I develop mathematical, computational, and conceptual models to study natural selection and the evolution of organismal design. My work makes testable predictions on topics ranging from microbial life history to sociality to cancer. My study of particular topics leads to syntheses of natural selection, robustness in relation to biological design, and the commonly observed patterns that emerge from information flow and scale.
To make progress on the broad topics of natural selection and organismal design, I alternate between analyzing specific puzzles and developing broad empirical and conceptual syntheses.
Several of my articles call attention to novel puzzles that had not previously been noted. In Developmental predisposition to cancer (86), we argued that somatic mutations early in development cause somatic mosaicism and a significant impact on cancer risk (115, 142). Recent advances in genomic technology have led to many recent articles implicating early somatic mutations as causes of late-life diseases, such as cancer or neurodegeneration.
In Mitochondria and male disease (50), we introduced a novel consequence of maternal transmission. Males represent an evolutionary dead end for mitochondria, thus selection of mitochondria is neutral with respect to the male-specific consequences of mitochondrial genes. This selective sieve allows male-specific deleterious mutations to accumulate in mitochondrial genomes. Several subsequent studies designed to test our idea have shown that mitochondria do carry deleterious mutations with male-specific effects (132).
In other studies, I began with well-known puzzles, and then developed novel explanations that have often led to new empirical tests. Examples include the sequential dominance of antigenic variants by parasites during long-lived infections (64, 103), the variable relations between the dosage of pathogen inoculation and the probability of successful infection (104), the patterns of genetic divergence and hybrid incompatibilities between species (22), and the spatial complexity of genetic polymorphisms causing male sterility in hermaphroditic plants (13).
Solution of any one of those puzzles is by itself of significant value. In addition, I use the individual topics to build my deeper interest in all aspects of evolutionary process and biological design and to test my progress on those deeper problems.
Empirical and conceptual syntheses
The path from particular puzzles to broader impact leads to synthesis of key topics. For example, I have particularly focused on two biomedical problems: infectious disease (76) and cancer (109). The great progress in modern genetics and biomedical research has opened special opportunities to learn more about organismal design and to apply the concepts of evolutionary biology to topics of biomedical importance. My cancer work illustrates this approach.
In Dynamics of Cancer, I linked molecular process to disease outcome (109). In particular, I developed a comprehensive mathematical framework to connect the rates of genetical and biochemical changes in cells and tissues to the age-specific rate of cancer onset. I used my framework to make specific, testable hypotheses about how genetic changes alter the course of disease progression and shift the age-specific onset of cancer. I tested my theories by application to existing data on the rates of cancer in the retina and in the colon (98) and by experimental studies (97).
I also developed a novel approach and broad synthesis of the empirical literature with regard to the rate of cancer onset at different ages and in different tissues (109). I found a characteristic shape for such age of onset curves: linear on log-log scales through most of life, with a late-life decline in the slope of the incidence curve. To understand those curves that describe the ultimate outcome of disease progression, one must understand how a complex set of rate processes combines to affect, in the aggregate, the outcome of observable tumor formation.
That relation between aggregation and outcome is more general than just the cancer problem: mortality curves for different causes of death follow similarly regular patterns (96). The constancy with regard to the shape of biological failure curves suggests generic features of aggregation that apply broadly. Some of my recent work explores the connections between aggregation and common patterns (114).
Life history in relation to conflict and cooperation
In earlier work, I studied theories of life history in relation to conflict and cooperation. By the early 1980s, W. D. Hamilton had already established the genetic structure of populations as the key theory of biological cooperation. Although Hamilton correctly identified the metric that balances competitive and cooperative aspects of evolution, he was unable to use his own metric to solve his most important problems.
For example, in Hamilton’s famous sex ratio studies, males compete against brothers for matings with nearby females. Such local competition between relatives alters the way in which natural selection values males and females and changes the sex ratio. This particular problem became the touchstone by which one evaluated conceptual approaches to the study of adaptation. Yet Hamilton never successfully used his theory to explain the forces involved and to develop a broad predictive theory that could be widely applied. In consequence, the subject was wildly controversial and confused in the early 1980s (135).
I developed the theoretical foundations by which one can analyze life history evolution in complex demographic and genetically structured situations, including problems of sex ratios (59). My formal synthesis combined the complexities of genetics and demography with the inherent economic and game-like problems that concern allocation of scarce resources in potentially competitive or cooperative situations (with Peter Taylor, 47). I could apply my theory to nearly any natural setting by expressing the essential features in simple equations and using a novel method to obtain a testable prediction.
