Wednesday, April 20, 2016

Subtle Sexism Self-Evaluation

Sexism in science has been discussed for as long as I can remember being a part of science. Until I joined social media, however, nearly all of the discussion I can remember was related to hiring. In particular, considerable attention is always paid to trying to increase the representation of women faculty through the hiring process. These efforts are manifest as encouraging women to apply, making sure a number of qualified women are interviewed, and giving preference to women in situations of relatively equal qualification and ranking.

Since I joined social media, the range of topics and contexts, and the frequency with which they are discussed, has become much more apparent to me. In particular, a lot of scientific discourse on Twitter and Facebook is focused on gender equality and fairness and opportunity in science. I have not been particularly active in these discussions, as I might have been by – for example – tweeting or retweeting the many comments about gender biases in academia. 

Probably the main reason for not being too active here was that I didn’t want to seem opportunistic or casual or unthinking in any advocacy. Rather, I wanted to restrict myself to a few posts that might thereby indicate some sincerity.

As time has gone on, however, I had a growing sentiment that it would be appropriate and valuable for me to say something more specific and focused on the topic. The final impetus that motivated this post was my current geographical location – I am on sabbatical at UC Berkeley, where a number of high-profile sexual harassment cases have recently come to light. In particular, the “resignation” (presumably under pressure) of Dr. Geoff Marcy, following exposure of the university’s cover-up of his violation of their sexual harassment policy, hit close to home. The reason was that – just before the allegations were made public – I happened to have lunch with a bunch of scientists, including Dr. Marcy. At the time, I had no idea of the allegations, nor of the cover-up, nor of the resignation that would soon follow. It was distressing to realize that I had no idea of the overt sexism of someone with whom I was interacting.  

I am pretty certain that I am not overtly sexist. However, reading various articles highlighted on Twitter has made it clear to me that sexism is not just overt but also subtle and subconscious. That is, a professor might treat men and women differently without realizing it – unless it is specifically pointed out to them. Instances of my own potential subtle sexism have been pointed out to me before, most publicly in the tweet below about a symposium I organized. I therefore decided to undertake my own “subtle sexism self-evaluation” by assessing potential quantitative indicators of how I interact with women in science.


Do I collaborate more with men than with women?

To answer this question, I tallied the male and female coauthors of every paper I have published (a person was counted each time I published with them – so the same person can be counted a number of times). Of my 490 total coauthors, only 142 (29%) were women. This proportion differed dramatically from before to after my appointment as a professor (see graph below). Most dramatically, only 1 of the first 17 papers I published had a female coauthor. Considering all papers before I became a professor, only 10% (8 of 79) of my coauthors were women. 


Considering all papers after I became a professor, 33% (134 of 411) were women, and at least one woman has been an author on each of my last 9 papers. Why the shift through time? One possibility is that my graduate work was in a Fisheries department, a field traditionally dominated by men (e.g., < 30% of “fisheries” professors are women: figure below), whereas my faculty position is in Biology, where the sex ratio is less biased (although not in my department). Another possibility is that participation of women in research is increasing through time. A final possibility is that I altered my behavior to specifically collaborate more with women – although I don’t remember consciously doing so.

From Arismendi and Penaluna (2016) via @TrevorABranch

Of course, even my post-hiring rate of 33% women collaborators could still indicate a bias, depending on the proportion of women among peers and trainees in my field. Let’s start with graduate students. Since starting as a professor, women have comprised 60% of my MSc students, 47% of my PhD students, and 30% of my postdocs (total across all three categories = 46%). And what about my peers? Schroeder et al. (2013) reported that 22–39% (depending on career state) of professors in evolutionary biology were women.

My self-evaluation of these numbers is that I do not show any apparent bias when it comes to collaboration with my peers – other professors in my field. However, why are 46% of my students/postdocs women but only 33% of my post-hiring coauthors women? To further consider this apparent discrepancy, I divided all my coauthors into students/postdocs versus peers, revealing that 42% of my student/postdoc coauthors were women – about the same as their frequency in my lab.

However, the data do show at least one situation involving very strong bias. In particular, I have been a part of two “perspective-type” papers with more than 15 authors. One of these had only 1 female author out of 16 and the other had only 5 female authors out of 15 – and I was first author on both papers! However, the participants in these two papers were determined by participation in the workshops that spawned the papers – and I didn’t organize either of those workshops or invite people to them. I merely took the lead on writing the papers that came from them. Regardless, it seems that I could do a better job of encouraging participation of women in the working groups that generate such papers.

15 men and only 1 woman

Do I invite more women than men?

Above, I noted a bias toward men in large working groups that generate big “perspective” papers, while also noting that I was not in charge of inviting participants to those groups. Thus, I wanted another way to assess my own possible bias in inviting attendees. Indeed, my initial thinking was that I am likely to have such a bias given the earlier-noted tweet about a symposium I organized where all 6 speakers were men. Moreover, Schroeder et al. (2013) noted a strong bias toward men in invited (as opposed to contributed) talks at the European Society of Evolutionary Biology (see figure below).

From Schroeder et al. (2013).

