In 1983, following up on a little-known paper published by Karl Pearson eighty years earlier, Russell Lande and Stevan Arnold, developed statistical methods to measure selection based only on changes in population phenotypic traits within a generation. Thirty-six years after the paper was published, I asked Stevan Arnold about the making of this paper, his collaboration with Russell Lande, and what we have learnt since about the topic of this paper.
Citation: Lande, R., & Arnold, S. J. (1983). The measurement of selection on correlated characters. Evolution, 37(6), 1210-1226.
Date of interview: Questions sent via email on 15th January, 2019; responses received via email on 28th December 2019.
Hari Sridhar: I would like to start by asking you about your motivation to do the work presented in this paper. This paper seems to have come around the time when you started becoming interested in evolutionary theory. Before this paper, most of your work was empirical. Could you tell us what got you interested in theoretical evolutionary biology at this point in time?
Stevan Arnold: Shortly after I moved to the University of Chicago (UC) in 1974, Jonathan (now Joan) Roughgarden came through to give a seminar and I showed him around my lab. I was in the early planning stage of comparing feeding scores in newborn garter snakes and estimating heritabilities in two populations. Jonathan remarked that the heritability idea was the most interesting to him and that advice helped steer me in that direction.
HS: How did this collaboration between Russell Lande and you come about and what did each of you bring to the paper?
SA: Lande was hired at UC soon after I was and we struck up a friendship. Monty Slatkin, who was also interested in evolutionary theory, lived in the same apartment building as me, and was in the same department as Lande. As a consequence, I was used to having conversations with both of them on the projects they were working with. Likewise, my friendship with Mike Wade developed into a kind of mentoring. In retrospect, those conversations were a kind of mentoring for me, since I had no formal training in either population or quantitative genetics.
Here’s an excerpt from my slide show on measuring evolution that is relevant to the second part of the question: “This project began with a recreational examination of the selection gradient described in Lande 1979. In the course of writing out the terms for the two-trait case, Arnold realized that the terms in Pinverse *s seemed to represent a partial correlation. Lande was told about this epiphany the next day and quickly determined that beta was actually a vector of partial regression coefficients. The paper quickly fell into place over the next several weeks.”
The significance of the partial regression realization was that it moved the selection gradient from being completely theoretical to something that could be estimated from data.
HS: Stepping back a bit, how did you get interested in biology and evolution?
SA: As a high school student in southern California I was part of small group of reptile enthusiasts who hunted snakes in the deserts. In the course of those trips, I became familiar with geographic differences in the size and coloration of several snake species in California. As an undergrad at Berkeley, I read E. Mayr’s 1963 book (Animal Species and Evolution) and especially the parts describing how geographic variation represented the early stages of evolution. This kind of micro-evolution was real to me and my field experience helped guide me to a study of micro-evolution in garter snakes as a young Prof at UC.
HS: You say “To demonstrate the utility of our approach, we analyze selection on four morphological characters in a population of pentatomid bugs during a brief period of high mortality. We also summarize a multivariate selection analysis on nine morphological characters of house sparrows caught in a severe winter storm, using the classic data of Bumpus (1899).”
How did you choose these particular datasets?
SA: In 1983, relevant multivariate datasets (scores on multiple traits and a component of fitness in a reasonably large sample of individuals) were rare. The complete Bumpus dataset was available in the literature and had already been used to illustrate methodology in a couple of papers. For these reasons, it was a natural dataset to use. On the other hand, the sample was not huge and the selection acting on the event was not strong, so it did not do a good job illustrating the potential of our approach. Consequently, we were looking for a second dataset when I stumbled onto the pentatomid samples.
HS: You say “On the morning of April 19, 1981 adult pentatomid bugs (Euschistus variolarius PB) were found along the shoreline of Lake Michigan in Porter Co., Indiana. The bugs, along with several species of carabid and chrysomelid beetles, were common in the zone of waft debris, about 1 to 3 m from the water’s edge. The first 94 bugs encountered in this zone were collected and of these only 39 were alive. The 55 dead bugs were in excellent condition and had apparently died recently. We suppose that the bugs were knocked down into the lake during a storm as they attempted to migrate north over Lake Michigan and were then washed ashore. There was a light rain during the morning when the collection was made, and live bugs were inactive and apparently incapable of taking flight.These circumstances suggest that we obtained an unbiased sample of bugs washed ashore during the storm.”
Could you please provide some background to this intriguing natural history encounter, which has played such a crucial role in theoretical evolutionary biology?
SA: A friend (Jerome Harding) and I decided to escape from the urban Chicago and made a trip around the tip of Lake Michigan to the Indiana Dunes one weekend morning. We were walking along the lakeshore, we noticed the abundance of insects among the debris above the water line. It suddenly occurred to me that we were looking at a Bumpus-like sampling situation. I quickly showed Jerome, a computer programmer, how to capture and save two samples of the most abundant insect, the pentatomid bug.
HS: Could you give us a sense of how long this work, including the writing, took, and when and where did you and Russell Lande do the work?
SA: I don’t have the folder for this project with me, so I will have to guess. After the partial regression realization, the project moved rapidly. I think only several months transpired before we submitted a paper. Most of the work was done in Chicago at UC, but I remember talking about a manuscript with Lande at my field site at Eagle Lake in northern California. Lande had come along on that trip just for fun, but we spent some time talking business around the campfire.
HS: Could you tell us a little more about the people you acknowledge – how you knew them and how they helped?
