His research involves the synthesis and development of biologically active small molecules, conducting research into chemical biology as well, involving the use of organic synthesis bioorthogonal click chemistry, and biochemical proteomics analysis to generate novel platforms for modulation of protein-protein interactions. In his talk, Dr. Tom will describe the use of interdisciplinary techniques in the development of molecules to combat cancer and antibiotic resistance. Specifically, he talks a little bit which reduces growth and is preactivated in certain cancers. Another topic of the talk, it will be into modern medicine on antibiotic resistance. In order to overcome bottlenecks in biological testing, Python-automated assays were established for compound analysis and machine learning was applied to accelerate optimization of biomimetic systems. And without further ado, let us put our hands together for Dr.

Zhang. Thank you very much. Good afternoon, everyone. I'd just like to start by thanking the organisers for the invitation to come and talk to you today. So as I mentioned, my talk is going to cover the use of chemical biology in both antibiotic and cancer drug discovery. Okay, so just to start off with, this is my research group. We're basically interested in the use of an application of chemical methodology to try and understand biological processes and also to develop new small molecule therapeutics. So we have projects that are targeting hedgehog signaling in cancer and bacterial DNA repair mechanisms, and we're also now doing quite a bit of work into trying to identify new cardioprotective targets for small molecules. I should just like to start by acknowledging as well the group of people on the left who are— who were the first people actually to join my lab, which is quite a brave step to join a lab with no money and no members in the middle of a pandemic. Not sure I would have done that. And the people on the right who are the ones who have actually conducted much of the work that I'm going to tell you about today. So we'll start off by looking at hedgehog signaling in So for anyone who's not come across the hedgehog pathway, this is one of those fantastically weird and wonderful biological pathways that controls growth and development. So this is an example of a cat that has had its hedgehog pathway overactivated, and this has basically caused overpartitioning of the cat's face. So this is what's called a Janus cat, which is obviously named after the Roman god of two faces. Conversely, you can have the opposite situation.— so this is a man whose hedgehog pathway has been inhibited, and so this has prevented partitioning of the face. And the compound that causes this inhibition is called cyclopamine, which is obviously named for the cycloptic phenotype. So I think it's really interesting to see the biology underpinning classical mythology. As this pathway has evolved in growth and development, it's obviously switched off in most healthy adult tissues. And perhaps unsurprisingly, we also see that the hedgehog pathway is then reactivated in certain cancers. Sorry, my pointer isn't really working. So on the left, N is normal tissue and T is gene tissue. And overexpression of these hedgehog pathway genes tends to correlate with a very bad prognosis in cancer. So for example, in pancreatic cancer, where the 5S5R is only about 5%. So the most famous hedgehog signalling protein is named after arguably the most famous hedgehog, which is Sonic hedgehog. So I'm going to be talking about Sonic today. And if you look at how this biological pathway actually fits together, so Sonic is expressed as a precursor protein, and this then undergoes autocleavage and attachment of a a cholesterol group at the C-terminus. So the N-terminus of the protein is then palmitoylated by an enzyme called heteroglycerol transferase. So this is attached to the bottom of a 16-carbon fatty acid. This gives the mature signalling protein, which is then secreted from signalling cells, and this then goes on to dock at signal-receiving cells at the membrane receptor patch. This releases which inhibits and smooths and activates the expression of Hedgehog target genes. There's been a lot of interest in the pharma company in driving this pathway, so Bisumograjab was approved in 2012 for treatment of basal cell carcinoma. However, its efficacy is rapidly compromised by the emergence of resistant mutations. So there's therefore a need to identify new inhibitors of this pathway, and our interest has been in inhibiting HAT. So if we look at the mechanism of HAT's action in a little bit more detail, so it attaches acetic acid from palmitic acid onto the N-terminal cysteine site, and initially it's a thioester, and this then rearranges by an ester-N-acyl shift to give you a final amide product. Now, HAT, so itself is a transmembrane protein, so this makes it particularly challenging to study because it's often very difficult to express, purify, and analyze these kind of proteins. So we did a lot of work initially just by inserting protease cleavage sites into HA to try and figure out what its membrane topology looks like, so how it threads through the membrane. One of the interesting things from this work that we found is that actually the key residues that are required for catalysis, so histidine, SPARTA-339 were actually located on opposite sides of the membrane. And one of the final pieces that slotted into place a couple of years ago was actually how this enzyme acts as a site for reassembling. So a bacterial family member related to HAP was solved, and this actually shows that those transmembrane helices kind of assemble into a donut with the active site in the middle. So there's a lot of things that make HCaP actually a very attractive drug target. So firstly, palmitoylation is critical for hedgehog signaling. Secondly, if you can block this palmitoylation, you're actually stopping the