ESB offers a wealth of opportunity for both philosophy of science and philosophy of biology to advance some of their long-running deliberations on scientific practice. The issues we focus on are explanation, modeling, and theoretical synthesis. In addition to enriching philosophy, philosophical analyses of ESB probe and potentially make contributions to several of the issues that are central to ESB’s own development.
Explanation and Prediction
Philosophy of science has a particularly strong historical investment in the nature of explanation. Despite earlier efforts to give a mutually entailed account of explanation and prediction within a “covering law account” (Hempel and Oppenheim 1948), philosophers began to think of the two activities quite separately in the 1950s. Evolutionary biology was in fact one of the examples used by philosophers to show that explanation is not necessarily predictive (Scriven 1959). Explanation took over as the philosophical focus, with prediction becoming its lesser, dubiously related cousin (e.g., Rescher 1958). Today, explanation—and especially mechanistic explanation—rule the roost. However, the standard philosophical formulation of mechanistic explanation asserts that mechanistic explanations generate predictions, and that failed predictions can help refine the mechanistic explanatory model (Machamer et al. 2000; Douglas 2009).
While it might be reasonably straightforward to posit mechanistic explanations as the background to the foregrounded predictive aspects of systems biology, will this be the case for ESB? A perennial question of evolutionary biology has been its capacity for predictiveness, which—prior to the advent of large molecular datasets—was often decreed to be impossible (e.g., Mayr 1961; Gould 1990). ESB gives evolutionary biology new means by which to generate evolutionary predictions. The mechanistic models that underpin network biology are set out as mathematical equations. These models can generate predictions about future system states given additional knowledge about ecological conditions over time (Soyer 2010; Papp et al. 2011). However, philosophy of mechanism has so far tended to focus on concrete and easily visualized mechanisms, meaning philosophical accounts of explanation relate best to graphical representation and qualitative causal reasoning. This might work well for some molecular-physiological models and proximate explanation, but becomes problematic for probabilistic population-level phenomena and ultimate explanation (Matthewson and Calcott 2011; Levy 2013). Evolutionary models are very commonly concerned with these aggregated phenomena and their explanations. How then, do mechanistic models fit what is going on in evolutionary biology? Exploring this question is one of the primary challenges for philosophical analyses of ESB, and ESB research itself.
While systems biology’s very rationale is to combine explanation and prediction, evolutionary biology is currently framed by ultimate explanatory aims (Mayr 1961). By integrating systems biology, which requires precise models of mechanisms, ESB is attempting to model those mechanisms as they operate across stretches of evolutionary time. In its efforts to develop mechanistic explanations of evolutionary processes and outcomes, ESB brings in causal details to generate a fine-grained understanding of evolution. But it also aims to predict phenotypic consequences of genetic and environmental changes on evolving lineages (Papp et al. 2011). What ESB can do, therefore, is give philosophers a variety of ways in which to think about the explanation-prediction connection, including that of explanation on different timescales, with special reference to modeling strategies.
Modeling and Explanation
Because ESB arises at the confluence of different approaches and research questions, modeling tensions can emerge between different levels of explanatory model (Calcott et al. 2015, this issue). These tensions are especially strong between detailed mechanistic and law-based explanations, as one of the articles in this collection discusses (Green et al. 2014, this issue; see below). The philosophy of modeling has undergone an enormous upsurge in philosophy of science as it has replaced adherence to law-based theory by a focus on mechanistic models. Although there are diverse discussions of modeling in philosophy of science, one area that has developed very rapidly in the last decade is that of how models explain. This is of particular importance when different modeling strategies are used for the same research questions—something highly characteristic of ESB. In systems biology, the aim is to go beyond phenomenological descriptions and qualitative causal models, and towards prediction and control via mathematical modeling. But in addition to prediction, explanation is sought: “the structure of the model should somehow reflect the underlying mechanisms in the biological system” (Cedersund and Roll 2009, p. 904).
