Contributed Papers X: Methods, Models, and Perspectives

Well methods, models and perspectives. I’m the moderator, Andrew Henry from McGill university Uh, the questioner who’s monitoring the question board. Uh, we expect to be, uh, Inbar Mayan, but, uh, you can also direct your questions to Michael Whitlock, uh, either through the Slack channel or on zoom. So the first speaker is Thomas Hitchcock, uh, who will speak to us on a jeans I view of sexual antagonism Go for it, Thomas. Cool. Thank you. Let’s see if this works. Is that all good? Yeah Okay. Uh, yeah. So I’m Tom Hitchcock and, uh, I’m going to be talking about some work that we’ve been doing recently on sexual antagonism and in particular with a focus on different forms of haplodiploidy. So, before I begin, I just want to thank some people who’ve been involved in this work. Um, Andy and Laura, who are the kind of coauthors of this antsy based at the university of St. Andrews and Laura based over at the university of Edinburgh, as well as Jack, Sam, uh, the rest of the gardener group and the theory lab for really helpful discussions and the university of St. Andrews for getting money. Okay. So males and females have distinct reproductive strategies, and this may lead them to having distinct Optima over range of traits, whether it be morphological, physiological pay for all or so on. And so here, we’ve got two particularly striking examples on the left. We have the rusty tussic moth with these slight winged mails and these slightly plumper wingless females. And on the right, we have the humpback anglerfish, which have this display, this extreme sexual dimorphism in size kind of main fish. You can see as a, is a female And I’ve ringed here the much smaller parasitic, two alpha male, but nonetheless, although males and females might have distinct phenotypic Optima over a range of traits, they nonetheless share essentially the same genome. And this may lead to trade offs between attaining high fitness in females and in males. And this trade-off is a commonly known as section on Satanism or sometimes on the name of interlock as sexual conflict. So most of the work that has been done on this has focused on diploid organisms, but today I’m going to focus mostly on haplodiploidy. So these organisms account for about 15% of all animal species and span a wide range of genera, uh, everything from their, a nematode groups that happily deployed to rotifers and here we have some springtails fungus gnats, wasps scale insects thrips, and even Embracia beetle they’re in the bottom, right? And it’s all these organisms are United by this asymmetric inheritance system where males only pass on them, maternal origin genes, but this can come in slightly different forms So in particular, there’s kind of two forms of haplo diploid. And I’m going to talk about today on the left, we have our Anatole key, which we might think of this kind of classic haplodiploidy where females are formed from fertilized eggs. So contain both maternal origin, which I’ve got in there kind of block color here and paternal origin genes, which are the kind of the strike B gene left on here. In contrast to this males are formed from unfertilized eggs. And so only contained maternal origin genes when they produce the gametes correspondingly females can produce eggs containing both maternal origin genes and paternal origin genes, or smells can only produce sperm containing paternal origin genes in contrast to this, uh, another group of haplodiploid, uh, organisms to have a system called paternal genome elimination. So initially both males and females are both produced from fertilized eggs. And so initially contained both maternal origin and paternal origin genes, but at some point during development, it might be early on during embryogenesis or even late during spermatogenesis males will eliminate that paternal genome, hence the name. And so subsequently females will produce eggs containing both maternal origin genes and paternal origin genes, but males will only produce sperm containing paternal origin genes So that’s the transmission genetics that unite these organisms, but they often displayed distinct sematic and transmission genetics, but also like the genetics. A lot of these organisms display characteristic, social ecologies, and in particularly often are associated with chronic breeding. So a nice example of this here, we have the date stone or button beetle here, uh, which as you might guess, feasts on date stones. Uh, and so here we have a female will typically find a date stone she’ll borrow into it. She’ll take her gallery and she’ll lay a brood over offspring. And these offspring will then often interbreed before the females disperse again So that’s an example of kind of, there’ll be high levels of kind of brother system eating in the system. So the kind of key questions we’re interested in is firstly, how to these different combinations of transmission and somatic genetics affect haplodiploidy and sexual antagonism. Secondly, how to the different

meeting apologies if some of these species of and finally, how may this lead to conflicts oversexed antagonistic traits. So firstly thinking about the transmission of smattered genetics, we’ve considered this fitness scheme here where we’re considering the invasion of a section antagonistic, a Leo. So we have two scenarios, firstly, a male beneficial female costly scenario, and secondly, a female beneficial male costly scenario. So here T represents the cost and S represents the benefit So we can see here are three possible female genotypes, and I’ll kind of three groups of kind of possible male, uh, Fitnesses and the corresponding genotypes. So from this fitness scheme, we can then compute the invasion conditions and I’ve plotted these out here. So the sorted line corresponds to the invasion boundary for female beneficial Leo and the dotted line for a male beneficial deal. So anything below the below these lines represents when the ilial can invade from rarity. So starting on the left, we can see that for both male and female beneficial ILS invade on exactly the same circumstances. I there’s, um, there’s no particular bias towards either sex on the diploidy and the classic deputy, which I put a little towards this hair to represent that in contrast on the male PG, what we see is there’s a zone of feminization, and this is because for those per turf, under paternal genome animation, there’s a kind of twofold waiting upon the fin. It’s the females because females contribute twice as much to the ancestry of the population as males do. And so we expect subsequently kind of feminization of the genome under any Turkey. Not only do we see this two-fold fitness effect of waiting a for wasting on the fitness of females, where they see the second effect, which is because males are haploid and females are deployed, typically fitness effects are greater in males than females. And this is particularly acute when mutations run average recessive. And so we see as a subset, as a consequence of this, this kind of zone of masculization in the blue, but as mutations were on average dominant, then we see kind of feminization, uh, once more. So we can see that these different somatic genetics alter the, um, also the effects of, um, also how sexual antagonism plays out on the PG systems and the arenas focus ones So we can kind of simplify this somewhat and just plotting out the, uh, female beneficial, uh, boundary We can kind of call this the potential for feminization. So if this is above one, we expect feminization of the genome is below the one we expect masculization of the genome And I’ll follow this, uh, from now on just kind of simplistic. So the second thing we’re going to be thinking about how some of them may see colleges might alter sectional antagonism So we’ve kind of devised this kind of life cycle, the simple life cycle to kind of capture some of the ecology of these organisms. So initially juveniles are born onto a patch Then a proportion of these of the females of this patch will meet with their sibs. This proportion is S um, the rest of the females will then meet with the global pool of males The females will disperse and find there competition for breeding spots before we start the life cycle once more. So what we see as if we now allow for sip mating, then we have plotted out here or in a TOKY male, PG and deployed And so we see that as we increase the amount of sip mating, we get increasing amounts of feminization. So we can see that sip mating promotes feminization of the genome on the all three of the mating system. All three of the genetic systems we’ve looked at here are on a PG and deployed and actually under additivity. We can see that for our own psyche and deployed here kind of actually identical. So why is this? Well, we can split this into kind of three types of effects The first is the increasing amounts of submitting generates local ma local make competition So brothers are disproportionately likely to compete with other brothers for meeting opportunities, and this discounts, the fitness benefits kind of male beneficial ILS to males Secondly, when there’s sit mating, females could also confer fitness effects upon their brothers. So say there’s a female beneficial, ill, not only will it benefit the female that that Aliyah relates within, but also she mates with a brother that brother will benefit from the increased fitness of his mate. And finally, an effect that kind of only has an effect that only falls upon our auto cause. Organisms is the increased in broadness also will promote feminization. So I won’t go into any more detail here, but different types of inbreeding will have kind of slightly different effects They’ll definitely modulate the relatedness and competition amongst males and females, but broadly we can see that generally under a wide range of different kinds of inbreeding systems. It’ll often promote feminization, although to slightly different extents and slightly differently on drug, different genetic systems. And the final thing I want to focus on is how sexual tigers and Whitely to conflicts between different actors. And in particular, I want to think about how it can lead to parent often flicks and in conflict between

parents and their offspring. So we see kind of in lots of other traits that parents and offspring might disagree about the traits that their offspring should express, and also the kind of other traits as well. We see the difference central disagreement between mothers and fathers, because they place different valuations of prompt sons and daughters in their breed. And particularly this, a lot of this work has kind of been relevant to kind of sex allocation. The here, I’m going to think about this in respect to sexual antagonism So kind of slight different type of model this time. Imagine we have kind of this calcium fitness function for males and females, where we have a theme, a fitness optimum here for females and a fitness optimum to make for males. If we assign control of this trait to either offspring, mothers, or fathers, how does this trait evolve and where do we buy? We get to, so here we have the female optimum at one and the male optimum at minus one. And here I’ve assigned control either to the offspring, to the mothers in orange or to the fathers in green. So what we can see is that if on the diploidy with kind of increasing amounts of sip, mating, mothers, offspring, and fathers are all increasingly feminized, have we see two different accents, mothers and fathers, more feminized than there are. And this is very similar to an effect we see again in sex allocation, that was first noted by trippers, which is that parents often often favor kind of more female by sex ratios than, than their offspring And this is because under increasing arts as sip mating, the brood is kind of, um, it’s more successful. Um, if those kinds of increased amounts of feminization, but nonetheless, an individual would rather be fit to themselves, even at some cost to the rest of the brood However, this pattern isn’t identical under are a different genetic systems. Aaron’s a male PG. We see that initially. Um, we see that fathers, uh, uh, favor, very high levels of feminization when there’s no submitting because of course fathers don’t contribute anything to their sons in the long run. Um, so we’ve kind of brought a much more feminized trait value. A mother’s would rather a kind of equal trait value. And then offspring, as we saw above a kind of a more feminized, however, this changes with increasing that analysis of mating, as we see this interesting pattern where actually the extent of conflict changes direction. And finally, we see this pattern and I ran a Turkey where I just asked with diploidy both mothers and fathers favor increasing feminization, their offspring, but to different exp different extents with all of the father’s face, uh, favoring increased amounts of feminization compared to why the mothers or offspring. Okay. So yeah, in summary, we’ve kind of considered three broad kind of things that might affect sectional antagonism Firstly, how different combinations of transmission and sematic genetics in particular, focusing on a Toki versus male PG. Secondly, how different Macio colleges were also section second of them. And finally thinking about who controls the tray again, altering the, the, um, uh, the fate of sections I can escalate. So yeah Uh, if we have any questions, I think I’ve kept the time. Okay. Uh, yes. Thank you, Thomas So Yon will be handling the questions, um, make sure if you have questions, you, you send them to Yon, sorry. Inbar uh, you send them to Inbar uh, on the chalk or on Slack We’ll just wait a little bit longer. Um, Inbar you’re there, right? I am. Yeah. Unfortunately I came in a little bit late, so normally I would have many questions for you. Um, but I had some technical difficulties, but I’m very happy to pass on anybody else’s questions Okay. Well, Thomas, I doesn’t seem like there’s any questions right now, but Sam would like to know how does the feminization interact with selection for dominance? I see. So, um, so do you mean in terms of, so the scenarios are kind of, if I go back a little, do you mean in these scenarios, how this relates

to whether or not there’s activity for dominance? Um, you know, we, it, because you showed an effect of dominance versus recessive newness, is that, so will there actually be selection for dominance or recessive nest in these different scenarios? Um, I’m not sure Yeah. Um, I know there’s been some models where you’ve yeah. You consider things like the evolution of reversals of dominance or, or something, but that’s not something I’ve explicitly considered here, here. We’ve kind of treated dominance as though it’s a parameter, um, rather than something that necessarily evolves itself, but I suppose it’s, yeah, it’s, it’s possible that dominance itself will evolve as a trait and I’m not quite sure how that will necessarily play out under these different systems that we’ve looked at does not send me answer your question. Yeah. Thanks Okay. Well, no, as lost, that’s another question I’ve got one, which is these two different, um, uh, forms, um, men, male PGE versus a rennet Toki, uh, does one evolve from the other or do they evolve independently? And yeah, I think one, I think that’s a kind of semi open question and there are better people than meats placed. I think, to answer how the systems evolve. I think, I think it’s unclear in some cases where the one evolves from the other, um, uh, I think the, the, the, the assumption would be that our ENA Tokyo evolves from male PG, but I don’t necessarily think that there’s, it’s certain that that happens in every case or has called it. That’s always been the direction of travel. Yeah Then do your models say anything about, uh, you know, about how these things are going to evolve? What, what the conditions are would be that would favor one form of haplodiploidy over the other or, you know, when they’re going to evolve? Yeah. So for this we’ve, we’ve treated these as though these are kind of, the systems themselves are fixed and they don’t evolve themselves. It’s almost like that’s the kind of constant, and then we’ve looked at how sexual antagonism is expected to play out, um, kind of given those systems effects, but you’re right. In reality, there’s the potential that these will co-evolve with the systems as well. And maybe that’s a kind of another question to proceed, but at the moment we’ve just treated these as though they are kind of they’re fixed. And I think at least kind of short term, um, kind of more recent evolution, maybe that’s, that’s kind of a reasonable assumption, but you’re right This will cover with things like section I’m thinking that’s something that is worth exploring Thank you. I don’t see any more questions coming in right now. So if anybody has as theirs and they want to get it in the last minute, uh, go ahead. And, uh, while we’re seeing, if another one comes in, I’d like to remind everyone that you can, um, since there’s sometimes a delay time and typing in your question, uh, you can send, uh, Thomas a question via the chat or, uh, potentially the Slack. And he can answer that afterward In addition, for those of you watching along, uh, you can sort of type a question into the chat, but not hit return until the talk is over that way. You can get your question ready and queued up. Um, so, so that you can get it in right after the speaker finishes. Yeah. And hit me up then just email me or hit me on my Twitter and I’ll loop around here for a while as well Yeah. Perfect. Thank you. Uh, Thomas. Okay, thanks. Okay. So in the minute that we have coming up, we’ll switch over to Molly, all Becker, who is going to talk to us about a novel analytical framework to study relationships among genetic and environmental influence on phenotypes. Go for it, Molly. Okay. Can everyone hear me? Okay. Great. All right Well, I’m happy to be here. I want to thank the organizers for putting on such a wonderful conference so far. I’ve really had a nice time watching a lot of these talks, and I’m excited to talk to you about some of the work I’ve been doing as a part of my postdoc with the research coordinated network for evolution and changing seas with Katie Lauder house and Jeff Trussell. So do you graphic variation and Clines and phenotype have long served

as the observational basis for ecological and evolutionary investigations. And generally at a very basic level. What we’re interested in is understanding the genetic and environmental drivers of these climbs. So we will use experimental designs like reciprocal transplants, where we go out to an environment. We transplant individuals across populations, into different environments and measure their phenotype, or we use, uh, an experimental design called common garden where we take individuals out of nature, bring them to the lab and expose them to some shared gradient of conditions and measure phenotype. And then by fitting reaction norms to these phenotypes can get a good understanding of whether the environment alone is driving these Clines and phenotype, whether genotype alone is driving climbs and phenotype, or whether it’s some interaction of genotype and environment. And I want to take a moment to really focus on these genotype and environment interactions, because these are particularly important. Uh, these interactions indicate that each genotype is responding differently to the same environment or that genetic variation in plasticity exists. And in our modern age where climate change urbanization all sorts of different environmental changes are really drastically changing. The environments that organisms inhabit understanding these G by E, uh, interactions can really help us understand how organisms are going to be impacted by environmental change. And what G by E ultimately means is that we cannot predict the phenotype with only one piece of information. We need both. We need information on genotype and environment in order to make an accurate prediction, but G by E or genotype by environment interactions are not the only way that these two features interact, um, to, uh, affect phenotype. We know that co-variants can exist between genotype and environment and formally. This means that Gina typic effects on phenotype are non, randomly associated with environmental effects on phenotype and a little bit more simply, it just means that genotype and environment are both affecting phenotype. So Levins was one of the first ones to document patterns of co-variants in pattern, uh, Indra Sophala along an altitudinal gradient in Puerto Rico and bourbons Keith bourbons followed shortly thereafter, following similar patterns in frogs. Um, and in the eighties, Conover found patterns of co-variants in Atlantic silversides along a latitudinal gradient. And it was really Conover who popularized the ideas of co-variants and really, um, drove the ball down the field for a lot of this work. And I, myself found patterns of co-variants in the frogs that I studied during my dissertation So we have known about co-variants for over 50 years now. Um, at least we have abundant example nature across a variety of taxa, and it may be as important as GBE and understanding how organisms are evolving and what their responses to climate change will be. So why aren’t we all measuring it? Well, probably the chief reason is because we don’t have a way to measure it. Um, this is not for lack of trying. So in 1989, falconer in his now classic book intro to quantitative genetics suggested that in order to measure trait variation or VP, uh, phenotypic variation here, we must measure all these different other areas of trait variation So variation due to environment variation due to genotype and variation due to G by E and then any variation that’s left over, we can call co-variants. Um, unfortunately his approach requires inbred lines and various other controls that are not always tractable in wild systems. So this is largely, um, been unused in, uh, research. So if there’s no way to measure it, um, what were leavens bourbon, Conover and myself calling co-variants? Well, typically researchers would identify visual patterns in reaction norms, across environments in genotypes. And if it conform to certain expectations, they would call it co-variants And these patterns they were looking for, there were two different types. One is a positive covariance. So I’m showing here two reaction norms of. One is green it’s native to environment One in green, G2 is blue, it’s native to environment

two in blue. And so what I’m showing here is that, um, positive covariance is also called co gradient variation. And this just means the influence of genotype and environment are moving in the same direction on phenotype So here I’m showing a positive slope indicating that, um, environment has a positive effect on phenotype. And because G2 is above G one and therefore has a greater effect on phenotype It also has a positive effect on phenotype And if we look at the phenotype of each genotype in its native environment, we do tend to see an exaggeration of phenotypic differences across environments. And this is in contrast to negative covariance, which is called counter gradient variation, or what 11 is called Contra gradient variation. Um, and this is occurs when, um, we have the gene of typic and environmental influences on phenotype that oppose each other So again, I’m showing the same plot, and again, we see a positive influence of environment on phenotype, but instead now genotype genotype type two has a smaller impact on phenotype than genotype one. So that is, uh, now opposing the influence of environment. And again, if we look at the native environments or if the phenotype of each genotype in their native environments, because they are opposing each other, it cancels out any phenotypic changes we might expect to see across environments Um, and so we’ll have been talking about this and publishing about this for a really long time, and there are real consequences of not having a way to quantify it. So first of all, is visual detection has its limits. We know that when we rely on visual cues, we may be missing instances of co-variants. If there are say larger designs, or if GBE is occurring in the data, for example, in this pattern, is there covariance in these, in these reaction norms? Um, hard to say, uh, and second of all, we cannot statistically test whether these patterns we’re detecting are, are real Um, and finally, and perhaps most importantly, we can not answer these bigger questions involving covariance covariance has implications for our understanding of evolution, how organisms will respond to climate change. And right now we don’t even have any idea and how common it is in nature. So I, along with Katie Waterhouse and Jeff Trussell have spent the last couple of years, um, addressing this problem. And in the process, we have built a method to measure co-variants estimate error and perform hypothesis tests. So for the rest of the talk, I’m just going to give ’em an idea for the intuition of the approach, the equation, uh, specifics, how it can be used, and then also some, uh, guidelines on how to design experiments that are best able to detect co-variants So for, uh, thinking about the intuition behind the analysis, let’s imagine the same two genotype, two environment design, where you go out and you collect phenotypic data across an environmental climb, you fit a reaction norm to each genotype and you get these results. So the first question that we would ask to measure co-variants is what are the gene typic effects on phenotype? So the genotypic effects are the average phenotype of each genotype across environments. So for genotype one, that would be approximately here and on the right. I’m just showing a relative measurement of M G one and G2 relative to each other. And for G2, that would be approximately here. The next question would be to ask, what are the environmental effects on phenotype? So this is sort of the flip side of the gene and typic effects. So this is what is the average phenotype of each environment across genotypes. So the average phenotype for environment one is approximately here and the average environment for, um, excuse me, the average phenotype for environment two is approximately here and what I’d like to draw your attention to is this plot on the right, that shows now that we have matched each genotype to its native environmental average, and the covariance is the relationship between those points. So this equation, uh, allows us to measure that joint variability between gene a typic and environmental effects on phenotype. And so in this equation, this term here, this Y bar I minus Y bar, this describes the Gina typic effects and this Y bar J minus Y bar, this describes the environmental effects on phenotype. I’d like to take a moment and discuss this. I variable. So this is what we’re calling the indicator variable. And we put this in, because if you were to say,

