Volume 78, No. June 2003 The Quarterly Reviewpeople.umass.edu/lsadler/adlersite/kiers/DarwinianAgriculture1.pdfwheat (Triticum aestivum), rice (Oryza sativa), and maize (Zea mays)

Volume 78, No. 2 June 2003

145

The Quarterly Review of Biology, June 2003, Vol. 78, No. 2

Copyright � 2003 by The University of Chicago. All rights reserved.

0033-5770/2003/7802-0001$15.00

The Quarterly Reviewof Biology

DARWINIAN AGRICULTURE: WHEN CAN HUMANS FINDSOLUTIONS BEYOND THE REACH OF NATURAL SELECTION?

R. Ford Denison

Agronomy and Range Science, University of California

Davis, California 95616 USA

e-mail: [email protected]

E. Toby Kiers

Agronomy and Range Science, University of California

Davis, California 95616 USA

Stuart A. West

Cell, Animal & Population Biology, University of Edinburgh

Edinburgh, EH9 3JT, United Kingdom

keywordsagriculture, natural selection, evolution, crop improvement, yield potential,

adaptation, tradeoffs, agricultural ecosystems, agricultural biotechnology,group selection

We have seen that man by selection can certainly produce great results, and can adaptorganic beings to his own uses, through the accumulation of slight but useful variations,

given to him by the hand of Nature. But Natural Selection, we shall hereafter see, is apower incessantly ready for action, and is as immeasurably superior to man’s feeble

efforts, as the works of Nature are to those of Art (Darwin 1859:76).

146 Volume 78THE QUARTERLY REVIEW OF BIOLOGY

abstractProgress in genetic improvement of crop yield potential has slowed since 1985. Simultaneously, more

sustainable management of agricultural ecosystems is needed. A better understanding of natural selec-tion can help solve both problems. We illustrate this point with two specific examples. First, the geneticlegacy of crop plants has been refined by millions of years of natural selection, often driven by compe-tition among plants. We therefore suggest that most simple, tradeoff-free options to increase competi-tiveness (e.g., increased gene expression, or minor modifications of existing plant genes) have alreadybeen tested by natural selection. Further genetic improvement of crop yield potential over the next decadewill mainly involve tradeoffs, either between fitness in past versus present environments, or betweenindividual competitiveness and the collective performance of plant communities. Eventually, we maydevelop the ability to predict the consequences of genetic alterations so radical that they have not yetbeen tested by natural selection. Second, natural selection acts mainly at the level of genes, individuals,and family groups, rather than ecosystems as a whole. Consequently, there is no reason to expect thestructure of natural ecosystems (diversity, spatial, or temporal patterns) to be a reliable blueprint foragricultural ecosystems. Natural ecosystems are nonetheless an important source of information thatcould be used to improve agriculture.

ADECADE AGO, Williams and Nesse(1991) proclaimed “The Dawn of Dar-

winian Medicine” and applied modern evo-lutionary theory to problems ranging fromhuman aging to the evolution of virulence inpathogens. Here we discuss ways in whichagriculture might also benefit from a greaterunderstanding of natural selection. Darwin(1859) used examples from plant and animalbreeding in developing his theory of naturalselection, and there have been some worthyanalyses of agricultural problems from a Dar-winian perspective. Natural selection is oftenignored or misunderstood in agriculturalresearch, however. This is unfortunatebecause in agriculture, as in medicine (Wil-liams and Nesse 1991) and biology generally(Dobzhansky 1973), there is much that makessense only in the light of evolution.

The evolution, in agricultural pests, ofresistance to pesticides and nonchemical con-trol methods has already received muchattention (Barrett 1983; Alstad and Andow1995; Heap 1997) and is not the focus of thisreview. Instead, we consider the implicationsof past natural selection for crop geneticimprovement and for the overall design ofagricultural ecosystems.

We present two main hypotheses. Our firsthypothesis is that natural selection had ampleopportunity, before the wild ancestors of ourcrops were domesticated, to test alternativesolutions to problems that limited individualfitness under preagricultural conditions. Planttraits such as efficient enzymes, competitive-

ness, stress tolerance, and any broad-spectrumdefenses against pests would have consistentlyconferred individual fitness (survival andreproduction relative to alternative genotypes)in preagricultural environments, so furtherimprovement of these traits is likely to be dif-ficult. Instead, opportunities for furthergenetic improvement of crop yield will mainlyinvolve tradeoffs between plant adaptation toagricultural versus natural conditions, orbetween the competitiveness of individualplants and the collective performance of plantcommunities. Breeding crops resistant toevolving pests will continue to be an importantongoing activity, as will breeding to accom-modate changing consumer preferences.

Our second main hypothesis is that naturalselection is the only reliable source ofimprovement (by any definition relevant toagriculture) in natural ecosystems that oper-ates on a time scale longer than the lifetimeof individual plants. Natural selection acts atthe level of genes, individuals, and familygroups, not communities and ecosystems.Therefore, our second main hypothesis isinconsistent with the suggestion (Soule andPiper 1992; Lefroy et al. 1999) that agricul-tural ecosystems whose structure (speciescomposition, spatial and temporal patterns)is based on natural ecosystems will be consis-tently more efficient, sustainable, and pro-ductive. Although this hypothesis rejectsmindless mimicry, natural ecosystems, prop-erly understood, are nonetheless a valuablesource of ideas for agriculture.

June 2003 147DARWINIAN AGRICULTURE

Implicit in these hypotheses is the need todetermine when natural selection has actedin ways congruent with human goals, andwhen it has not. If a plant trait has alreadybeen improved (by human criteria) throughmillions of years of natural selection, theremay be little opportunity for further improve-ment of that trait through either traditionalor molecular plant breeding. But when pastnatural selection and present human goalsconflict, it may be possible to improve onnature. For example, crops may be improvedby deliberately reversing the negative effectsof past natural selection on the collective per-formance of plant communities. Similarly, wemay be able to design agricultural ecosystemsthat maximize resource-use efficiency or con-trol pests in ways not seen in natural ecosys-tems. A consideration of natural selection canexplain past successes and predict whichapproaches are most likely to succeed in thefuture.

Hypothesis 1 and Crop ImprovementOver the last 10,000 years, humans have

made substantial genetic improvements incrops, by human criteria. Modern cucumbers(Cucumis sativus), for example, are less bitterand therefore more palatable to humans(and to some pests), but our emphasis hereis on crop yield. Yield potential—i.e., produc-tion of the harvested product per unit area,with adequate water and nutrients and in theabsence of pests or disease (Evans and Fischer1999)—has increased roughly twofold forwheat (Triticum aestivum), rice (Oryza sativa),and maize (Zea mays) since 1930 ( Jenningsand de Jesus 1968; Ortiz-Monasterio et al.1997; Duvick and Cassman 1999). We suggestthat virtually all of this progress in yieldpotential has, often unknowingly, exploitedtradeoffs between individual plant competi-tiveness versus plant community (i.e., crop)performance, or between fitness in past(including preagricultural) versus presentenvironments.

Our first main hypothesis implies that cropgeneticists are unlikely to improve on naturalselection anytime soon in solving those prob-lems that have limited the fitness of individualplants for millions of years. More recent prob-lems, or cases in which yield or other factors

of value to humans lack a positive correlationwith individual fitness, offer more opportu-nities for improvements. Leading crop phys-iologists have made various statements consis-tent with this hypothesis. Loomis (1993)suggested that “natural selection has alreadyfound efficient solutions to traits such as pho-tosynthesis that lend individuals success incompetition with their usual neighbors” (p584). Similarly, Evans (1993) wrote that pho-tosynthesis “has been subject to intense andprolonged natural selection and plant breed-ers have yet to improve upon it” (p 186).

Although our first hypothesis does notexclude the possibility of substantial increasesin yield potential, we tend to agree with theprediction attributed to Donald Duvick, asenior statesman of plant breeding, that newmolecular methods alone “will not producesharp upward swings in yield potential exceptfor isolated crops in certain situations”(Brown and Kane 1994:25), at least for thenext decade or two. This contrasts with opti-mistic statements from some biochemists(Zelitch 1992), molecular biologists (Ku et al.2000), and biotechnology enthusiasts (Con-way 1998).

Before discussing the evidence for our firsthypothesis, and its implications for cropgenetic improvement in the future, we willconsider recent progress in genetic improve-ment of crop yield potential. We emphasizethe grain crops upon which most humansrely, directly or indirectly, for most of theirdietary protein and energy, although ourbroader hypotheses also apply to horticul-tural crops and domesticated animals.

yield potential of major cropssince 1985

World grain yields per unit area increased26% between 1965 and 1975, and 31%between 1975 and 1985, but only 9% between1985 and 1995 (Brown et al. 1998). Changesin global markets can affect total production,but do not explain the slower growth in yieldper unit area. As a result of continuing popu-lation growth, grain production per personfell more than 10% between 1985 and 1995.Some combination of decreased populationgrowth, more equitable distribution, andincreased crop yields will be needed to


improve food security without additionalexpansion of the land under agricultural pro-duction. Most ecologists are opposed tomajor expansion of agricultural land area,which would likely involve draining wetlandsor clearing tropical forests. What then are theprospects for further yield increases?

Even the slow yield increase since 1985largely reflects more widespread adoption ofthe best available cultivars and other technol-ogies (such as fertilizer) developed prior to1980. Countries such as Korea, which hadalready adopted these technologies by 1980,have achieved little or no increase in yieldssince then (Cassman 1999). Any substantialincrease in world food production will there-fore require increases in crop yield potential,which sets an upper limit on the yields thateven the best farmers can obtain.

Measuring progress in genetic improve-ment of yield potential is not simple. In adirect comparison of crop cultivars developedover a period of decades, IR8, an early GreenRevolution rice cultivar released in 1966,yielded about 7,000 kg/ha, whereas the high-est-yielding cultivar tested, released 30 yearslater, yielded almost 10,000 kg/ha (Peng etal. 1999). Peng et al. noted, however, thatyields of 9,000 to 10,000 kg/ha were reportedfor IR8 in the 1970s. Yield declines for IR8over the last 30 years presumably reflect evo-lution of pest and pathogen populations.These data therefore provide little evidencefor real improvement in yield potential.Hybrid rice may offer some yield advantages,but it is unlikely to repeat the yield advancesof the Green Revolution. On average, hybridrice cultivars may outyield inbred (i.e., non-hybrid) cultivars by 9% or so (Peng et al.1999), but in one two-site comparison, thehybrid cultivar that yielded best in China hadyields identical to a good inbred, when testedin the Phillipines (Ying et al. 1998).

Reviewing similar data and results fromyield contests, Cassman (1999) concludedthat for “tropical rice and temperate maize. . . there has been no detectable increase inyield potential” in recent years. Similarly, ananalysis of yield trends for wheat in Mexicofound an increase of “0.9% y�1 . . . for culti-vars released from 1960–1981 [still less thanpopulation growth], and zero for the period

1981–1988” (Bell et al. 1995:60). Althoughnot all assessments are this pessimistic, it isclear that progress in yield potential duringthe past 20 years was much slower than in theprevious 20. For example, a yield increase of9% from hybrid rice would be less than one-third of the 31% increase achieved withinbred rice during the decade from 1975 and1985 (Brown et al. 1998).

