Organismal development and evolution are complex, multifaceted processes that depend intimately on context. They are subject to environmental influences, chance appearance and fixation of mutations, and numerous other idiosyncrasies. Genomics is detailing the molecular signature of effects of these mechanisms on phenotypes, but because numerous distinct evolutionary explanations can produce a given genomic pattern, the molecular details, rather than elucidating process, typically distract from explanatory insight and contribute little to predictive capability. While genomic research has burgeoned, direct study of evolutionary and developmental processes has lagged. We advocate for reinvigoration of direct study of process, along with refocusing of attention on questions of broad biological import, as more productive of urgently needed insights, which genomic approaches are not providing.

Biological complexity demands explanation, which, in turn, entails understanding the causes and consequences of biological change over time. Major advances in the 19th and 20th centuries identified the processes of evolutionary change, dramatically advancing explanation of the diversity of life. Population and quantitative genetic theory arose directly from reconciliation of evolutionary mechanisms with Mendelian genetics, providing conceptual and mathematical frameworks for understanding biological complexity (Provine 1971). Speciation, aging, and even the previously surprising amount of molecular variation, among many other general biological phenomena, can readily be understood using simple evolutionary genetic models.

Numerous challenges remain. For example, morphology has long attracted keen interest, but most studies of morphological evolution and development are either descriptive or taxon specific (Wittkopp and Kalay 2012). Advances in quantitative models, incorporating life-history theory, fitness measures, and community context have been slow, despite the obvious influence of these and other aspects on morphological evolution. One difficulty is to account for the operation of evolutionary mechanisms over different temporal and spatial scales, a problem common to most current topics in evolutionary biology.

Currently predominant methodology is to search for molecular bases of phenotypic difference. Rapid expansion of access to the tools of molecular biology has released a flood of detail documenting molecular change within evolutionary lineages. Enthusiasm for the burgeoning of molecular detail has inflated expectations of its explanatory power, yet the more complex is the phenomenon demanding explanation, the more molecular details obscure understanding. Fifty years ago, Mayr (1961) distinguished the goal of determining how a biological form or function comes about from that of elucidating why it exists. We here argue that molecular analysis will fall far short of achieving both goals: the first, because the complexity of influences on trait variation makes full accounting of its molecular basis intractable; the second, because even a complete accounting of the molecular basis of particular phenotypic differences does not address questions of why they arose and persist. As has often been acknowledged, pattern does not reveal process because multiple, perhaps many, scenarios produce similar patterns. Yet, direct study of evolutionary process is increasingly neglected, in favor of genomic analysis.

We review diverse examples illustrating that intensive, molecular approaches tend to yield modest information about how differences at the genetic level confer differences in organismal form and function. The emphasis on determining molecular detail often impedes progress in explaining organismal phenotype by diverting research from directly evaluating the roles of evolutionary and ecological processes that affect phenotypic change. We acknowledge that for some biological processes molecular approaches provide the best evidence, for example, parentage from one generation to the next and branching among diverging genes and species over many millions of generations. Molecular approaches also illuminate molecular structures and inform understanding of molecular processes; these insights have proven valuable in many contexts, ranging from elucidating how energy is stored through photosynthesis to drug development (Ghosh 2009). However, the goal of explanation of organismal structures and functions requires elucidation of mechanisms that favor their evolution (Mayr 1961).

Explanatory insight will require renewal of direct study of the processes that lead to organismal change. These include not only all the evolutionary processes, natural selection, genetic drift, mutation, migration, and nonrandom mating, each of which affects all the others, but also ecological processes, such as density-dependent population growth. To be effective, this research program must acknowledge the complexity of the biological properties under consideration as well as the processes that impinge on them. Limitations of molecular approaches to explanation of biological form and function are well exemplified by studies that aim to attribute differences in phenotype to particular genomic differences, genotype to phenotype (G-P) mapping, as we show below.