To address my potential “invitation bias” most directly in the currency that matters most, I tallied the invited authors of every published journal special issue that I have edited where I was at least partly responsible for the invitations. This proved a bit hard to parse because several people were often invited to work together on the same paper, and the invited person sometimes then invited someone else to be lead author. So – to be as simple as possible – I just tallied first authors of the 61 papers published in the 5 special issues made up of invited participants. Of the 61 first authors, 20 (33%) were women – about the same as the proportion of women faculty members in evolutionary biology, and somewhat higher than the proportion of women invited to speak in symposia at ESEB. But that “solidly average” outcome doesn’t mean I can’t do better.

Some of the great women scientists in my field.
Special issue: Women's contribution to basic and applied evolutionary biology

Do my female-associated and male-associated words differ in letters of recommendation?

A particularly insidious form of sexism in science that I recently became aware of (and tweeted about) was the use of different adjectives in letters of recommendation. As shown in the Wordle above, letters of recommendation (compiled from a job search) for men tend to use "standout" words such as fabulous, bright, exceptional, remarkable, extraordinary, and so on; whereas letters for women tend to use "grindstone" words such as conscientious, dependable, meticulous, hard working, and so on. I found this shocking (although some observers pointed out that the differences in the wordle appear greater than those reported in the original paper), and so I wondered if I had the same subtle bias. So I took every letter of recommendation I have written for graduate students or postdocs or colleagues (one – the most recent – per person), divided them into letters for men (N = 16) and women (N = 14), removed many uninteresting words (e.g., “student,” “PhD,” “ecology”, “evolution”) equally from both, and Wordled them. Here is what I came up with.


Words in my letters of recommendation for women.

Words in my letters of recommendation for men.

"Work", "paper", "published", and "important" seem to be the most common words in both, which isn’t surprising – but they are hardly informative with respect potential sexism. “Excellent” is about the same in both but “great” appears to be more common for men and “good” for women, “outstanding” appears more common for women than men, “hardworking” appears more common for women than men, and “confident” appears more commonly for women than men. Interestingly, none of the other standout (fabulous, bright, exceptional, remarkable, extraordinary) or grindstone (conscientious, dependable, meticulous) words noted above are common – perhaps because I make sure I don’t re-use such words multiple times in the same letter.

I therefore also searched for the standout and grindstone words that appeared in the "shocking" wordle comparison I tweeted earlier. I tend to use these words rarely - probably because they don't sound sincere and, in fact, I tend to use them twice as much for women than men. However, this got me to wondering about "great" and "good" again. So I searched for those and found a HUGE bias. I used "great" 41 times for men and only 7 times for women, whereas I used "good" 16 times for men and 25 times for women. Looking closer, I see that I used "great" a bunch of times in one sentence as a literary device. "As a final key observation, he is the most collegial person I have ever met. He is a great person, a great collaborator, a great friend, a great colleague, a great … well … everything. Every loves him. I can honestly say I have never met any who had a single bad, or even equivocal, word to say about him. In fact, everyone professes great admiration and affection for him. He is, quite simply, a wonderful person." But even excluding this exceptional case, I do seem to have a bias with respect to the standout words "great" and "good", which I clearly need to be aware of.



Where should I go from here?

I am partially reassured by my “subtle sexism self-evaluation” in that none of the indicators appears particularly egregious. Nevertheless, I can see several areas for improvement. I can work to achieve a better balance of participation in working groups, and I can pay attention to my use of “great” vs. “good” in letters of recommendation. Moreover, there are many other contexts where my subtle sexism might be apparent but that I can't easily quantify based on hard data. Many other ways exist to improve opportunities and equality for women in science - and many of these I can implement myself. I will save more details for another post as this one is already quite long but I also hope that others reading this post will provide suggestions below. Finally, while this post has focused on sexism in science, a similar evaluation would be useful for minorities and other under-represented groups in science.

More resources are here.

My own "subtle sexism self-evaluation" was very interesting and revealing and I encourage everyone to conduct their own. It took me about a day (much of it spent on the letters of recommendation) and the outcome was rewarding and motivational.


Notes:

It has been argued that women are done a disservice by being asked to participate equally in various committees and working groups when they are not equally represented in the field. As a result, women could end up spending more time on external activities that take their time away from research. With the context of working groups and symposia, however, any invitee can say no if they wish to (and this is common - Schroeder et al. 2013), and so it is not exploitative for an organizer to work toward equal representation.

The word "sexual" appears in the wordles because some of my students – coincidentally more women than men – studied the process of sexual selection in guppies.

Monday, April 11, 2016

The small population paradigm: The be-all and end-all for small populations in nature?

[ This post is by Jackie Wood; I am just putting it up.  –B. ]

What is the fate of small populations in nature?

This question (and its answer) has critically important implications for the conservation and management of wild species considering that many natural habitats now exist only as small and isolated fragments due to human activities.

With only a cursory glance at the scientific literature, we might understandably become convinced that we absolutely, without a doubt, know the answer to this question. There are a suite of well-known threats that include habitat loss and fragmentation, exploitation, and now, increasingly, climate change that have conspired to yield widespread reductions in the population size of many natural species. As these threats continue to exact their toll on the population size (N) of an isolated population (isolation is a key point since such populations no longer benefit from an influx of genetic material from neighboring populations), breeding between related individuals becomes more common (inbreeding) and genetic variation that enables a population to respond to environmental change is lost more rapidly (genetic drift). Together, these can potentially lead to reduced reproduction and increased mortality, resulting in further decreases in N. This, in turn, results in even more inbreeding and greater genetic drift – and just like that, the population is caught in a downward spiral of ever-decreasing N until a point at which the population is so small that a chance event such as a bad winter can deliver the coup de grâce that finally drives the population out of existence. This process, called the “extinction vortex” by Gilpin and Soulé in 1986, is a hallmark of Caughley’s conservation genetics small-population paradigm; it is cited in virtually every textbook and taught in every undergraduate course on conservation biology. There are certainly several well-known cases of natural populations that have suffered the grip of the extinction vortex, but how pervasive is it in general?