I applied my conceptual framework to develop an explicit foundation for the study of sex ratios in all organisms, synthesizing and extending much work by others. Current studies of sex ratios, dispersal and diverse aspects of life history and sociality frequently begin with my theoretical insights and analytical tools, leading to further theory and specific empirical tests. I pulled together those insights and technical advances in my book Foundations of Social Evolution (59).
Other conceptual syntheses
In addition to my books (59, 76, 109), I have recently worked on three conceptual programs. First, I wrote a series of articles on natural selection, including syntheses of past work and new analyses of major issues. Second, I unified key ways in which large-scale pattern arises from aggregation of small-scale process (114, 121, 122, 123). Third, I analyzed the synergism between robustness and evolutionary decay, which I believe pervades evolutionary history (94, 108, 138).
My conceptual syntheses often lead back to particular applications. For example, I applied my framework for demography and life history to several problems of microbial adaptation. In one case, I analyzed the harm, or virulence, parasites cause to their hosts. My article on parasite virulence extended earlier work to set the foundation and current agenda for thinking about microbial evolution in terms of demography and life history (43, also 111). This area of research remains very active. Applications range from the harm caused by different malaria strains to HIV evolution to life history problems in rhizobia, mycorrhizae, and the diverse symbionts of insects. In other studies, I analyzed the fundamental tradeoff between rate and yield in microbial metabolism (118) and the tradeoff between secretion and uptake of extracellular factors (117).
Virulence and mechanisms of pathogenesis
The problem of parasite virulence has developed into a major topic in evolutionary biology and infectious disease studies. There are two reasons for this. First, it is important to understand why parasites vary so much in the degree of harm they cause their host. Second, the concepts needed to analyze the evolution of virulence arise from fundamental aspects of sociality and life history. On the social side, parasites compete within hosts for limited resources. Outcompeting neighbors provides a benefit, but intensive competition may degrade the resource. In this case, degrading the resource means harming the host. On the life history side, harm to the host affects the parasites’ survival, altering the parasites’ balance between reproduction, measured as transmission to new hosts, and longevity, measured as how long the parasite survives before the host dies or before being cleared from the host. My 1996 review Models of parasite virulence brought together the problem of parasite virulence and the fundamental aspects of microbial sociality and life history (43).
Those theories of virulence arise from a basic understanding of evolutionary process. But the connection between abstract theories and how parasites actually make hosts sick remained vague until recently. Over the past few years, much has been learned about the biochemical mechanisms by which pathogens manipulate host tissues and host immunity. Those biochemical mechanisms of pathogenesis mediate most aspects of virulence. So the challenge becomes much richer. How do the general evolutionary processes shape the biochemical mechanisms of pathogenesis, which in turn play the direct mechanistic role in virulence? We do not know the answer, but this sort of connection between evolutionary concept and biochemical mechanism is the exciting prospect of modern biology.
We analyzed the relation between mechanism and evolutionary process in two preliminary papers. Our overview predicts how evolutionary process will shape the timing of expression of various pathogenic mechanisms in relation to the amount of harm those mechanisms cause to the host (111). We also apply this line of thought to one of the unsolved mysteries of infectious disease: why can some pathogens start an infection with a very low dose (number of cells or particles in the inoculum), whereas other pathogens require a high dose to start an infection (104)?
We predict that pathogens requiring only a low dose depend primarily on pathogenic mechanisms that act locally during invasion. For example, direct injection of molecules into neighboring host cells manipulates aspects of host physiology and immunity. This sort of local action requires only a few pathogens to alter host physiology. By contrast, we predict that pathogens requiring a high dose depend on pathogenic mechanisms that act at a distance. For example, secreted molecules must build up to sufficient concentration over long distances in order to manipulate the host immune response. Such manipulation may arise by altering host cytokine signaling or similar immune regulatory processes. Distant action typically requires larger pathogen populations during invasion to build up the concentration of a diffusible secretion. In general, the problem of relating biochemical mechanisms of pathogenesis to the population biology and evolution of infectious disease will be an important topic going forward.
Microbial life history
The theory of virulence is an example of the broader problems of sociality and life history. The application of sociality and life history concepts to microbes has grown rapidly in the past few years. For example, many bacteria secrete molecules to take up iron, which is often a limiting resource for growth. Similarly, many pathogens secrete molecules to manipulate host immunity or to create biofilms.
The secretions by an individual microbe typically act externally and benefit all members of the local group. Secreting imposes a cost in terms of growth, so cheaters that do not secrete gain by sharing the benefits without paying the costs. Microbial cheaters have been observed in several experimental and natural settings. Given that cheaters grow faster than secretors when in direct competition, what maintains the widely observed patterns of secretion?