cancer cell from signaling to the healthy surrounding cells. So that means that, for example, you have to inhibit less cells than you do with a smooth end inhibitor. And lastly, HCaP and Sonic are proposed to in an exclusive enzyme-substrate relationship. So what that means is that the only substrate for HAT is SONIC, and the only enzyme that pumps away SONIC is HAT. So potentially this could give us quite a low chance of mechanism-based toxicity. So as with any good idea, once you've got a good idea, someone else has normally already done it. So Marilyn Rash's group were the first to publish a series series of HAT inhibitors which they'd identified via a scintillation proximity assay screen. So it's these compounds here, RUSK41, 43, and 201. So in collaboration with a group of Ed Taylor and Cho-Li He, we set about trying to firstly identify, synthesize, and characterize these inhibitors, and secondly to use them as probes to actually see what we could understand about HATs. Structure and function. So I'm not going to talk about the synthetic chemistry, how we made these, it's all detailed in these papers at the bottom if anyone feels the urge to look at those. So instead I would like to talk about biology and how we can actually analyze these compounds. So one of the first things we did was we set up what was called a like-2 signaling assay. So this has 5,5-luciferase under control of the gliatin activated hedgehog transcription factors. So essentially, when the hedgehog pathway is activated, it produces luciferase, and you can measure that lightly. And what we find is that when we take cancer cells secreting parmitolase in sonic and then we dose in our RUSTI inhibitors, they do indeed inhibit the 5'-luciferase fluorescence. So that all makes sense. What we were quite surprised surprised to see those, that when we used a Smoothened agonist, so one of these compounds that will activate the Hedgehog pathway, that two of the Rouski compounds still actually inhibited signaling at higher concentrations. And this, we shouldn't be seeing this because Smoothened is downstream of where those compounds should be acting. Whereas Rouski 201 actually turned out to not inhibit the pathway under these conditions. And as it turns out, this is because that's general nonspecific cytotoxicity of the compound showing roughly at the concentrations where it starts losing the pathway. So having identified RUSP-201 then as an on-target tool molecule, we next used what's called palmitoylation profiling. So here we take a 16-carbon fatty acid that incorporates a palmitate. So the cells will take this, they'll process it, and out of the would do the native lipid, and they'll incorporate it into different palmitoylated proteins. We can then use a click reaction, so bioorthogonal reaction, to label our alkyne probe with any enzyme functionality. What that allows us to do is to selectively, over all of the other proteins in the cell, enrich the palmitoylated proteins. And we can digest these and sequence the peptides to identify the which proteins were pulled down. And we can also combine this with our Hedgehog isotransferase inhibitor. So what this allows us to do is basically build up a global picture of how every phosphorylated protein in the cell changed when we treated it with an HAT inhibitor. And remarkably, the only protein that phosphorylation status changed was the SONIC, which might seem quite surprising that that's so selective about absence was entirely what we were predicting, that the only enzyme that was acting on sonic would go with the hedgehog. So this meant that we now had an on-target, non-toxic chemical probe that we could use to block hedgehog signaling from cancer cells. Now one of the other key things that we needed was a biochemical assay to allow us to look at the activity of purified hedgehog. Now, one of the things that I realized here, when HCAP attaches a 16-carbon fatty acid to it, so for example a peptide from sonic hedgehog, that lipidating product is not going to be water soluble. It's not going to float around in a buffer solution. But if you have a detergent in there, the lipid is going to insert into the detergent micelle. Now, the reason why this is important is because Obviously that small peptide in solution is tumbling very quickly. The detergent micelle is massive, it's about 50 or 60 kilodaltons, tumbling very slowly. So when you shine plain polarized light into a fluorescent peptide, when you have attachment of the lipid to that peptide, you can measure this directly just by looking at polarization of fluorescence. And just to show that this works, so this is synthetic substrate sonic peptide in synthetic product, solid peptide. And what you can see is that you get an increase in fluorescence polarization only in the presence of detergents that have formed micelles. Now the reason why this is so useful is because it literally means if you want to look at a lipid transferase enzyme, you just mix it together with a fluorescent peptide and you can monitor that in real time. So this is an example of H-HAP palmitoylating solid peptide. And this then opens you up because it's a very versatile platform, you can do all sorts of analysis. So this is just an example of using different solid constructs to inhibit the labelling of the fluorescent peptide. Now one of the reasons why this is so powerful is because obviously that's just an enzyme and a lipid mixed together. We kind of asked the question inside, well, how general, how generally applicable is this principle? So we tried it for an enzyme called N-mirastole transferase, which attaches to 14-carbon fatty acid, and you again see an increase in polarized fluorescence. You can do the same with CAX boxes. So these are prenyltransferases that work on proteins like the brown sphagnum proteins. And what you see is that, for example, farnesol transferase will only give you an increase in fluorescence