Although the notion “bandwagon” technically only means something that is rapidly growing in popularity, calling a scientific research program a bandwagon carries several more connotations. These include the idea that it will crash (people will abandon it) and that people are piling in because they perceive the research program as a way to do something that is “easy” (or even formulaic) but still get in a good journal (i.e. the proverbial something for nothing). Popular and easy are of course two of the worst reasons to choose a research project, but that seems not to matter in the bandwagon phenomenon.
There is little doubt that functional traits are a bandwagon research program right now:
The use of the phrase “functional trait*” (per Web of Science) is rising exponentially with a doubling time of less than 4 years. In less than two decades, there are almost 3000 total publications cited 56000 times, 14000 times last year alone (with an astonishing average citation rate of 19 times/article and an h-index for the field of over 100).
For better and worse, I am probably one of a fairly large group of people responsible for this bandwagon due to this paper which came out simultaneously with a couple of other papers arguing for a trait based approach, although (as likely true of all bandwagons) the idea has been around much longer and builds on the research of many people.
By calling functional trait research a bandwagon, I am implying (and now making explicit) two things: 1) The popularity of the functional trait program is in part due to the fact that people see it as a simple way to do something trendy. I think there is no doubt of this – there are a lot of papers being published right now that just measure a bunch of functional traits on a community or guild and don’t do much more. 2) That this party is about to come to an end. I predict we will see multiple papers in the next two years talking about how functional trait research is problematic and has not delivered on its promise and many people bailing out on the program.
You might think I am worried about the impending crash, but I am not. I actually relish it. Its after the bandwagon crashes that we lose all the people just looking for a quick paper and the people who are really serious about the research field stay, take the lessons learned (and identify what they are), build a less simple, more complex but more realistic, productive world view. In my own career I have seen this with phylogenetic systematics, neutral theory of biodiversity, and – if we go back to my undergraduate days – neutral theory of genetics and island biogeography.
In an attempt to shorten the painful period and hasten the renewal, what follows are my ideas/opinions about what is being ignored right now on the functional trait bandwagon (although by no means ignored by the researchers I expect will still hang around after the crash and I have tried to give citations where possible), which I predict will become part of the new, more complex view of functional traits version 2.0 in 5-10 years down the road.
(As an aside – I wanted to briefly note as a meta comment on how I think science proceeds, that: a) I think probably many other people are thinking these thoughts right now – they’re in the air, but as far as I know nobody has put them down as a group in ink (or electrons) yet and b) my own thinking on this has been deeply influenced by at least a dozen people and especially by Julie Messier as well as Brian Enquist & Marty Lechowicz – more full acknowledgements are at the bottom c) its not as easy to assign authorship on these thought pieces as it is on a concrete piece of experiment or analysis – if this were a paper I could easily argue for just myself as author or 1 more or 3 more or 10 more)
So without further ado, here are 9 things I think we need to change to steer the bandwagon:
- What is a trait? – there are a lot of definitions (both the papers linked to above have them). But the two key aspects are: 1) measured from a single individual and 2) conceivably linked to function or performance (e.g fitness or a component such as growth rate). The 2nd is not a high bar to clear. But a lot of people right now are ignoring #1 by taking values that can only be tied to a species or population (such as population growth rate, geographic range size, mortality rates) and calling them functional traits. They’re not. They’re important and interesting and maybe science will someday decide they’re more important than things you can measure on individuals. But they’re not functional traits if you can’t measure it on one individual. The functional trait program is going from function (behavior and physiology) to communities or ecosystem properties. Its where a lot of the excitement and power of the idea comes from. It is actually in a subtle way a rejection of the population approach that dominated ecology for decades.