Spoiler alert: it turns out that this classic extinction vortex might not be so pervasive after all!


King of the Cape: a brook trout at Cape Race, the system in which we studied these questions (as described below).  Photo credit: Cyndy Desjardins.

A key factor driving the extinction vortex is “genetic stochasticity”: the random loss of genetic variation necessary to enable a population to adapt to environmental change. In populations that are small and declining, neutral genetic variation (i.e. genetic variation that is not currently contributing to the fitness of individuals and is therefore not affected by natural selection) is predicted to be lost increasingly more rapidly, an expectation that has been confirmed in both laboratory and natural populations. But in the case of adaptive potential, what we are more interested in is the genetic variation that underlies traits that are related to fitness, which is known as quantitative genetic variation (QGV; which can be estimated using one of several different metrics including VA, h2, and CVA). The specific connection between QGV and adaptive potential originates from the breeder's equation (R = h2S), where response to selection (R) is directly proportional to the strength of selection (S) and the QGV (h2) of a trait (or multiple traits using the multivariate form of the breeder’s equation, the G-matrix).

In theory and under certain simplifying assumptions, loss of QGV is predicted to occur at the same rate as loss of neutral variation with decreasing N, and this has been found to be the case in laboratory populations for certain kinds of traits. In the nature, however, the relationship between QGV and N remains unresolved, and there are a number of reasons why this is the case.

One reason is that many traditional assays of genetic variation in natural populations have used neutral genetic variation as a surrogate for QGV even though the correlation between these metrics has been found to be weak, and the studies also didn’t account for how selection acts on variation. And this does matter; for example, stabilizing selection can actually favour the maintenance of genetic variation within small populations. Similarly, loss of QGV might be buffered to some extent if small populations can respond to changing environmental conditions in the short term by plasticity (the ability of a single genotype to produce different phenotypes in different environments), a trait which might reasonably be favored under certain types of environments occupied by small populations, but this possibility has often been overlooked. Laboratory studies of QGV and response to selection relative to N have largely been restricted to a small number of species including bacteria and fruit flies, and both the environmental conditions used in these studies and genetic characteristics of these species may not adequately portray what happens under many conditions in nature. For example, the relationship between genetic variation and N might reasonably differ between fruit flies and vertebrates because vertebrates can exhibit behaviours such as inbreeding avoidance and complex mate choice. Laboratory experiments also tend to use similar, simplified, or benign starting conditions across populations even though this is unlikely to occur regularly in nature. Among small natural populations, conditions immediately after habitat fragmentation likely vary, and this will ultimately influence the fate of each population in terms of QGV and response to selection.


Typical weather conditions at Cape Race, our research site.  Alternate title: “Lost in the fog of quantitative genetics.”

How the process of habitat fragmentation might alter environmental conditions as fragment size and N is reduced has potentially important evolutionary consequences because this will determine the kind of selective regimes that populations are exposed to, and hence will influence the adaptive genetic make-up of fragmented populations (see diagram below). One possibility, which represents the more or less classic view of habitat fragmentation, predicts a systematic directional change in environmental conditions with decreasing fragment size and N (the Directional Hypothesis). In the Directional Hypothesis, large fragments are typically assumed to be high quality, whereas small fragments are expected to represent marginal habitat due to increasing edge effects including increasing parasitism, predation, and more widely fluctuating climatic conditions.

However, this might not be the only possibility. An alternative prediction is that evolutionary effects of habitat fragmentation might depend on initial starting conditions within the fragments themselves (the Variable Hypothesis). Simply put, where you end up will depend quite a bit on where you started. Here, large fragments are presumed to be similarly complex and consist of multiple habitat types, and small habitat fragments are expected to simply be random samples from the larger ones. For example, microclimate conditions or distribution of food in a large fragment may differ both spatially and in time, such that fragmentation will result in patches that become increasingly dissimilar as they are reduced in size – possibly also resulting in increased variability in selective regimes. Collectively, the Variable Hypothesis suggests if that we were to compare multiple large habitats, they would be similar in a containing a large number of different habitat types, whereas if we compared multiple small habitat fragments there would be less variability in habitat types within the fragments (since logically a small fragment would have less room for many different habitat types), but more between-fragment variability. For example, the most extreme version of this hypothesis, and perhaps the most easily visualized, is a scenario in which we have many very small fragments with each fragment containing only one type of habitat, and the habitat type is completely different for each fragment.


Two alternatives for the effect of habitat fragmentation on environmental conditions within and among fragments occupied by populations of differing N.

In 2010 our lab initiated a multi-year research program aimed at testing the assumptions of the conservation genetics small-population paradigm in nature. Using the Directional Hypothesis and the Variable Hypothesis as our points of departure, we set out first to quantify the relationship between N and variability in habitat conditions for a series of pristine, naturally fragmented, and isolated brook trout populations located at Cape Race, Newfoundland, that varied widely in N.


