Scientist Stories: George Perry, Evolutionary Medicine

Hi everyone, welcome to the eLife Evolutionary Medicine Symposium. Thank you so much for joining us. We have 4 speakers today, and these speakers are, are going to be presenting their work, which are papers that are included in a special issue that is out in eLife, and I encourage you to encourage you to check out and to keep checking, keep coming back to check out further, because there are 37 published articles so far, but more in the works that will continue to be published. I'm George Perry, a senior editor at eLife and associate professor at Penn State University in the United States, and we developed this special issue with the excited understanding of the importance of using evolutionary theory to advance our understandings of modern human health. And there's a number of different areas of research in which this is done. For example, studying antibiotic resistance or considering the evolutionary history of humans in terms of how that affects modern human health and even dynamics of cancer progression within the lifetime of an individual. So again, please check out the special issue. We're here today to celebrate some emerging scholars who led some amazing papers as part of that special issue, and I hope that you enjoy the talks. So our speakers today are Stephanie Gann from Johns Hopkins University, Geronimo Rodriguez Beltran from Ramón y Cajal University Hospital, Josh Deeth from Imperial College London, and Juan Manuel Vázquez from the University of California, Berkeley.

Hello everyone, my name is Maria and I'm part of the ELIFE executive staff team. I would like to welcome you all to today's session. So on behalf of ELIFE, I'm very excited to hear today's presentation and to see so many of you here joining us today. Before we start, I would just like to go through some participation guidelines. And so we want everyone to enjoy today's meeting. So, to ensure that that happens, we will ask all participants to abide by EWAD's code of conduct. So, some examples of behaviour that contribute positively to our communities include showing empathy and kindness towards others, as well as being respectful of any differing opinions, viewpoints, and experiences that we may hear today. And acceptable behaviours will include making it difficult for others to speak or participate, for example, through repeated interruptions or disruptions. At the end of each talk, there will be roughly 5 minutes for Q&A. So you can submit your questions through the chat function at any point, and we will do our best to address all the questions within the allocated time. We'd appreciate it if you are asking any questions to also indicate your name and affiliation after your question and indicate the speaker you're asking the question to. For your reference, this session is being recorded and will be available to view after the event. A live transcription is also available by the Otter.ai app. Throughout the session, if you experience any technical difficulties, please message my colleague Anya Stasz of the live stream. Without further ado, I'll hand it over to George to introduce our first speaker.

Absolutely. Thank you, Maria. And thank you, Maria and Anya, for all the work that you have done to put this symposium together. And thank you Thank you again to our speakers for presenting their work here. And our first speaker today is Stephanie Yan from Johns Hopkins University, and her title is Local Adaptation and Archaic Introgression at Human Structural Variant Loci.

All right, I'll get started. Okay, hello everyone, I'm Steph. I'm a PhD student at Johns Hopkins. I'm in the lab of Rajiv McCoy, and today I'm excited to tell you about this project. So, um, studying patterns of genetic variation is a useful way for us to get insights into human evolutionary history. And so from this research, we know that anatomically modern humans originated in Africa, and then from Africa they migrated through the world where they encountered new environments, and then they experienced selection for variants that were adapted to these environments. So one really common form of this local adaptation phenomenon in human populations is selections on alleles that are related to immunity. So one really famous example is positive selection on sickle cell alleles that confer malaria resistance in humans. And so another key element of human evolutionary history is that during these migrations of human groups out of Africa, they encounter groups of archaic hominins like the Neanderthals and Denisovans, and they interbred with these archaic hominin populations. So this admixture event left all of the descendants of these migrating human populations with a small amount, about 2 to 4% per person, of introgressed archaic DNA. Most of the research on these events in human evolution has focused just on one form of genetic variation, which is single nucleotide variants or SNPs. People tend to focus on SNPs because these are the easiest kinds of variants to discover with short-read sequencing. But it also means that we're overlooking the impacts of all other forms of genetic variation. In this talk, I'll be focusing on structural variants or SVs, which are any larger insertions, deletions, or inversions that are also widespread throughout the genome. So the reason why SVs are usually not as studied is because they're hard to find with short read sequencing. So short reads are usually shorter than the length of an actual structural variant, and so short reads are relying almost completely on abnormalities in alignment to detect these kinds of variants. So for example, short reads tend to be especially bad at discovering insertions, where most of the insertion sequence doesn't actually align to the reference. So you just don't see most of these insertion reads. And then they also find it difficult to look for variants in repetitive regions of the genome that are generally more difficult to align to accurately. So because of these limitations, we've only been able to comprehensively study structural variants using long-read sequencing methods. So here the reads are actually long enough that you can sequence the full length of most of these SVs. But the main drawback of long reads is that they are much lower throughput and a lot more costly than short-read sequencing is. And this means that we still don't really have data about structural variants on population scales. And as a result, the role of these variants in human evolution is still not well understood. So in our study, we overcame this challenge by using a hybrid approach where we discover structural variants using long read sequencing, and then we genotype these variants in short read sequence datasets. And so we used a method for this called graph genotyping that was implemented in this tool called Paragraph that was developed by Illumina. And how this works is that instead of aligning your sequencing reads to a linear reference like you normally would do, you construct these local graph representations of known structural variant loci. So at these sites, the graph has one path that has the SV in it and then one path without the SV, and you can align your reads to the graph along the path of best fit, and it lets you get accurate genotypes for your sample at all of these SV loci. So you can scale this up, run it on thousands of samples, and you can genotype them for all of these SVs that you've discovered with long reads. So in our study, we started with 15 long read sequence samples from this paper by Evan Eichler's lab, and we used the samples to discover about 100,000 structural variants in these 15 individuals. And then we use graph genotyping to genotype these variants in 2,500 individuals from the 1,000 Genomes Project. So this is a really large and diverse human dataset that includes 26 populations from around the world. And so we did some quality filtering and this left us with about 92,000 SVs that we had genotyped in these individuals. So we then used the set of genotyped structural variants to look for variants that were under selection in human populations. So to do this, we use the method called Ohana. This was developed by Rasmus Nielsen's lab, and the first step of this method involves modeling ancestry in your dataset by representing all of the individuals that you have as combinations of some number of ancestry components. So the actual number of components that you choose is a little bit arbitrary, but here we chose 8 components total, which lets us look at more continental-scale patterns. And so here I'm showing these admixture results for all of the individuals in the dataset. So you can see that this is nicely separating out individuals into the 5 continents that are represented in the data. And it's also showing us known instances of admixture, like these 2 African American populations in the upper left, which have both African ancestry in purple and also European ancestry in these pink and green lines. So this first ancestry step allows this method to work effectively with admixed samples like these African American populations. And it also means that we're not relying on population labels that were manually assigned to individuals when maybe their actual ancestry is a little bit more complicated than just the label. So the second step of this method is that Ohana looks for SVs that were under selection by finding variants that have extreme allele frequencies in one ancestry component. So this pattern should occur with locally adaptive variants, because if something is undergoing selection in just one or two populations, they're going to have a higher allele frequency in that population specifically.. And so I'm representing this process visually with this tree here, where this one in particular is showing a genome-wide average distribution of allele frequencies, which we are assuming to be a null distribution where there is no selection that's occurring. And then there is a hypothetical model where a variant is under selection in just one ancestry component. So here I'm showing ancestry component 3, where it has this longer branch here. So you can compare these models, this genome-wide null and this selection model, to the variant's actual allele frequencies that you observe., and this lets you get a selection likelihood statistic. So we use this approach and we found 220 unique SVs in our dataset that exhibit this significant deviation from genome-wide patterns of allele frequency. So today I'll be talking about just one of these variants. So on this plot, I am showing the selection results for the ancestry component that corresponds to the Chinese Dai and the Vietnamese populations of this dataset that we used.. So the x-axis is the log-scaled length of all the structural variants that we're looking at. The y-axis is this likelihood statistic that I just talked about. And so the strongest signal of selection in this population that we found is this short insertion in an intron of a gene called IGHG4. So this gene codes for the constant domain of an immunoglobulin protein. So this was really exciting for us. We know that immune-related genes are common targets of local adaptation in humans. So this is a really cool result that we ended up finding. So these immunoglobulins, they mediate the adaptive immune response, and they have a variable domain which is in gray here. So in B cells, this variable domain undergoes somatic recombination and hypermutation to generate a diversity of epitopes that can recognize antigens. And then these immunoglobulins also have a constant domain which is in blue, and this domain determines which type of immunoglobulin it is. So this insertion SV that we identified is in one of the genes that codes for this constant domain. And another thing that's interesting is that this region of the genome that these immunoglobulin constant domain genes are in is at the very end of chromosome 14, and it's a pretty repetitive region. So we did a lot of kind of internal validations to sanity check our genotyping, and I won't describe them in a lot of detail, but this is a really hard region to study just with short reads, and we think that it's a case where graph genotyping or similar approaches can really make a big difference in these kinds of situations. Okay, so this IGHG4 insertion SV shows dramatic differences in allele frequency between populations. So you can see this from this map where I'm showing these pie charts of allele frequency where the insertion SV is in purple and then the alternative no insertion allele is in yellow. And so the insertion is almost fixed in the Chinese population. It's at really high frequency in the Vietnamese population. The other Asian populations have it at an intermediate frequency, and then it's almost absent in populations outside of Asia. We then started looking through the literature to look for other examples of research on these immunoglobulin constant domain genes. We found this paper by Sharon Browning that reported a haplotype in this region of the genome that was introgressed from Neanderthals into humans. Then we wanted to check whether the adaptive haplotype that we had identified in our study was this same Neanderthal introgress sequence. So what we did is that we looked for allele matching between Neanderthals and humans. So on this plot, the variants are ordered by position on the Y-axis, and then on the X-axis are individuals from a bunch of different modern human populations. And then we also have 3 Neanderthals and 1 Denisovan. And so the sequences for every individual are these vertical lines, and they're colored white at sites where they carry the reference allele and then black at sites where they carry the alternative allele. So from this you can see that all 3 of the Neanderthals, but interestingly not the Denisovan, carry this haplotype of almost all alternative alleles in black. And the haplotype is also shared by the majority of individuals in this Chinese Sai population and also the Vietnamese population. It's at lower frequency in the rest of Asia, and then it's almost absent outside of Asia. So this is matching the pattern that we see for the insertion SV that we found. This really high level of allele matching between the Neanderthal sequence and the human adaptive haplotype is confirming that the sequence was actually Neanderthal introgressed. The last thing that we did was that we used some simulations to further investigate the selection event on this immunoglobulin insertion SV. We used a 5-population demographic model that was based on these 2 papers over here to try to infer the time that selection began and also the selection coefficient. With these simulations, we inferred really strong and recent selection on this SV, where the selection coefficient that we inferred is actually on a comparable magnitude selection on traits like the sickle cell allele that confers malaria resistance. Then we also inferred that the onset of selection occurred within this gray range over here, which is right after this Dai and Vietnamese population split off from the rest of the populations. In Asia that are in the simulation. And so we aren't actually including the Neanderthal introgression event in the simulation. This is both for simplicity and also because the introgression would've occurred like around here. So that's before the simulation begins and it would've mostly served to just introduce this SV allele into the human populations. But it's interesting to note that based on the selection timing that we've inferred from the data, it suggests that this adaptive haplotype was introgressed into humans from Neanderthals many, many years ago during these Out of Africa migrations. And then the haplotype stuck around at some low frequency for tens of thousands of years. But then when these two populations in Southeast Asia migrated into a region where this haplotype was not advantageous, it rapidly swept almost to fixations. And so possibly Neanderthals that were living in Eurasia had already undergone some forms of adaptation to these to this trait, and then they contributed these adaptive alleles back into the human populations during admixture. In conclusion, we used graph-based methods to genotype structural variants at population scales, and we use these methods to identify an adaptively introgressed Neanderthal haplotype in the immunoglobulin locus. This is really interesting because it suggests that there are multiple levels of selection that are acting to increase diversity on these immunoglobulin genes. So they have this variable domain that is undergoing mutation recombination somatically, and then also a constant domain that we've shown undergoes selection to maintain diversity, not only between different human populations, but also between different hominin groups. And so together, this is demonstrating that structural variants are unexplored potential contributors to human local adaptation, and that we can hopefully begin to unlock the evolutionary impacts of these variants now with long read sequencing. So in conclusion, I'd like to thank my lab, especially Rajiv, my advisor, all the people here in bold who've contributed directly to this work, and then our collaborators. And thank you all for listening, and I'm happy to answer any questions.