Why has yield potential stagnated? Whatwas the basis of past progress? Whatapproaches offer the greatest hope for fur-ther progress? To answer these questions, weneed to consider the relative merits of humaningenuity and natural selection.

the power and limitationsof natural selection

Natural Selection Vs. Human Ingenuity

The superiority of natural selection tohuman ingenuity can be illustrated by a briefexcursion into engineering. A recent exam-ple is the successful development of enzymesmade from DNA, which may be valuable inmedicine (Santiago et al. 1999). Mostenzymes in nature are made from proteinand a few from RNA, but no one knew howto design an enzyme made from DNA.Instead, Breaker and Joyce (1995) generated�1013 randomly varying DNA molecules anddevised an automatic selection regime (notinvolving evaluation by humans) wherebyonly those DNA sequences with the greatestability to catalyze the target reaction wereretained from one “generation” to the next.Significant enzyme activity was obtained in sixrounds of selection, and a useful enzyme wasproduced after adjustment of the selectionprocess and an additional seven cycles ofautomated selection. An analogous processwas recently used to simulate the evolution ofcrawling robots (Lipson and Pollack 2000).The resulting designs were built, and per-formed as predicted.

The growth of biomimetics (Ball 2001) andevolutionary computation (Harik et al. 1999)reflect increasing recognition by engineers ofthe limitations of human imagination, rela-tive to natural selection and analogous pro-cesses. An optical computer mouse based onvisual information processing in honeybees


(Giles 2001), drag-reducing coatings thatmimic the tiny ribs on shark skin (Ball 1999),fracture-resistant composite materials basedon conch shells (Kamat et al. 2000), andpolarized image systems that work in muddywater (Rowe et al. 1995) are a few recentexamples of reverse-engineering by humansof natural selection’s inventions.

Agricultural Examples of Natural Selection

The rapid evolution of weedy grasses illus-trates the power and rapidity of natural selec-tion in improving the adaptation of plantsthat, like crops, are adapted to agriculturalenvironments. Evolution of herbicide resis-tance, sometimes in as little as four years (Hillet al. 1994), may involve relatively simplegenetic changes, comparable to those beingintroduced into transgenic crops. But otheradaptations are more sophisticated thanthose typically attempted by molecular biol-ogists today.

A prostrate foxtail (Setaria lutescens) ecotypesurvives in frequently harvested alfalfa fieldsby keeping much of its leaf area below thecutting height (Schoner et al. 1978). Barn-yardgrass (Echinochloa crus-galli) has adapted totemperature conditions from Quebec to Mis-sissippi in less than 300 years (Potvin andSimon 1989). Flooding-tolerant watergrass(Echinochloa sp.) and flooding-intolerant barn-yardgrass are believed to have a recent com-mon ancestor. Hand weeding by humans overhundreds of years imposed strong selection onwatergrass—seedlings that can be confusedwith rice are less likely to be killed—but it isstill remarkable that watergrass evolved toresemble traditional rice cultivars in physicalappearance (based on a multivariate analysisof 15 quantitative characters) more than itdoes barnyardgrass (Barrett 1983). Wouldcrop molecular geneticists have known, inadvance, what genes to transfer to achievethese phenotypes? If we are unable to dupli-cate what natural selection has done overdecades or centuries, it may be overoptimisticto assume that we will soon solve problems thatnatural selection has been working on for mil-lions of years. Of course, evolution of weedsalso illustrates the important point that utilityto humans is not a criterion for natural selec-tion.

Natural Selection Does Not Share Our Goals

Given the power of natural selection, itmight seem that we could rely on naturalselection to further improve crop yield poten-tial. This approach would be effective only tothe extent that traits conferring individual fit-ness also increase crop yields.

We suggest that the wild ancestors of cropswere already well adapted, perhaps nearlyoptimal (Williams 1992), with respect to anytraits whose effects on individual fitness werefairly consistent across the range of environ-ments to which they were exposed. We do notnecessarily expect optimality (even in termsof individual fitness) in new or variable envi-ronments, but we do assume “convergence toan optimum or an evolutionarily stable strat-egy” when “the regime of selection acting onthe trait . . . remains invariant” (Eshel andFeldman 2001:186). For example, we assumethat natural selection has always favored moreefficient enzymes, because they would consis-tently have enhanced competitiveness, butthat rooting patterns of some crops may notyet have adapted fully to the higher soil nitro-gen typical of modern agriculture.

Simply allowing natural selection to oper-ate under agricultural conditions (by sowinga random subset of seed harvested the previ-ous year) can improve adaptation to localbiotic, and especially abiotic, conditions. Iflocal adaptation is initially poor, as may be thecase when a cultivar is grown in a new envi-ronment, then natural selection, operatingon a genetically diverse population, canincrease yields significantly over a few gener-ations (Suneson 1956).

But Suneson’s (1956) “evolutionary plantbreeding method,” also popularized by Allard(1988), has never resulted in cultivars thatoutyield the best modern cultivars developedthrough human selection. For example, amixture of barley lines, subject to 45 genera-tions of natural selection under agriculturalconditions, had only 85% the yield of a com-mercial cultivar (Soliman and Allard 1991),even when grown in narrow (four-row) plots.Due to interactions with neighbors, narrowplots tend to overestimate the relative yield ofthe more competitive genotypes favored bynatural selection (Fischer 1978). Gustafsson(1951) summarized studies dating back to


1912 that showed that the genotypes thatincrease in frequency under natural selectionimposed by intraspecific competition are notnecessarily those that maximize yield per unitarea. Earlier, Suneson and Wiebe (1942) sug-gested that natural selection can be useful inimproving adaptation to cold and disease, butpointed out that “in the absence of such fac-tors valuable material is likely to be lost as aresult of competition” (p 1052). Similarly,Jennings and Aquino (1968) identified selec-tion for competitive ability as “a basic causeof limited breeding advance in most tropicalimprovement programs” (p 541). Naturalselection is powerful, but it will not select forcommunity performance (e.g., higher seedor biomass yield per unit land area) when thisconflicts with individual competitiveness.Only humans can do that.

If our first major hypothesis is true, thenopportunities for improving crop yield poten-tial may lie mainly in exploiting tradeoffsrather than simply the “continuation of cropevolution by other means” suggested by Sim-monds (1979:v), at least if the latter impliescontinuation in the same direction. Sim-monds probably did not intend to imply thatplant breeding continues evolution in thesame direction as natural selection, however,as most of his examples of changes thatoccurred during crop domestication, includ-ing reduced height, less branching, elimina-tion of toxins and spines, nonshattering(reduced seed dispersal), and reduced seeddormancy, would have been opposed by nat-ural selection in preagricultural environ-ments.

If natural selection had ample opportunity,before the wild ancestors of our crops weredomesticated, to test alternative solutions toproblems that limited individual fitnessunder preagricultural conditions, then cropgeneticists will be unable to beat naturalselection at its own game anytime soon. Theproblem is not a lack of genetic variability;any large population will vary. The problem,for traits that would enhance both crop yieldand individual plant fitness, is that rare vari-ants remain rare precisely because they areless fit. Natural selection has never had max-imizing the oil or protein content of seed asa goal, so it is not surprising that there has

been prolonged response to directional selec-tion by humans for these traits (Dudley andLambert 1969). Similarly, selection by humanscan alter the balance between vegetative andreproductive growth away from that favored bynatural selection, in order to favor seed pro-duction in grain crops or biomass productionin forage crops or trees grown for biomass.However, as discussed below, many of theimprovements suggested by molecular biolo-gists involve traits that have already beensubject to long-term improvement by naturalselection and are currently under stabilizingselection.

Even with molecular tools, “magic isunlikely” (Simmonds 1998:139). However,that still leaves the following nonmagical ave-nues for genetic improvement of crop yield:

1) Red Queen breeding (keeping ahead inevolutionary arms races with pests).

2) Accelerating adaptation to new physical orchemical conditions (e.g., different climatesor greater soil fertility).

3) Reversing the effects of past natural selec-tion for traits that enhance individual com-petitiveness but limit community-level perfor-mance, either through deliberate selection ofless competitive plants with specific traits orthrough human-mediated group selection.

4) Eventually, perhaps, designing truly novelphenotypes not previously tested by naturalselection.

These approaches will be discussed, inorder. But first, we explain why, even with newmolecular tools, the problems that naturalselection has been unable to solve over mil-lennia will not easily be solved by humans.

“magic is unlikely”With molecular methods, it is now feasible

to make minor genetic changes in specificgenes of many crop plants. However, any newphenotype that could be achieved throughminor genetic changes—usually there will bemany routes to the same phenotype—is likelyto have been tested by natural selectionrepeatedly in the past. Therefore, any minorchanges that would consistently haveimproved individual competitiveness in pre-


agricultural environments (e.g., tradeoff-freeimprovements in photosynthesis) will alreadyhave been incorporated by natural selection.Minor changes will therefore contribute toyield potential only in cases involving trade-offs between past natural selection and pres-ent human goals, as discussed below.

Major genetic changes may not have beentested by natural selection (at least in ances-tors of a particular crop), so we cannotexclude the possibility that some majorchanges would increase yield potential. Mak-ing major changes at random will almostalways decrease yield potential, however. Pre-dicting the effects of any specific majorgenetic change is far beyond our current abil-ities, so we do not expect major geneticchanges to contribute to crop yield potentialanytime soon. A standard for distinguishingbetween minor and major changes is pro-posed below.

Minor Changes

So long as we are limited to minor varia-tions on existing genes, we are tinkering, notengineering. By “tinkering” we mean suchminor changes as: 1) modifications of theactive site of key enzymes, hoping to improveefficiency; 2) overexpressing one or moreenzymes believed to limit crop yield potential;or 3) converting an existing inducible geneto constituitive expression.

Trial-and-error tinkering is also the main-stay of natural selection ( Jacob 1977), butnatural selection has already tested incom-parably more variants than any conceivableresearch program. “How fleeting are thewishes and efforts of man!” wrote Darwin(1859), “[H]ow short his time! and conse-quently how poor will be his results, com-pared with those accumulated by Nature dur-ing whole geological periods!” (p 94).

Major Changes

It is at least conceivable that there could bea radically different genotype, producedthrough major genetic modification, thatwould outperform any genotype that existstoday. If every evolutionary path to this hypo-thetical genotype would require many succes-sive mutations, some of which would signifi-cantly reduce fitness, then the improved

genotype might never evolve through naturalselection, despite its ultimate superiority.

How novel does a genotype need to bebefore we can reasonably entertain the pos-sibility that it, or other genotypes giving simi-lar phenotypes, have not already been testedby natural selection? The evolution of C4photosynthesis from conventional C3 photo-synthesis may be a useful standard againstwhich the complexity of other geneticchanges may be compared.

