 Hello, my name is Uttarorek and I would like to present you today very quickly our quest for small molecule cancer immunotherapy projects together with the bioinformatics and the experimental tools that we use for the design validation and classification. So the target that I will be speaking about today is endolamine 2,3-dioxygenase 1 or idea 1. Idea 1 is an enzyme that breaks down tryptophanes leading to an immunosuppressive environment. It has been shown that idea 1 plays a role in immunity, neural function and aging. And since 2006, idea 1 inhibitors have been developed mainly with a focus on oncology in order to overcome resistance to immunotherapy. It has been shown that this works very well in preclinical models. There's a strong biologic irrational behind it, and everything looks great. Until about 2019, the first results of a clinical trial of an idea 1 inhibitor, apocytostat, was shown to be inefficient in melanoma. So this put a big question mark on the field. But on the other hand, it's still a very active environment field of research and new aspects of idea 1 biology are continuously being discovered. So today, I will speak mainly about idea 1 and show what we have done to address its role in cancer. But I'm confident that most of the tools and approaches that I'm showing can also be applied to other drug targets. And so this presentation will center on the questions on how to design, validate and classify inhibitors for therapeutic applications. And I will speak first a bit about the validation and classification and come to the design towards the end of the presentation. So a few years back, we did an experimental high throughput screening in order to find new idea 1 inhibitors. And two of the hit compounds that we obtained are shown here. We then of course had to validate these compounds and we found different things. For the first one, we found that it showed a very nice dose response curves in the standard essay shown in black here. But when we added a nucleophile to the essay solution, it lost all of its activity, showing that this compound was very probably a false positive acting through chemical reactivity with idea 1. The second compound, which looks chemically very similar to the first one, already showed a somewhat suspicious dose response curve in the standard essay because this curve was very steep. And indeed, when we added a detergent to the essay solution, it lost most of its activity again. And this is a very strong indication of an aggregator, so false positive, which acts through aggregation. Of course, these experimental validation, these experiments are quite time consuming and tedious to do. And so we turned in the following to use cheminformatics filters in order to try to predict compounds that would act through promiscuous mechanisms. And there are a lot of these in the literature right now. And one of the most famous ones is probably the Pan-Assay Interference Compounds Filter or PAINS filter. When, with this experience in the back of our heads, we turned to have another look on the compounds that had been published in the literature as being idea 1 inhibitors. And a lot of these compounds, just a few examples are shown in this slide, look suspicious to us as potentially being chemically reactive, redox cycling, iron chelators, aggregators, or dyes. And so we decided to validate also those compounds. And in order to do so, we developed this radar plot for each compound where we assigned a value of one for each past filter here on the upper three corners of the Pentagon and a value for zero for each failed filter. And on the lower half, we have two kinds of experimental data points using ligand efficiency in the enzymatic assay in the cellular assays. And for each compound, we could develop such a Pentagon and the bigger the red surface of a compound, the better its properties. In addition to this, we added recommendations, pitfalls to avoid and validation procedures to this perspective that we published in 2015. So going back to the four compounds that I had shown before, we can see that indeed, those have very small red surfaces in this validation Pentagon, while one compound that we published in 2012 shows very good properties in this assessment. And also apacados that the clinical compound that I mentioned before just fails one of the three filters that looks rather good in this test. And I think this validation that we did really helped to clean up a bit the literature and to focus on more valid drug like structures in the in the years following the publication of this review. The second point I want to discuss is the classification of inhibitors. And I would just like to shortly motivate why we think this is useful and necessary. So what people usually do when they develop a new enzymatic inhibitor, they look at the kinetics that is following. And if the kinetics are following this red line, I will say it's a competitive inhibitor, which binds to the substrate binding site, while if it follows this green curve will be a non competitive inhibitor, which is supposed to bind to an allosteric site. This classification, however, does not work very well for ideal one, because it has been shown many times that inhibitors with a non competitive profile with respect to the substrate L tryptophan still bind to the same binding site. And so we used, again, a combination of experimental techniques, such as absorption spectra and inhibition assays, and more in silico approaches, such as quantum chemical calculations, analysis of 3D structural data, and a thorough literature review to reassess the classification of ideal one inhibitors. And without going into much detail about the different idea one reactions, I'd just like to show that using these techniques, we could define inhibitor types that explain really at which point or in which way these inhibitors interact with ideal one. So the type one is the type that is really competitive with respect to the substrate L tryptophan, but there are plenty of other inhibitors than that of different types, such as epacadostat, which is competitive with respect to oxygen. There are other inhibitors that interact with the oxidized form of ideal one, which we call type three inhibitors. And finally, there are also inhibitors that actually compete with the hemeco factor for binding to ideal one. So this classification that we developed clarified a lot of confusion in the literature about observed kinetics. It helps to understand the different in vitro properties of