 almost start, still 15 more seconds they told me because they have to start the recording. Okay, I think I can start. So welcome everybody. I hope you enjoy the social dinner. Oh no, this is another story. My name is Alessandro Lio and I'm chairing the first morning, well this morning session, the first speaker of this morning is Professor Giorgio Colombo, who is Professor of Organic Chemistry at the University of Pavia in the head of Biocomputing Group. So Giorgio, we look very much forward to your talk. You have half an hour, which will be followed by 15 minutes of questions. I will warn you when your time is almost finished. Please Giorgio, go ahead. I would kindly ask all the other participants to switch off the cameras. When the questions session will start you can either rise your hand and then might make a question directly or you can write your question on the chat and I will pass your question to the speaker. So please Giorgio. Okay, thanks Alessandro and thanks to all the organizers for this chance of speaking here. This is one of the first times for me to give a talk on Zoom so I hope everything goes smoothly. I hope you can see my screen. All right, so what I want to talk to you today about is what we try to do in the lab to try and use protein dynamics and the study of protein dynamics and simulations to help us design new molecules with potential functions in biology. So one of the things we try to do is to develop new computational chemistry methods to investigate how a certain protein will interact and will recognize with partners. In particular, you know, in cells you might have a protein, a biomolecule in general, which has to interact differently with different partners to carry out functions, regulations to carry out the tasks that are needed for the cells to survive, to proliferate and to live correctly. When these functions do not work properly, diseases come up. And what we want to do is to develop small molecules that help us fix these functions, these malfunctions, or that help us understand what mechanisms and what types of surfaces or what types of interfaces are mostly used by proteins to interact with other partners. So we want to develop small molecules that by modulating the dynamics also modulate functions. And also we want to design mimics of these differential interfaces so that we can target the formations of these particular complexes, of each of these particular complexes, with selective precision and with such that we can relate the disruption of a complex with a certain biological effect. So in order to do that, we generally try to use molecular dynamics, which is a technique that you have seen described for at length in the thoughts of Giovanni Bussi of Alessandro Magistrato and Alessandro Laio. So I want to add too much on to that. But what we want to do is to try and extract from this messy motion of the protein that occurs on short time scales the summation of motions on shorter, on longer time scales that is related to functional and functional aspects of how the protein works. So in this case, once we have understood the way a protein works, we can identify potential sides on the protein that when targeted by a small molecule will send the conformational signal around throughout the molecule that is related to a certain functional or biological response. And we generally try to do that by using as a test case a particular protein, which is a molecular chaperone that I will tell you more about later. Also, these types of motions are responsible for how a protein interacts with a partner. So different types of protein conformations might select different partners and by reshaping the potential interface, the protein might interact with different partners at different times in the cell cycle. So what we want to do in some in many of these cases is to try and predict which are the potentially interacting parts of the proteins and then mimic them by interaction mimics, which are synthetic molecules that we make with synthesize in general in the form of peptides. So I will first start by trying to give you an idea of what we have done in trying to design these functional small molecule functional regulators of protein functions. So as I said before, what we want to do is to devise or to design small molecules that by binding two sites alternative to the active site will change the dynamics of the molecule of the protein so that we obtain a modulated response or a different response than the response that you would obtain normally in the absence of the synthetic molecule. We applied these concepts to the study of this molecular chaperone, which is called HSP90, which is a very complex protein composed of two monomers. It's a dimer composed of these two monomers whose functional dynamics you see is very complex and goes through a cycle of ATP binding, which brings the protein to the closed state like a big molecular clamp that closes and then this molecular clamp acts on other proteins and folds them to the correct and functional state. So once these other proteins which are called clients are folded, they will carry out their function. In cancer these molecular chaperones and in stress conditions these molecular chaperones are highly over expressed and in general they help cells like cancer cells which normally are living in a hostile environment to adapt to that environment and to survive to that environment by overexpressing, by helping the overexpression of say kinases, glucocorticoid and hormone receptors and so on, which are needed by cells to survive. So if we can interfere with this type of mechanism then we have a good chance to interfere with cancer pathways. HSP90 is very ubiquitous in cells. It's in the cytoplasm and lately it has been found to be present also in the mitochondria and the mitochondrial homolog is called Trap1 and the interesting thing is that Trap1 has got this interesting asymmetric dimeric structure which crystallizes in this asymmetric dimeric form which is different from the symmetric structure that you see in the cytosolic homolog. What is also interesting is that in some kinds of tumors this Trap1 is highly overexpressed in other types of tumors Trap1 is highly downregulated. So what you want to do is to develop molecules that might target Trap1 only where it is overexpressed without interfering with HSP90 which is normally used by other cells to carry out to survive. So in order to do that one strategy would be to target the ATP binding site which is shared by all these chaperones but as you know ATP binding sites are very conserved so you wouldn't be able to target HSP90 sorry Trap1 while living HSP90 unscaded or untouched. ATP sites are conserved ATP is very abundant and so it's very hard to to compete with it but this type of asymmetry here gives us a chance to think about possible targeting of these sites by small molecules that can differentiate between this structure here and this structure here. So the question now becomes how do we find a draggable pocket in here which is which is the most likely draggable pocket in this region that might interfere with even with ATP binding and then the functions of the whole protein. So let's go back to the mechanism so the mechanism is the following so ATP binds the protein gets into the closed state and into this sort of asymmetric state okay and then one ATP gets hydrolyzed the this region here changes its conformation and reshapes the client protein in a mechanical way. So what we want to do is to find the residues in this region which are most responsive to ATP binding and hydrolysis so that if we put something in here if we like putting like a wedge or a small molecule in here we can send a conformational signal back up interfering with ATP signaling in trap one exclusively and then interfering with mitochondrial processes that are at the basis of tumorigenesis. So in order to find these residues which residues we need to target in this region then we need to find what is it that communicates best with ATP and the ATP site. So in order to do that we use this simple coordination propensity measure which we developed a few years ago together with with Christian Nicoletti also in which basically what you look at is the low distance fluctuations between every residue i and every residue j in the protein and then when we focus on which residues in in the protein are mostly connected to the residues that are in the ATP binding site what we can