This post was written by Vincent Fugere, a postdoc who was the driving force behind the paper - and elements of the text likely appear elsewhere on the web, such as in press releases. The pictures are by me (Andrew Hendry) and are intended to represent a number of situations in which humans and animals are likely interacting in a way that would be expected to generate changes in natural selection.
Quantifying evolutionary impacts of humans on the biosphere is harder than it seems @mcgillu https://t.co/QQFQhp9VJH pic.twitter.com/L2Lnpx1fT4— NSERC / CRSNG (@NSERC_CRSNG) October 15, 2018
This paper is the first large-scale test of the hypothesis that human activities lead to stronger selection. Human activities are not only driving rapid ecological change (eg changes in biodiversity), they are also driving rapid evolutionary change within species. For example, many fish species are becoming smaller as humans selectively harvest the largest individuals. Many weeds and insect pests are becoming resistant to pesticides, as do many pathogens to medicines. Some fish are losing their coloration in lakes that are becoming murky because of erosion and nutrients from agricultural activity. There are a lot of examples of trait change generated by human activities. Further, some data compilations have shown that on average, rates of trait change (e.g. % change in body size per generation) are indeed faster in populations facing human disturbances than in populations in more natural environments (work by A. Hendry and colleagues). In this new paper, we test the hypothesis that natural selection is also stronger in populations facing human disturbances.
Natural selection is the relationship between the traits that individuals possess (eg how big an individual is relative to other individuals in its population) and the fitness of these individuals (eg how many offspring are produced over a lifetime). The strength of selection is the steepness of that relationship – when particular trait values have much lower or higher fitness than others, then selection is strong, and one would also expect rapid change in the mean trait value of the population as unfit trait values are ‘weeded out’. One thus measures selection strength by sampling a bunch of individuals in a population, measuring one or more traits on each individual, recording fitness of each individual, and linking the two (trait values & fitness) with a trendline on a scatterplot (more or less – technically we use a statistical tool called a linear regression). The slope of that trendline (regardless of whether the trend is positive or negative) is the strength of selection. If the line is flat, then individuals have the same fitness regardless of their trait value, and thus there is no selection operating on that trait. A lot of evolutionary biologists have measured selection on hundreds of species and traits in nature, and some have recently started doing that in populations facing human disturbances.
One might suppose that populations in nature are reasonably well adapted to their environments since they have experienced selection in this environment for a long time. For example, if being larger is better in colder climates, then populations of the same species should be larger, on average, in colder than in warmer environments (this is indeed generally the case and is in fact known as the temperature size rule). Thus, background levels of natural selection should not be too strong – in the previous example, selection on body size should be relatively weak in all populations along the temperature gradient because individual body size is already well matched to local environmental conditions (temperature). But then come humans. Our activities are inducing drastic environmental change in many places. We make some places warmer, for example, and in those places, it might no longer be useful to be larger. We would then expect renewed selection on body size, ie stronger selection now that the environment has been changed by humans than before when the body size-temperature match was good. This stronger selection should weed out all large individuals from the population and bring the mean trait value to the new optimum in the new (now warmer) environment. As mentioned above, there is indeed evidence that rates of trait change are faster in populations facing human disturbances, confirming that selection might/should be stronger in human-disturbed populations as well.
Some case studies of individual species show that indeed, human activities strengthen selection. Other studies, however, found the opposite. I collected data from all studies available and asked if there was an overall trend across study. This is what’s called a ‘meta-analysis’ of the scientific literature on a particular topic. I reviewed thousands of scientific papers and retained only those that provided estimates of selection strength in both human-disturbed and natural (‘control’) environments, for the same traits and the same species. For example, this could include a weed species found in grasslands sprayed or not sprayed with herbicides. Or a lizard species found on islands with and without an invasive species of ants introduced by people. Or a shark population before and after the construction of a beach resort that induced major changes to the local mangrove habitat (this one particular example was based on the work of Joey DiBattista, a former PhD student at McGill under the supervision of Andrew Hendry). I found 40 publications that provided such estimates, and from them I collected a total of 1366 estimates of selection strength measured on 102 traits from 37 different species. I then used statistical models to test if there was an overall trend across studies of stronger selection in human-disturbed conditions.