different inhibitor types. And we believe that this might also translate into different vivo properties. And it could be instrumental for helping to choose the right inhibitor for a certain application. For example, if you would reconsider testing a compound against melanoma and cancer, you would probably try to choose an inhibitor that is as different as possible from epacadostat. Okay, with this, I come a bit to our core business, which is the design of new inhibitors. And based on our experience and our previous data that we were aware of, we were searching for novel heme binding scaffolds. And we decided to test these 36 compounds shown here, which belong to six different heme binding scaffolds in combination with six different substituents on the phenyl ring. And when we synthesized and tested all these compounds, we had very interesting results. First of all, we found six compounds with nanomolar IC50 values, meaning that these were very active small molecules. In addition to that, we could show that within the same heme binding scaffold, we could span more than four orders of magnitude in difference of activity by just changing a few small substituents on the phenyl ring. And we could also show that the same substituent on the phenyl ring could have very different influence on the activity depending on what heme binding scaffold it was attached to. So in the following, we used again bioinformatics tools to try to understand the relationship between the structure and the activity of these compounds. So in order to explain the in vitro and cellular inhibition assays that we've done, we used expert crystallography to determine complex structures. And based on these, we used ligand protein docking, quantum chemical calculations, and classical binding free energy simulations in order to correlate structure to activity. And as a result, we could come up with a very simple equation that very well described the different activities of these different compounds spanning four orders of magnitude. And I would just show one example in this slide. So this is one of these highly efficient inhibitors, MMG 0706, which has a high affinity heme binding group, this 124 tetrasol. It makes favorable van der Waals interactions with the protein through its chloride substituent. And it is able to make both intermolecular and intermolecular hydrogen bonds through its amino group. So when we compare this compound, the efficiency of this compound to the ones of all published idea one inhibitors, we see that we're really among the very best and most efficient compounds. So this is great. But actually, there's a drawback for these with these compounds. And actually, that comes into play when we try to rationally extend these compounds into larger chemical compounds. Because, as you know, a drug needs not just to be active, but also it needs to have a lot of other good properties for making a good drug out of it. And these very small compounds do not provide enough flexibility to address these other properties. So in order to understand why we were not able to extend our compounds, we went again back to structural data that had been deposited in the meantime in the Protein Data Bank and to try to understand this issue. So here, I show that in 2006, we had just two structures of idea one that had been published, but starting from 2014, more and more structures became publicly available. And today, we have 62 structures of idea one in total. Some of them are without ligand, others are with substrate analogs, and a lot of them are bound to different inhibitors of different types. And in orange, I'm just showing the structures that we have resolved ourselves. So when we look in detail into the idea one active site, we had already defined in 2010 two sub-pockets of the active site, one that we called pocket A, that is located directly above the HIMCO factor, and that is filled by all known idea one inhibitors. And extending from this pocket A, we have a second pocket that returned the pocket B, and that is located more towards the entrance of the active site. And so our challenge has been for a long time to extend our compounds from pocket A to pocket B. And I will try to quickly explain in the next slide why this wasn't working so well. So here, I show a top view of the active site, and here the first example, the substrate L tryptophan binding to idea one. And we see here in Sayano, again the A pocket, and in red with a red arrow, I marked the passage from the A pocket to the B pocket. So here, this is how the substrate is binding to idea one. The second example is, again, a top view of epacadostat binding to idea one. You see that the passage is more or less in the same region as for L tryptophan, why for Navoximod and all our azole compounds that we developed ourselves, they are actually oriented in a slightly different way binding to idea one. And actually, this makes them not optimal for passing from the A pocket to the B pocket, because by their geometry, they cannot pass here in the optimal way from one pocket to the other. So in summary, the result of our structure-based drug design for idea one is that idea one is a very challenging target. There are very subtle structural requirements, such as the one shown in the previous slide, that are very difficult to predict correctly by classical docking approaches. And the second challenge is the presence of the iron, the heme iron in the active site, which makes it necessary to resort to quantum chemical calculations in order to accurately predict binding energies. And we have developed a few approaches to tackle this challenge. With this, I come to the end of this presentation. And in summary, by taking idea one as an example, I hope I could show you how a combination of bioinformatics and experimental approaches has been able to move ahead this academic drug design project and to generate knowledge beyond just the development of new active compounds. And I'm very confident that these approaches that we used are also transferable to many other drug targets. On the other hand, we're also ready for renaissance of idea one, which might become interesting again in the future as a target, either in cancer or in other pathologies. With this, I would like to acknowledge my co-workers on this project. Of course, people from the SIB and Lausanne and many laboratories, both from the University of Lausanne and the EPFL. And last but not least, also the Brussels branch of the Dietrich Institute for Cancer Research. Thank you.