Recent theory has emphasized the genetic structure of populations, in which secretors tend to associate spatially with other secretors, reducing direct competition and allowing highly secreting groups to share mutual benefits. Such kin selection can be a powerful force favoring cooperative traits. In my recent papers (116,117,119), I argue that, although kin selection is a factor, demographic processes often dominate in determining the relative fitness of secretors versus cheaters when measured over the full cycle of microbial life history.
Key demographic factors include the local density of microbes at which secretion significantly alters the environment, the extent to which secretion enhances microbial growth and maximum local density, and the ways in which secretion alters colony survival and dispersal. I also show that, in long-lived colonies, competition of progenitors with their descendant mutants significantly alters evolutionary dynamics and degrades the level of cooperative secretion favored by natural selection.
Secretion versus nonsecretion is just one example of the kind of tradeoff shaped by demography, life history, and the genetic structure of populations. A similar tradeoff occurs between metabolic rate and metabolic efficiency (Pfeiffer et al. 2001). Increase in rate typically requires additional energy to drive metabolic reactions, thereby reducing the net yield of biomass production per unit of energy intake. My initial studies on the tradeoff between rate and yield in microbial metabolism suggest that demography and the genetic structure of populations may strongly influence how evolution shapes variations in metabolic flux (118).
A simple prediction is that short-lived resource patches may favor rapid and relatively inefficient metabolic flux, whereas long-lived resource patches may favor slower and relatively more efficient metabolic flux. That prediction arises from classical life history theory, which quantifies how natural selection weights the relative valuation of rapid reproduction versus long-term survival.
The specific prediction about metabolic efficiency under simple alternative demographic assumptions is easy to obtain and requires little theoretical development. However, there has been neither a systematic application of such thinking to the actual variations in metabolic design nor a deeper analysis of how particular evolutionary processes interact to shape variations in metabolism. Other tradeoffs, such as uptake versus synthesis of complex organic compounds, lead to similar issues. Uptake versus synthesis may be particularly interesting, because that tradeoff has the potential to structure the patterns of resource flows through multispecies microbial communities.
This subject provides an opportunity to connect the fundamental concepts of sociality and demography that I helped to develop in my book, Foundations of Social Evolution, with many interesting and important microbial traits such as virulence (111) and metabolism (118). Because microbes can often be studied experimentally in controlled settings, there is much opportunity for testing theories and refining the way in which we understand microbial ecology, evolution, and virulence.
My studies of particular problems, such as microbial life history, often require development of fundamental conceptual issues. Many of the conceptual issues relate to natural selection. For example, how does environmental variability influence life history traits? How does the genetic structure of populations interact with the spatial and temporal scaling of resources to shape aspects of competition and cooperation between individuals? My work has led me to many such problems. In consequence, I have devoted much time to developing foundational principles of natural selection and evolutionary process.
I wrote a series of seven articles in Journal of Evolutionary Biology on the theory of natural selection. Those articles bring together some of the fundamental issues that arose in my earlier work. The topics include a unified approach to natural selection in variable environments (126), the role of phenotypic variability and the Baldwin effect in accelerating evolutionary rate (127), the consequences of different time scales and levels of selection in shaping phenotypes (129), the Price equation as unified mathematical expression of natural selection (130), an information theory interpretation of natural selection (133), a path analysis approach to partition the distinct selective pressures and causal pathways on traits (134), and the history and interpretation of kin selection in light of the path analysis and causal perspective of natural selection (135).
Information, probability, and the common patterns of nature
The patterns of natural history tend to follow a few common probability distributions or their relatives: normal, gamma, power law, and so on. Within ecology and evolution, there is a long and developing line of neutral theory to explain the commonness of these patterns. The neutral theories show how small-scale random processes of mutation, birth, and death can, in the aggregate, lead to the observed distributions. These neutral theories do often come close to matching observations. The question is why. Are the small-scale processes of mutation, birth, and death truly neutral and purely random? Or, are there other reasons why the observed aggregate patterns follow these apparently neutral distributions, in spite of non-neutral processes acting at smaller scales? These questions touch directly on how we understand the patterns of nature.
I used the well developed concepts of entropy and information theory to discuss why the commonly observed aggregate distributions are in fact so common (114). We know from the central limit theorem that the summing up of many small-scale processes, each non-normal in distribution, leads nonetheless to the normal distribution at the aggregate, or summed, level of measurement. In effect, the non-normal fluctuations at the small scale cancel in the aggregate, leaving as the only signal, or information, the standard normal pattern. The same sort of canceling of non-neutral fluctuations at the small scale likely happens in many cases in which we measure at the aggregate level, for example, of nucleotide substitutions, species numbers in communities, number of ecological connections between species, and so on.