Thank you so much, Steph. And yes, if anybody has questions, please enter them in the chat. While people are potentially typing their questions, I'll get us started. Steph, so are there any theories from that Cell paper or, you know, from your further work in terms of the functional significance of the variant that's been under selection?

Yeah, that's a great question. It's something that we're really interested in as well. We've tried to answer this in a lot of ways and we haven't come to any real satisfying conclusions, which is the short answer. But we've looked at things like Giovottis has a bunch of RNA sequencing data for some 1,000 genomes individuals. So you try to look for expression associations, but we're running into this issue where there is not a ton of phenotype data for East Asian populations. A lot of this RNA-seq data phenotype data is for mostly Europeans. And so we just don't really have the sample size to be able to satisfactorily try to get a phenotype. But if anyone has any datasets that they recommend for this or are working on, we are super interested in exploring this further. But yeah, it's a great question. No real concrete answers yet, though we're really curious about it as well.

Thank you. So Stephanie, a question came in, two questions. One is, why do we see these geographically distinct genetic structural variants? And can this be because of the genomic manifestations of environmental selection?

Yeah, so I think these structural variants are— I'll pull up the allele frequency chart again. So I think these structural variants, they are all on haplotypes of variants that are undergoing both neutral drift and then selection in some populations. And so I think in this case, through simulations and other forms of data, we've concluded that this is almost definitively a selection event and not just neutral drift. We also have in our model our selection scan method, because it is also incorporating the neutral allele frequency distributions that you just see genome-wide. This is inherently accounting for allele frequency differences that just result from drift between these populations, because you're comparing it to all the variants where you don't expect selection on all of them, and you're comparing just one variant's distribution.

Hopefully that answers the question. And another question is, so what are your next steps with this work? And I guess we could ask both on this specific locus that you're highlighting, but then the broader worldwide findings.

Yeah, so I think for this specific locus, I didn't have time to go into it in a lot of detail here, but it actually has a really interesting pattern of allele frequencies between populations.. And so I'll show this plot that I have in my extra slides. And so the adaptive haplotype that I was talking about most of the time is this like kind of yellow region in these like East Southeast Asian populations up here. And it looks really complicated here because there's like a bunch of different kind of forms of this haplotype. And some of them are kind of reaching higher allele frequencies in some of the European populations as well. And so it's really interesting. We also see a lot of recombination in this region, like at least 3 recombination events that have kind of split up this haplotype at different places. And so it's really interesting. We've been thinking about doing like more complicated simulations that are trying to— that would maybe be able to explain how we get this really interesting different patterns of haplotypes at different frequencies in different populations. So that's definitely one follow-up that we're interested in doing. And then on the broader dataset. Yeah, so I think more broadly, we're also interested in using this data to just more broadly look for evidence of introgressed structural variant alleles. And so like when we look for introgression in human populations, we are again mostly just focusing on SNPs, but I think it's also really interesting to ask like For SVs, like, are the distributions of introgression like different for these variants? Like potentially, like, are there inversion SVs that are inhibiting recombination and so they're like reduced for introgression at those sites? And so I think like using this set of population-wide SV allele frequencies to ask questions about structural variant introgression in addition to just like SNP and indel introgression would be really interesting. Awesome.

I agree.

Okay, uh, all right, so, uh, one more question. And the question is, uh, thank you— well, first a statement which I, I concur with— thank you for a great talk. Uh, what kind of medical implications does this have? And for example, would the findings impact drug discovery or personalized medicine tactics?

Yeah, so I think, um, the medical implications really depend on being able to get at a phenotype from this genotype data somehow. Either really directed wet lab studies with cell lines that carry the insertion versus without, or some more phenotype-based study of associations where you have a dataset that does have a lot of representation from East Asian and Southeast Asian individuals that have this insertion. Then you can start to ask, what phenotypic effects does this have Does it increase resistance to some specific form of infection or something? We did actually look for if there are any reports of Southeast Asian epidemics 66,000 years ago or something. Couldn't really find anything. But I think any future downstream medical implications will have to get at the phenotype associated with this haplotype first.

Thank you so much, Stephanie. Great talk and great paper. Thank you for being here and presenting your work. We will now proceed to our next speaker, who is Geronimo Rodriguez-Beltrán from Ramón y Cajal University Hospital in Spain. His talk is Collateral Sensitivity Associated with Antibiotic Resistance Plasmids.