do a four loop in R and just chug these data through a four loop, it would give you every combination of genotype with both environments And that’s as indicated here, not what we’re interested in. We’re only interested in those measurements when they are correctly matched with their, uh, proper native environment And so this indicator variable, um, allows us to make sure that only those that are correctly matched are counted towards the covariance estimation. So if it’s incorrectly matched, it’ll be multiplied by zero, which cancels it out for the, um, for the estimation And this also brings up a really important point is that we need to know which genotypes are native to which environments in order for this analysis to be successful. Um, unfortunately in the literature, um, oftentimes this information is absent and so it makes measuring co-variants a little difficult. Uh, we wanted to standardize our measure of covariance because this would allow us to compare estimates across species or across study systems. We tried correlation which standardizes by the standard deviation of genotypic and environmental data, but that didn’t preserve the effects of the relationship because co-variants, uh, is non-independent So to account for this non-independence, we just standardized by the standard deviation of one or the other, depending on which one was bigger. We use bootstrapping to generate 95% confidence intervals around our sample estimates, by sampling phenotype, with replacement, from each genotype and environment group And then we use permutation for hypothesis testing in which we re sample phenotype across genotype and environment, uh, without replacement to generate a null distribution. And so in the end, this gives us a way to generate sample estimates for co-variants or as we call Cove GE that range between negative one and one And as a reminder, anything less than zero is counter gradient variation, and it gets stronger. The pattern gets stronger. The further from zero, you go and above zeros, co gradient variation Now a major reason behind choosing this approach is because it allows us to measure co-variants based on reaction norms produced from common garden and reciprocal transplant designs Um, so we simulated data reflecting a variety of scenarios across these two experimental designs. And so here, I’m showing just a classic reciprocal design with four genotypes in four environments, and then we could vary the amount of GBE seen in those reaction norms. And then also I want to change this example of the fruit fly with our common garden, just to show a paired common garden design and which you collect different genotypes that share the same environment. So you collect multiple cold environment populations and multiple warm environment populations. And then this would be, might be what their reaction norms look like. And when you apply our approach, what you see as each of these, I have selected specifically because they are all very good examples of strong co gradient variation So each of these is nearing one positive one, and they are all statistically significant No, I cherry pick those three, uh, just to give an idea of what the analysis would do and what the simulations look like. But in reality, I actually simulated over 30,000 different combinations of experimental designs and effect sizes to really identify which designs best detect co-variants in nature So I’m going to show you two heat maps, um, where on each tile I’m gonna show two numbers. One, the top number is the total sample size. So the number of genotypes, the number of environments and the number of samples. And then, then the, um, the number below it is this statistical power, which we define as one minus the false negative rate. And so to show the results from the full reciprocal transplant, if we are trying to design our experiments so that we have 80% power to detect moderate, so not too strong, but, you know, moderate levels of co-variants, what we need is 256 samples And there’s some wiggle room on whether or not those samples are collected with gene, um, whether you heavily sample genotype or, uh, replication within a genotype and for paired common garden, it’s the same result where 256 samples are needed for adequate power. And I’d like to also draw your attention to what’s going on at these two samples size or these two genotype scenarios, because these were really, um, common and early studies that identified co-variants. Because, like I said before, when you rely on visual cues,

you need a simple design to detect those visual cues. But what our analysis actually found is that unfortunately those designs are underpowered So really what we want to know is does this approach work on empirical data? So I volunteered my data that I, um, found patterns that are consistent with counter gradient variation and Jeff, who is a co-author on this paper also Okay. Thank you, four minutes. Okay. Got it Thank you. Um, Jeff, also a coauthor on this work volunteered some of his data in which he also identified co gradient variation in a population of Marine snails that, uh, are adapted to different flow regimes. And so when we apply our metric, we do find that with a high list scenario, they are, um, they do show significant patterns of counter gradient variation. And, uh, with Jeffs, we also see actually stronger patterns of Coker counter gradient variation, but I would like to draw attention to the fact that his was not significant, and this is likely an artifact of the, um, fact that he only used two genotypes. So it’s less likely that the pattern itself is not real, but just the fact that it’s underpowered to truly detect these differences. So hopefully I have convinced you that covariance is really important and we need to be studying it more, but research has been hampered for good reason We didn’t have a way to measure it. Uh, we have now built a quantitative method to measure it in nature using, um, data that is available from GB or excuse me, um, reciprocal transplant and common garden designs. And we are really hoping that now that we have the tools, we can really, um, investigate this further across a variety of TEXA and systems. And so I’d really like to acknowledge my lab mates who helped out so much the research coordinated network for evolution changing seas and except for funding that RCN and Northeastern for hosting me as a post-doc And then please do contact me for any inquiries on code, um, to implement this metric. Thank you very much. Awesome. Thanks so much, Molly Um, there was a question from Andrew Henry, does the method ignore variance within the populations or samples? And does that matter it, are you talking about replicate, like the common garden design where you have replicate samples? No, I mean, it takes it, it’s looking at only the it’s fantastic by the way. Um, I can, I can’t wait to use it, um, but it looks only at the means. Right. And I’m just wondering if, if the, the precision of the estimates of the individual means mean anything from an evolutionary or statistical perspective, because if they do, they’re not considered as a part of this approach. Yeah. Um, we did not look at that explicitly. I can say that if you, if you are interested in that variation, the more sample that you collect, the better precision you have in the estimate itself, which, you know, um, but we did not look to see whether the variance itself affects the estimate, because as you said, we’re just looking at the means. Um, and then also, how does canalization affect the estimates? Canalization um, that’s a good question that would, that should affect well, it’ll affect everything Um, that’s essentially a similar question, I believe, cause that should be reduced variation Yeah. So I guess I would just repeat the same The same thing is we did not look explicitly at variation. That might be something I’ll talk to Katie and Jeff about, um, after this But, uh, we would assume that less variation would lead to a more accurate estimation of the mean, um, did I say that right? More accurate estimation of the main, um, but I’m not sure evolutionarily how that would affect things Okay, Molly, thank you very much for your talk. Uh, we need to move on to the next talk Uh, the next talk is a prerecorded talk, uh, by Julio mano Morimoto that Mike Whitlock will play for us. Um, there won’t be time for questions at the end of this one because the prerecorded talk takes almost the entire timeframe, but, uh, hopefully Julio can respond,

um, well on the chat, but also particularly on the Slack channel. Okay. So let’s go ahead and play the hi everyone. I’m Dr. Giuliana Morimoto thank you for coming happy new year Uh, I’m here to talk to you about applying the theory of justice to academia. Can we build justice in fair academic community? And this is a conceptual model, uh, that I want to present to you today. Please ask the tough questions. If not at the end, you can send me an email. I think I want to encourage this to be a collaborative effort moving forward. So before we start just some clarification of scope, this is a sensitive topic. And as such, I would like you to dissociate my delivery from the idea. Of course, I put a lot of effort into co you know, creating this topic, but I might not have delivered in the way that you please, please you. So please the associated delivery from the idea itself, and let’s discuss the idea moving forward. I have no intentions to advocate benefits to a particular group, as we shall see late in the end. And instead, what I want to do is to stimulate ideas in generate debate in strife, in the future for true equality of opportunity to swallow anything, the end, everything I present here with disregard after discussion after discussion, I’m fine with it. This is science, right? Uh, so let’s get into it. The basic takeaway message of this talk is that we should incorporate individuals as ecological contests during the academic selection processes. And this is all needed for our location of grants fellowships and, and you know, all their distributive income that we say, um, in, in academic communities and in doing so, we will generate true equality of opportunities to grow. And I hope to convince you by the end of the tour today. So if you’re, we all know that academia is unjust and he has inequalities, if you don’t believe in that, then you know, it’s not going to be one or two slides that will convince you. So I will assume that everyone here understands that academia, it’s not fair in a fair community, and it doesn’t promote equality of opportunities to all, there are several papers about it So this is the assumption that I’m going forward with. And I asked the question, can academia and academic institution strive for true quality opportunities? And if so, how? So there are two ways basically that w widespread that we attempted in a sense to generate a true equality of opportunities. The first one is the discretization of career path. Okay? And this is a standard across the world, but you have, you finished your PhD. You have ears, post PhD, where you are eligible for fellowships, let’s say, okay, this is just an example of a fellowship in the UK where I am based right now, these landmarks change, but they all relative to the year post PhD. The problem with that is that PhDs vary widely across countries, right? Just an example, PhD in the U S takes roughly almost twice as long as the PhD in the UK. And therefore these differences in, in how PhD works in each country, differentiate the opportunities that each individual have in each of these particular countries ending the system as a whole. So it’s difficult to create this fixed landmark of post PhD, because in a sense, it doesn’t take into account the variation between countries the second and perhaps most innovative, and it hasn’t been implemented everywhere, but it’s truly innovative is the lottery system in New Zealand, where you have a pool of applicants, you have an initial screen and you have the lottery award and the lottery is supposed to be a sarcastic event. And therefore is supposed to introduce fairness in the system. Now, the problem with that is that if the pool of applicants is skewed, then a fair lottery system will still reproduce the skewed distribution of the pool of applicants, unless they initial screen somehow mitigate that effect. Right? I particularly don’t know. I couldn’t find this information, whether the initial screening, uh, you know, mitigate any skewness, but the point here is that in fairness can emerge even from a fair process, what I want to convince you, or at least present you, the idea here in this talk is a third option where we could implement concepts of the theory of justice, uh, to generate, to equality of opportunity, to all let’s dive into it a little bit more So the theory of justice is based on the theory of justice, the book by John Ross, which, who was a philosopher, an American philosopher, and he wrote a book called the theory of justice, where he equates justice as fairness, and amongst all these things, all the things that he wrote, three main concepts are important, and they are called sips off the original position, the veil of ignorance and the decisions

made in the original positions under the veil of ignorance Right? So basically put original position would be imagined that we have souls before you get to the body in your plant, in the planet, right? So the souls are all equal They are aware that there are there exists inequalities in the planet, but they don’t know where they will be in days, spectrum inequalities, right? So this is the veil of ignorance. So they are ignorant to where they’re going to lie. And basically these souls, they have to come up with rules that dictate how this society will work. And what genre propose is that individuals in this original position, underdeveloped ignorance will make decisions and deciding how society is going to work You know, just men are simply because they don’t know where they’re going to end up with So they, there is no favoritism in that respect And that’s the veil of ignorance. And this idea that justice is blind to inequalities is the whole purpose. Why we represent justice, I can with a blindfold, right? So this is for example, uh, in Greek mythology, we have a representation of justice. That is blindfolded problem is that in reality, the original position is an attainable and the veil of ignorance, in fact, doesn’t exist. So in a sense, we know that justice is not blindfolded. Justice is not blind. Then how can we meet to gate that, how can we overcome the fact that we cannot be in a position where we don’t know where we’re going to end up in the society and therefore make just decisions one way that I think we should move should progress moving forward is that we need to identify the sources of inequality. And here, perhaps an example would, you know, we’ll illustrate a point better here. I took the percentage of 25 to 34 year old with their shared education by level Uh, and this was extracted from the education at gloss published in 2020. And I picked two countries that represent two opposites, basically is Levenia and Brazil, which is the country that I come from. So in is Levine at 37% of this cohort, have their share your location, 7% of which at the PhD level in Brazil, on the other hand, 21% have tertiary education only, and less than 1% at a PhD level. Now that means in itself that PhD amongst this cohort, it’s more than seven times higher in his Lavinia compared to Brazil. And this in itself translates into a whole other ecological spectrum of things that favor individuals to get a PhD in Slovenia. And in mine, I’m just using Slovenia as an example, right? I’m not, I have nothing against Slovenia or nothing in favor of Brazil, uh, is just for the purpose of illustration. Now, if you go to, for instances of India, social attitudes, source for the education might be different in those in Brazil, social expectations opportunities to get a PhD degree and so on and so forth might be different. So for the sake of this example, would it be fair to assume that candidates obtaining a PhD Slovenia or in Brazil have undergone the same challenges? Now I know the PhD is supposed to have a standard, but we know as we talked before that during the PhD itself, there are very, very Instructure and therefore there might be each candidate may have different opportunities during that PhD. So is it fair to assume that they have undergone the same challenges? Now you might be thinking, okay, this is just a cherry picked example. Let’s look within Finland, okay Feeling is even a better country in that education and gloss graph. It has a higher percentage of that cohort with tertiary education hired in day. You ever a J in fact, so let’s look at Izzy. Now we’ll extract from a paper published by Helion collaborators in 2019, which is a longitudinal study, um, with the cohort from 64 66. And they assess whether parental level academic education affected the chances of, of this cohort to progress in their academic achievements and extract it from the paper itself. Because I think illustrates perfectly the point here is that finished professors born in his cohort, uh, among those that have parents that lack post-secondary education about wanting ten one on one in 110 became professors while the same number was one in 140 among master degree holders with at least one university educator parent. This means that there is a difference of two point 75 times the likelihood of the, the chances of the odds in a sense, in this case of you attaining professorship, if one of your parents had a university educated in Finland, which

is a developed country in that respect, the authors, interestingly went further into their assessment and they claim that for instance, individuals of non-academic backgrounds, who nevertheless became professors were more likely than others to have been advantaged in other ways. So in a sense, the academia in this case finish academia community is even more socially selected. And then the results may suggest thus, there’s no reason why academia everywhere is supposed to be this way, right? Socially selected and therefore academia as a whole is an environment that lacks true equality of opportunities. And why is that? Why do we lock to equality of opportunities? Why is academia socially selected? And here, I want to explain to you the cost of exclusive and inclusive opportunities. And if you have a district knowledge, I use, I created food, the purpose of this stock, please feel free to make suggestions on how I can change it before we published this. So let’s go into exclusive opportunities. So exclusive opportunities, let’s say you have a distribution of a trait It has its meaning the population we select the outstanding bays on how far they are from the mean, and we claim that those are the best phase of merit, right? Exactly. All of these is the Olympic games We don’t care who is the tallest or heaviest All we care about is who runs the fastest, who jumps the highest and so on. So this, in this exclusive opportunity context, outstanding is relative to their population distribution, but you notice diversity and focus on the utilitarian outcome. That’s a strength in the Olympic games, a unit of effort it’s equal for each individual, you know, to increase from in this distribution, uh, but impulses this exclusive opportunity and poses a lot more effort to the least advantage so they can make the cut to the outstanding. So basically if you are on the opposite side of the curve from the outstanding, you have to put a lot more effort to get there. And if you don’t, the assumption is that you did you fail because you lack merit. Now the inclusive opportunity does not, it still has the population distribution, but assumes that within the distribution, you have types, okay. And each of those types have their own mean. And then you applied this outstanding, not to the population as a whole, but for each type within the population This could be for instance, socioeconomic background or the parents or parent level education, as we saw in Finland, in Finland and by doing so, we select the best in the outstanding, relative to the opportunities available to those types. Okay. In this case, outstand these relative to the type of opportunities available for that particular cohort promotes diversity and a unit of effort is equal across individuals, but does not impose a hugely more costly effort to the least advantage in this whole population is in the population as a whole. Okay. So in the sense inclusive opportunities is, is what we want to generate diversity. Now, what do we need to go to inclusive opportunities? The difference between exclusive and inclusive opportunities to their collection of data, data that makes us understand the sources of inequality. So it is impossible, no matter how much we claim, it might at least in my opinion, for academia and academic institutions to adopt these inclusive opportunities, if they don’t have the information about the sources of the inequality. And here, for example, imagine that in the exclusive opportunity to scenario, you have a blind approach, you have an application, you have to CVS, you don’t know where to come from. You have no information apart from what he’s written in that particular application, this assumes that both candidates had the same opportunities and differences are based solely on merit Now for an inclusive opportunity, it’s a data-driven approach, right? You have the CVS, but you also have the ecological context from where those CVS were generated. And this not only provides more context to the CV itself, but I can knowledge, is there deep opportunities and differences in those CVS may not have been solely based on merit. Now, how can we do that in reality? And this is where I think the recent advances in technology can allow us, you know, to, to push this forward. And this is particularly true for big data and machine learning. So basically what we could, and this is a conceptual model, couldn’t test it because the COVID is to generate a database of global ecological information. Let’s say we take percentage of PhDs, tertiary education As I showed at India education at glance inequality scores and so on. And we applied to a model in this case, a machine learning algorithm that will standardize and provide a standard of each of these candidates based on the, this database will be collage green information, uh, it’s core for each of those relative to what is expected from the population where

it is CVS or those candidates come from. Once we do that, then we have a true equality of opportunities, obviously comparison between the two CDs together, repeat review, which I don’t think you will, and, you know, go away anytime. So, and I don’t necessarily think you should go away, but we know it’s bias. If we combine a fair ecological score with the peer review score on the merits of the project itself, then we have a final marking, which is, if it’s not fair, at least it’s fair. Then what it is today. Now I have a serious cause that in machine learning, I know that machine learning algorithms today, they are not fair. They are unfair. We have biases, but I am assuming, and perhaps I’m too optimistic that in the future, we can overcome those biases, whereas unconscious bias, for instance, in human behavior, we, I don’t think we can do the, you know, no matter how best, how good is our effort. So I know the caveat here, the machine learns are biased. I just want to make the point that I believe this can be me to get it in the future. Now are some of the potential immediate criticism of this. And one of these is that, okay, so will selection processes become too personalized, basically. In other words, where do we draw the line? How much data we need to collect, right? And for this, I suggest us to take the pragmatic approach Let’s consider factors known through scientific knowledge to affect academic achievement In potential. For instance, we had the example in Finland, in Finland, um, Parenteau education is a factor that affect potential for career achievement. So these should be a variable that we could collect from and to make things fair. And of course, we need to research more, right? Uh, um, about what socioeconomic and other cultural factors that can affect academic achievement. So we truly understand all the surroundings, the ecological context of the selection process. Now there are a lot more of ecological factors that influence, uh, uh, academic achievement. And for that, I suggest that the literature is off Urie Bronfenbrenner, and more recently, Jay Belsky and collaborators on the ecology of human development in, you know, these, this broader field of ecological, uh, effects on human potential. Now, I don’t have time to dive into each specifics here, but I believe that exampling feeling, um, illustrate that very well. Now, just to reiterate this, the purpose of this talk was to convince you that we should incorporate individual’s ecological context during the academic selection processes, thereby generating true, rare quality of opportunities to, well, now, what can we do? This may seem farfetched now, all the models that are presented here, what can we do then in the immediate, like right now, uh, to, to, to change in generate a true equality of opportunities. First one is to create a positive motto and test what I just presented to you. And we are in the personal, but you know, more, more immediate We can be empathetic and be kind, and we can acknowledge that others particularly from developing regions may not have had the same opportunities that we had in, in other countries I acknowledge that we cannot, even with the best of our intentions objectively assess someone’s potential from economic backgrounds and cultures that are very different than ours. And therefore we should seek true their, their city of inputs from different socioeconomic backgrounds, cultures, countries. And so on One thing that we are trying to do here at the university of Aberdeen is to incorporate an ecological statement where to come to that end, the supervisors explained the circumstances around the kinds of teacher. So providing an ecological context for why that candidate it’s suitable for the position that they are applying for. Not only relying on the CV and with that, I would like to thank you. I know it was run, but it was squeak, but I had limited time. Thank you very much. Again, ask the hard questions, uh, play the devil’s advocate Uh, I think this is a collaborative effort that we all responsible for. True equality of opportunity is the responsibility of all of us, uh, and please get in touch and stay safe. And I see you soon. Thank you. Bye bye Thank you very much. Uh, Julio mano, um, we don’t have time for questions. Uh, we have to move on right to the next speaker, but remember, you can send questions to Julio nano on the Slack channel, but also he provided his email on that last slide, which you can revisit, uh, on the YouTube feed if you need to. Okay. So next, we’re going to go to, uh, Rafael Mora. Who’s going to talk to us about assortative mating in space and time patterns and biases. Rafale you’re muted. Can you hear me now? We can try again. Yeah. You need to

share again, your in a city, the presentation that’s good. Oh, nice. Thank you very much for that. But Trinity is really happy to hear My name is FAO and I will represent a lecture So take meeting space and time. So I started meeting image scribe as a deviation from radar maintain that can image for our attendance and making preference for phenotype similar or similar partners. So we can measure the sorts of meeting using your correlation, uh, of, uh, meeting pairs or forgetting traits of meeting types. When score relations positive, we call it as a positive assaulted meeting with it’s negative. We call it a negative associate mates. It’s very important to identify partners of associate mating nature, because they can provide, uh, pressures, informations about how selection is affecting, affecting that. Given trait. For example, if there are sorted mates as positive, uh, maybe this trait is on the roots selection in which the extreme phenotypes are favored by selection. So, um, sympatric speciation is likely to occur. However, when the negative associate meeting, uh, some license to be favored, the average type. So those speciation is expected. That’s what’s it meeting thing be classified also depends on the traits and their selection. For example, a young Getchell plus five, the types of assaultive plating as age behavior and cold conditions, echo time technology, site, structure, and visual. However, they separate what we call a social team meeting direction on two different types. They call size as the measures of the overhead size, for example, where he learned And it’s true, true as a measure of a body part, for example, the kind of password on phase five-year-olds I mean site, however, here I will rephrase it. Say sorted meeting as in type of measure of a bird sites, because they are frequently correlated and the literature you read, you will readily see some kind of extreme deflecting So research is used to practice to investigate assortative mating nature, but they may have some problems that are in SKUs here. It was one is if you want to study, let’s say mating two populations are two different, different breeding terrorists. You may measure to size of individuals in Coppola, and you can find Ciocca relations, for example, two kinds of negative assortative mating. However, if you observe the populations, you’ll see that the average size of individuals is different So when you put this data, what happens is a posted by his investment of assaultive mate, we can call this Simpson’s paradox or in the literature for summative meeting scale, you have choice effect here. The universities we found almost 46% of the studies before was kind of prep. A second practice is children, broad conclusions for the whole population, or even the whole species based on a single measure of assortative mating is can be very problematic, especially if assortative mating, varying, spacing type. So in this example, you can see that there is a possible source It made some money on other moments or any other population, another population you can obtain a different partner. We found, uh, almost 30% of the studies in this database before it’s kind of practice from broad conclusions using single measures. So our core, what was revealed here, sorry, was to use, uh, uh, correlations between, uh, meeting pals, um, should form a systematic review in a mirror analysis to evaluate the positive bias of, especially in temper all the data or the size So team meeting patterns, how consistent patterns of space in time in general partners of asserting mating across and within the Tasha of head

