Open bradduthie opened 3 years ago
One thought is an analysis of the relative merits of spatial vs temporal heterogeneity in delaying / preventing resistance evolution. Following our discussions, it would be fair to predict that the answer will be sensitive to geneflow - and that (I think) under infinite geneflow temporal variation in biocontrol will be superior?
One pedantic thought on your text Brad, is that for our fungal biopesticides the larvae don't eat them. The mode of action is all by spores germinating on the cuticle and penetrating the cuticle. (though there is some speculation that Beauveria spores may also be able to have effects through the gut, but this is not confirmed). I think this is just a 'language' thing in the description of the trait because if you replace 'consumption of biopesticide' by 'sensitivity to biopesticide' then I imagine it works just the same. Obviously if you wanted the model to investigate transgenic BT plants then consumption of the active ingredient would be spot on.
One question we could investigate would be the extent to which variation in the magnitude (and direction) of covariance between resistance to different biopesticides / performance on different crops influences the effectiveness of heterogeneity in delaying resistance evolution / maintaining pest suppression. This then validates our approach - how strongly negative do genetic covariances need to be to make landscape heterogeneity worthwhile; AND how much evolution in the g matrix (and therefore alternation in the covariance) is necessary for the heterogeneity benefit to break down.
A comment, rather than a question for testing. However, I wondered to what extent is the outcome of the model runs sensitive to population size (and the balance with mutation rate). Is the outcome constrained by the flow of novel mutations? This might be relevant both for considering the minimum population size needed in the simulations, and what the biological effect is of supressing the pest population to small population sizes? Can this be used to ask interesting questions about the relative effects of de-novo mutation and pre-existing genetic variation in resistance evolution?
Would it be possible to 'test' run the model by modelling the impact of 'standard' refugia (eg GM) on resistance evolution? (or is this tricky due to dominance issues?).
We could look at varying the number of susceptible immigrants per generation to see if there is a threshold required to maintain susceptibilty in the entire population.
We could vary the likelyhood of laying eggs on the same crop that each individual fed on as a child to see if persisting crop preferences change the distribution of resistant and susceptible genotypes. This could be good as an additonal layer over the spatial vs temporal heterogeniety simulations.
The success of H. armigera populations is in part due to their broader host range of and persistence in cropping areas from year to year. Is something we can try play with? If there are alternative hosts available in the area, with growers using alternative pesticides: how might this impact the development of resistance? Will the population experience a’ holiday’ from the fungal biopesticdes selection pressure?
I wonder if we can input or play around with the timing sprays for when H. armigera larvae are feeding or moving in the into protected feeding locations like the flowers, or pods or cotton bolls. I suspect the growers already choose pesticide application times that maximise this. I’m unsure if Brazil has restrictions on pesticide application timing as occurs in Europe but my guess is its unlikely.
Again, timing sprays to ensure H. armigera larvae are at an appropriate size to be most the most susceptible stages to the product. Or can we put in a request to reassess if a spray application is delayed more than 2 days.
I also wonder if the economic damage of a plant population is important? Is there an option to only spray if the value of the crop saved is more than the cost of spraying?
Can we include the natural enemies in the model? It’d be nice to show we’re aware of the presence of beneficial arthropods in the crop, and factor their likely impact into management decisions?
I think may have spoken about this, excuse me if we have. Overwintering of H. armigera pupae and the carryover of insecticide-resistant individuals from season to season. Some regions in Australia carry out pupae busting.
An option during the non-growing-season is to remove all alternate hosts in and around crops can prevent the build-up of H. armigera.
Many thanks for the feedback @MattTinsleyStirling @rosemckeon and @RosieMangan -- I'll work through these comments one by one, but first to follow up on one of the questions @MattTinsleyStirling had at our meeting regarding the partial effect sizes of each locus on each trait. I should have realised right away that we already have this information, for free, with each call to mine_gmatrix
. It's the fifth element in the output list of the function. It also should have been intuitive to me that the effect of each loci on a trait should take a half normal distribution. This is difficult to see with a small number of loci, but I ran build_gmatrix
with 256 and 4 traits. Here's a histogram of the results for each trait.
Am I correct in thinking that if you were to do this for a real quantitative trait that you would get a similar shape, but likely with a much longer tail?