In my first contribution, I presented a broad synthesis of these concepts from other disciplines (114). I presented the ideas in a way that may help in applying the theory to biological problems and to understanding the role of neutral theories in ecology and evolution. Within that first synthesis, I also began to identify some of the weaknesses in the conceptual structure as it currently exists in physics and information theory.
I am now pursuing those weaknesses. In particular, we recently developed a new approach to measurement, information, and probability (121). An important component of the variation in observed pattern arises directly from the fundamental way in which measurement changes at different scales. As one brief example, measurements are often linear at small magnitudes but grade into relative or logarithmic scaling at large magnitudes. Such linear-logarithmic scaling leads naturally to normal distributions at small scales, power law distributions at large scales, and Student's distributions at intermediate scales.
More generally, there appears to be a simple way to relate the scaling of measurement to the common patterns of nature. We have used the relation between measurement scale and probability to provide a classification of common probability distributions (123). Our classification emphasizes that simple aspects of measurement and aggregation determine the form of the common probability distributions and their relations to each other. This understanding of the natural history of pattern provides an essential background to nearly all scientific studies. One must know the natural contours of pattern set by measurement and aggregation in order to analyze how particular processes cause deviations from those natural contours. I have illustrated the way in which measurement sets expected pattern by analyzing the simple biological example of species abundance distributions (122).
Epidemiology and evolution of infectious disease
The integration of mechanistic and evolutionary analyses is the theme of my book, Immunology and Evolution of Infectious Disease, and of my various studies on host-parasite interactions. At this point, I have finished much of the conceptual background and development of the fundamental quantitative principles. I also developed a collaboration with an experimental laboratory that studies antigenic variation, using the mathematical and computational models to design experiments (103). That study illustrates the long-term goal of understanding how various mechanistic components determine complex aspects of host immunity and parasite escape, and how evolutionary processes have shaped the underlying mechanisms. In another preliminary study, we developed a theory for why some pathogens, such as influenza, vary so much, whereas other pathogens, such as the measles virus, vary so little (105).
Cancer, age-specific disease, and somatic genomics
Cancer might be described as an evolutionary process within individuals that changes the normal regulatory interactions governing cellular birth and death rates. This is a fascinating problem for an evolutionary biologist, because one must understand the short-term evolutionary processes within individuals that disrupt normal regulation, and the long-term evolutionary processes that have shaped the normal network of regulatory controls on cellular dynamics.
The great progress of modern biology at the genetical, biochemical, and cellular levels of cancer provides an opportunity to link mechanistic models of complex phenotypes to the evolutionary processes that shape mechanisms. My early work on cancer focused on epidemiological data, linking mechanistic, quantitative models of progression within individuals to rates of incidence in populations (93, 96, 98). In my recent book, Dynamics of Cancer: Incidence, Inheritance, and Evolution, I start with incidence rates in populations and then develop a deeper understanding of mechanistic details, quantitative models of phenotypes (cellular regulation and cancer progression), and evolutionary processes.
Recently, my interests in cancer have turned to two related topics. First, somatic mutation must be common in all individuals, because a human body contains approximately 10^13 to 10^14 cells, all descended from a single-celled zygote. To understand the relation between the number of somatic mutations that occur and the number of cells that carry a mutation, we must think of the body in relation to the lineage history descending from the single ancestral zygote and how mutations accumulate in that lineage history. This view of somatic evolutionary genomics led us to propose that a significant fraction of adult onset cancers may arise from somatic mutations early in development (86). New high-throughput genomic technologies are just opening up the possibility of directly measuring somatic variability and evolution (115). This new work will be one of the great biomedical topics in the coming years. I remain interested in placing the new data within the essential context of cell lineage history.
The second topic arises from thinking about cancer as the disruption of the normal checks and balances in the regulatory control of cells and tissues. This view leads to the more general problem of robustness and regulatory networks.
Robustness and regulatory networks
Our cells and tissues have multiple repair mechanisms and checks and balances on growth. Those multiple mechanisms buffer against most mutations and physiological disruptions, leading to a form of robustness against perturbation. Cancer arises only after the final protective mechanism fails. This sort of robustness leads to an evolutionary paradox.
Each additional protective mechanism reduces the impact of any single hereditary mutation and therefore allows the accumulation of more mutations in the population. The additional mutations allowed by robust buffering lead to significant mutational decay in the protection provided by each buffering component. In general, a robustness mechanism decreases the sensitivity of a character to perturbation and therefore reduces the intensity of natural selection on that character. Reduced selective pressure on the character may often favor a less costly, lower performance trait (108).