loss when the pool of different size are excluded So we follow the treatment protocol or not quite the details, but are present here. Basically we chose chokey worlds, which is size assaultive insights assaulted, and those three data based So initially a more than 50,000 results in our last survey was forum 20 April, 2018 Uh, in total, we obtained the largest data set about the suicide assaultive meeting published to the date. So, uh, 1,827 effect size from 457 studies using 321 species. So we have been fine, uh, apartment, uh, truck out the decades represent new studies published in each lactate and the black dots and the lines represent the number of publications. So as you can see, uh, so I think mating is a ongoing, uh, field of study, which is growing in each decade when we’d stand or the types of putting practice that we classified as partial cooling, temperature cooling, or no cooling, and even both winning practice. We saw a tendency here that mostly studies published, um, partners without using the kind of practice. However, the number of studies using temporal pulling practice has increased. So basically we screen each study, uh, looking for, uh, business correlation, coefficients, and when data was presented in graphs, we extract the data that they agreed, the data points using the web part, digital software in complicated are one population coefficient or transport, or the statistics to the art coalition graph is infusing the metering company. So we use all this, uh, there’s measures of correlation, coefficient to own, to transform in efficiency, which we used as effect, size and efficiency. We used this kind of statistics because efficiency presents global distribution of residuals So we also added the weights for each year effect size, which was the investor variance, uh, that is most populated using the number of samples. So for each way, we formed a several multilevel meta-analyses using the metaphor packaged in included as Reno variables, they effect size ID, they study it, they species ID and affiliation, ledger relationships We also added as moderator in the analysis, the pudding practice. Yes. And though, if there is any kind of public practices or no, in what we also disentangle the effects of different kinds of practice, special theme from the interaction in compared to openly practice, we measure it. They’re generally using Sigma square and I square, which measured the presentation, but total variance attributed to the random variables. So we calculated the obtaining, the foolish natural relationships using the opening tree of life dataset. And we structured the methods of a tree and pruning the tree, taking it a fusion, natural relationships for the species, including the nights. We also classify the animals in different sub Okay. Uh, here you can see a full watch with all the results separated by the types of public practice. And we identify the posting publication bias, but we investigated different source of the bias, um, such as the emotionality of the trade, which could be moved three B when the dimension or even technician others, that type of study. If the study was before at the laboratory or at the field in the year of publication, however, none of these, uh, moderators explaining this, uh, positive bytes So we assume that this may represent a true, a symmetry in assortative mating uncertainty nature. When we investigated the effect of practice, we found slightly positive bias in studies that before practice compared to an open practice, and these results remains consistent after separating the types of public practice as a spatial and temporal sector, the higher average effect size of to the practice

And there was no interaction. So there is a positive bias, uh, pulling in the year They compare the iteration of always studies included in our meta-analysis and models We found our moderate-risk Nat. However, when we exploded the studies that before any kind of practice, we filed a significantly in West, particularly less lowered iteration. However, the impact of these new data set on the average effect size was not. So in January we filed a moderately positive, uh, every team meeting. April’s always, uh, we also identify using the sun bias dataset, a partner of, uh, of average effect size for each to be field. So my Malia was the only one that exhibit the matings while the others, the esoteric meeting was positive in that, uh, exhibit the highest plan, the iterative nature. This model was the lowest of our models that we presented here. So our average FODMAP buttons reported in Sonata medicine, very consistent eight gross systems. We formed the same type of analysis using two different kinds of data set that presented Spicer and paper replicates So we extracted the residuals of these better analysis to obtain their non bias, uh, response variable that we include the interpretability test. So we use this random variable, the special temporary replicates, and truly paying the reputability measure points, kind of putting off, uh, replicated. So here we have the resource April’s populations, the average part that exceed the low repeatability, uh, also exhibit repeatability, but most Starship such as Malia PBA and Krista said, there’s not, did not exceed any reputability between, uh, eight cross populations. And between reading period, we found similar results because the average part that will also exceed a low repeatability and obvious an example that presented similar results, but a failure can be set out in crystals There’s not exhibit, uh, the kind of pockets and conclusion assortative mating can vary in space and time. So, uh, so sorts of meeting estimations should be careful in our eyes if I’m drawing conclusions about their potential mechanisms, focal segments, indeed providing positive information about meeting relationship, but they must be interpreted exponential as punctual evaluation. So in a teacher extends his conclusions or a different populations in grades. The parrot can be misleading, especially temperature variation. Assortative mating is more common with the prejudice stock in the commercial conditions may have important roles in drive sets variation. So if the same species we can farm operation, choke out diagnostic space, we should look for some ecological completions that can affect such consistent story variation. This is especially important because there is a need for a new hypothesis about the underlying mechanism of spatial temporal variation in associate mating patterns to the data. When you only have hypothesis that predict the occurrence are not assertive, but knowing of hypothesis available to explain how our sorting made sure it varied between populations or breeding parents. So I wanna, uh, I want attention my cooperators without hope this study will not be possible and all funding agents. So thank you very much and I’m available for questions. Thank you. Great We have a few minutes for questions. Uh I’ll I’ll start with a question Um, so, uh, Raphael, that that’s really exciting and it, it, it sent me, uh, immediately searching off Recology ladders where it’s not quite out yet. So I’m, I’m anxiously waiting for its for its emergence. Um, the question I had was that, um, David Green re recently analyzed this pooling problem using and urine assortative mating database in American naturalist, and found that there was actually a rather large effect of this, uh, grouping this pooling

problem. And whereas I see from your analysis, it seems that it’s much smaller. So what, what what’s your take on that are, and you’re in special in this regard or is the analysis different or something, our analysis similar, but I also sit at a work of greens paper and, but he used the data set for anterior and here we are using a much larger database So as you can see this results here, sorry, let me come back a little bit for reputability importance. You can see our high variation between so subfield. So if you look for a CEDIA here, there is no relatability, uh, across populations and break between breathing barrier. So I would expect that pulling data would be more problematic for this particular study field compared show, for example, examples that accept a higher repeatability, increased populations in between breathing berries So the work of green is really nice and inspiring It helped me a lot, but his data set is limited short failure. Um, while we wait for more questions that I have one too, um, were you surprised to see, you know, basically a sort of meeting be very common across all your tax? I think you said, except for mammals Um, and if so, you know, what, what do you think that says about the sampling bias and studies? No, if I understood the question, you, you are trying to look for some explanation for why my Malia has a rental car. Yeah. Why is mammalia so different from the other taxa? And did you find that surprising? What do you think that says about sort of the sampling bias? Well, I think we can, we can look for explanations here, but I think in the difference of my data site compared to others is that I included humans here so young again and collaborators, and also another matter analysis prohibition, 2019 by collaborators excluded almost sapians from the analysis. So I think one of the reasons is because as a species, our species here, but other important questions is how, um, memos interpreting the significance of sites in may choice for insects, for example, size health, several implications for how they, they compete. They choose mate, but this is not necessarily true for there’s other conditions that can be more important than size into memos. So I think that could be one explanation, but my, the focus of my study was not sure explored this assymetry between Sushila, but just to report it, someone bias estimation for each dash, they don’t answering your question. Yeah, yeah, no, I think it’s, it’s, uh, I was surprised by that. And then I was thinking whether I should be surprised. So, um, but you’re saying that a large, large part of that is because you included humans as mammals or at least some part of it. Yeah. Which is also interesting that one, one tax on can have that effect Um, I think we have one more minute for questions, really curious to explore why, for example, some sort of filler exceed higher values of assortative mating while others do not. For example, Jenny also had a paper in which he showed that, um, the strength of assortative mating does not explain the number of species that evolved that appear in EG. It’s a feel, very think his approach. He has, some could have some improvements if you look for some variability in size sorted mate, which was not his goal. So I think we can discover a lot of new things here. If we look as far as we look chose, assortative mating law, not as a start King pattern, but something that can vary. Okay. Thank you very much, Rafael. Um, and we should move on to the next speaker. Uh, so next we’re going to hear from, uh, Kyle Cole blends, who will talk to us about ecological boundaries, determined evolutionary trajectories on adaptive landscapes. Go forward,

Kyle. All right. Um, thank you everyone for tuning in, I’m really excited to talk to you. I’m excited to talk to you all today about, um, some work that we’ve been doing, looking at ecological boundaries and how they might constrain, um, evolutionary trajectories of populations. So extinction is a powerful evolutionary force and perhaps one, a pretty spectacular example of this has been the effects of mass extinctions. Um, so for mass extinctions over its history have radically altered, uh, and shaped and reshaped, uh, the diversity of life on earth over time. And perhaps one of the, uh, most well-known examples of this is the extinction of, uh, non AP and dinosaurs that the end Cretaceous, uh, which might be, uh, partially the reason we’re all here as it’s thought that, um, mammals have this large adaptive radiation following that extinction Um, but ecologists that evolutionary biologists have also identified the importance of evolution on much smaller scales also. So for example, in evolutionary biology, there’s this idea of Darwinian extinction, um, also known as evolutionary suicide. And, uh, this is the idea that natural selection itself, uh, can actually cause species to evolve to their own extinctions. And so we walk through an example here, uh, using the example of a predator prey interaction. We have a graph here where we have a predator fitness on the Z axis, and then this plane with the predator mean attack rate and predator populations. And so we can start say with our predator, the attack rate here, and there’s selective pressure for the predator to increase its mean attack rate. However, as the predator does that, um, this fitness surface drops. So we see an increase in the mean attack rate of the predator. Um, but we also see the in doing so the predator actually evolves to a smaller population size, and this is because it’s a more efficient predator. So it decreases the prey density, which decreases the predators density itself. And we can let this continue to play out. And what we eventually see is that the predators mean attack rate evolves to a point in which the predator population actually goes extinct. Um, and this suggest that there must be some sorts of constraints acting that, um, keep people, uh, these populations from evolving to go extinct or that, uh, we just don’t see these, um, in nature because all of the species that have undergone this process don’t exist anymore. Now, ecologists have also, um, come up with a similar idea about how, um, traits and species might be constraint. And, uh, this is an idea of stability selection and stability selection says that well, the way communities are and the way see them in the world is because they’ve undergone a process where unstable configurations of species have just gone extinct. And so what, what our end result is is that we see mostly these stable configurations of species. And so an example of this comes from an idea of a persistence domain. There was introduced by, um, autoette owl in 2007. And so what they did was, uh, to look at some models of try trophic interactions and look at body sizes and allometric models. Um, so the parameters in these models were based on, uh, body mass ratios of Faisal to intermediate species and the, uh, intermediate species to the top predator And so we can look at this plane and they found this red area in which, uh, they call the persistence domain in which all of the species in these tri trophic food chains were able, were able to persist. And so, uh, what they did was they also took, um, data from actual empirical food webs and the associated body sizes. And they plotted these onto this graph is these black dots. And lo and behold, the vast majority of these data points landed in this persistent study. And so this suggests ecologically that there also the extinction might also, uh, create these constraints on traits that species are able to, um, exhibit And so we were thinking about this idea in terms of, well, if there are these sorts of constraints, how might this play out, um, with contemporary eco evolutionary dynamics and sort of think about this. Um, we went directly to, um, this classic evolutionary

metaphor of the adaptive landscape. And so here we have, um, a landscape where we imagine a population with two traits, um, and there’s some, uh, fitness measure. And we can say start our population at this white point. Well, traditional evolutionary theory tells us that that population should evolve towards higher fitness and that should follow the steepest selection. And so with this idea, we would expect that, um, our population is just going to evolve right towards that higher fitness. However, if we think about this same situation, but to say that there are these ecological boundaries that are likely to lead to extinction, we might get a very different picture where for example, if this population were to evolve directly along the steepest selection ingredient, they would all just go extinct. And so this suggests that there might be some other processes might be really important in determining how species are able to actually reach areas of high fitness and that without thinking about these extinction boundaries, uh, w we might species might end up taking very different trajectories than what we might expect. And so this seems like an interesting idea, at least hypothetically, um, it seems to make intuitive sense. Um, but we, um, wanted to see whether we could just do have a proof of concept model of whether this, we might actually see this. And so what we did was look at a classic predator prey model, uh, specifically the Rosenzweig MacArthur, uh, predator prey model. And so let’s walk through this real quick. Um, so our here is our resource, uh, and it’s growth is logistic growth. And you might look at this equation and say that doesn’t look like logistic growth. Um, but what we’ve done is we’ve just made births and deaths explicit. Uh, so B being first D being deaths in these Q terms are the interest specific density dependence. Uh, and this does have the exact same dynamics as the sort of traditional logistic growth equation. All right. So our prey also suffer mortality from being eaten by sea, our consumer, uh, and in the Rosen’s Mike MacArthur model, uh, this happens, um, is described by a type two functionary response. So a is the, uh, predators attack rate age is the handling time. And so we also have a, uh, an equation for our consumer, um, same type to functional response describing the predators feeding rate, uh, with the attack rates in handling times. Now we have this conversion efficiency turning prey into predators, um, and we have some density, independent mortality rate, and we chose this predator prey model for specific reasons. Well, a couple of them, uh, and one is that there are these well-known boundaries that do exist, uh, for this predator prey model. Uh, and they depend on the predators handling times and attack rates. So this first boundary here is, uh, is, um, a feasibility boundary. And what that means is that if, uh, the predators attack rate in handling time, fall below this, then it can’t eat enough to sustain a population And so below this line, the consumer will go extinct. Now, there’s also another boundary, this upper boundary, which is a stability boundary. And so for any attack rate handling time combinations that fall above this line, the system will exhibit predator prey cycles. And, uh, although the populations might be able to persist when they’re cycling, they’re also likely to run into these areas with either really low consumer or really low resource abundances, uh, where the populations are likely to be very susceptible to extinction. And lastly, another reason to choose is we know where the highest predator fitness is, and that’s when the predator has low handling times in high attack. Um, so that’s when we’ll be eating the most and having the highest. Alright, so how did we actually model evolution? I won’t go into a lot of detail about this, but the way in which we did it was to use these two SP eco evolutionary models, uh, which piggyback on, uh, the, this Gillespie algorithm, which is a way to simulate ordinary differential equations in a stochastic manner. Uh, and so piggybacking off of that, um, what happens in these models is that we can take the parameters in these ordinary differential equations and model them as heritable traits, uh, of individuals and in doing so, um, this method, uh, for modeling evolution incorporates, um, things like demographic

stochasticity drift and trait variation, and allows for extinction, which some other methods of modeling evolution don’t really allow, uh, for these, uh, these eco evolutionary feedbacks to happen as well. Um, and the stochastic component is really important here. As I mentioned earlier, with the sort of cartoon where of species were to directly follow, um, the strongest selection gradient, we know they’re just going to fault themselves into extinction. All right So I’m going to go over a couple of results that we got from doing this. Um, the first one is to look at those, uh, stability and feasibility boundaries, and just see what happens with persistence of populations when there’s no evolution. So this will give us a sort of background of whether these boundaries actually describe well, um, whether populations are likely to persist or not. And then I’ll just show an example of predator evolution occurring along these ecological boundaries, uh, and how that influences the evolution evolutionary trajectory. Um, so again, we have our handling times and attack rates and we have these deterministic boundaries, but what does it look like if we actually have, um, populations within these boundaries? And so, uh, for a bunch of different parameter values, I, uh, simulated a hundred populations at each of them, uh, with no evolution and just asked how many of those persistent given the demographic stochasticity is happening And what we see is that, um, partially these boundaries are pretty well filled in, um, but at low handling times and high attack rates, the predators actually so efficient, um, at eating and pray that the densities are so low. Um, but they’re still likely to go extinct even within these boundaries. And we can look at different, uh, Ks or carrying capacities. And we see that this remains true across these, although maybe we’re starting to fill in, um, a little bit here, uh, at those low handling times And it seems like there are at least some, uh, unstable cycling populations that are all right, so let’s add in the evolutionary component. So we have this background of where species are able to persist in this handling time attack, rate plane, and we’ll just pick a starting value to start our predator population So if we do this, we can track over time, uh, whether these, uh, populations go extinct or whether they remain extent. So if we look at this, uh, we see that, um, predicted decently well by this persistence, uh, we see that populations at evolve, uh, these higher attack rates end up going extinct, whereas our persistent populations remain at this, uh, at these lower attacks. And so we can look at this another way by looking at, um, the actual time series of these evolutionary dynamics. This shows pretty clearly that it is these populations of all being high attack rates that are the ones going extinct. So we can also look at the average trajectories of these two different groups, uh, the populations that went extinct and the ones that didn’t. And so with this magenta line, we see that it both, um, was more steep in terms of evolving towards higher attack rates Uh, whereas the extent populations ended up at, again, this lower attack rate, but also had less deep of an evolutionary trajectory All right. So what does this all mean? Uh, first it suggests that, um, when we think about adaptive landscapes and, uh, eco evolutionary pathways, there might be some that are more viable than others. So these buyable eco evolutionary pathways that avoid these ecological boundaries might be the ones along which species are actually able to persist. It also suggests that there’s a real importance of constraints and stochasticity. So for example, again, if this population were to adjust and evolve, um, towards the strongest selection gradient, they would inevitably go extinct, but because of demographic stochasticity and genetic drift, we do end up having these populations that are able to persist over time. Uh, moreover, if there are constraints that keep species, uh, saved like a trade-off that keeps species from evolving towards these high attack rates, um, the species in which those constraints exist may be better able to persist. Lastly, it has some ramifications for interpreting past evolution. For example, if these, uh, bound, if these ecological boundaries are changing the evolutionary trajectories populations that are persistent, um, it’s, we can’t really say that, uh, the environment in the past

was maybe pushed species to evolve along some trajectory if there is this component of chance and avoiding these boundaries. Um, so anyway, really excited about this idea and excited to share it for the first time. Um, with that I’ll take any questions anyone may have. So, uh, we have a question from Andrew Hendry Um, so beyond theory, how often do you think such extinctions actually occur? That is how often do you think the constraints arise in reality? Yeah, I think that’s a really tough question and something that I think, um, particularly is, uh, in the say Darwinian extinction and evolutionary suicide literature. There’s a lot of talk about this, just how it’s hard to observe these things because it has involved extinction. Um, so we do have some ideas about, um, maybe some ways it’s a proof of concept show that this might happen in the lab. Um, but yeah, I think it is just really hard to tell how often this might be the case. Uh, and you know, here we don’t have any trade-offs built into this, right? So species are just freely able, uh, to evolve these lower handling times and higher attack rates would say for something like body size. Um, it was in the ecological stability example, but yeah, that’s going to trade off with so many different things that species are likely constrained in some ways where they might not actually be able to reach these boundaries. Cool. Um, we also have a question from Brian lurch What’s the timescale for which this can explain ecological persistence. You said if I’m understanding the model correctly, every population will eventually go extinct. Uh, so is this more buying time until evolutionary suicide will occur again in this system? Um, there are, uh, with the lack of trade-offs that we are built into it, then, um, most, either species will evolve to be extinct or actually in these models. Um, trait variation can be degraded so much that there’s not enough variance left in the population fruit to really evolve much And so that’s another way in which at least in these, uh, Mo in this modeling framework, you could, you could get that stoppage before you reach some parameter space in which you’re likely to see, and then even if the population succeeds in growing you’ve lost that variation Right. So that, that can’t happen. Um, do we have time for one more question? I think we do. Um, so as asks, it seems that the extent that the extent populations in your simulation are still doomed to extinction at some point in the future, since the selection gradient points toward the top left region, right Would it perhaps be interesting to consider situations with alternative stable States where you can actually escape ultimate extinction by taking a suboptimal path that leads the population closer to an alternative attractor? Yeah. Um, and this particular predator prey model, uh, uh, there are no alternative Sable States, right? You’d have to, you definitely have to use some more model with more complex, uh, ecological stuff going on. But yeah, I, I would say this work does suggest that, um, you might end up seeing maladaptive, uh, or, or traits that look maladaptive just because they are traits that allow a population to persist over time. Alright Stop sharing them. Yeah, we should. Uh, we should move on to the next talk. Thank you very much, Kyle, for a cool talk. I sent you another question on the, on the comments Um, the next talk will be by Zachary, uh, Lubbock, and it’s a biologist guide to model selection and causal inference, something we all need go for a Zachary. Sorry. I’m muted Um, cool. Um, so are folks able to, to see this my screen? Yeah. Okay. Um, so yeah, my name is, um, Zach Lebow and I wanted to thank everyone for taking the time to listen to this talk today. So this talk is based on a collaboration between some EDB researchers and epidemiologists, um, and on paper, that’s impressed titled a biologist guide to model