I found that on average, selection is not stronger in disturbed environments – in fact it was slightly weaker, but that finding is highly uncertain. Thus, the conclusion that we emphasized is that overall, selection is not stronger in disturbed conditions, as one might have expected. There were some exceptions – for example two fisheries (a natural cod fishery and an experimental pike fishery) showed very strong human-induced (harvest) selection. There is also plenty of evidence that fisheries experience rapid rates of trait change (eg for body size), confirming that this human activity is likely to have large evolutionary effects.
But then we needed to explain the lack of human impacts on selection in other studies. Note that this refers to the average effect – several studies showed stronger selection in human-impacted environments, but several others did not, leading to no net effect on average. One possible explanation is a taxonomic bias. We expect many species to do poorly in human-impacted environments and thus to go extinct in these environments (this is known as ‘population extirpation’). Rates of population extirpation have never been higher, so we know that this happens often. However, evolutionary biologists cannot measure selection in populations that go extinct, as a sufficiently large number of individuals need to survive for fitness and trait values to be measured. Thus, part of the explanation for our main conclusion is that there is likely a taxonomic bias in the literature on human-induced selection, where researchers might be studying species that fare relatively better in such environments than the average species on Earth. We are suggesting ways to alleviate this bias, for example conducting experiments with populations that can be tracked even if they eventually become extirpated, or monitor natural populations expected to undergo a disturbance and measuring selection at the onset of the disturbance (again, even if the population eventually becomes extirpated).
The other potential explanation for our results is that the opportunity for selection might be weaker in human-disturbed environments. The opportunity for selection is the variance in relative fitness in a population – if there is no variance, then selection cannot be strong. That is to say, if you are fitting a line through a bunch of data points with trait values on the x axis and fitness on the y axis, some individuals need to have low or high fitness or else the line will be flat (and again, the slope of that line is the strength of selection). We collected estimates of the variance in fitness from roughly 60% of studies in the database and found that some studies did report lower variance in fitness in human-impacted populations. We also found a trend: when a given human disturbance increased the variance in fitness, it increased the strength of selection, and when a disturbance decreased the variance in fitness, it also weakened selection. We cannot say for sure why some disturbances increase the variance in relative fitness while others lower it. However, we speculated that some form of human-induced environmental change might be beneficial for some species – for example warming or higher CO2 can stimulate growth in some species even if it is detrimental to others. If researchers just happened to study such combinations of species-disturbances in which disturbance had a positive effect on the population, then it is normal that we do not find strengthen selection in human-disturbed environments.
We concluded the paper by discussing the challenges of studying human impacts on selection – including the taxonomic bias discussed above, but also problems with measuring fitness (which is very hard in many species). Despite all of these challenges, we suggest that measuring human impacts on selection will reveal some disturbances that consistently induce large evolutionary effects, and that identifying such disturbances is key to understand and mitigate human impacts on biodiversity. We suggested that fishing is one such disturbance, in fact the only one that we could identify in this synthesis exercise.
The future step for this work is to keep collecting data as they arise in the scientific literature – our dataset is large, but all the data come from ‘only’ 37 species. This might be one criticism that other scientists will have of this work (although we did collect all data that currently exist!). There are billions of species on Earth, and many human disturbances for which we could not find data. More data will also allow splitting studies into various categories (for example those focusing on climate change vs. habitat degradation, or those on animals vs. plants, etc.), which will help identifying disturbances with large impacts, taxa that are most vulnerable, and so on. We tried to do this, but data were too few to reach robust conclusions (this might also be a criticism of our study). All of this calls for additional studies of selection in human-altered environments!
This work is important because it involved a large literature review to build a synthetic dataset in human-induced evolution, a sub-discipline of biology for which there is currently much interest. Our analysis complements other meta-analyses that have focused on populations in more natural environments. The dataset was also uploaded to a data sharing website (FigShare) and is in open access, such that anyone can use it to replicate or extend our analyses. We also reached a surprising result; our guess is that most people would have predicted stronger selection in human-disturbed environments, but we did not find that. We expect that this negative result and our tentative explanations for it will generate some discussion in the evolutionary biology community working on human-induced evolution. Finally, we are not arguing that human disturbances do not cause evolution; in fact, I certainly believe the opposite. However, quantifying the overall impact of humans on selection across the biosphere is obviously very challenging. These challenges served as the scaffold for the discussion section of the paper, and what we outlined could help guide future studies of selection.
The paper is HERE
The pictures are HERE