In theory you might get a good tail, but in practice these are probably hard to detect. See this figure from UG text on evolution, concerning QTLs for sternopleural bristle number in Drosophila:
Huh, thanks @luc-bussiere -- those actually look quite reasonable in relation to the simulations. I should add that I took the absolute value to show the effect sizes on the x-axis in the figure above. Am I too optimistic to think that the Drosophila QTL that you show looks about right in terms of shape?
Nope, @bradduthie I agree, this is not too optimistic. Your sims look about right!
Thanks @luc-bussiere -- always exciting and a bit amazing when the model finds the reasonable pattern by itself!
Just going through my initial thoughts on each of these. I'm going to do a bit of work on the package today and try to add an option for immigration in the simulations, then maybe a bit of work on the documentation.
I agree with @MattTinsleyStirling on ideas for spatial versus temporal heterogeneity, and this is probably relatively straightforward to set up. Geneflow can simulated by both immigration and pest movement. We could simulate both spatial and temporal heterogeneity, the former being just a static landscape fo crop and pesticide options that do not change, and the latter being a uniform crop and biopesticide application across the landscape that changes every tau timesteps.
Good point @MattTinsleyStirling on the mechanics of the biopesticide. I agree that this could just be a language thing, but maybe my assumption about the positive correlation between biopesticide accumulation and feeding rate isn't correct then?
I like the question about how the strength and direction of covariance affects the efficacy of heterogeneity and crop or pesticide rotation. I'm not even sure what the shape of the relationship would be, and I think that alone would be worth exploring. E.g., is the curve concave down (reasonably effective at low magnitudes of correlation, with diminishing returns) or concave up (not very effective at all until you get to high magnitudes of correlation, but then rapidly improving). Or is the relationship linear? It would be good to think about this a bit in advance to form some kind of intuition, but simulating this just to know, as a starting point, the nature of the relationship between genetic covariance and effectiveness of heterogeneity would be super useful. This would also help address how much evolution in the gmatrix could lead to heterogeneity benefits breaking down.
The constraint of population size is a tricky one, and there is only so much we can do to increase population size while still keeping the model individual-based. We can, however, control mutation rate with the argument mutation_pr
in run_farm_sim
. At the moment, the default is just 0; the traditional value in most IBMs is 0.001, but we can set it to whatever we think is reasonable if we want to model a realistic amount of variation introduced within the timescale of landscape change. We can even vary it while keeping the rest of the starting conditions constant to see how mutation rate affects the evolution of resistance.
I think we can model refugia (adding a landscape area with something like a refugia on it isn't too difficult). Dominance relationships will be a bit trickier. Since the model allows for a diploid genetic architecture, this could be possible, but then we need to decide how dominance works in practice (e.g., do we then need a pre-determined, finite, number of alleles, with some kind of dominance hierarchy among them?).
Increasing immigration rate to find the number of immigrant required to maintain susceptibility will be doable once I finish the immigration function @rosemckeon -- this shouldn't take too long, and I think would just be introducing new individuals into the population initialised in the same way as the starting population (i.e., same genetic architecture, but random normal alleles that have not yet been under selection in the local environment). Would be good to test!
Likewise with the constraint on crop preference, though this will require a bit more tweaking of the code to make the movement and settling rules for this apply. But the probability of accepting an alternative crop to the one on which a pest developed could be a fixed number or evolving trait once coded.
This is definitely something we can play with @RosieMangan -- I think just by tweaking the covariance matrix and crop and biopesticide number. I'm still struggling with thinking about host generalisation versus specialisation though. Is it known the extent to which individuals or (sub)populations of H. armigera specialise on crops? One way to do this might be to limit dispersal and make farm sizes bigger, so some pests essentially have no other option than to locally adapt.
We can definitely play around with the timings @RosieMangan -- as I have been running the simulations so far, there has been quite a lot of demographic overlap so pests in a population are not all larvae, feeding, or moving at once. But maybe this is unrealistic, and it would be better to simulate more demographic synchrony in the population instead? This might actually be an interesting part of the question -- timing of both pesticide spraying and synchrony of life-history stages in the pest.