selection and, um, causal inference. So a common goal in biology is to understand how or why. And we usually start with an observation and then next, we might ask a question. So the question here being, why did the chicken cross the road? And then we often formulate a hypothesis, um, a Fox causes a chicken to cross the road. And so to answer our question question, and test our hypothesis, we collect data and then, um, use models to summarize the data and to draw inference. But biology especially observed in natural settings is rarely simple. So what variables do we include in our model and how should we interpret our results, our results, and ensure the answer is it depends. So it depends on our research question and what analysis should follow. So here I’ll share a strategy commonly used by epidemiologists to analyze observational data and the draw causal inference using an example from, so this is around, or this diagram is a roadmap to data analysis. And as you can see, it’s broken into four steps beginning with asking a research question and ending with results interpretation. So the first and key step to an analysis is to divine define the research question, and then decide on which data analysis task is appropriate. So there are four types of data analysis tasks, and these are description prediction association and causal inference So each task will help answer different questions and entails different modeling approaches For example, when doing a descriptive study, the goal is to generate a quantitative sign of a variable or a few variables variables of interest prediction involves identifying a collection of independent variables that explain a maximum amount of variation and a dependent variable. This discovery steps should then be followed by a validation and an independent sample or study population, both association and causal inference begin by identifying an explanatory variable X and an outcome. Y the cause of relationship between these variables is included in the directed acyclic graph or a dag. And I’ll discuss this in more detail later, what these tags are, the task of association proceeds with unadjusted models while causal inference typically involves models that include additional covariates that may confound modify or mediate the effect of X on Y. And I’ll go over this with an example, an example momentarily, but this figure is just meant to summarize a systematic approach to data analysis. Yes, we’re getting into more specifics about methods and models I think it’s worth mentioning what motivated this work. So I work on the Maura hyena project, um, and like, um, other long-term field studies generates large amounts of data. So I’m the one investigating questions with these data The researcher faces the decision, what variables should be included in their model. And this can be challenging, especially when the research goal is to test causal hypothesis using observational data. So in the remaining slides, I’m going to walk through an example involving spotted hyenas and show how various data analysis tasks from our roadmap are applied in practice And I’ll also discuss an important tool that’s widely used by epidemiologists, which is those direct to basic click graphs that I mentioned earlier. So this is it’s a generic, dag and tags are used or yeah, or directed acyclic graph. I’ll just call them tags from here on out. But, um, dogs aren’t use to explicitly define study questions to map out potential causal relationships between independent and dependent variables and to summarize an analytical strategy. So variables are with boxes and relationships between variables are represented with single direction arrows that indicate the flow of time and the direction of causation. So we’re often interested in an unbiased effect of an independent variable on a dependent variable until obtain this and an unbiased effect We must control for confounding. So confounder is a shared common cause of your independent and dependent variable of interest and not accounting for such a variable in the analysis can lead to biased and or spurious associations We may also want to control for, um, precision variables, which helped them minimize extraneous variation and either the exposure or the outcome And finally, an intermediate variable occurs in between the independent and dependent variables and as part of the causal pathway. So this is a dag representing the presumed causal and temporal relationships among multiple variables that we, um, have hypothetically measured in our study population of spotted hyenas. So for example, C one is pointing at X and Y implying that it’s a potential cause of X and a potential cause of Y. Whereas the arrow pointing from X to S indicates that X is a potential cause of pass. So like biology, this is, this is pretty complex. However,

just because we have these data doesn’t mean we should include every single variable in our analysis. So how we decide to model these variables and the inference that we draw, all depends on our research question. So when first studying the type of animal, we may begin with the task of description here, the goal is to provide a quantitative overview of the data. We typically focus on a single variable, for instance, X, the social connectedness or Y immune function. And for this task, we might be asking, what is the central tendency and variation of T-cell count in these wild spotted hyenas. We might also ask what do individuals, social connectedness networks look like over the course of development from time one at a time three. So for, for description, no prior or expert knowledge about the variables of interest is required. That’s an important piece to know another D data analysis task is prediction. So here we are asking the question, what sort of social and ecological factors explain maximum variation in the wild hyena immune function. So for prediction we use prior or expert knowledge to select variables that may re maybe related to our outcome, and then we can use any number of, um, step-wise or automated approaches to identify a set of variables that maximize the explained variation in our outcome of interest. And these are, um, these are largely data-driven approaches that do not focus on the causal or temporal structures between our explanatory variables So once the algorithm has been optimized with the training data set, a key component of prediction is to test the models, accuracy, and an independent validation dataset. And there are many methods available, including machine learning approaches for, and this is a, it’s an active area of development in EDB, um, and other fields as well, the data analysis task of association. So we again use expert knowledge to formulate the research question. And here we are asking, is social connectedness associated or correlated with immune function in wild hyenas to answer this question, we model the simple association between our explanatory variable and our outcome So we might control for precision co-variants like storage time, which can improve position of our estimate of association. We might also look at simple associations within strata of key covariates like sex. However, this task does not require controlling for confounders since we’re not trying to control bias in order to make causal inference here, we’re simply trying to understand for all associations So while association does not allow us to make causal inference, this task has the benefit of requiring the researcher to make fewer assumptions when constructing their model And this type of analysis is often used as an informative first step prior to co causal inference analysis, um, or, or it can be implemented when the data and or analytical techniques required to make causal inference are not feasible. Okay. Finally there, the, the task of causal inference and there, there are two broad end points of interest for causal inference models, including models that quantify the total effect of X on Y and models that partition the, the total effect into direct and indirect effects. So in both cases, the goal is to test a causal hypothesis and hear expert knowledge regarding the interrelations and the temporality of X, Y, and various third variables like confounders and mediators are required. So if our research question, um, yeah, if we were interested in a research question involving the total effects of excellent life, we might ask, what is the effect of social connectedness on a few on immune function in wild hyenas And in order to answer this question and test this hypothesis, we need to build a model that quantifies the total effect of social connectedness on immune function while controlling for bias. So this means that we must control for confounders up to the relationship of interest. So let’s assume for example, that there is no effect of social connectedness on immune function, but the geographical region where a hyena lives is a determinant of both X and Y here, a shared common cause of X and Y like geographical region opens what is known as a backdoor path between the social connectedness and the immune function, and can lead to spurious associations between the two variables. So this is that represented by the red arrow that, that actually came in a little bit early Um, so for instance, you can imagine that the association between social connectedness and immune function may transpire from the fact that hyenas in a certain geographical region have lower social connectedness and lower immune function due to human disturbance, but that’s it, the association that transpires from the confounding effect of human disturbance

is not a causal one. It’s one that occurs due to geographical region, but we are interested in the effect of social connectedness on immune function. So that’s when we need to control for geographical region in order to block the backdoor path and isolate the causal effect of social connectedness on immune function Okay. So what about these other variables? For example, there are, there are additional backdoor paths through the confounder C2, social rank, and through C2 and C3 in which diet is a direct descendant of social rang And by examining in our deck, we can see that social rank is a shared common cause of social connectedness and diet. So in our model, we can block the backdoor path between social connectedness and immune function simply by controlling for social rank. So this will also block the path that goes through C3 and thus controlling for any effect of diet. And in this case, it makes most sense to control for social, right, given that it’s upstream of diet. And because social rank is more reliably measured, um, more reliable measurement than say, um, endless diet. And that brings us to the second broad category of causal inference models. And here the goal is to quantify direct and indirect effects. And what is commonly known as mediation analysis. So as the name implies, mediational analyses often investigate, um, biological mechanisms that link are explanatory and how come variables, uh, for an example, Dr. Um, Liz Lange presented on this topic yesterday. So you should check out her empirical work. Um, in our spotted hyena example, we might ask to what extent is the effect of social connectedness on immune function, mediated by stress hormones and wild honey. So for simplicity, let’s remove all other variables, except our explanatory variable, our potential mediator and our outcome of interest in order to partition the variation in Y that is explained by our explanatory variable and our mediator We can use this equation in which the total effect of social connectedness on immune function is the sum of the indirect effect that passes through our mediator cortisol, plus the direct effect of social connectedness on immune function that is not due to cortisol, and we can start, but by estimating the total effect of X on Y next in a model that a clue that includes X, M N Y, we can condition our mediator and estimate the direct effect of X on Y that does not pass through them. Then the difference in the estimates for our total effect minus our direct effect gives us this indirect effect of X on Y the passes through M. So for example, if there’s complete mediation between social connectedness and immune function by cortisol, the direct effect would be zero. Now, there are multiple methods that can be used for conducting, um, mediation analysis, which each require various assumptions, but in the interest of time, I’m not going to discuss those, um, or those methods today It’s also important to note here that mediation analyses do not ignore confounding variables and that these would need to be controlled for in our models. So, including, I think it’s important to reiterate the causal inference does not imply control for everything. So you’ll likely notice variables in our original deck that had been skipped over or only briefly mentioned. And we’ll address these under the premise that our research question is again about total effects. So here for this final example, we’re asking, does social connectedness cause differences in immune function in wild hyenas? When our research question is about total effects, we should not control for mediators as this will lead to no bias estimates. Second, we should not control for reproductive state or if, or affiliative preference. So as indicated by the dare, the arrows in this stag, both social connectedness and immune function influence reproductive state, which make this variable, what is called a Collider and conditioning on a Collider opens a backdoor path between X and Y that otherwise was closed. So this can lead to Collider bias, which results in flipping the estimate of interest or, or, um, a spurious association between X and Y even if Knossos no such association exists. And then finally we can exclude or include, um, uh, precision covariates like sample storage time that only affect our outcome or explanatory variable based on whether or not inclusion of these variables improve the position of our estimate And the main point here is that, um, the statistical significance of associations between third variables and either explanatory or outcome variable of interest alone does not justify including that variable in the model as this can introduce bias, just like omitting a confounding variable can also introduce bias. Yes. Um, like epidemiologists Eby researchers are often

presented with large observational data with measurements on many variables. So given a collection of variables that we need to ask, how are we, we should decide how we choose the model, the data. And this depends on, um, clearly defining our study question next, we should decide on an appropriate data analysis, task or tasks. So it’s worth noting that these tasks can be used together in a research project We should also use director, basically grass to map out hypothesis and to incorporate expert knowledge into our analysis plan. Diags also inform our model specification and importantly, these can help identify sources of measured and unmeasured confounding. And then finally we should interpret and describe results based on our research question, um, and which data analysis tasks that we, that we ended up using So here we should discuss the potential sources of bias and alternative Hyatt hypothesis as highlighted by our dead. You have four minutes. Perfect. I’m almost, I’m almost done Thanks. Um, yeah, these methods will not replace randomized experiments in seeking to understand causation, but these can provide a viable option for testing causal hypothesis, um, especially when using observational data and when experiments are not feasible or, or ethical So if there ends up being any time for questions before we get to those, I just want to acknowledge, um, some funding sources, um, and, um, the work of, of numerous causal inference, um, epidemiologists, including, uh, Dr. Jessica Young, whose research and ideas helped to shape our paper. And lastly, I wanted to say, thanks for everybody for taking the time to listen. Awesome. Thanks. That that was great Um, all are waiting for questions I wanted to ask you. So it seems like the, the beginning point and the ed point of your dads are pretty well-defined. And then in the middle, you have a lot of things. How, how do you make the decision between classifying different pieces of that middle section as, you know, either confounders or intermediate variables or any of those other types of that you were discussing? Yeah, that’s a really great, great question. Um, so this, I guess this is, um, I’ll just go to this, the stack in particular, this is where, um, pre previous, um, sort of expert knowledge and, um, work hasn’t been done in on the system is required. Um, so you, you might, yeah, if you just start studying the system, you might not know the temporality or how some of these, these variables in the middle are related to each other, but hopefully as the system becomes sort of a more developed, especially in some of these long-term studies, you can you get a sense for, um, you know, what, what proceeds, what, um, in terms of, and that’s a, that’s a big component of it, I guess you can’t overlook the temporality aspect of it, for sure. Um, thank you. Then we also have a question from Bob. Bob says, awesome talks. Zachary, does it make sense to include, uh, causal feedback? I E removing the a and dag, for example, feedbacks between ecological and evolutionary dynamics? Um, sure. Um, I’m not sure I’m fully understanding, but I’ll, I’ll go with what I’ve gotten, but I’m happy to chat with, um, Bob afterwards So I think some of these, um, yeah, some of these, these variables could, could be certainly on an ecological or evolutionary scale, the, the, whether or not there’s an actual feedback loop or a double-headed arrow that, that does not come into, um, into, to the, the model is yeah. Becomes very challenging to, to model So the diags or explicitly, um, one directional, um, yeah. And would have to identify that, I guess you would have to be explicit about this possibility of a feedback loop if that existed. Great. Thanks so much for that. Great. That was an awesome talk. Thanks Okay. So it’s time to get started again. Uh, again, this is the contributed, uh, paper section 10 from the virtual, a civil Mar American society of naturalists meeting. Uh, and this contributed paper session is titled methods, models, and perspectives. And so we have the last set of talks, uh, for the conference prior to the natural history trivia. And the first of those will be by, uh, Matthew stra talking about non-directional cryptic female choice, maintaining variation in the Jack Hewlett traits. Go for it, Matthew, thank you. Um, can people hear me and see the cursor and desktop looks good? Cool, perfect. Um, yeah, so thank you everyone for taking the

time to watch this and for all the organizers and moderators for making this possible. And my talk today is auto model that I developed with my advisor, Suzanne Alonzo, ask questions on the maintenance of variation and post CAPA Tory sexually selected traits. And I really first became interested in this question because of the diversity of sperm. And so we actually look across the animal kingdom, sperm are some of the most diverse cells and rapidly evolve excels. And so while we’re familiar with sort of the tadpole shapes for himself, we also have cells that sperm cells that lack flagellins have multiple flagellins and have all sorts of different, crazy shapes and sizes And the vest diversity of sperm suggests that sperm traits such as we’re faulty, but other things are under strong selection. And we might expect that strong selection should degrade genetic variation However, when we look within species and within populations, we often find that there’s tends to be a decent amount of heritability and variation in sperm traits. And so for a question that’s sort of outstandings how’s this variation maintained, and one proposed mechanism that’s been verbal, but not formally tested. Using theory is something called Don directional cryptic female choice. And to back up a bit, we mostly think of female choice. We think of. Um, so females choosing to made on a male based off of some sort of trade, like an ornament However, this is only half the story and many species females mate, with more than one male, and this can result in post-conference her sexual selection or selection and traits that aid in fertilization success. And this is where cryptic female choice happens. It’s when females bias fertilization towards specific males and by non-directional cryptic female choice. I mean that there’s this male by female interaction. So say for example, if one female mates with two different males, no one, a male two, um, with, in this certain area, we see that male to consistently sires 80% of offspring. However, if there’s a different female with the same males, we might see that paternity is reversed. So male one SAR is a majority of the offspring consistently And so these male by female interactions are what we call a non-directional cryptic female choice. And there’s been growing evidence that this process happens and why for a, a ride for a wide variety of species, including species with both internal fertilization and external fertilization. And so I developed a model trying to ask the main question being, does these male by female interactions or non-directional cryptic female choice, does that help maintain genetic variation compared to other types of selection? Do trade-offs help maintain genetic variation and other proposed mechanisms? So say sperm count trading office firm size, and finally interested in how risk of sperm competition. So the probability of female mates with more than one male influences us as well as the strength of selection and particularly the, how they interact with one another. And so to do that, we developed an individual based model. And so the individuals in this model are either male or female. And so males express some sort of nondescript male sperm trades. So we could think of that maybe potentially as sperm length, and it’s completely determined by the genotype. And so for each trait or each trait, we have two different low side and because they’re sexually deployed, they get a copy from both parents, but it’s only the male traits only expressed in the male. Then we have a female choice trait, which only comes into play under the non-directional cryptic female choice part And once again, completely sex bias expression, only express females, but both males and females carries that. And in terms of what traits these might be in a biological real example, we might think that male sperm trait could be, say, sperm length and the female, and the choice trait could be the reproductive length, which at least comparative studies show that these often correlate evolution with one another. And then finally also modeling sperm number. And this is so we can get at that trade off. And so when there is a trade off, I’m assuming that sperm number is inversely proportional to male traits. So if you have higher value of male sperm trait, you’re going to have lower values of sperm number, um, with the no trade-off model, I still cleaning sperm number, but it’s separately controlled genetically. And so once again, the parameters of interest, so I’m going to very risk of

sperm competition. So the probability of female mates with more than one male, and then we’ll look at four different types of selection So it can imagine it being a fair raffle So where there’s no direct selection on the male sperm trait, all that matters is which male produces the same the most from number finally, um, just a shorter, I’m going to S matching selection, which is the non-directional cryptic female choice. I call it matching is because you can think of each female having an optimum and the best, um, the male that matches that optimum has higher fertilization success, directional selection. So always larger than that. Male sperm trait value is better and stabilizing selections Is there some sort of fixed optimum, and then I’m going to look at the strength of selection And so what this looks like, I’m just going to go over all of them. Um, so this is all assuming that sperm count is held constant And so on the Y axis, we have the probability of fertilization. Next axis is different values of the focal male trait value on the red dash line is the optimum trait value. Blue is the competitor trait value. And then the different lines represent the different strengths of selection. So in this cartoon example, in this example, the male competitor is at 40 The optimum is around 99. And so say focal males around 50 under weak selection, the focal male has around like a 70% probability of fertilization holding sperm County. Um, the same under mater selection, that same difference results in 80%. And then under strong selection, that’s almost a hundred percent probably of fertilization stabilizing selection is very similar except the same equation, except for that optimum to set sorta at that starting average of the populations And if you can notice here is that it’s symmetric So it doesn’t matter whether or not you’re greater or less than it just matters how close you are relative to your competitor. And the way, like I mentioned before that we can think of matching selection or this non-directional cryptic female choice is that each female with the female choice trait value creates own optimum And so this is the model process. So it start population of equal sex ratio at 500 individuals And we also tested, um, the a thousand, but results are called tentatively similar. So I just focused on 500 and then each female randomly mates with the least one male with some probability, which is that risk of sperm competition. Second male, each female produces 10 offspring. And then the paternity of wedge is determined by the two males, um, both by the sperm count and the sperm trait using those functions that I showed for the way sperm count comes in, is that the values from that, from those equations that I should wait the sperm number. And then finally, the offspring genotype is determined by making quote unquote sperm and egg. And so I’m also allowing mutation in this model to take place. And then we returned to the original population size and sex ratio by randomly sampling from the offspring. And then I’m repeating this for 500 generations I did run a subset to a thousand generations, but things generally stabilized by 400 generations And then, because there’s a stochasticity in this model, I modeled five 50 different populations per parameter combination to get at some of that variation. And so right now, I’m just going to show a subset of what this might look like, the evolutionary dynamics, um, Mrs. With weak Mrs, with a moderate strength of selection and 50% risk of sperm competition. So each female, it has a probability of mating with another male, 50% of the time, and there’s also no trade off. And so up here is the average trait value, red being the female choice, trait value, blue being the male sperm trait value. And this is the coefficient of variation in that trait value and then on the X access to the generation. And so we can see with the matching selection that the female and male trait tend to co-evolve with one another, they match each other, which is what we expect and variation generally, declines males. And then next directional, we see that the male trait value evolves quickly. And once again,