when it's matched with farnesol pyrophosphate. And the same with the of the transference. And we also thought, well, can we do the reverse? And we look at removal of lipid from peptides. So this is an example then, if you're just taking your palm phthalate peptide and cutting the lipid off, and you see suddenly you get a decrease in fluorescence polarization. And similarly, you can do this with the actual endogenous Asar protein thioesterases. So it seems that this is a general strategy that is broadly applicable for any lipid transfer centre. So with this assay in hand, we were then able to start digging into the structure-activity relationship of this inhibitor. So what we're looking at here at the top is the solution 01, so 2 micromolar IC50, and then when we remove this side chain, we lose all the inhibitory activity. Few methylate this, you recover some of that activity. And if you shift nitrogen along on position side chain, so you've got a urea linkage there, and you lose a lot of the activity. Whereas on the other side of the side chain, you can kind of shift that little tail around, it doesn't really change too many things in terms of potency. But if you N-methylate this, then you do look like you lose a lot So what that said to us was that basically this inhibitor kind of needs a not too sterically incumbent nitrogen followed by some sort of alcohol chain. So what we could do therefore is we could insert this diaziridine into the side chain of BRU21. So the diaziridine, when you shine UV light on it, it will covalently crosslink to anything that's bound in proximity. And obviously the alkyne just allows us to use bioorthogonal click chemistry to functionalise the protein afterwards. We can then test those probes using the XR-CLIP assay, and we see happily that it did maintain high inhibitory activity as we had hoped. And then lastly, we can take purified hHat, mix that with the probe, covalently cross-link to it, and then attach the fluorophore. So what you see here is that in the presence of UV, Hap has become fluorescent, indicating the probe has caught onto it. So why are we going to all this trouble? How is this going to get us any true information about Hap? Well, so the next step is that we can then take Hap that's been labelled with this probe and we can digest it with a host of different proteases, so we separate it out into its amino acid peptides., and we can then analyse those by mass spectrometry and actually identify which peptide has been modified by our cross-linking probe. So in this case, we found a modification to a peptide at histidine 215, and this was quite surprising to us because, as I've said, the active site residues are known to be histidine 379 and aspartate 339. So what we did in this case was we actually took that bacterial structure that had been solved, and we made a homology model of ATRAP based on that. So what we're looking at here, red is predicted transmembrane domain regions, at the top is the cytosol, and at the bottom would be the ER lumen. And you can see that these assemble nicely into a cluster, and in the centre of that we have the two catalytic residues. And sat just above those in blue at the top is the peptide that was modified by our chemical probe. So not only does this give us an indication of ACE-TAT structure now, but also suggests a potential mechanism of inhibition with the inhibitor sitting just above the active site. And this model was done by our collaborators at Oxford, so Christian Seaborg and Claire Hope. So at this stage then, one other thing just to mention is that we can separate out our— what is our best performing inhibitor and separate that out to two enantiomers. So we've got a really potent inhibitor now with 0.75 micromolar sensitivity. So given where our chemical probe bound, which is on the site selected by the active site, it suggests that maybe these compounds are going to be competing with palmitoyl-CoA binding. And indeed, when we look at the effects of varying concentrations of palmitoyl-CoA CoA on the inhibitor potency, we do indeed see that this is a palmitoyl-CoA competitive inhibitor, and it's actually a highly potent one with a Ki of about just below 40 nanomolar. So just to say, we were very happy with that, that everything was internally consistent. And then subsequently, Christian and Claire actually solved the structure of HAP by cryo-EM. There were some very— some surprising things in it. For example, there was a keen group bound to the protein that we really didn't expect to see and were very surprised by. But Claire was also able to solve the structure with one of our best performing inhibitors bound. It does indeed bind next to aspartate 339 and histidine 379, and the structure actually shows that it rotates and trips the FAN into the palmitoyl-CoA binding pocket. So it's nice to see the chemical and structural biology both arriving at the same conclusion. So lastly, now that we've set up all these synthetic routes, we've got an improved molecule, we then set about a med chem campaign, so synthesised about 50 different analogues of these compounds and managed to determine what the pharmacophore required for each atom position was. So we were kind of getting close to a tool molecule that we could then take in vivo to see what the effect of blocking the tobacco was. However, before we can do that, we need to look at how metabolically stable these molecules are. And it turns out they're completely unstable. So when you mix them with microsomes taken from liver cells, within 30 minutes, 98% of the compound is metabolized. And if you look at the position of these oxidative modifications, say for example on the thiopyridine on the A side chain, you see that actually those are the parts of the molecule which require the target inhibition. So these compounds are never going to make it into an in vivo model to actually show what HAT inhibition will do. Luckily though, we