Most Cape Race streams end in waterfalls emptying into the ocean so we can confidently say the populations are isolated! But, like, watch your step.

#squad

From detailed habitat surveys wherein we collected data for a range of habitat parameters across 14 and 19 streams in two consecutive summers, we observed a little evidence for increased variability in mean habitat parameter values among small compared with large populations, but much more evidence that habitat parameter CVs (a measure of variance around the mean) were more variable at small than large N. Taken together, our findings suggested at least some support for the Variable Hypothesis in Cape Race streams. These patterns naturally raised the possibility that the small trout populations might be subject to more varying selective regimes than large populations. If true, and if such patterns in environmental conditions were stable over time, we hypothesized that population-level characteristics including phenotypic plasticity and QGV might also vary more among small than large trout populations.

To test this, we conducted two common-garden experiments, but we found no evidence for consistent differences in plastic response to changing temperature from large to small N for 8 Cape Race populations, and there were also no systematic differences in the level of QGV (for any of three different QGV estimators: VA, h2, and mean-scaled evolvability) in relation to N for the populations. Using genome scans, we also looked for signals that adaptive genetic differentiation was related to N (adaptive genetic differentiation simply refers to how far apart two populations are in terms of genetic markers that are linked to fitness-related traits). In contrast to the plasticity and QGV studies, here we found evidence to suggest that small population pairs tended to be more different than pairs of large populations, not only suggesting that selection is at work in small Cape Race populations, but also providing another line of evidence that suggests that small populations might be exposed to a variety of environmental conditions. In this study we also found that small-large population pairs were more different than pairs of large populations, so there was support for the Directional Hypothesis as well! The result of increased adaptive differentiation among small populations is notable given that we also found that neutral genetic variation increased with increasing N, which is consistent with pronounced genetic drift in the small populations.

Our results were clearly not concordant across the board; they provided support for the Variable Hypothesis in terms of habitat conditions, for both the Directional and Variable Hypotheses in regards to adaptive differentiation, and for neither hypothesis for plasticity and QGV in relation to N. Nevertheless, they are encouraging in suggesting that (i) populations of 100 adults can be as genetically robust as populations with 1000–10000 adults in some contexts, (ii) small populations may have the ability to deal plastically with environmental change to some extent, and (iii) selection may very well operate at small N resulting in adaptive population differentiation even at very fine scales. In other words, many small populations may not be evolutionary dead ends, as commonly thought, but rather, may constitute locally adapted populations that contain potentially unique adaptive genetic variation (by virtue of inhabiting unique environments and being exposed to unique selection regimes).

Of course, one thing we can’t rule out is that the small populations we study are merely the lucky ones that are left; perhaps other small brook trout populations were present at some point in time but were extirpated due to the suite of genetic problems usually associated with small N. We’ll never know.  Still, our general observations of Cape Race populations are that brook trout are quite resilient little fish as long as their key habitat requirements are met. Brook trout can be found in virtually any isolated ditch or tiny pond on the Cape, and seem to do just fine at small N provided there are cold-water refugia to escape warm summer temperatures, and upwelling groundwater for spawning and egg incubation.


Yes, there is a stream in there. Yes, it contains a naturally reproducing brook trout population.

If there’s a main point to be made based on the results of our work at Cape Race, it’s the critical importance of habitat quality for population persistence at small N; our results also suggest that in many cases environmental stochasticity might be a more immediate threat than genetic stochasticity for small brook trout populations. On the other hand, the idea that brook trout are fairly robust to the effects of small N may not be all that surprising given their capacity to colonize new habitats, an affinity for small headwater stream environments, and a partially duplicated genome which may act to buffer the loss of genetic diversity at small N.

What about other types of species? How do they do when populations are reduced to a finite number of individuals? It’s hard to say; we can certainly make inferences based on what we know about species ecology, life history, behaviour, and conservation-genetic theory, but the truth is that there are currently no other natural vertebrate systems that have been studied as extensively as Cape Race in this regard. In a recently published meta-analysis, we attempted to shed some light on this question by extracting data from previously published articles dating back to 1980 which estimated h2 or selection (strength, direction, and form) for natural populations for which N data was also available. Using this data, we looked for evidence of systematic differences in h2 and selection with N, pooling data across a variety of species. Looking at h2 and selection together also enabled us to infer adaptive potential in a very crude sense using the breeder’s equation (recall that R = h2S; we couldn’t formally test N vs. adaptive potential since h2 and selection data weren’t typically available for the same population). What we found was, well, not much.  There was little evidence for differences in h2 or the extent of natural selection across the various species and N that we included in the analysis. As for brook trout at Cape Race, these results suggest, at least indirectly, that while genetic drift and selection operate simultaneously in nature, drift may not always overwhelm selection at small N and that populations of varying size may respond to environmental change. Evidence supporting the efficacy of selection despite small N has also been found in a recently published study by Labonne and colleagues for a brown trout population originating from only three founding individuals.