Thank you very much, George. Thank you for having me. It's a real pleasure to be here. Today I'm going to present our work characterizing the collateral sensitivity responses associated with the acquisition of antibiotic resistance plasmids. But first, a little bit of introduction. Antibiotic resistance is one of the most pressing problems facing humanity. Indeed, recent estimations predict that by 2050, antibiotic resistance might become the leading cause of death in the world, with a staggering death toll of 1 person every 3 seconds. So these figures highlight that we need to act urgently to tackle this pressing problem. But how do bacteria become resistant to antibiotics in the first place? Well, as most of you know, the development of antibiotic resistance is a simple example of Darwinian evolution. In every large bacterial population, bacterial resistance to a given antibiotic may emerge through mutation just by chance, and these bacteria will take over the population once antibiotics are introduced into the environment. These bacteria will take over because sensitive bacteria will be eliminated or killed by the antibiotic, or if the concentration of the antibiotic is low,, it will at least select for these resistant bacteria that will not grow the sensitive bacteria. So this is the classical Darwinian selection, but a crucial factor for the development of antibiotic resistance is that bacteria can transfer antibiotic-resistant genes through a process known as horizontal gene transfer that I'm sure you heard about. So this allows antibiotic resistance to rapidly spread through initially sensitive bacterial populations and is one of the main problems we face when trying to control antibiotic resistance. One of the main vehicles for this horizontal transfer of antibiotic-resistant genes are plasmids that, as you know, are self-replicative DNA molecules that are able to coexist with the bacterial chromosome. Crucially, plasmids are able to move between different bacteria through a process known as conjugation that I am showing you here in this beautiful picture. Plasmids are one of the main drivers of antibiotic resistance, and therefore the strategies that we may come up with to fight the antibiotic resistance crisis should also work against plasmids. One such strategy is collateral sensitivity, is the use of collateral sensitivity. Collateral sensitivity occurs when the acquisition of resistance to one antibiotic causes an increased susceptibility to a second antibiotic.. So let's see an example because perhaps it's easier to understand here. So imagine this blue bacteria here which is resistant to antibiotic A, but it's somewhat sensitive to antibiotic B. So collateral sensitivity occurs when the evolution of resistance towards antibiotic B makes this bacteria less resistant to antibiotic A. So in some sense, this bacteria cannot become resistant to both A and B antibiotics because of this collateral sensitivity. This concept or this theory behind collateral sensitivity is really interesting because it could be used as a new therapeutic approach to fight antibiotic resistance. In principle, one could use a couple of antibiotics showing collateral sensitivity responses and design treatments that cycle through both antibiotics to prevent or at least reduce the emergence of resistance. However, although there is a huge body of literature of papers showing these collateral sensitivity effects, most of them are focused on chromosomal mutations. And as I told you before, to be useful, these approaches should work for the main mechanisms for the spread of antibiotic resistance, which are, of course, plasmids. And also, a second factor that we need for these approaches to be useful is that they should work in phylogenetically diverse bacteria and not only laboratory bacteria.. So in this work, we decided to test these two ideas. And to that end, we took these 6 natural plasmids that you can see here. These are plasmids that were isolated from different Enterobacteria that were isolated from patients. They are clinical Enterobacteria. These plasmids all encode antibiotic-resistant genes that you can see here colored in yellow. And these plasmids had different sizes from 8,000 base pairs to 147,000 1,000 base pairs. They had different mobility. They were conjugative, mobilizable, or even non-transmissible. They have different incompatibility groups. So this diverse set of 6 plasmids kind of represents the diversity of plasmids that we found in hospitals, right? So we introduced these plasmids as models of regular plasmids into E. coli, a laboratory strain of E. coli, which is called E. coli MG1655. And we just performed dose-response experiments with 13 antibiotics. In these experiments, for those of you who are not familiar with this kind of experiments, we measure the level of resistance of a given bacteria by recording the minimum antibiotic concentration that is able to kill that bacteria. This concentration is called the minimal inhibitory concentration, or MIC. For instance, in this example, this would be the MIC. We have the antibiotic being diluted here in these multiple multiple plate, and here the bacteria is unable to grow. So this would be the MIC, right, the minimal inhibitory concentration. So as I was saying, we determined this MIC, this minimal inhibitory concentration, using this method for the wild-type bacteria and bacteria carrying each of the 6 plasmids that I showed you before. So in this heat map, you can see the results, and here in every column represents the results from one plasmid that you can see here, and every row represents one antibiotic of these 13 different antibiotics that we used from 8 different class families of antibiotics. And the results that we show here is the fold change in MIC between the plasmid-carrying and the plasmid-free bacteria. So red colors represent an increase in resistance, and as I told you before, the plasmids were carrying resistant genes, so it's not unexpected that the bacteria showed some resistance to some antibiotics, like for instance cefotaxime.. But crucially, we found 6 instances of collateral sensitivity response to different antibiotics, which are represented here in blue and also with this asterisk. So for instance, the pOXA48 plasmid, which is a really clinically relevant plasmid which spreads epidemically and confers resistance to last resort antibiotics. So the acquisition of this plasmid induced collateral sensitivity to 2 antibiotics, which are azithromycin acitromycin and colistin. Or in other words, when bacteria acquire this antibiotic-resistant plasmid, they become less resistant to other antibiotics, which are in this case acitromycin and colistin. So we validated these results using different methods that I'm not going to show you here today. But at the end, we concluded from this part of the experiment that indeed plasmid acquisition can induce collateral sensitivity, opening the door for for developing these, these strategies. So next, we move to the second step of the talk, which is trying to— for these strategies to work, we needed that they were— that this response was conserved across the diverse phylogeny. So what we did was to take this pioxa48 plasmid, and we introduced it into 9 E. coli strains of E. coli distributed— that are distributed across the E. coli phylogeny. So here you have a phylogenetic tree of E. coli.. And I have highlighted here the E. coli strains that we use. So as you can see, they are quite diverse and they are distributed across the phylogeny. So then we are just as before, we measure the levels of resistance of the plasmid-free and the strains carrying this pOXA48 plasmid by using this diffusion assay. So these assays are a little bit different and are really simple. We just place an antibiotic-containing disc on the top of an agar plate, antibiotic, and we just let the antibiotic diffuse from the disk into the agar. So at some point, the antibiotic concentration is low enough so that bacteria can grow, which is what we see here with this white lawn of bacteria. So the diameter of the inhibition halo is proportional to the level of resistance. So here in this example, you can see that the plasmid-free bacteria produces a smaller halo than the plasmid-carrying bacteria, or the bacteria that carries the pioxa48. We use this technique for the 9 strains that I showed you before, and I know there is too much information in this plot, but let me walk you through it. First, regarding azithromycin, which was one of the antibiotics in which we found this collateral sensitivity— sorry, there you go— I have highlighted here in gray the instances where this azithromycin collateral sensitivity response was conserved, and you can see that in 8 out of the 9 strains the introduction of the plasmid pOxa48 produces collateral sensitivity to azithromycin, right. So this response is remarkably conserved. On the contrary, collateral sensitivity to colistin was only conserved in 2 cases, which is not that good, but nevertheless all strains show collateral sensitivity to at least 1 antibiotic, which suggests that collateral sensitivity can be used to selectively kill plasmid-carrying bacteria.. So next we move to test that idea, if we can selectively remove plasmid-carrying bacteria, antibiotic-resistant bacteria, from mixed populations using this experiment that I'm going to explain to you right now. So we designed 4 treatments, 4 antibiotic treatments, in which we— in 2 of them we maintain the antibiotic constant. So during the course of today, we treated the bacteria with azithromycin or acitromycin, yeah, or only with colistin over the course of 2 days. But we also designed 2 antibiotics, 2 treatments in which the antibiotic cycles. So we started with acitromycin, and the second day we changed the antibiotic to colistin, and the opposite the other way around. So we propagated a number of bacterial populations in these 4 antibiotic treatments and just recorded the number of populations that were able to survive, right? So here you have the results. So for the plasmid-free bacteria, which are highlighted here with these gray lines, right, and the bacteria carrying the pioxa48 plasmids, which are these red lines. So you can see that in all four treatments for this particular strain, the four treatments were able to kill preferentially or selectively kill or kill more bacteria carrying the pioxa48 plasmid than the plasmid-free bacteria, right. In some cases, both bacteria both the plasmid-free and the pOXA48 plasmid bacteria died. But in most cases, the bacteria carrying the plasmid died faster. And this is also true for— we tried, we repeated this experiment with 5 strains of those representing the E. coli phylogeny, and we found the same results, so that we can selectively kill plasmid-carrying bacteria in 11 out of the 20 treatment strain combinations. So suggesting that effectively collateral sensitivity strategies can be used to selectively eliminate plasmid-carrying bacteria. So this was the paper. So it was really nice. It's a little bit preliminary in the sense that we don't think that these responses are ready to be used in the clinic, but for sure they are a stepping stone in the development of these strategies to counteract the plasmid-mediated antibiotic resistance. Resistance. So, well, just to conclude, the take-home message would be that the acquisition of antibiotic-resistant plasmid induces collateral sensitivity, and that this collateral sensitivity is conserved, at least for the pOXA48 plasmid, and that we can use this collateral sensitivity to selectively kill these plasmid-carrying bacteria. So this was the paper, and I just wanted to thank the members of my group, and particularly Cristina Herencias and Álvaro Sanmillán, who who were the main authors of this paper.