in this small, there’s no selection acting on the female choice, right value. And then with coefficient of variation, there’s a rapid decline. And then whenever there’s a new mutation, you do see peaks of coefficient of variation during that selective sweep under stabilizing selection, because I sat at the optimum to the average trait value, doesn’t really change the average trait value amongst the different populations. But we see do, we do see that coefficient variation rapidly declines, and then finally fair raffle. You can sort of think it as like a base case in this case, because there’s no trade off, there’s no selection going on. And so both traits are just scripting and putting it all together just as just like one subset of our analysis. We see this is what we just saw. And then we can see that it looks like variation tends to be higher matching selection. But what we wanted to do is look all the different parameters we ran. I took the average and then bootstrapped the average so we can compare it. And so for now until walk through some of the main results. And so on the Y axis, we have that coefficient of variation of that male sperm trait on the XSA accesses, different risks, sperm competition. So 0.5 would mean each female has a 50% chance of baiting with another male. And the different types of line indicates whether there was a trade-off with the data or straight line is no trade off And so just starting off with fair raffle So fair raffle without a trade-off once they, like I said before, meant that there’s no direct selection acting again, there’s no selection at all. It’s just scripting. And what you can see is that spur risk of sperm competition. It doesn’t matter. You get that interaction with a trade-off because you actually have indirect selection for smaller male trait values, because remember it’s inversely proportional, would you see over here? And then my life matching selection, we see that matching selection generally maintains more genetic variation, then both stabilizing and directional and somebody else we can note here is that the presence of trade-offs except for unfair raffle, when we expected it to be genuinely, doesn’t make that big of a difference in terms of the maintenance of variation. Um, do you see this interaction effect between risk of spread competition? So it doesn’t seem to risk of sperm competition. Doesn’t generally matter for directional selection, this darker blue, but it does for stabilizing and matching and because, um, stabilizing selection, do you see a sharp decline when there’s risk of sperm competition increases, um, at least, uh, low risk of sperm competition stabilizing tends to maintain more variation than directional selection. And I said that, and so then we can look at moderate selection and we see that it’s pretty, it’s a very similar pattern The main difference is that directional selection now tends to maintain more genetic variation than stabilizing selection. And then when we look at strong selection, we can see that stabilizing selection, pretty much everything goes to fixation. And we do see that matching and directional selection tend to maintain about the same amount of genetic variation Okay. And then when we put it all together, um, Oh, and then so with we, and then putting it all together, we can see that as strength of selection increases, um, for both directional stabilizing and matching, we tend to see a decline in variation. And there’s also this interesting interaction in terms of the effect of risk of sperm competition. So under weak selection, we do see that as risk of sperm competition increases at least for stabilizing and matching selection. And there is a decline in genetic variation. However, that slope, as you can notice, goes away under strong selection where risk of sperm competition doesn’t seem to matter in terms of understanding the amount of variation maintained. And so main conclusions are that male by female interactions generally can help maintain genetic variation, especially under weak and moderate selection. Um, surprisingly trade-offs do have a limited effects on the maintenance of genetic variation. At least

the trade-offs I explored here are four minutes left. Thanks. And then genetic variation tends to decline with risks from competition, but it does affect weekends as the strength of selected increases. And finally, I’d like to acknowledge my advisor’s Suzanne Alonzo national science foundation, my lab mates, and folks at the biology of sperm conference that helped put inputs when I was at the beginning of developing this model. And I’d be happy to take any questions. I have one bar if there aren’t any in there, there was one for Maddie Um, it will large variation in sperm traits is only seen in a fair raffle situation. Oh, sorry, my Slack, my all the way or for a sec Oh, there we go. Okay. Uh, how common is that fair raffle in nature? Uh, Oh, I would say generally it’s unlikely. That’s for traits That wouldn’t matter. Um, but there are definitely in some examples where at least experimentally, it seems like spring loss It doesn’t matter as much as sperm count Cool. Okay. Um, Matthew, I’ll ask a question Um, and it’s, of course it’s appropriate that the last question came from Maddie because her work is also looking at another mechanism, maintaining variance within populations. Uh, and so, um, what I’m curious about is, um, to what extent the classical mechanisms suggested to maintain variance, despite selection might alter the outcome of these, uh, simulations For example, if you add a negative frequency, dependence selection or temporal variation selection, or spatial variation, and coupled with gene flow, would those things just overwhelm the results you’re finding here? Or would these emerge even in the context of these other forces? Um, I mean, that’s a great question and that’s definitely something I want to explore more, um, in terms I can’t, I honestly can’t make a great guess on that. Um, yeah, but that’s, that’s definitely a great feature direction. Yeah. It’s one of these classic things where it’s, it’s like, uh, a question with too many solutions almost, and both you and Maddie have advanced further ones. Um, it’s just really interesting that so many things can contribute to this process and we have difficulty knowing which are the most important. Um, so if you will indulge me, how do sperm move without flagella expert at that? Uh, I believe that it’s like an amoeba Um, does it mean like shapes sort of like crawl, but that’s, I find that very weird, but, and then, um, yeah, I believe a lot of those ones are from crustaceans. I don’t know the exact species, but okay. Thank you, Matthew Uh, we’ll move on to the next talk now. So the next talk is by John Harper and he will be talking to us about sexual antagonism in humans, low PSI and evidence. Oh, um, yeah So yeah, this is my talk. Um, before I start, uh, I is 11 o’clock here in England. So my apologies if I’m a bit, seem a bit tired, a bit slow. Um, so yeah, so my talk is about section tokenism humans. Um, so I am a PhD student at the university of Sussex and Ted Mara is my supervisor. Um, so I’m going to start, we already heard about what social antagonism is from Thomas’ talk earlier, but, uh, in case you didn’t make the case for people who weren’t around, um, sex and tokenism is when you have a trait which has beneficial if possessed by one sex, the deleterious in the upper, and what are the big horn sheep is a classic example of this males would want to have bigger homes because their corns are useful for competing against other mates Um, when the, when they’re trying to, um, men not competing is sexual selection. However, these horns are not particularly useful in dealing with predators. Whereas females are want smaller horns, smaller dagger like horns So they’re able to actually stab predators

to ward them off their young. Um, so for a male to have small ones, they would not do very well. It would be unable to breed. They’d be able to fend off other males and Nate, whereas for females, I mean big coins, they’re kind of useless for dealing with predators So what you have here is one trait in different directions being beneficial and deleterious to different sexes. So what we really want to find out for, from an evolutionary perspective is how common actually is sexual antagonism And there’s been a lot of work done on this Um, and it’s widely believed that it’s very widespread, pretty much every taxer where it has sex is found to have sexual thing to a degree in various different predictions from modeling and genomic studies. Despite this, there aren’t very many examples of specific low PSI. So within genes or particular basis in the genes, in the DNA where we know that they have a section antagonistic effect. So the Boston paper revealed that there was a salmon example and there have been a few discovered under software, but for the most part, there aren’t very many concrete examples of sexually antagonistic low side. Um, and at the same time, uh, there was plenty of evidence of sexual antagonism and of course sexual dimorphism is known to be very widespread. Um, pretty much every species which has any kind of everything sex at all, um, has done offers into a different extent. Dimorphism. They’re not always, um, physically obvious as well. There can be many, um, less obvious changes in biochemistry It’s not just morphological traits. So is there a section time there’s many humans? So there is definitely sexual dimorphism in humans. And, uh, there are very big differences in sex in diseases, particularly look at cancers So this study looked at the different instances of human cancers. And if you look at the top cancers in either sex, you see a really strong dimorphism, like you have ovary cancer, free government females and prostate cancer. They’re both pretty big cancers in males and females, but they’re virtually absent in the opposite sex. So we see this huge dimorphism in humans So could they also be low side? We don’t know of any actual examples of social antagonistic human though site, or at least that used to be true because we search out to search for examples of human, low side human illegals, all those sites, which were sexually antagonistic So why don’t, why don’t we know of any, why are we not finding any? So we had an idea We decided to look at this in two different stages. The first stage would be looking for these low side using search terms used by evolutionary biologists. So terms like sex antagonism, sexually antagonistic and intro Lucas, sexual conflict. So if thought about how evolutionary biologists would describe sexual antagonism, and we filtered for humans, uh, the search returned about 44 papers, but none of them actually named specific, although, sorry, as we expected, because we didn’t believe any has actually been found, but then we had another fall What if people have found sexually antagonistic, low PSI, but they didn’t know that’s what they were biomedical. Science is a huge field and lots of different studies are going on at any one time. Could any of them have found such an antagonistic low site and not knowing what it was? So we started to think about how said people might describe sexual antagonism If you didn’t know sexual, what sex antagonism was as a concept, how would you describe it? So we started putting together words like sex, gender, male, and female, but not just that men and women as well, and boys and girls, because these are human studies. They wouldn’t necessarily just say humans in the script or male or female labor. And we combined those with any kind of word that she would refer to the locus low side gene pulling off a specific genetic structure. So this search returned many, many more papers in pub med 881. And we screened and went through all these papers and screened them. And we discovered that 42 of these papers describe low side in total There were 51 which had solid evidence of a different effect in each section. So not just an effect in each sex, but effect an opposite direction. And what were these low side that we discovered? Uh, there were 21, which we decided we called complex traits. And these would be what we mean by complex traits would be things like waste hit ratio. So the rationale

for splitting these is because we wanted to be sure that these were actually having an effect on fitness, complex traits. We couldn’t be sure you could infer perhaps that having a high BMI is detrimental for fitness. You’re more likely to be unhealthy. You’re more likely to die young, um, and similar things with high blood pressure, but you cannot really say that these are definitively selection related. It’s difficult to actually make that inference. You can perhaps argue it could be, but there is room for ambiguity and the others less. So, so we had 19 disease risk and severity traits across 21 different low side. So these would be things like cancers So a gene as locus or an illegal, for example, which would increase the rate of cancer in one sex, but decrease it in the other or perhaps increased severity or, um, make prognosis The prognosis of said cancer worse. So that’s what we mean by disease risk severity in no case where these referred to as sexually antagonistic And in one example, a paper had discovered in effect in different directions, but it dismissed it as a false positive, which we think might mean that actually there’s even more that we’ve not discovered and I’ve never made it to publication simply because researchers have looked at their results and gone, these are rubbish and bend them. And also, unfortunately none of these were validated or independently replicated by another study. So no case where were we able to observe the effect of the same Lucas in several different studies? So this first thing I want you to take away from this is these exists. They’re around, they’re out there. Next is going to require a bit more thinking, um, because we wanted to have a look at whether or not these illegals behaved in a way which was predicted by any of the past models of sexual antagonism. So there’ve been lots of theoretical studies that the sexual antagonism and how these antagonistic ideas might behave. So we decided to try and test some of these. So the metric we’re really interested in is what we call effect size ratio. Um, so each sexually Antonis is illegal by beneficial, by definition has a beneficial effect in one sex and the deleterious effect and the other sex. The metric we used looked at the ratio between these, we divide the positive effect, which should be a positive number by the negative effect, which is going to be a negative number to produce a negative number. Our ask is always going to be the case for a sexually antagonistic Lopez on this graph. If you look at the X axis, which is the effect size ratio, the LDLs, which are more to the left will have a net positive effect So there’ll be more negative number, but a bigger magnitude number as you move left These are going to have a stronger, positive effect and a smaller negative effects as a consequence. So our hypothesis was that these illegals, uh, as would make us, would just kind of make logical sense because their beneficial effects are bigger than the deleterious effects And they’re going to be more like beneficial deals overall, they will be balanced a bit by the negative effect, but they’re going to be more like positive. And therefore we would expect that they would move to a higher frequency. Conversely, we have negatively, which I’ve got a larger negative effect and a smaller, uh, beneficial effect. Their effect size is going to be, um, relative. It’s going to be, this magnitude is going to be smaller, but it’s going to be less negative is going to be more over to the right of the access we have here. And minus one, the point at which you’ve got equal effects sizes is going to be sort of like the tipping point. So everything shaded blue is going to be more like a bare appreciated, white, sorry, it’s going to be more like a beneficial deal and everything to the, in the red zone is going to be more like a deleterious Lille. So is that what we see? So we examined the streets, uh, if you, first of all, by, um, stratifying them according to their trait class, unfortunately we didn’t really find anything. There’s not enough points in either one to find, uh, any statistically significant relationship, which is really a power issue from the lowest sample size. But if you combine both treat classes together, once which conferred disease risk and severity traits and complex traits, we find quite a strong negative relationship, as you would expect more beneficial heals, I’ve got higher effect of or frequencies, whereas more negative ones have got lower

Um, and these points are weighted as well So we weighted according to, um, variance, so points which have got a lower variance, I’ve got high waiting, cause we’re more sure of where they are. Um, and you can see us as a six down here, we had quite a significant P value. So we actually have this really solid trend. Um, and we tried several different modeling methods to try and find the best fit. And it was significant in many different ones. So that’s the second thing I want you to take away from this talk is that not only did these illegals exist, but they also behave in accordance with predictions made, um, from previous models and previous statistical models and simulations. So really there is kind of a wider point as well to be made. It’s the fact that there’s different terminologies prevented these examples for me, communicated by evolutionary biologists have been looking for human sexual interest, sexual sexually antagonistic low site in humans for a long time. And haven’t and discovered them because there’s been this terminology gap here because they’ve been not been thinking about how medics would perhaps a scrub. Um, however, we do need to be really, really sure that these are sexually antagonistic, low side. We need some more independent validation, which would have to come really from the biomedical scientists So all of this information are more is available in the pre-print overall Medoc I’ve. Um, so if you are interested, please go and check that out. Um, now that brings me to the end of my talk. So thanks very much to Ted Moore and Tim Johnny King. So my supervisor and the collaborator on this project and also to the people who commented on the manuscript, Jessica rabbit, Tim Colin, and Phillip Brazuca Uh, you can contact me on that address there and thanks everyone for listening. Um, both zoom and also YouTube and people from the future on YouTube watching this back fast for watching, um, and your questions. Great Um, so we have, uh, we have a question from Andrew. Andrew, do you want to verbalize it or should I read it? I can say it out loud That’s fine. Um, so I’m, I’m well, I mean, it’s really fascinating that you have, um, just you changed the search terms to find, you know, a bunch of missing literature, but then also your inference that what you’re finding seems to suggest that basically there’s a publication bias against sexually antagonistic low PSI Um, so I’m just wondering if there’s any way to formally assess the magnitude of that bias So for example, you know, people use funnel plots and meta analyses and other sorts of things like that, to try to assess the extent of publication bias. Is there any way that you can see formerly doing that to get sort of a better estimate of the average overall effect and number of sexually antagonistic will say that’s a very tricky question. Um, because even if we were able to go over everything, we’d still have the issue of people who have, um, because there’s always a pressure to find results, to find positive results, um, which should drive people who discover why they don’t struck to find by sex who discover sex checkstand sinus fat and go, well, this is rubbish. It can’t be opposite in males and females. This is either going to be a disease, the risk gene, or it’s not. Um, we did try looking into other databases, um, but it’s really quite a mammoth task really, because this is not just even though the most metaphylaxis, we’ll be looking for one statistic, one value to come out of it. But the problem is that none of these studies are looking at precisely the same thing. They’re looking at their own candidate genes and candidates illegals and the effects of settlers. So I’m not really sure how one would go about actually trying to assess the bias itself. Um, but I imagine it wouldn’t be a very easy task at all, but it would be very, very interesting though We have plenty of time. So I’ll ask a up question Um, so another way that one might potentially get a hint of this is again, seeing if studies that had, you know, smaller sample sizes are less likely to report these sexually antagonistic, uh, uh, alleles are low PSI. Um, and another way is to look at the publication, um, like effect size variance through date of publication, but because people may be more willing or less depending to accept the sexually antagonistic results and publish them as time goes on Yeah. So we did, we did look at the kind of

the years of publication. Um, there wasn’t anything massively meaningful we could confer can, could, could sort of take from it. But, um, generally we do see that. So it’s more common that you would see these papers. Some, a lot of these papers would be forums, sort of the 2010s and later. Um, there weren’t very many at all, um, sort of before 2000, it’s quite a recent real thing that people have actually been, uh, perhaps because of more awareness in gender differences in medicine, but it’s only fairly recently that they’ve actually really been properly documenting these things. Um, and the power issue is quite a big one. So, um, anyone who’s doing a study into a gene, if they stratify by sex, then they’re cutting their sample size in half or depending on the sex ratio So they’re going to lose power from that, which is something that’d be less likely to do. And as a result of that, then also makes it less likely that they would find and report, um, uh, sex, different effect or a sexually antagonistic. Those are all really good points, John. Um, and also there’s, I know in a lot of animal studies, often only males are studied, for example. Um, I wonder if that’s in humans too, so we have a few more questions. Uh, hopefully we can get to all of them. So first we have one, this one is on Slack from Andreas Um, have you looked at all at the geographic distribution of these sexually antagonistic illegals and are they limited to just some populations? Do you have any sense of that? Yeah. Okay. Um, so sort off as the answer to that one, I guess, um, each study discovered in Leo and they won’t have their own study populations. And the vast majority of said studies would be looking at one population in one area. So, um, they’d be looking at speed dating success in, um, I can’t remember what that one was actually, but that was a schizophrenia and Koreans, for example, study, um, various things like that. So they were focused on particular populations. However, um, we looked at Aaliyah frequencies and different populations off these off snips, um, and compared them sort of different trended. So not quite geographical areas, but sort of different populations, different, uh, races, I suppose Um, and we find that for the pretty much every ilial is present in every population to some degree the frequencies do vary. Um, I don’t think there were very many which massively varied infrequency, um, but it would be interesting to do a more detailed analysis of that to actually find out whether or not because the environments are going to be different, the selection pressures will be different. And it’d be interesting to see if these effects are that actually, um, persistent over different communities and different societies. We have another question from Thomas. Um, he wanted to know what was the relative proportion of low side that you found on the X versus on the autisms? Um, so we actually found zero on the X chromosome, um, which is quite interesting Um, there was no, so from the rest of the autism just quickly there w we didn’t find any particular enrichment in any specific area. Um, but there were no, there were no examples on the X chromosomes or Y chromosome by extension. Um, but it could be because of, uh, the publication bias and the, uh, sex bias that we mentioned earlier. Um, because again, if you’ve, when you’ve got a dizzy, if you’ve got a disease leader, which is only on one sex chromosome, then that’s perhaps going to be more complicated for people to study. Uh, and we wouldn’t necessarily be able to, in some cases, look at sex sexually antagonistic effect. Um, we should, um, we should transition to the next speaker now to make sure that we stay on time. Thank you very much, John, you have a few questions on the Slack channel, so you can check that out. Okay. Right. Thank you. Okay. Thank you again, John. The next talk is by Samuel scar Pino, and he’s going to talk about the effect of space hierarchy and metapopulation structure on the shape of COVID-19 epidemics. Okay, great. Thank you very much. And you see the slides, everybody, or at least everybody that’s here. Yes. Okay, great. All right, well, um, thank you all very much for being here this evening or morning or, or whatever it is Um, you know, I know that person, I wish we were all together in a silver bar, um, for a whole variety of reasons, but I think continuing

to do these kinds of events where we can come together and talk about science and give people opportunity to share what they’ve been working on are perhaps even more important now than they’ve ever been. So, uh, to the organizers, I certainly understand the effort and dedication that takes to pull something together like this. So thank you all very much. Um, and today I’m going to talk a little bit about some work we’ve been doing on the spacial hierarchy and metapopulation structure of COVID-19. Um, this is heavily motivated by, uh, some classical work in ecology having to do with the effect of crowding, uh, on animal populations or really the effect of crowding, uh, as, as first proposed by Lloyd in 67 on perhaps nearly everything that we think about in, uh, equal evolutionary dynamics before we do that, um, I want to mention just very briefly a project that we’ve been working on since about this time last year. And this is also motivated by one of the things that I think is incredibly important about this community, uh, and that we can bring out into the broader community of people that are working on, uh, eco evolutionary dynamics, whether that’s in human populations, as it pertains to COVID, or whether that’s an animal or plant populations as made the talks you’ve heard about today, and that’s the importance of data, uh, and, and large comparative data sets that we can use to answer questions, or at least ask questions, uh, about, uh, things that are of relevance either to human populations or in general to science And so we’ve been working with about a hundred volunteers now supported by a variety of, of foundations, uh, and technology companies to curate a large anonymized individual level data set of COVID-19 cases that we were fortunate to have this covered by New York times magazine back at the beginning of the summer. Um, now, uh, so we did, at the time of printing, we had 142 countries with a little over a million cases of COVID-19. Uh, this number now is up to about 5 million cases, and these are all hosted and have been on an open, uh, open database, uh, since, since the very beginning, uh, as we’ve been capturing these data and fortunate to have the support of a diverse international consortium of colleagues now that are working under the auspices of an organization that we’re calling global.health, uh, to continue to build out these data and make them available internationally. So that researchers and public health officials can learn more about COVID and, and respond to it. The talk that I’ve been getting about COVID-19 starts with a slide that says COVID-19 became a pandemic because the world doesn’t understand complex systems, and I’m often talking to individuals interested in complex networks. So I put networks and apprentices We could replace this with, uh, the world, doesn’t understand fundamental theory and ecology and evolution. I put an asterix here because I certainly don’t either. Uh, one of the things that I have realized as I was going about this is that I should’ve paid attention, uh, in some of the classes in graduate school a little bit more carefully, because it turns out that a lot of this stuff that we’re grappling with now was, was fairly well understood from a theoretical perspective 50 years ago or more. And so I think one of the things that’s really important for us to remember it, and a lot of you have contributed on the COVID front, uh, either as a volunteer, uh, or, or leading a research efforts. One of the things that’s important for us to remember is that when people talk about, uh, researchers staying in their lane or epistemological trespass, what they’re really talking about is people being jerks and not coming into a field with, uh, an open interest in collaboration, uh, and, and integrative science. They’re not talking about, uh, keeping barriers between ourselves. They’re not talking about how interdisciplinary science is not the path forward for, for COVID and so many other things. So I really, I want individuals on this call that may be have, or on this little listening that has thought perhaps that, that they can’t, or shouldn’t contribute, uh, to research around COVID-19 because it’s not in their area of expertise Uh, it certainly wasn’t in mind, uh, until I was a post-doc. I started, uh, working with professor Linda Delvin, Indiana on quantitative genetics, uh, in the field and in the lab And, and initially worked with Parker, Patrick, uh, as an empirical population, geneticists, not surprisingly that, that didn’t last very long, uh, and, and moved into more theoretical work. So really it’s important that we remember that a lot of what we learned as, uh, ecologists evolutionary biologists, uh, has quite a bit to say about COVID, uh, and, and many other things. So what about, what do I mean by this? Well, it turns out that some of the things that we have gotten wrong, and I’m not going to say who we is here, uh, I’m happy to do that over beers, but that we have gotten wrong about COVID-19 go kind of all the way back to the very beginnings of things that had been worked on in ecological theory in particular around population growth and population dynamics. So what do I mean, well, here’s a picture of an individual with 1918 flu Here’s a picture of individuals responding to the Ebola outbreak in West Africa. One of the most common numbers that we might use to describe these outbreaks is the, are not,