C4 photosynthesis is a complex adaptation.C4 plants have evolved sophisticated CO2-concentrating mechanisms that can increasephotosynthesis and essentially eliminatelosses to photorespiration, which significantlyreduces net photosynthesis in C3 plants. InC4 plants, the initial uptake of CO2 is by phos-phoenolpyruvate carboxylase rather thanrubisco, the usual CO2-fixing enzyme. Theinitial product is a four-carbon molecule(hence the name “C4”) such as malate. Mal-ate is transported to the interior of special-ized structures known as bundle sheaths,where it is decarboxylated to release CO2.Suberin layers prevent the CO2 from leakingout again. Rubisco is concentrated inside thebundle sheaths, where elevated CO2 concen-trations increase its efficiency.

If C4 photosynthesis had only evolved onceor twice, then we could speculate that it wasbut one of a large number of possibleimprovements on conventional C3 photosyn-thesis. In that case, it could be reasonable tolook for an even better solution. But C4 pho-tosynthesis, with minor variations, has evolved(and persisted) at least 31 separate times(Kellogg 1999). This is a conservative esti-mate, since it is not known how many times aC4 variant arose, only to die out in competi-tion with C3 conspecifics. For example, C4photosynthesis may be less beneficial underflooded conditions (Sage 2000). The fact thatnatural selection keeps finding variations onthis same solution suggests that any bettersolution is even more different from C3 andC4 photosynthesis than they are from eachother. If so, then any change in a plant’sphotosynthetic system less radical than C4photosynthesis is unlikely to improve yieldpotential.

The repeated evolution of C4 photosynthe-


sis “provides very strong support for the argu-ment that evolution of Rubisco has reached abarrier” (Long 1998). Long went on to argue,however, that the rubisco found in Rhodo-phyta is significantly less prone to photores-piration than that found in angiosperms.Given that the homology between these tworubiscos is only about 50%, it is conceivablethat Rhodophyta rubisco is a solution that wasnever tested by natural selection in the imme-diate ancestors of crops. We suggest, however,that even such a major change to rubiscowould evolve more easily (in terms of the num-ber of intermediate genotypes and their rela-tive fitness) than would the entire C4 system.Furthermore, there appear to be significanttradeoffs between the specificity of rubisco forCO2 (reducing losses to photorespiration) andits specific activity (see Table 1 of Uemura etal. 1997). We therefore doubt that Rhodo-phyta rubisco represents a real opportunity toincrease crop photosynthesis, but agree that itmight be worth trying. On the other hand, wethink it is safe to assume that any proposedchange much less radical than C4 photosyn-thesis or Rhodophyta rubisco is likely alreadyto have been tested and rejected by naturalselection.

Designing Truly Novel Genotypes

Designing truly novel genotypes is beyondour current abilities. Berry (1975:649) notedthat the repeated evolution of C4 photosyn-thesis and the difficulty of breeding new C4plants in the lab “illustrates the massive dif-ference in scale between the processes ofnature and laboratory efforts at geneticmanipulation” (p 649). Given the biochemi-cal and anatomical complexity of this adap-tation, converting a C3 crop to a C4 would bedifficult or impossible with current moleculartechnology. Similarly, actually designing animproved rubisco (as opposed to just tinker-ing with it) would be impossible today.

So far, we cannot even predict the tertiarystructure of an enzyme, much less its activity,from its sequence, except by comparison withsimilar enzymes. Most of the antibiotics weuse are just copies, often with modifications,of those fungi and bacteria use to kill eachother. Designing a new photosynthetic enzymethat would outperform those in existing plants

would be much more difficult than designinga new chemical to interfere with bacterial cellwall synthesis.

True genetic engineering will also requirean ability to predict the overall phenotypes ofgenotypes substantially different from anythat have previously existed. Random geneticchanges, whether the result of mutation orgenetic manipulation by humans, will rarelybe beneficial. Predicting which major geneticchanges will increase crop yield potential iswell beyond our current capabilities.Although there have been post hoc attemptsto explain the near-normal phenotype ofgenetically modified mice without myoglobin(Garry et al. 1998; Godecke et al. 1999),nobody would have predicted this result inadvance. Predicting the effects of geneticmodification on crop yield potential will beequally difficult.

We doubt whether today’s molecular biol-ogists could even duplicate, by gene transfer,what has already been achieved, throughselection, by traditional plant breeders.Could they, for example, convert wild kale(Brassica sp.) into cauliflower without usinggenes from cauliflower?

Can Rice Photosynthesis Be Improved?

As an example of the limitations of humantinkering, consider current attempts todevelop rice with the photosynthetic poten-tial of C4 crops. Ku et al. (2000) transferredthe gene for phosphoenol pyruvate carbox-ylase (PEPC) from maize to rice, hoping toachieve some of the ability, in C4 maize, toconcentrate carbon dioxide and therebyincrease photosynthesis. Tests found, how-ever, that “photosynthetic CO2 assimilation oftransgenic rice decreased with increasingPEPC activity” (Matsuoka et al. 2000:173),consistent with our hypothesis that photosyn-thetic traits are under stabilizing selection. Kuet al. nonetheless claimed a significantincrease in rice yield, which they attributed,post hoc, to greater stomatal conductance.

Increasing stomatal conductance has infact been an important contributor toincreased yield potential in irrigated wheat(Amani et al. 1996) and cotton (Cornish etal. 1991). Greater stomatal conductanceincreases photosynthesis, but also transpira-


tional water loss from leaves, a tradeoff whichmay be beneficial in irrigated agriculture butnot in dry, unpredictable environments. Ricehas been grown under irrigation for thou-sands of years, however. Greater stomatal con-ductance is a trait that could presumably beachieved through many different simplegenetic changes that would increase eitherstomatal opening or number. It thereforeseems safe to assume that mutants withgreater stomatal conductance have occurredrepeatedly among the ancestors of present-day rice cultivars. Presumably, natural selec-tion has already tested a wide range of sto-matal strategies, i.e., genetically determinedrules that adjust stomatal number and open-ing appropriately for the range of conditionsto which rice was exposed. We consider itunlikely that a single gene that causes a con-stituitive increase in stomatal conductancewould improve on these strategies.

We are therefore skeptical of claims thatrice that contains a single maize gene has 30–35% higher yield (Ku et al. 2000). We predictthat independent tests in several countries, inplots large enough to avoid edge effects,would fail to confirm that this transgenic ricesignificantly outyields the best locally adaptedcultivars.

Eventually, it might be possible to transferthe entire C4 system to rice, as proposed bySheehy (2000) and others. By the standard ofthe previous section, this might meet the min-imum requirements for a novel phenotypenot previously tested by natural selection.Given the frequency with which C4 photosyn-thesis has evolved, it is not clear whether C4rice is an untested variant, or one that hasalready been tested and rejected by naturalselection. Even assuming that C4 rice wouldhave higher yield, developing it would be along-term enterprise, beyond our current sci-entific and technological abilities. Mean-while, further increases in atmospheric CO2

during the course of this project would tendto reduce the photosynthetic advantage of C4rice relative to C3 rice (Sage 2000).

what has workedIf natural selection had millions of years,

prior to crop domestication, to improve planttraits that enhance individual fitness, how can

we explain the dramatic progress made byplant breeders? What additional opportuni-ties remain?

Past progress in genetic improvement ofcrops has usually either sacrificed individualplant fitness in the pursuit of human goals(e.g., better flavor or higher yield per unitarea), or dealt with traits whose fitness con-sequences under present agricultural con-ditions are different from those in the past.For example, natural selection will haveimproved some aspects of pest resistance,but will not have generated defenses specif-ically targeted against pests to which theancestors of a crop have not been exposed.

Red Queen Breeding

The arms race (Dawkins and Krebs 1979)between plants and pests predates agricul-ture. Effective pest control is not a substitutefor increases in yield potential, but it is essen-tial to achieving that yield potential. Plantbreeders make major contributions to yieldby developing crops resistant to current pests.As pests continue to evolve and new pestsarrive from other parts of the world, breedersrace to develop cultivars resistant to the newpests. Lewis Carroll’s Red Queen describedthe dilemma perfectly: we must run as fast aswe can to stay in one place.

Because the pests that attack today’s cropsmay be different from those that attackedtheir ancestors, genetic tinkering may some-times provide short-term protection againstrecently arrived or rapidly evolving pests. Forexample, constituitive expression of existingdefenses (Broglie et al. 1991; Cao et al. 1998)or gene overexpression (Liu et al. 1994; Caoet al. 1998) can sometimes increase disease orpest resistance.

But if constituitive expression of defensessometimes increases pest resistance, then whyare inducible defenses so common in nature?Many different simple mutations could con-vert a plant defense gene from inducible toconstituitive expression. Mutant plants withconstituitive defenses must have arisen fre-quently and competed against those withinducible defenses. Therefore, the preva-lence of inducible genes suggests that theywere valuable in the past, at least in terms ofindividual fitness. In addition to any meta-


TABLE 1Transgenic crop petitions approved by USDA orwithdrawn by applicant, as of 30 August 2002

Trait Approved Withdrawn

Herbicide tolerance 24 6Insect or virus resistance 21 12Fruit ripening or seed composition 11 3Male sterility 5 1

Some applications withdrawn may subsequently have beenresubmitted and approved. No applications were denied. Datafrom http://www.aphis.usda.gov/ppq/biotech/petday.html.

bolic costs of producing defensive chemicals,the benefits of inducible defenses mayinclude reduced chemical attraction of spe-cialist herbivores, spatial variability thatreduces herbivore performance or increasestheir rate of movement (thereby attractingpredators), slower evolution of adaptation byherbivores, and decreased autotoxicity (Agra-wal and Karban 1999). The question iswhether these benefits are as important toagricultural productivity today as they were toindividual fitness in the past. If pests are pre-dictably more common now, then maybe con-stituitive defenses will be useful. But then weshould at least ask why these pests havebecome more common.

In a narrow sense, transgenic crops thatproduce insecticidal toxins of bacterial origincould be considered an innovation not pre-viously tested by natural selection. However,insects have a long evolutionary history ofexposure to a wide variety of plant toxins, andminor evolutionary changes in pests may besufficient to confer resistance to these “new”toxins. Even limited use of bacterial toxins assprayed insecticides has led to the evolution ofresistance in some insects (Tabashnik et al.1997). There is no reason to expect transgenicsolutions to be any more durable, as a class,than those generated by traditional plantbreeding, although resistance-managementstrategies may delay the inevitable (Alstad andAndow 1995). Some transgenes may lack valu-able control sequences (introduced by tradi-tional breeding) that trigger defenses onlywhen their benefit to the plant exceeds theircost. Traditional plant breeding may alsotransfer several defense genes that functionsynergistically, which could extend the usefullifetime of the defense genes.

Most of the transgenic crops approved bythe U.S. Department of Agriculture throughAugust 2002 fall under the general categoryof Red Queen breeding. Of 61 cases, 45involved crop tolerance to herbicides, orresistance to insects or viruses (Table 1). Theremaining cases involved fruit ripening, seedcomposition, or male sterility—traits whichmay be useful to humans and which wouldnot have been favored by natural selection.None of the traits would increase yield poten-tial or tolerance of abiotic stress.