In philosophical accounts of mechanism, explanatory models describe in detail mechanisms and their behaviors. More abstract dynamic mathematical modeling is deemed by some philosophers to be explanatory only when it describes a mechanism in a very concrete way (Kaplan and Bechtel 2011; Kaplan and Craver 2011). While many biologists are entirely focused on such mechanisms—for good reason, because such models enable intervention—a more abstracted level of modeling (e.g., dynamical systems theory, which allows generalizable inferences to be made about biological systems over time) is also at work in ESB (for other fields, see Fagan 2012). Some workshop participants saw ESB’s goal as one of being able to combine these modeling approaches in order to give general high-level accounts that are of relevance to experimentalists. A key challenge for any integration of such models is to account for the influence of evolutionary forces on cellular and subcellular processes. Most approaches and tools from dynamical systems theory assume time invariance of the environment and hence constant parameter values. There are few practical tools for the analysis of nonlinear time-invariant dynamical models. From a philosophical perspective, particular formulations of dynamical systems theory, such as mathematical general systems theory (Wolkenhauer et al. 2012), may offer opportunities for philosophers of science to develop accounts of non-mechanistic explanation.
Philosophy of Evolution: Explanatory Synthesis
As well as philosophy of science issues, ESB addresses some of the core concerns of philosophy of biology. Historically, philosophy of biology has focused on evolutionary biology, and especially evolutionary theory in the form of the conceptual machinery of Darwinian evolutionary theory (rather than population genetics). Particular attention has been paid to conceptualizing units of selection and species. ESB is not of particular relevance to those efforts, but it does tread a path that is just as well worn in philosophy of biology. This shared interest has to do with evolutionary explanation. ESB highlights several important explanatory conflicts in evolutionary biology. They include those between non-adaptional and adaptive evolutionary explanation (see de Visser et al. 2003; Lynch 2012; Landry et al. 2013; Siegal 2013), proximate and ultimate explanation (see above), optimality and non-optimality explanation (Flamholz et al. 2013), and engineering versus evolutionary explanation (Calcott 2014).
A general theme that is of great relevance to the philosophy of evolution is whether and how ESB contributes to the quest for a revised Modern Synthesis of evolutionary biology. This contribution does not need to be in the sense of overthrowing established knowledge, but may simply have the aim of enhancement. A conservative way of thinking about ESB is that it merely adds new molecular tools and analyses to existing evolutionary approaches. A more radical view would hold that in this very integration of systems and evolutionary biology, the old framework changes into something different. But does this simply involve throwing a bit of evolutionary theory into systems biology, and a bit of systems biology into evolutionary biology? A point raised very strongly at the meeting was that evolutionary analyses in ESB need to be, as one participant said, “integral, not just tacked on.” What might this mean in terms of integrating a population genetics approach with a systems one?
Lately, additions to the Modern Synthesis have been described by philosophers and philosophically oriented scientists as an “Extended Synthesis,” in which there is an “ongoing shift from a population-dynamic account to a causal-mechanistic theory of phenotype evolution” (Pigliucci and Müller 2010, p. 12). This rebalancing is in part what is driving ESB, but not quite in the way these commentators suggest. Our view of ESB research so far is that it is not a matter of going from population-dynamic accounts to causal-mechanistic ones, but of integrating them or working out their relationships to a particular research goal. This is what is of central philosophical importance in ESB, and requires an elaboration of how to interpret Fig. 1.
Population and quantitative genetics offer formal models that describe average fitness in populations and statistical associations between genotypes and phenotypes, whereas systems biology provides quantitative models of the function of intracellular networks in individual organisms. There is no hard line between the two approaches, however, and crossover work already exists between systems biology and population genetics. Ultimately, the aim of integrating the two approaches is to have complete understanding of the fitness effects of variation in molecular networks at a populational level (Loewe 2009). For example, combining large-scale mutational analyses with mechanistic models would allow evolutionary predictions to be made at different levels of biological organization (Papp et al. 2011; Landry and Rifkin 2012). Very often, these sorts of proposals are made on the basis of metabolic networks, for which there is a great deal of data and characterization, as well as assumptions about steady states.