or the R zero. It’s approximately the average number of secondary infections that has a much more specific definition, uh, which means it’s almost never relevant for diseases in finite population sizes. But you can think about this as the average number of secondary infections. It’s approximately the same two for both of these diseases, with the important caveat that I do understand that saying something that’s non-linear is approximately the same as is approximately meaningless, but it’s about to however, 1918 flu half a billion people, a bowl in West Africa, devastating to Sierra Leone, Guinea Liberia to our global economies. Uh, 30,000 people is a lot, uh, in terms of mortality, but it’s not, it’s not 500,000. And so why is this? Well, it turns out that we have an answer and we’ve had this answer for some time, which is that when we use the average number of secondary infections to project out to the final size of an epidemic, most of the theory that is commonplace in epidemiology, and certainly on the front page of the newspapers now assumes that we can model the offspring using a post on with Lambda equal to the are not. And so that we can really neglect the importance of super spreading events. We can neglect most of the importance of variability because the average, uh, and the variants are equivalent under these assumptions, foundational paper by Lloyd Smith at all relaxing this assumption, uh, that we can describe the dynamics of an epidemiological process using only the average and instead uses a negative binomial distribution to describe the offspring, the number of secondary infections for an epidemic process, and comes up with much more realistic expectations about the way in which these dynamics are going to play out. Now, for those of you that have not encountered the negative binomial distribution before, I will tell you that one of the rules for publishing on the negative binomial distribution seems that each new paper has to invent a new set of notation. And so I will often have to spend 30 minutes with Wikipedia and a bunch of conspiracy theory strings behind me trying to connect up the dots between whatever someone’s notation is and the notation that I personally understand Uh, but the way that I think about this is that there is a second parameter Kappa that describes the importance of super spreading in this case. And let’s not also get confused back to the fact that some ecologists would call over dispersion as being randomly distributed, where I would think of over dispersion as being super spreading, meaning there’s clumping or clustering, but setting that aside, large values of Kappa mean that the distribution of secondary infections or offspring gets closer and closer to Pusan, and his capital goes to infinity. Uh, it becomes plus on with Lambda equal, to are not in practice capita’s of about 10 are functionally Infiniti. So flu has a capital of about 10. It basically acts like a mass action, random mixing, uh, with a Pusan distribution, uh, of, of secondary infections that are not up to as capital gets closer and closer to zero. Uh, it becomes increasingly dominated by rare, super spreading events. So you can think about a disease like Ebola. It has an average of two, but that’s because just about everybody, in fact, zero other people, and occasionally someone will infect 30, 40, 50 other people. And so the entirety of the dynamics of the Bola are driven by to first approximation. That’s weird, I’ll say first approximation is the second moment, right? It’s all about the variability. And so this plot from the Lloyd Smith at all paper shows the proportion of infectious cases ranked on the X axis based on the expected proportion of transmission events. So if we just think about this in a discreet time model where everybody that’s infectious today will recover tomorrow It’s the proportion of infectious individuals infected today ranked by the proportion of infectious individuals that they infected when these infectious individuals become infectious tomorrow. And so this one-to-one line is a homogeneous population of facade models, influenza And as we see these lines blowing up into the left, we have increasing importance of super all the way up to SARS, which is almost entirely dominated by, uh, by super spreading And I can label this with a bowl and influenza And so, uh, most of the difference between a bull, uh, infecting 30,000 people and being eminently controllable, public health measures and flu marching along pretty deterministically until there’s a vaccine has to do with the importance of super spreading, uh, and the role of super spreading on that dynamics Now it turns out the COVID is actually sitting at kind of a sweet spot. It’s just a little bit less reliant on super spreading and SARS and Ebola, probably a little bit more maybe about the same level of measles. And what that means from an applied public health perspective is that it can be controlled with aggressive, highly targeted interventions, even without a vaccine. But as soon as it gets a little bit out of control, it basically flips over from us, the Catholic regime into a deterministic regime, and it behaves more like influenza and really cannot be controlled without wholesale

lockdowns, uh, in the absence of a vaccine So one of the reasons that I say that if the world understood, uh, ecological theory a little bit better, I called complex systems theory or complex networks is that we can see right away. We had estimates of our and Kappa about this time last year, almost, uh, that we could see right away that this tells us very, very quickly that this is a disease that is controllable, but only with rapid aggressive, thoughtful, targeted, uh, public health interventions, and very, very quickly, uh, we did some work. That’s now impressed the general real estate interface, where we take the Lloyd Smith at all framework, and we extend it so that we can look at the effects of higher moments on the distribution of secondary infections. And so it turns out that in very interesting ways, uh, the effect of the kurtosis depends on the lower moments. And so as you go higher and higher and higher, you have this dependence on the final size, on the rate of spread on the epidemic profitability Uh, we also look at the effect of stochasticity So we’ll use basically a branching process model to include stochasticity. And this turns out to be really important because one of the things that happens with COVID, so I’ll show you here. So this is the proportion of susceptible individuals infected, uh, at time infinity and an infinite time model, infinite population model. In this case on the left are our estimates that include, uh, the second moment, the variability and secondary infections on the right, uh, is an estimate that just uses the first moment And the variability here is not from uncertainty, it’s from different estimates of the Arnott and capita that have been published in different papers. You see that there’s really two different regimes here, and it turns out the regime on the left is largely dominated by stochastic die outs early on. So about 60% of the time that COVID finds itself in a new population, the chain sputters out and dies before it actually is able to take hold. So the way you think about this is that COVID has to have an early super spreading event before it can flip over. And I’m, I’m obviously anthropomorphizing and massively simplifying what’s going on here, but before it can flip over into the deterministic regime. So the difference here really has to do with some populations where either the population structure or the public health interventions were able to prevent COVID from exiting from this earliest, the Catholic phase and others on the right that close circle on the right is a fishing boat, or it’s a cruise ship, or it’s Manaus where you basically end up with COVID sweeping through the population. And so I’m just going to layer up a bunch of other diseases here, uh, not with the hope of overwhelming you, but instead, what I want everybody to just kind of squint at is look at the red bars, which are all over to the right here and look at how for all these other pathogens where we’re able to pull cut K capita’s are, are not out of the literature, smallpox, flu, flu SARS, MERS. We’ve got a bunch of others in the paper, the red largely overlaps with the rest of them, meaning that, you know, it matters that we take into account the higher moments, and it really matters in some populations, but not like for COVID where you get this big separation between what happens when we account for super spreading or over dispersion and what happens when we don’t. And so it really does seem to be the case that COVID is sitting in this kind of special parameter space where small variability in what it experiences when it arrives in a population can lead to large organizational differences, uh, in terms of the outcome. And again, that’s why I say that the kinds of eco evolutionary dynamics that we’re all studying are so important for bringing to bear on the COVID problem, because these sorts of dependence on spatial scales in this case for COVID, it doesn’t appear as though you can core screen away the effects of lower levels on the emergent properties. Neither can you kind of find like you have to account for the higher level of properties as you go back down again. And these are the kinds of things that are a part of all of our training, understanding of many different disciplines that we can bring to bear on COVID. So then very quickly it turns out that all of this is not even that simple. So if we want to say predict how long it’s going to take for COVID to sweep through a population, we just can’t use the first moment. And second moment of the secondary infections. It turns out that the metapopulation structure that COVID finds itself in is also very important. And this has actually been one of my favorite pastimes of, of the recent kind of five years, is it, it turns out that a lot of individuals that work on complex networks and a physic style approach to epidemiological dynamics are rediscovering things that population geneticists, uh, published, uh, in the early to late 1990s, a lot of which having to do with metapopulation dynamics. And so there’s this kind of interesting, uh, rediscovery of some of the important things from population genetic theory around the funky things that metapopulation structure can, they do, can do to the dynamics of something that’s spreading on the effects of infectious disease dynamics And we’re seeing them play out in real time, uh, with COVID-19. So what do I mean? Well, if we have a nice close population like this, the epidemic curve kind of looks like this it’s mostly symmetric. It’s not totally symmetric, but it goes up and it comes back down. However, if we have this metapopulation structure where households are embedded in

neighborhoods, and we have complex processes that described the ways in which individuals move between these households, we don’t really know what these curves are going to look like important foundational theory from Duncan Watts at all, showed what some of these look like in a simulation model, where you have a hierarchical organization realizing I’m coming up on time here, four minutes, you have a single wave in one population. And as you have increasing layers in the metapopulation, you get this bigger smear. And the smear is actually because of the stochasticity you have N minus one waves are in a well-organized hierarchy, we’re in as the number of layers and very quickly, uh, and I’m just going to fly over this. This is maybe my favorite paper that was ever published by, uh, professor Lisa satin spiel and professor Deon herring, where they use log book data from the Hudson Burford trading company during the 1918 flu to show that 1918 flu behaves this way with respect to metapopulation dynamics, where you get these multi wave outbreaks as a result of the metapopulation structure of the fair trading networks and in what’s now Northern Canada. All right. So I’m going to fast forward here. Uh, I’m going to skip over what I said in Twitter was going to be about, uh, fundamental ecological theory, but it turns out that Lloyd’s mean crowding, uh, is a very strong predictor of the width and duration of COVID-19 epidemic curves over multiple orders of magnitude across countries from China to Italy all over the globe. So if you just take the human population distribution, you use that to calculate Lloyd’s mean crowding It is highly predictive of the shape and duration of COVID-19 curves globally in particular, that large urban areas like San Paolo had these very long, broad epidemic curves as it bops around from deem to deem in the metapopulation that is San Paulo. Whereas in Manassas, you get this quick sir wave light curve, as it sweeps through this relatively well connected, uh, population. It also turns out that, and for those of you that haven’t read this paper by a weight at all, uh, they go through in great and very fascinating detail, all the ways in which Lloyd’s meat crowding is literally related to everything, but it turns out the numerator of being proud. It can be mapped onto the expected number of contacts in a network model, and there’s a direct connection to the are not the epidemic threshold, the outbreak size. And so what we showed is in China and in Italy, uh, it’s highly predictive of the intensity of, of these epidemic curves And so what I’ll do, we did this with some simulation models. Uh, it turns out if you want to flatten the curve, you have to decide whether you’re in a metapopulation or not, because actually, if you’re in a metapopulation and you introduce a lockdown, you compress everybody into the households where the per contact probability of transmission is higher, and you actually get an intensification of the curve. Whereas you only get the flattening if you’re in a strict single layer metapopulation model Um, and I’m going to skip over this. And with that, I will, uh, thank again, the organizers And I think I have a minute or so for questions Thank you, Sam. So where we can take any questions that people have, and we have about a minute, um, I have one, uh, Samuel, when you were presenting it, it’s kind of like you’re talking about like showing the different curves, uh, for, you know, super spreader type curves versus homogeneous populations. Um, it’s like, you’re talking about them as a necessary property of the diseases themselves, as opposed to the current society in which we find ourselves where super spreading, it could be much more likely now because you have, you know, airplanes and big mass, uh, events and things like that So I was wondering to what extent it’s a property of the disease itself versus the society structure that we now have. Yeah, absolutely. So, um, I didn’t mean to give that impression if the, the longer version of this talk, it has a slide that says that these are properties of the pathogen and the system that it finds themselves in, and actually turns out with SARS, the missed estimation of the pandemic risk, which is published by Lauren Ansul Meyers was because they took crowded apartment buildings and they extended those parameters to the rest of the global population. Right. And unfortunately this is still happening for COVID-19 is that really people don’t understand as much as they should, that the population structure matters as much or more, uh, as, as the properties of the pathogen. Well, I, I think I took up the entire question time I’m sorry about that. Um, we have to move on to the next speaker. Uh, thank you, Samuel So, uh, the next speaker is Orlando Shorey, who is going to talk to us about the right tool for the job is my phylogenetic diversification model. Go for forward Orlando. Thank you Um, yeah. Hi everyone. I’m Orlando. Uh, thanks for showing up for my talk. Uh, do some ongoing research from a current postdoc with Emma Goldberg, uh, about, um, model advocacy. Um, so the kind of models and question here, which

we’re trying to address the adequacy of our, um, models that are used to study managed diversification. So real quick, what we want to do is we want to describe whether there are differences in the accumulation of species richness, uh, this, these differences vary in space or time or between languages And we also want to explain what causes those differences and whether maybe attributes of the environment or attributes or the organisms themselves, uh, might have something to do with that. Uh, our weapon of choice here, as you might have realized though, radio models that, uh, allows to estimate speciation and extinction rate dynamics, and we feed those with phylogeny and depending on the model also with, um, all the relational factors, for example, trade, which are thought to maybe affect those speciation and extinction rates Now these, um, models have been very popular and a lot of research has been done across the board, addressing a lot of very interesting questions about what might be affecting diversification in different groups of organisms. But as we went along, it became, we became increasingly aware that there are different kinds of issues that affect these inferences. And if we do not, if we are unaware of what those weaknesses are and we don’t account for them somehow, uh, they could seriously affect our inference results. So what we, um, what we want to do, usually what we do is that we compare our models and we try to go for a model that has the best fit. But the problem is we don’t know whether our best model is still actually a good model for our data to begin with. So what we need is a way to find out whether a model is adequate for a data that is adequate to answer the question of your author. And so therefore that’s what we’re were setting up to do here, developing addicts, the test for those models, and hopefully use the result of this to be able to pinpoint what the reasons are, why our model might not be adequate to have some kind of idea of what we need to do to improve now real quick. I want to give you some kind of intuition about how model adequacy tests work. So let’s assume we have some kind of pilot data here, and you can see the numbers leading up to 12. So maybe you think my 12th standard time, I’d be a great model for this. And of course it will give you an answer, but this is an answer that you can rely on. Well, what we can do is we can roll this time. Any times it’s simulate data, we’ll find this like uniform distribution across cell values. So then we could, for example, look at the distribution and the data and see that we actually have some kind of more of a bell shaped kind of distribution So clearly something is different here. Maybe upon reflection one, like get to the realization, Oh, maybe we should actually use to fix dyes and use those. And if we simulate under those, we might get to a distribution that is much more similar than the distribution you have here in our data. So that is really kind of what we do when we do more to see, we know that data generated under different processes will have different properties. So essentially the data will look different and by comparing, but at the data under the model looks similar to our actual data. We will be able to judge how suitable the model is. It was of course, because we wrote seven, the Rover will come and steal half of our fees, but, uh, this is a story for a different day. So of course the idea of mold tests is an entirely new and equal. I’ve done similar things before, both for stuff and nucleotide substitution models for trade evolution, follow the non-tech models, and even for simpler than diversification models. Um, what we’re employing here is what’s called a post your predictive simulation approach to PTs. And what we’re doing is essentially if we have an ethical tree that we tried to study, and we have a model that we think would be useful for this, we will estimate model parameters for our data under this model in a patient way. So we’ll come out with the distribution of model parameters, for which, for example, will be destinations who were speciation and extinction rates. Uh, now we will draw from these distributions to simulate a load, more trees under this model to get this like, uh, distribution of different fellow

Denise, uh, under this model and Google measure, what we call summary statistics, uh, which are different properties of those trees that describe what those trees looks like, look like we do this, we’ll get different distributions for each summary statistic that shows what most of our trees look like. And let me estimate, if you measure the same summary statistics for our empirical data as well, we can compare it to the simulator ones. And we can see, for example, that some summary statistics will, uh, the empirical data will fall outside of the simulated distribution, which would indicate that our model might be inadequate Whereas if our empirical tree essentially looks the same as our simulated data, that would rather point towards it being adequate So the actual big challenge here is of course, um, finding the summary statistics that show us the relevant differences, right? We wanna, we wanna figure out what really makes a difference here. Uh, and so we are using a lot of summary statistics right now, throw everything at it. Um, that relates to a lot of, um, aspects of allergies that we think are important here, like branch lines, typology, uh, face trait for construction. We’re going to try and figure out which ones of those are useful for us Um, real quick, uh, to have a look at the kind of models that we’re using, outcome based models. So what we have here is that we have, uh, a constant rate model where we assume that there is a trait that has two different States, indeed And we have, uh, a speciation, right, that will generate new languages. And we have an extinction rate that will take lineages away And, uh, you can see on the constant rate model, we will actually not have a different rate between the two traits States. And so the trait does not affect, uh, the diversification unlike for a state dependent model, the, uh, which is essentially the same, except that the rates are allowed to be different between the two traits States. So just to give you some kind of intuition, what that could mean for, um, the trees is that for example, we could have a slow, uh, diversifying green trade green clay, and the faster the bushes diversifying perfectly now, slightly more involved. We have a state independent model, or you can have, are two traits. And we also have two different, um, classes of diversification, but they’re not actually dependent on, uh, the traits they’re separate. So we might get a tree that looks very similar to this one here, where we also have to like a flow and a fast clay, but they are actually referring to the, it like those internal rain clouds And I’m not related to the train. So going forward, I will refer to these as the constant model, the BC model and facility to model just for brevity. So if we simulate the load of trees under models that we know is that we know the answer, and then we run this procedure and we ended up with something like this, where for every tree and the rows, we will end up, uh, with a lot of, um, summer’s synthetics, which will either tell us, Oh yeah, this is kind of looks the same, or raise the red flag and say, Oh, this looks kind of different This might be an inadequate model. And for different pairs of model inference versus generating, we might get different patterns here. Some of them essentially going to do now is I’m going to look for each tree, how many red flags do we have? And these will be the kind of floats that I’m going to be using, going forward, where on the x-axis we have the number of significant summaries that, that tell us, Oh, this might be adequate poetry. Now was at first to see whether this is a reliable, we’re going to look at the case where the generating model is the same as the inference model. So basically we’re asking our models adequate for themselves, which hopefully they would be. And in fact, what we’re finding is, yeah, there’s some outliers and we will have to curate our set of summary statistics a little bit, but roughly we’re finding what we’re expecting, that they’re all like close to zero, which would mean adequate Now, what if we want to run this for, uh, the whole combination of those three models

and what will be there? So if we want to use the concentrate model on those trees simulated under those different models, of course, again, we will expect that the model is adequate for itself, but then we would expect that a for trees generated under a more, uh, complex, complex model that the simple concentrate model would not be adequate at that. It would be similar out here for the state dependent model. Again, we would assume that it’s adequate for itself And since the concert model is nested within this, it would assume that it’s also adequate to describe those concentrate trees, but it will not be adequate for the state independent model, which is more complex. And then finally, we could maybe expect that the most involved model here is able to actually describe all of those trees, uh, adequately. Now, what are we finding drum roll success age. So we’re actually doing quite a, but there are a bunch of cases where we’re off with our expectations First off, it seems that concentrate models are adequate for, uh, our complex trees under, uh, under a state independent, uh, process, which shouldn’t really be the case the same for our state dependent model for those same trees. And finally, it seems that the state independent model is not adequate for our trade dependent penetrates. So what is going on here? Well, essentially, um, but I think is going on up here is that the summary statistics that we’re currently using are not able to find the signature along, uh, one versus two sets of rate categories in the trees. And, uh, we’re really, we can only find those if they are linked to the traits and then kind of highlight about the traits. And I think we just need me summary statistics that allow us to do that. And we have some ideas on both. We might be able to use their finally down here. Um, it turns out this data dependent model isn’t actually nested in the state independent bottle. So if you have a trained dependent tree and you unleash the state, didn’t have animals on, it will actually give you the right answer in a way, because it will possibly identify that the two rate categories are where the two, um, traits are. But of course, if you simulate under this model, nothing has a model So those should be aligning. So we’ll the simulate, the will be all over the place So in a way, this is actually a good thing because it allows us to distinguish between the same pattern versus the same process So now what do we want to raise the bar a little bit and come up with more complex scenarios? So our first two are somewhere we, uh, have trade dependent diversification, but more so for these two models here, we have, uh, two different sets of speciation rates for the two different traits States, but they are additionally also decreasing over time in one case linearly in one case exponentially. Um, and secondly, we have those bam trees that are very messy trees, where we draw, um, range from very different rate categories, like much more than two. And we have an independent, uh, uh, neutral traits simulated over it. So the trades stage would have nothing to do with the diversification rate. The reason we use those is because, uh, or both can Goldberg for those Berry trees that train dependent models are very likely to fall asleep claim that there is trade dependent speciation going on here. So again, about what to expect, we would of course expect the model to be, uh, adequate for itself. And then we would assume that the more we go away from a pure train dependence, the more inadequate we’re getting So more for linear, more for financial independence, and then for the batteries, we should be far up again. What are we finding? Well, kind of, so we can’t really tell apart the linear time dependent, uh, trees from the pure trade dependent ones. We know really, but we can see him for the exponential bonds, both for the bam trees for the complex series. We can actually, we’re doing a pretty good job telling those apart. So what does that mean? First