Overall, however, it’s extraordinarily difficult to draw definitive conclusions regarding the status of small populations in nature in general based on our findings from collating data from scientific articles. That’s because the other take-home message from our meta-analysis is that there are several glaring information gaps that hamper our ability to say anything with certainty.  For example, we weren’t able to account for the degree of isolation of populations (due to a lack of information on gene flow); this is a critically important point since even small amounts of gene flow can help small populations to retain genetic variation that would otherwise be lost by persistent genetic drift and inbreeding depression. Data was also biased towards widespread, generalist species. We found no studies which estimated selection or h2 for amphibian populations of known N, despite being one of the planet’s most threatened taxonomic groups. Other rare, specialist species (species that might reasonably be most likely to fall prey to the small-population paradigm), or entire taxa (reptiles, invertebrates) were similarly poorly represented in our databases. There is also a need for greater adoption of multivariate approaches to studying evolution in natural populations. Most studies focus on single traits; however, many traits do not evolve independently, but rather are correlated with other traits. This means that direct selection resulting in an increase in the value of one trait can result in indirect selection on a correlated trait to increase it, decrease it, or even constrain it; thus, focusing on single traits may not adequately detect constraints to evolution in small populations.

Keeping these gaps in mind, we very cautiously suggest that response to selection at small N might be more extensive than previously assumed in evolutionary and conservation biology. But even if we could be sure about this, it still wouldn’t mean that small populations are equal to large populations. Even if some small populations have been lucky enough to land in habitats that can support them and retain the capacity to cope with changing conditions to some degree by plasticity or evolution, they still have a greater risk of extinction from environmental and demographic stochasticity. Only large numbers can buffer against such threats. And of course, other small populations won’t be so lucky, and will end up in marginal environments where they will inevitably fulfill the expectations of the conservation genetics small-population paradigm.

What is the fate of small populations in nature?

Well, it is probable that things simply become more unpredictable at small N, including evolution. This means that we may only be able to address this question on a case-by-case basis. As suggested by the late David Reed in regards to the minimum amount of genetic diversity required for populations to persist for a given length of time, the answer will intimately depend on factors such as habitat quality, the rate and magnitude of environmental change, and the history and genetic characteristics of the population or species of interest.


Just completely casual, totally normally lounging. Nothing out of the ordinary happening here.

Literature Cited:

Wood, J. L. A., Yates, M. C., and D. J. Fraser. 2016. Are heritability and selection related to small population size in nature? Meta-analysis and conservation implications. Evolutionary Applications. doi: 10.1111/eva.12375.

Labonne, J., Kaeuffer, R., Guéraud, F., Zhou, M., Manicki, A., and A. P. Hendry. 2016. From the bare minimum: genetics and selection in populations founded by only a few parents. Evolutionary Ecology Research 17(1):21–34.

Wood, J. L. A., Tezel, D., Joyal, D., and D. J. Fraser. 2015. Population size is weakly related to quantitative genetic variation and trait differentiation in a stream fish.  Evolution 69:2303–2318.

Wood, J. L. A., and D. J. Fraser. 2015. Similar plastic responses to elevated temperature among different-sized brook trout populations. Ecology 96:1010–1019.

Fraser, D. J., Debes, P. V., Bernatchez, L., and J. A. Hutchings. 2014. Population size, habitat fragmentation, and the nature of adaptive variation in a stream fish. Proceedings of the Royal Society of London B: Biological Sciences 281(1790): 20140370.

Wood, J. L. A., Belmar-Lucero, S., Hutchings, J. A., and D. J. Fraser. 2014. Relationship of habitat variability to population size in a stream fish. Ecological Applications 24:1085–1100.

Reed, D. H. 2010. Albatrosses, eagles, and newts, Oh My!: exceptions to the prevailing paradigm concerning genetic diversity and population viability? Animal Conservation 13:448–457.

Caughley, G. 1994. Directions in conservation biology. Journal of Animal Ecology 63:215–244.

Gilpin, M. E. and Soule, M. E. 1986. Minimum viable populations: processes of extinction. Pages 19–34 in: Conservation Biology: The Science of Scarcity and Diversity (ed. Soule, M. E.). Sinauer Associates, Sunderland, MA.

Tuesday, April 5, 2016

Announcing SLiM 2.0: flexible, fast, interactive forward genetic simulations

After finishing my Ph.D. with Andrew at McGill, I more or less vanished from view.  For almost two years, I have toiled in the darkness, chained to my desk, unseen by any in the land of the living.  But now – now, I emerge into the light!  And today, I am announcing the fruit of my labors: SLiM 2.0!


OK, possibly this comparison would be taking things a little bit too far.  :->

SLiM is a forward genetic simulation package.  The old SLiM, which ended at version 1.8, was written by Philipp Messer of Cornell University.  Then Philipp hired me into his lab, and – cue quick-cut sequence of toil, chains, and darkness – I have rewritten it and greatly extended it.  We're very excited about it, because it brings a new level of flexibility, power, and usability to the realm of forward simulations.

The first thing to note about SLiM 2.0 is that it has an interactive user interface called SLiMgui, for Mac OS X.  You can run SLiM at the command line, too, so it can be used on Linux, on computing clusters, and so forth; but when you're developing a new model, the easiest way to do it is in SLiMgui, which looks like this (click to see at full size):


This shows a six-subpopulation model midway through a run.  The six subpopulations have been set up as an "island model", connected by migration to form a chain with gene flow predominantly in one direction (like sites along a river, with gene flow tending to go downstream, for example).  SLiMgui is showing the population structure in the subwindow at lower right; if you click to see the image at full size, you can see that arrows of different widths indicate the strength of migration between the subpopulations.  SLiM is an individual-based simulation package; every individual in the six subpopulations is shown in the six arrays of colored blocks at top.  The colors indicate fitness; in this model, a mutation has swept through the population, but there is spatial variation in selection that makes that mutation beneficial in some subpopulations (green) but deleterious in others (red).  Running a simulation like this in SLiMgui is completely interactive; we can step forward a generation at a time, inspect individual organisms or even individual mutations, execute commands that modify the simulation even as it's running, and even pull up graphs showing things like the evolutionary trajectories of mutations.

