Mauricio NRG 2001

370 | MAY 2001 | VOLUME 2 www.nature.com/reviews/genetics

R E V I E W S

The world has recently secured the first completegenetic blueprint of a plant. The mouse-ear cress,Arabidopsis thaliana, a small mustard plant, hasrecently joined the increasingly less exclusive club oforganisms for which every gene has beensequenced16. In the next three years, the completesequence of a plant of worldwide agricultural impor-tance, rice (Oryza sativa), will be available7.Undoubtedly, the dawning of the genomics age willhave a great impact on agriculture, human health andmolecular genetics. This wealth of genetic informa-tion is beginning to have just as profound an impacton the study of evolutionary biology, particularly onunderstanding one of the most enduring problems inevolution and molecular biology the genetic basisof complex traits.

The complexity of these phenotypic traits,particularly of those involved in adaptation, probablyarises from segregation of alleles at many interactingloci (quantitative trait loci, or QTL), the effects of which are sensitive to the environment810. Recentand continuing advances in molecular genetics and statistical techniques make it possible to identify the chromosomal regions where these QTL are located. Ultimately, an understanding of adaptive evolution will require detailed know-ledge of the genetic changes that accompany evolutionary change.

A needle in a haystackModern quantitative genetics was born from thefusion of Mendelism and biometry, the mathematicaltheory that surrounds the science of heredity (BOX 1).The conceptual basis for the genetic dissection of com-plex traits is relatively straightforward. At its mostbasic level, QTL mapping simply involves finding anassociation between a genetic marker and a phenotypethat one can measure (FIG. 1). For example, if all the tallplants among 500 individual corn plants of varyingheight have a particular allele of a genetic marker, thenthere is a very high probability that a QTL for planttallness is associated with this marker in this popula-tion of plants.

QTL mapping in both plants and animals involvesjust a few basic steps11. The primary requirement is fortwo parental strains that have differences betweenthem in the alleles that affect variation in a trait. Theparents need not be different in the mean phenotypicvalue of the trait as different allelic combinations canyield the same phenotypic mean. A polymorphicgenetic map allows the two strains to be distinguishedgenetically. The more detailed the linkage map (that is,the greater the number of markers), the better themapping resolution. The parental alleles are then shuf-fled by creating a large mapping population, in whichthe phenotype and the multilocus genotype of eachindividual are measured.

MAPPING QUANTITATIVE TRAITLOCI IN PLANTS: USES AND CAVEATSFOR EVOLUTIONARY BIOLOGYRodney Mauricio

Gregor Mendel was either clever or lucky enough to study traits of simple inheritance in hispea plants; however, many plant characters of interest to modern geneticists are decidedlycomplex. Understanding the genetic basis of such complex, or quantitative, traits requiresa combination of modern molecular genetic techniques and powerful statistical methods.These approaches have begun to give us insight into understanding the evolution ofcomplex traits both in crops and in wild plants.

Department of Genetics,Life Sciences Building,University of Georgia,Athens, Georgia 30602-7223, USA. e-mail:[email protected]

2001 Macmillan Magazines Ltd

NATURE REVIEWS | GENETICS VOLUME 2 | MAY 2001 | 371

R E V I E W S

In practice, several crossing schemes are used togenerate this mapping population12. In all of these,the parents are mated to generate an F

1population.

In one approach, recombinant inbred lines can becreated by selfing (self-fertilizing) each of the F

1

progeny for several (usually eight) generations (FIG.1)13. In an F

2design, the mapping population is gen-

erated by mating the F1

progeny to each other (FIG. 2).In a backcross design, the mapping population isgenerated by crossing the F

1progeny to either, or

both, of the parents (FIG. 3). Several variations onthese crossing schemes have been designed to maxi-mize the shuffling of parental alleles12,14. Once allindividuals in the mapping population are scored forphenotype and multilocus genotype, the actual QTLmapping can begin. The statistical tools at the foun-dations of QTL mapping have been used for manyyears (BOX 2). In fact, in 1923, Karl Sax mapped a QTLfor seed size in the bean, Phaseolus vulgaris, by statis-tically associating it with a Mendelian locus for seed pigmentation15.

Today, we generally have much more detailed geneticmaps available. For example, Arabidopsis thaliana has1,262 genetic markers, which consist of restriction frag-ment length polymorphisms (RFLPs) and singlenucleotide polymorphisms (SNPs) that vary betweenthe two parental lines of the most commonly used set ofrecombinant inbred lines (FIG. 1)16,17. Cereon Genomicshas identified and made available a collection of 28,117SNPs, which include 15,674 insertion/deletion poly-morphisms that are polymorphic between the two par-ents of those same recombinant inbred lines.