So thank you very much, and I'll gladly take your questions. Outstanding, thank you. Yes, please, please again put your, uh, put any questions that you have in the chat box, and I will read them aloud. And while people might be typing, I'll go ahead and ask. So in terms of I know you said that this is preliminary in terms of application, but what would a broader strategy look like from the point of identification of antibiotic-resistant strain to then testing to then implementation?

What theoretically would the process be? Well, I think the ideal situation would be that to identify just one antibiotic that works with with as many plasmids as we can find, right. So in our study, we didn't find that. So the collateral sensitivity responses seem to be like quite specific. So one plasmid promotes collateral sensitivity to one antibiotic, but we don't find like general patterns there. So for this to be applicable and useful, we would need something like that. But in any case, we could design specific treatments for specific plasmids like the pOxa48 plasmid that I've shown you here for concerning plasmids And then what we would need to do is first— well, what we are doing actually is trying to understand why this collateral sensitivity happens, what's the molecular mechanisms behind these collateral sensitivity patterns that we observe. And if we can identify like general patterns there, we can maybe find drugs, specific drugs that will enhance the killing or make the differences like larger. So I think that would be the way to go with this approach. Because otherwise you are stuck with just searching for a specific plasmid which antibiotics will give this collateral sensitivity, and then that may be a little bit too burdensome to be effective in the clinic.

Okay, let's thank Geronimo and the great work here. And again, thank you for being here to give the presentation and for also submitting your paper to the special issue. And congratulations. Thank you very much. Yeah, thank you. Okay, Josh. Our next speaker is Josh Deeth from the Imperial College London. And his talk is Intraspecies Spread of AMR Elements via Recombination Within Pneumococci.

Great.

Thanks, Josh.

Can you see my slides all right? We can. Thank you.

Great.

Yeah, so hello, good afternoon everyone from London. My name is Josh and I'm a PhD student at Imperial College London, and today I'll be talking about my paper looking into antibiotic-resistant pneumococci lineages and how they evolved resistance. So a bit of background first. So Streptococcus pneumoniae is a gram-positive opportunistic pathogen that is a member of the Firmicutes, or now to be known as the Bacillota. It is a leading cause of mortality worldwide, especially among infants. Worryingly, there are numerous multidrug-resistant lineages that have spread globally, and this has led to the pneumococcus being one of the WHO's priority pathogens for research, for further research on antimicrobial resistance. So in general, resistance in the pneumococci can be gained by alterations to the core genes such as the persin-binding protein genes, PPPs, the persin resistance, or through acquisition of gene cassettes on mobile elements, which I'll focus on in this talk.. So in the pneumococcus, there is a limited plasmid repertoire, so integrative conjugative elements, ICEs, drive much of the resistance gene spread. And these can integrate within the host chromosome, but they can also encode their own conjugation machinery to spread between cells. And the first identified ICE, and one of the most abundant, is the tn916 element, shown here both in its linear and circularized forms. And this encodes the tetracycline resistance by the tetM gene. Present. So TnI/VI can integrate within the host chromosome, but it can also excise and circularise to move via conjugation, as depicted here, and as John Murray went over as well. So conjugation is whereby a pore is linked, is formed between two cells, and this allows for the stable transfer of an element into a new host. And this element can then integrate into the host chromosome by site-specific recombination in the case of ISes.. And so while there is evidence for TNO16 and ISes moving between species via conjugation, there's no in vitro evidence, however, for conjugation between pneumococci, which leads to a question of how these common elements disseminate within a pneumococcal population. So the other mechanism which these MGEs might be using to spread along pneumococci is transformation. Transformation is the uptake of exogenous DNA into a cell and its incorporation into the host genome via homologous recombination. This requires cells to be in a competent state, and it also typically favors the import of short sequence length and is biased towards sequences from closely related cells, which is due to the physical structuring of bacterial populations and the nature of homologous recombination. Pneumococcus is a naturally transformable species that exerts tight control over the regulation of its competence machinery. Transformation is key for major adaptations gained by Pneumococcus, with serotype switching to evade vaccine response mediated through gene exchange by transformation and homologous recombination. Also, common penicillin resistance is also gained by transformation, with numerous studies, some highlighted here, showing resistant PBP genes in human pneumococcus often originate from other streptococcal species, and with mosaic resistant fragments imported by transformation and homologous recombination. We also observe this in our paper with resistant PBP genes in the global collection of pneumococci with we look into, often originating from other commensal streptococcal species. But today I just want to focus primarily on the transfer of MGEs between cells, where there is less evidence for large MGEs moving via transformation. So the aims of this work were to firstly try and understand how ISes move within pneumococcal populations, secondly to try and get an idea for how widespread these genes are in a large, big collection of pneumococcal genomes, and then lastly to assess where these genes are inserting from, if they're reconstructed to have inserted gene recombination events in the host chromosome. So the results. We looked for 2 common MGEs among a global collection of pneumococcal: TN016, which I've just mentioned in the introduction, and TN1207.1, just shown here. This is a shorter 7-kilobase element, not an iS, but a transposon, which encodes macrolide resistance via the MefA efflux pump, shown in the middle. Both TN1207.1 and TN016 elements can often be found together in a larger construct such as the T9, T20, and T29 elements shown here, or even in even larger elements, sometimes up to 50 or 60 kilobases long, but can also contain other resistance genes. So the dataset we looked into was the Global Neumococcal Sequencing Project dataset, also known as the GPS dataset, and this has over 20,000 genomes collected and sequenced, mainly in the last decade.. The majority of these, 65% of which come from Africa and Asia, as seen in the map on the right. These isolates have been subdivided and clustered by PopPock into 621 different clavulonemococcal sequence clusters, which are sort of akin to bacterial lineages. So in terms of presence/absence of the MGs under investigation, tnl16 was much more present in the collection than TN1207.1. It was detected in over 5,000 isolates in 134 different GPSCs. TN1207.1, on the other hand, was present in over 2,000 isolates across 64 different GPSCs. TN916 was also present on its own in 3,860 isolates, while TN1207.1 was present in 635 isolates. TN1207.1 was actually more frequently observed as part of a larger TN916 element, such as the TN2010 or TN2009 element I showed previously. This large construct occurring in 1,150 isolates. So you can see here that these resistance encoding elements are reasonably common among a large global collection of pneumococci, but to understand how these elements move between isolates and where they come from, we need to identify how often these elements insert within an image, and where in the chromosome do they tend to insert. So to do this, we came up with an insertion analysis pipeline. Which looks firstly to subdivide MGE insertions based on where they insert within the genome and how large they are, to try and identify unique insertion types. And once these are identified, we go across all the GPSCs with an MGE present and reconstruct the number of insertions. These are then checked against predictions of recombination events across the genome to see if these MGEs are likely to be by homologous recombination transformation, as opposed to site-specific recombination and conjugation.. The insertions within recombination blocks then have their flanking regions extracted to test which species these insertions likely came from. So here's the reconstruction of all the different insertion sites for the TN1207.1 element. This circle represents the Pneumococcus genome, the genes with hits proximate to them identified. This is a reference genome without any MGEs present within, and in total 93% of the hits were able to be reconstructed, although some hits were in smaller contigs. So flanking regions weren't able to be reconstructed. But in total, there were 50 unique insertion types across 27 different insertion loci around the genome, these inserting independently across the collection over 200 times. The element was most commonly present within a TNR1-6 element, but not shown on this graph, but the second most common was within the TAG gene. An example of that insertion is shown here, with TN1207.1 in the pink here at the bottom, compared to the reference genome on the top, and the bars in between the two genomes represent the sequence percent identity between the hits. So this insertion appears to split the TAG gene, which is a methyladenine glycosylate involved in DNA base repair, and this depicts how MGEs can carry useful cargo genes such as macrolide resistance in the case of Tn27.1, but here they can also be disruptive to host cells upon insertion. TN916, there's a much larger diversity in insertion type of over 400 unique types, over 100 different loci in the genome. It was most frequently observed appearing as a TN-2010 element inserting near the FTSE gene, which is involved in cell division. We reconstructed over 1,000 independent insertions across the collection, with the most frequent insertion being TN916. Present within a larger 40-kilobase element, which inserted close to the ZMPA gene, which was a metalloprotease gene, and that occurred 29 times across the collection. So once we had these insertion types, the next step was trying to identify the MGEs inserting via recombination. So on the right, we have the distributions of the length of recombination events identified through Gubbins, and also the SNP density for these recombination events. Those events that imported in MGE highlighted in blue against the background of the rest of recombination events for the GPS collection. And so for TN27.1, 55% of the insertions were within putative recombination events, while for TN906 this dropped down to only 8% of insertions. So this probably reflects TN906 coding its own conjugation machinery as a way to move between cells. These MGE importing recombination events highlighted in blue here tended to be much longer and also much more SNP dense, which suggests these imports might be arriving from non-pneumococcal species, as they're slightly more divergent from pneumococcal donors. So to test this, we looked at the flanking regions for these insertions. And so here we plot the closest species match, the flanking regions in solid lines, of control regions, which are orthologous regions around an insertion from, and isolate with no MG present, and these are represented by dotted lines. So coral here represents matches to the Pneumococcus, whereas green represents Streptococcus mitis, blue Oralis, and purple Pseudohyponiae, and these are all common nasopharyngeal commensals which are closely related to the Pneumococcus. And so we can see for TN1207.1 here, the upstream regions in particular, and the top left-hand of the graph, match much more closely to mitis here, compared to control. For TNR1-6, the top hit is also Metis at early flanking links, but this soon converges back into pneumococci. But yeah, so this likely indicates that these elements are still inserting into pneumococcal populations from other non-pathogenic species. And in the paper, we highlight one really interesting example of interspecies transfer of TN1207.1,. And this occurs within the MDR pbin lineage shown here, which is a subset of the GPS dataset. And so the tree and there's some associated metadata here. And so in particular, we look at the German clades of isolates through the tree, and this is so named because the majority of isolates come from the German Reference Center for Streptococci. These all have TnTollio7.1 insertion, they're all of serotype 14, which is included in the PCV7 vaccine, and they're all generally penicillin-susceptible. So when we looked further into the insertion site of TN12S1 in this clade, we saw a splitting of the commIc gene, shown here in sort of cyany blue on a control, and then we see the TN12S1 insertion up here in pink in the M200 genome. This gene is necessary for cells to enter a competent state in order to uptake DNA through through transformation. So this insertion prevented these cells from acquiring DNA by this method, and this seems sort of deleterious. Yet the clade in Germany is quite large, it's the largest clade within PMN9, coming in at 250 isolates. So we wondered what could be causing the spread of this clade in Germany. Germany, especially over the periods of the late 1990s and early 2000s when these isolates were collected, tended to have a low rate of antibiotic consumption compared to the rest of Europe. However, the ratio of its rates of macrolide consumption, which TN12S1.1 confers resistance to, and persin consumption, which these isolates are susceptible to, are quite high and generally the highest in Western Europe over this time period. So we wondered if this unique pattern of antibiotic consumption could explain the relative success of this clade during the period. So to answer this, we formed some phylogenomic analyses focusing on the German isolates. And so we reconstructed the effective population size through time, shown in this middle plot, while incorporating the ratio of macrolide to beta-lactam consumption into the reconstructions. And so we observed that there was a significant effect of this ratio on the growth rate of lineage, which is shown in the bottom right here. So this is the growth rate of the effective population size over time, and we see that this peaks around the same time as the peak in macrolide to beta-lactam consumption for this lineage. So this could explain how T916 might have been able to grow in Germany during this period. However, we see that these isolates start to decline towards 2005 and they're not really present in Germany after 2006-2007. In 2006, the PCV7 vaccine was widely introduced in Germany and that covers serotype 14, which is all of these isolates. And so given that these isolates could not acquire DNA via transformation due to the insertion of Tn1207.1, they could not switch serotypes to avoid the effects of the vaccine. So this could explain their loss from the population. So in conclusion, we can see that these resistance encoding elements are quite common in this pneumococci population, that they insert often. We've also seen that Tn016 appears to not move by transformation that often. There is some uncertainty in this, while Tn1207.1 often moves through this pathway. We've also seen that while interspecies transformations are rare, selection pressure from antibiotic consumption can enable the cross-species spread of these resistance-coding elements, which has important implications in terms of preventing the spread of antibiotic resistance among important pathogenic species. Finally, we've also seen that the case of TAG and the COMI-C insertions, these MGEs inserting can often be deleterious to hosts, but again, such is the strength of selection for antibiotic resistance these insertions can often spread within a population. So thank you for listening. I'd just like to thank all my co-authors, particularly my supervisor Priya Ornith-Croucher, and my funder, the Wellcome Trust, as well.