We may not be as good at evaluating long-term risks of biotechnology as we are at seeingshort-term benefits (Snow and Palma 1997).If widespread use of transgenic crops leads tomore consistent selection for resistant pests,or if pollen from crops transfers fitness-enhancing genes to weedy relatives, they mayworsen pest problems in the long run. Molec-ular biology may speed the search for newpest control solutions each time an old onefails, but it is important not to confuse thisshort-term benefit with either permanentpest-control solutions (possibly an insur-mountable challenge) or long-term improve-ments in yield potential. Thousands of yearsfrom now, dwarf wheat will still have higheryield potential than taller cultivars, whereasongoing evolution of weeds and insect pestsis likely to diminish the value of today’sherbicide-tolerant and insect-resistant cropswithin a decade.

Changing Conditions Change Tradeoffs

During the millions of years prior todomestication, the ancestors of our cropplants were exposed, frequently or occasion-ally, to drought and nutrient-poor soils. Theyevolved numerous sophisticated adaptationsthat conferred resistance or tolerance tothose stresses. As a result, it may be difficultto improve on natural selection with respectto the ability to survive and reproduce whensoil water and nutrients are limiting. Oppor-tunities may lie in the opposite direction,namely in developing crops that take advan-tage of agricultural improvements in waterand nutrient supply. Accelerating adaptationto increasing atmospheric CO2 might offersimilar opportunities.


Prior to modern agriculture, uptake of soilnitrogen by the plant community often leftlittle nitrogen in the soil by the time it wasneeded to make protein in seeds. Rubiscoand other proteins in extra leaves (i.e., moreleaves than needed to intercept the availablesolar radiation) may serve as reservoirs fornitrogen, allowing plants to take up nitrogenwhen it is available and then transfer it toseeds after soil nitrogen becomes scarce. Itmay make sense to breed cultivars with fewerleaves, but only if they also have sufficientnitrogen uptake capacity late in the season. Areliable and economic way to supply nitrogenduring grainfill (e.g., from aircraft, or in irri-gation water) would also be needed.Although crops that can use additional soilnitrogen, phosphorus, or water when avail-able may be a key to increasing yield poten-tial, crops that perform poorly without addi-tional inputs could increase dependency onresources whose future on-farm availability(at reasonable cost) may be uncertain insome regions.

Reversing Past Selection for Competitiveness

So far, the most successful approach toincreasing crop yield potential has been to dosomething that natural selection never does,namely, to select for characteristics that ben-efit a plant community at the expense of indi-vidual competitiveness. This may also be trueof other measures of community perfor-mance, such as resource-use efficiency andinterannual stability. When community per-formance and individual competitivenessconflict, natural selection favors competitive-ness (Harper 1977). To a give a specific exam-ple relevant to agriculture, Zhang et al.(1999) have considered the conflict betweenindividual plant and the plant communitywith respect to resource allocation to rootsunder water-limited conditions. The produc-tivity of a community of identical plants wasshown to be greatest with a moderate invest-ment in root biomass: just enough root tocapture the available water. But a populationwith this optimum root allocation patterncould be invaded by a genotype that allocatedmore resources to roots, thereby capturing alarger share of the limited water supply. Theyconcluded that “optimal resource partition

maximizing a crop’s yield is never evolution-arily stable” (Zhang et al. 1999). Similarly,Schieving and Poorter (1999) showed thatthe leaf morphology that maximizes totalphotosynthesis by a plant community doesnot maximize individual competitiveness.

Modern wheat and rice cultivars owe theirhigh yield potential largely to their short stat-ure, relative to traditional cultivars. Short stat-ure reduces lodging (falling over) andincreases growth of grain at the expense ofstems, but selection for these traits hasresulted in decreased individual competitive-ness. Forced to compete with traditional, low-yielding cultivars, modern high-yield rice cul-tivars completely disappear within as few asthree generations ( Jennings and de Jesus1968).

How can we reverse past natural selectionfor individual competitiveness and select forthe collective performance of the crop com-munity instead? Donald (1968) proposedselection based on an “ideotype”: a pheno-type designed to maximize community per-formance, often at the expense of individualcompetitiveness. His specific suggestions,when soil nutrients and water are not limit-ing, included short stature, less branching,erect leaves, earlier flowering, and fewerleaves. Less branching and more erect leavesboth require greater densities (plants/hec-tare) to produce increased yields. A high den-sity of nonbranching plants will achieve aclosed canopy (i.e., one that intercepts 95%of the available sunlight) earlier than a lowdensity of branching plants, thereby increas-ing seasonal photosynthesis. Nonbranchingplants waste less energy contesting space withtheir neighbors. Crop plants with more erectleaves use the light they intercept more effi-ciently, due to the nonlinear response of pho-tosynthesis to light. Earlier flowering andfewer leaves are both associated with an ear-lier switch from vegetative to reproductivegrowth. Once a crop has enough leaves tointercept the available sunlight, any furtherleaf growth contributes little to canopy pho-tosynthesis (but may store nitrogen), andcomes at the expense of grain growth.

Donald’s specific suggestions regardingtraits to enhance community performanceare generally consistent with the available


data. For example, Austin et al. (1980) com-pared wheat cultivars released between 1900and 1980, and found a negative correlationbetween yield and both height and branch-ing. Experimental manipulations of leafangle and branching in soybean (Glycine max)also gave results consistent with Donald’shypotheses (Kokubun 1988). Higher-yieldingcotton (Gossypium barbadense) cultivars haveless leaf area per plant (Lu et al. 1994).Increases in crop yield potential haveinvolved, and will continue to involve, trade-offs between individual fitness and commu-nity performance, and selection (deliberateor incidental) favoring the latter.

It is possible, though far from certain, thatthere will be synergistic interactions amongthe approaches discussed. If the biotech-nology and chemical industries could keepfinding new combinations of herbicides andherbicide-tolerance genes fast enough to stayahead of evolving weed populations, thatmight allow crop geneticists to exploit trade-offs between crop competitiveness and yieldto a greater extent than would otherwise bepossible. This approach would entail somerisks, as the consequences of faster-than-expected evolution of herbicide resistance orunexpected loss of herbicides (e.g., due toincreasing regulation or due to disruption oftransportation infrastructure by natural disas-ter or war) would be more serious. These risksmight be considered acceptable in somecountries, if major increases in yield potentialcould be achieved through moderate reduc-tions in competitiveness, but not if increasesin yield potential are relatively small.

what else might workPractical Group Selection

There are two general approaches to fur-ther exploitation of the tradeoffs betweenindividual competitiveness and the perfor-mance of the crop community. We couldextend Donald’s (1968) ideotype approach,applying human intelligence (perhaps aidedby computers), to the design of high-yieldingideotypes. It may also be possible to developmechanized field protocols that select for com-munity performance. These two approachesare not necessarily mutually exclusive.

As an example of the first approach, con-sider the optimum timing of spring growth inperennial crops, which involves multipletradeoffs. Growth earlier in the spring offersthe potential reward of greater seasonal pho-tosynthesis, but increased risk of losing leavesto late frost. Natural selection may have over-weighted an additional risk, however, namelythe risk of being shaded by a neighbor thatstarted growing earlier. In a given environ-ment, therefore, the optimum starting timefor spring growth of the crop community maybe somewhat later than that favored by indi-vidual selection. Computer models withenough physiological realism and detail tosimulate tradeoffs mechanistically (Cock etal. 1979) might be useful in identifying addi-tional tradeoffs that could form the basis ofideotypes (Loomis 1979).

The second approach would impose selec-tion for community performance, withoutadvance knowledge of the plant traitsinvolved. Harper (1977) suggested that“group selection which is believed to beextremely rare or absent in nature . . . may bethe most proper type of selection for improv-ing the productivity of crop and forest plants”(p 892). The conditions under which naturalselection can favor traits that enhance groupperformance, when it conflicts with individ-ual competitiveness, are quite restrictive, asdiscussed below. But humans can, and have,selected for crop yield and other group-leveltraits. For example, six generations of selec-tion among groups of hens (Gallus domesticus)based on collective egg production led todecreased aggression among hens, decreasedmortality, and increased production of eggs(Muir 1996).

It is possible that large-scale screening forcrop yield would identify genotypes whoseyield advantage comes from traits not yetunderstood. Machines that measure yield ateach location in a field during harvest are nowcommercially available for several crops. Sucha harvester could be modified to segregateharvested seed from the highest-yielding loca-tions within a field into a container separatefrom the bulk of the harvested grain. If onlyseeds from high-yielding locations werereplanted in successive years, that could selectfor higher yields, even at the expense of indi-


vidual competitiveness. Several problemswould need to be solved before this approachwould be ready for testing, however. Yield dif-ferences across a field could result from dif-ferences in soil properties unrelated to cropgenotype. Uniform fields would help, but itmight also be helpful to compare yields to acomputerized yield map from a previous year,when a genetically uniform crop had beengrown. The field would also need to beplanted in such a way that genotype variedwith location in the field; simply planting awell-mixed batch of genotypes would notwork well. Ideally, the field would be a mosaic,with each patch containing progeny of a sin-gle plant. The scale of yield measurement ofcommercial harvesting equipment is toocoarse to detect such small patches, but simi-lar capability could be added to small-plotharvesters.

Given these challenges (and others notidentified), automating group selection foryield would be an expensive project, but notcompared to major DNA sequencing projectsnow underway. A few such tests might suffice,either to identify valuable new ideotypic traits(by looking for consistencies among thehighest yielding lines), or to show that wehave already identified the most importantyield-enhancing traits among available plantmaterials. Automated screening for othergroup-level traits, such as more efficient wateruse, might be more difficult. Methods basedon remote sensing of leaf temperature as anindication of water use can be automated tosome extent, however (Amani et al. 1996).

the future of novel genotypesWe are defining truly novel genotypes as

those that are so different from any previouslytested by natural selection that we cannot pre-dict their performance. In particular, we can-not exclude the possibility that they wouldoutperform existing genotypes in ways thatmight contribute to crop yield potential. Wesuggest, however, that novel genotypes suchas those discussed in the following para-graphs are unlikely to be achieved within thenext two decades. It is nevertheless interest-ing to consider some future possibilities.

Because there has never been a plant thatcould assimilate N2 directly from the atmo-

sphere (without microbial symbionts), wecannot conclude that this phenotype wastested and rejected by natural selection. It istherefore at least plausible that nitrogen-fixing rice or wheat would have significantlyhigher yield, at least under some conditions.The evolutionary persistence of symbioticcooperation in legumes and some otherplants suggests that the photosynthate cost ofsupporting symbiotic rhizobia was often agood investment for plants, at least underpreagricultural conditions, when soil fertilitywas usually low. Nonsymbiotic nitrogen fixa-tion by a plant without microbial symbiontsmight be an even better investment, becauseconflicts of interest among unrelated rhizobiainfecting the same plant may limit the effi-ciency of nitrogen fixation (Denison 2000).The development of crops that fix N2 withouta symbiotic partner would not be easy, how-ever. It is not just a matter of transferring thekey bacterial enzyme, nitrogenase, to a cropplant. Nitrogenase is quite sensitive to O2, butthe ATP and reductant needed to drive N2

fixation come from respiration, whichrequires O2. Legumes have sophisticatedmechanisms, including a variable-permeabil-ity gas diffusion barrier (Denison 1998) andleghemoglobin-facilitated diffusion of O2

within infected cells, that regulate O2 supplyin their nodules. Duplicating this system innonlegumes is beyond our current abilities.Recently, however, an oxygen-insensitivenitrogenase has been reported (Ribbe et al.1997). If confirmed, this would solve some ofthe technical problems that limit the expan-sion of N2 fixation to new crops. Even so,there will be other technical challenges, andwe consider it unlikely that agronomicallyuseful nitrogen-fixing rice or wheat could bedeveloped within the next decade or two.

