vol . 1 9 3 , no . 1 the amer ican natural i st january 20 19

American Society of Naturalists Address

From the Past to the Future: Considering the Value

and Limits of Evolutionary Prediction*

Ruth G. Shaw†

Department of Ecology, Evolution, and Behavior, University of Minnesota, Twin Cities, Saint Paul, Minnesota 55108

Submitted April 27, 2018; Accepted June 22, 2018; Electronically published October 30, 2018

abstract: The complex interplay of the multiple genetic processesof evolution and the ecological contexts in which they proceed frus-trates detailed identification of many of the states of populations, bothpast and future, that may be of interest. Prediction of rates of adapta-tion, in the sense of change in mean fitness, into the future would,however, valuably inform expectations for persistence of populations,especially in our era of rapid environmental change. Heavy invest-ment in genomics and other molecular tools has fueled belief thatthose approaches can effectively predict adaptation into the future. Icontest this view. Genome scans display the genomic footprints ofthe effects of natural selection and the other evolutionary processesover past generations, but it remains problematic to predict futurechange in mean fitness via genomic approaches. Here, I advocate for adirect approach to prediction of rates of ongoing adaptation. Followingan overview of relevant quantitative genetic approaches, I outline thepromise of the fundamental theorem of natural selection for the studyof the adaptive process. Empirical implementation of this concept canproductively guide efforts both to deepen scientific insight into the pro-cess of adaptation and to informmeasures for conserving the biota in theface of rapid environmental change.

Keywords: adaptation, experimental evolution, fundamental theoremof natural selection, quantitative genetics.

Introduction

Evolutionary biologists recognize immense impediments tospecific, quantitative evolutionary prediction about popula-tions in nature. The challenge stems from the size and com-plexity of genomes, along with the multiple processes thatcan change the genetic composition of populations: naturalselection, gene flow, mutation, and genetic drift. In mostspecific cases, each of these processes is profoundly difficult

* Ruth Shaw received the 2017 Sewall Wright Award. The Sewall WrightAward, established in 1991, is given annually and honors a senior but still ac-tive investigator who is making fundamental contributions to the society’sgoals, namely, promoting the conceptual unification of the biological sciences.† Email: shawx016@umn.edu.

Am. Nat. 2019. Vol. 193, pp. 1–10. q 2018 by The University of Chicago.0003-0147/2019/19301-58432$15.00. All rights reserved.DOI: 10.1086/700565

This content downloaded from 128.1All use subject to University of Chicago Press Term

to characterize in detail. In addition, environment, whichmediates much selection, varies over time and space, oftenidiosyncratically. The dependence of a population’s evolu-tionary change on its history (e.g., Travisano et al. 1995)compounds the challenge. Together, these characteristicsof evolutionobviate accuratepredictionof apopulation’sde-gree of adaptation—that is, its mean absolute fitness manygenerations into the future (Gerrish and Sniegowski 2012)—and, even over the short term, frustrate prediction of manyattributes of interest.All of these reasons for doubting the possibility of de-

tailed evolutionary prediction raise two key questions: Whatevolutionary changes, if any, are currently feasible to pre-dict? And even more pertinent: Would it be worthwhile tohave these predictions? It is important to be clear aboutthe meaning of the word “prediction.” In this essay, I am us-ing “predict” in the common language sense of foretelling at-tributes of a population sometime in the future, an objectivethat must draw not only on theory but also on data from thepopulation(s) of interest. I am not considering retrodictionor hindcasting, that is, inference of past states. Nor am I re-ferring to qualitative prediction, the province of evolutionarytheory developed as proof-of-concept models (Servedio et al.2014), nor yet to prediction in the statistical sense of valuesfitted from data in accordance with a statistical model. Eachof these kinds of effort is valuable in its own right, but theyare distinct from predicting states of populations in the fu-ture. One kind of evolutionary prediction for which all ofthe necessary approaches are well established is predictionof change inmean fitness, that is, the rate of adaptation. I herepropose that this research program merits extensive imple-mentation. Beyond its intellectual interest, evaluating the rateof adaptation now looms in practical importance as environ-mental change at extreme rates threatens the persistence ofpopulations and species globally.

Can Genomics Predict Change in Mean Fitness?

With current heavy investment in genomics and other mo-lecular tools, some have opined that those approaches can

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

2 The American Naturalist

effectively yield predictions of evolutionary adaptation inthe sense of change in mean fitness. I challenge this claim.Genome scans yield the genomic historical record of the ef-fects of natural selection, along with all of the other evolu-tionary processes, accumulated over many generations.Distinguishing from this record the loci that contributedto past adaptation remains problematic because of historythat generates population structure and linkage disequilib-rium (see, e.g., Tiffin and Ross-Ibarra 2014; Schrider et al.2015; Hoban et al. 2016). Moreover, loci whose effects areindividually small but that can collectively account for muchof the adaptive change are unlikely to be detected (e.g., Lau-rie et al. 2004). Thus, loci or genomic regions that can bediscerned as possibly subject to selection generally repre-sent a small fraction of those that actually contribute to se-lection response, and even these may often be mistakenlyidentified (Schrider et al. 2015). More importantly, genomicapproaches do not generally elucidate rates of change inmeanfitness except, at best, extremely indirectly (see Bay et al.2017, especially the numerous caveats therein). The tempo-ral scale of the genomic signal of adaptation is that of manygenerations, and the spatial scale of most genomic studiesfar exceeds that occupied by an interbreeding population.In contrast, adaptation of a population in nature proceedson the timescale of generations and can continue over thevast sweeps of time into the future only in populations thatadapt to the vagaries of environment that confront themgeneration by generation and thus maintain an absolute fit-ness sufficient to persist. Adaptation over this timescale war-rants direct study prospectively. I propose that quantitativegenetic approaches offer evolutionary predictions that wouldbe of great value, both in deepening scientific understand-ing of evolutionary process and in informing measures toaddress issues of pressing societal concern. This line ofstudy is challenging, but I urge that the payoffs warrantthe effort.