The Elusive Genotype-Phenotype Map

For well over a century, but particularly since Johannsen (1911) noted the distinction between genotype and phenotype, geneticists have worked to determine the relationship between them. This research program has illuminated the complexity of this relationship. The original conception of the genotype–phenotype map was Mendelian: a one-to-one mapping of differences in genotype to the corresponding differences in phenotype. Early studies of inheritance ruled out this simplistic notion. Yule (1902) demonstrated that typical continuous distributions of variation in traits could arise from the influence of multiple genes. Soon after, Castle's studies (Castle 1906) on polydactyly of guinea pigs and Nilsson-Ehle's (Nilsson-Ehle 1908) on pigmentation of grain demonstrated that multiple genes influence expression of these traits. A clever biometric technique (Castle 1921) yielded estimates of the number of loci that contribute to phenotypic differences for many traits. This approach required numerous restrictive assumptions, for example, that the lines chosen to cross are homozygous at each locus affecting the trait. It is unlikely that this and other assumptions are ever fully satisfied, but when they are violated, the approach estimates a lower bound on the number of contributing loci (Castle 1921; Zeng 1992).

Despite its limitations, the biometric approach resoundingly demonstrated the polygenic nature of variation in many traits (Wright 1934a,b). This finding, which long predated the elucidation of the molecular basis of inheritance and the discovery of copious molecular variation within populations, profoundly advanced understanding of the relationship between genotype and phenotype by making clear that a given phenotypic difference can arise in many distinct ways at the level of genotypes.

Molecular studies have reinforced these conclusions about the genotype–phenotype map. Variable nucleotide sequences can be localized to chromosomal positions; they thus serve as genomic landmarks, and variation in phenotypic traits can be statistically associated with them. This quantitative trait locus (QTL) mapping approach yields information on the approximate genomic locations of genes that affect traits, as well as estimates of their number and the effect of each. Nevertheless, many loci go undetected, while the effects of those that are detected are generally overestimated (Beavis 1994). The capacity to detect many, even all, molecular variants does not alleviate these biases. The role of recombination is paramount in dissociating effects of different genes on the traits of interest; thus, the number of genes whose effects can be distinguished depends heavily on the prevalence of recombination in the design of a study.

Moreover, for the greatest precision and accuracy of mapping of genes that influence phenotypes, studies generally focus on artificially homogeneous lines in unrealistically controlled conditions. Such environmental control achieves statistical power for detection of genetic effects, but, especially when the study population had not been selected under those experimental conditions, conclusions that the inferred genetic effects generalize over populations and conditions are likely to be spurious (Travisano 2009). Despite methodological stringency, molecular studies have not altered fundamental understanding of the relationship between genotype and phenotype; studies of sufficient scale to detect many genes have amply confirmed that many genetic factors importantly contribute to variation in most quantitative traits (Hill 2005, 2010; Rockman 2012).

The overwhelming evidence of extreme, inherent complexity of the mapping of G-P, involving very many genes and, moreover, strong dependence on many aspects of environment, discredits the view that the mapping of G-P is simple enough that knowledge of the genes underlying variation in traits will reliably predict phenotypes and, further, have practical and explanatory value. A possible alternative view is that, relative to other approaches, G-P mapping studies are a more efficient and powerful means to explanation of phenotypic diversity. Yet, even though the cost of genomic studies has plummeted, with corresponding increase in their number and scale, fresh explanatory insights of broad biological import have not emerged, as our case studies below illustrate. Rather, with few exceptions (Colosimo et al. 2005; Steiner et al. 2007), the commonality of highly polygenic determination of trait variation has been increasingly validated (e.g., Weber et al. 1999, 2001; Mezey and Houle 2005; Burke et al. 2010; Ehrenreich et al. 2012), reinforcing evidence of the complexity of genetic influence (Hill 2005), as demonstrated seven decades ago. Similarly, recent studies have confirmed the dependence of gene effects on both genetic and environmental context (Wilczek et al. 2009), long established via evidence of interactions between genes (epistasis) and between genotype and environment (Clausen et al. 1940; Gupta and Lewontin 1982). As Lewontin (1974) explicitly noted at the dawn of the molecular revolution in evolutionary biology, deciphering the relationship between genotype and phenotype is likely to yield little explanatory insight beyond the understanding of the degeneracy of the relationship and the consequences of complex environmental effects on phenotype, as elucidated by quantitative genetics (Lewontin 2002).