- Where’s the variance? – I believe that the first step in any domain of science is to know at what scales and levels of measurement variation occurs. Only then can you know what needs to be explained. There has been an implicit assumption for a long time that most of the variance in functional traits is between species and/or along environmental gradients. There is indeed variation at these two levels. But there is also an enormous amount of variation between individuals in the same species (even population). And there is way more variation between members of a community than between communities along a gradient. Finally, although the previous statements are reasonably general, the exact structure of this variance partitioning depends heavily on the trait measured. Functional traits won’t deliver as a field until we all get our head around these last three facts. And learn a lot more than we already know about where the variance is. A good intro to this topic is Messier et al 2010 and Viollet et al 2012 (warning I’m a coauthor on both).
- Traits are hierarchical (can be placed on scale from low level to high level) – we tend to lump all traits together, but traits are hierarchical. Some are very low level (e.g. chlorophyl concentration per leaf volume), one level up (e.g. light absorption), and going on up the ladder from this one trait we have Amax (maximum photosynthetic rate), leaf CO2 fixation/time, CUE (or carbon use efficiency or assimilation over assimilation+respiration), plant growth rate, and fitness. Note that each trait directly depends on the trait listed before it, but also on many other traits not listed in this sequence. Thus traits are really organized in an inverted tree and traits can be identified at any tip or node and performance sits at the top of the tree. We move from very physiological to very fitness oriented as we move up the tree. One level is not more important than the other but the idea of different levels and being closer to physiology or closer to fitness/performance is very real and needs to be accounted for. And we need to pick the right level for the question. All traits are not equivalent in how we should think about them! And learning how to link these levels together is vital. A depressing fact in phenotypic evolution is that the higher up the hierarchy a phenotypic character is, the less heritable it is (with fitness being barely heritable), but so far we seem to be having the opposite luck with functional traits – higher level traits covary more with environment than low level traits (there are a lot of good reasons for this). A good intro paper to this topic is Marks 2007.
- Traits aren’t univariate and they’re not just reflections of 1-D trade-offs – How many papers have you seen where trait #1 is correlated with environment. Then trait #2 is correlated with environment, and etc.? This is WRONG! Traits are part of a complex set of interactions. If you’re a geneticist you call this epistasis and pleiotropy. If you’re a physiologist you call this allocation decisions (of resources). If you are a phenotypic evolution person you call this the phenotypic covariance matrix. Of course we are finding that one trait low in the hierarchy is neither predictive of overall performance nor strongly correlated with environment. It is part of an intricate web – you have to know more about the web. The main response to this has been to identify trade-off axes. The most famous is the leaf economic spectrum (LES) which basically an r-K like trade-off between leaf life span and rate of photosynthesis. Any number of traits are correlated with this trade-off (e.g. high nitrogen concentrations are correlated with the fast photosynthesis, short life end). And several of the smartest thinkers in traits (e.g. Westoby and Laughlin) have suggested that we will find a handful of clear trade-off axes. I hate to contradict these bright people, but I am increasingly thinking that even the idea of multiple trade-off axes is flawed. First the correlations of traits with the LES are surprisingly weak (typically 0.2-0.4). Second, I increasingly suspect the LES is not general across all scales. And the search for other spectra have gone poorly. For example, despite efforts, there has not yet emerged a clear wood economic spectrum that I can understand and explain. So to truly deal with traits we need to throw away univariate and even trade-off axes and start dealing with the full complexity of covariance matrices. This is complex and unfortunate, but it has profound implications. Even the question of maintenance of variation simplifies when we adopt this full-blown multivariate view of phenotype (two nice papers by Walsh and Blows and Blows and Walsh). For a good review of the issue in traits see the newly out just this week in TREE Laughlin & Messier
- Any hope to predict the performance consequences of traits requires dealing with the TxE (traitXenvironment) interaction – Does high SLA (specific leaf area or basically thinness of leaf, a trait strongly correlated with the rapid photosynthesis end of the LES) lead to high or low performance? The answer blatantly depends on the environment (e.g. causes lower performance in dry environments or environments with lots of herbivory). Too many studies just look at trait-performance correlations when they really need to look at this in a 3-way fashion with performance as a 3-d surface over the 2-D space of trait and environment. Presumably this surface will usually be peaked and non-linear as well (again see Laughlin & Messier 2015)
- Theory – the field of functional traits is astonishingly lacking in motivating theory. When people tell me that natural history or descriptive science is dead, I tell people its just been renamed to functional traits. I personally see descriptive science as essential, but I also see theory and the interaction between theory and description as essential. Key areas we need to develop theory include:
- How exactly filtering on traits works – one of the appealing concepts of traits is that we can move from simply saying a community is a filtered set of the species pool to talking about what is being filtered on. But we aren’t thinking much about the theory of filtering. Papers by Shipley et al 2006 and Laughlin et al 2012 are good starts but not referenced by most workers in the field. And nowhere have we got a theory that balances the environmental filter that decreases variance with the biotic competition filter that increases variance within a community (and yes Jeremy, other possibilities are certainly theoretically possible per Mayfield & Levine 2010, but for good empirical reasons, I believe this is the main phenomenon happening in traits).