The second thing to note about SLiM 2.0 is that it's scriptable.  The old SLiM ran a model that was specified in a structured input file that listed things like the genetic details and the population structure.  Anything not explicitly supported by the input file was impossible to do (without going in and editing the C++ code in which it was written, anyway).  SLiM 2.0, on the other hand, is scriptable; you write a simulation script in a language called Eidos (which is very similar to R, so if you know R there will be very little learning curve there), and SLiM runs the script.  This means that SLiM 2.0 is now almost completely open-ended.  We've written a manual for SLiM that consists, in large part, of "recipes" for different sorts of simulations, to show what's possible.  We've got recipes for such diverse topics as:
  • complex genetic structure – different types of mutations, different genomic regions, variation in the recombination rate along the chromosome, etc.
  • complex population structure – subpopulations, migration, exponential growth, cyclical changes, context-dependent changes
  • complex selection – spatial and temporal variation in selection, frequency-dependent selection, epistasis, polygenic selection, kin selection
  • complex mating systems – modeling hermaphrodites versus sexual organisms, selfing, cloning, sex ratios, assortative mating, sequential mate search, even gametophytic self-incompatibility
  • complex temporal structure – modeling the introduction of mutations at particular times, detecting the establishment or completion of selective sweeps, making particular events trigger in particular generations or upon particular conditions, and producing customizable model output at particular times.
Since you can do anything you want in your Eidos script, the sky is really the limit; we even have a recipe for a model of social learning of cultural traits that influence fitness, and a model of a "gene drive" powered by the CRISPR/Cas9 genetic modification mechanism.

The third thing to note is that because all this flexibility is controlled through a simple scripting language, all of the recipes mentioned above are less than a page of code.  SLiM models are generally very short and easy to understand.  For example, to set up a six-subpopulation island model like the one shown in the screenshot above, we could do this:

initialize() {
   initializeMutationRate(1e-7);
   initializeMutationType("m1", 0.5, "f", 0.0);
   initializeMutationType("m2", 0.5, "f", -0.1);
   initializeGenomicElementType("g1", m1, 1.0);
   initializeGenomicElement(g1, 0, 99999);
   initializeRecombinationRate(1e-8);
}
1 {
   for (i in 0:5)
      sim.addSubpop(i, 500);
   for (i in 1:5)
      sim.subpopulations[i].setMigrationRates(i-1, 0.001);
   for (i in 0:4)
      sim.subpopulations[i].setMigrationRates(i+1, 0.1);
}

The first code block sets up basic simulation parameters like the mutation rate and the genomic structure.  The second block uses for loops to create the six subpopulations and then connect them through migration.  This model simulates only neutral mutations.  To then add a new mutation that is under selection, we can add this code:

100 late() {
   mut = p0.genomes[0].addNewDrawnMutation(m2, 10000);
   p0.genomes[1:49].addMutations(mut);
}

That creates a new mutation in generation 100, and then adds 49 more copies of the new mutation to other individuals, perhaps simulating gene flow from an external source that introduced the new allele into the system.  We can track the fate of our new allele and print a message if it fixes or is lost:

100:10000 late() {
   if (sim.countOfMutationsOfType(m2) == 0)
   {
      fixed = any(sim.substitutions.mutationType == m2);
      cat(ifelse(fixed, "FIXED\n", "LOST\n"));
      sim.simulationFinished();
   }
}

Finally, we can set up spatial variation in selection for the introduced mutation, making it beneficial toward the lower (downstream) end of our subpopulation chain, and deleterious toward the higher (upstream) end:

fitness(m2) {
   return 1.5 - subpop.id * 0.15;
}

This all pretty much follows a recipe in the SLiM manual's "cookbook" (in section 12.3).  That recipe then goes on to add a CRISPR/Cas9 "gene drive" that drives the introduced mutation through the whole system despite the fact that it is strongly deleterious in the upstream subpopulations.  That scenario is actually what is pictured in the screenshot above; that is why the introduced mutation is fixing even in the subpopulations where it is deleterious.  Without the "gene drive" – as modeled by the code given here – you get a nice simulation of migration-selection balance in an island model.

Of course many details of this model are probably unclear to you now; you can check out SLiM's manual for further details.  Here, the point is to illustrate how simple it is to write a model like this – with complex population structure, both neutral and introduced mutations, monitoring of a selective sweep, spatial variation in selection, and custom output! – in just a few lines of code.  Remarkably, you pay almost no speed penalty for all of this added power and flexibility.  In fact, SLiM 2.0 is faster than SLiM 1.8 was, across the board, and it is often faster than the competition, too.  (We will soon publish a paper with speed comparisons to SLiM 1.8 and a couple of other forward genetic simulation packages, but – spoiler alert – that will be the punch line of the comparison.)

