Why functional trait ecology needs population ecology

I have an embarrassing confession: I’m just not that into you trait-based ecology.

Which doesn’t feel like confessing a murder, but does feel like confessing, I dunno, not liking Groundhog Day.* It’s slightly embarrassing. For years now trait-based ecology has been one of the biggest and fastest-growing bandwagons in ecology. Plenty of terrific ecologists whom I really respect are really into it. Which doesn’t mean that I have to be into it too, of course–but which does mean that if I’m not into it, I’d better have a good reason.

Which is a problem, because honestly I’m not sure why I’m not into it. In a field like ecology, where there’s no universal agreement as to what questions are most important to ask or exactly how to go about answering them, I think it becomes more (not less) important that each of us be able to justify our chosen question and approach, in terms that others can appreciate if not necessarily agree with. And also justify not liking any questions or approaches we don’t like. It really bugs me when people object to my own favorite approach for weak reasons that don’t stand up to even casual scrutiny. So I’m embarrassed to admit that there’s lots of trait-based ecology that I just vaguely think of as uninteresting or not likely to go anywhere, even though honestly I don’t know enough about it to really have an informed opinion. It’s embarrassing to not have an informed opinion on what’s probably the most popular current approach to topics that I care a lot about (e.g., species diversity, composition, and coexistence along environmental gradients).

This post is my attempt to do better. I want to think out loud about what I like and don’t like about trait-based ecology. My selfish goal is to clarify my own thinking, and to get comments that will teach me something and help me think better. My less-selfish hope is that buried somewhere within my half-formed thoughts are some useful ideas that trait-based ecology could take on board.

Here’s my plan: I’m going to talk about a body of work in trait-based ecology that I actually do know well and that I do like a lot. Then I’m going to go back to Brian’s old post on where trait-based ecology is at and where it ought to go and see how this body of work stacks up. How do my reasons for liking this particular body of trait-based ecology line up with what an actual trait-based ecologist–Brian–looks for in trait-based ecology?

The body of work I’m going to talk about is from Elena Litchman, Chris Klausmeier, and their group at Michigan State. They’re phytoplankton community ecologists. Which immediately raises the question “What ‘traits’ do unicellular algae have and how on earth does one measure them?” Glad you asked, Imaginary Reader!

For decades, one of the core tools of phytoplankton ecology has been the chemostat: a continuous flow-through culture. Culture medium flows into the culture vessel, and the medium and organisms in it flow out at the same rate:

Chemostat. From nyu.edu.

You grow a population of algae in the culture vessel, vary key aspects of the environment (e.g., the concentration of limiting nutrients in the inflow medium), and measure parameters that describe algal population growth. Those parameters are “traits”, or at least can be thought of as traits (I’ll come back to the issue of whether those parameters are “really” traits). For instance, under constant conditions (meaning: no fluctuations over time in flow rate, the inflow nutrient concentrations, the temperature, etc.), algal per-capita growth rate is an increasing saturating function of the inflow concentration of the limiting nutrient, traditionally denoted by S. That increasing saturating function is described by the Monod model, which can be completely specified by two parameters. The choice of parameters is somewhat arbitrary (there are multiple mathematically-equivalent ways to write the Monod model), but one traditional choice is to use the half-saturation constant Ks (the nutrient concentration at which per-capita growth rate is half its maximum possible value), and the maximum per-capita growth rate μmax:

Illustration of the Monod model, from montana.edu. Per-capita growth rate μ is an increasing saturating function of S, the concentration of the limiting nutrient in the inflow medium.

Those parameters vary interspecifically. For instance, here’s the fit of the Monod model to chemostat data for the growth rate of two algal species as a function of phosphorus concentration in the inflow medium:

Per-capita growth rate of two diatom species as a function of phosphorus concentration in the inflow medium. One species (open circles) has a higher μmax than the other (filled circles).  From Steiner et al. 2012.

There are also other traits one can measure in chemostats, such as cellular nutrient quotas (basically, the minimum amount of nutrient X needed to make a new algal cell). Cellular nutrient quotas are parameters in a more complicated growth rate model (the Droop model) one can use to describe algal growth in a chemostat in which flow rate and/or inflow nutrient concentrations vary over time. You also can measure parameters describing growth rate as a function of light levels. You can measure composite parameters like Tilman’s famous R*, the equilibrium level of the limiting resource in a chemostat in which the flow rate and inflow nutrient concentrations are held constant. You can even measure a trait that all you macrobial ecologists would find familiar: body size (i.e. cell size). And one could of course drill down and measure various “low level” physiological, biochemical, and gene expression traits.