see is something interesting so if we consider the protein in the 2 ATP state you can see that the protein is kind of highly connected it's very rigid these are the residues connected to ATP site for most of the of the simulation and you see that this sort of allosteric communication pathways go down to the far domain of the protein down to this region in both protomers when ATP is bound to one arm to the buckled arm and ATP is bound to the straight arm then you only see this region here that is highly connected to the ATP binding sites and when you have the when you switch ATP for ATP between the buckle and the straight protomers then you also see this region here so this region in the middle part of the protein always appears to be involved in some sort of strict or strong communication with ATP binding sites interestingly this region here for for track one if you look at this measure calculated locally so only on on on the short stretches of residues and we compare track one versus the homologous residues in HSP 90 the red versus the the black lines you can see that the local dynamics also of these residues is very different between track one and HSP 90 so these residues may be representing a sort of you know point of differentiation between the two proteins so we can plot them onto the three-dimensional structure and then use a collection of structures that we sample from the simulations from MD simulations to look at the possible structures and conformations that need to be targeted by small molecules and that we can actually use to host small molecules so we look at these residues the ones that are most connected and that are related to the presence of a possible pocket that is small enough to host a small molecule a drug like small molecule and then what we do is we look at what kind of functional groups these residues display to possible potential binding with a with a with a small molecule we do that by developing a sort of pharmacophore model in which we ask the model to display a certain functionality that is complementary to the one presented in the site that we want to target so let's say the protein presents a glutamate structure a glutamate functional glutamate sidechain with a the carboxylate function then in our in our pharmacophore model we'll need to have a charged a positively charged group to form a plus minus charge with a glutamate which will give us a good electrostatic interaction if we have hydrophobic interactions here we also present hydrophobic interactions and so on so once we have this model we can use this model to screen for a small molecule databases which are available for instance from the zinc database or we can screen for for commercial databases and we did that and we came up with about 11 with about 35 candidates 11 of which we could buy and test and of which we noticed we first looked at the possibility for them to block ATPase activity to block the hydrolysis of ATP in trap one versus Hsp90 so what you can see here here in blue we have trap one and in red we have Hsp90 you can see that the blockage of the ATPase activity always happens with the small molecules in trap one but not in Hsp90 so we have sort of selectivity versus the isoform that we are interested in in terms of in terms of targeting the mitochondrial species in some cases even some compounds are inhibitors for trap one and activators for Hsp90 but we still have to look into that also with these molecules we also have a nice dose response profile with IC50s that are in the low micromolar which is not so fantastic for a drunk but it's very good for a chemical tool that we can now use to you know mimic the effects of shutting down selectively trap one while leaving Hsp90 function in the cells so this is what we did in collaboration with Andrea Rasola at the University of Padua and he did the following experiments so in we know that when trap one is active it will bind SDH succinate to hydrogenase which is a protein involved in respiratory phenomena in the cell and by targeting by by being activated it will act as a break on SDH so if you ablate trap one you knock down trap one or you inhibit trap one succinate the hydrogenase is released from the complex and becomes activated and its activity will go up so trap one ablation will bring up SDH activity and they did that by using a CRISPR-Cas technique so when we when we look at the effects of the molecules you can see that by using a normal inhibitor which doesn't distinguish between trap and Hsp90 you have of course you have this is an ATP competitive inhibitor you have the activation of SDH the same way as by using trap one ablation this orange curve is trap one ablation by biology here on different lanes you have the effects of the small molecules all of which sort of produce a phenotype a copy of the phenotype introduced by the ablation and the knockdown of trap one also when we test these molecules on tumorigenesis you can see that while tumors grow nicely in the presence of trap one in the presence of the small molecules at higher and higher doses they will basically this fochi of tumor growth will disappear and the SDH effect is also then observed in vivo in zebrafish larvae models so we have a nice chemical tool molecule five to start doing more experiments to investigate the functions of the proteins differential their differential functions of two very highly homologous proteins by you know doing more experiments and by developing this molecule five even more using the models that we have generated so far so to summarize this this part we have exploited structural and dynamic asymmetry as the basis to discover new isophone specific allosteric inhibitors of trap one it is the analysis of the nucleotide dependent motions that allows us to identify the structures that are probably mostly involved in regulating these motions and then that we can target to perturb these motions and we from the therapeutic point of view we here we have selectivity versus one one isoform which is highly overexpressing certain types of tumors and not expressed in others versus the various isoforms of Hsp90 which is you know present all over the place and you don't want to inhibit it indiscriminately not to have side effects or off target effects and then we can use this data that we have generated for a modulation for the mechanism rationalization and so on to define novel leads and to design design new ligands so this brings me to the second part of my talk in which we would like to interfere with the activities of these chaperone proteins by blocking their interactions with other proteins so so far i've talked to you about the way that the chaperone interacts with ATP and goes through the cycle as an isolated molecule but in reality it will go through this cycle by interacting with a number of other proteins and a number of other proteins forming multi protein assemblies which are highly dynamic and which involves different players so Hsp90 binds ATP gets into the sort of closed form and then a co-chaperone which is called cdc37 binds up and brings with it a client protein indicated by cl here the client is typically a kinase and by bringing the kinase onto the complex it will start the way that Hsp90 works on the complex to reshape the client so you know these are very dynamic complexes dynamic structures and they have been finally shown in their full atomic detail by cryo em by the group of the agar a few years ago in a great science paper which they showed that Hsp90 here this x in the middle binds to cdc37 that goes throughout uh all around the protein all around Hsp90 and to a client protein which is here which is a kinase what's interesting here is that the kinase is sort of unfolded and part of it goes through the lumen of Hsp90 and this large long two helices are part of of of cdc37 so we have this client which is unfolded partially unfolded we have the co-chaperone we have the full length chaperone Hsp90 so how do we interfere and this is a you know this is a general mechanism that would involve other clients other proteins because Hsp90 reacts on many other proteins so how do we develop ligands or reagents that selectively detach one client from the protein from from the chaperone while leaving others untouched so two and and once we we we can predict how to do that what do we design we should design you know unfolding region mimics so this is the region that unfolds here we need to find which regions are unfolding