Case Studies

The persistence of the simple view of the genotype–phenotype map follows directly from a reductionist stance that a systematic description of each component of a biological system is essential to explain its evolution (Rosenberg 2006). Yet, findings about the molecular basis of difference in organismal phenotype are typically incomplete, despite their intricate detail, and lack generality. Below, we show, for cases drawn from a broad taxonomic range, how simplification of the biological complexity of the relationship between molecular genotype and phenotype can fundamentally mislead. Moreover, the evolutionary mechanisms underlying the phenotypic change remain obscure because they are subject to little direct study.


The first successes in molecular genetics appeared in microbial model systems. Jacob and Monod (1961) identified the general mechanism for transcriptional gene regulation within a decade of the discovery of the structure of DNA. By the mid-70s, hundreds of phenotypically important loci had been mapped in multiple bacterial species (Taylor 1970). Research applying genetic mapping to elucidate adaptation in Escherichia coli illustrates both the potential and limits of molecular genetic approaches.

For E. coli, 310 phenotypically relevant loci had been localized to approximately ± 3% precision by 1970 (Taylor 1970). The short generation time, huge population sizes (> 109), and tightly controllable recombination rate made possible rapid placement of genes via identification of phenotypes, primarily losses and gains of function, and assessment of recombination of these traits. Identification of the loci underlying these phenotypes made possible determination of biochemical pathways and thereby a “map” of the genetic information required for bacteria to acquire nutrients, grow, and reproduce. Further empirical and theoretical investigation led to metabolic control theory, which mechanistically explains pleiotropic, epistatic (Kacser and Burns 1973), and dominant (Kacser and Burns 1981) gene action in biosynthetic pathways. This theory was extended (Dykhuizen et al. 1987) to predict outcomes of competitive interactions in single and multiple nutrient laboratory environments (Dykhuizen and Dean 2004). Thus, in the E. coli model system, mapping of G-P via biochemistry has yielded conceptual advances and has promoted study of the ecology of fitness differences and molecular mechanisms underlying them in controlled laboratory environments. This success results from E. coli's genetic, physiological, and ecological simplicity. Its relatively small genome has coordinately regulated genes, which are physically associated into modules (operons) and very rarely contain introns. Its life history is more complex than originally thought (Stewart et al. 2005) but is far simpler than that of many multicellular organisms, by virtue of limited development and complete absence of sexual reproduction.

The largely successful discovery of loci that are required for function of E. coli in specific environments has proven of limited benefit in predicting adaptive evolution of its physiology. In a study of E. coli spanning over 50,000 generations, phenotypes of replicate populations changed largely in parallel (Lenski et al. 1991; Lenski and Travisano 1994; Cooper and Lenski 2000; Cooper et al. 2003; Cooper et al. 2008; Barrick et al. 2009). In contrast, molecular changes underlying these responses ranged from parallel to divergent (Cooper et al. 2003; Woods et al. 2006; Barrick et al. 2009). Parallel evolution of phenotype masking divergent genomic changes has often been documented in microbial selection experiments (McDonald et al. 2009; Nguyen et al. 2012). The spectrum of genotypic responses demonstrates that multiple, distinct genotypes can meet the demands of the imposed selection regime. Moreover, the molecular changes observed do not match the adaptations predicted a priori (Travisano and Lenski 1996) from the E. coli G-P map. These findings, even for a unicellular organism, undercut hopes of reliable prediction of evolution at the level of individual genes. Genome-scale computational G-P mapping has shown some promise for predicting metabolic states of a given genotype under particular conditions (Lewis et al. 2010), but persistent effects of a lineage's genetic changes (Travisano et al. 1995; Woods et al. 2006; Blount et al. 2008) indicate that prediction of molecular changes in response to selection is inherently intractable even in organisms as simple as bacteria (Zhong et al. 2009).