These analyses can be extended to molecular phenomena with unknown fitness effects, such as “noise” in gene expression. Noise refers generally to molecular fluctuations in biological processes; with regard to gene expression it is the stochastic variability of messenger RNA and protein levels in cells with identical genomes and environments. Very little is known about how such noise percolates through regulatory networks, although it is likely to be attenuated or amplified by various nonlinearities in those networks (see, e.g., Raj et al. 2010). Noisy processes can be captured by mechanistic models of molecular activity on cellular biomass production (e.g., Wang and Zhang 2011), whereas the classic abstract models of population genetics do not incorporate effects that arise from intracellular noise. These classic models also shy away from nonlinear processes—even deterministic ones. In addition, mechanistic models can make predictions about those effects and what their alterations mean for evolving systems. The inevitability of noise and its effects on fitness make it clear that “molecular stochasticity must be included in a comprehensive evolutionary theory” (Wang and Zhang 2011, p. E74). Importantly, however, integrating mechanistic models into population genetic theory might not lead to complete incorporation of one sort of model by the other, but to the mutual revision of both population-genetic and mechanistic models.
The key place where traditional population/quantitative genetic theory and ESB intersect is the genotype-phenotype map. Theorists have always made simplifying assumptions about the nature of this map. This is understandable because without simplifying assumptions, the models would be mathematically intractable and generalizations would be hard to come by. The danger, of course, is that if the assumptions are unfounded then so are the generalizations. Yet it must be said that in general, population genetics is not in crisis; indeed, it is experiencing a massive resurgence as genotype data on populations of humans and other species are being generated at a rapid and accelerating pace. Population genetics has been crucial to extracting meaning from these data by revealing, for example, recent adaptive events or demographic histories.
The situation for quantitative genetics is a bit different. The outpouring of human data from genome-wide association studies (GWAS), although leading to exciting discoveries of genes associated with important human traits, has revealed a consistent problem of “missing heritability.” This problem refers to the realization that the genetic loci found to contribute to variation in a particular trait typically account for a minor fraction of the variability expected to be explained by genetic causes. This gap raises the major problem of just how explanatory quantitative genetic explanations of traits are (Turkheimer 2011). The reason or reasons for the missing heritability remain to be determined, but several hypotheses concern aspects of the genotype-phenotype map that are central to ESB, such as robustness and epistasis (Zuk et al. 2012). Other proposed explanations for the missing heritability focus on the low power of (even extremely large) GWAS to detect all relevant loci: the missing heritability could be explained by genetic architectures in which very many loci of very small effect contribute to phenotypic variation (Rockman 2011). These explanations are not mutually exclusive. For example, the small effects could be context dependent and small only on average. This is a place where ESB has the potential to contribute substantially to evolutionary theory—by forcing the adoption of more complicated models of the genotype-phenotype map but at the same time constraining the potential forms those models could take.
The abstract connection in quantitative genetics between genotype and phenotype might not, however, be neatly supplemented by systems biological approaches. Much successful systems biology takes a modular approach, in which genes with a common function are conceptualized as having some independence from genes that do not contribute to that function (Wagner et al. 2007). The way in which modules of genes are experimentally mapped to functional phenotypes greatly reduces the complexity of the mapping process (Landry and Rifkin 2012). But unfortunately, module-based mapping is not the same sort of mapping that quantitative genetics aims at, due to the mutational complexities it addresses via statistical associations. Systems biology and quantitative genetics thus deploy different idealizations of gene interactions such as pleiotropy and epistasis. The modular approach of systems biology is unlikely to capture the subtleties of quantitative trait maps, even if the latter maps do not causally describe multiple levels of biological interaction (Landry and Rifkin 2012). However, despite the imperfect correspondence between such mapping approaches, there is considerable optimism in ESB that integrated mapping is feasible and valuable, even if it is more difficult than inserting one mapping methodology or result inside the other.