Other examples of truly novel genotypeswould include crops that, like fungi, canextract energy and nutrients directly from soilorganic matter. Or perhaps new crops couldbe developed that, like some Acacias, houseand feed ants in exchange for protectionfrom herbivorous insects and other pests.Transgenic crops that produce drugs, includ-ing illegal ones, could be highly profitable.All of these examples would presumablyinvolve new tradeoffs that might limit yield


potential, or could create serious social orenvironmental problems beyond the scope ofthis paper. We mention them only as exam-ples of phenotypes novel enough that theymay never have been tested by natural selec-tion.

Hypothesis 2 and AgroecosystemDesign

The idea that agricultural ecosystemswould be more productive or sustainable ifthey were more similar to natural ecosystemsis widespread. One influential paper ( Jack-son and Piper 1989) suggested we should“model agroecosystems on nature’s stan-dards,” basing agriculture on natural com-munities “beyond complete human compre-hension” (pp 1591–1592). Similarly, it hasbeen suggested that agricultural pest manage-ment strategies should be modeled on natu-ral ecosystems (Altieri 1987). These ideas stillpersist today, as evidenced by a recent booktitled, Agriculture as a Mimic of Natural Ecosys-tems (Lefroy et al. 1999).

should agriculture mimic nature?The argument for mimicking natural eco-

systems is unclear. Natural ecosystems mayhave “withstood the test of evolutionary time”(Ewel 1999), but the test criteria were notstated explicitly. No natural terrestrial ecosys-tem has exported as much human food perhectare, over millennia, as rice fields in Asia.Some agricultural ecologists appear toassume that natural ecosystems somehowhave been optimized, perhaps by naturalselection. Natural ecosystems have beendescribed as “stable, self-sustaining . . .” per-haps due to “[s]election through time . . . ofthe most complex structure biologically pos-sible within the limits set by the environment”(Gliessman 1990:5,7). Others apparentlyassume that natural selection operates, if notfor the benefit of an entire ecosystem, at leastfor the benefit of a species as a whole. Theonly reference to natural selection in theindex of a recent book on Ecology in Agriculturestates that while natural selection may conflictwith the goals of plant breeders, it “operatesto continue the survival of species” ( Jacksonand Koch 1997:3). Similarly, it has been sug-

gested that the apparent stability of many nat-ural ecosystems is due to each species having“evolved so that it does not overexploit itsresources” (Gutierrez and Daxl 1984:200).Some of these statements may not fully reflecttheir authors’ current views, but they are rep-resentative of the implicit assumptions madeby many who have attempted to apply ecolog-ical principles to agriculture. Most evolution-ary biologists, however, would reject theseassumptions as inconsistent with modern evo-lutionary theory.

Our second major hypothesis is that natu-ral selection is the only process that consis-tently improves the performance of anythingin nature. If this hypothesis is true, then“nature’s wisdom” should be manifest only inthose entities that have been subject to natu-ral selection. Individual plants and animalscontain genes that have repeatedly passedthrough the screen of natural selection, buthas anything analogous to natural selectionalso improved the structure and function ofthe ecosystems where those species live?

natural selection and ecosystemstructure

Natural selection acts to maximize trans-mission of some alleles relative to alternativealleles (Dawkins 1976), usually by maximizingthe reproductive success of individuals car-rying those alleles. Some plant genes, such asmaternally inherited genes that increase seedproduction at the expense of pollen produc-tion, enhance their own transmission at theexpense of the plant’s overall fitness (Dom-inguez 1995).

Classic group selection, the expectationthat individuals would often sacrifice theirown reproduction in the interests of the spe-cies as a whole (Wynne-Edwards 1962), wasdiscredited in the 1960s (Hamilton 1964;Maynard Smith 1964; Williams 1966). But, asGoodman (1975:261) predicted of the diver-sity-stability hypothesis, this form of groupselectionism has persisted “as an element offolk-science,” among some agricultural ecol-ogists. Kin selection in animals can favorcooperation or even self-sacrifice to helpclose relatives. Sterile worker bees are a dra-matic example, although egg-laying workerbees can sometimes exploit this system (Mar-


tin et al. 2002). There is no clear evidence forkin selection in plants (Cheplick 1992) andits importance in general has been overesti-mated (Griffin and West 2002).

The conditions under which natural selec-tion will favor an individually costly trait (e.g.,restraint in exploiting resources) that bene-fits a group of unrelated individuals are quiterestrictive (Dawkins 1976). Individually costlytraits that increase the ability of a group to“reproduce” (generate another group) maybe favored by differences in the survival ofgroups, but only if each group contains fewerthan about 25 individuals, and only if an aver-age of fewer than one individual changesgroups during the time it takes a group tospawn another group (Table 2). Theseextremely restrictive conditions can be metfor group selection imposed by humans. Forexample, Muir (1996) used groups of fourhens, and prevented movement amonggroups. But these conditions are rarely metin nature.

The agricultural ecologist de Wit (1978)pointed out that “there is nothing in the pro-cess of evolution that has any aspect of com-munity behaviour as a goal” (p 405). Even thefew evolutionary biologists who think naturalselection may operate on groups of plantsgenerally agree that natural selection isunlikely to consistently improve ecosystemfunction, when the collective interests of spe-cies in the ecosystem are in conflict with indi-vidual competitiveness (Goodnight 1990).

Conflicts between individual competitive-ness and collective benefits are also apparentin the evolution of pathogenicity in parasites.

It may be true that an obligate fungal parasitethat produces very large numbers of spores atthe expense of a host plant is likely to goextinct because “in the long run, this wouldtend to destroy the host” (Browning1981:166). However, the claim that therefore“pathogens have responded by evolving ratesof sporulation” low enough to eliminate the“risk of domination” contains two serious fal-lacies. First, natural selection is blind to long-term consequences. Like a river, which rarelyfollows the most direct route to the nearestocean, natural selection responds only to cur-rent conditions. Second, natural selection isdriven mainly by competition (often indirect)within a species. If two fungal genotypes differonly in the number of spores they produce, aconsistently more prolific strain will almostalways outcompete and displace the otherstrain, whatever the long-term consequencesfor the species as a whole. If each host plantis infected by a single clone of parasitic fungi,there may be selection for keeping individualhosts alive longer, but this is simply within-host kin selection, unrelated to the long-terminterests of either species. There is clearempirical evidence that parasites maximizetheir own transmission and not the survival oftheir hosts (Herre 1993). Selection amongindividual plants may indirectly improvesome aspects of ecosystem function—a forestmainly containing insect-resistant trees will beless subject to massive defoliation—but thereis no reason to expect the overall structure ofeven ancient natural ecosystems to be optimalunless they have been improved by some pro-cess other than natural selection.

TABLE 2Conclusions of key papers that examined the conditions under which natural selection will favor the interests

of a group over those of those of the individuals that compose that group

Author Conditions For Group Selection to Work

Levin and Kilmer 1974 Groups consist of fewer than 25 individuals, and usually closer to 10. Rate of gene flow bymigration is lower than 5% per generation (i.e., one individual).

Maynard Smith 1976 The number of successful migrants produced per group is one or less. Successfulmigrants are individuals who leave a group and leave descendents in another.

Leigh 1983 Assuming N groups which each contain n individuals and last for L generations: (a) eachgroup is founded by a single individual; (b) the groups never exchange migrants; (c) N� n.

All results assume unrelated individuals within groups (i.e., kin selection is not at work).


have other processes improvedecosystems?

Are there processes other than naturalselection that consistently improve naturalecosystems in ways that would make themappropriate models for agriculture? We thinknot, but this question may merit additionalresearch.

Succession

Succession may improve some aspects ofecosystem function but make others worse,by human criteria. Like most modern ecol-ogists, we reject old analogies (Odum 1969)between ecological succession and individualdevelopment. Communities and ecosystemsare not superorganisms (McIntosh 1998).Developmental changes of individual organ-isms are part of the organism’s phenotype,controlled by genes subject to natural selec-tion. There is nothing analogous in naturethat operates over millennia to improve theprocess of ecological succession.

Successional changes in species composi-tion of plant communities may improve someaspects of ecosystem function, by human cri-teria, but other aspects typically deteriorate.For example, productivity, nutrient retention(as opposed to loss to rivers), and greenhousegas absorption by forests decline as successionproceeds (Gorham et al. 1979; Robertson etal. 2000). A comparison among islands differ-ing mainly in successional stage found thatmicrobial biomass and microbial activity, twoparameters well-loved by some ecologists,were both lower later in succession (Wardleet al. 1997).

Some ecologists might find changes duringsuccession counterintuitive. Succession oftenleads to increased dominance by a small num-ber of species (Paine 1966)—the Sequoia-dominated oldgrowth forests of the northernCalifornia coast are a dramatic example—butwe would hesitate to conclude from this thatagriculture would always benefit from areduction in crop diversity.

The longer an ecosystem has been in exis-tence, the greater the opportunity for coevo-lution. Mutualisms requiring complex coor-dinated adaptations might therefore be morecommon in ancient natural ecosystems thanin agroecosystems of more recent origin. But

greater ecosystem duration also providesenough time for more sophisticated negativeinteractions to evolve, such as parasitism or“cheating” by symbiotic partners (Letour-neau 1990).

Complexity

We have seen no convincing theoretical orempirical evidence that complex entities notsubject to natural selection spontaneouslyorganize themselves in ways that are optimumin any reasonable sense of the word. Ourexperience with human organizations andwith computer operating systems suggeststhat complexity does not necessarily enhanceperformance.

Ecosystems, like individual organisms, mayhave homeostatic mechanisms (Lewis et al.1997). However, organism-level homeostaticmechanisms have been fine-tuned by selec-tion among organisms, whereas ecosystem-level homeostatic mechanisms have not beenfine-tuned by natural selection among ecosys-tems. Certain kinds of trophic interactionsmay promote stability (McCann et al. 1998),but it is not clear why natural ecosystemswould be especially likely to have just thosesorts of interactions. Assuming that someoptimum number of species will maximizecommunity productivity or stability in a givenenvironment, is there some mechanism thatadjusts the actual number of species in anatural ecosystem upwards or downwardstowards that optimum? We know of no dataor convincing (i.e., not group selectionist)theoretical argument showing that this is, orshould be, generally the case.