of this is kind of a success because remember those are the trees where four minutes, four minutes, Fernando, thank you. Uh, those are the trees where the model comes up with the wrong answer a lot. So now they’re able to find those wrong answers, which is a good thing. I would say, not to things that are going on here. First off. One thing might be that we’re getting to the limits of the approach and maybe those trees are just not different enough from trees generated under the actual trade dependent model that we could tell the difference. And secondly, if you think about it, um, the model tells us that there is trait dependent diversification going on. And if you remember those models will is, there’s just not only that going on. So maybe if we explore the summary statistics, but more than I’d be able to tell whether there we have adequacy for the model can describe our data. Exactly. And maybe we adequacy for, uh, we can get the right answer out for our question. So this is a very, to me, a very exciting thing to look into, but just real quickly. So we can kind of, uh, tell models apart. We can detect, uh, some of the false positive inferences. You will definitely have to do some more sensitivity testing and some more, um, development on the summary statistics to be better able to find what the actual reasons for modeling CR, uh, but eventually, hopefully this is going to be a practical tool for people to be able to identify which models they should use and maybe also, which models they should use as an ally processes So they are not comparing their models against an alternative that is clearly off the charts anyway. So I’d like to thank a bunch of people Of course, my advisor and the lab, uh, my current host lab and, uh, an Idaho and people who have already moved on since then, uh, appropriate provided input Alamos national lab and people back, uh, Tennessee and giving you with this meme that actually just came up yesterday at 6:00 PM about this problem Um, I would thank you all for listening and be ready for any of your questions. It’s a great meme. All right. So I think we have a couple of minutes for questions. I was curious to know whether you run into any situations where you kind of get a circularity where your parameters that you use to model your expected distributions kind of circle back Does that make sense? Sure. What you mean by that? Um, we’re not actually modeling the summer Institute mystics that the tribal shape of the trees is. So obviously they will be affected by it. Hopefully, hopefully they will be affected by it because they’re not different between the different models, then they can’t say anything. Um, but I think there shouldn’t be any circularity otherwise, because we’re not actually, uh, simulating the tree shapes, uh, targetedly. Yeah. Let me know if that doesn’t really answer. Yeah. I didn’t think that there would be, but I was curious to know if you would, if you ran into that, we did have moments where we thought about this a little bit, because could sometimes seem like it, but I think would say it’s there. Uh, I think that’s the end of our time. Thank you very much, Orlando Thank you. So the next talk, uh, will be by and it’s about evolutionary rescue under demographic and environmental stochasticity. You did Uh, thanks. Uh, I’m calling you from the university of North Carolina at chapel Hill and my talk is on the. Uh, so is the framework for starting adaptation, which combines both abolition and ecology. So people mainly consider a population that recovers from initial demographic decline through genetic evolution after even to change And here are some examples of emperor average and re-asking even wild. So to illustrate,

uh, I would take the most population, an example So in these bigger, the Y axis is population size and acts axes is a generation. And before the industrial pollution, most malls are wide, but after industrial pollution, the white most will be slapped against and public has dies with decline in the meanwhile, the black will be favored and increase in frequency And after some time the population may rebound to the regional, uh, publishing side and most malls would be black, but when public inside the very low, uh, fluctuation of the population size may leave publishing to go extinct. Therefore, the stochasticity of the publishing size that directly pop patient extinct, but there are two types of stochasticity. Uh, one is diamond graphic stochastic state, which due to a STEM thing, our in videos operating number of, for example, some mother may have a single baby, but some mother may or may have twins or even trap it. The major type is the classic is mental stochastic state, which due to the temporary temperature, fluctuation and main fitness, uh, for example, for the plant population, the reproductive output of the sea, it will depend on the temperature fluctuation upon native surveys, which is influenced by the weather or rainfall and things. The publishing of the Bible rates will depend on its demographic dynamics. There are two major factors that influenced his dynamics. Why is, does lack of intensity? So a stronger action will cause the population to evolve fast, faster, but also will cause more Slack with deaths if the harvest reduction and make the population to initially decline most severely. So if we can draw a prod for the survival probability change with time, a stronger selection may cause lower survival rate initially over the short term, compared to a week and action, but cause a higher survival rate, uh, over the longer term, the other factor is genetic variance. Similarly connect variance will increase the Aboriginal rate, but also impose the business cost under stable lightning production. And for this, uh, Florida survival probably has changed with time. Uh, higher genetic Darren’s may have may lead to a higher survival rate over the shorter term and lowest survival rates over the longer term. So based on these privates analysis, we mainly ask three questions. The first question is out how the staffing tends to influence population survival rate and sack and, and how these facts, uh, differ under demographic and environmental. Stochasticity the stack from classroom doubts, the effects of genetic variance and how it differs under the two types they’ll cast this T and for the third question, uh, I, I, I’m interested in knowing the factors that are crucial for populations, the vibe on the demographic versus environmental stochasticity. Is it due to initial hires, connect battery or initial hire fitness or initial larger patient size that makes population to survive over the hell? Everything we’re asking and these questions I many can feeders the context of phenotypic to action. So in these two panels, the ax backstage is being a ticket value. And in the opera panel, the y-axis is the fitness So you can see that there is an optimum level of phenotypic value that gave the highest of fitness and, uh, and the omega-3 would be nodes, the scans and production and correspondingly the population will always have a frequency distribution of the finished pig value with, uh, tentative diaries of Figma P square, and authors have to change the phenotypic optimum or shift from DVR to the zero plus the video, which drives the population to evolve to a higher phenotypic value. So based on these data, we can write out how the fitness changed with time. And I also assumed as a phenotype again, the genetic variance is constant during the evolution as many priests model deed, but later I would do individual based in relation to relax this assumption. So based on the previous model, we can write out the, uh, stochastic differential equation for publication size change. And the determinants part is representing the first term in these two equations and am is the main fitness population. And

the second term in these equations are this noise due to stochasticity and the key difference of demographic and that that noise term would depend differently on the population size So here’s the readout. Uh, first for the selection intensity on population survival probability, we find that the higher snatching tasks we will always cause lower survival rate. So in this bigger, the Y axis is a survival probability and the act act phase is, is lacking intensity You kind of see that, uh, the smack would probably just decline on Tunica leeway with the latching tans state, but more importantly are, uh, under difference of action tends to your population will have different demographic modes. Uh, so for example, in region one, we have to act in Tanzania is low. The population will, will increase with time. Uh, if there’s no noise, we answered action task to is intermediate in region two. We returned back to the classic demographic curve in which population first decrease and then increase with time. Any readings, three with lacking very strong and the publication looked decline, always decline with time and from the emperor, a perspective to detach the selection, to detect the facts of flashing tends T the key is that we need that the Bible probably that change with the lacking County, which is in this region because in this region on Slack, it’s about a previous one and we cannot attack And if that’s lacking pants, yeah, but when the alarm to shift very small, most of these differential survival probably happened in the third region. In region three, we are popping in size, will always do decline, was time. And it’s only when the image shifts is large. Then most of the differentials probably happens in region two, which is a U shaped curve. This resulting canes that, uh, the average and rescue should not be restrict to restrict to a usual demographic curve breast should be considered in all demographic demographic modes. And furthermore, we also find that under environmental stochasticity, the differential survival probability probability will happen at a whole range of strategy intensity. And which is, which means all the demographics, demographic models, one for the effects of, of genetic virus on publishing survival rates to compare, to compare between different populations We assume that position will have the same interim two variance so that we can use Howard splitty as an indicator of the level of genetic variance I’m here with Claude. That’s the Bible probably that changed with time. So the y-axis is population and that’s have different data of how its ability, which is different novel of genetic variance. And actually it’s this generation and the Connor Raptor, Dan says the Bible property and with find out there exits and optimal NAB of how the ability, which gives the highest as the Bible probability, but this level would change with time. We ended up in the shifts is wrapped in the know this optimal increased time as a white line show, but when the interim to shift large, this optimum level of how people take, take decrease, decrease with time, which means that a higher initial genetic variance gave a highest survival property over the short term. The term used a quick evolution, but the rapid lower genetic directions will get higher spiral quality over the longer term due to a lower fitness load. And to answer his third question of the key factors, adapt cost published and survive under the two types of stochastic fatigue. Aye, aye, aye. I track the genetic and demographic parameters of populations from simulations. And here is a typical figure that I will show later. And in this bigger wax, this is genetic variance and act phase it just generation and the theories of the orange and yellow lines are the Rafika population that, uh, finally survive. And the great theories of gray lines are those that go extinction. And I calculate the average of the, the average from this, the Bible and extinction group, which is the peak, the, by the, uh, bold red and bold black lines. So under demographics they’ll cost the fee we find out the key factor

that makes pupation to survive is that the survival group will have a much higher initial genetic variance. Then extinction group, as you can see from the last panel and this higher initial Jack veteran’s can be due to two factors The first is so Casta in genetics, which is grieved and the stacks and fats are, is because the survival group will have a slightly higher population size than the extinction group due to demographic stochasticity. And in the right panel, it shows, uh, the survival group has higher initial Dem probably in Dan extreme extinction group, but it’s hard to disentangle between these two contributions because they mutually facilitate each other. So a higher genetic variance will cause populations evolve faster. So which means a higher and you show population size and a higher population size, but also makes it means a higher population size also means a drift is weaker. So which leads to a higher genetic variance. But I wonder in mind, environmental stochasticity, we find that the key difference between the two groups are out those, the Bible group will have a much high initial mean girls rates Then that extinction group, uh, is it is showing the middle panel and the genetic variance and genetic balance between the two groups are not quite different as that on the demographic stochasticity and the black and the blue dash lines are the predictions from the model without stochasticity. So the survival group who will also have a higher initial gross rate than the deterministic predictions. So as a conclusion, uh, in this Dar day, we first investigate, we first first investigate the fact of flacking 10 state to find that stronger Snapchat always cause high extinction. And, but on the damn graph, the cast to see differential vagal rails can happen at different demographic modes with depends on both the level of the incremental change and the intensity of seduction and on that in your answer. So Kelsey differential survival rate can have an ad all re all the demographic models. And for the fact that we connect with Ariens, we find that there is existing optimal level of Jack Darien’s, which, which gave the highest probability and this optimal, and actually Aryans will increase the time when the shape is small, but decrease with time when the shift is large And we further to find out these optimal level checks, the RNs will not differ on the demographic and even rental stochasticity, which means that the types of stochastic does not affect the effects of the generic without Renaissance, the Bible. And at last for the key factor guys cost per patient to survive. We find that under demographics, the cost, a high initial and popping and thighs is key. If we’ll pump patient to survive over the intravascular process, but under the a high initial fitness is crucial for us, the Bible. And I want to thank the peers, uh, and faculty members from UNC for their benefit, your comments and suggestions, and thanks for watching, and I’m happy to pay and questions. Great. Thank you. That was, that was awesome. Uh, we have a question from Andrew Hendry, which range of parameter space for genetic variance is closest to empirical best at to empirical best estimates of actual additive, genetic variance and fitness. I can’t answer because, uh, though, so what I do for the Jack variants is that is, as I can tell the mutation pressure. So the Jax membranes would depend on the mutation, how large the mutation pressure is and how many low sides the trade is controlled. So, and yes, so because I’m as theories, I, I think I don’t quite know the entry code, uh, lead

to return, um, the acrobatic generic to the areas. Yeah. Um, we also have a question How were the optimal genetic variances values calculated? Okay. Uh, so four days, uh, for the, for the, uh, demographic demographics, the cast to see based on this past take differential equation, you can grab a, write out the, uh, an expression for us, the Bible priority, and you can derive the Margaret poverty with generic barons We ended the first direct to Barrow. You can gas the highest check balance debt. Uh, you can get the optimal generic burning staff gave the highest the property, but for the you can not write out the survival, right? So I use, uh, Numerica stimulations and I just increased the genetic variance by, uh, various most apps and finds the optimal Dabo optimal genetic variance under environmental stochasticity. So in a related question, uh, can the model incorporate both demographic and environmental stochasticity? Uh, yes So to incorporate bows, you can just add a third terms, a four as a nice term, just combine the two terms. And actually in, uh, in the real population, a population will have both the demographic and the two types of the calcium would be further. And, uh, when they are both, the cast is T the, uh, even stimulation. And you find that it’s going to be an intermediate between both like when there’s only demographics is key for the Bible, but when you add in, but it was a higher track to barriers and a high initial growth rate will be key for survival, but they’re in quarters with, depending on the proportion of demographic, any month of the Cassidy. All right. We have a couple more questions. Um, did you consider competition among species in this framework and with differential genetic variances? No, I, I, I can feed are just wild, wild population, so there’s no competition. And then, um, how did you track the evolutionary dynamics of the species? So for stimulations, I just tracked the calculates, a genetic variance, uh, main fits knees and pop and rack core, the puppy, hand size, uh, every generation So, so that I can track the connection and demographic parameters. Yeah. So basketball motto, uh, you can drag the guides out the dynamic equation, so you can know the fitness as the Bible property. Cool. Thank you. I think that’s all the time we have for questions Okay. Thank you. Now we’re going to go to a prerecorded talk, uh, by one of our European colleagues, Matt barber, well, currently European, uh, and so we’ll play that, uh, Matt doesn’t appear to be here, but if one of his coauthors is, of course you can speak up and, uh, maybe we’ll have some questions for you. Uh, so we’ll go ahead and play that video. Now, this is Mike Barber talking about a Keystone gene underlies the persistence of a food web. Hi everyone. My name is Matt barber. I’m a, post-doc at the university of Zurich, and I’m also an associate editor at the journal of animal And today I’m going to share with you some work that I’ve done in collaboration with Dan steam steam. the looks at the role of genetic variation within species and shaping community stability. So one of the grand challenges that if you will, we face today as a is isn’t understanding how biological processes scale from genes, all the way up to complex ecosystems And over the past 30 years, we’ve seen some fantastic examples of these cross scale linkages Much of this body of work was nicely summarized by Andrew Hendry and his book on eco evolutionary dynamics. Now, while this framework has produced some fantastic examples, demonstrating these linkages, I think that it actually hinders our ability to not only understand, but predict how these biological processes scale and that’s

because at least, especially at the community level, studies often treat a community as a collection of species that can be well-described by its richness or composition, but we know, but what we know is that a community is far from a collection species, but that it’s structured by interactions between species and the strength and organization of these interactions that determine its dynamics. And so one of my favorite examples of this comes from Bob pain’s classic experiment, where he showed that removing sea stars from an inner title, food web altered the structure of feeding interactions resulting in just a few species out, competing others for limited space and then title. And so this experiment showed that C-stores Keystone species and that they determine the structure and stability of this eclipse community. And so rather that rather than approaching the problem like this, what I think we need is a more explicit view of how genetic variation within species structures, interactions in a community context. So there’s an experimental food web that I’ve developed at the university of Zurich, each, each circle corresponds to a population of the different species. And the DNA ring represents the fact that there’s genetic and phenotypic variation within these populations with two arrows, between with arrows, between species representing the interaction, solid arrows or positive effects and dashed arrows are negative ones. And so what I hope this diagram makes clear is that any property of a community such as its richness or composition emerge from this network of interactions between species. And so what we need is a more explicit view of how genetic variation shapes interactions between species in a community context. And so for the rest of my talk today, I’m going to show you how I’ve been using an experimental approach to understand how genetic variation within species scales up to sh a shape, the structure and stability of a food web. And to do this, I’ve been using the experimental food web, which is a subset of the one that I showed you earlier at the base of this food web is a rabid Opsis Italiana. And it’s fed upon by two different species of aphids that are in turn attacked by a parasitic wasp Now this food web is naturally associated with the rabid opposites in nature. And one of the reasons why I decided to rabbit Opsis is because we have a lot of information about the genotype to phenotype, um, map in this species. In other words, how genetic variation shapes ecologically important phenotypes. One of the phenotypes that we suspect is important in this system is the, um, is the secondary metabolites rabid Opsis produces called glucosinolates. Now you may not be familiar with these compounds, but I’m sure you’ve tasted them. Uh, it’s what gives cabbage and brussel sprouts, their bitter tastes, and also that spicy flavor to mustard and wasabi. One of the fortunate things is that the genotype to phenotype map of these glucose scintillates has been rather well characterize. And so there’s a simplified schematic of what this, uh, what the biosynthetic pathway looks like for these glucose and like compounds. So as you can see, there’s three key genes, ma’am one and JSO H the determinant, much of the qualitative variation in these chemical phenotypes. And so what I did is I leveraged the existing natural accessions and transgenic lines that knock out, um, these key genes along this biosynthetic pathway to determine much of the qualitative variation that we see in, um, in this rabid opposites, uh, chemical diversity that we actually see in nature. And so by using these different genotypes, I can isolate the effects of, uh, Lila differences at each one of these genes on interactions. Okay, well, what I did using these four different genotypes, I created experimental plant populations that either had just one, two or four different genotypes And by creating this gradient of genetic diversity, this allowed me to not only test the effects of specific illegals at each one of these genes, but also more general effects of genetic diversity on the species and tractions And so I grew these plants for four weeks in the greenhouse before adding the experimental food web. So here’s a short little video of me adding the aphids to the Sood web. I did two adult, uh, aphids of each one of the species to each of these pots and what this allowed And then I allowed these Asian populations to grow for two weeks before adding the parasitoid wasp. So you can see I’m just adding the final eighth into this pot before I add it to this experimental cage, which is where we’re going

to be tracking the population dynamics of this, of these experimental food webs over time. He was just a video of one of the 60 cages, um, experimental cages that we created And so together we had 60 experimental food webs across to two different climate chambers And one of the reason, one of the reasons we took advantage of these different climate was to simulate how the effects of warming might alter the effects of, uh, specific Leos and genetic diversity on, on the food web And so in one of the chambers, we kept at 20 degrees Celsius, which reflects the daily mean temperature during the summer and this part of Switzerland and the other chamber we held at 23 degrees Celsius to represent the warming. We expect these insects to experience in the next 25 to 50 years now, just to let you know the, for the, my results, I’m just going to focus on the genetic effects, importantly, that they were robust to this experimental warming, but I’d be happy to answer any questions about, um, the effects of warming later. So every week we would take out the cages and we count the abundance of aphids, as well as the parasitoids And so for the parasitoids, we would count the adult individuals that are flying around the cage. And we’d also look for these mummies, which you can see are these, these little Brown shells, and that’s what the adult or pairs of joints turned the aphids into before they emerge. And then we repeated this procedure for four months tracking the population dynamics of all three species. And so for the first series of results that I’m going to show you, I’m going to, I’m going to focus on food web persistence. And this is so we’re gonna look at how genetic variation and specific, uh, Leos each of these genes affect food, web persistence, which is the proportion of species surviving until the end of the experience So in this graph here, each of the gray circles corresponds to the average, uh, food, what persistence for particular genetic composition, which we have 11 in total. And what you can see is that with increasing the number of genotypes in the plant population, increasing genetic diversity, increased food, web persistence, and for every genotype at increased food web persistence, by as much as 38%, but we’re also interested in the effects of, uh, specific genes. And so in this graph here, the dotted line represents the effect of the average Lele on food web persistence after we control for the effects of genetic diversity. So what you can see here is that for man wanted JSO H different illegals at each one of these genes has typical effects on food. What persistence, in other words, it doesn’t seem to differ from the average effects, but in contrast, a opt to, depending on which a Lele is in the plant population as a strong effect on food web persistence. And in particular, if we add a Knoll to the plant population, increase food web persistence, by as much as 28%. So analogous to Bob pain’s experiment, where he showed that sea stars function as a Keystone species. What we see is that appears to function as a Keystone gene in this food web, in that, depending on which a Lila is present in the population determines the stability of this, of its associates, what that’s an interesting result. But right now we don’t really know exactly how is affecting the food web. So to kind of dive into this a bit more, I looked at critical transitions in the food level over time. So a critical transition refers to a local extinction that simplifies the food web and this diagram on the right shows you all of the critical transitions that we observed in this experiment. And so I looked at the effects of on each one of these critical transitions. And so in this graph here, this shows you the effects of adding a functional on the risk, uh, on the risk of each one of these critical transitions. Okay. And the main thing I want you to see here is that all of these arrows between the critical transitions is that they’re all gray, which means that there were statistically unclear effects for nonsignificant effects. But instead when we add a functional of the population, we see that it alters the risk of a critical transition in the dominant food chain with just the aphid