Prediction via Quantitative Genetics

Whereas anticipation of evolutionary change is largely lim-ited to qualitative prediction, quantitative genetic studystands as a key exception. Its approaches have long enabledquantitative prediction of change in one or more trait meanphenotypes in response to selection via the breeder’s equa-tion, R p h2S (Falconer and Mackay 1996), and its multi-variate form, D�Z p GP21S p Gb (Lande 1979; Landeand Arnold 1983). Here, for a set of traits of interest, D�Zis the vector of the change in their means from one gener-ation to the next (R for a single trait); G is the matrix ofadditive genetic variances of and covariances between thetraits; P is the matrix of their phenotypic variances andcovariances; S, the selection differential, is the phenotypic

This content downloaded from 128.1All use subject to University of Chicago Press Term

covariance between fitness and each of the traits; and b,the selection gradient, as a measure of selection directly oneach trait, is the partial regression of fitness on the traits. Inthe single-trait case, h2 is the trait heritability, the ratio ofthe additive genetic variance of a trait to its phenotypic vari-ance. Reliable prediction is here favored by considering inaggregate the many polymorphic loci that collectively con-tribute to the standing genetic variation in traits and to thegenetic covariance between fitness and traits. These equa-tions, built on the regularities of Mendelian transmission,yield predictions for the change in trait mean from onegeneration to the next, a modest timescale.Artificial selection experiments in controlled conditions

have demonstrated the utility of the quantitative predic-tions, even though realized responses to selection vary amongreplicate populations and generations, reflecting random-ness of population sampling and genetic transmission. Sher-idan (1988), in his review of selection in livestock and lab-oratory populations of Drosophila and Tribolium, deemedselection responses to differ substantially from predictions.However, Hill and Caballero (1992) offer a critique of how thecomparisons were made (see also Walsh and Lynch 2018,p. 607) and a more encouraging view of agreement betweenpredicted and realized responses to selection. Realized re-sponses often show high repeatability and accord well withpredictions (e.g., Enfield et al. 1966; Carey 1983; Conner et al.2011). Moreover, even though h2 is expected to change withthe changes in allele frequencies implied by the observedresponse to selection, in practice the rate of change in a traitor traits under artificial selection sometimes holds far beyondthe initial generation—for 10–30 generations (Yoo 1980) ormany more for large populations (Weber 1990; Weber andDiggins 1990).Within highly inbred lines, in which the stand-ing genetic variation is nil (or very nearly so), the response toselection is, as expected, not detectable initially (López andLópez-Fanjul 1993; Mackay et al. 1994; Keightley 1998).Thus, in accord with the theory for predicting selection

response, empirical studies in the laboratory and the green-house have demonstrated the general reliability of predic-tions of trait change under artificial selection. But is it rea-sonable to expect such reliability under natural selection?One reason to believe so is that the equations for predictingresponse to selection apply equally for natural selectionwhen all of the traits under selection are known and takeninto account (Lande 1979). However, artificial selectioncontrasts with natural selection in a crucial way. In the for-mer the experimenter chooses the trait or traits as the basisfor selection, whereas in the wild many traits may be sub-ject to natural selection, and we do not know which onesa priori. To address this problem, Lande and Arnold (1983)presented theory and methods for inferring the directionand magnitude of natural selection on each of a specifiedset of correlated quantitative traits.

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

Predicting Adaptation 3

Challenges in Predicting Trait Changedue to Natural Selection

Lande and Arnold’s (1983) influential paper inspired manyfield studies to assess the relationship between a measure ofindividuals’ fitness and measures of phenotypic values ofmultiple traits (the approach is outlined in fig. 1A; studiesreviewed by Kingsolver et al. 2001). Such studies yieldestimates of the strength and direction of selection directlyon individual traits, b, the vector of selection gradients.They may also indicate curvature, estimated as the matrixg, in the relationship between fitness and traits, providingan estimate of the adaptive landscape in which the popula-tion is evolving. As in the examples Lande and Arnold gave,many such studies have based inference of selection on asingle component of fitness (e.g., survival over a particularinterval). Moreover, phenotypic selection studies have gen-erally considered a small number of traits because studyingselection on many traits poses at least two important chal-lenges. First, there are practical problems with obtainingmeasures of many traits, along with fitness records, on eachof many individuals. Automation has made trait measure-ment feasible on very large study populations for single(Weber 1990; Weber and Diggins 1990) and numerous (e.g.,Houle et al. 2003) traits, enabling artificial selection on largescales. However, not all traits that may be targets of selectionwill be amenable to automated measurement, which is likelyto be largely restricted to laboratory studies. Second, the de-mand for data to ensure statistical power and precision in-creases dramatically with the number of correlated traitsconsidered (Mitchell-Olds and Shaw 1987). Thus, for a givennumber of individuals the precision of estimates of selectionon each trait declines markedly as the number of traits in-creases. Moreover, a crucial limitation of the approach is themissing character problem. The particular traits chosen forstudy need not (and are unlikely to) account for all, or evenmost, of the variation in fitness. Omission of traits that are,in fact, under selection and are correlated with traits underconsideration biases the estimates of selection on those traits.Furthermore, environmentally induced covariation betweenfitness and traits leads to spurious inference of selection(Mitchell-Olds and Shaw 1987; Rausher 1992).

In contrast to the body of phenotypic selection studies,few studies of selection in nature have been designed toyield estimates of the genetic parameters, G, needed forquantitative evolutionary prediction. Ideally, such estima-tion employs experimental designs that place progeny offormal genetic crosses in the wild as the individuals whosefitness and traits are recorded (fig. 1C; e.g., Rausher andSimms 1989; Campbell 1996; Tiffin and Rausher 1999; Et-terson and Shaw 2001). Thus, studying expression of fit-ness and traits in the context of a quantitative genetic ex-periment conducted in nature yields all of the components

This content downloaded from 128.1All use subject to University of Chicago Press Term

required to predict evolutionary genetic change from themultivariate breeder’s equation (or directly, as the additivegenetic covariance between fitness and traits, covA(w, z);e.g., Etterson and Shaw 2001). Observational studies of ped-igreed populations of vertebrates have also yielded estimatesof the parameters required for prediction of response to nat-ural selection on traits (fig. 1B; e.g., Grant and Grant 1995;Kruuk et al. 2002; Sheldon et al. 2003). In some of thesecases, the realized change in the traits aligned well with thepredictions (e.g., Grant and Grant 1995). In their review ofthe vertebrate studies, however, Gienapp et al. (2008) foundthat this is often not the case, noting the vexing issue ofdistinguishing genetic response to selection from pheno-typic plasticity in response to changing environment (seealso Pujol et al. 2018). Pemberton (2010) emphasizes otherimpediments to accurate prediction from observational stud-ies. Paramount among them is the general problem of con-founding of environmental influences on organisms withgenetic influences, which stymies the effort to ascertainthe genetic contribution to fitness and traits. This is the clas-sic problem of discerning the effects of nature versus nurture(Lewontin 2000). Ironically, comparison of predicted to re-alized responses to selection are far scarcer for the experi-mental studies of annual plants, which are less subject tothese impediments. As one instance, however, Franks et al.(2007) contemporaneously grew generations collected be-fore and after drought and thus conclusively demonstrated agenetically based response to natural selection on floweringtime that accorded with prediction.Whereas the challenges for evolutionary prediction from

observational evidence are clear, the examples of experi-mental studies tantalizingly intimate the predictive promiseof quantitative genetics that is more widely achievable. Im-plementation ofD�Z p GP21Swill, however, always entail afraught choice of the characters to consider, of which thereare many more possible candidates as targets of selectionthan it would be feasible tomeasure. Moreover, the strengthof selection on them and even which characters matter forfitness seem likely to vary as environment changes overtime, although limits on statistical precision may often makethat difficult to prove (Morrissey and Hadfield 2012).

