The concept of multilevel selection

When the Strong Outbreed the Weak: An Interview with William Muir by William Muir and David Sloan Wilson The idea that people should be bred for desirable characteristics, like farm animals, has a long pedigree. Darwin’s theory wasn’t required—after all, it was Darwin who relied upon animal breeding practices to explain how nature plays the same role as the farmer. Nevertheless, some of the most famous people who followed in Darwin’s footsteps, including Francis Galton and Ronald Fisher, were unabashed eugenicists. Even Darwin conceded that people could be bred for desirable characteristics and counseled against it only because it would violate our moral instincts, which he regarded as the most distinctive product of human evolution. Almost everyone who thought about eugenics at that time unquestionably assumed that creating a better society was a matter of selecting the ablest individuals, or “hereditary genius”, as Galton put it.

Against this background, consider an experiment conducted in the 1990s by William M. Muir, Professor of Animal Sciences at Purdue University. The purpose of the experiment was to increase the egg-laying productivity of hens. The hens were housed in cages with nine hens per cage. Very simply, the most productive hen from each cage was selected to breed the next generation of hens. If egg-laying productivity is a heritable trait, then the experiment should produce a strain of better egg layers, but that’s not what happened. Instead, the experiment produced a strain of hyper-aggressive hens, as shown in the first photograph. There are only three hens because the other six were murdered and the survivors have plucked each other in their incessant attacks. Egg productivity plummeted, even though the best egg layers had been selected for each and every generation. The reason for this perverse outcome is easy to understand, at least in retrospect. The most productive hen in each cage was the biggest bully, who achieved her productivity by suppressing the productivity of the other hens. Bullying behavior is a heritable trait, and several generations were sufficient to produce a strain of psychopaths.

In a parallel experiment, Muir monitored the productivity of the cages and selected all of the hens from the best cages to breed the next generation of hens. The result of that experiment is shown in the second photograph. All nine hens are alive and fully feathered. Egg productivity increased 160% in only a few generations, an almost unheard-of response to artificial selection in animal breeding experiments. Muir’s experiments reveal a tremendous naiveté in the idea that creating a good society is merely a matter of selecting the “best” individuals. A good society requires members to work together to create what cannot be produced alone, or at least to refrain from exploiting each other. Human societies approach this ideal to vary degrees, but there is always an element of unfairness that results in some profiting at the expense of others. If these individuals are allowed to breed, and if their profiteering ways are heritable, then selecting the “best” individuals will cause a cooperative society to collapse. It’s a good thing that the early eugenicists did not have their way! Muir’s experiments also challenge what it means for a trait to be regarded as an individual trait. If by “individual trait” we mean a trait that can be measured in an individual, then egg productivity in hens qualifies. You just count the number of eggs that emerge from the hind end of a hen. If by “individual trait” we mean the process that resulted in the trait, then egg productivity in hens does not qualify. Instead, it is a social trait that depends not only on the properties of the individual hen but also on the properties of the hen’s social environment.

Ever since Muir’s experiments were published, I have been using them to illustrate the concept of multilevel selection and as a parable for thinking about human social evolution. However, their implications for animal breeding practices are important in their own right. Very few domestic animals are housed as individuals. This means that selection for bullying traits might be taking place even when it isn’t intended, resulting in decreased productivity from the human perspective and increased suffering from the animal perspective. Parenthetically, plant breeders face a similar problem. A corn plant that produces big ears by suppressing the productivity of its neighbors is little different than a hen that produces many eggs by bullying her cage mates. My interview with Bill is centered on animal husbandry practices, but we also touch upon the parallels between human genetic and cultural evolution. DSW: Welcome, Bill, to This View of Life. WMM: Thank you for this opportunity to comment on those experiments. I have been working on the concept of individual vs. group productivity since I was a graduate student and found parallels in insects, plants, and many domesticated animal species. However, the idea of using those concepts on humans never crossed my mind, as with Darwin, I consider the concept of eugenicists against our moral and ethical principles. DSW: I don’t propose that we breed people like farm animals in any sense! Let’s begin with that great experiment of yours. In my introduction, I observed how surprising it is against the background of the early proponents of human eugenics. How surprising is it against the background of Animal Science and what gave you the idea to do it? WMM: It was not very surprising to me at all. In fact, I expected the results.

Earlier experiments with Tribolium titanium (flour beetle) during my Ph.D. thesis at Purdue University showed that selecting beetles that could survive on minimal amounts of food in a group environment did so at the expense of others. The selected population evolved to grow faster so they could pupate first and eat the slower-growing pupae, i.e. they became cannibalistic and the winning strategy was to grow the fastest with the slower fueling the growth of the survivors. After graduating and teaching for a few years at the University of Kentucky, my first job as an animal breeder was back at Purdue University as a poultry geneticist. The curious result with beetles motivated the experiments with poultry. The only result that was really surprising to me was the amazing rate of response of the group-selected birds, which showed a realized response greater than the selection differential, i.e. the realized heritability was greater than 1 for the first few generations. All of these results were beyond “classical theory” taught at the time but were found to be due to associative effects, i.e. the trait “group productivity” capitalized on social effects, of which there are many. Cumulatively those social effects were much more important than that of the individual. I have since demonstrated the same finding when selecting for growth rate in quail and swine.