Molecular studies to determine the G-P map of plants began in earnest in 1987. QTL studies initially focused on crop plants (maize, Edwards et al. 1987; tomato, Paterson et al. 1988), and soon on Arabidopsis thaliana, a convenient model organism (small genome, rapid life cycle, naturally inbreeding, Jansen et al. 1995), and on species chosen for their significance from evolutionary considerations (e.g., Mimulus spp., Bradshaw et al. 1995). The limits of inferences of estimates of both the numbers of genes that influence a trait and of the magnitudes of effects of identified genomic regions are clearly exemplified by Laurie et al.'s (2004) study of the genetic basis of difference in oil content of maize seeds from lines divergently selected for 70 generations. This study succeeded in locating 50 genomic regions, (each of approx. 2–3 cM), all of them making small, roughly equivalent contributions to the difference. To detect so many QTLs is a remarkable accomplishment, attributable to the extremely large scale and ingenious design of the study. Yet, approximately half the genetic basis of divergence in this trait was not localized, strongly indicating an important role for genes of small effect in sustained response to selection. Complete gene identification for any flowering plant remains remote, and a complete G-P map far more so. Of the 74 genomic regions affecting plant quantitative traits listed in 2009 (Alonso-Blanco et al. 2009), many remain unidentified at the level of individual genes, and for very few has the magnitude of the effect on phenotype been assessed in natural conditions.

Like the microbial investigations preceding them, these studies have amply confirmed the degeneracy of the G-P map, while also demonstrating limits of gene identification. Beyond this, these studies have yielded a wealth of molecular detail about model genotypes. They have illustrated molecular mechanisms by which new genes arise and change evolutionarily (e.g., Baumgarten et al. 2003, Mathews 2010). We recognize these accomplishments as substantial, but distinct from explanation of evolutionary processes underlying change in phenotype.

Recently, the approach of genomic selection (GS) in plant and animal breeding has emerged in recognition of the importance in trait expression of genes that defy identification because their individual effects are so small (Meuwissen et al. 2001). GS abandons the goal of identifying genes that contribute to phenotypic differences, while acknowledging that genes of individually small effect importantly contribute to selection response (Hayes and Goddard 2001). GS takes advantage of the efficiency of selection directly on phenotype, supplementing it, in generations when expression of the phenotype is unreliable (e.g., when the population is grown in an atypical environment or when it is selected before the trait of interest is fully expressed) with selection on molecular markers associated with the phenotype of interest, regardless of the weakness of the association. Whereas phenotypic selection generally outperforms GS on the basis of per-generation response to selection, the benefit of increasing the number of generations per year can give GS the advantage in absolute rate of response to artificial selection (Wong and Bernardo 2008). As the associations between markers and traits change in the course of selection, they are recalibrated periodically. In this way, GS can accumulate unidentified alleles of subtle effect that escape detection via QTL mapping or appear inconsistent in their influence on a trait (Hospital et al. 1997; Bernardo 2008). This large class of genes is critically important in explanations of the duration of response to selection (Weber 1990; Weber and Diggins 1990; Dudley and Lambert 2004; Laurie et al. 2004) as well as its precision (Weber 1992; Weber et al. 2008). These crucial aspects of selection response cannot be addressed through study of effects of alleles at individual loci.


Numerous studies of G-P mapping have been pursued in Drosophila species. Despite early recognition that the contributions of multiple loci, environmental effects, and developmental interactions greatly complicate determination of effects of individual loci (Kempthorne 1960), G-P mapping was considered possible, in principle, with sufficient labor (Thoday 1961). Early studies suggested that relatively few major loci contribute to variation in wing size, bristle number, and other traits (Robertson 1966; Spickett and Thoday 1966; Cavicchi et al. 1981). However, this apparent genetic simplicity was not supported in further investigations showing influences of numerous pleiotropic loci whose effects depend on sex, environment, and alleles at other loci (Mackay 2009; Frankel et al. 2011). No fewer than 100 loci contribute to natural variation in bristle number (Mackay and Lyman 2005). Similarly, Weber (Weber et al. 2008) identified several hundred loci affecting wing shape using a microarray analysis and showed that loci contributing to variation in traits likely differ among populations. Key improvements distinguish the later studies that detected a far greater genetic complexity from the early studies that identified few loci: particularly, increase in the scale of the experiments and, hence, in precision linking phenotype to genotype. Even with such success in gene identification, verification of effects of alleles in nature has proven elusive (Genissel et al. 2004; Macdonald and Long 2004).