Predicting Adaptation: Fisher’s Fundamental Theorem

An alternative and potentially more robust way forward isdirect prediction of the rate of adaptation via the funda-mental theorem of natural selection (FTNS; Fisher 1930):D �W p VA(W)= �W . Here, the evolutionary (i.e., geneticallybased) change in mean fitness from the current generationto the next is predicted by the ratio of the population’s cur-rent additive genetic variance for fitness to its mean fitness.Fitness, W, can be defined operationally as the number ofoffspring an individual contributes to the population, that

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

Figu

re1:

Schematic

diagram

ofapproaches

tostud

ying

naturalselectionin

thewild

high

lightingtheirsimilarities

anddifferences.Observation

alapproaches:(A

)stud

yof

phenotypic

selectionon

traitsand(B)stud

yof

phenotypicselectionon

traits,joinedwithpedigree

inform

ationperm

ittingestimationof

thematrixof

genetic(co)variancesof

thetraits,G

,andthus

enablin

gprediction

ofchan

gein

traitmeans.E

xperim

entalapp

roaches:(C)form

alcrossesensure

that

progenyarederivedby

rando

mmating;progenyaredistribu

tedat

rando

mover

the

environm

ent,ensuring

independenceof

geneticandenvironm

entaleffects.Study

ofph

enotypicselectionon

traits;use

ofpedigree

toestimateGandthus

enablin

gprediction

ofchan

gein

traitmeans.D

,Implem

entation

ofthefund

amentaltheorem

ofnaturalselection(FTNS).A

sin

C,formal

crossesensure

that

progenyarederivedby

rando

mmating;

progenyaredis-

tributed

atrando

mover

theenvironm

ent,ensuring

independ

ence

ofgenetic

andenvironm

entaleffects.Fitnessrecordsas

completeas

practicalob

tained

onindividu

als.Estim

ates

ofV

A(W

),� W,andV

A(W

)=� W

topredictchange

inmeanfitness.Asterisks

indicate

that

estimates

aresubjectto

concernsabou

tconfoun

ding

betweengenetic

andenvironm

entaleffects

(A,B)andmissingcharacters

(A–C).Primeanddo

uble

prim

esymbo

lsindicate

distinct

coho

rtsfrom

thesamesetof

progeny.

This content downloaded from 128.101.134.127 on January 07, 2019 07:34:03 AMAll use subject to University of Chicago Press Terms and Conditions (http://www.journals.uchicago.edu/t-and-c).

Predicting Adaptation 5

is, the per capita population growth rate. Thus, �W p 1 yieldsthe expectation that the population will remain the samesize, and a magnitude of �W greater or less than 1 indicatesthe rate at which the population is expected to grow or de-cline. The ratio VA(W)= �W predicts the change in the pop-ulation’s growth rate resulting from genetically based dif-ferential contribution of offspring to the next generation.Importantly, in cases with an estimate of �W ! 1, the magni-tude of VA(W)= �W offers a prediction of the rate of adap-tation toward �W p 1 and, in particular, whether a singlegeneration of natural selection can be expected to result inevolutionary rescue, stabilizing the population’s persistence.Although the interpretation and even the validity of theFTNS were unclear at first, the derivation has now been ver-ified as quite general, and the interpretation has been eluci-dated (e.g., Ewens 1989, 2004; Frank 2012). A crucial pointis that the FTNS predicts the change in mean fitness re-sulting strictly from genetic changes due to selection. Thisis the rate of evolutionary adaptation. The total change inmean fitness may deviate from this substantially becausethe environment in which the offspring develop may differin ways that directly cause their mean fitness to differ fromthat of their parents (i.e., even in the absence of any geneticchange; phenotypic plasticity with respect to fitness).

The utility of FTNS in empirical contexts has not yetbeen demonstrated; no published empirical research, tomy knowledge, has applied it, although we have such stud-ies under way (Eule-Nashoba 2016; Sheth et al. 2018; R. G.Shaw and S. Wagenius, unpublished data [see the overviewin “Experimentally Applying FTNS” below as well as Shawand Etterson (2012) and Gomulkiewicz and Shaw (2013)]).Fisher (1930) himself likely dampened enthusiasm for empir-ical application of FTNS by suggesting that concomitant en-vironmental deterioration (e.g., from competitive pressuredue to an increase in population density as adaptation pro-ceeds) would generally cancel out any increase in the meanabsolute fitness of individuals. This has a logical appeal. Itis clear that population growth cannot exceed 1 indefinitely,but how tight is such a governor? Others have suggested thatVA(W) would be readily exhausted by incessant selection onfitness itself (e.g., Charlesworth 1987). However, recent theo-retical work by Zhang (2012; see also Ellner and Hairston1994; Shaw and Shaw 2014) has made clear that various re-gimes of environmental change can maintain VA(W), suchthat it would be available as a basis for response to selec-tion. Moreover, evidence now emerging from genomic stud-ies indicates that past adaptation has generally proceededthrough subtle changes in allele frequency at many simulta-neously segregating loci (Pritchard and Di Rienzo 2010; seealso Boyle et al. 2017), in accordance with classical quantita-tive genetic thought. Even when adaptation has resulted inextremeallele frequencies, this may have occurred primarilyvia “soft sweeps” from standing genetic variation as op-

This content downloaded from 128.1All use subject to University of Chicago Press Term

posed to “hard sweeps” from de novo mutation (Messerand Petrov 2013).Thus, a strong reason for optimism that FTNS can yield

informative predictions of adaptive rate is that the magni-tude of the change in mean fitness depends on a popula-tion’s current overall standing genetic variance for fitnessin its current environmental context. Genetic contributionsto variation in fitness are likely to be at least as highly poly-genic as manymore readilymeasurable traits, such as oil con-tent of corn kernels (Laurie et al. 2004) and human height(Yang et al. 2010). Moreover, at any given time it is not nec-essary to consider mutations that have newly arisen. Thosethat could eventually contribute to selection response are notlikely to contribute importantly just after they originate, whenthey are rare.To return to the key questions above: What evolutionary

