 over to you if you're there and I was muted so let's see if I can share the screen here we go can you see it I can see it I mean it's still it's not in present I mean yeah I can see all of your slides they're not just the main one yeah here we go so basically thank you thank you so they're organizers for inviting me basically this is the the talk the project that gave me the idea to start this idea of making this workshop so I wanted to present it even if it's two years later at the time I was working at Imperial and I got the money from Imperial now I'm working at UK HSA so and basically I don't have to make an introduction of what plasmids are they still so nice so easy but I wanted to just for a moment focus your attention on the fact that plasmids can be vertically inherited and horizontally inherited and what this does mean for the for the fast from how fast they can spread because of course vertically inherited spreads really is low because each generation they are inherited but horizontally and it it can go very fast and also we can have conegative and mobilization plasmids and the machinery length and I won't use any other name than machinery because Fernando will not like it I think it changed and okay I was just pointing out this also I don't have to tell you that plasmid dynamic is not so simple so they can inherit pieces from everywhere but I like this this image so I wanted to show you so the original question in this pro project was how much do horizontal and vertical inheritance contribute to the spread of antimicrobial antimicrobial resistance and the approach was basically comparative phylogenetics that try to relate to the evolution of a host bacterium the plasmid and the antibiotic resistance gene contained in the plasmid the really cool thing that I think hasn't had enough relevance the relevance that it needed to have and that maybe some of you will want to use this data set it's 116 plasmid sequence from 97 hosticola is and the cool thing is that they were sample from human and animal and from both pathogenic and commensal bacteria and that they were spanning a lot of and different as you see here in the pie chart the ESBL resistance alleles let's call them alleles also the other interesting thing that I put here on perspective on the on perspective so is that this plasmid or vesicoli samples were sampled from a very long very long between very long for plasmids very long time frame so there were three that in the end resulted to be just one E. coli from the Murray collection and then some that some quite a lot that were sampled before extend the spectrum that was in the 90s. The other cool thing about this data set is that it's extremely well sequenced so they were sequenced because it's a very old data set they were sequenced with the 454 so it's long sequence and some of them the long ones I think were electrocuted out and then put in the K12 and then sequence with the K12 so there was no problem whatsoever in reconstructing the plasmid because when there was something when there were ants in the sequence Catherine who is the main author everywhere in his paper went and sequenced all the gaps so you can trust these sequences in any way possible and she also annotated them manually and there was an automatic pipeline but then she checked every single gene manually and every single inter non-coding space if there was a gene that the automatic annotation was missing so it's a really well sequenced data set and you can find and download it if you need it for anything you want to try. The problem at that point was that we had the MLSPs and so we could reconstruct the hospacterium tree then we could in silico isolate the antibiotic resistant gene and use it to build the tree but the problem was what's the plasmid tree like because of course the ideas we had was like if the plasmid tree is more similar to the hospacterial tree that the plasmid is mainly is mainly vertically inherited while it was similar so if it's different there is significant plasmid conjugation among bacteria at the same time or yeah if the plasmid tree is very similar to the resist the plasmid to the antibiotic resistant tree then it means that the antibiotic resistance stays in the plasmid and just the plasmid conjugates while if the tree is different horizontal transfer of genes of resistant genes is happening heavily among the plasmids. So the problem was how to build the plasmid tree because we didn't have any one single gene that can compass the whole data set so there was no way of building a tree with the entire data set so we needed to result to subtree but how did we obtain the most informative and reliable subtree possible so informative meaning including as many plasmid as possible so that we could reconstruct a nice history and reliable meaning these trees relied on as many genius possible because of course you couldn't use just one gene that is shared among most of the plasmid but then it's not so reliable and then I came out with this data driven network that is kind of inspired by network theory to build the most informative and reliable subtree possible first I described the method and then I show the graph the method is called giant component analysis and is a standard analysis in network theory and so I started from a very strict threshold for defining shared genes between plasmid which meant that they blasted them with a very strict value then I computed the jacquard distance between plasmid in terms of shared gene so the jacquard distance is one minus the percentage of shared genes and the shared genes if I mean probably is too much is the number of shared genes divided the total number of genes present in both plasmids and then for each threshold for different thresholds in jacquard distance I determined but there is a command in a graph the jacquard distance I determined the size of the biggest connected component of plasmid and this is the graph that you obtain and you are looking at the purple dotted line so what are you seeing here in the left hand side so towards one we are connecting plasmids even if they share one single gene and we end up with a totally connected network of plasmids and we connect all of them even if they share all different genes on the left hand side sorry I have a problem with left and right we are connecting the different plasmids if they share almost all the genes and there as you can see we manage to connect very few of them and most of them stay there as single atoms and this is really you don't get any way of telling a history of the evolution