I hope this whets your appetite for more!  If so, surf to the SLiM 2.0 home page at http://messerlab.org/slim/ and download SLiM itself (as an installer app, if you're on Mac OS X, or as a source code archive if you're on Linux), manuals for SLiM and Eidos, and some other goodies.  There are links there to mailing lists that you can join to get announcements (slim-announce) or to ask us questions, start discussions, log bug reports, and so forth (slim-discuss).  SLiM is open-source and free to use (we ask only that you cite our soon-to-be-published paper), so enjoy it – and let us know how it goes for you!

Friday, April 1, 2016

Sex, Guppies and Parasitism

By: Christina Tadiri

Most of my work focuses on understanding guppy-Gyrodactylus interactions in the lab, and I’ve talked about some of my previous work with them here. These parasites are interesting because they don’t fit neatly into the macro-microparasite dichotomy that most ecologists use. On the one hand, they’re worms like a macroparasite, but on the other, they reproduce directly on the host and transmit via direct contact, like a microparasite. The important thing about them is that they cause epidemics, which can have eco-evolutionary implications and are therefore worth understanding a bit better.


Figure 1: A female and male guppy (L) and their ectoparasite Gyrodactylus turnbulli (not to scale, obviously). Photo credit Paul Bentzen (guppies) and Joanne Cable (Gyrodactylus)

My PhD thesis involves looking at how differences among hosts can impact epidemic dynamics at the population and metapopulation level. For my first study, I focused on sex, and if sex-based parasitism could influence the way these parasites spread among groups. Since guppies are sexually dimorphic (males and females look really different) there's reason to believe that the sexes might respond differently to the pressures of parasitism parasitism (as shown here), which could affect parasite dynamics at a larger scale. There have been a few other studies that looked at differences in epidemics of this parasite between males and females (this one and this one, for example), but a lot of the results were inconsistent, and these studies didn’t necessarily look at the implications at multiple levels of analysis.

First, I had to determine if sex was associated with a difference in parasitism, so I isolated individual male and female guppies and infected them all with parasites and counted how many parasites they got over the course of their infections. Then I also assembled and infected single sex groups of male and female guppies, to determine if any differences in parasites by sex translated to the population level. Finally I assembled mixed-sex groups, to see how they parasite dynamics played out once males and females were put together. You can read a bit more about the methods and analysis in the paper, but I’ll just go over our most interesting results here.
Figure 2: Total parasite population numbers over the course of the experiment in group tanks for male groups (a), female groups (b) and mixed-sex groups (c). Data are square-root transformed for graphing purposes, but were not transformed for analysis. For all panels, “Day 0” indicates the day on which at least 3 parasites were first found in the tank.

First, we found that isolated female guppies did have much higher parasite loads than isolated males. This could either be due to their larger size making it easier for them to harbor more parasites, or a difference in innate immune response. However, when we put fish in groups, we found no difference between males and females in the number of parasites per fish, nor the total number of parasites in the population, nor any of our other epidemic parameters. This of course meant that we also found no difference between our mixed and single-sex groups, since our single-sex groups were the same anyway. We also found that no matter what, parasites reached their peak earlier on males than on females (graph below).  We also found that relative condition (the metric we used for fish size) only influenced parasite load on isolated males (with larger ones having more parasite than smaller ones).

Figure 3: Mean peak burden (±SE) vs. mean time to peak (±SE) for individual fish in each treatment. SF: single (isolated) females, SM: single (isolated) males, FG: female groups, MG: male groups, MIX: mixed groups.

While these results seem a little underwhelming, we were surprised to find that differences that had mattered for isolated fish (sex, size) had no effect on parasite dynamics at the group level. These findings highlight the importance of studying host-parasite relations at a broader level, and can help to inform future experiments and models with this system. Of course, the pattern we observed may differ with fish from other populations, and further work is necessary to understand the dynamics of this unique host-parasite system. For the full story, you can check out the paper in Parasitology here.

Monday, March 28, 2016

(Genetic) Natural Selection in the Wild

A few days after I moved to Montreal to start my PhD at McGill, I met with Rowan Barrett, my co-supervisor at McGill, at one of Montreal's many shamelessly hipster third-wave cafes (Pikolo, where the coffee, to be fair, is excellent) to discuss plans and projects for my first year. A common bit of advice for first-year PhD students is to write a review or meta-analysis about some aspect of their chosen field1: it leads you to read a bunch of papers (that, frankly, you probably should be reading anyway), figure out the current state of the field, and evaluate how your proposed research fits in2. Hopefully, the review forms the basis for a thesis chapter and (ideally) a published paper. 


The authors, taking a break from meta-analyzing to do some
fieldwork in the Bahamas. 
We discussed a few possibilities that would fit nicely with my planned thesis and settled on an idea which Rowan had had in mind for a few years: a meta-analysis of genetic selection coefficients in natural populations. Biologists have been measuring phenotypic selection gradients and differentials since the 1980s, when Lande and Arnold developed the multiple regression methods used to quantify phenotypic selection. Influential meta-analyses of these data (e.g., Kingsolver et al. 2001, Rieseberg et al. 2002, Hereford et al. 2004, Siepielski et al. 2001 and 2013, and others) have told us a lot about how natural selection operates at the phenotypic level. At the genotypic level, however, fundamental questions remain: how strong is natural selection on allelic variation? How does selection vary through time and space? How are selection coefficients distributed? That these questions remain largely unanswered is partially due to technological limitations: biologists couldn't directly measure selection on alleles until the development of molecular genetics techniques in the 1980s3. The technological advances in DNA sequencing over the last decade have made it feasible to link genotype with fitness in natural populations of organisms. By 2013, when I was starting my PhD, a number of papers had reported estimates of selection coefficients (s) for specific genetic variants. So, we began tracking down all the estimates of s in natural populations that we could find. 