But you don’t have to drill down to the underlying physiology and biochemistry when ecologically-relevant effects of those low-level traits are so well-summarized by the “high-level” parameters like μmax, R*, etc. I mean, look at the great fit of the Monod model to the data in that last figure (which is a typical example, by the way)! To which you might reply, pfft, whatever, that’s for a controlled lab environment. There’s no way the laboratory-assayed parameters of the Monod model or the Droop model or etc. will tell us anything about what happens in complex natural environments. Where there are other algal species, and zooplankton, and water turbulence, and evil demons that constantly change the laws of nature just to make life hard for field ecologists.**

You might say that, Imaginary Interlocutor I Invented For Rhetorical Purposes, but you’d be wrong. Elena Litchman, Chris Klausmeier, and their group members have been compiling phytoplankton trait data from the literature. Which leads to several very nice empirical results. First, there are strong cross-species correlations and trade-offs among different parameters:

Three-way trade-off between competitive abilities for N (x-axis) and P (y-axis), and cell size (indicated by size of the circle). The measures of competitive ability are composite parameters reflecting how well species take up N and P, relative to their cellular requirements for N and P. Each point is data for one algal species. Filled circles are freshwater species, open circles are marine species. From Edwards et al. 2011.

Those correlations and trade-offs are interpretable in terms of lower-level physiological and biochemical mechanisms, which is why it’s legit to think of those high-level parameters as “functional traits”. For instance, smaller species are in the upper-right of the above plot (where better nutrient competitors are) and larger species in the lower left because of surface area-volume ratio. The most competitive species for limiting nutrients are the ones with the lowest requirements for those nutrients (i.e. lowest cell volume, all else being equal), relative to their ability to meet their requirements via nutrient uptake through uptake proteins on the cell surface.

Second, many traits are taxonomically or phylogenetically conserved, though of course with substantial variation within groups. So if you see species’ traits as a window into the ultimate evolutionary-historical determinants of current community structure, well, here’s what the view through that window looks like for phytoplankton:

Boxplots of light level at which per-capita growth rate is maximized. Each box summarizes data for species from a different taxonomic group. Modified from Schwaderer et al. 2011.

Third, because the traits for which we have data are the parameters of theoretical models describing population growth, we can use those theoretical models to predict which species with which traits should competitively dominate in which environments. We can then go out into nature and see if species are present or abundant where we’d predict they’d be competitively dominant, and absent or rare where we’d predict them to be outcompeted. This approach famously worked for David Tilman (1977). It still works now that the Litchman-Klausmeier group has compiled data on traits and presence/abundance for many more species in many more lakes. For instance, cyanobacteria do in fact have lower relative abundance in low-light lakes, as you’d predict from their light use traits as compared to the light use traits of other species:

Ok, that’s not the most impressive R^2 ever. But given that we used one laboratory-assayed trait to predict relative abundance in the field, and then just looked at snapshots of relative abundance and a crude measure of light availability to test that prediction, any R^2 significantly >0 is something. Plus, this is just one illustrative example. From Schwaderer et al. 2011.

Anyway, you should totally spend an afternoon reading up on the Litchman-Klausmeier (“LK”) group’s work if you have any interest in trait-based ecology. It’s outstanding work.

But what makes it so outstanding? Why do I love this stuff but not lots of other trait-based stuff? To answer that, I’m going to go back to Brian’s old list of 9 things that trait-based ecology needs to do to fulfill its potential. The LK work does some of them–but also kind of does the opposite of others! My list is numbered to correspond to Brian’s:

1a. LK use an unconventional definition of “trait”. Brian argued that it’s not a functional trait if (i) it’s not something you can measure on an individual organism, and (ii) it’s not linked to individual performance. He complained that too many people are ignoring (i). Well, sorry Brian, but LK’s work ignores (i), and it still works great. The reason being that individual morphology/physiology/biochemistry/gene expression/etc. only matters for species distribution and abundance insofar as it affects per-capita growth rates. If you can summarize the effects of all that low-level morphology/physiology/etc. on per-capita growth rates with a few high-level parameters in a mathematical model, you should totally do it. This is an example of what philosophers of science call “screening off”. If A only causally affects C via its effect on B, you only need to know about B in order to capture A’s effect on C. B is said to “screen off” A’s effect on C. Here, “A” is “cellular biochemistry, etc.”, “B” is “parameters of the Monod or Droop model”, and “C” is “algal abundance under specified environmental conditions”.

1b. LK embrace population ecology rather than trying to avoid it or skip over it. Brian refers to the focus on individual organisms as a “subtle rejection of the population approach”. Which as LK’s work illustrates is maybe one reason why the functional trait bandwagon needs some steering. As discussed in the previous paragraph, LK’s work treats parameters of population growth models as traits! Which seems to me like a really good idea if what you’re ultimately interested in explaining is the distribution and abundance of species. And if those models don’t exactly describe any particular natural system (which they don’t), well, as I just illustrated they still generate testable predictions from which we can learn. And note as well that caring about the parameters of population growth models doesn’t mean not caring about the underlying individual-level morphology/physiology/etc. Chris Klausmeier for one has done some eco-evolutionary modeling of algal physiology and biochemistry to try to explain interspecific variation in population growth parameters. I think one take-home message of LK’s trait-based work for trait-based ecology in general is to quit trying to cut out the “middle man” of population ecology. Quit insisting on always and only trying to go straight from individual-level morphology/physiology/etc. to explanations of distribution and abundance. Instead, go from individual traits to parameters of population ecological models, and from parameters of population ecological models to testable predictions of species’ distributions and abundances.