And I'm happy to take any of your questions. Thank you, Josh, so much. And first, this goes to all the speakers, just, you know, so impressed and so inspiring to see you as early career scholars doing such impressive work and leading these, you know, helping to lead these papers. So just, you know, really, really great, really great moment to see. Okay, there is one question already, and if anybody else has other questions, please enter them as comments in the chat. This question is regarding the large diversity of insertion sites.

Could insertional gene inactivation events be leading to adaptation independently of antibiotic resistance phenotype?

So could you repeat the question again? I can, and I think it's also visible to you in the chat if you, if it would help to read it. But yeah, I'll read it for others as well. So regarding the large diversity of insertion sites, could insertional gene inactivation events—

so could it be insertional gene inactivation events that are leading to adaptation independently of the antibiotic resistance phenotype? That's a good question. Yeah, we haven't really looked into that in terms of how— so yeah, so lots of these insertion types we didn't get a proper chance to properly dig into in terms of such a great diversity as you can see in Tn016. So we can only pick up the big examples for Tn12.7.1 in terms of the obvious disruptions of COMEC in terms of competence, which prevents transformation. But yeah, that's something definitely that we could look into in further work in terms of trying to understand the interplay between MGEs, both in terms of how they can be beneficial to a host in terms of conveying antibiotic resistance genes, but then also the costs to the hosts that they incur in terms of disrupting metabolic pathways and disrupting core genes as well.

But that's a good question. Yeah, it's something we definitely need to look into more.

Agreed. Thank you.

So, you know, fascinating, fascinating work, including on the, you know, the connection to the, you know, history of vaccines and so on in Germany. So, you know, you have some timing information on these events.

My question was, is the timing inference only from the years at which, you know, data were collected, or based on, you know, do you have resolution to estimate, you know, certain events based on, you know, the phylogenetic tree and so on? Yeah, that's a good question. So we have the Germany, I think, with our earliest ice that was collected in 1992, and then most recent ice that was collected in 2007, so we had like a 15-year span. And so from that we've been able to sort of look at the effective population size over time, going back into sort of the origin of this clade, which was around the 1960s. So we can sort of see where it first emerged, and we're lucky in the case that we had this data from Germany because it's all been sort of systematically sampled, and it's all from the National Reference Center, so we were able to to properly employ these sort of phylogenetic methods and have a reasonable guess at the dates of emergence and things like that. And this was just quite a nice story that jumped out in terms of this linking between antibiotic consumption and growth of this clade. And we did look at sort of how just penicillin consumption affected the growth rate of this clade, and that sort of had some negative effects as well, which is nice to see, and macrolide resistance as well, which had the positive effect, which is sort of what we'd expect.

But this sort of ratio had the largest effect. So yeah, that was always nice to try and reconstruct with this data. Yeah, things like the, you know, your comments about antibiotic consumption variation, you know, that's something I lived. My daughter was born in Germany and then we moved back to the United States. You know, until you experience it, it is remarkable just the kind of cultural, you know, cultural medicine differences worldwide.

And then that's, you know, very informative with respect to these patterns and evolutionary processes in evolutionary medicine. So well done. Yeah, it's quite useful, I think, especially on a population level. There's lots of studies that, you know, even in Western Europe, lots of heterogeneity in terms of prescribing rates and how that's affected local rates of resistance as well.