But ongoing attempts at this integration still leave untouched the epistemic challenges posed by the missing heritability problem: if robustness, epistasis, and/or weak causation are widespread, is there a limit to what we can know about the causes of phenotypic variation even with extremely large studies? Although this is a fascinating conundrum for philosophers to reflect on further, we suspect the biologists confronting this challenge will continue to attack it regardless of any theorized limit to knowledge. In fact, that limit is not clear. Missing heritability is comprised at least in part by hidden heritability (Gibson 2010) and what may still be missing are the appropriate methodologies to reveal that hidden fraction. And indeed, if ESB has one message for observers of the life sciences, it would be that problems that are understood as methodological limits to evolutionary biology are merely the best starting places for novel approaches in ESB.
A similarly challenging situation arises when evolutionary theory and ESB intersect in attempts to explain features of system organization. These are both global features, such as scale-free architectures, and local features, such as over-represented regulatory motifs. So far, many of these explanatory efforts have been adaptationist, with network structures and properties (e.g., robustness) being explained by selective advantage (see Siegal 2013). What is still lacking is a fully elaborated “non-adaptional theory”—incorporating but not restricted to neutral theory—of the evolution of biological systems. Some important work is already pointing toward ways in which to do this (e.g., de Visser et al. 2003; Solé and Valverde 2006; Soyer and Bonhoeffer 2006; Lynch 2007a, b; Crombach and Hogeweg 2008; Wagner 2008b), but further development of appropriate null models for the evolution of complex networks is likely to become a major goal not only for ESB but also evolutionary biology as a whole.
However, the residual general issue embedded in questions about modifications to the Modern Synthesis is the fact that for some evolutionary biologists, the very idea that anything “revolutionary” could be underway with regard to the Modern Synthesis is suspect.
I think the theoretical tools—the idea of natural selection and mathematical population genetics—are already at hand, and although we may get new analytical tools (such as genomics!), the rest is just hard grinding work trying to understand speciation and natural selection in the wild. I may be wrong, but I don’t scent revolution in the wind yet. (Coyne 2009)
EvoDevo is sometimes cited as a truly novel contributor to the Modern Synthesis (e.g., Müller 2007; Carroll 2008), but according to many other commentators, its inclusion does not greatly perturb the core conceptual apparatus of evolutionary biology (e.g., Sterelny 2000; Minelli 2010). Here, we are not particularly interested in that debate as it stands; far more compelling is the question of how EvoDevo itself can be understood in light of ESB. Some very promising results have been achieved along those lines already, with—for example—the development of a conceptual framework that has the scope for understanding phenotypic variability both quantitatively and dynamically (Jaeger et al. 2012). One of the articles in this collection will delve more deeply into epistemic questions about EvoDevo and dynamic mechanistic models (Jaeger et al. 2015, this issue—see below).
In addition to these advances, it is clear that developmental biology on its own is increasingly incorporating systems biology approaches, as the former increases its attention to modeling and genome-scale quantitative data (see Jaeger et al. 2015, this issue). Figure 2 expands Fig. 1 to represent the interactions between these approaches. This trend in itself is worth watching, to see the transformations brought about in developmental systems biology, and how it becomes “evolutionized” (Busser et al. 2008; Wunderlich and DePace 2011). This outcome, of developmental and systems approaches becoming integrated with evolutionary analyses, is also reached by developmental approaches that are already evolutionary (EvoDevo), and which integrate systems biological approaches to achieve multilevel explanations of particular developmental phenomena (Fig. 2). But as in the population-genetic and network biology intersection discussed above, the adaptiveness of organizational features in development cannot be taken for granted. Developmental systems drift and null models of genome evolution will have to be ruled out by rigorous analysis (True and Haag 2001; Siegal 2013), and consideration of constraints at different levels taken into account (de Visser et al. 2003).