or are most prone to unfolding and then we should design mimics of these regions so structural mimics of or sequence mimics of these regions which can adapt to the surfaces and interfere with the process of of full assembly so what we did then was okay we don't know the unfolding pathway of or the full unfolded state or of complex proteins like kinases or hormone receptors and so on so what we can do is to find the regions on these proteins by computation that are most prone to unfold so in other words the regions on the protein that are not so well connected or energetically coupled to the most stable regions of the client so imagine this is a kinase or receptor which has to bind to Hsp90 one part of it will fold automatically by itself and will constitute the folding core of the protein one part which is the ones that may be in contact and may be recruited by chaperones and coshaperones is not so stable and will be of course recruited by the coshaperones and blocked by Hsp90 so we need to predict this how do we do this we do this by predicting which regions on the client protein are mostly most likely to be uncoupled to the folding core of the client and we do that by running MD simulations of the client in isolation we calculate all the possible non-bonded interactions between all the residues then we simplify this matrix by eigenvalue the composition and we filter it by the contact matrix and we isolate all the couples on all the pairs of residues which are on the surface or close to the surface of the protein and that are not well coupled or not strongly coupled to the interior of the protein for us these regions which we call epitopes are the regions that are most likely to undergo unfolding because they are not strongly attached to the rest to the rest of the protein so when we apply two kinases for instance we decided to apply first to this ABL kinase which is a tumor driver and it's different from the kinase that we have seen in the in the agar structure we predict this region here in the envelope in red as the region that is most likely or most prone to undergo unfolding transitions or to undergo unfolding we don't know anything about the unfolded state we just predict that this part of sequence in red will be the most likely to undergo unfolding transitions so when we look at the complex of Hsp90 with the other kinase and we superimpose the folded parts the c-terminal domains of the original kinase in the cryo-em structure two hours you can see that the predicted epitope really traces the surface of Hsp90 that is really then in touch with the unfolded part so if we can realize this red part here in the form of an isolated peptide synthetically in the lab and we then throw the peptide into the mixture of this chaperon plus the rest of the chaperon machinery we should be able to break up the chaperon machinery and block its action on the folding of ABL okay so that's what we do we realize peptides that mimic these regions by synthetic normal synthetic methodologies and then we test them with NMR we put in fluorinated compound fluorinated residues to run fluorine 19 NMR spectrum and so this is the peptide that we realize mimicking this region in isolation when we add Hsp90 there is a shift of the signal indicating interaction and then when we use this peptide to detach Hsp90 from ABL from Hsp90 you can see that ABL is not present in the complex anymore here you have Hsp90 in the presence of the different of different sequences and different variants other proteins CDK4 which is the original kinase is not touched is always there but ABL in the presence of the different peptides is completely absent in the cells is not folded anymore as well as this cochaperon AHA1 which is detached from the complex then you know this is one case to check whether this could be generalized to others we carried out our predictions on other kinases like BRF which is hard to drug it's one of those undruggable proteins CDK4, SARC, GR1, etc which are all related to tumorigenesis we predicted the regions and then we decided to also synthesize them as synthetic peptides and look at their interactions again with fluorine NMR the different lanes indicate the different peptides these are the signals to or bind into CDC37 AHA1 which are cochaperons and Hsp90 so you see the different peptides may bind to different parts of the chaperon machinery to different kind of players in the chaperon machinery and then what do they do they do exactly what we want so they detach their cognate protein BRF in the case of the BRF peptides CDK4 SARC and so on what's interesting also is that some of the peptides in particular reform more minutes yeah I'm done basically so this CDK4 for instance induces the expression of this caspase which is related to apoptosis in the cells so by disrupting these complexes we are also inducing apoptosis and cell death in Hsp90 dependent mechanism so we basically have found a way to target selective and difficult protein-protein interactions to induce the degradation of difficult to target by small molecules proteins and we also have a number of different sequences that we can use to evolve into potential better chemical tools or even drugs but that's a longer a much longer way to go what to just summarize so you know it is possible then to combine computational chemistry methods the analysis of biological mechanisms in biological chemistry and develop new chemical tools to you know combine all these three areas and develop molecules or small molecules that help us playing around with biological systems and reshape the functions or better dissect the intricacies of the functions of this of these systems so I got to the end hopefully in time and I would like just to thank all the people that work on this project that you see in in this slide all the collaborators in particular I'm pretty and really grateful to Andrea Rasola and Carlos Sanchez Martin at the University of Padua besides all my group and I would like to thank all the finance know all the funding and all of you for your attention so I'll be happy to take questions thanks very much Giorgio thanks very much also for staying really really you know strictly on time so start questions which are start appearing in the chat I will directly make the questions just to save a little bit of time so the first question is by Davide Bassani the question is what is the length required at the required length of a dmd simulation which is necessary to depict the allosteric activity of trap one okay so these these simulations have all been done in multiple replicas and are in general between half a microsecond to one to one point something microseconds each so for each for each system we have carried out pretty extensive simulations in replicates and so we try to filter out only the signals that are consistently made consistently coming up in the original paper of the trap one study we did also simulations were still a little shorter than that which we use them for the compound screening and however we use different force fields also so we used gromos and number force fields to compare the data and the results and they were you know overall consistent okay then there is a second question for the screening of compounds yeah so the question is for the screening of compounds the flexible docking program algorithms so um i'm sorry i haven't been very clear on that we don't use stocking okay so we try not to use that at least at the beginning for the selection of the compounds so once we have identified the site i don't know if i'm trying to use my hands but once we have identified the the site where the molecule could bind we look at the chemistry and at the conformations of the of the potential binding site then what we do is we make a sort of mold of this binding site and we look you know when we look at the chemistry we identify the hot spots that are needed to bind so if you have a group donate you will need you will have on the protein it's an anacety charge on the protein so you need on a molecule a positive charge to bind there so we just represent that as a function as a plus charge in one position with a certain distance from say another point which has another characteristic complementary to the potential binding site and then we screen uh we screen do we screen these for uh we use this pharmacophore model to screen for molecules and then we only do docking afterwards at the really end of the of the of the process now