changes, if any, are currently feasible to predict? FTNS of-fers a strong conceptual basis for direct, quantitative pre-diction of the rate of adaptation, in the sense of change inmean absolute fitness. The methodologies of quantitativegenetics enable its implementation, and experimental ap-proaches of population biology can guide assessment of theextent to which the predictions are realized. Would it beworthwhile to have these predictions? Quantitative predic-tions of the rate of genetic adaptation warrant the effort toobtain them for both fundamental and practical reasons.Direct study of this kind would strengthen our under-standing of the process of adaptation, including feedbacksbetween evolutionary change and population dynamics(community genetics; Antonovics 1992; also known as eco-evolutionary feedbacks). It would also inform efforts to pro-mote population persistence and guide those efforts to wherethey are most needed.Among the specific empirical questions that follow are

these: What is the magnitude of VA(W), in particular natu-ral populations? What is the rate of adaptation that itpredicts? How closely do realized rates of adaptation accordwith predictions? If they align closely, over how many gen-erations? If they differ substantially, why? These questionshave not been answered, not even for populations in the lab-oratory—nor have they been answered for wild populationsin relatively undisturbed conditions, nor especially for popu-lations whose fitnesses (hence, whose likelihood of persis-tence or increase) are compromised by rapid changes in en-vironment.

Experimentally Applying FTNS

Implementation of FTNS requires estimation of the pop-ulation parameters VA(W) and �W. Experiments to do so(fig. 1D) can draw on the general approaches of quantita-tive genetics for estimating the additive genetic variance,

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

6 The American Naturalist

as laid out in textbooks (Falconer andMackay 1996, chap. 10;Lynch and Walsh 1998, chaps. 18, 20). In brief, estimationof additive genetic variance of any trait, including absolutefitness, entails measures of the trait on a pedigreed set of in-dividuals that is representative, both in the genetic samplingand in the environment in which it is grown, of the popula-tion of interest. For the purpose at hand, it is appropriate toconsider a population in the sense of a deme (or, more spe-cifically, a gamodeme; Gilmour and Gregor 1939), a localgroup likely to be exchanging genes. Because of a deme’s typ-ically restricted spatial extent, most of its members may besubject to many (if not all) of the same ongoing selective pro-cesses.

Eventual interpretation of the experiment depends criti-cally on avoidance of confounding between genetic and en-vironmental effects. To ensure this independence, sets ofexperimental individuals can be generated from randomchoice of a large number of parental individuals by inter-crossing the parents in a formal, randomized mating scheme.In a common approach, individual paternal parents are eachmated to a distinct set of multiple maternal parents, as de-scribed by Falconer and Mackay (1996, chap. 10), to obtainprogeny in paternal half-sib groups that include subgroupsof full sibs. Any mating scheme that results in groups of pa-ternal half-sibs will allow estimation of additive genetic vari-ance unbiased by variance due to dominance or maternal ef-fects. The progeny are set out, randomly arrayed over the (oneor possibly more) focal environment(s), to develop and ma-ture in conditions as similar as possible to the conditions un-der which rate of adaptation is the focus. Raising the progenyat the site fromwhich the parents were sampled offers the op-portunity to predict the ongoing rate of adaptation of thepopulation under the conditions prevailing in situ.

The progeny are then followed, ideally throughout the lifespan, and measures of components of fitness are obtainedfor each (see, e.g., Campbell 1997; Stanton-Geddes et al. 2012),to accumulate records of lifetime fitness. With these datain hand, it remains to estimateVA(W) and �W. The frequencydistribution of lifetime fitness does not lend itself to stan-dard statistical analysis; as a compound distribution arisingfrom the serially expressed components of fitness, lifetimefitness conforms to no standard probability distribution (Shawet al. 2008). To alleviate this problematic aspect of fitnessdata, Geyer et al. (2007) developed aster modeling, whichtakes into account the dependence structure of arbitrarilymany components of fitness to obtain, via maximum like-lihood, usefully precise inferences about fitness (see alsoShaw et al. 2008). Its extension to accommodate random ef-fects, including random genetic effects (Geyer et al. 2013),enables estimation ofVA(W) and ofVA(W)/ �W, the predictionfrom the FTNS of the increase inmean fitness of the next gen-eration over the current one due to the change in allele fre-quencies resulting from selection in the current generation.

This content downloaded from 128.1All use subject to University of Chicago Press Term

It is important to acknowledge challenges. For example,whereas individual fitness is, in principle, simple to defineas the number of offspring an individual contributes to thepopulation, the biology of fitness expression, whichmay spana single year or many, often complicates the measurement offitness in practice. Apart from the issue of life span, it is of-ten most convenient to assess fitness through maternity,ignoring paternal fitness variation because of the difficultyof attributing offspring to particular sires. It is now, however,feasible to assign paternity with considerable confidence (e.g.,Kulbaba andWorley 2012; Briscoe Runquist et al. 2017).Withsuch information, FTNS could be applied using estimatesbased on either maternal fitness or paternal fitness.

Comparing Realized Adaptation to Prediction

Predictions of the evolutionary change in mean fitness arewarranted in their own right because they would directlycharacterize a population’s expected capacity for ongoingadaptation to its current environment. They would be ofeven greater value in comparison to the realized change inmean fitness. To estimate the realized change in mean fit-ness, a representative sample of the next generation mustbe grown and its mean fitness evaluated. As noted above,the environments in which the parental and offspring co-horts develop are likely to differ, and this could directly con-tribute to the difference in mean fitness between parentaland offspring cohorts, in addition to the genetically basedchange in mean fitness. To isolate the latter, the parentaland offspring cohorts must be grown in a common envi-ronment—that is, a genetically representative sample of theoriginal pedigreed cohort must also be grown and its fit-ness evaluated, along with the offspring cohort. This ap-proach parallels that of mutation accumulation experiments,with similar rationale (e.g., Schultz et al. 1999; Shaw et al.2000).At the outset, it is important to recognize that the realized

change in mean fitness may often poorly match the predic-tion, even when the direct effect of change in environmentis addressed as described above. The overall change inmeanfitness also depends on genetic variation in sensitivity tosuch environmental change (i.e., the interaction betweengenotype and temporally varying environment), which sig-nifies temporal change in genetic selection, whether thischange is erratic or due to more regularly fluctuating selec-tion. Moreover, even in experimentally controlled labora-tory environments, rates of change in mean fitness can varysubstantially among replicate microbial populations subjectto the same selection regime (e.g., Travisano et al. 1995; Ra-miro et al. 2016), although it is important to note that thesestudies typically span 1,000 generations or many more. Thematch between realized and predicted change in mean fit-ness for wild populations—and explanations for deviations

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

Predicting Adaptation 7

between them—are empirical questions whose answers cancome only from direct experimental study.