have an assay for HAT enzymatic activity in this fluorescence polarization assay. So I went up to the University of Dundee to form a high-throughput screen to identify new inhibitors of ACEHAT. And the way that we did this is actually to use the cheminformatics-based approach. So there's around about 250,000 compounds in the library at Dundee, and what we do is essentially, if you can think about visualizing a molecule with different descriptors, so like shape, C log P, molecular weight, you can essentially take a subset of of the compounds in the library cover the full chemical space of those 250,000 compounds. So we then screen those using— screen those 20,000 compounds using fluorescence polarisation data. So this is just some data from the screen. So this is what the actual screening output looks like. That's the positive controls— sorry, negative controls are light blue, so very efficient. Positive controls, 100% efficient. The vast majority of the compounds in the library don't do anything, but we're really interested in the 8 compounds that are showing some inhibition here. So we then take those 8 compounds— these are our active inhibitors in the library— and we use chemoinformatics and we essentially select every molecule that was nearby to those hit molecules in chemical space. So that gave us around about 330 compounds, and we then screened using dose-response. So this is just an example. We used RUSK-201 as a tool inhibitor just to check our assays were working, and this got us out 15 new hits across 6 distinct series. So we repurchased and resynthesized those hit molecules just to check that they actually worked, and then we screened them in a real-time assay at 15-20 micromolar. So this is the data, and you can see here RUC2 at 1-25 micromolar is not inhibiting, you know, particularly well compared to our hits. And when we measure the IC50s, so we've actually got 6 series that are around 10 times better potency than the RUC2 series now. We've successfully run a MedChem campaign on this. It's got us to 3 chemical series that are now sort of single-digit nanomole potency. We've got cryo-EM structures, cross-linking tools that allow us in photocrosslink to actually validate that we're engaging HAT cells, and suitable stability for in vivo testing. And my take-home point from this is that HAT was an incredibly challenging enzyme to work on, and it's this interdisciplinary research and the application of chemical biology that has essentially taken HTAP from a highly challenging target to a highly tractable one. Okay, so with that, we're going to shift gears now and move on to antibiotic resistance. So I'm sure you've all heard about this in the news, and it's often called the antibiotic apocalypse, which feels kind of dramatic almost. But then when you think about it, antibiotics do fundamentally underpin a lot of modern medicine, right? So without antibiotics, you have no surgery, you have no cancer chemotherapy. And if you think about the knock-on effect of losing those treatments, then what happens if this affects life expectancy? Well, then things like neurodegeneration, are they such a medical priority? Things like diabetes, if people aren't reaching old age to suffer from these conditions, So you suddenly start to see that actually losing antibiotics is a really massive deal, especially because bacterial infections, you know, they don't just affect old people. We all get things like food poisoning, right? This is not going to be the case like, you know, the recent COVID pandemic where it was really elderly and vulnerable people at risk. This is going to be across the whole of society. And to be honest, if the last pandemic should have taught us if anything, it's that we have to proactively work to address these emerging threats. We can't wait for the crisis to come to you. And sadly, when we look at antibiotic development, there's really been nothing in the pipeline for the last sort of 50 years or so, since what was called the golden age of antibiotics. And that's not good because trying to bring a drug through is a sort of 10 to 15 year process, so we need to be acting on this now. And there's a further challenge that obviously any new drug that hits the market, resistance emerges. And if you look at the more recent cases, so for example daptomycin, which is on the market, resistance emerged almost as soon as that was approved. And that brings us to another layer of this problem, which is that essentially there is no financial incentive for making a new antibiotic because you can't sell it. So the last companies that have released antibiotics onto the market, for example, Akiogen, was valued at $500 million, received FDA approval to put its antibiotic onto the market, and the company went bankrupt 3 months later. And it's not just Akiogen, this happens repeatedly. So we've got this weird dichotomy because we have to keep an antibiotic as an agent of last resort, right? The arguably most valuable medical product on the planet, there is no financial incentive for making it because you can't ever sell it. So this is where my group is trying to contribute, and what we're trying to do is that we're trying— we're not trying to develop a new antibiotic, we're trying trying to develop antibiotic adjuvants. So these are compounds that can be paired with existing antibiotics to either reverse resistance where it exists or slow the emergence of resistance where bacteria is susceptible. So our target is the bacterial DNA repair processes. So if you think during an infection, bacteria is subject to DNA damage, so this could either be attacked by the host immune system or by the action of traditional antibiotics such as Prilosec. So bacteria have obviously evolved mechanisms to survive this DNA damage, but not only does this lead to the survival of bacteria, it also activates what's called the SOS response, and this literally sends the bacteria into a state