we are studying the small molecules within the protein context and we have used flexible docking algorithm to uh to put the small molecules in to have good stocking structures under the next there is another question which is more technical which tool do you use for pharmacophore predictions so in this case we have used uh we have used a combination we are very deterministic on that so opportunistic also we use both the ones available in the programs and the ones available in Maestro we try to generate a consensus of the different algorithms and generally they you know work pretty well one thing that is also important is some sort of chemical sensitivity so you know you really want to avoid having large uh huge hydrophobic areas uh which will make your molecule unsoluble or too many different stereogenic centers and so on clear okay so the next question is more general and it's also connected to another question that I had myself is well can you please explain again how to compute the coordination propensity and how to exploit it for comparison of two proteins my question which is really was related and I wanted to ask is also if you can generalize this measure to take into account some how the causality in a way if a specific contact is responsible for the change of another contact somewhere else uh for in this case not really these are on average uh properties from the trajectory so we kind of lose the the causality yeah we are trying to look at the coordination propensities on different uh length scales of the simulations and trying to analyze them in using different types of techniques but um it's kind of hard uh more going back to the question which is more general maybe you can just okay so uh what we do is we calculate um from the simulation you have every possible fluctuation between uh the atom i and atom j so they have a distance so this distance is fluctuate right so um we decide that two residues are uh in communication mechanical communication if they're if they they're distant the distance fluctuation is below a threshold which is defined on the basis of a local uh parameter um and then uh we find all the pairs of residues that have this uh fluctuation in the distance below this threshold which is typically the average fluctuation that you calculate between every residue i and i plus four which is one turn of a helix basically so if you have a high coordination at rest for residues that are very you know physically separated that might not be a trivial um a trivial uh signal that might be something related to how the protein is packed and how different secondary structures are organized to transmit a signal from say an active site to another part of the protein how do we use it to to compare two proteins um we you can uh calculate it uh for for different proteins and then realign them by sequence and then um we can either calculate the sum of the of the of the columns of the of the matrix that we obtained for the coordination propensity matrix or look at some or transform it into a local fluctuation uh parameter like the one i showed in talk and then look at the uh realign these graphs on on the basis of the sequence alignment so that we can compare um two or more different proteins um yeah that's what uh what you do yeah so actually there are three more questions all related to the coordination propensity and more and much related within each other so i'm going to read these three questions at the same time and we can see in the last three four minutes you can give us a hint well could the coordination propensity be useful for evaluating well for discovering potential regions of the protein for allosteric inhibition question that's accepted uh well then the second one is there any connection between the coordination propensity and conformational thermodynamics question number two and third question still also on the same uh topic uh well yeah the threshold by just uh the threshold you just mentioned so these are mentioned um the coordination propensity and conformational thermodynamics um it's probably something qualitative that we can say at the moment so if you know if you have regions that are highly coordinated or that constitute highly coordinated domains you may imagine that these uh move uh in um in um in um um how do you say in uh um in a coordinated way as well or in uh in um i miss the word uh i'm sorry um concerted in a concerted exactly in a concerted faction so you may extend it to identify some sort of uh to predict some sort of uh overall or large-scale motion of all these regions that are conserved or you can display i mean if they are subdivided by flexible regions or unconnected regions you can imagine that they these unconnected regions will be hinges but we cannot really use it to to to really get any quantitative data on conformational thermodynamics it's more uh qualitatively really all right we still have one one two minutes there is a question which is not related to that is well since most of the pockets that you find in your analysis are highly flexible how do you actually choose the shape of the pocket on which you perform docking how do you so um so um we actually really do not perform docking to screen for the compounds but we uh use different pocket conformations uh that we sample from the simulations to develop adaptive let's say uh pharmacophore uh models that are complementary to the different shapes of the pocket that we identify and then um we try to build a sort of a concessus for macophore that recapitulates all the all the chemical functions that are consistently required in all the different structures to bind productively to all these different structures i don't know if uh if that's um uh that's clear uh i think it's clear enough well actually george now i think we have to move to the next speaker but there are other if i i think two or three questions which i didn't mention which are more general like md details so you can maybe reply directly on the on the yeah so i there's no other surprises in the in the in the talk so you can find the methods and the coordination capacity also for the they want a publication about the coordination so just email me i'll send you the links and sorry clap my hands on the behalf of everybody and i think we can uh now move to the next speaker with uh marco de vivo uh marco de vivo is uh is uh uh uh working at i it italian institute of technology he's a research lab director and working on well molecular modeling and drug discovery so topics which are very much related to the issues which were already discussed by georgio so marco please go ahead with your hi good morning everybody thanks alexandro for the introduction and thanks to all the organizers for the invitation i really do hope to meet you all in person soon and maybe for a second edition of these nice nice workshop and school okay so um in the spirit of the title of the conference and talks i've seen and you will see in my talk a number of examples where i think we've put a good use in molecular dynamics either classical or quantum to study biological processes that are quite challenging so actually oops okay in the group we have a number of um interest uh on pharmaceutical relevant targets going from membrane bound proteins actually up to uh functionalized gold nanoparticles and what i wanted to mention is that everything we do here is in terms of computations and molecular dynamics mostly but then we translate this knowledge into the actual synthesis of small molecules within the group so we are lacking a way so we can put our information into uh the rational design of small molecules that are used then to initiate their discovery up to in vitro in the group when possible and then of course we rely on external collaborators to move to um animal models and before starting the the talk which today will be on uh as you know uh dna and rna uh processing metallo enzymes i'd like to have a snapshot on the drug discovery part of the room which has quite uh uh has been quite active over the last few years so with a couple of fleet components that are active in vivo in neuroscience and and cancer as you see and we hope to move forward these compounds in advanced preclinical models very very soon so just to give a snapshot of how then we use molecular dynamics but today we will stay on computations on molecular dynamics of these type of enzymes that process dna and rna actually we look at a bunch of them in terms also