SA: Some of the people on the list below must have read the manuscript for us but I don’t know which ones. I may have info in the project folder. Will let you know what I find there.
J. Antonovics – I heard Janis use the phrase ‘genetics without tears’ in a seminar.It struck a chord with me because it symbolized the courage one needed to adapt theory to data in data analysis. In other words, he helped inspire,
J. Bradbury – He had attended a Dahlem Conference with me at which we had wide-ranging conversations about evolutionary biology. I can’t remember the details of his contribution.
M. Bulmer – I’d just done a one term sabbatical leave with Michael at Oxford in 1983. I must have discussed the manuscript with him.
D. Burdick – A statistical consultant at Duke, I believe. Lande talked to him.
T. Clutton-Brock – I visited Tim on the Oxford trip and had lots of conversations with him. The work with Lande must have come up.
G.E. Dickerson – An established figure in quantitative genetics, Lande must have talked with him.
P.R. Grant – He was in the process of doing his famous work on selection in finches in the Galapagos. Either or both of us could have talked with him about our selection manuscript. In either case, he and his students became early and influential adopters of our method.
T. Nagylaki – A colleague of Lande’s in Biophysics. Lande probably talked to him – as he often did – about arcane mathematical details.
L. Van Valen – Lande talked to him about the project.
Jean Gladstone – My laboratory assistant, charged with handling the project data files and running the analysis programs at the campus computer center (no desktop at the time).
John Steadman – An undergraduate working in my lab. He did the measurements on the pentatomid bugs.
Daniel Summers – An entomologist who, I believe, worked at the Field Museum. He identified the pentatomid.
HS: Was Evolution the obvious choice for this paper? Did you consider other journals? What do you remember of the peer-review process of this paper?
SA: Evolution was the obvious choice, especially given the technical detail in the paper. Evolution, then as now, was popular journal for both theoreticians and empiricists and we wanted to reach both audiences.
HS: At the time this was published, what kind of attention did it receive in academia?
SA: A primary ingredient for being able to apply our methods was scores of one or more fitness components along with trait measurements. Such datasets were still relatively rare in 1983 even though David Lack had pioneered a field approach focused on fitness a few decades before. In other words, the population of potential users was relatively small and hence reaction to the paper was slow at first. The paper was more widely discussed by theoreticians early on. I remember Lande relating to me the 1983 reactions of Alan Robertson (the leader of the quantitative genetics group at Edinburgh). When Russ asked what he thought of our findings, Alan said he wasn’t surprised by the linear selection gradient result (beta). Russ asked what he thought of the analysis of stabilizing selection (gamma). “Oh, that’s completely new” he said.
HS: Could you reflect on the impact of this paper on your career and future research trajectory?
SA: The paper certainly had a major effect on my visibility, particularly as its methods became widely used. I also used my understanding of the foundations of the paper to write other papers along the same lines. Some of these (e.g., Arnold 1983 AmerZool) help make the original paper with Lande more accessible, especially to field workers.
HS: Today, 35 years later, could you reflect on the validity of the study’s main takeaway in relation to current theory:
“Following an approach pioneered by Pearson (1903), this analysis helps to reveal the target(s) of selection, and to quantify its intensity, without identifying the selective agent(s). By accounting for indirect selection through correlated characters,separate forces of directional and stabilizing (or disruptive) selection acting directly on each character can be measured.These directional and stabilizing selection coefficients are respectively the parameters that describe the best linearand quadratic approximations to the selective surface of individual fitness as a function of the phenotypic characters.”
SA: All of that is still true, but many of these statements have important provisos that are detailed in the paper itself.
HS: You say “Some important questions which can be approached with the techniques we have described include the following: Do selective surfaces commonly have multiple peaks? How frequent is maladaptation of the mean phenotype in a population? What is the magnitude of geographical and temporal variation in selective surfaces? Studies designed to answer such questions should be done on a number of animal and plant populations.Only then will we have adequate information on the intensity and pattern of phenotypic natural selection.”
Could you tell us to what extent these questions have been answered since the time of your paper?
SA: Nearly all of these issues have been addressed in the many papers that cite Lande & Arnold 1983 and implement its methods (currently about 4500). Probably the best place to get an overview is in the papers by Joel Kingsolver and colleagues that compiles estimates of beta and gamma and discuss them. The one exception on your list is the multiple peak issue, which is still a source of disagreement if not controversy.
HS: In the 35 years following this paper, have you ever read this paper again? If yes, in what context?
SA: I am writing a treatise on Phenotypic Evolution and often need to consult Lande & Arnold 1983. Some results were derived by making assumptions about the strength of stabilizing selection or normality of trait distributions. Those dependencies I find difficult to remember offhand.
HS: Would you count this as a favourite among all the papers you have published and why?
SA: Definitely a favorite, if not the favorite. This paper stands out in impact, as measured by citations, for example.
HS: What would you say to a student who is about to read this paper today? Would you guide his or her writing in anyway? Would you point them to other papers they should read alongside? Would you add any caveats?
SA: Early on, Peter Grant told me that we should write a short follow-up paper to make the methods more accessible to the average reader. Unfortunately, we did not take that advice and the paper is, today, still a very difficult read. Arnold 1983 AmerZool comes close to providing what Peter was asking for, but it does not discuss the details that can snag a user. Stinchcombe et al 2008 Evolution discuss a misunderstanding of the method that has snagged many users. I often think that if we had followed Grant’s advice we might have saved hundreds of users from going astray.
0 Comments