and one, or would just one of the aphids in the parasitoid. And it reduces the risk of this food chain collapsing, too. Um, so that there are no insects by 66%. Okay. So now we’ve identified where AI tool is acting in the food web, but we still don’t know, okay, how exactly is it promoting food, web persistence? So to do this, I leveraged theory on the structural stability of ecological communities. And the reason I did that is because depending on if we have information on how species, their interaction interacting, as well as their intrinsic growth rates, this enables me to make predictions for whether the system will persist or not. And so I use my, uh, the data we had on the population dynamics of all the interacting species to estimate the strength of these interactions, as well as the intrinsic growth, right? This next graph on the X axis, we have the intrinsic growth rate of the aphid. And on the Y axis, we have the intrinsic growth rate of the parasitoid The gray area represents the range of intrinsic growth rates that are compatible with the aphid and parasitoid coexisting. And this gray area is determined by the interactions between the aphid parasitoid. And so what you can see here is that rather than altering, uh, this gray area, it increases the intrinsic growth rates of the aphid and the parasitoid okay, which are denoted by this arrow. And so if there’s a in the plant population, it increases the intrinsic growth rates of the aphid and parasitoid, and moves it into the region where they’re both able to coexist, okay. Compared to if there’s only a functional in the population. So this is what appears to be this positive effect on aphid and parasitoid intrinsic growth rates is appears to be what’s underlying the persistence of this food chain And now for AI to what we think is underlying its effects on the aphid and the parasitoid is simply ineffective. on plant growth rates So these are the results from an additional experiment, except we didn’t add any insects And we did it over a much shorter timescale, but otherwise everything was the exact same as before. And what you can see is that adding annoy up to a Lele to the plant population and a strong effect on plant growth rates. And we think that this pleiotropic effect of on plant growth is what’s underlying these positive effects on aphid and parasitoid intrinsic growth Okay. So to wrap up here, what we’re seeing is that genetic diversity within the plan increases from wet persistence, and that this is determined by a single Lele and a single gene that increase the intrinsic growth rates of species across multiple trophic levels And importantly, these effects were robust to a three degrees increase in water. Okay Now, one of the, I think dire implications of this result is that if there is a Keystone gene in, um, a natural population, and given that the fact that we’re losing, losing genetic diversity at an alarming rate, that this could actually cause abrupt and catastrophic shifts in the functioning and persistence of ecosystems, okay. It makes them potentially they’re much more susceptible than we currently realize, but I also think that there’s some more hopeful implications. For example, we could leverage this information on how genetic variation shapes the structure and stability of a food web to leverage this in assisted migration, where we are not only interested in optimizing or promoting the persistence of species and future climates, but also fostering bio-diversity in these future clients. And we could even think about how this, uh, this general approach could be used for, uh, for breeding novel plants that optimize both the crop yield and bows and diversity, if we want to create more sustainable agricultural systems. And with that all take any questions Thanks so much. Okay. I think that, uh, Matt is not here to answer questions, which is quite reasonable of course, because I think it’s a 2:00 AM where he is. And also, um, his co his coauthors are there as well. So you’ll just have to shoot him an email, check out his paper on bio archive. And, uh, I’m sure he’d love to have the feedback. And in the meantime, we have a few minutes before the last talk, which will be by my colleague, Anna Hargreaves, uh, and, uh, Michael have an announcement here, I think, uh, to help us with that, that time. Sure. Uh, hi everybody

Um, uh, we have an unplanned event on the schedule that we’ve just set up and I put it in the chat for everybody a while ago Um, but we, we set up, uh, an extra zoom session that will start immediately after this, uh, uh, contributed papers session. And we’ll go just for just under an hour and we’ll stop right at six. So I don’t, uh, uh, get in don’t clash with the trivia thing tonight, but for about an hour, we’ll have a zoom room set up where it was kind of a virtual, uh, social, uh, bring a polar bear, but, uh, like Andrew has there and, um, uh, just hanging out and we’ll probably we’ll set up some breakout rooms, so we can have the option of moving into smaller rooms if there’s enough people that want to do that. Uh, but, uh, sorry for the last minute notice, but, uh, it’s one of the suggestions that came through today is that it’d be nice to have more social times though. We’re, we’re trying to do that. Uh, the second announcement is the, uh, it’s too late if you haven’t signed up before, but if you have, of course, remember that the natural history trivia run by the ASN, uh, grad students is tonight and, uh, that’s, um, going to be awesome as it always is. I’m sure. And then I actually think Anna had an announcement she wanted to make, uh, about the, um, well, so I’ll tell you, uh, hi everyone. Uh, so this year I am heading the committee for the ASN student research awards. And, um, usually they’re due January 31st at the end of the month, but because it’s been such a wild start to 2021, we thought that, uh, we would extend the deadline by a week. So, um, if you’re a student and you’re planning to submit an application, uh, definitely do it. Um, and, but you have an extra week to do it. And so if you’re a PI, um, just let your students know. I think we’re going to announce it officially tomorrow Um, and the website should be up to date tomorrow as well. And Judy commented in the chat. I just passed the piece into there too, that she just signed up for the trivia night. So as, uh, as the president elect, she may have got special treatment who knows, but, um, uh, she said she just signed up for the trivia night, so it may be possible to still to still sign up. Uh, so, uh, if you’re interested in that and haven’t signed up, check it out and speaking of the graduate student, there’s the store and you can buy merch. I received my shirt today, so it lots of nice designs So I it’s on red bubble, I think in the link out, sent out a few times. Uh, and I think the funds go towards supporting initiatives by the graduate student council. All right So support diversity initiatives. I think they get a nice shirt and support diversity initiatives that seemed like a win-win right Uh, Kelsey tells us that they’re accepting everyone. So, uh, for the trivia right back to that. Um, so it’s still possible, I think, and since we probably don’t have like a, a final wrap-up session, um, it’s a good opportunity for me to remind everybody, uh, that membership in ASN, uh, affords many benefits, uh, both just from a social perspective, but also to an interaction networking. And it’s extremely cheap, particularly for graduate students. So I strongly encourage graduate students to sign up for it. Um, back in the day when I was younger, I got purchased a lifetime, uh, membership, uh, which will pay itself off, I think, in about 18 years. So if you plan on being in the business for awhile, it’s a great way to not have to worry about to renew your subscription. So, uh, the younger you are, the quicker you can get, uh, you can get your money back. Okay. Um, are there any other quick announcements? Okay. Okay So we’re going to go straight into a talk by, uh, Anna Hargraves from McGill university, my colleague, and she will be speaking to us about geographical patterns and the importance of biotic interactions, Anna. Okay. Okay So everything’s showing up as it should. Great Okay. Thanks everyone so much for coming to this very, very last talk and what has been a really wonderful conference? Um, I don’t know about you, but I really welcomed a change from the new cycle of the last week. It’s been really nice to think about science instead of everything else. So today, instead of telling you, um, I’m just gonna see if I can move this. There we go. Instead of telling you about several results from one study, I want to share with you a one result from each of three studies that each touch on this question of whether there are predictable geographic patterns and the importance of species interactions. Okay. So we know that interactions between species can be powerful,

ecological and evolutionary forces. We know that they can drive population dynamics, um, and population cycles. We know that they can strongly affect individual fitness and that by doing so they can select on the traits that are involved in interactions. Uh, we know that when there’s spatial variation in interactions, this selection can lead to local adaptation where organisms adapt to, um, their local biotic environment. And we know that this local adaptation can in fact, eventually lead to speciation. So one of the world’s most famous examples of adaptive radiation is driven by a biotic contraction of food availability. And finally, we know that, uh, interactions can also scale up to define species distributions. So the example here is from MoMA in Quebec where sea predation was the single most important factor determining the range limit of sugar Maples. So there’s no question that interactions can be really important, but incorporating them into a systematic understanding of spatial patterns in ecology and evolution has been tricky. So if I are you, uh, want to see how some, a biotic feature of the environment relates to whatever it is we study, uh, you can download, um, really fine scale. Long-term incredible data on a biotic variables, uh, in a matter of minutes. And, um, this is because for a biotic variables, we measured them in standardized units. We often have standardized equipment and ways of measuring them. And we increasingly do this at a huge, sometimes global scales. And there’s just no equivalent for systematically measuring the strength or importance of biotic interactions Uh, in fact, you know, as a community, we would be pretty stoked if we could even just map where individual species occur in the world. Uh, and we’re not even quite year there yet let alone trying to map their interactions So while we have lots of compelling case studies that illustrate that interactions can be really important, we have less predictive ability of when they’re important. Uh, in other words, we also don’t know when we need to go out and collect the data on interactions to inform our ecological and evolutionary predictions So this leads to this question, um, that I have been thinking about on and off for about, uh, 10 years, at least. And I know that many of you, um, in the audience think about it as well, and like any good question in ecology and evolution, we are not the first people to be intrigued by it. So Doug Schatzki in a famous paper called evolution in the tropics proposed among other things that species interactions are more intense towards low latitudes. So in tropical ecosystems and therefore more evolutionary evolutionarily important in the tropics in a similar vein Darwin and no less than the origin of species propose that species interactions are more intense, both towards low latitudes, but also towards low elevations And that they therefore, uh, are more likely to limit the low latitude and low elevation edge of the species distribution. So to, uh, Seminole thinkers, among others who have predicted that interactions are predictably more important, uh, um, tropics and lowlands. So I’m going to tell you about three vignettes. Um, each of these is a separate study, a different dataset, and I’m going to present you one result from each that touches on this question So first from an ecological perspective, uh, where we will test whether interaction strength does in fact become stronger towards the tropics and, um, towards low elevations. Second, from an evolutionary perspective, uh, we’ll look at how often interactions drive local adaptation and whether, um, in line with dub Schatzki’s prediction, they drive local adaptation more often in the tropics and in the temperate zone. And finally, from a biogeographic perspective, we will, uh, look at how often biotic interactions influenced species distributions and test Darwin’s predictions that they should do this more often at the low latitude and low elevation edge of species ranges. So the first study I’m going to tell you about it’s a standardized experiment and what’s different between this and the other two things is that here we actually collected the data in the field ourselves So the interaction that we used as seed predation, see pronation is a great interaction for this purpose, because it has really strong fitness and demographic consequences for plants. And also because seeds are stable, so you can ship them around. Although it turns out this is actually not as easy as I first thought it was when I set up this experiment. But if you persevere, you can

in fact ship seeds internationally, which means that you can use the same biological material at all of your experimental sites So, uh, the experiment that we did was really simple. We would go to a site and we would set up 30 depots of seeds using one of these two agricultural species. And then we just come back in 24 hours and see how many have been eaten or removed. The trick is replicating the experiment, the biogeographic scale, that’s relevant to this hypothesis. So every triangle that you can see in this map shows you one mountain where we ran the experiment in is a transect transactive sites spanning, uh, 800 to 2000 meters of elevation. And so this enables us to disentangle the effects of latitude and an elevation when does not do this type of experiment alone. And I worked with a really fantastic group of collaborators from seven countries and many invaluable field assistants to pull this off. So the first result that I want to show you is just, um, to highlight the, uh, large amount of variation in the final dataset. So here, I’m just showing you spatial variation. We also had temporal variation, um, as we repeated the experiment at different mountains, each line in this messy figure shows you the elevational pattern at one of those mountains. And the take home message I want you to get from this slide is that there’s no point testing for global patterns at local scales, um, at the local scale, there’s just way too much variation And so you really need to be testing these patterns at the appropriate scale. Um, and because we had such a good spread of data, we were able to sort of Pierce through this variation and find that there was actually a really consistent pattern that supported the prediction. Um, Betsy predation was strongest in the tropics. So this is low latitudes and decreased towards high latitudes up in the Arctic. Uh, this was true whether we measured total seed predation here or prediction just by invertebrates and the effect size is about 18%. So who you as a seed are 18% more likely to be eaten, uh, on, in one day in the tropics than you are in the Arctic. And I’m not showing you the data, but we actually found the same effect size for elevation. So, um, an 18% decrease in sea predation from sea level to higher elevations at the time that we started this experiment, it was the most ambitious, um, test for gradients in interaction strength And, uh, and we got a really robust signal that interactions at least seed predation is stronger at low latitudes and elevations is predicted. And there have been a few studies that have come out, um, in the meantime, also running really big standardized tests of this, uh, hypothesis. And so far, the data generally supported in at least terrestrial ecosystems, at least for consumption interactions The next study I’m going to tell you about, um, is, uh, looking at local adaptation. So for this study, um, we did a meta analysis of common garden experiments, um, to refresh your memory. Uh, local adaptation is the idea that if you have spatial variation in the environment under the right circumstances, populations will adapt to, uh, increase their fitness in their local environment. And the prediction is that if you then move individuals from different populations to a common environment, you expect the local genotype to out-compete the foreign genotypes. Uh, and we have, uh, decades of this type of common garden experiment But of course, everybody does this experiment in slightly different ways. So some people will do their common garden experiment in a completely natural setting. And these studies are really testing for local adaptation, the full suite of a biotic and biotic factors that, um, that individuals experience, but it’s also really common for researchers to alter the environment for their transplants in a way that, um, usually increases transplant performance, uh, but in a way that often reduces biotic interactions. So for example, if I was going to set up a C transplant, I might clear away all the local vegetation first, um, which would reduce competition So if biotic interactions are commonly, the thing that’s driving local adaptation, then we should expect to find a stronger signal of local adaptation in these studies that leave interactions intact and a weaker signal of local adaptation. And these studies that remove some of those interactions. So in this meta analysis, uh, we did way more than I’m

going to tell you about here. Um, we did look at the strength of local adaptation as well I’m just going to tell you about our data on the frequency of local adaptation. So, um, we deemed local adaptation to be found as long as the local genotype outperformed the foreign genotypes and to test a Schatzki’s prediction. Uh, we divided the data for this analysis into data from temperate zones, uh, or the temperate zone and, um, data from the tropics. So first we’ll look at the temperate data. This made up more than 90% of the data in our dataset. And we find that when experiments are done in natural settings, we see a strong signal of local adaptation. So local adaptation is detected in a more than 50% of the times that we test for it. And so of course our prediction, if interactions are important is that we should see a lower frequency of local adaptation being detected in studies that remove these interactions. And this is totally not what we found. So, uh, no matter how we slice the data, um, certainly for the temperate zone, there was no indication that biotic interactions were generally driving local adaptation. This result is super surprising, um, both because of our own personal like beliefs and the power of local adaptation as coauthors, but also because within this dataset, we, um, could detect a strong fitness effective interactions in the temperate zone. So we have this weird result where, uh, interactions are strongly affecting fitness, but they don’t seem to be systematically driving local adaptation Hi, if you can’t think of anything else to ask me about in the question, period, asked me about this one. Um, it’s still a bit of a puzzle, so, uh, we find the spear pattern, but we also find that there’s an interaction between the latitude zone, where the study is done and the effect of removing interactions on the signal of local adaptation. So when we look at our tropical data, the first thing to point out is how little we have. So we did do our literature search in both English and Spanish, but we still only found 13 studies that met our inclusion criteria. So we really don’t have the type of rich data that we need to make robust conclusions about local debt at local adaptation in the tropics that said from these 13 studies, we see much more of the pattern that we expected where studies that leave interactions in Intacct are detecting local adaptation more often than those that are ameliorating them. Although this is not significant, obviously our data suffers from a sample size, no problem. So, uh, we got a surprisingly weak signal of interactions affecting local adaptation overall. Um, but from the data that we can find it does seem to do so more commonly in the tropics as predicted. So the third study that I will tell you about is, um, this assessment of how often biotic interactions contribute to species distributions and whether they do so more often at species warm range limits. So this study is a systematic review of the causes of species, um, range limits, where we scoured the literature for any study that assessed any biotic or a biotic factor and its importance at a species. Cool. So high latitude or high elevation or warm, so low latitude or low ILO, uh, low elevation range limit. We found, uh, about a thousand assessments, um, of each type of range limit The data are, um, relatively well distributed geographically. Although there are definitely, um, more dense in the area where the data are almost always more dense than meta-analyses Uh, and we found about 500 tests or assessments, the biotic factors, and almost three times that many assessments of a biotic factors So, um, as a community of people interested in brains limits, we are spending way more time and effort assessing the importance of a biotic factors, um, which why I would argue, we understand relatively well versus the effect of biotic factors, which I would argue we understand relatively less well. Um, so what I’m going to tell you about today is just this result. So just the results for the biotic factors. And so here on the Y axis is the probability that biotic factors influence a given range limit. So we, um, for every factor that was assessed, we just scored it as influencing that range limit or not according

to the study results. So first I’m going to show you just the complete data set, all different types of, um, field experiments, lab experiments models. And of course the prediction from Darwin is that we should be getting a stronger influence of interactions in the warm range limit, uh, and, uh, less a weaker influence of the cool range limit. And unlike the last study, that is exactly what we find. So, uh, um, interactions are more important at species warm range limits since BC’s cool limits Um, and in fact, interactions are supported more than 50% of the time that they’re tested at warm limits, um, as a field biologist, uh, I’m well aware that not all types of studies are equally able to establish causation. So we also looked just at field experiments that are, um, less reliant on correlation, um, to, to assess the importance of different factors. And we find, um, the same pattern, but even stronger. So, um, there is a ton of work left to be done on this question Um, but I think, I think these initial results, um, are really encouraging, um, in that we have these different datasets, um, different methods, but they are so far relatively consistent in that. Um, we are seeing a signal that interactions are predictably stronger in some ecosystems than others. And this is nice because it gives us a starting place for where we should be looking to, um, to collect the next generation of data. This is just a plug for students in my lab that are working on a related questions. You can find out more about their projects on my website, and if there’s any time left, I’m happy to answer questions. Great. Yeah, we have a couple minutes left. Um, so we have a question from Easton, uh, how often, how often does local adaptation occur if it’s not? So how often local adaptation occurs is not affected by latitude, but the strength of local adaptation that would occur already is affected by latitude or another words. Um, so whether or not local adaptation occurs is not affected by latitude, but the strength of local adaptation that does occur is affected by latitude. I, so I’m not, um, I think that might be just a request for clarification. I’m not sure, but what I can tell you is that, uh, I didn’t show any data, but the strength of local adaptation, but we don’t. Um, we don’t actually pick up the same signal, uh, in the measure that we used for strength. So we get that difference between, um, the tempered zone and the tropics is only significant when we look at the, um, the frequency of local adaptation. So, uh, and again, I think this really comes down to sample size, um, because our ability to pick out patterns is limited by the amount of tropical data that we have Did you notice any sort of indication that there was publication bias in terms of when people found local adaptation? And I know you had the slide though, is like, if you have nothing else to ask, ask about the slide, but that’s, that’s kind of what I was really curious about is, and that slide, it seems like that could possibly explain the pattern or at least, yeah. I mean, so we did the standard, like meta analysis funnel plot, um, and we didn’t detect a publication bias in, you know, in studies that were, um, more likely to be published if they found local adaptation or found stronger local adaptation. But I thought about this a lot, and I actually think that we are super biased as a community in when and where we look for local adaptation. So, you know, if a grad student comes to you and says, I really want to work on local adaptation, I’m not going to tell them to go pick two random populations. I’m going to tell them to go look for, uh, populations where we think there’s an underlying difference and test local adaptation there, and the easiest way to see an underlying differences along in a biotic gradient. So, um, I think there’s, I don’t think there’s publication bias once the work is done, but I think there’s a huge bias in how we select systems. And so, you know, there’s lots of studies that say, look, a lot of teaching is super important. You know, it’s like 75% of the time. And I think that the like caveat that we have to keep in mind is like, we find it seventy-five percent of the time where we think that it already happens. Okay. Um, that is the right time to call an in to fantastic talk by Anna. And I’d like to thank everybody

who contributed to this session, a series of really fun talks. And it was great to use them as a lead into, uh, the impromptu beer zoom session that, uh, Mike has sent a link to in the comments and presumably on Slack as well. Uh Mike’s un-muted so I think he has something to add to that. No, no, just it’s on the Slack and it’ll of course the chat will disappear in a couple of seconds So, um, hopefully we’ll see folks there for a beer and then it will be into the natural history trivia thing. So once again, thanks everybody for participating in the conference, especially to Mike, Mark, and other incredible organizers of this insane amount of work went into this, uh, very much unheralded, even though I’m heralding it. Now, it should be heralded much more as I know how much it must take. So thanks everybody Thanks. All the moderators. Thanks to the speakers, technical coordinators and just the people who had the nerve to, to say we were going to do this thing. I think that’s incredible too important to recognize them as well as Susan Dan and everybody else involved Okay. Um, by everybody’s session ends go over to this zoom beer session, uh, and I will see you there at some point. Bye bye