One detail that often gets left out of the advice to write a meta-analysis as a first-year student: actually performing a meta-analysis is a ton of work! Scanning thousands of papers for suitability, inputting hundreds of estimates of selection, and scouring bibliographies for candidate studies took much, much longer than I anticipated. But, the advice also proved to be true: I read widely4, learned a ton, and we (eventually) turned the meta-analysis into a publication, now available online at Molecular Ecology. So, what exactly did we learn? The paper has all the nitty-gritty details, but here are some of the highlights:

  1. It's an exciting time to be studying natural selection at the genetic level! We were really surprised at the breadth of research: biologists calculate selection coefficients in many different ways, in a wide variety of organisms, using many different types of data. DNA sequencing is more accessible than ever before, and this has made it much easier to incorporate genomics into field experiments. Most of our estimates of selection came from two such studies5, and we believe that genomics-enabled field experiments of adaptation will lead to a lot of new discoveries about how evolution proceeds. But researchers estimated selection in all sorts of other ways. My favorite off-the-beaten-path example: Robinson et al. 2012, who used Bayesian epidemiological modeling to quantify selection at the PRNP gene (which influences susceptibility to prion diseases) in white-tailed deer. 
  2. Given how many different ways there are to estimate selection, we were surprised to find that selection coefficients are rarely estimated. We searched thousands of plausible papers, many of which inferred natural selection at the genetic level, but a relatively small percentage of papers (~3.5%) actually quantified selection. There are, of course, many reasons why this might be the case (and we speculate on some of them in the paper), but, whatever the cause, we encourage biologists to quantify selection more often! As our paper hopefully shows, selection coefficients can help us answer some very interesting questions, and more data will lead to better conclusions. 
  3. We found that the distribution of selection coefficients was roughly exponential, with most estimates of selection being quite weak, but a few being quite strong (there were many estimates of s > 0.5!). This distribution relates to all sorts of theoretical questions (what are the dynamics of adaptive walks to a fitness optimum? What is the extent of local adaptation? How are positive and negative selection related?) and practical issues (when trying to measure selection, what values of s are possible? In Bayesian models, what sort of priors should we put on selection coefficients?). The distribution of phenotypic selection coefficients is also roughly exponential. This similarity led to some novel theoretical work, done in collaboration with Dr. Sally Otto at UBC, about how genetic architecture may (or may not) influence how these distributions are related.
    The distribution of the 3416 estimates of s included in our
    quantitative analyses.  
I could go on, but check out the paper (http://onlinelibrary.wiley.com/doi/10.1111/mec.13559/full) for a much more thorough discussion of these issues and many others, including thoughts on how selection varies through time, the prevalence of overdominant selection, and the best ways to report and interpret selection coefficients in manuscripts. 

1. See Andrew's post on how to succeed in grad school.
2. Another good reason, at least in my case: I had to write a meta-analysis as a term paper for the evolutionary ecology course I was taking.
3. Before that, studies of selection at the genetic level were limited to Mendelian traits/alleles, such as melanistic alleles in the peppered moth, where genotype could be reliably inferred from phenotype.
4. When was the last time you read a paper from the Russian Journal of Human Genetics?
5. Anderson et al. 2014, on local adaptation in a species of mustard, and Gompert et al. 2014, on selection in Timema stick insects. 

Related papers:
Anderson JT, Lee C-R, Mitchell-Olds T (2014) Strong selection genome-wide enhances fitness trade-offs across environments and episodes of selection. Evolution, 68, 16-31.

Gompert Z, Comeault AA, Farkas TE et al. (2014) Experimental evidence for ecological selection on genome variation in the wild. Ecology Letters, 17, 369-379.

Hereford J, Hansen TF, Houle D (2004) Comparing strength of directional selection: how strong is strong? Evolution, 58, 2133-2143.

Kingsolver JG, Hoekstra HE, Hoekstra JM et al. (2001) The strength of phenotypic selection in natural populations. The American Naturalist, 157, 245-261.

Lande R, Arnold SJ (1983) The measurement of selection on correlated characters. Evolution, 37, 1210-1226.

Rieseberg LH, Widmer A, Arntz AM, Burke JM (2002) Directional selection is the primary cause of phenotypic diversification. Proceedings of the National Academy of Sciences, 99, 12242-12245.

Robinson SJ, Samuel MD, Johnson CJ, Adams M, McKenzie DI (2012) Emerging prion disease drives host selection in a wildlife population. Ecological Applications, 22, 1050-1059.

Siepielski AM, DiBattista JD, Evans JA, Carlson SM (2011) Differences in the temporal dynamics of phenotypic selection among fitness components in the wild. Proceedings of the Royal Society B: Biological Sciences, 278, 1572-1580.

Siepielski AM, Gotanda KM, Morrissey MB, et al. (2013) The spatial patterns of directional phenotypic selection. Ecology Letters, 16, 1382-1392.

Thurman TJ, Barrett RDH (2016) The genetic consequences of selection in natural populations. Molecular Ecology, 25, 1429-1488.

The Null Hypothesis is Always Wrong

 No two populations are identical for any trait. No two communities have the same species composition. No detectable phenotype is ever compl...