Potential Breadth of Application and Variants on It

Unquestionably, experiments of this kind require consider-able effort, and, even more obviously, some kinds of organ-isms are better suited to such experiments than others. Inparticular, plants and other sessile organisms “stand stilland wait to be counted” (Harper 1977, p. 515). Even studiesof these organisms alone would be highly informative, butothersmay also be good candidates. Via’s (1991) confinementof aphids on host plants in the field to obtain full fitness eval-uations demonstrates that there is much greater potential forsuch studies. More vagile organisms would have to be taggedand reliably recoverable for fitness assessment, as has beendone with several experimentally manipulated vertebrate pop-ulations (e.g., De Lisle and Rowe 2015; Bolnick and Stutz 2017).

Organisms whose lives span decades pose obvious chal-lenges. Nevertheless, for a few perennial plants with lifespans up to a decade, investigators have conducted formalquantitative genetic experiments in the field, obtainingcomplete lifetime fitness measures (Campbell 1997; Camp-bell et al. 2008; Shefferson and Roach 2012), and othersare in progress with plants of still greater life span (R. G.Shaw and S. Wagenius, unpublished data). A priori, for re-liable prediction from FTNS it seems necessary to obtainfitness records over the full lifetime, particularly in caseswith strong genetically based trade-offs between early- andlate-life components of fitness. However, it is an empiricalquestion whether useful predictions of the rate of adaptationcan be obtained from incomplete fitness records. An en-abling condition would be that, after reproductive maturity,the genetic variance in fitness and mean fitness change to-gether through the life span such that the ratio remainsroughly constant. This condition would promote the feasi-bility of quantitative genetic studies to predict adaptation ofmuch longer lived organisms (e.g., trees; Yeh and Heaman1987; Aitken and Adams 1996; Warwell and Shaw 2017).

The foregoing outline represents a general approach onwhich many variants can be envisioned to address ques-tions related to the central ones. To evaluate how consistentselection is over time, it would be worthwhile to growsubsets of individuals from the same pedigrees at successivetimes. The predicted rate of adaptation may differ, but evenif it hardly differs, comparison of genetic effects on fitnessbetween cohorts could reveal interaction between genotypeand temporally varying environment such that the geneticcomposition favored by selection differs between them. Thespecifics underlying predicted and observed changes in meanfitness (which traits, which alleles, and what aspects of envi-ronment played key roles?) will often be elusive, simply be-cause of the enormously high dimensionality of each of these

This content downloaded from 128.1All use subject to University of Chicago Press Term

aspects. Even so, measurement of traits of individuals, in con-junction with their fitness components, would illuminate se-lection on those traits and suggest their roles in the adaptiveresponse. Inclusion of trait values could also bear on the roleof trait plasticity in adaptation (see, e.g., Chevin et al. 2010).Similarly, measurements of environmental attributes near in-dividuals could suggest roles those features play in selectionand adaptation. Because environmental attributes are gener-ally highly intercorrelated, conclusive evidence about theirroles as agents of selection would entail experimental manip-ulation (e.g., increasing CO2 concentration; Lau et al. 2007).As noted at the outset, it is not selection alone that

changes the genetic composition of populations. Geneticdrift, gene flow, and mutation also play important roles.The stochasticity of drift implies that it undermines the pre-dictability of change in mean fitness. The extent to whichit generates discrepancies between predicted and realizedchange in mean fitness could be directly studied by exper-imentally varying effective population size of study cohorts(e.g., Newman and Pilson 1997). Likewise, gene flow couldalso be varied experimentally to assess its interplay with se-lection. It would be of interest to vary the amount of geneflow via both haploids (e.g., pollen) and diploids, for whichthe genetic and demographic consequences differ, impor-tantly affecting the evolutionary dynamics and outcomes(Aguilée et al. 2013). It would also be of value to vary the pop-ulation sources (e.g., their distances or the disparity of thehabitats). Such studies can inform the pressing debate aboutthe merits and risks of human-mediated gene flow (managedrelocation, assisted migration) for enhancing the persis-tence of populations threatened by environmental change(Richardson et al. 2009; Aitken and Whitlock 2013). Newlyarising mutations are unlikely to substantially influence thechange in the mean fitness of the immediately subsequentgeneration due to their initial rarity. It would be of great in-terest, however, to conduct a study as outlined here on setsof mutation-accumulation lines as a basis for augmentingunderstanding of the contribution of newly arisingmutationsto ongoing changes in mean fitness in the wild.To predict a population’s rate of adaptation to different

conditions, the same reference population can be grownat different sites (e.g., a chronosequence representing climatechange; Etterson and Shaw 2001). Even within a site, the se-lective environment (sensu Antonovics et al. 1988) may vary.The approach outlined here could be further extended to clar-ify the spatial scale of selective heterogeneity (although at nofiner scale than could be addressed in an informatively repli-cated experiment). Fitness expressionmay depend on geneticcomposition of neighboring conspecifics or their density (e.g.,Shaw 1986; Campbell et al. 2017). The nature of such “soft se-lection” and how it bears on adaptation could be investigatedthrough experimental manipulation of neighbor density andrelatedness. To evaluate the role of a partner species in medi-

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

8 The American Naturalist

ating ongoing adaptation, exposure to the partner could bevaried (e.g., microbial inoculation; Alexander et al. 2017;Ossler and Heath 2018). Yet further, it is of interest to learnabout how capacity for adaptation changes over time. TheProject Baseline archive of seeds representing numerous spe-cies in substantial seed collections from populations occupy-ing diverse habitats over the period 2012–2017 (Etterson et al.2016) is an unparalleled resource for such studies. As experi-ence with this approach advances, it will eventually be possi-ble to address the extent to which generalization about ratesof adaptation beyond individual populations or over longerperiods is warranted.

Conclusion

Among the profound aspects of evolution is its playing outover enormous sweeps of time, from more than 4 billionyears in the past onward into the indefinite future. Evolu-tionary biologists, through a multiplicity of approaches,have woven countless insights into a fabric of understand-ing about organisms in the remote past and their changesup to the present. These considerations of evolution overunimaginable spans of time have tended to dwarf interestin evolutionary change on the timescale of organismal gen-erations. Darwin’s (1859) view that adaptation would pro-ceed via imperceptible changes accumulating exceedinglyslowly surely influenced this emphasis in the early periodof evolutionary study(Antonovics1987).Now,however,nu-merous retrospective studies have detected adaptation inthe wild over tens of generations or fewer (see above). Evenso, it is also clear, from field observations (Bradshaw 1991)and from experiments in the field (e.g., Newman and Pilson1997) and laboratory (e.g., Bell and Gonzalez 2011; Lindseyet al. 2013), that adaptation under natural selection may notsuffice to maintain a population in a habitat where its fitnesshas been severely compromised. Can we predict evolutionaryrescue and adaptive persistence versus failure to adapt, result-ing in ultimate extirpation for particular populations in thewild (Gomulkiewicz and Shaw 2013)? This is an empiricalquestion that has importance both for advancing basic evolu-tionary understanding and for informing measures intendedto conserve the biota as the environment changes so rapidly.