classes so these are polymerases nucleases so polymerases elongation of nuclear acids poly nucleases they cut nuclear acids ribozymes which are rna enzymes rna based enzymes and topoisomerases which regulate the topology of uh nucleic acid tna now all these enzymes amazingly have conserved the same catalytic site the same architecture of the actually catalytic site which is a so-called two metal ion active site so all the drugs that act on the catalytic site of these enzymes target the same type of architecture now this is uh important because it goes back to a seminal paper that was published years ago by thomas tides and obela rate and and his wife and john tides it's a very seminal paper on these particular mechanisms that i suggest those young in the audience to read because it captured all the uh essential aspect of these mechanisms based on the very first structure where they saw this particular architecture where there is one ion that stabilizes the nucleophilic oxygen the other ion that stabilizes the living group and they together stabilize the transition state for either for the forceful transfer reaction that's the name of it okay so i started to look at these type of action times ago and the approach basically that we use is the following we first study through classical and the targeted enzyme and for these of course the challenging aspect is the metal centered catalytic site where we use a non-bonded approach for the for the metal ligand interaction we have a distribution of the charges keen to distribute the charges to stabilize the site at times and then of course once we want to look at bond forming and breaking events we have to switch to quantum simulations in the in most of the cases we use the carbon yellow coordinates uh q m m m implementation as alessandro in fact implemented with Ursula years ago and we still use that in most of the cases and of course then we have to couple these simulations to an asset sampling techniques where the trick is always to find the right coordinate reaction coordinate or collective variable to describe your reaction the reaction you are interested in now how do you know if the reaction if the profile you get out of these analysis sample simulation is the reasonable or a possible pathway of your catalytic mechanisms well there are three criteria that you have to satisfy the first of all of course you want to see the reaction that you have in mind so you have to generate the product as expected as you know it has to be formed second one the free energy barrier for this reaction to happen has to be better meaning lower than the uncatalyzed reaction in water of course because you want to see the catalytic effect of the enzyme and third one you can compare the energy you got out of your analysis sample simulations with the k cut and then if all these criteria are satisfied you can say with a reasonable uh confidence that your reaction mechanism is a potential one and the correct one so we have used this approach in a number of cases starting several years ago on a ribonuclease H catalysis because it is a two metal ion site and it is promising a promising target for for hiv we have compared these mechanisms with a one ion mechanism another enzyme same reaction but with only one magnesium we have had a paper years ago where we proposed the presence of a third ion and you will see this is quite relevant for the following of the presentation this was predicted only on based on md simulation at the time then topoisomerases and more recently DNA polymerases and this is a particular polymerases that we are very much interested in and i'll tell you more about that okay so but let me make a point that is going to be useful for the rest of the simulations uh for the rest of the presentation sorry so we um try to integrate our view on taking into account the structure the dynamics and the function for catalysis RNA and DNA so this is a snapshot i took few days yesterday or the day before from the pdb data bank and this is the amount of structural information we have in the pdb data bank now we are used to run a few uh uh model systems but actually you see we are over 170 000 structures deposited in the pdb and in particular in my case because of what i'm interested in in this case protein nucleic acid complexes there are about 8 000 more than 8 000 uh complexes including those very amazing complexes that now are resolved with clio yam so we have to take advantage of this information apart from those from those few that we can actually simulate and that's the point i want to make here and showing you a few cases where we generate a mechanistic hypothesis based on an analysis of these information available today okay so the very first example regards the DNA and RNA polymerases now the catalytic site uh the catalytic cycle that is here actually reviewed in these uh perspectives and perspectives if you are interested starts with the deprotonation of the 3 prime OH and then there is the incoming nucleotide that has to be attached to the growing strand of the uh growing filament of DNA here and the nucleotide has to be added so uh i'm gonna go fast on these because then i want to reach the more recent stuff but this is very interesting so we looked at all the ternary complexes in the pdb data bank which means the complexes that have the enzyme the uh protein of course so the enzyme the DNA and the incoming nucleotide and if you look at all the possible ternary complexes so from prokaryotes eukaryotes viruses you will always find that an entering nucleotide which is this one here has this particular conformation which is not the conformation that has been reported in solution it likes to stay linear and with one magnesium ion it doesn't fold in this way forming this intramolecular hydrogen bond now this is also where this is a collaboration at that time with uh with paulo cologne in germany and if you overlap all these incoming nucleotides taken from the uh ternary structures available in the pdb you will see this particular conformation where you have always the formation of these intramolecular hydrogen bond so this triggered at that time this unprecedented post observation which was not reported uh it triggered our proposal for a new reaction mechanism in order to have the addition of this incoming nucleotide to the growing DNA so the the reaction is this one here we have to form a new bond and break this so i will jump to a movie where we have cut and paste so it's a merged of classical and the quantum simulations this is the starting system equilibrated in md so you you see here now this is the nucleophilic oxygen the protonated this is the incoming nucleotide and this is the intramolecular hydrogen bond that is formed and this tabuli maintain there and which helps this conformation to be kept within the catalytic side at this point the quantum simulations and this is q m m so this is the q m part that you see uh bold and liquorice there is the reaction happening so this is the classical s and two reaction for with the inversion of the umbrella for the phosphoryl transfer reaction here the proton starts to be shared kind of shared but of course there is no proton transfer because of the k a reasons so the proton remains on the uh sugar but at this point this proton will become this one here so this has to um partial translocate and when we have this partial translocation and we switch to q m then thanks to the presence of the magnesium which lowers the p k a of these oxygen then we can have the deprotonation and here you fall within uh two point five i think angstrom from the uh co-crystal of the product state and so here we see that the protonation and partial translocation is a process that is concerted and these are we explain now this mechanism which is a closed loop of sequence of chemical steps so nucleophilic attack proton transfer and physical step the partial translocation that leads to the addition of the incoming nucleotide and we call this the self-activated mechanism because of this intramolecular hydrogen bond so this is an example of how we used these information available in the many polymer structures that we had to generate a mechanistic hypothesis another example is a collaboration with the structural biology group of marco marcia the mbl with whom we put together a curated data set of structure with the integer metal ion in the center and these includes over 140 enzyme 49 3d structure including the ribozyme