Comparing Natural andAgricultural Systems

“[W]e have no idea how evolution adaptsecosystems,” wrote Leigh (2001), since theyare “not units of selection.” Leigh argued,however, that the productivity of natural eco-systems is usually affected adversely by humandisturbance, and therefore that undisturbedecosystems must somehow be structured inways that enhance productivity. In principle,we accept the argument that if human-induced changes in ecosystem structure usu-ally have an adverse effect on specific ecosys-tem functions, that would provide strong


evidence that the structure of undisturbednatural ecosystems is well-suited to thosefunctions. But do undisturbed natural ecosys-tems actually excel in those functions relevantto agriculture, such as the ability to exportfood sustainably? The greater year-to-year sta-bility of nut production by almond orchards(7,500 to 12,800 kg/ha) relative to oak forests(0.4 to 55 kg/ha) seems inconsistent with thishypothesis (Feldhamer et al. 1989; Hutmacheret al. 1994). In one study, total seed yield ofnative prairie ranged from 7 to 525 kg/ha overthree years (Brown 1943), a small fraction ofthe 1988 average world grain yield of 2480 kg/ha (Loomis and Connor 1992). Improved pas-tures typically have greater productivity or car-rying capacity for livestock, relative to nativevegetation (Phillips and Coleman 1995; Rosi-ere and Torell 1996). Few, if any, comparisonsbetween natural and agricultural ecosystemshave thoroughly controlled for the effects ofsoil, climate, or external inputs, but we con-sider it unlikely that natural ecosystems willconsistently outperform agricultural systems,by most agricultural criteria.

We conclude that “Nature’s wisdom” residesmainly in the sophisticated adaptations of indi-vidual plants and animals, not in ecosystemstructure. Any given feature of a natural eco-system may be there “for a reason,” in the samesense that earthquakes happen for a reason,but that does not mean they are there for apurpose. Mindless mimicry of natural ecosys-tems reveals ignorance of, not respect for, eco-logical principles.

the true value of natural ecosystemsThere are many reasons to preserve natural

ecosystems unrelated to their potential asmodels for agriculture, including recreationalvalue, beneficial effects on air and water qual-ity, and the “ecosystem services” such as polli-nation or biological control of pests they pro-vide to nearby agricultural areas (Constanza etal. 1997). Furthermore, despite the limitationsdiscussed above, natural ecosystems may be arich source of ideas for agriculture.

Natural Ecosystems As Case StudiesIf we were trying to improve our educa-

tional system, we would be wise to study theeducational systems of a variety of countries.

Some might serve as models to imitate, whileothers might be equally valuable as examplesof mistakes to avoid. We should take a similarapproach when looking at natural ecosystemsas a source of ideas for agriculture. Althoughthere is no reason to expect all natural eco-systems to be optimal, some are likely to bemore stable or efficient than others. Differ-ences among natural ecosystems may be asinformative as differences between naturaland agricultural ecosystems. Comparativestudies, preferably based on long-term datasets, may well provide information useful indesigning agricultural ecosystems. But blindlycopying a natural ecosystem would be as illog-ical as advocating ethnic cleansing because“they’re doing it in Europe.”

Copying Time-Tested Individual Strategies

A second major way in which the study ofnatural ecosystems may contribute to improv-ing agriculture is more directly linked to ourgene-centered view of natural selection. Nat-ural selection optimizes genes, if anything(Dawkins 1976, 1982). But plant genes thatpersist and spread usually code for functionsthat benefit the plant. Seed collections andother gene banks can preserve valuablegenes, but to understand their function wewill often need to preserve, and refer to, theecosystems in which they evolved.

Consider, for example, plant defensivestrategies. Wild potatoes repel aphids byreleasing a volatile chemical functionallyequivalent to aphid alarm pheromone (Gib-son and Pickett 1983). Volatile chemicalsreleased by other plants, when attacked byherbivorous insects, attract predators andparasitoids (Stowe et al. 1995; Turlings et al.1995) Other plants recruit defenders by pro-viding shelter or other resources to predatoryarthropods (Agrawal and Karban 1997) orants (Brouat et al. 2001). We suggest that itwould be nearly impossible to decipher thephenotypic effects of the plant genes respon-sible for these types of indirect defenses, inthe absence of the other species (herbivores,predators, or parasitoids) involved.

Understanding the strategy may be moreimportant than understanding the gene.Transferring genes from wild to cultivatedplants has often been useful, but this may not


always be the best way to implement the wildplant’s strategy. Understanding costs andbenefits of defenses against aphids based onchemical alarm signal mimicry could be use-ful when considering analogous defensesagainst other pest species, even if the specificchemicals were different. Understandinghow plants maximize benefits from one root-associated microbial species may have appli-cation to other rhizosphere mutualisms (Den-ison et al. 2003). Some plants produceclusters of ephemeral roots that solubilizeand take up soil phosphorus that would nor-mally be unavailable (Dinkelaker et al. 1995).Transferring this trait to new species might bedifficult, but understanding the spatial andtemporal distribution of these “proteoidroots” might suggest new approaches to fer-tilizer placement that would differentiallybenefit crops relative to weeds.

deploying crop diversityAs an example of how a Darwinian per-

spective might contribute to the design ofagricultural ecosystems, consider the use ofcrop diversity in the control of specialist agri-cultural pests. Many natural ecosystems con-tain intimate mixtures of plant species. Naıveecologists might see this as sufficient evidencefor the benefits of such mixtures (i.e., inter-cropping) in agriculture. There may some-times be advantages to growing mixtures ofcrops. Some pests may be less common withintercropping, although a yield benefitlinked to reduced pest populations with inter-cropping has rarely been demonstrated(Risch et al. 1983). A mixture of two or morespecies may sometimes use resources morethoroughly than one (Loreau and Hector2001), although agricultural studies thatclaim this result have often failed to show thatthe single-species reference crop was grownat optimum density (Fukai 1993). One recentstudy showed that a serious disease affectinga high-value rice cultivar could be controlledby growing it in single rows surrounded by 4to 6 rows of a resistant cultivar (Zhu et al.2000). This study also reported 18% highertotal yield with intercropping, but it appearsfrom their Figure 1 that this may have beenan artifact due to suboptimal density of theirmonoculture control.

Attempts to mimic natural ecosystems canblind us to strategies for deploying the avail-able crop diversity that would have evengreater benefits. For example, assume thatthere are three crop species adapted to agiven region and in sufficient demand thatthey could profitably be grown in quantity.We could grow a mixture of all three specieseach year (Figure 1A), hoping that by mim-icking diverse natural ecosystems we wouldconfuse herbivorous insects enough to pre-vent serious crop losses. A Darwinian per-spective might lead to the opposite conclu-sion. Herbivorous insects that evolved in themidst of diverse vegetation are presumablyadept at finding their preferred food plantsin a mixture of many other species; a mixtureof three species is unlikely to be a formidablechallenge.

One problem with growing an intercroppedmixture of all available species is that itexhausts the available crop diversity. Maximiz-ing small-scale spatial diversity leaves noremaining crop diversity to be deployed at thelandscape scale. For pests with effective disper-sal differences on the order of 10 to 100 m, apatchwork of fields, each containing a singlecrop species, might be more effective in lim-iting pest dispersal than an intimate mixtureof species.

Deploying crop diversity on the landscapescale is also more compatible with crop rota-tion (Figure 1B), which deploys diversity overtime. Darwin (1859:119) used the metaphorof a “simultaneous rotation” in discussing spe-cies mixtures in nature, but the effects onpests of diversity in time may be differentfrom those of diversity in space. Pests adaptedto considerable spatial diversity (as found innatural ecosystems) may be poorly adapted todramatic changes in vegetation from one yearto the next, because this sort of change is rela-tively uncommon in many natural ecosystems.A three-year rotation, growing each crop insequence, might therefore provide moreeffective pest control than growing the samethree-crop mixture each year.

Pest evolution can overcome relatively sim-ple crop rotations. A two-year rotation ofmaize with a nonhost was once effectiveagainst northern corn rootworm (Diabroticabarberi), but a mutant with two-year diapause


Figure 1. Two Alternative Ways of Deploying Crop Diversity in Space and TimeTime from planting to harvest is shown on the horizontal axes for three successive years, with intervening

periods when nothing is grown (e.g., severe winters). The vertical scale may differ between A (Intercropping)and B (Crop rotation). The vertical extent of each species shown in A could correspond to a single plant or atransect across one or a few rows of the same species, as opposed to whole fields in B. Numerous variations arepossible, including use of perennials and multiple crops per year as well as systems that incorporate features ofboth intercropping and crop rotation. The main point is that, in evaluating alternative spatial and temporalpatterns, we should not limit ourselves to patterns that resemble the nearest natural ecosystem.


survives until maize is again available (Krysanet al. 1986). A maize-soybean rotation wasonce effective against western corn rootworm(D. virgiferea), but some rootworms now leavethe maize field where they matured and laytheir eggs in adjacent soybean fields in fall,emerging the next spring when maize is sown(Onstad et al. 1999). These pest strategiescould perhaps be defeated by a synchronous,three-year crop rotation over an area largerthan the effective dispersal distance of thepests. If no maize were grown in a large areafor two years, any maize-dependent pest withlimited mobility could become locally extinct.This approach would be less effective withhighly mobile pests and would require con-trol of any weeds or wild plants that wouldserve as alternate hosts. Such an approachwould present logistic problems that may notbe immediately evident to ecologists butwhich would be obvious to farmers. Whetheror not this particular strategy could prove suc-cessful is unclear, but it illustrates the pointthat, in designing agricultural ecosystems, weshould consider the full range of possibilities,not just those that resemble natural ecosys-tems.

ConclusionsWe conclude that human ingenuity is no

match for natural selection, when the latterhas been working on a problem, worldwide,for millennia. Prospects for tradeoff-freeimprovements in traits like photosyntheticefficiency or drought tolerance are thereforepoor, because these traits would haveincreased individual fitness in most environ-ments. On the other hand, natural selectionhas little or no power to optimize the perfor-mance of plant communities or ecosystems,when collective performance conflicts withindividual fitness. In designing crop plantsand agricultural ecosystems that maximizecommunity performance in ways that arebeyond the scope of natural selection, thereis still ample room for human ingenuity.

acknowledgmentsWe thank Bob Loomis, Ted Foin, Cindy Tong, AndyHector, Albert Fischer, and Egbert Leigh, none ofwhom necessarily agrees with our conclusions, forhelpful comments. We received support fromNational Science Foundation grant 0077903 (RFD),an NSF graduate fellowship (ETK), BBSRC (SAW),NERC (SAW), and Royal Society (SAW).

REFERENCES

Agrawal A A, Karban R. 1997. Domatia mediate plant-arthropod mutualism. Nature 387:562–563.

Agrawal A A, Karban R. 1999. Why induced defensesmay be favored over constituitive strategies inplants. Pages 45–61 in The Ecology and Evolution ofInducible Defenses, edited by R Tollrian and C D Har-vell. Princeton (NJ): Princeton University Press.

Allard R W. 1988. Genetic changes associated with theevolution of adaptedness in cultivated plants andtheir wild progenitors. Journal of Heredity 79:225–238.

Alstad D N, Andow D A. 1995. Managing the evolutionof insect resistance to transgenic plants. Science268:1894–1896.

Altieri M A. 1987. Agroecology: The Scientific Basis of Alter-native Agriculture. Boulder (CO): Westview Press.

Amani I, Fischer R A, Reynolds M P. 1996. Canopytemperature depression association with yield ofirrigated spring wheat cultivars in a hot climate.Journal of Agronomy and Crop Science 176:119–129.