of this plasmid the thing is that if you now focus on the on the purple line you would expect it to raise nicely from 0 to 100 while as you can see it's almost flatish and then it jumps all together and this is suggestive as similar of a process in network theory that is called the percolation and you don't expect this to happen everywhere it's something it's a peculiar process and it tells you something about the dataset characteristics and what we think this is telling us is that up to a certain point which is here you have information so you connect the plasmids and the genes that you are using to connect these plasmids are actually a signal of a shared evolution while from there on you connect plasmids just because by for the restricted probably size of the common pump plasmidome they end up picking the same gene but it does not contain any evolutionary information on that or evolutionary information so once we choose the threshold here we draw a a graph connecting the plasmid so here the shapes are if they are if they are mobilized the rounds are mob and the squares are conjugative and the stars are the stars so the the circles the stars and the triangles are different types of mob while all the squares are conjugative and when you color them by the amr gene that they share you clearly understand that they are not clustering by antimicrobial resistance while now that we connected the edges we color the edges by their incompatibility group you see that it's more like it that they are sharing that they are connecting by the incompatibility groups which is in a sense not so surprising because for can you get it plasmid then compatibility groups are big but the thing is that this paired with the the percolation like signal it's kind of telling us that uh something else that I have still to figure out completely so in the end I was able I isolated the here is just an example I isolated the genes that made the clustering and I was able to build the host bacteria tree the plasmid tree and the antibiotic resistant gene tree and here if you look at the red samples which are always the same you see that while in the antibiotic resistant gene tree they cluster all together in the plasmid tree two cluster and one is on the other side and in the host bacteria tree that it is they are everywhere so these trees don't resemble the example at all so the thing is here in this asset we have high mobility of plasmid among bacteria but also high mobility of antibiotic resistant genes among plasmid and now I wanted to finish with uh some seeing that I thought before but then after today's talk I think it makes a lot of sense to discuss it with you so I thought we all started this um this method by matching gene genes for similarity in sequence and the implicit assumption in this matching for similarity in sequence is that since uh the the data set spans a short time frame I mean if you think of the distance between human and sheep or whatever this is less than 50 years but also the plasmid is short so you cannot expect too many mutations in properly vertically inherited plasmids or gene and so the the more similar they are the more likely they are to be vertically inherited and therefore they are useful useful to tell us the story of the evolution of these plasmids but are all these genes that I used for the plasmid tree um vertically inherited but most probably not and were some of them lost and reacquired I think most probably yes and then can we define or is this method a way a good way to define uh I don't know a local uh in time and plasmid type um characteristic core plasmidum that we can trust that is mostly vertically inherited or should we give up the idea of using tree based um no of building trees based on molecular evolution to tell the plasmid evolutionary history and we should instead develop methods based on gene gain loss which seems to be what most of the people talking today are trying to do or are aiming to do but then the thing that puzzles me and on which maybe you can help me understand is that the phylogenetics work so well because there is an intrinsically random process at the base of it so like the mutation happens randomly it's a fair assumption in that too is a fair assumption to assume that mutations are up and randomly and do so my question is do we understand enough of the biological mechanisms that have plasmid gain and loss gene so that we can develop a properly reliable model that we can use because most of the most of the methods we saw today are very good at defining the plasmid the tree like the shape of the tree but phylogenetic does not only consist in the shape of the tree but also in a time frame so it's timing the tree so yeah I want to leave you with this question um and nothing this is the the people who worked on this project and the paper is down here and that's all that was a fantastic way to end the day that was great and lots of interesting questions for us um it's a bit odd separating questions from this talk from the general discussion but let's do it have we got any immediate questions for Alice before we open up for a wider discussion everybody's sleeping I'm sure they're not sleeping and I mean I uh oh hold on no sorry that's that's really I'm sure I'll say that um um I mean I think you raised the key thing which is that um we we got lucky with you know snips and um species chromosomes sootamine because it looks you know you can model things with a with a with a ticking clock yeah and uh the differences that we're seeing between plasmids I I don't believe that there's anything clock-like in there so it seems to me unlikely that we're going to be able to models from which would directly correlate with time with the events that we're seeing please anyone if they disagree please just feel free to talk um so I think I think we might be we might need to use these models to infer slightly different things than what we're doing with standard phylogenetic trees um I also think uh as Liam suggested at the end of his talk we could really do with more longitudinal data um yeah if anyone's got 25 years to spare and they want to start a long-term evolution experiment of uh of plasmids I'd be you know I'd be very happy to send a sporting letter um yeah but then people argue that like lenskis uh lenskis evolution experiment is an experiment and you cannot translate it into uh like clinical or like sure but I'd rather have some data than no data yes of course I mean um anybody else want to pipe up no really only two opinionated people here find out how to do