so the RNA based enzyme nucleases polymerases and then we compare structurally what they conserved around the two metal ion that was still present in the structural alignment and what you see here is that there are preserved spath spatial localization where positive charges in this case is the ribosome so there are no amino acids and we have a potassium ion here and a potassium ion up here these are the two magnesium ion and where you have in the RNA based structure these two ions in proteins you have lysines or arginines okay so there is an evolution replaced in a way these potassium ions this positive charge with lysines or arginine which are preserved in this particular position and these are nucleases including cast 9 increase per cast 9 and polymerases here of course then you have to we should see the overlap but another way to look at these are the conservation of these positive charges which are second shell positive charge around the two metal ion is here so these are a sphere of four angstrom where you have k1 and what we call k1 and k2 and these are about four angstrom five angstrom from the substrate okay so they are in touch with the substrate either on one side or on the other and so in this way we have an expansion of what was believed to be only the two metal ion site and with additional positive charge that are located in the surrounding of the a concert in the surrounding of the metal ion center and this is a plot where you see the distance from the substrate so you have restricted restriction and the nucleases other nucleases polymerases and the two ions are always within distance for an H bone basically from the substrate so three angstrom now we have taken our enzyme at that time we were still interested and we are still interested in fact in polymerase eta and we mutated these two particular residues and what you see is that indeed in the reactant scent in the reactant complex if you go from the y type to either k1 or k2 or both are mutated into alanine then you have the disruption of the optimal mycali complex with the the deviation for the optimal number of Watson and Cricage bonds here that is present in the crystal structure but also what we call the d stands for distance new bond which is the distance of the new bond information so you have an elongation of the over the right here of the reaction coordinate which means that it's less prone to undergo catalysis of course we cannot look at the barrier in this case but at least we see the stability or not of the mycali complex what's useful this for also for example here is a complex structure of the spliceosome which is related to the ribosome and you will see here that in this particular structure there is a paper where they report that where the response to potassium binding on the spliceome remains to be elucidated here on the on the bottom so they don't know where the potassium here but the potassium is needed in order to have the catalytic action and here if you use our model you can place a potassium iron there which fits quite nicely and of course could maybe was not there because of the resolution of this particular structure so that's another example let me give you another additional example now this is one of these particular residue that are that is concerned this is actually a first shell meaning that is in direct contact with the nuclear with the entering nucleotide this residue here is either an arginine or or allysine and these are the alpha carbon of these residue in a multitude of PDB that we have retrieved from our collection of structures of the ternary complex and you see here all the enzymes and these are 28 representatives in the figures here and this particular residue was actually already known to be important for the chemical step for example in the case of polymerase eta it is an arginine it can be in different conformation and depending on the conformation it helps or not the catalytic step stabilizing the transition state and this is also what we have seen for the a conformation pointing down versus the other type of conformation so we know the electrostatic effect of this arginine is important for the reaction to happen what however we noticed looking again at the structural information available to us is that there are a few structures of polymerases that are reported in paper like in this one in nature to have the uggsten base painting instead of the watson and creek where one of the bases fleet 108 degrees in this case the arginine and if you look at the crystal structures you will see and we noticed incidentally that most of these structures have either these particular residue that i told you about mutated or it's pointing somewhere else and at that point there is the flipping of the base and therefore the formation of the uggsten base painting which is unusual in this type of polymerase so unusual in fact that they were able to publish such paper in very early in pop journals so the hypothesis we generate is that actually these residue here helps the bay the correct base pale for the watson and creek base pale to be formed so we came up with a collective variable to look at the correct base painting and we have the wild type where we have the energy between the two minima the watson and creek and the uggsten and then you see that the watson and creek is much more stable because the presence this arginine the stabilize the painting if you remove that and then you calculate the difference between the watson and creek and the uggsten that well basically is actually what is important is the energy of the two states which become iso energetics so you can have either uggsten on watson and creek and that's how then we propose that there is a strategically located residue there which promotes the correct base painting for nuclear acid biosynthesis in polymerases so again based on the observation one last example it's this particular lies in here in these exonuclease which is believed to be the protonated because it then it needs to act as a general base to get the proton from the water this is a water that will attack on the phosphate and what we found is that indeed when this is protonated after the protonation oops after the protonation of the water molecule then you form a cation by interaction with this concert tyrosine on the side at that point this interaction allows the tyrosine to bring out the so there is an oscillatory dynamics here of this couple and that prefers to go out in in such a way that the lysine can release the proton in solution and be back again to initiate the cycle so you see here the simulations where the lysine is in neutral prefers to stay next to the water of course ready to act as a general base but once we simulate the protonated state the cation by interaction brings out the lysine which then we show through here qmm and simulations the reaction to happen and here you move the proton outside that can be jack and jump actually on an aspartate or a water nearby so in this way again we were able to form a closed loop sequence of steps that allows the reaction to happen and to form again the new system once the proton is being released to undergo catalysis so again another example of how we have used this information to generate mechanistic hypothesis okay then one more point that regards the presence of a third ion i mentioned these are the very beginnings this was a big finding in the field because there is the two metal ion now there is a third metal ion as you see here it ended up in in science and we put together in this paper here a table of the pdb that are now showing the third ion presence present in the two metal ion structures and there is a particular paper that came out on these human exonuclease one enzyme where there is a transitory third ion that is found in the reactant but not but not in the product so these are increased tallow studies which means that we have several crystals along the reaction of the enzyme starting from the pre-reactive to the post-reactive states and we run our simulation in this case this is the enzyme and this is what the enzymes does it cuts the last nucleotide of the five prime strand in case of processor related to repair of DNA and so on so this is what is called exonuclease because it cuts the extreme of the strand okay so here we made an observation when the third ion is present we have this glutamate that is folded toward the third ion that is here but when it is not present the