Austin R B, Bingham J, Blackwell R D, Evans L T, FordM A, Morgan C L, Taylor M. 1980. Geneticimprovements in winter wheat yields since 1900

and associated physiological changes. Journal ofAgricultural Science 94:675–689.

Ball P. 1999. Shark skin and other solutions. Nature400:507–509.

Ball P. 2001. Life’s lessons in design. Nature 409:413–416.

Barrett S C H. 1983. Crop mimicry in weeds. EconomicBotany 37:255–282.

Bell M A, Fischer R A, Byerlee D, Sayre K. 1995. Geneticand agronomic contributions to yield gains: a casestudy for wheat. Field Crops Research 44:55–65.

Berry J A. 1975. Adaptation of photosynthetic pro-cesses to stress. Science 188:644–650.

Breaker R R, Joyce G F. 1995. A DNA enzyme withMg2�-dependent RNA phosphoesterase activity.Chemistry & Biology 2:655–660.

Broglie K, Chet I, Holliday M, Cressman R, Biddle P,Knowlton S, Mauvais C J, Broglie R. 1991. Trans-genic plants with enhanced resistance to the fun-gal pathogen Rhizoctonia solani. Science 254:1194–1197.

Brouat C, Garcia N, Andary C, McKey D. 2001. Plantlock and ant key: pairwise coevolution of an exclu-


sion filter in an ant-plant mutualism. Proceedings ofthe Royal Society of London Series B 268:2131–2141.

Brown H R. 1943. Growth and seed yields of nativeprairie plants in various habitats of the mixed-prai-rie. Transactions of the Kansas Academy of Science46:87–99.

Brown L R, Kane H. 1994. Full House: Reassessing theEarth’s Population Carrying Capacity. New York:W. W. Norton.

Brown L R, Renner M, Flavin C. 1998. Vital Signs 1998:The Environmental Trends That Are Shaping ourFuture. New York: W. W. Norton.

Browning J A. 1981. The agroecosystem-natural eco-system dichotomy and its impact on phytopatho-logical concepts. Pages 159–172 in Pests, Pathogens,and Vegetation, edited by J M Thresh. Marshfield(MA): Pitman Publishing.

Cao H, Li X, Dong X. 1998. Generation of broad-spectrum disease resistance by overexpression ofan essential regulatory gene in systemic acquiredresistance. Proceedings of the National Academy of Sci-ences of the United States of America 95:6531–6536.

Cassman K G. 1999. Ecological intensification ofcereal production systems: yield potential, soilquality, and precision agriculture. Proceedings of theNational Academy of Sciences of the United States ofAmerica 96:5952–5959.

Cheplick G P. 1992. Sibling competition in plants. Jour-nal of Ecology 80:567–575.

Cock J H, Franklin D, Sandoval G, Juri P. 1979. Theideal cassava plant for maximum yield. Crop Science19:271–279.

Constanza R, d’Arge R, de Groot R, Farber S, GrassoM, Hannon B, Limburg K, Naeem S, O’Neill R V,Paruelo J, Raskin R G, Sutton P, van den Belt M.1997. The value of the world’s ecosystem servicesand natural capital. Nature 387:253–260.

Conway G. 1998. The Doubly Green Revolution: Food ForAll in the Twenty-First Century. Ithaca (NY ): Com-stock Publishing Associates.

Cornish K, Radin J W, Turcotte E L, Lu Z, Zeiger E.1991. Enhanced photosynthesis and stomatal con-ductance of Pima cotton (Gossypium barbadense L.)bred for increased yield. Plant Physiology 97:484–489.

Darwin C R. 1859. The Origin of Species. Reprint. NewYork: Collier, 1962.

Dawkins R. 1976. The Selfish Gene. Oxford: Oxford Uni-versity Press.

Dawkins R. 1982. The Extended Phenotype: The Gene Asthe Unit of Selection. Oxford: Oxford UniversityPress.

Dawkins R, Krebs J R. 1979. Arms races between andwithin species. Proceedings of the Royal Society of Lon-don Series B 205:489–511.

Denison R F. 1998. Decreased oxygen permeability: auniversal stress response in legume root nodules.Botanica Acta 111:191–192.

Denison R F. 2000. Legume sanctions and the evolu-tion of symbiotic cooperation by rhizobia. AmericanNaturalist 156:567–576.

Denison R F, Bledsoe C, Kahn M L, O’Gara F, SimmsE L, Thomashow L S. 2003. Cooperation in therhizosphere and the “free rider” problem. Ecology84(4):826–833.

de Wit C T. 1978. Summative address: one. Pages 405–410 in Plant Relations in Pastures, edited by J R Wil-son. East Melbourne (Australia): CSIRO.

Dinkelaker B, Hengeler C, Marschner H. 1995. Distri-bution and function of proteoid roots and otherroot clusters. Botanica Acta 108:183–200.

Dobzhansky T. 1973. Nothing in biology makes senseexcept in the light of evolution. American BiologyTeacher 35:125–129.

Domınguez C A. 1995. Genetic conflicts of interest inplants. Trends in Ecology & Evolution 10:412–416.

Donald C M. 1968. Breeding of crop ideotypes. Euphy-tica 17:385–403.

Dudley J W, Lambert R J. 1969. Genetic variability after65 generations of selection in Illinois high oil, lowoil, high protein, and low protein strains of Zeamays L. Crop Science 9:179–181.

Duvick D N, Cassman K G. 1999. Post-green revolutiontrends in yield potential of temperate maize in thenorth-central United States. Crop Science 39:1622–1630.

Eshel I, Feldman M W. 2001. Optimality and evolu-tionary stability under short-term and long-termselection. Pages 161–190 in Adaptationism and Opti-mality, edited by S H Orzack and E Sober. Cam-bridge: Cambridge University Press.

Evans L T. 1993. Crop Evolution, Adaptation, and Yield.Cambridge: Cambridge University Press.

Evans L T, Fischer R A. 1999. Yield potential: its defi-nition, measurement, and significance. Crop Science39:1544–1551.

Ewel J J. 1999. Natural systems as models for the designof sustainable systems of land use. Pages 1–21 inAgriculture As a Mimic of Natural Ecosystems, editedby E C Lefroy et al. Dordrecht: Kluwer AcademicPublishers.

Feldhamer G A, Kilbane T P, Sharp D W. 1989. Cumu-lative effect of winter on acorn yield and deer bodyweight. Journal of Wildlife Management 53:292–295.

Fischer R A. 1978. Are your results confounded byintergenotypic competition? Page 731 in 5th Inter-national Wheat Genetics Symposium, edited by SRamanujam. New Delhi: Indian Society of Genet-ics and Plant Breeding.

Fukai S. 1993. Intercropping—bases of productivity.Field Crops Research 34:239–245.

Garry D J, Ordway G A, Lorenz J N, Radford N B, ChinE R, Grange R W, Bassel-Duby R, Williams R S.1998. Mice without myoglobin. Nature 395:905–908.


Gibson R W, Pickett J A. 1983. Wild potato repelsaphids by release of aphid alarm pheromone.Nature 302:608–609.

Giles J. 2001. Think like a bee. Nature 410:510–512.Gliessman S. 1990. Agroecology: researching the eco-

logical basis for sustainable agriculture. Agroecology:Researching the Ecological Basis for Sustainable Agri-culture, edited by S R Gliessman. New York:Springer-Verlag.

Godecke A, Flogel U, Zanger K, Ding Z, HirchenhainJ, Decking U K M, Schrader J. 1999. Disruption ofmyoglobin in mice induces multiple compensatorymechanisms. Proceedings of the National Academy ofSciences of the United States of America 96:10495–10500.

Goodman D. 1975. The theory of diversity-stabilityrelationships in ecology. Quarterly Review of Biology50:237–266.

Goodnight C J. 1990. Experimental studies of com-munity evolution. I. The response to selection atthe community level. Evolution 44:1614–1624.

Gorham E, Vitousek P M, Reiners W A. 1979. Theregulation of chemical budgets over the course ofecological succession. Annual Review of Ecology andSystematics 10:53–84.

Griffin A S, West S A. 2002. Kin selection: fact andfiction. Trends in Ecology & Evolution 17:15–21.

Gustafsson A. 1951. Mutations, environment and evo-lution. Cold Spring Harbor Symposia on QuantitativeBiology 16:263–281.

Gutierrez A P, Daxl R. 1984. Economic thresholds forcotton pests in Nicaragua: ecological and evolu-tionary perspectives. Pages 184–205 in Pest andPathogen Control, Strategic, Tactical, and Policy Models,edited by G R Conway. Chichester (UK): JohnWiley and Sons.

Hamilton W D. 1964. The genetical evolution of socialbehavior. I. Journal of Theoretical Biology 7:1–16.

Harik G R, Lobo F G, Goldberg D E. 1999. The com-pact genetic algorithm. IEEE Transactions on Evo-lutionary Computation 3:287–297.

Harper J L. 1977. Population Biology of Plants. London:Academic Press.

Heap I M. 1997. The occurrence of herbicide-resistantweeds worldwide. Pesticide Science 51:235–243.

Herre E A. 1993. Population structure and the evolu-tion of virulence in nematode parasites of figwasps. Science 259:1442–1445.

Hill J E, Smith R J, Bayer D E. 1994. Rice weed control:current technology and emerging issues in tem-perate rice. Australian Journal of Experimental Agri-culture 34:1021–1029.

Hutmacher R B, Nightingale H I, Rolston D E, BiggarJ W, Dale F, Vail S S, Peters D. 1994. Growth andyield responses of almond (Prunus amygdalus) totrickle irrigation. Irrigation Science 14:117–126.

Jackson L E, Koch G W. 1997. The ecophysiology of

crops and their wild relatives. Pages 3–37 in Ecologyin Agriculture, edited by L E Jackson. San Diego(CA): Academic Press.

Jackson W, Piper J. 1989. The necessary marriagebetween ecology and agriculture. Ecology 70:1591–1593.

Jacob F. 1977. Evolution and tinkering. Science 196:1161–1166.

Jennings P R, Aquino R C. 1968. Studies on competi-tion in rice. III. The mechanism of competitionamong phenotypes. Evolution 22:529–542.

Jennings P R, de Jesus J. 1968. Studies on competitionin rice. I. Competition in mixtures of varieties. Evo-lution 22:119–124.

Kamat S, Su X, Ballarini R, Heuer A H. 2000. Struc-tural basis for the fracture toughness of the shellof the conch Strombus gigas. Nature 405:1036–1040.

Kellogg E A. 1999. Phylogenetic aspects of the evolu-tion of C4 photosynthesis. Pages 411–444 in C4

Plant Biology, edited by R F Sage and R K Monson.San Diego (CA): Academic Press.

Kokubun M. 1988. Design and examination of soy-bean ideotypes. Japan Agricultural Research Quarterly21:237–243.

Krysan J L, Foster D E, Branson T F, Ostlie K R, Cran-shaw W S. 1986. Two years before the hatch: root-worms adapt to crop rotation. Bulletin of the Ento-mological Society of America 32:250–253.