A candidate exception to the complexity of G-P mapping is the Hox gene cluster. The structure of Hox genes is similar across virtually all animal lineages, and major morphological differences among taxa can readily be traced to evolution of the Hox gene clusters. The roughly one-to-one mapping of genotypic differences to differences in head-tail axis demonstrates that major genes affect this phenotype. In Drosophila, alterations of Hox gene expression result in stunning morphological changes, such as the transformation of antennae into legs (Gehring 1993). In some species such as the Amphioxus Branchiostoma floridae, localized gene expression is approximately collinear with the arrangement of the single 14 gene Hox cluster (Kmita and Duboule 2003). In others, extensive Hox gene evolution has occurred, with expansion to four Hox clusters in mammalian species (Maconochie et al. 1996) and up to seven in some teleost fish (Hurley et al. 2005). Studies of closely related species have identified individual loci that underlie phenotypic differences (Simpson 2002).

Nevertheless, closer examination has revealed extensive genetic complexity involved in trait differences, even ones for which single locus regulatory evolution had previously been inferred (Randsholt and Santamaria 2008; Crocker et al. 2010). Complexity is apparent even for Hox gene structures among closely related Drosophila species, in which little phylogenetic information is apparent for segmentation genes (Yassin et al. 2010). The emerging pattern is one of morphological change within selective constraints, with Hox gene evolution providing broad outlines of body plan control that structural and regulatory changes can dramatically alter (Di-Poi et al. 2010). Most importantly, variation in Hox loci is not associated with morphological variation within species. While Hox loci do provide a coarse schematic for understanding G-P mapping across a vast phylogenetic sweep (Ronshaugen et al. 2002), they have provided little insight at finer scales of morphological evolution.


In briefly reviewing a small fraction of the prodigious efforts to map G-P, we emphasize the extreme entanglement of the effects of numerous genes and of environmental influences on phenotype. Beyond this, organisms alter their environments, which reciprocally affect the organisms’ own phenotypes, as well as those of surrounding organisms (see also Lewontin 2002; Laland et al. 2011). Consequently, complete knowledge of a genome's loci and existing and potential allelic variants cannot, in principle, account for the phenotypic variation of multicellular organisms, except under exceedingly restrictive, unrealistically simplified genetic and environmental conditions. Even in microbes, evolutionarily predictive mapping of G-P has proven elusive (for an exceptional case, see Box 1). Understanding and prediction of phenotypic evolution has largely not been advanced by attempts to determine a G-P mapping, even with detailed, prior understanding of molecular mechanisms (see also Roff 2007).

Box 1. Genomic successes in understanding microbial pathogenesis.

The molecular basis for microbial disease outbreaks, major host shifts, and rapid host range expansions can often be discerned by retrospective genomic analysis, owing to the relative simplicity of these systems (Levin and Bergstrom 2000). Microbes have small genomes, low rates of recombination, and simple development, and they reproduce primarily or exclusively clonally. These characteristics favor informative G-P mapping. As important, dramatic expansions in host number or host range, like selective sweeps, greatly reduce genetic variation not tightly linked to the causal genetic changes. Consequently, specific molecular differences can be identified as causative factors in disease outbreaks.