of hypermutation. It promotes horizontal gene transfer, causes bacteria to shut down and form dormant cystic cells, so So there's a whole host of pro-inflammatory phenotypes. But this is where we think if we can develop molecules that inhibit this DNA repair process, we've got a lot of potential advantages. So firstly, these potentially promote immune clearance, and that would leave the healthy gut microbiome unaffected. That means there's less chance for a secondary infection, something like Clostridium difficile. We could also sensitize resistant bacteria to DNA-damaging antibiotics, or potentially even decrease the rates of mutations of bacteria that are not yet resistant to an antibiotic. Because these targets are quite highly conserved, this should be a very broad-spectrum approach, and obviously these compounds, DNA repair inhibitors, don't need to be kept as an agent of last resort because they can be compared with all existing antibiotics. So hopefully then that gets us around at least part of the challenge of the limited financial market. Now I should say that AMR is a multifaceted problem. This isn't going to be the only solution, but we need multiple different routes being progressed. So in terms of the target, you're going for in bacterial DNA repair, this is an enzyme called ADAMAB in Gram-positive or Gram-negative bacteria. So essentially it's a helicase-UTase enzyme. So this is the structure. Basically what it does is it binds onto the second strand of DNA and essentially it pulls apart the DNA duplex, so over a central pin here, and feeds that through to the nucleases, which chop up the DNA. So in terms of the actual double-strand break repair pathway, so So when bacteria has a double-strand break, AddA/B/C/D load onto this and they digest the DNA back until they encounter what's called a kai site, and this causes the complex to rearrange and generate a strand, a single strand of DNA on the 3' end. And it's that single strand of DNA that then has a protein called RecA loaded onto it, and RecA is the protein that actually initiates this molecular combination of DNA. There. Now at the same time, RecA interacts with a protein called lexA, and lexA essentially controls the expression of different genes. So when lexA fuses after interaction with RecA, that releases this SOS box, and under control of the SOS box are all the hypermutation, so the aeroplane polymerases and so on, proteins involved in gene transfer. Now these complexes these are monsters, they're really challenging, very large, they're about 350 kilodaltons, multifunctional enzymes. But that does give us the chance that because there are so many functions, there's a lot of potential where we can inhibit these proteins. So just to show some biological data first, so to actually show that these targets might be good. So this is some data with an APPB transposon insertion mutant that just basically needs a complex knockout. And so top left is showing killing by whole human blood. So when you lose— when bacteria loses AdB, it becomes more sensitive to killing by human blood and immune cells. Similarly, we can look at rates of mutation. So mutation just being induced by oxidative stress, and then when we knock out AdB, bacteria lose the rates of mutation. So it's now. We can also test for how sensitive these bacteria are to the DNA-damaging antibiotic. So in this case, ciprofloxacin. And what we can see in green is that the, the AddAB knockout mutant is more sensitive to the antibiotic. And actually, quite surprisingly, when we test this mutant against a whole host of different antibiotics from across every major class that we had, we actually see that the, the AddAB knockout mutant is sensitive to a really wide range of antibiotics, and almost surprisingly, some of these aren't even known to be involved in DNA damage. You know, things like cephalosporins and, you know, beta-lactam antibiotics shouldn't be acting on DNA damage. So we set about to try and identify inhibitors of this enzyme. As I've said, if you've got a good idea, someone else has already done it. So this has been done by the lab of someone who's now a very close collaborator of mine, Jerry Smith, and his group identified this compound ML321. So the part of the molecule highlighted in red is actually what's called pyrimidine acid. So this is a known inhibitor of DNA gyrase and DNA polerase, and it functions via intercalating into the DNA double helix. Now this obviously gave us a potential mechanism by which these compounds are they sticking into the DNA and jamming up the enzyme that way? So what I meant is that potentially you could be testing scaffold compounds. So this section of the molecule shown in blue, which is called ciprofloxacin, so this is basically an analogue of piperidine acid that is 130 times better at intercalating. And then we also did some classical med chem just to optimise that trifluoromethyl substituent position, usually what's called the top of the substituent tree. Okay, so then testing data. So we take compound ML328 and we use this, and we use this in a killing assay to see what concentration is required for killing of MRSA. So that's a hospital-derived superbug. Compound has an IC50 of about 6.9 micromolar., whereas when we give bacteria a sublethal dose of DNA-damaging antibiotic, the IC50 decreases to 1.1 micromolar. So nothing particularly exciting at this stage. We tried it with our optimised compounds, this compound called IMP1700. Compound on its own treating MRSA, 3.8 micromolar IC50, but you give the bacteria a sublethal dose of ciprofloxacin to reduce DNA damage, and the potency of our compound increases to 5.9 nanomolar. Now that's just about one of the most potent kills that I've ever seen in an assay, and that's killing a hospital-bred superbug that was resistant to ciprofloxacin. We can also do what's called quantitative drug synergy analysis, and essentially this just allows us