Efforts using this direct approach to predicting and eval-uating evolutionary adaptation require humility and mod-est expectations. From the outset, it is important to ac-knowledge that predictions may often be imprecise evenwhen they are accurate. Characterizing the magnitude ofuncertainty is itself worthwhile. Predictions of the geneti-cally based change in mean fitness will not likely be metin all cases. Even when they are not, well-designed experi-ments that enable both these predictions and evaluationof the extent to which they are met will directly illuminatethe adaptive process.

This content downloaded from 128.1All use subject to University of Chicago Press Term

Acknowledgments

I am deeply grateful to the American Society of Naturalistsfor honoring me with the Sewall Wright Award and for in-viting this contribution. I thank my mentors, especially JanisAntonovics and Joe Felsenstein, who introduced me to theseideas and career-long colleagues who helped develop them.My thanks to G. May and M. Travisano for thoughtful dia-logue; H. M. Alexander, J. R. Etterson, E. R. Grosholz, M.Kulbaba, A. C. Love, S. N. Sheth, P. Tiffin, and A. Weis forcomments on themanuscript; and participants in the Biolog-ical Interest Group of theMinnesota Center for Philosophy ofScience for probing, constructive discussion of these ideasand their presentation. D. Bolnick, P. Nosil, and an anony-mous reviewer provided comments that helped clarify keypoints. I gratefully acknowledge support from the NationalScience Foundation and the Environment and Natural Re-sources Trust Fund of Minnesota.

Literature Cited

Aguilée, R., F. H. Shaw, F. Rousset, R. G. Shaw, and O. Ronce. 2013.How does pollen vs. seed dispersal affect niche evolution? Evolu-tion 67:792–805. doi:10.1111/j.1558-5646.2012.01816.x.

Aitken, S. N., and W. T. Adams. 1996. Genetics of fall cold hardinessin coastal Douglas-fir in Oregon. Canadian Journal of Forest Re-search 26:1828–1837.

Aitken, S. N., and M. Whitlock. 2013. Assisted gene flow to facilitatelocal adaptation to climate change. Annual Review of Ecology,Evolution, and Systematics 44:367–388.

Alexander, H. M., E. Bruns, H. Schebor, and C. M. Malmstrom.2017. Crop-associated virus infection in a native perennial grass:reduction in plant fitness and dynamic patterns of virus detection.Journal of Ecology 105:1021–1031.

Antonovics, J. 1987. The evolutionary dys-synthesis: which bottlesfor which wine? American Naturalist 129:321–331.

———. 1992. Towards community genetics. Pages 426–449 in R. S.Fritz and E. L. Simms, eds. Ecology and evolution of plant resis-tance to herbivores and pathogens: ecology, evolution, and genet-ics. University of Chicago Press, Chicago.

Antonovics, J., N. C. Ellstrand, and R. N. Brandon. 1988. Environ-mental variation and genetic variation: expectations and experi-ments. Pages 275–303 in L. D. Gottlieb and S. K. Jain, eds. Plantevolutionary biology. Chapman & Hall, New York.

Bay, R. A., N. Rose, R. Barrett, L. Bernatchez, C. K. Ghalambor, J. R.Lasky, R. B. Brem, S. R. Palumbi, and P. Ralph. 2017. Predicting re-sponses to contemporary environmental change using evolutionaryresponse architectures. American Naturalist 189:463–473.

Bell, G., and A. Gonzalez. 2011. Adaptation and evolutionary rescue inmetapopulations experiencing environmental deterioration. Sci-ence 332:1327–1330.

Bolnick, D. I., and W. E. Stutz. 2017. Frequency dependence limitsdivergent evolution by favouring rare immigrants over residents.Nature 546:285–288.

Boyle, E. A., Y. I. Li, and J. K. Pritchard. 2017. An expanded view ofcomplex traits: from polygenic to omnigenic. Cell 169:1177–1186.

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

Predicting Adaptation 9

Bradshaw, A. D. 1991. The Croonian lecture 1991: genostasis and thelimits to evolution. Philosophical Transactions of the Royal SocietyB 333:289–305.

Briscoe Runquist, R. D., M. A. Geber, M. Pickett-Leonard, and D. A.Moeller. 2017. Mating system evolution under strong pollen lim-itation: evidence of disruptive selection through male and femalefitness in Clarkia xantiana. American Naturalist 189:549–563.

Campbell, D. R. 1996. Evolution of floral traits in a hermaphroditicplant: field measurements of heritabilities and genetic correlations.Evolution 50:1442–1453.

———. 1997. Genetic and environmental variation in life-historytraits of a monocarpic perennial: a decade-long field experiment.Evolution 51:373–382.

Campbell, D. R., A. K. Brody, M. V. Price, N. M. Waser, and G. Al-dridge. 2017. Is plant fitness proportional to seed set? an experi-ment and a spatial model. American Naturalist 190:818–827.

Campbell, D. R., N. M. Waser, G. Aldridge, and C. A. Wu. 2008.Lifetime fitness in two generations of Ipomopsis hybrids. Evolu-tion 62:2616–2627.

Carey, K. 1983. Breeding system, genetic variability, and response toselection in Plectritis (Valerianaceae). Evolution 37:947–956.

Charlesworth, B. 1987. The heritability of fitness. Pages 21–40 in J. W.Bradbury and M. B. Andersson, eds. Sexual selection: testing thealternatives. Wiley, Chichester.

Chevin, L.-M., R. Lande, and G. M. Mace. 2010. Adaptation, plastic-ity, and extinction in a changing environment: towards a predic-tive theory. PLoS Biology 8:e1000357.

Conner, J. K., K. Karoly, C. Stewart, V. A. Koelling, H. F. Sahli, and F. H.Shaw. 2011. Rapid independent trait evolution despite a strong pleio-tropic genetic correlation. American Naturalist 178:429–441.

Darwin, C. 1859. The origin of species. J. Murray, London.De Lisle, S. P., and L. Rowe. 2015. Ecological character displacement

between the sexes. American Naturalist 186:693–707.Ellner, S., and N. G. Hairston Jr. 1994. Role of overlapping generations

in maintaining genetic variation in a fluctuating environment.American Naturalist 143:403–417.

Enfield, F. D., R. E. Comstock, and O. Braskerud. 1966. Selection forpupa weight in Tribolium castaneum. I. Parameters in base popu-lations. Genetics 54:523–533.

Etterson, J. R., S. J. Franks, S. J. Mazer, R. G. Shaw, N. L. Soper Gorden,H. E. Schneider, J. J. Weber, K. J. Winkler, and A. E. Weis. 2016.Project baseline: an unprecedented resource to study plant evolutionacross space and time. American Journal of Botany 103:164–173.