Ku M S B, Cho D, Ranade U, Hsu T P, Li X, Jiao DM, Ehleringer J, Miyao M, Matsuoka M. 2000.Photosynthetic performance of transgenic riceplants overexpressing maize C4 photosynthesisenzymes. Pages 193–204 in Redesigning Rice Pho-tosynthesis to Increase Yield, edited by J E Sheehy etal. Amsterdam: International Rice Research Insti-tute and Elsevier.

Lefroy E C, Hobbs R J, O’Connor M H, Pate J S, edi-tors. 1999. Agriculture as a Mimic of Natural Ecosys-tems. Dordrecht: Kluwer Academic Publishers.

Leigh E G. 1983. When does the good of the groupoverride the advantage of the individual? Proceed-ings of the National Academy of Sciences of the UnitedStates of America 80:2985–2989.

Leigh E G. 2001. Adaptation, adaptationism, and opti-mality. Pages 358–387 in Adaptationism and Opti-mality, edited by S H Orzack and E Sober. Cam-bridge: Cambridge University Press.

Letourneau D K. 1990. Code of ant-plant mutualismbroken by parasite. Science 248:215–217.

Levin B R, Kilmer W L. 1974. Interdemic selection andthe evolution of altruism: a computer simulationstudy. Evolution 28:527–545.

Lewis W J, van Lenteren J C, Phatak S C, Tumlinson JH. 1997. A total system approach to sustainablepest management. Proceedings of the National Acad-emy of Sciences of the United States of America 94:12243–12248.


Lipson H, Pollack J B. 2000. Automatic design andmanufacture of robotic lifeforms. Nature 406:974–978.

Liu D, Raghothama K G, Hasegawa P M, Bressan R A.1994. Osmotin overexpression in potato delaysdevelopment of disease symptoms. Proceedings of theNational Academy of Sciences of the United States ofAmerica 91:1888–1892.

Long S P. 1998. Rubisco, the key to improved cropproduction for a world population of more thaneight billion people? Pages 124–136 in Feeding aWorld Population of More Than Eight Billion People,edited by J C Waterlow. Oxford and New York:Oxford University Press.

Loomis R S. 1979. Ideotype concepts for sugarbeetimprovement. Journal of the American Society of SugarBeet Technologists 20:323–342.

Loomis R S. 1993. Optimization theory and cropimprovement. Pages 583–588 in International CropScience I, edited by D R Buxton et al. Madison: CropScience Society of America.

Loomis R S, Connor D J. 1992. Crop Ecology: Productivityand Management in Agricultural Systems. Cambridge:Cambridge University Press.

Loreau M, Hector A. 2001. Partitioning selection andcomplementarity in biodiversity experiments.Nature 412:72–76.

Lu Z, Radin J W, Turcotte E L, Percy R, Zeiger E. 1994.High yields in advanced lines of Pima cotton areassociated with higher stomatal conductance,reduced leaf area and lower leaf temperature. Phy-siologia Plantarum 92:266–272.

Martin S J, Beekman M, Wossler T C, Ratnieks F L W.2002. Parasitic Cape honeybee workers, Apis melli-fera capensis, evade policing. Nature 415:163–165.

Matsuoka M, Fukayama H, Tsuchida H, Nomura M,Agarie S, Ku M S B, Miyao M. 2000. How to expresssome C4 photosynthesis genes at high levels in rice.Pages 167–175 in Redesigning Rice Photosynthesis toIncrease Yield, edited by J E Sheehy et al. Amster-dam: Elsevier.

Maynard Smith J. 1964. Group selection and kin selec-tion. Nature 201:1145–1147.

Maynard Smith J. 1976. Group selection. QuarterlyReview of Biology 51:277–283.

McCann K, Hastings A, Huxel G R. 1998. Weak trophicinteractions and the balance of nature. Nature395:794–798.

McIntosh R P. 1998. The myth of community as organ-ism. Perspectives in Biology and Medicine 41:426–438.

Muir W M. 1996. Group selection for adaptation tomultiple-hen cages: selection program and directresponses. Poultry Science 75:447–458.

Odum E P. 1969. The strategy of ecosystem develop-ment. Science 164:262–270.

Onstad D W, Joselyn M G, Isard S A, Levine E, SpencerJ L, Bledsoe L W, Edwards C R, di Fonzo C D, Will-

son H. 1999. Modeling the spread of western cornrootworm (Coleoptera: Chrysomelidae) popula-tions adapting to soybean-corn rotation. Environ-mental Entomology 28:188–194.

Ortiz-Monasterio J I, Sayre K D, Rajaram S, McMahonM. 1997. Genetic progress in wheat yield and nitro-gen use efficiency under four nitrogen rates. CropScience 37:898–904.

Paine R T. 1966. Food web complexity and speciesdiversity. American Naturalist 100:65–75.

Peng S, Cassman K G, Virmani S S, Sheehy J, KhushG S. 1999. Yield potential trends of tropical ricesince the release of IR8 and the challenge ofincreasing rice yield potential. Crop Science 39:1552–1559.

Phillips W A, Coleman S W. 1995. Productivity andeconomic return of three warm season grassstocker systems for the southern Great Plains. Jour-nal of Production Agriculture 8:334–339.

Potvin C, Simon J-P. 1989. The evolution of cold tem-perature adaptation among populations of awidely distributed C4 weed: barnyard grass. Evolu-tionary Trends in Plants 3:98–105.

Ribbe M, Gadkari D, Meyer O. 1997. N2 fixation byStreptomyces thermoautotrophicus involves a molybde-num-dinitrogenase and a manganese-superoxideoxidoreductase that couple N2 reduction to theoxidation of superoxide produced from O2 by amolybdenum-CO dehydrogenase. Journal of Bio-logical Chemistry 272:26627–26633.

Risch S J, Andow D, Altieri M A. 1983. Agroecosystemdiversity and pest control: data, tentative conclu-sions, and new research directions. EnvironmentalEntomology 12:625–629.

Robertson G P, Paul E A, Harwood R R. 2000. Green-house gases in intensive agriculture: contributionsof individual gases to the radiative forcing of theatmosphere. Science 289:1922–1925.

Rosiere R E, Torell D T. 1996. Performance of sheepgrazing California annual range. Sheep and GoatResearch Journal 12:49–57.

Rowe M P, Pugh E N, Tyo J S, Engheta N. 1995. Polar-ization-difference imaging—a biologically inspiredtechnique for observation through scatteringmedia. Optics Letters 20:608–610.

Sage R F. 2000. C3 vs. C4 photosynthesis in rice: eco-physiological perspectives. Pages 13–35 in Redesign-ing Rice Photosynthesis to Increase Yield, edited by J ESheehy et al. Amsterdam: Elsevier.

Santiago F S, Lowe H C, Kavurma M M, ChestermanC N, Baker A, Atkins D G, Khachigian L M. 1999.New DNA enzyme targeting Egr-1 mRNA inhibitsvascular smooth muscle proliferation andregrowth after injury. Nature Medicine 5:1264–1269.

Schieving F, Poorter H. 1999. Carbon gain in a mul-tispecies canopy: the role of specific leaf area


and photosynthetic nitrogen-use efficiency in thetragedy of the commons. New Phytologist 143:201–211.

Schoner C A, Norris R F, Chilcote W. 1978. Yellow fox-tail (Setaria lutescens) biotype studies: growth andmorphological characteristics. Weed Science 26:632–636.

Sheehy J E. 2000. Limits to yield for C3 and C4 rice:an agronomist’s view. Pages 39–52 in RedesigningRice Photosynthesis to Increase Yield, edited by J ESheehy et al. Amsterdam: Elsevier.

Simmonds N W. 1979. Principles of Crop Improvement.London: Longman Group.

Simmonds N W. 1998. “Shades of Green.” Review ofThe Doubly Green Revolution: Food For All in theTwenty-First Century, by G Conway. Nature 391:139.

Snow A A, Palma P M. 1997. Commercialization oftransgenic plants: potential ecological risks. Bio-Science 47:86–96.

Soliman K M, Allard R W. 1991. Grain yield of com-posite cross populations of barley: effects of natu-ral selection. Crop Science 31:705–708.

Soule J D, Piper J K. 1992. Farming in Nature’s Image.An Ecological Approach to Agriculture. Corelo (CA):Island Press.

Stowe M K, Turlings T C J, Loughrin J H, Lewis W J,Tumlinson J H. 1995. The chemistry of eavesdrop-ping, alarm, and deceit. Proceedings of the NationalAcademy of Sciences of the United States of America92:23–28.

Suneson C A. 1956. An evolutionary plant breedingmethod. Agronomy Journal 48:188–191.

Suneson C A, Wiebe G A. 1942. Survival of barley andwheat varieties in mixtures. Journal of the AmericanSociety of Agronomy 34:1052–1056.

Tabashnik B E, Liu Y-B, Finson N, Masson L, HeckelD G. 1997. One gene in diamondback moth con-fers resistance to four Bacillus thuringiensis toxins.

Proceedings of the National Academy of Sciences of theUnited States of America 94:1640–1644.

Turlings T C J, Loughrin J H, McCall P J, Rose U S R,Lewis W J, Tumlinson J H. 1995. How caterpillar-damaged plants protect themselves by attractingparasitic wasps. Proceedings of the National Academy ofSciences of the United States of America 92:4169–4174.

Uemura K, Anwaruzzaman, Miyachi S, Yokota A. 1997.Ribulose-1,5-bisphosphate carboxylase/oxygenasefrom thermophilic red algae with a strong specific-ity for CO2 fixation. Biochemical and BiophysicalResearch Communications 233:568–571.

Wardle D A, Zackrisson O, Hornberg G, Gallet C.1997. The influence of island area on ecosystemproperties. Science 277:1296–1299.

Williams G C. 1966. Adaptation and Natural Selection.Princeton (NJ): Princeton University Press.

Williams G C. 1992. Natural Selection: Domains, Levels,and Challenges. Oxford and New York: Oxford Uni-versity Press.

Williams G C, Nesse R M. 1991. The dawn of Darwin-ian medicine. Quarterly Review of Biology 66:1–22.

Wynne-Edwards V C. 1962. Animal Dispersion in Relationto Social Behavior. Edinburgh: Oliver and Loyd.

Ying J F, Peng S B, He Q R, Yang H, Yang C D, VisperasR M, Cassman K G. 1998. Comparison of high-yieldrice in tropical and subtropical environments. I.Determinants of grain and dry matter yields. FieldCrops Research 57:71–84.

Zelitch I. 1992. Control of plant productivity by regu-lation of photorespiration. BioScience 42:510–516.

Zhang D Y, Sun G J, Jiang X H. 1999. Donald’s ideo-type and growth redundancy: a game theoreticalanalysis. Field Crops Research 61:179–187.

Zhu Y, Chen H, Fan J, Wang Y, Li Y, Chen J, Fan J X,Yang S, Hu L, Leung H, Mew T W, Teng P S, WangZ, Mundt C C. 2000. Genetic diversity and diseasecontrol in rice. Nature 406:718–722.

Documents

Volume 78, No. June 2003 The Quarterly Reviewpeople.umass.edu/lsadler/adlersite/kiers/DarwinianAgriculture1.pdfwheat (Triticum aestivum), rice (Oryza sativa), and maize (Zea mays)