crystal glutamic acid points outside towards the solvent so here we run simulations with and without the third ion here you see one of the simulations in the product state in the reactant state this it is stable there if we remove it we see sometimes a potassium ion sometimes a sodium ion going there so there is clearly an electrostatic trap that attracts ions from the bulk that slowly diffuse there here you have a case where in the product state we have a weather we do see the third ion going there in this position here so this is the frequency the probability density of having the glutamate either out or in depending on the presence or not of the magnesium and what basically this says that if there is a third ion profess to stay in if the reason then flips out like looking fishing for an ion nearby to bring close to the other two and here is the geofar of the third ion in the wild type which is here so it is present and then if we mutate in our simulation the glutamate to another and then you don't have any more the presence of this third metal ion so again there is a functional or there is a way to call to select and recruit this ion and then we started to hypothesize the possible role of this ion of course we cannot say yet the role on the catalytic step but what we saw in our unbiased MD is that once the third ion is there in the product state the nucleotide that has been cleaved at that point would like to leave starts to leave it doesn't leave completely because we assume at that point maybe there is a barrier or maybe anyway it's not able in this time scale to live but we clearly see in two of the replica that the nucleotide that is being cleaved starts to you know go away from the leave the catalytic side this is in the absence of the third ion the nucleotide even if cleaved stays there it doesn't leave so at that point again we use this reaction coordinate to look at the possible pathway to see the energy for the nucleotide that is now leaving the catalytic side to exit the side completely and here you see the energy that has been estimated for the two processes in presence in yellow in the presence of the third ion or not and then you see that of course with the contour ion because of course this is negatively charged the exit is much more favored and it fits it matches the overall k-cut and therefore is compatible with the process we are looking at so in this particular case we suggest the third ion for the physical step of leaving the departure is one of the players so it helps to shuttle out the the the leaving group and this is a movie that should convey the overall mechanism I described and here you have the protein here at the center you see the two ions and now there will be the third ion coming in with the recruitment of the glutamate so again here we are merging different parts of the simulations a bit about kions the third ion goes there but now you will see that you will see the glutamate switching it's here you see it went close to the other to the other two at that point we have the hydrologists we don't see that we jump in the post catalytic state and here we have an acid sampling simulations and at that point you will see that actually a nucleotide leaves together with two ions and we are not forcing that that's one of the simulation what we see and it passed through the arch that is particular arch that is formed there so in this case we have seen again this trafficking of a third ion that seems crucial for catalysis one well also in this case actually you can extend now the relevance of these particular mechanisms looking at other structures where there is either the outer or the inner conformation this is another exonuclease and these are the other very similar proteins where again there is always an aspartate in this case or and glutamate here that is and glutamate also here that is positioned in the same way in order to likely act in the same way as in exonuclease we have I have shown you and this is the work of Elisa Donati in my group okay one more example if I have the time to do it three four three four more minutes okay so I'll I'll I'll reach the end so the last example is the group two introns so this is the system that Alessandra introduced Monday morning and it's very important for many reasons in general expression it's an ancestor as I already mentioned the splice zone and this proposes a genetic tool and we are lucky to have these very nice collaborations with these structural biology groups that generate structures for us and here it's a complex structure because it is a 360 nucleotides that are folded and they again form these two metal ion center with these are the two magnesium and here you have k1 and k2 as you already seen in the past what however is here important is that there is this particular cytosine that is critical for catalysis every time is replaced with another base that is not an adenine the enzyme doesn't work anymore and what we have noticed this is simple chemistry that these are the only two bases that can be protonated so maybe this is a role and there is indeed a role for that I'll show you and then there is a second observation that the crystallographers did they trapped the enzyme in a structure that is called toggling state it's actually a structure where these junction is called a junction at the catalytic site changes a bit this conformation and now this is complex to study on the other end is very local is at the catalytic site and we have knowledge of the starting point and the ending point of this conformational change we don't know however what's for why this is important of course it has to happen but at some point during the transition from the first step to the second step of splicing now this is the step that has been already clarified by lorenzo and alessandra in this nice paper here with the attack of the water and now we have now to locate this transition here that has been reported in the structures we placed our set in a very good situation because we had access to four new crystal structures where this particular cytosine has been mutated and every time it was mutated the good thing is that it was a structural impact the enzyme so the enzyme that didn't break because of the mutation however the assay the assays demonstrated that the mutations were less effective in generating the product of the first step of splicing so we know that that particular site cytosine is important okay so what triggers now the toggling formation so here i make the story very short we run simulations of the pre hydrolytic state the post hydrolytic state we have crystal structure of these and then we clean what we call the cleave state which is the the reactive system where we cut the bond that has to be broken during the action and we see what what happened when we do this and indeed we have simulations with the cytosine unprotonated and with the cytosine protonated and what happens is that when this cytosine is protonated and we believe this may be for example the proton that comes from the nucleophilic oxygen that has to be the protonated we have QMM simulations for that in the paper what we see is that the protonation of this cytosine destabilize the cathodic site you have salvation of the cathodic site and at that point one of the potassium ion is immediately released at that point when it is released this also happens in the case of the cytosine if it is not protonated but in our simulation always takes much more time this is always instead immediate basically so we propose that the protonation of the cytosine induces the release of k1 which at that point allows only without k1 you can have the toglic formation this is from the crystal structure so this is a just to show you the moment where the potassium ion leaves spontaneously and at that point we have a structure that can go the pink one here is the junction that has to toggle so we have used adaptive with pathogenomics to look at that point at the structural conformational change of the of the toggling going from the pre-reactive to the leaving of the k1 so here so we have the leaving the k1 and at that point we have the process that can take place and this is useful to go to the first to the second step of splicing so that's the positioning of the toggle conformation that they were able to observe in crystals and the energy is compatible again with the k cut and therefore this could be a possible explanation and visualization of the transition