Etterson, J. R., and R. G. Shaw. 2001. Constraint to adaptive evolu-tion in response to global warming. Science 294:151–154.

Eule-Nashoba, A. 2016. Fitness and adaptive capacity in a Minnesotaprairie. PhD diss. University of Minnesota.

Ewens,W. J. 1989. An interpretation and proof of the fundamental the-oremof natural selection. Theoretical Population Biology 36:167–180.

———. 2004. Mathematical population genetics. I. Theoretical in-troduction. 2nd ed. Springer, New York.

Falconer, D. S., and T. F. C. Mackay. 1996. Introduction to quanti-tative genetics. 4th ed. Longman, Essex.

Fisher, R. A. 1930. The genetical theory of natural selection. 2nd ed.Dover, New York.

Frank, S. A. 2012. Wright’s adaptive landscape versus Fisher’s funda-mental theorem. Pages 41–57 in E. Svensson and R. Calsbeek, eds.The adaptive landscape in evolutionary biology. Oxford UniversityPress, Oxford.

Franks, S. J., S. Sim, and A. E. Weis. 2007. Rapid evolution offlowering time by an annual plant in response to a climate fluctu-

This content downloaded from 128.1All use subject to University of Chicago Press Term

ation. Proceedings of the National Academy of Sciences of theUSA 104:1278–1282.

Gerrish, P. J., and P. D. Sniegowski. 2012. Real time forecasting of near-future evolution. Journal of the Royal Society Interface 9:2268–2278.

Geyer, C. J., C. E. Ridley, R. G. Latta, J. R. Etterson, and R. G. Shaw.2013. Local adaptation and genetic effects on fitness: calculationsfor exponential family models with random effects. Annals of Ap-plied Statistics 7:1778–1795.

Geyer, C. J., S. Wagenius, and R. G. Shaw. 2007. Aster models for lifehistory analysis. Biometrika 94:415–426.

Gienapp, P., C. Teplitsky, J. S. Alho, J. A. Mills, and J. Merilä. 2008.Climate change and evolution: disentangling environmental andgenetic responses. Molecular Ecology 17:167–178.

Gilmour, J. S. L., and J. W. Gregor. 1939. Demes: a suggested newterminology. Nature 144:333.

Gomulkiewicz, R., and R. G. Shaw. 2013. Evolutionary rescue be-yond the models. Philosophical Transactions of the Royal SocietyB 368:20120093.

Grant, P. R., and B. R. Grant. 1995. Predicting microevolutionary re-sponses to directional selection on heritable variation. Evolution49:241–251.

Harper, J. L. 1977. Population biology of plants. Academic Press, Lon-don.

Hill, W. G., and A. Caballero. 1992. Artificial selection experiments.Annual Review of Ecology and Systematics 23:287–310.

Hoban, S., J. L. Kelley, K. E. Lotterhos, M. F. Antolin, G. Bradburd,D. B. Lowry, M. L. Poss, L. K. Reed, A. Storfer, and M. C. Whitlock.2016. Finding the genomic basis of local adaptation: pitfalls, prac-tical solutions and future directions. American Naturalist 188:379–397.

Houle, D., J. Mezey, P. Galpern, and A. Carter. 2003. Automated mea-surement of Drosophila wings. BMC Evolutionary Biology 3:25.

Keightley, P. D. 1998. Genetic basis of response to 50 generations ofselection on body weight in inbred mice. Genetics 148:1931–1939.

Kingsolver, J. G., H. E. Hoekstra, J. M. Hoekstra, D. Berrigan, S. N.Vignieri, C. E. Hill, A. Hoang, P. Gibert, and P. Beerli. 2001. Thestrength of phenotypic selection in natural populations. AmericanNaturalist 157:245–261.

Kruuk, L. E. B., J. Slate, J. M. Pemberton, S. Brotherstone, F. Guinness,and T. H. Clutton-Brock. 2002. Antler size in red deer: heritabilityand selection, but no evolution. Evolution 56:1683–1695.

Kulbaba, M. W., and A. C. Worley. 2012. Selection on floral designin Polemonium brandegeei (Polemoniaceae): female and male fit-ness under hawkmoth pollination. Evolution 66:1344–1359.

Lande, R. 1979. Quantitative genetic analysis of multivariate evolu-tion, applied to brain∶body size allometry. Evolution 33:402–416.

Lande, R., and S. J. Arnold. 1983. The measurement of selection oncorrelated characters. Evolution 37:1210–1236.

Lau, J. A., R. G. Shaw, P. B. Reich, F. H. Shaw, and P. Tiffin. 2007.Strong ecological but weak evolutionary effects of elevated CO2

on a recombinant inbred population of Arabidopsis thaliana.New Phytologist 175:351–362.

Laurie, C. C., S. D. Chasalow, J. R. LeDeaux, R. McCarroll, D. Bush,B. Hauge, C. Lai, D. Clark, T. R. Rocheford, and J. W. Dudley. 2004.The genetic architecture of response to long-term artificial selectionfor oil concentration in the maize kernel. Genetics 168:2141–2155.

Lewontin, R. C. 2000. The triple helix: gene, organism, and environ-ment. Harvard University Press, Cambridge, MA.

Lindsey, H., J. Gallie, S. Taylor, and B. Kerr. 2013. Evolutionary res-cue from extinction is contingent on a lower rate of environmen-tal change. Nature 494:463–466.

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).

10 The American Naturalist

López, M. A., and C. López-Fanjul. 1993. Spontaneous mutation fora quantitative trait in Drosophila melanogaster. I. Response to ar-tificial selection. Genetics Research 61:107–116.

Lynch, M., and B. Walsh. 1998. Genetics and analysis of quantitativetraits. Sinauer, Sunderland, MA.

Mackay, T. F. C., J. D. Fry, R. F. Lyman, and S. V. Nuzhdin. 1994.Polygenic mutation in Drosophila melanogaster: estimates fromresponse to selection of inbred strains. Genetics 136:937–951.

Messer, P. W., and D. A. Petrov. 2013. Population genomics of rapidadaptation by soft selective sweeps. Trends in Ecology and Evolu-tion 28:659–669.

Mitchell-Olds, T., and R. G. Shaw. 1987. Regression analysis of nat-ural selection: statistical inference and biological interpretation.Evolution 41:1149–1161.

Morrissey, M. B., and J. D. Hadfield. 2012. Directional selection intemporally replicated studies is remarkably consistent. Evolution66:435–442.

Newman,D., andD. Pilson. 1997. Increased probability of extinctiondueto decreased genetic effective population size: experimental popula-tions ofClarkia pulchella. Evolution 51:354–362. doi:10.2307/2411107.