- What is the multivariate structure of trait covariance – This is partly an empirical question but there are many opportunities for theory to inform on this too. In part by thinking about …
- Causes of variation – we know variation in traits are due to a combination of genetic variation and adaptive plasticity and that these respond to environments at many scales. But can we say something quantitative?
- Individuals – we are very caught up in using traits as proxies for species but I increasingly think that filtering happens on the individual level and that we need to shift away from thinking about traits at the species level. The same given trait value (say the optimal value in some environment) can be provided by any of several species, each species of which shows consider variability in traits and therefore having significant overlap in the trait distributions between species.This idea can be found in Clark 2010 and Messier et al 2010 among many others. This might seem subtle, but it is a pretty radical idea to move away from populations to individuals to understand community structure.
- Interaction traits, reproduction traits and other kinds of traits – most of the traits studied are physiological/structural in nature. This is probably because one of the major roots of functional traits has been seeking to predict the ecosystem function of plants (e.g. CO2 fixation, water flux). But if we are going to develop a fully trait-based theory of ecology we need to address all aspects of an organism including traits related to species interactions (e.g. root depth for competition, chemical defenses for herbivory, floral traits for pollination and reproduction, and even behavioral traits like risk aversion).
- Traits beyond plants – the trait literature is dominated by botanists. There is a ton of work in the animal world that deals with morphology and behavior. And some of it is starting to be called “functional traits.” The hegemony of one term is not important, but the animal and plant people thinking about these things (whatever they’re called) need to spend more time communicating and learning from each other.
So there you have it. If you want to predict outcomes (e.g. invasion, abundance, being found at location X or in environment Y, and etc) based on traits, its easy. You just have to recognize that it happens in interaction with the environment and many other traits (many of which we haven’t even started studying) and figure out what the appropriate level of traits to study for the scale of the question. Sounds easy right? No, of course not. When is good science ever easy? That’s the problem with bandwagons. Anybody want off the trait bandwagon before we get to that destination? Anybody want on if they know that is the destination?
What do you think? Are traits a bandwagon? Is it about to crash? What will be the story uncovered by those picking up the pieces? Anything I forgot? Anything I should have omitted?
PS – I don’t usually do acknowledgements on informal blog posts, but it is necessary for this one. My thinking on traits has been profoundly influenced by many people. First among them would be Julie Messier who is technically my student but I am sure I have learned more from her than vice versa. And she currently has shared with me several draft ms that make important progress on #2, #4 and #5. I also have to highlight my frequent collaborators, Marty Lechowicz and Brian Enquist. Also influencing me greatly at key points are Cyrille Violle, Marc Westoby, Evan Weiher. And this field is advancing by the work of many other great researchers (some of whom I’ve mentioned above) who were there before the bandwagon started (and many before I got on) and will still be there after it crashes but whom I won’t try to name for fear of leaving somebody out. Despite it being a bandwagon right now, there is no lack of smart people trying hard to steer constructively!