But really, in saying everything in 1a and 1b, I don’t think I’m saying anything different than what Brian said in his #3: traits are hierarchical, and “higher level” traits covary more with the environment than lower level traits. Which is exactly what one would expect (right?) And which I think is a strong reason to use lower-level traits to explain higher level traits, which then in turn explain abundance. I’m just highlighting that one can and should go to an even higher level than, say, whole-plant carbon use efficiency or whatever. One can go up to the level of parameters in population growth models–Ks values, R* values, etc. in the case of phytoplankton–and still be doing trait-based ecology.

[Aside: There are other examples I could’ve used to illustrate the same points in 1a and 1b. For instance, one can explain a substantial fraction of variation in relative abundance of plant species in N-limited grasslands via their R* values for soil nitrate (Harpole and Tilman 2006). One can of course also explain interspecific variation in R* via interspecific variation in lower-level physiological traits (Craine et al. 2002). I’m not arguing that R* values in particular will always be the high-level traits of interest. It’s just that the best examples that happened to occur to me as I was writing this post all happened to involve R* values.]

2. Embracing variation. LK’s data definitely could be used to look at interspecific trait variation within sites, though as far as I recall they haven’t gone down that road very much yet. They don’t have any data on intraspecific trait variation as far as I know. But I don’t see any obstacle in principle to collecting such data. Just do chemostat trait assays on a bunch of different asexual lines of the same algal species. That would be work, of course, but nobody said ecology was easy.

3. Traits are hierarchical. I talked about this already.

4. Traits aren’t univariate and don’t just reflect 1-D trade-offs. LK’s work is a nice illustration of Brian’s point; see e.g. the 3-D trade-off plot above.

5. Embrace trait x environment interactions. In other words, different traits or trait combinations are best in different environmental contexts. LK’s work is a nice illustration of this point.

6. Theory. As discussed in 1a and 1b above, LK’s work is a powerful illustration of what embracing theory can look like in trait-based ecology. I share Brian’s sense that a lot of trait-based ecology is far removed from theory and that it would really benefit from tighter links with theory. But where I perhaps differ from Brian–and again, this may reflect my own ignorance of much of trait-based ecology–is my sense that the required theory (or at least many of the ingredients for it) already exists in population ecology. Now, embracing theory as they did required LK to take advantage of the accumulated decades of trait measurements by many algal ecologists whose trait measurements have been motivated by the Monod model, the Droop model, etc. LK were lucky that those data were already available (though I’m sure compiling those data was still a lot of work!) But that just illustrates what I’m sure is a familiar point to any trait-based ecologist: you want good trait data for lots of species. Ideally you want data for the traits that theory says are most relevant, rather than for whatever ol’ traits happen to be easiest to measure. If those data don’t exist, you either have to put in the grunt work to collect them, or go work on something else.

This kind of brings up a larger point. My admittedly-vague sense is that one major motivation for a lot of trait-based ecology is a desire for scalability. We want to be able to predict or explain something about any (every!) community. But I get the sense that scalability is somehow supposed to be easy, at least compared to the alternatives. We want a scalable approach that lets us predict or explain every community in one go, because the alternative of studying each community one by one is too much work. LK have scalability in the sense I’ve defined–but only because lots of algal ecologists spent decades doing finicky chemostat assays of theoretically-relevant growth parameters. Being able to work “at scale” in the sense I’ve defined isn’t easy–it often requires a massive up-front investment in data collection.***

7. Individuals. LK only have species-level data, not individual-level data. I already argued that individual-level data isn’t essential for trait-based ecology in general, though there may be specific questions for which it’s essential.

8. Traits besides physiological/morphological traits. LK’s work is a great example of trait-based ecology using traits that are directly relevant for the outcome of species interactions.

9. Traits beyond plants. Phytoplankton are primary producers, but in other ways they’re very different from terrestrial plants. So I’d say LK’s work addresses this need.

So what do you think? Is the LK work a model for trait-based ecology to follow? Good work, but not an example anyone else could emulate? Good work, but far from unique because lots of trait-based ecology already does the same sort of thing just as well? Not really trait-based ecology at all? More broadly, what do you think is the most exciting trait-based ecology being done right now? And are there any popular lines of trait-based research that we could use less of? Really looking forward to your comments.

FYI: I’m going to stay out of the thread for the first few hours to let the conversation develop its own momentum before replying to anyone. I want to take the time to reflect on the first wave of comments and make sure I reply thoughtfully, and I don’t want to dominate the conversation.

*Hypothetical example. I actually love Groundhog Day.

**Here’s a picture of one of the evil demons. 😉 (UPDATE: see here if you don’t get the–extremely obscure–joke. Now you know why I failed to make my college’s improv comedy group despite trying out four times.)

***I also think “scalability” in the sense I’ve defined is overrated. But that’s another post.