Ossler, J. N., and K. D. Heath. 2018. Shared genes but not shared ge-netic variation: legume colonization by two belowground sym-bionts. American Naturalist 191:395–406.

Pemberton, J. M. 2010. Evolution of quantitative traits in the wild:mind the ecology. Philosophical Transactions of the Royal SocietyB 365:2431–2438.

Pritchard, J. K., and A. Di Rienzo. 2010. Adaptation—not by sweepsalone. Nature Reviews Genetics 11:665–667.

Pujol, B., S. Blanchet, A. Charmantier, E. Danchin, B. Facon, P.Marrot, F. Roux, et al. 2018. The missing response to natural se-lection in the wild. Trends in Ecology and Evolution 33:337–346.

Ramiro, R. S., H. Costa, and I. Gordo. 2016. Macrophage adaptationleads to parallel evolution of genetically diverse Escherichia colismall-colony variants with increased fitness in vivo and antibioticcollateral sensitivity. Evolutionary Applications 9:994–1004.

Rausher, M. D. 1992. The measurement of selection on quantitativetraits: biases due to environmental covariances between traits andfitness. Evolution 46:616–626.

Rausher, M. D., and E. L. Simms. 1989. The evolution of resistanceto herbivory in Ipomoea purpurea. I. Attempts to detect selection.Evolution 43:563–572.

Richardson, D. M., J. J. Hellmann, J. S. McLachlan, D. F. Sax, M. W.Schwartz, P. Gonzalez, E. J. Brennan, et al. 2009. Multidimensionalevaluation of managed relocation. Proceedings of the National Acad-emy of Sciences of the USA 106:9721–9724.

Schrider, D. R., F. K. Mendes, M. W. Hahn, and A. D. Kern. 2015.Soft shoulders ahead: spurious signatures of soft and partial selec-tive sweeps result from linked hard sweeps. Genetics 200:267–284.

Schultz, S., M. Lynch, and J. Willis. 1999. Spontaneous deleteriousmutation in Arabidopsis thaliana. Proceedings of the NationalAcademy of Sciences of the USA 96:11393–11398.

Servedio, M. R., Y. Brandvain, S. Dhole, C. L. Fitzpatrick, E. E.Goldberg, C. A. Stern, J. Van Cleve, and D. J. Yeh. 2014. Not justa theory—the utility of mathematical models in evolutionary biol-ogy. PLoS Biology 12:e1002017.

Shaw, R. G. 1986. Response to density in a wild population of theperennial herb Salvia lyrata: variation among families. Evolution40:492–505.

Shaw, R. G., D. L. Byers, and E. Darmo. 2000. Spontaneous muta-tional effects on reproductive traits of Arabidopsis thaliana. Ge-netics 155:369–378.

This content downloaded from 128.1All use subject to University of Chicago Press Term

Shaw, R. G., and J. R. Etterson. 2012. Rapid climate change and therate of adaptation: insight from experimental quantitative genet-ics. New Phytologist 195:752–765.

Shaw, R. G., C. J. Geyer, S. Wagenius, H. H. Hangelbroek, and J. R.Etterson. 2008. Unifying life-history analyses for inference of fit-ness and population growth. American Naturalist 172:E35–E47.

Shaw, R. G., and F. H. Shaw. 2014. Quantitative genetic study of theadaptive process. Heredity 112:13–20.

Shefferson, R. P., and D. A. Roach. 2012. The triple helix of Plantagolanceolata: genetics and the environment interact to determinepopulation dynamics. Ecology 93:793–802.

Sheldon, B. C., L. E. B. Kruuk, and J. Merilä. 2003. Natural selectionand inheritance of breeding time and clutch size in the collaredflycatcher. Evolution 57:406–420.

Sheridan, A. K. 1988. Agreement between estimated and realised ge-netic parameters. Animal Breeding Abstracts 11:877–889.

Sheth, S. N., M. W. Kulbaba, R. E. Pain, and R. G. Shaw. 2018. Ex-pression of additive genetic variance for fitness in a population ofpartridge pea in two field sites. Evolution (forthcoming).

Stanton-Geddes, J., P. Tiffin, and R. G. Shaw. 2012. Role of climateand competitors in limiting fitness across range edges of an annualplant. Ecology 93:1604–1613.

Tiffin, P., and M. D. Rausher. 1999. Genetic constraints and selectionacting on tolerance to herbivory in the common morning gloryIpomoea purpurea. American Naturalist 154:700–716.

Tiffin, P., and J. Ross-Ibarra. 2014. Advances and limits of usingpopulation genetics to understand local adaptation. Trends in Ecol-ogy and Evolution 29:673–680.

Travisano, M., J. A. Mongold, A. F. Bennett, and R. E. Lenski. 1995.Experimental tests of the roles of adaptation, chance, and historyin evolution. Science 267:87–90.

Via, S. 1991. The genetic structure of host plant adaptation in a spa-tial patchwork: demographic variability among reciprocally trans-planted pea aphid clones. Evolution 45:827–852.

Walsh, B., and M. Lynch. 2018. Evolution and selection of quantita-tive traits. Vol. 2. Oxford University Press, Oxford.

Warwell, M. V., and R. G. Shaw. 2017. Climate-related genetic var-iation in a threatened tree species, Pinus albicaulis. AmericanJournal of Botany 104:1205–1218.

Weber, K. E. 1990. Increased selection response in larger popula-tions. I. Selection for wing-tip height in Drosophila melanogasterat three populations sizes. Genetics 125:579–584.

Weber, K. E., and L. T. Diggins. 1990. Increased selection response inlarger populations. II. Selection for ethanol vapor resistance inDro-sophila melanogaster at two populations sizes. Genetics 125:585–597.

Yang, J., B. Benyamin, B. P. McEvoy, S. Gordon, A. K. Henders,D. R. Nyholt, P. A. Madden, et al. 2010. Common SNPs explaina large proportion of the heritability for human height. NatureGenetics 42:565–569.

Yeh, F. C., and J. C. Heaman. 1987. Estimating genetic parameters ofheight growth in 7-year old coastal Douglas-fir from disconnecteddiallels. Forest Science 33:946–957.

Yoo, B. H. 1980. Long-term selection for a quantitative character inlarge replicate populations of Drosophila melanogaster. I. Re-sponse to selection. Genetics Research 35:1–17.

Zhang, X.-S. 2012. Fisher’s geometrical model of fitness landscapeand variance in fitness within a changing environment. Evolution66:2350–2368.

Editor: Daniel I. Bolnick

01.134.127 on January 07, 2019 07:34:03 AMs and Conditions (http://www.journals.uchicago.edu/t-and-c).