The 2011 E. coli outbreak in Germany provides an excellent example of the possibilities for genomic explanation in epidemics. Starting in May and continuing through to the beginning of July, an outbreak of gastroenteritis, bloody diarrhea, and hemolytic-uremic syndrome (HUS) occurred in Europe, primarily in northern Germany with over 3800 cases. Screening of E. coli from stool samples for Shiga toxin and its encoding gene led to preliminary identification of the disease causal agent (Askar et al. 2011). Shiga toxin has multiple effects on host cells, including reducing protein expression by ribosome modification and inducing cell death. Subsequent analysis (Brzuszkiewicz et al. 2011) identified that the Shiga toxin producing E. coli (STEC) genotype is an uncommon serotype (O104:H4) with alleles from enteroaggregative and enterohemorrhagic E. coli, and carrying a plasmid encoded extended spectrum beta-lactamase. Rapid, whole genome sequencing technology indicated that the disease causing genotype differed from a presumptive progenitor strain by only 24 out of 1144 core genes; a phylogenetic analysis suggested that the outbreak genotype arose from a linear series of gene acquisitions and losses (Mellmann et al. 2011). Clinical patient data indicated that the higher rates of HUS induced by the outbreak strain make it exceptionally more virulent than other STEC strains, even though it lacks an adherence factor gene (eae) found in 97% of STEC strains producing HUS in children (Frank et al. 2011). The strain was spread via fenugreek sprouts, due to fecal contamination of the seeds (Buchholz et al. 2011). The combination of genomic, clinical, and epidemiological research elucidated the origin and spread of the outbreak. Similar analysis has been performed on other recent microbial disease outbreaks, such as a 2006–2008 tuberculosis outbreak in British Columbia (Gardy et al. 2011), the 2009 H1N1 influenza outbreak (Smith et al. 2009a), and the Haitian 2011–2012 cholera epidemic (Reimer et al. 2011). The evolution of some microbial pathogens over much longer time spans (e.g., tuberculosis, Smith et al. 2009b) has also proven amenable to elucidation via genomic approaches, albeit with less certainty about the causal mechanisms and phylogenetic relationships. In these and similar cases, the combination of reduced complexity, rare recombination, and relatively rapid host expansions simplifies the molecular genetic basis of adaptation, and was essential for substantive success involving genomic approaches.

To be sure, there are instances of genes or genomic regions that strongly contribute to differences in traits of interest in particular environmental and genetic contexts (e.g., Abzhanov et al. 2006). Yet, it is now clear that extrapolation from these major genetic effects is misleading. The special cases are of great interest, but they are not representative, and their effects are frequently overwhelmed in realistic genetic and environmental contexts (Connelly and Akey 2012, Rockman 2012). The expression of most organismal traits reflects the action of many genes. The more assiduous are efforts to unravel the complexity, the more idiosyncratic is the detail that emerges (Martin and Willis 2010), as has been demonstrated in human studies (Pritchard and Di Rienzo 2010; Pritchard et al. 2010).

It is commonly claimed that sequence data are the key to explanation for phenotypic evolution and that this research program is ‘on the cusp’ of yielding these explanations (examples in Rockman 2012), but the time has long passed for expecting that the manifold contributors to variation in trait expression can be resolved into the molecular details that confer organismal phenotype, whether to obtain broadly applicable insights, or specialized applications, such as therapies for medical conditions of complex etiology. In efforts to explain evolution of phenotypes, focus on particular alleles at specific loci is misplaced. The severe limitations on the prospects for the research program to relate phenotype to molecular genetic detail have prompted calls for rethinking of approaches to advancing explanation of phenotypic change (Keller 2000; Lewontin 2002; Erickson et al. 2004; Weiss 2008; Keller 2010; Grosholz 2011).

This reorientation is urgently needed because many pressing questions demand answers. For populations subject to novel selection as environment changes, for example, under ongoing climate warming, how rapidly can adaptation proceed and over how many generations can it continue? What evolutionary mechanisms account for limits of species ranges?Escherichia coli populations adapt in the laboratory to conditions beyond their apparent upper thermal limit if density declines sufficiently to allow weakly competitive, but thermal-tolerant, genotypes to increase in frequency (Mongold et al. 1999). This demonstrates that, as theory has indicated (Gomulkiewicz and Holt 1995, Gomulkiewicz et al. 2010), feedback between demography and genetics can critically influence evolutionary rescue from extinction as environment changes. To understand potential for evolutionary rescue in nature, extensive experimental research is required (see also Box 2).