to show that particularly for IM-1700, we're seeing interactions. So this is the two compounds amplifying each other's potency. It's not an additive effect between them. Now, one of the things that eagle-eyed amongst you might have noticed is that the IMP1700 molecule contains ciprofloxacin, and we're also using ciprofloxacin to reduce DNA damage. And that might well lead people to say, well, haven't you just made ciprofloxacin again? And that's certainly a question that we you've been asking yourselves a lot. So ciprofloxacin inhibits two enzymes called DNA gyrase and DNA topoisomerase. So gyrase essentially supercoils DNA to package it up. So we can actually just take assays for these enzymes that are cool. So this is DNA supercoiling assays. So what we see is that in the presence of DNA gyrase, we're forming DNA supercoils. We treated ciprofloxacin we inhibit the enzyme, whereas IP1700 and ML328 don't inhibit this enzyme. And we can do the same for topoisomerase. So topoisomerase relaxes DNA supercoils, ciprofloxacin inhibits that, our DNA repair inhibitors don't. We can also go a step further now and actually set up a reporter assay for the SRS response. So here we're just hijacking the fact that LexA controls protein expression. So we put a green fluorescent protein under control of LexA, and when that cleaves, we can measure GFP fluorescence as a readout of the SOS response. And what we see is that in red, when you treat with cyclophloxin, you activate the SOS response, whereas when you activate the SOS response with cyclophloxin and then add in the DNA repair inhibitor, you then see that we can inhibit the SRS response back down to pretty much baseline zero levels. And lastly, just to try and get some evidence of target engagement, so we can take IMP1700 and attach this onto a set of those beads, and then we can essentially use a pull-down assay. So where we incubate our compound attached to the beads with cellular lysate that's very overexpressing AdAB. What we see is that blank bead doesn't pull down anything, whereas when we attach our conservative ligand to the bead, we can successfully pull the target out to solution. So that all seems pretty good, I'm pretty happy with that as a start point. So we can potentiate CYPLOXIN activity in clinical MRSA strain at nanomolar concentrations. We don't have any gyros where some raise off-target activity. We can inhibit the mutagenic SRS response, and we can also engage in our putative target analysis. We went a step further and we just showed that this compound is also not toxic to human cells and that it also inhibits both purified ADB and purified RecPcd in syncytiation proximity assay. However, we really don't understand anything about the structure-activity relationship is serious simply because we only have to make 6 compounds following that rational design strategy to get to an antibiotic potency inhibitor. There is also a black box toxicity warning that is associated with quinolone antibiotics. So anytime I've spoken to a pharma company about whether there's any interest in pursuing this, as soon as I say the word quinolone, they say no, they're not interested, and walk away. We also don't see any activity in gram-negative bacteria. So gram-negatives are often more challenging to get small molecules into because they have two membranes protecting the cell. And lastly, also, we don't really understand the mechanism of action of these compounds. The fact that we can engage a target with a lysate to me is not compelling evidence that that's what's happening inside a cell. I would like to see something far more robust to convince me that we're doing what we think we're doing. Part of the reason why I feel that's so important is because antibiotics in general, there's been a field where people have developed toxic molecules and just thrown them into cells and not cared what they're doing or how they're killing bacteria. And obviously if you end up in situations where you're activating hypermutation, that's not strategy for a long-term solution. Yeah, so we set out trying to address this cell penetration issue first. So there was a paper that came out in Nature in 2017 from Paul Pergenroth's group, and they had basically tried to identify common characteristics in molecules that allowed them to get to bacteria. And essentially this came down to putting an ionizable nitrogen these molecules seem to help uptake. There's also these compounds shown here which are called siderophores, and essentially these are molecules that are secreted by bacteria to bind iron during an infection setting and then bring it back into the bacterial cell. So what we're thinking here is we can just take a siderophore, tack it onto our DNA repair inhibitor, and then hopefully we can hijack that import mechanism into the cell. And similarly, the paper came out this year that was actually saying NR328 and IP1700 aren't just DNA gyrase inhibitors after all. So essentially what the authors have done is that they have taken the compounds at an excessively high concentration, incubated that with bacteria, and seen where a mutation that conferred resistance had emerged, and that mutation happened to be in DNA gyrase. Oh, I thought I'd taken this out, but okay, we'll talk about some synthetic chemistry. So in terms of synthetic routes, this was developed by Jacob Fang, who was a Jacob first-year PhD student from the snakes and ostriches group. So essentially we take diethylfoxy-methylene malonoate and attach this onto our amine, cyclize that in diphenyl ether, and that gives us our pre-ring 4. And then we can alkylate to introduce diversity there. We can then do an SNAr to attach the perazine ring, hydrolyze off the ester, and then attach on an isothiocyanate. That gives us our sort of final analogues. We can also functionalize, so for example we can alkylate on the nitrogen at the thigh and rear We can obviously use that carboxylic acid