between the first step and the second step and this is the work of Jacobo in my group that just came out a few weeks ago okay so i'm finished i hope i was able to show you how to integrate our the view of structure dynamics and function for catalysis in this case on RNA and DNA we have generated new mechanistic hypotheses based on the role of key residues we have expanded the functional implication of the two-meter lion to a second shell that is now recognized to be important and then you'll see now trafficking of cations at the reaction center is also important for this type of reaction that are conserved in many many enzymes with that our direction now of course is to translate this information i hoping in enzyme design of course that i've discovered this is what we are doing now without this information and i want to thank the people involved in my lab all the lab where you see Pietro, Elisa and Jacobo who have been involved in part of the different parts of the work that i show you and then i thank you and uh be happy to take questions hello Alessandro? Alessandro you have to unmute i'm muted sorry thank you very much Mark it was a very very clear talk there are already a couple of questions on the chat i start with the first one for the simulations you use amber both as force field and this program or you perform the simulation with Gromax OpenMM or something else yeah usually we use amber as a force field but then we perform simulation with Gromax there are people in my group that also like to use amber time to time but most of it is done with with Gromax then there is a second question which is also well related to a question which i had myself um this is a question from Ali standard gga functionals in BFT are well known to make island bonds very sticky due to the lack of dispersion can you comment a bit on the role of treating this interaction correctly and well how would using a different functional eventually change the results very difficult that's a very good point so i have to say that of course this relates to catalysis at the qm level so we are bound to the fact that in the methods we use typically carparenello as i mentioned the level of theory is dft and the functional is most of the time dft even though now there is the implementation but it's very super costly so and not really affordable at the end of the better functionals but of course now in our case we are lucky in a way if you if i wasn't if i'm not wrong you say iron right the metal you he was mentioned so that are more difficult cases because you have the open spin system and therefore you have to care about the oxidation states in our case we also we always have a closed systems and so magnesium calcium calcium and yeah not do not present these challenges so in a way we are lucky in that case otherwise i think yes if so sorry marco just i mean i wasn't even thinking of those types of exotic ions i mean just standard hydrogen bonds yeah are known to be very sticky and you know that the standard trick is to to include the semi empirical corrections like grimmer dispersion correction so it's with no with essentially zero cost and this helps so i'm just curious about whether this has been checked or whether this is known understood how much it affects things let me ask you back because i'm just for me to understand whether are you also considering the same empirical methods you would suggest or it's a correction you know it's a it's a it's a post-dft empirical correction no we haven't checked ourselves these effects to be honest at the end this is well started and it's giving the results of care so in the sense that matches the barrier we know the barrier at times is underestimated and that's accepted especially you know if you deal with the proton transfer then you also have other type of questions related to how that happens and the barrier for it but yeah the the the short answer for you is that no i don't know about these corrections enough to tell you if would help or not in the in the scheme we have all right thanks well i have actually a question which is a related which is in a way how much you can trust classical molecular dynamics when you start having ions in the business in the sense that i remember from the old times when i was also doing similar things it was always a little bit of an issue deciding how to treat the the metal ions now things are better you can trust the force field at least qualitatively how does it work yeah i i do see the challenge i'm well aware of that i would divide these in our case in two different problems we have the problem with the catalytic ions which are in the site and stay there and that's a problem related to the stability of the coordination and we show in the past that if you really distribute the charge of those two ions around the ligands there are ways in which you can stabilize these two octahedral coordination so the two ions so in a way if you have a good crystal then you stick to that coordination and you are fine because in a way we assume that there are no ligand exchange apart from the living group that is living i would say is more challenging when you have these binding and unbinding events of ions because in that case well first of all this diffuse diffusion of ions is not easy to see in at least to this scale we care in plain md and then you also have a ligand exchange the salvation salvation that is would say more challenging now what we do we use the the parameters for those ions those that are in the solvent we use the parameters that are now available the math or the parameters and of course what we don't care at that point very much on on the coordination of the ion when it comes and goes because we know that it's going to be very flexible but we care about the location and how many times we see this happening in our simulations all right so we still have two questions maybe we can move to the next two questions because otherwise we'll run short of time well the next question is how do you derive a most probable mechanism from bias simulation methods like a metadynamics this well that's a very interesting question because at the end you need to know the reaction and use your chemical intuition to come up with the reaction coordinated describes the chemical event under investigation so if i have an s and two reaction most likely the bond information the length of the bond information or the length of the bond breaking or the difference between the bond information and and and breaking some combination of the degrees of freedom the system should describe your reaction that's how we started and then it's a trial and error process so you start with the core reaction coordinate or collective value and you see if that leads the system where you want with a decent energy profile and then you repeat this till you find the correct crash coordinate which is never the true reaction coordinate most of the time it's a approximation that allows you to capture the chemical event you are interested in and that's how we we start basically and then we refine it okay so we now take the last question which is well i saw in the last example you use the 126 lander jones model for the divanet metal ion do you think that the use of 1264 lander jones model developed by li et al could produce different results so yes i know these new models that came out from mercer and these are very recent and now actually they are available in number and at that time we used what we had now this is a new model and i can't answer if this would change or not of course we have to try in the future to implement also now these new parameters for the ions and see i tend to think maybe i'm optimistic that yes we may see some differences in the energy but not so much to you know be at the end is a qualitative match what we look at is not really we want to in fact we can't really match with the decimals the barrier that we see and we don't know exactly if that barrier is for chemical step physical step so the answer is no we haven't tried and yes we have to try and of course it will be better most likely the overall picture i would say won't change much but of course it's interesting all right so this was the last question and it's also the last talk of this morning session so i would like to thank both speakers again for their very clear and interesting presentations we reconvene this afternoon at two p.m so see you this afternoon thanks for following bye bye bye