Figure 2 depicts general research approaches, in which methods, aims, and interests are shared in various configurations. Practitioners located in any general research area (developmental biology (DB), evolutionary biology (EB), systems biology (SB)) can combine elements of approaches from their neighbors. We see ESB, evolutionary developmental systems biology (EDSB), and developmental systems biology (DSB) as comprised of these overlaps rather than being some sort of fixed area into which practitioners move. EvoDevo, however, has a somewhat more defined identity. Moreover, we do not mean to indicate any hard boundaries between ESB and EDSB. We are merely suggesting that not all ESB is going to be concerned with metazoan (and sometimes plant) development, and that issues arising in EDSB—or for that matter, EvoDevo and DSB—may not always be shared with EB or SB. In this article, however, we have focused on issues we think are common to both ESB and EDSB.
Philosophy of Evolution: Scope
The final broad question raised by ESB is whether all biology must necessarily be evolutionary. Theodosius Dobzhansky’s famous quote about all biology being understood in the light of evolution can be interpreted as biological understanding being achieved “only in light of evolution.” However, this interpretation is a rather unnuanced claim that that can be easily resisted. The other view is more difficult to reject: that non-evolutionary research approaches have made progress and will continue to do so without any consideration of evolutionary processes, thus refuting Dobzhanksy’s claim. For example, research into physiological function is often thought to be exempt from evolutionary questions, in that the focus is what a system does now rather than how it came to be that way (e.g., Boogerd et al. 2007). Once a good mechanistic account of such a system has been developed, this view agrees that there may be some epistemic and therapeutic benefits from applying evolutionary analyses to that system, in order to understand malfunction. However, according to this interpretation, evolutionary analysis is not necessary to achieve explanations of current function.
We think this position is balanced by numerous biological phenomena in which both mechanistic and evolutionary accounts are valuably combined in order to grasp current function as well as dysfunction. A classic example is that of the heart as a pump. There is a fairly complete understanding of the function of the heart from a “plumbing” perspective, but high levels of heart disease require an integrated evolutionary perspective that addresses dysfunction within the context of explaining the evolution of healthy function. The increasing rates of heart disease in industrialized societies is one of the prime examples used in discussions of evolutionary medicine (e.g., Nesse and Williams 1998). A more molecular but still familiar illustration can be drawn from adaptive immunity and the production of antibody diversity, which is understood in deep molecular detail. Despite this detail, the existence of a diversity-generating mechanism for antibodies requires an evolutionary explanation, and understanding the evolutionary forces shaping that diversity has important clinical implications (James and Tawfik 2003; Burton et al. 2005).
Heart function and immunity are both helpful examples because they are familiar to most people. In ESB, examples are more complicated but they illustrate this explanatory interdependence even more convincingly. For instance, mechanistic explanation of a process known as stochastic switching in gene regulation—whereby intracellular networks flip between two phenotypic states—turns out to require an evolutionary explanation and vice versa. Fluctuating selection, which arises in relation to variable environments, is not enough on its own to drive the emergence of stochastic switching. Noise or stochastic variability in gene expression is what tips intracellular systems into one state or the other, and this capacity to switch has consequences for fitness and evolvability (Kuwahara and Soyer 2012). Or, viewed from the perspective of robustness, mechanistic explanations alone cannot adequately characterize how systems respond robustly in different environments, and what the role of noise is in those responses (Félix and Wagner 2008). Only by combining both these explanations can we understand the switch mechanism, the contexts in which it will operate, and the fitness consequences for the entities involved.
Overall, ESB can be understood as a meeting ground from which the mutual enrichment of systems and evolutionary approaches can result. Some of these benefits are to do with breadth. ESB enables attention to a greater diversity of evolved systems, and this comparative potential leads inevitably to evolutionary analyses across different levels of variation. But there is also the advantage of depth. By providing detailed mechanistic explanations of system properties that are embedded in a dynamic landscape, greater explanatory depth is achieved—not to mention predictive potential. But in addition, the discussions above demonstrate very clearly that some of the central issues in ESB are also core issues in philosophy of science and philosophy of biology. This alignment of interests indicates a potentially productive interplay between ESB and philosophical analysis.