to attach on esters and amides. We can also hydrolyze the ester first and then basically take dienes where one of them is Boc protected, so we can attach that on by an SNAr to vary the pairing of the linker, and then obviously take off the Boc group. We've done a lot of work validating different methods for styrogears as well. And then to dig back further into the quinolone core, so this is from cyanocoumarin to aminobenzoic acid, and yeah, obviously we can attach onto that acetyl-CoA esters and amides to give us some diversity. We can also use isocyanate cyclized round to give us the quinaciline dione either with or without, um, protonization at this position to introduce carboxyl cation. And we can also do condensation with amides just to form this quinazinone core, and then obviously we just do the same SNAr on each of those to introduce the left-hand side of the molecule. Yeah, we can also remove the carboxylic acid altogether. So this time going after a slightly different methylene malonate analogue that basically means once we cyclize, that then we lose the carboxylic acid disposition. This chemistry does mean that then we end up with a mixture of these two regioisomers that we couldn't separate, but fortunately when we then took that into the SNAr reaction, what we found was that the para position reacted preferentially, and that allowed us to get rid of the other read about it. So a lot of this work was done by Gabriele, and he also looked into just other electron withdrawing substituents, that para position that you could access quite easily. So that gets us to our compound panel. So we synthesized about 90 analogs of these just to try and understand the structure-activity relationship. Now, one thing that we're very keen on doing in our group is is actually trying to figure out where the bottlenecks are in our process. We then asked ourselves, what is the one thing that we can do that will make all of our work run 10 times quicker? And we realized pretty rapidly that we were relying on a lot of other labs to test our compounds. This meant that we could synthesize analogs, but then we wouldn't get data back for 3 months. So Jacob is joined with a lab of Professor Tim Walsh, who leads the within the Oxford Institute and was really instrumental in getting— rather than going from sort of 96-well, very simple assay, typically antibiotic assays, you're basically looking at a plate by eye and seeing how bacteria are growing there or not, which seems a little archaic, to say the least. So what Jacob did was he switched this onto an automated liquid handler, so this thing was Biomech i7, shifted the assay format to 384-well plates, and switched to reading those by optical density measurements in the plate reader. And that meant that we went from needing to wait 3 months to get data to being able to run roughly 5,000 samples per day. And that's now what the Institute— that's their go-to strategy for doing this, which is nice to see. So this is what the data from those compound panels animals look like. So just to kind of talk you through, because I know this is a bit of a noisy graph. So what we've got on the x-axis, so this is basically the ability of compounds to potentiate DNA damage. So how many times more potent do the compounds get in the presence of DNA damage? And that's obviously on a log scale. So for example, these compounds up here, you're looking at 500-fold more potent DNA damage. On the y-axis is the SRS response, right? So basically what we're looking for there is the higher up on the y-axis something is, the better it is at inhibiting the SRS response. And what we were expecting was that, you know, we kind of see a, you know, sort of pretty linear correlation. The better something is at inhibiting, you know, potentially DNA damage, the better it's going to inhibit the SRS response. And it's kind of— it's what you can see that sort of up in the top right of the got several compounds that potentially DNA damage in the protein SOS synthesis. And these compounds are coloured depending on where the modifications were made in the structure. So you can kind of see up in the top right, we've got a lot of cyclopropyl analogues and some thiogrear analogues as well. What we're really interested in was what happened here, where we've got a couple of thiogrear analogues and non-chlorophyll analogues.

Way.

Those are compounds now where they show no cellular toxicity, even with or without sublethal DNA damaging and arachidonic acids. But they are very potent inhibitors of the SLS response. And this kind of— this raised the question to us, does that mean these molecules are inhibiting two targets, or does that mean maybe they're inhibiting two functions of the same target? It's something we just don't know, but we're quite surprised that now we have one set of compounds that are entirely non-toxic that can block the emergence of resistance. So yeah, just to show you some examples, so these are some of the compounds. So OXF077 is roughly twice as potent as the best compound IP1700, whereas other compounds such as OXX045, this is where we've introduced modifications to improve the stability of that thiol. This is an example of a compound, so OXF114, so where we've still got inhibition of the SRS response with this compound. This is one of the compounds that shows no toxicity regardless of whether there's DNA damage or not. And interestingly, you can see that we've actually managed to remove that quinolone group that was the one associated with that So we're still, we're running in silico analysis of these compound datasets now to try and better visualize the pharmacokinetics. However, we still haven't managed to get any activity in gram-negative bacteria, and we're assuming this is still due to the lack of cell penetration. So Adam, who is a postdoc in the group, has been setting up a mass spec-based assay And this is essentially just to allow us to see, are