But yeah, it's nice to see that link evolution-wise. Yeah, okay, we will move to our final speaker. Who is Juan Manuel Vázquez from the University of California, Berkeley.

And his talk is Of Mice and Elephants: Tradeoffs of Tumor Suppressor Duplication and Body Size Evolution in Afrotheria. Take it away. All right, thank you very much. So today I'll be talking to you about a different kind of evolutionary problem. So this is a problem that's intrinsic to all of life. It's a problem of how— of all multicellular life, and it's a question of how we actually grow and develop, like, our body size, but more importantly, how we avoid cancer. And so of mice and elephants is because of the differences in size between mice and elephants. And in order to understand this problem, we need to understand cancer, and the story of cancer starts with a single cell. So all of our bodies are comprised of various different cells, and every cell has an intrinsic risk of becoming cancerous. Cancerous. And this happens as cells acquire different mutations, which leads to them becoming this precarcinogenic and eventually like a cancer cell, which then forms tumors. And this is described by the multistage model of cancer formation, which was originally described by Armitage and Doll in 20— sorry, in 1954. And what this model basically dictates is that if you take any cell and you give it some arbitrary cancer risk, there's— it will acquire mutations over a period of time, and eventually you'll end up in this precancerous state where it has all the mutations necessary minus one before it becomes cancerous. And then with that one extra mutation, it will become a cancer cell which can divide indefinitely. And what this model says is that there's two real things that dictate your cancer risk, given all else equal. It's the number of cells you have that can become cancerous. Cancerous, and the time you have in order to develop cancer. And so if I walk you through it, the number of cells is because between individuals, the size of a cell is dictated by its function. So liver cells between humans are all the same size, heart cells are all the same size. So the difference in height, for example, between a short individual and tall individual isn't actually because of different size cells. It's because of the different number of cells. Cells in your body. And so that means that over periods of time, every time they acquire these mutations, there's more cells in the tall person which can acquire mutations and more likelihood— there's a higher likelihood that any one of those cells acquires a pro-cancerous mutation, which means over the course of the lifetime of the short person and the tall person, the tall person's expected then to have a higher likelihood of any given cell in their body acquiring all the mutations necessary to actually form a tumor. And if you were to look at a meta-analysis, you would expect to see that there is a meta-analysis of cancers. You would expect to see this positive correlation between the cancer rates and the body sizes. And that's actually exactly what you see. So in this study shown here, there was a meta-analysis of various different cancer cells across the world. And what you can see is that the hazard ratio relative to height is generally greater than 1 for all types of cancers. In other words, In other words, with your increasing height in a population, you have an increased risk of cancer. And the overall mean tends to be 1.12, which means that a tall person has a 12% likely, uh, greater likelihood of, uh, cancer relative to someone below. I think in this study it was like the median height was like 5'5". Sorry to all the tall people in the talk. Uh, so the other thing I talked about is time, right? And this one's a little bit more intuitive. When you're born, you're kind of a fresh slate, and over time then you acquire more mutations that can then lead to cancer. If you die relatively young, you're not likely to have, like, the time necessary to actually develop cancer. On the other hand, in old age, you do have that time frame. And so many of us are sadly familiar with this in that older people have a higher incidence rate of cancer than younger people. This is data from the National Cancer Institute SEER study, which then shows you that different, different age cohorts, the incidence rates of all cancers increases in humans with age. As I've just said, either being long-lived as a human or being taller as a human leads to this increased cancer risk relative to your peers. But this isn't just a human phenomenon. This has also been shown to be true in mice. It's been shown to be true in cats and dogs. Even in elephants, there's some data suggesting this is true. This is one of those phenomena that within every species, it holds very well. The question is, what happens between species? Because if you look at different species of mammals, for example, there's a huge range of body sizes and lifespans that we see. This is data taken from the Haggard database of all the published information of body size and lifespans for all mammal species they can get a hands-on. You can see that you have mice and bats over here, you have dogs and other carnivores in the middle, And then you have humans as a giant outlier, elephants and whales being enormous. And compounding this problem is this positive correlation between body size and lifespan that you see across all mammals. So if you stop and think, well, OK, what's the cancer risk that we expect here, right? It should be higher on the right side, higher on the top side. So that means that if you were to try to imagine what the data showing that cancer and body size correlates in these species, right? You'd expect to see a graph kind of like this, where if you were to be able to take animals and then just look at them at time of death to find tumors in their bodies, right? So you're looking at percent necropsies with tumors, and you look to see how that correlates to body size and lifespan. You would expect to see this kind of positive correlation, right? It's where, like, mice and small short-lived species kind of fall over here, and then they're positively correlated cancer risk. So elephants and whales are just constantly dying of cancer.. But of course, I wouldn't be here today if that was actually true. As it turns out, this is data now, it's a little old from 2015, but there's now more convincing data from this group published in 2020, basically showing that if you actually take real-life necropsy samples both in the wild and captivity from all these different species plotted here from all these different orders, there's actually no correlation between your cancer risk and your body size and lifespan. So this paradox has become known as Pitoh's paradox, named after Sir Richard Pitoh, who talked about this between mice and humans, of mice and man, in 1975. And so the paradox has kind of trivially true solution. Obviously, over the course of evolutionary history, evolution works to decouple the cancer risk between body size and lifespan. And so large species evolve countermeasures to the increased cancer risk that comes with their size. However, we still don't know what the mechanism is, and that's kind of where, like, the mystery lies. Of course, there's already a genetic toolset common to all mammals of tumor suppressor genes, which serve to reduce your cancer risk. And so this works in a variety of different ways, either by killing off precancerous cells, repairing them so they become not so precancerous and fixing mutations before they arise, and/or just arresting the cell and preventing them from dividing any further and propagating these mutations. So evolution can act on this in one of two ways, either through sequence evolution— why, like, change the wheel? Just fix the wheel, make it better. Or if you're carrying a heavier load, have more wheels, right? So the 18-wheeler model basically duplicates the genes that are tumor suppressors in order to reduce your cancer risk. So in this study, when I started my PhD, we really started focusing on this gene duplication model as a kind of way way of reducing your cancer risk, um, in a very quick period of time. And so this is important in some species where suddenly there's like a large individual that emerges, right? Such as, for example, elephants. Um, so this here is the phylogenetic tree of all the members of Atlantogenata, which includes, for example, elephants. And what many of you may not know is that the closest relatives to the elephants shown here are manatees and the rock hyrax.. And so for a sense of scale, elephants are almost 5 metric tons, whereas like manatees are actually quite large, even though you, if you've never seen a manatee before, you may not have thought about it. They're also like 300 kilograms. And then the rock hyrax is about the size of a medium-sized dog. So this is a huge diversity in lifespan that just occurs over here in about like maybe 50 million years of divergence. If you then look at the rest of the tree, the rest of the tree is also quite small actually. So there are aardvarks, over here, not labeled. And then like all these other species, like the lesser hedgehog tenrec, Cape elephant shrew, and then over here you have sloth and armadillos. And so what's nice about this clade is that, you know, it seems like they're all kind of small, right? We have genomes for some of these. So like later on, like we'll be able to do genetic analyses. And importantly, we have ancient genomes of various extinct elephant species like mammoths and mastodons. But if you look at this tree, you kind of think, oh well, you know, elephants are large and everyone else is small, so like it was a sudden shift towards body size in elephants. And so that's kind of where we started on looking at gene duplications. However, you know, we need to find out whether or not this is actually true. And as it turns out, this monstrosity right here, the Titanohyrax, is the ancestor of the current tiny hyrax. And likewise, you all are probably familiar with like the giant sloth, which used to like basically dominate like the forests of South America, and it's the common ancestor of the current sloth, right? The, so clearly not historically speaking, it hasn't always been the case that the elephant is the only giant as species. Actually, historically speaking, all these different clades have undergone, seem to have large-bodied ancestors. So we really set out to see like, how does this actually look like on phyllogenetic tree? If you actually look to see and, and ancestrally reconstruct body sizes, what are the ancestral body sizes in this tree? Is it just that the elephants were always large or the elephants tiny? And as it turns out, that's kind of what ends up happening. So here I have the phylogenetic tree. Now the branch size lengths have been rescaled for how quickly body size has changed. And then the color intensity is how big they became on a log scale.. So as you can see here, there was this large shift in, for example, the sloths where like suddenly you have like the giant sloth. And that was a change that happened relatively quickly in evolutionary timescale. And likewise, you used to have like these giant armadillos as well, which happened. But for the tenrecs and for this Afro-soricida clan, there were some individuals that were much larger that evolved large body sizes suddenly., but as a general rule of thumb, they stayed small. And then if some of them became large, they became large quite like, you know, towards the end of fossil record. Then you have over here, like, you know, like this big shift in Pseudongulata. So the common ancestor of all these species shown here where they suddenly became