The paradox of robustness arises from evolutionary dynamics: enhanced robustness causes an evolutionary reduction in the adaptive performance of a character, leading to a degree of maladaptation compared to what could be achieved by natural selection in the absence of robustness mechanisms. Over evolutionary time, buffering traits may become layered on top of each other, while the underlying adaptive traits become replaced by cheaper, lower performance components.
I have established the basic logic of the paradox of robustness (108), and applied the idea in a simple model of cancer and the heritability of DNA repair defects (94). I believe the basic idea is quite general and has widespread implications for understanding organismal design. However, I am still searching for a clear way to connect this general concept of evolutionary dynamics to the empirical facts of evolutionary history.
Patterns of cellular stochasticity in gene expression may provide one way to connect robustness to reduced precision in phenotypic expression (77, 138). Robustness may be achieved by averaging inputs, which is a very general way in which to reduce variability in the expression of phenotypes. The greater the number of independent inputs, the less the variance in the average of the inputs. With a larger sample, perturbations to any input or component have less consequence for the overall output of the system. Robustness may reasonably be defined as reduced sensitivity to perturbation. Thus, averaging over multiple component inputs provides a powerful mechanism to achieve robustness of phenotypic expression and system performance.
How does an increase in system robustness affect the variability of components? Suppose that multiple component inputs are averaged to obtain a system output. The more independent component inputs, the less each component input affects the average. Therefore, as the number of inputs increases, the system can tolerate greater component variability to achieve the same level of output variability. If we assume that natural selection is acting on the system outputs, then increasing the number of inputs weakens the selective force acting on the variability of each component. This idea leads to a prediction: the more protections a system has against the variability of the underlying components, the more stochastic variability there will be in the underlying components. I have discussed ways in which that idea might be tested (138).
Through much of my career, I have been interested in coevolutionary interactions that lead to conflict or cooperation. These interactions occur at many different levels of biological organization: between members of the same population (social behavior), between different genetic elements within a genome, and between host and pathogen populations. This broad view led me to study coevolutionary interactions in a variety of systems.
Sex allocation and sociality
I have studied sex allocation, which is the division of resources to sons and daughters, or, more generally, to male and female reproductive function. Sex allocation has played an important role in the study of social behavior because (a) many aspects of reproductive competition affect the relative value of sons and daughters, for example, sexual selection or conflict among genetic relatives; (b) the methods required for analyzing sex allocation problems apply to several questions in adaptation, for example, frequency dependent selection or genetic constraints such as sex chromosomes; and most importantly, (c) numbers of males and females and investment in each are relatively easily measured traits. Perspectives on adaptation and social evolution can therefore be tested by studying sex allocation. My book Foundations of Social Evolution summarizes much of this work.
I have worked on theories of genomic conflict. Different parts of the genome are transmitted in different ways, and therefore may have conflicting reproductive interests. For example, mitochondria are transmitted mostly from mother to daughter, whereas autosomes are passed equally to both sons and daughters. Mitochondria and other matrilineal elements therefore favor a female-biased sex ratio, whereas autosomes tend to favor an equal sex ratio. Conflicting modes of genetic transmission cause many interesting traits, such as cytoplasmic male sterility in plants, sex ratio biasing bacterial and viral symbionts, and high frequencies of transposable genetic elements. Many aspects of genomic organization and patterns of genetic variation can only be understood in the context of genomic conflict.
I have developed theoretical models to analyze the coevolutionary genetics of hosts and parasites. Populations often contain high levels of genetic polymorphism for resistance to pathogens. The effectiveness of this resistance is limited because the pathogens are, in turn, widely polymorphic for host-range genes that can escape host resistance. In addition to the variability found within populations, the frequency of particular host and parasite genes may vary widely over small geographic areas (metapopulation dynamics).
Virulence, symbiosis, and sociality
The host-parasite work emphasizes evolutionary dynamics on the conflict side of my interest in conflict and cooperation. I have also studied the evolution of mutual harm or benefit in symbiosis. For example, the more a parasite harms its host (virulence) the more it damages its food supply. But the costs of virulence may be offset by increased competitive success against other parasites within the host or by greater transmission to other hosts. This problem of virulence can be generalized by considering parasite and host genes as replicators that live within a shared compartment (body). The different replicators have a shared interest in using the body's resources prudently, but also conflict over the distribution of resources. This general view of symbiosis applies to the evolution of protocells and genetic systems near the origin of life, to genomic conflict, to the evolution of ecological mutualisms, and to the evolution of group living and social behavior.
Click images for information, links, and downloading of my books and series