DSW: Tell me more about your educational background, which appears to have been unorthodox for an animal breeder. You anticipated your results, but were they more surprising to your colleagues with a more standard animal breeding training? Were standard animal breeding practices unknowingly selecting for antisocial traits in chickens and other farm animals? WMM: In many ways, life is a random walk, so too was my education. During the great “experimentation” days of quantitative genetics in the 1960-70s, everyone in academia was testing quantitative genetics theory, which means there was great doubt that the infinitesimal model was correct because of the many assumptions. The infinitesimal model assumes an infinite number of genes each with small effects such that the entire process can be modeled by statistics with means and variances, rather than genes with Mendelian segregation ratios. We knew the model was wrong in detail, but was it right in practicality. Of interest was long term selection and how many generations we could select animals before the beast ran out of genetic variation. The only way to answer that in real time was to use a model organism with a short generation interval. One such model organism was Tribolium castanium because it had a generation interval of a month and was easy to rear. My major professor, Dr. Earl Bell, was a pioneer in use of the beetle to study quantitative traits, such as pupae weight, egg number, and development rate. As a graduate student I was interested in how competition influenced the outcome of long term selection and set my experiments up to examine the results, i.e. when selecting for survival on limited amount of feed, did it matter how many individuals were in the group given that the total amount of food was adjusted to give a constant amount per individual. The results showed that classical theory based on non-interacting individuals was wrong. The number of individuals in the group did matter, profoundly. However, it was hard to convince animal breeders that these “bugs” had anything to do with animal breeding. In fact the state of the art for animal breeding was called BLUP, or Best Linear Unbiased Prediction. Statistical theory showed that this method would give the optimal response to selection, above all others, because it combined information from all sources and weighted them optimally. However, in order to use the theory phenotypes on 25 individuals were needed. So pig breeders could measure individual growth rate and make optimal rate of progress, or so it seemed. In contrast chicken breeders in the layer industry (eggs) could not measure performance of individuals in group cages because one could not tell from which bird the egg came from. To address this problem, they redesigned the testing facility to put birds in individual cages. Now they had individual performance data, but unfortunately they did not make much progress. This was about the time I started my research at Purdue. I saw that they set themselves up for what is called a genotype x environment interaction, i.e. the environment of selection (single bird cages) was different from the environment of production (colony cages). I reasoned that competition was the culprit; however, few believed me, and because birds could be beak trimmed to remove their “weapons” the problem could be largely addressed by management. However, I did not accept the management solution because it was costly and compromised the wellbeing of the bird; i.e. it hurt and continued to hurt. I suggested that group selection would solve the problem. Few believed me and several laughed openly because everyone knew that “group selection” has been debunked, i.e. early myth-busters claimed it could not happen, too many problems. Few bothered to examine what the problem with group selection theory was, only that the “experts” said it would not work. (Group selection is now referred to by the more general term “multi-level selection” of which group selection is a special case.) However, I saw that all the theoretical issues with group selection could be overcome in a breeding situation, i.e. we could control the group size and composition (relatives, and no migrants) as well as who mated to whom. When I started the experiment, my major professor said he thought it was a major risk, “I was gambling my tenure on the experiment”, he was not sure it would work convincingly enough to publish and gain the recognition needed for tenure. When I first presented the work at an international poultry wellbeing conference, the poultry company whose birds I used to demonstrate the results were outraged because I did the experiment was conducted in full light and did not beak trim the birds. I told them that this was the future, change was needed, and the theory they were using was wrong as it did not include indirect genetic effects. But I was a young pup and they did not believe me, nor did the company change, so they did not flourish. To me this is another example of group selection theory operating at the breeder level. Another company that was doing sire family (half-sibs) selection in group housing did flourish and is now the largest poultry egg producer in the US, this is largely due to improvements in livability due to group (family) selection. The company was trying to mimic maize (corn) breeding that uses a combination of inbreeding (to create family lines) and ear-to-row planting whereby the row was the unit of selection, not the individual. Plant breeders were by coincidence using group selection, not because they were interested in reducing competition, but because there is too much variability among individual plants. They found that the average yield of a row was much more stable and assumed this result was because of the law of averages, which is true but was also due to co-selecting groups that interacted well within families. Thus when the chicken breeders mimicked the maize breeding scenario, they also unknowingly adopted a selection method that guaranteed competitive behavior will be reduced and animal welfare improved. [sections cut on the topic of animal husbandry] DSW: Were your results more surprising for your colleagues with a more standard training in animal breeding? WMM: Yes, absolutely, most could not believe the rapid responses observed in the first few generations and concluded there must be a major gene. They could not conceive that the infinitesimal model could produce the same thing. Even recently I presented the work at a prestigious European university, and when I remarked the heritability of the trait was greater than 1 in the initial generations, the immediate response was that I clearly did not know what I was talking about because, by definition, the heritability of a trait cannot exceed 1. Again, they were caught in old school thinking. With MLS in the first 2 generations of selection there was more response to selection than the SD. This is exactly what is meant by the whole is greater than the sum of the parts. In totality (as a group), we got out more than we put in (as individuals). This is the magic of multi-level selection, changing the unit of selection gives more than it takes. So reducing indirect genetic effects (competition) can be much more impactful than increasing direct (performance of the individual). Old school thinking is based on performance of the individual (direct effects). Questions: 1.what stood out most to you in this article? 2. How was Muir’s experiment received by others?