quite large. And then some of them, like for example, the aardvark became smaller afterwards, but then you had another big shift in body size at Pananglia. Whalelets, so the common ancestor of the rock hyrax, manatee, and then the elephant. And this kind of makes sense if you think about it because they all, I showed you, they all have large body size-like ancestors. I did not show you Steller's sea cow, which was an elephant-sized manatee that used to live back in the day, and it's that branch here that's like very dark. And so what you can see is that there's like this accordioning, right? Like the body sizes increase and decrease and increase and decrease. And the cool thing is that at the common ancestor of elephants, they became about the size of large dogs. Um, and that is kind of like the common ancestor to all elephants. And we know that there were tiny elephants. We have evidence of like these dwarf elephants. But the cool thing is that there was like two coordinations that happened in a very short period of time. Uh, we have these different, um, large-bodied elephants that suddenly evolved, but in a very small period of time, it kept a coordination.. And so we kind of wanted to know, okay, well, how does this influence the cancer risk, right? As I've already talked to you about, there's a positive correlation between body size and lifespan, and the model predicts that you should be— your cancer risk is proportional to those two. We have ancestrally reconstructed body sizes, so we can now use this relationship between body size and lifespan and extent lifespan information to basically run a linear phylogenetic linear regression. To predict the lifespan for all these nodes as well. And so we can also then see how cancer risk evolves in this tree. And so shown here, it's a little bit non-intuitive, but cancer risk is predicted to decrease as body size increases, right, because the animal needs to exist. So even though the risk, the real risk increases, the cancer risk must decrease in order to compensate for the increases in body size and lifespans. So you can see that every time you have this dark long branch here, there's a corresponding white branch here of the decrease in cancer risk that needed to occur. And then you have an intensification of cancer risk in some species, which then suddenly decrease in size relative to how they were before. And so this kind of gives us a bunch of nodes which we can look at and be like, okay, well, if we were to predict some evolutionary response to cancer risk, it should occur at these nodes where there was a large increase in cancer risk. Therefore, there must have been, because the body size and lifespan, therefore there must have been compensatory decrease in cancer risk overall due to— like, in order for the animals to survive. And so in order to identify these, we basically searched the different genomes that I had shown before in this clade in order to see how many copies of tumor suppressor genes they had. And so in order to do this, we use the reciprocal best-hit BLAST method that I developed, a pipeline that I developed. The method is actually quite well done in the literature, And so basically the pipeline automates this process of taking your favorite gene and taking protein sequence of your favorite gene and looking in all these other genomes for all instances that you have that gene pop up. For all those forward hits, you then want to ask the question, is this really the gene that I searched for, or is it a close relative of the gene I searched for? Is it p53 or is it p63, for example? And so you then take all those forward hits and you search back the genome you started with. So in this case, we use the human genome and the human genome proteins as a reference, and we search all these other genomes with the collection of human protein-coding genes. And then for each hit that we found, we searched the human genome back again. We only called it a real copy of our gene of interest if the reverse search also returned our gene of interest as the top hits. And using this method, we found 13,700 or so genes that were conserved, that we were able to identify across all the different clades. And so it's worth noting here that it seems to increase in a stepwise fashion obviously there's not that many copies that were duplicated at the common ancestor and various different genes duplicated across the tree. However, you'll notice two things about this tree. The first one is that it seems like there's a reasonable number of duplications over here, and then suddenly there's like 2,000 are duplicated in the Cape elephant shrew. What we find is actually that the quality of these genomes over here were lesser than the quality of the genomes in the elephants, so there was already a bit of a confounding factor where we can't really say much about these gene duplications because of the, first off, the number of duplications that we have, but also the fact that relative to the tree I showed you before, we don't have very many species here. So this really serves to polarize the changes that we see on this part of the tree, the Afrofereans. And so this is where we directed most of our attention. It's also where we saw most of the body size coordinating occur. So we looked at them to see, well, we have all these gene duplications, right? What does that actually mean for us? We want to actually see what pathways these genes are involved in and to look to see in these gene pathways what's actually being enriched for. Are we enriching for tumor suppressor genes or for random other genes? And so we developed this UpSet plot, which I'll introduce stepwise. First, here is a histogram of the different number of genes that are enriched according to different pathways. Of pathways that were enriched according to the genes that they had duplicated using the Reactome pathway compendium. And so in black is just all the pathways, and in purple are the cancer-specific pathways. And so you can see is that for a lot of these body sizes where I mentioned there was an important— there was an increase, there was a decrease, you do see that there is an enrichment for cancer— like genes associated with cancer pathways. For example, in the African elephant, over half of the genes that are— like the pathways that are enriched that are enriched in the duplicated gene set for African elephants are enriched for cancer pathways. Likewise for Tephytheria, which is the common ancestor of elephants and manatees, and Panangulata then is the common ancestor of elephants, manatees, and the rock hyrax. When you then look to see how these different reactome pathways are enriched across the different clades. For example, this line would indicate pathways that are duplicated, where genes are duplicated both in the African elephant as well as in the aardvark, as well as in Tephitheria. And you then just ask the question, okay, well, what are— what is the— like, right, like how many of these have enriched cancer pathway intersections? You can see that a couple of these pathways, for example, proboscidea, has the most enriched cancer pathways amongst all of its groups, amongst all of its duplications. And likewise, then everything that involves the African elephant seems to actually have the majority like, you know, they have many of these cancer pathway intersections that are in common and shared between these different groups. When you look to see what that looks like, you find you get this word cloud where everything in purple is a gene duplication— it's a pathway that is enriched only in that clade, in this case Proboscidea. And then green is everything that's enriched only in Proboscidea and one other species. And blue is everything that's enriched only in tephitheria, so only between elephants and manatees. And so you can see these are all cancer pathway genes. These are not enriched for anything— like, I did not do any exclusion for other things. So it's just cancer-associated genes that are located in these gene sets, which means that gene duplications that occurred in elephants and in large common ancestors of elephants were all enriched in these pathways involving cancer resistance. Steps. Of course, this is all historical. Here I'm showing you a subset of the genes that we've found in elephants, and where basically you can see the duplications occur throughout the gene. Many of these duplications occurred very recently in the African elephant, which underwent a sudden increase from the common ancestor, which was small, to like the current large African elephants. But there are some duplications that also occur in the manatee, some duplications that also occurred at the roots.. And now these can just be artifacts, right? So you could have the retrogenes which are no longer functional. You could have premature stop codons and such. But what we found actually is that if you look at gene expression data in the elephants, these gene duplicates are actually all expressed in the elephants, in modern-day elephants. And therefore they need to have some functional significance, whether it's fully protein-coding gene duplications Or maybe they're truncated to mRNA mutations, and now, as evidence has suggested, p53 function more as like sinks for mRNA-level regulation of the master gene. And so all this together does suggest that like not these things that we're finding, they're not just different skeletons and like fossil record of like genes that used to be functional, but rather they're actually currently functioning gene duplications in the elephants. So in conclusion, I've shown you that body size and cancer risk has increased throughout Atlantogenatans shown here, but especially in this tephitherian lineage, red and green— red and yellow. And while these gene duplication events are relatively common and scattered across the tree, elephant lineage-specific duplicates are actually the ones that are enriched in cancer-related pathways, and they're conserved amongst elephants and they're still functional and conserved in modern-day elephants, indicating they're still functional in these elephants. There's a big open question to this, which is what my current work is now focusing on and what I want to leave you guys to think about for the rest of the day. These gene duplications are in tumor suppressor genes, which by definition inhibit growth. Yet I'm talking about animals which are gigantic. There's this new paradox which is formed, which is you somehow managed to become the size of a whale, while working as hard as possible to make sure that your cells don't divide. And so, and many of these other genes that are pro-apoptotic, the elephant's p53 duplications have been shown to be purely like working by cell death. And we know from like current model organism studies that these duplications of like these genes I'm showing you are duplicated in the elephant. They all have very negative like effects in mouse models when you overexpress them. Which indicates that there's a whole other side of this paradox that we're not looking at, which is genes that evolved in response to genes which evolved in response to body size, right? So it's this kind of like arms race between the pro-growth, pro— like the pro-growth and pro-large animal genes and these anti-growth, anti-tumor suppressor genes. And so whether or not— like how this— they orchestrate this dance for you to develop into a large animal,. And then when you see your adulthood lock down any further growth is kind of an open-ended question, which I want to leave you as future directions for this project. And with that, I want to acknowledge my PI who may be on this call from my grad student work, and he's currently at University of Buffalo, Vincent Lynch.