That's a good question -- at the moment there is no human agency; the application of pesticide is predetermined from the start of the simulation (or is randomly selected after a certain amount of time is passed). Hence, the application cannot react to the pest population itself. This could potentially be changed, either through a simple rule-based method (if the pest count is sufficiently high, then apply the pesticide), or through something more complex, as I've done with GMSE. Including natural enemies isn't possible at the moment either, but this could be relatively easy to implement as long as the ploidy of the natural enemies matched that of the pests, and the number of loci and traits were the same (traits could, potentially, differ if we allowed for some 'dummy' traits). For a more complex model, another array entirely could be introduced.
This seems especially important for a targetted model of H. armigera @RosieMangan -- at the moment, there are no natural 'winters' in the model. It would be good to think about what this should look like -- e.g., after some number of time steps, should all pests except the pupae die, then emerge synchronously later? This can definitely be done with the implementation of some kind of winter_cycle
argument that causes a major change every T timesteps.
Thanks all! I'm excited to continue building this model.
I had another thought regarding winters. I think it's typical for overlapping generations to be present in one population because of variation in development speed, pupation length and lifespan.
In the potato beetle this has caused complications in managing resistance, because not all of the active population at any given time have been subjected to the same control regime. Similar dynamics could be important for Helicoverpa too.
Could be fun to test overlapping cycles rather than having all individuals in a generation existing at the same time. Are any of these variables already covered @bradduthie?
At the moment @rosemckeon overlapping generations can occur because of the way the time steps are managed. The min and max age of reproduction can be specified so that an individual's reproductive output could span several time steps (they could even inbreed with an offspring under some parameter combinations). I've not actually tried this to compare overlapping versus non-overlapping generations though; it could be interesting for resistance evolution, particularly considering the scale of pesticide or crop heterogeneity.
That's a really neat way to parameratise that. Sounds cool!
Sorry to be so late in committing some ideas to this thread @bradduthie. I agree with the ideas posted by others above, and suggest approaching these more or less in order of tractability/easy-ness so that we can address the "low-hanging" fruit first. I also think the answers to these initial questions might help frame more complex ones, added below.
Of course I think some of the more complex ideas are also among the more interesting ones. Matt alluded to questions of dominance above, and I think there's loads of scope to ask how the genetic details in general affect the extent to which heterogeneity engenders fluctuating selection. Dominance is one aspect of this, as is the number of loci affecting a trait, and the distribution of trait effects. Having answers to some of these questions will be key to addressing reviewers who want to know how sensitive our findings are to the genetic system and your approach to genome creation, IMO.
Another more complex and perhaps more intriguing issue concerns how levels of pleiotropy and epistasis affect complex trait and G-matrix evolution. For example, it's obviously true that one can get strong genetic correlations with only disequilibrium, but that (absent some constraints on recombination) those correlations will crumble in the face of any selection on covariances. However, it's not clear how much pleiotropy or epistasis is needed. Is it a requirement that there are strong genetic trade-offs for sustainable fluctuating selection based on negative genetic correlations? Or can a large number of modest trade-offs do the trick? What about epistatic interactions? How many, and how strong do they need to be? I guess the answers to these questions won't be simple, but even getting a sense for the parameter space in which sustainable fluctuating selection is permitted would be fascinating.
Even more complex (maybe?) are questions related to condition-dependence. By this, what I mean is can we quantify how resource trade-offs might mediate investment decisions & trade-offs across habitats. IMO what would be needed here is some kind of latent "condition" metric, which would model the genic capture (sensu Rowe and Houle 1996) of many loci throughout the genome that affect how organisms acquire and convert resources into fitness. It's not clear to me that this would produce a pattern that is any different from a complex genome loaded with epistasis and pleiotropic effects, but in my view it's easier to conceptualise metabolic (rather than genetic) trade-offs if we can more clearly carve out how alleles produce "condition" in a context specific way. It may well be that this is just a different framing of the previous question (and so doesn't need a different set of models), but I guess it would need a different set of figures/interpretations/discussion. I fear I'm not being crystal clear about this, so please let me know if this seems interesting and anyone wants to discuss more...
And as a quick comment from me - after just being forced to write a progress report on our original grant - Brad, if it is possible, it would be great to make sure the outputs overlap with the original aims of the grant IN ADDITION to addressing some of these (more?) exciting ideas!
We can use this issue to propose and discuss potential ideas for simulations to address interesting scientific questions.