Box 2. Recent examples of advances in evolution and related fields via direct studies of process. These examples were chosen to represent diverse research goals. Numerous others could be included.


Accumulation of mitochrondrial defects (Taylor et al. 2002)

Evolution of host shifts (Antonovics et al. 2002)

Processes involved in stabilizing the legume-rhizobium mutualism (Heath and Tiffin 2009)

Ecosystem effects on adaptation (Bassar et al. 2012)

Evolution of multicellularity (Ratcliff et al. 2012)

Maintenance of reproductive isolation: sympatric, compatible fig spp. (Moe and Weiblen 2012)


Overcoming chestnut blight (Clark et al. 2011)

Novel breed development: honeycrisp apple (Luby and Bedford 1990)

Breeding for durability of resistance of crops to pathogens (Brun et al. 2010)


Bacteriotherapy for recurrent Clostridium infection (Khoruts and Sadowsky 2011)

Blocking the spread of dengue (Hoffmann et al. 2011)

Blocking the spread of hospital acquired infections (WHO 2010)

Predicting emerging infectious disease (Jones et al. 2008)

Vaccine deployment (Arinaminpathy et al. 2012)


Determinants of cancer (Hiatt and Breen 2008)

Environmental role in schizophrenia (van Os et al. 2010)

Maintaining cognitive function (Nagamatsu et al. 2012)

Patient-centered medicine (Krahn and Naglie 2008)

Direct, experimental study of process, at a level proximate to the phenomenon of interest, is a powerful means to accomplishing the goals of evolutionary explanation and prediction. Influential early examples include evolutionary studies of industrial melanism in the peppered moth (Kettlewell 1955, 1956) and of metal tolerance in grasses on mine spoils (Antonovics and Bradshaw 1970). Despite the current predominance of efforts to map genes for evolutionarily significant complex traits, there are also numerous recent examples of direct study of process (see Box 2). This approach will frequently require explicit considerations of scale, in spatial (Hauert and Doebeli 2004) or temporal (Williams 1992) dimensions, or levels of biological complexity (Okasha 2006). In the past, comprehensive study of natural selection in multicellular organisms has been stymied by the highly skewed, compound distributions characteristic of their variation in fitness, but statistical developments have recently alleviated this impediment (Shaw et al. 2008). Studies that experimentally evaluate the roles of evolutionary processes in phenotypic change minimize potentially misleading effects of uncontrolled and unanticipated variation in conditions, such as those that have always plagued inference of the human heritability of IQ (Lewontin 1975). Attempts to determine the how of trait expression, the G-P map, are typically undermined by conceptual and methodological difficulties, as we have seen; beyond this, because they focus on static phenotypic comparisons to assess the molecular basis of difference in phenotype, they cannot shed light on the why of past and ongoing phenotypic change.

Renewed focus on the process of phenotypic evolution would not only advance understanding, but would also support progress in applied research (examples in Box 2). The approaches of genetic engineering and synthetic biology are being used to modify microbes for the production of fuel and the degradation of pollutants. Progress is frequently slowed by unanticipated negative fitness effects and genetic interactions (Dunham 2007; Kwok 2010). Such impediments subside when the focus shifts directly to the phenotype of interest and to the evolutionary and ecological processes impinging on it (Bunka and Stockley 2006; Saxer et al. 2009; Hillesland and Stahl 2010). This shift of perspective will also hasten advances in fields such as infectious disease (Craft et al. 2009; Barton and Turelli 2011), psychopathology (Krueger and Eaton 2010), and other fields of human health. With the focus on phenotype, more general understanding of responses to treatments will emerge. A focus on the ecological and evolutionary processes that lead to change in phenotype, rather than molecular details that are generally not explanatory, promotes success.

Associate Editor: D. Fairbairn


We thank many colleagues for suggestions that greatly improved the manuscript, the Biological Interest Group of the Minnesota Center for Philosophy of Science for stimulating discussion, and NSF for support of our research programs. The authors declare that they have no competing financial interests.