This work was done when I was a grad student at University of Chicago, and I'd like to thank my current funding from NSF. And with that, I'll take any questions.

Outstanding.

Okay, we have 2 questions so far. And anyone else, you are please welcome to enter questions in the chat, and I'll read them out loud. The first question is, in the presence of cell death, the number of mutations does not necessarily correlate—

the number of mutations does not necessarily correlate with the number of cells. How does this affect the argument? The presence of cell death— so I'm going to interpret this as a mutation can occur which then kills off the cells. That is true, but it's still, in this case, there's a term that I did not talk about at all, which is just the C term, the proportionality constant. If you have cell death, you can underestimate the number of somatic mutations, but that doesn't preclude the fact that there will be cells that survive. That actually hold all these mutations. And that's the more important key part here, right? Even if you have an insanely high cell death rate, as with the elephants, for example, that can help deal with this problem. So actually, what Vincent's group had shown before, right when I joined the lab, and then what I worked on my PhD, is the elephants do kind of increase cell death turnover in order to kind of get rid of all these mutations. And that is considered pro-like, or rather an anti-cancer mechanism. So you just get rid of any cells with any mutations regardless of how severe it is because you're just really putting a lot of weight on the double-strand break and any mutations and shifting the response from repair to just apoptosis.

I think that answers the question. Thank you.

Okay, next question. Do you think there could be gene loss events in addition to gene duplications that might have conferred cancer resistance in these species? That's the question that kept me up at nights. So the problem is that in these genomes, we have very— I was not comfortable myself at considering gene losses because we're not talking about human-quality genomes. We're talking about genomes where I had to correct for fragmentation, and like fake heterozygote, like duplications to the heterozygosity, and many other things which I did not get into in this talk. So we focused specifically on gene gain because it was a lot easier to parse apart than gene loss. Did we not find it because the gene was just too different from the mother gene, so we wouldn't have identified it as like a real duplicate? Or was it, right, it would have failed like the percent identity thresholds? Or did we not find it because the genome was just awful? Or did we not find it because it literally isn't there? My guess is that yes, gene losses are going to be very important as well. It's just that with the current state of affairs, we don't have the power.

But I'd love to look into that as like VGP and many other projects are making better genomes for these species. Thank you. Okay. Next question.

Wonderful to see emphasis on the trade-offs of cancer suppression mechanisms any further ideas about how to study these, perhaps by looking at subsequent genetic changes associated with gene duplications? Yeah, so one thing that I kind of like, you know, finished my PhD right when because of the pandemic, I actually really wanted to look into like subfunctionalization, neofunctionalization in these gene sets, right? To kind of get at that question, like there's additional gene like changes that then results from these duplications. Whether— like, I think the next step for a lot of this work is more like doing like ex vivo work. So like working with elephant cells to actually like characterize these in like massively like parallelized assays or using organoid models to actually like see, for example, this is very broad, but like I said, and I think there's a new question that's kind of getting into this, there's a very strong developmental aspect to this that needs to be addressed as well.

And so I think looking at the gene regulation, like the gene regulatory context of these duplications and how they avoid just shutting down the animal massively, are probably the next steps here. Thank you. Okay, and regarding the, the new paradox that you presented, do tumor suppression genes inhibit growth or do they regulate growth.

If it is their job to regulate growth, then finer regulation is not necessarily paradoxical, right? Kind of, yes and no, right? So like, these genes kind of operate on a spectrum where I've shown that like one of them, these duplications, the duplication of leukemia inhibitory factor that happens in the elephants, is kind of like a kill switch that gets turned on by p53. It's a zombie gene. It was like it was a pseudogene that got refunctionalized by the evolution of p53 binding sites. And it just so happens that the new version of the zombie gene, when it gets turned on by p53, instantaneously kills the cells. So that would be straight up, like, not in neither inhibiting nor regulating growth. It would just increase cell death, right? But there are other genes that I looked at in my PhD, and I'm trying to wrap up the analyses now over Christmas, this must break, so stay tuned. Essentially what happens is that there are cell cycle genes which are duplicated, and then they, like, if you express the gene duplicates in any other cell type, like CHO cells, in human cells, in other species, related species cells, it just shuts down growth. Like the cells instantaneously, like, you know, within 18 hours they undergo senescence and that's it. So it seems like it's more of like this— it's like going to be gene-by-gene specific, and there's definitely going to be some cases where it's just a regulation of growth, in which case how they regulate growth and how strictly they do that is going to be the bigger question than like, you know, do they straight up stop growth. I think that like regulation is like the key word there.

Like we need to— this is showing that the repertoire of genetics exists for evolution to act on. Now the question is, how is it actually working? All right, thank you.

Another question: it would seem that the advantage of duplication to anti-cancer genes isn't to double the effect, but to provide extra assurance against loss of function if there were a loss of function mutation in anti-cancer genes, i.e., the duplication is more of an insurance policy. Reasonable? So I was tempted to agree with you at the beginning of my PhD, But this graph I show here is kind of like the counterarguments. So if I were to show you differential expression data between these genes and other closely related species, for example, the levels of Cas9 being expressed in these cells is much higher when you add up these two copies than the levels of Cas9 in other species. So if it were just to be a redundancy, you would expect that the species would either equalize two levels to normal expression of the two genes, so that way if you duplicate it, then you half expression of both of them, so you don't end up with this overdose of the gene. But in this case, there seems to be, for many of these genes, a straight-up overdose. And this is gene expression data from PBMCs, for example. And other genes and other tissues from elephants. We're also not talking about cell culture models, we're talking about relevance in vivo elephants, like gene expression. There is probably going to be an insurance policy aspect to this story, but there's probably going to be more of a functional, there's an active risk because it's also the risk of cancer isn't a passive one that sometimes pops up due to mutation, mutation occurs every time you have cell division.

So when you go from, like, you know, the zygote all the way to a giant elephant or giant whale, there needs to be, like, this constant surveillance, which I think is the more key part here. Okay, outstanding. Thank you for the questions and the discussion. Yeah, fascinating, fascinating subject. And I think there's a lot of opportunity to integrate your comparative species findings here with within-species patterns of variation and effects on cancer risk where we see, you know, different pattern, and then we've got all the population genetic tools as well to consider. So great work. Yeah, I want to, you know, quickly wrap up the entire symposium and again, just thank our speakers, thank all the authors, you know, other authors of papers in this special issue on Evolutionary Medicine and eLife are in the audience. Yeah, thank you for your participation. I think it's a fascinating time with— in terms of being able to bring a lot of tools to evolutionary theory, evolutionary data, evolutionary genomics, experimental tools to bear on integrating evolutionary concepts with medicine and human health. And you saw different examples here today of how we go about that. From studying antibiotic resistance to studying human evolutionary history in the context of human health, studying cancer processes, and also, you know, in Juan's talk, the comparative, you know, the kind of comparative long-term evolutionary biology model there. So, you know, one thing I'm excited about in the field writ large is to see these different groups and different types of studies come together and integrate in terms of methods methods and practices because, you know, their human evolutionary history is affecting also a lot of dynamics going on with antibiotic resistance and so on, right? So, you know, that's something that I'm excited, and I'm excited about our special issue, which again, you know, please check out. There's 37 articles published so far, but more coming through the pipeline over the next few months.. And yeah, thanks everyone again for participating today.

Really exciting to see you as early career scholars doing such great work, really inspiring for me and I'm sure others on the call. So thank you, thank you again. Maria has a few final words. Thank you very much, George. I wanted to thank our speakers for the great presentations today, our audience who took the time during the the busy period before Christmas to join us and listen to these talks. And PJ, for the great chairing, thank you so much. And equally for leading the work with eLife to promote evolutionary medicine as a topic of interest to the community. So we do hope that by signaling with the special issue that eLife is interested in kind of work that the people on the call will consider sending us their research for consideration.

At Healife.

We will be sharing the recording of the session online in a few days, and this will also be shared with you by email.

So without further ado, I wanted to thank you all for, for the time to join today, and I hope you have a lovely holiday break. Thank you so much.