these molecules getting into cells or not? So the way this works is that he takes the bacteria, incubates it with different compounds, he then isolates the bacterial cell, washes this, and then lyses the cell open and uses LC-MS to actually detect whether the compound has got into the cell, irrespective of whether it's actually doing to its proposed target. So this is an example of some known high-accumulating antibiotics. So ciprofloxacin tends to accumulate very well, whereas chloramphenicol doesn't. So we can see that recapitulated in Arabidopsis data on the left. And then when we look at accumulation of some of our design inhibitors, so for example IMP1700 doesn't get gram-negative cells. So that makes sense. We can make a modification following the entry rules. So for example, this compound, OXF025, and that does get us a bit more cell penetration. Not loads, but at least it's going in the right direction. But unfortunately, that compound doesn't produce any effect on gram-negative cells, and that kind of says to us, well, maybe it's not inhibiting the target. Now, switch the para-trifluoro-methyls. So the last part of this thing is that obviously we need now a high-throughput assay. Again, like I said, thinking about bottlenecks, we don't want to be shipping compounds to the United States to get data for RECCO-ECB, which we need to be doing in-house to run quickly. So Adam Winiforth, who's a master's student in the group, has been looking to set up a fluorescence-based assay for breath PCR. So a lot of assays in the literature, classically, they use radioactive radioisotopes like in DNA. It's the kind of stuff that is expensive, it's cumbersome, there's a hell of a lot of paperwork, it's not particularly safe. So we really want something fluorescence-based for this. And I put it to Randy when he was starting this project,— also think about where the bottleneck is in setting up a high-throughput assay. So when we think about quality in an assay, this is actually defined by what's called the Z prime. So the maths behind it isn't too important, but essentially it's a measure of the noise in your positive and negative controls relative to the signal of the assay. And as I said, Z prime over 0.5 signifies an excellent assay. So if you think conceptually about how you optimise an assay, what you probably do is that you, you take the factors that could affect the assay, say for example the volume, concentration of enzyme, substrate, solvent, and you fix the value for all of those factors and then you vary one of them and essentially you see whichever one gave you the highest Z', you then fix that value and you go on to the next factor and do same thing, change them one by one. So this is what's called one factor at a time optimization. And you do that— so for, I think, the screens that we set up, it takes you around about 2 months to go through and actually optimize all those different factors. So what Adam's doing is that he's using an approach called design of experiments. So this was something that I first came across in optimizing chemical engineering, so process development, which we've just been to increase yield. And we went to the company that produces the software to do this, and we said, would you mind if we just saw how many different applications in biology you can apply this to? And if it works, then obviously we'll tell people about it. And they were quite happy with that, so they gave out the software for free and said, go off, see what you can do. But what this allows you to do is basically vary all of those different factors at the same time. To almost take a random combination to cover the potential experimental space. But the outcome is that once you've got that data, you can then model and basically deconvolute how each of the factors work. And this means that in 2 hours Adam can do the same experiment that would normally have taken 2 weeks to do. And we've now actually started collaborations with the Centre for Medicine Discovery, so we've set Adam the challenge of they've been working on an assay for 3 years, haven't been able to get it working, and Adam's going to see if he can do it in one afternoon. So this kind of application of machine learning to biology, I think, is phenomenally powerful and something we should be thinking about, right? Because in biology, we're basically dealing with the most complex system in the known universe, and we're approaching this normally— our commonly used tools are a manual pipette and Excel spreadsheet, you know, need to be embracing new approaches. So with that, I'd just like to wrap up and sum up. So obviously AMR is a massive emerging global challenge, and DNA repair inhibitors could be an important means to contribute to addressing this. So the first-generation DNA repair inhibitors have got nanomolar potency against called GlamRSA, and we're working together with third ground natives. We've seen that these compounds now have a distinct mechanism and structure-activity relationship for both potentiating DNA damage and also inhibiting the SOS response, and that will allow us hopefully to avoid this black box toxicity issue. Something that I haven't really touched on at all, but this is the use of photochemical probes, which some of you now actively looking into to essentially allow us to profile every target that the inhibitor interacts with inside the lifestyle. And lastly, what I've shown you is that hopefully interdisciplinary research can accelerate drug discovery against these challenging targets. I'm a very firm believer that if the challenges we face as scientists are not confined to a single scientific discipline, then the solutions can't be either. So with that, I would just like to thank my group who conducted this work and our collaborators at Oxford Imperial and Craig Hutch, obviously to acknowledge our funders for supporting this work, and finally thank you all for your attention.