Ultimately, QTL analysis yields a statistical descrip-tion of the genes that underlie the phenotypes ofinterest. The statistical fog is not completely lifted,but we can see the shadows of those genes.

Box 1 | Quantitative genetics reborn

evolution is essentially a statistical problem, W. F. R. Weldon (1893)

The early history of evolutionary genetics focused on understanding complex traits, particularly those relatingto humans, such as intelligence, temper and artistic faculty99. Without the benefit of Mendels ideas, FrancisGalton and the mathematician, Karl Pearson, established that useful predictions of the evolutionary trajectory of complex traits could be made without recourse to an explicit understanding of inheritance99. This area ofquantitative genetics rested on the statistical properties of the MULTIVARIATE NORMAL DISTRIBUTION8,18.

With the rediscovery of Mendels theory of heredity in 1900, a conflict arose between this biometrical schoolof quantitative genetics and the discrete genetics of the Mendelian school99. By 1910, it had been shown thatcontinuous phenotypic variation could result from the action of the environment on the segregation of manyMendelian loci99,100. By 1918, Ronald Fisher convincingly reconciled the discrete inheritance of Mendelism withthe biometrical approach101. Modern evolutionary quantitative genetics is largely based on the same statisticalfoundations that were laid by Pearson and Fisher102108.

For most of the twentieth century, quantitative genetics had a crucial role in both agriculture and evolutionarybiology, but never seemed fully embraced by modern molecular genetics. There was (and perhaps still is) awidespread perception that quantitative genetics essentially ignored genetics, blanketing actual genes in what has been called a statistical fog109.

Mapping quantitative trait loci (QTL) the single or many genes that underlie quantitative phenotypes might represent an important part of the final synthesis of molecular and quantitative genetics. By working from the phenotype to the genotype, QTL mapping uses statistical techniques to localize chromosomal regions that might contain genes contributing to phenotypic variation in a complex trait of interest. Workingfrom the gene to the phenotype, molecular geneticists might be able to meet quantitative geneticists at somegenetic Promontory Point.

F1

a

b

cF8

Parent 1 Parent 2

F1...

Figure 1 | Principles of mapping quantitative trait loci.The basic strategy behind mapping quantitative trait loci(QTL) is illustrated here for a | the density of hairs(trichomes) that occur on a plant leaf. Inbred parents thatdiffer in the density of trichomes are crossed to form an F1population with intermediate trichome density. b | An F1individual is selfed to form a population of F2 individuals. c | Each F2 is selfed for six additional generations,ultimately forming several recombinant inbred lines (RILs).Each RIL is homozygous for a section of a parentalchromosome. The RILs are scored for several geneticmarkers, as well as for the trichome density phenotype. In c, the arrow marks a section of chromosome thatderives from the parent with low trichome density. Theleaves of all individuals that have inherited that section ofchromosome from the parent with low trichome densityalso have low trichome density, indicating that thischromosomal region probably contains a QTL for this trait.

MULTIVARIATE NORMAL

DISTRIBUTION

The central limit theoremassures a normal (bell-shaped)distribution for a variable thatis the summation of manyindependent, random inputs.This applies to single ormultiple variables.



R E V I E W S

a straw man these assumptions cannot be entirelycorrect as there are only a finite number of genesavailable in a genome. (Conversely, one of the mostimportant traits to an evolutionary biologist is fitness a very complex trait that is almost certainly con-trolled by many genes). On a practical level, once fiveor more loci contribute to a trait, it might be easier tomodel the evolution (or perhaps even function) ofthat trait with a mathematical abstraction rather thanattempting to explicitly take into account each indi-vidual locus. Nevertheless, QTL mapping might indi-cate to evolutionary biologists the direction in whichto proceed.

Genetics of adaptation. One of the most enduring con-troversies in evolutionary biology is the genetic basis ofadaptation19,20. R. A. Fisher concluded that mutations ingenes of very small effect were responsible for adaptiveevolution21. H. Allen Orr and Jerry Coyne stated that the neo-Darwinian view has triumphed, andthe genetic basis of adaptation now receives little atten-tion. Indeed, the question is considered so dead thatfew know the evidence responsible for its demise20.Orr and Coyne re-examined the evidence for thisFisherian view and argued that both the theoretical andempirical basis for it were weak20. They encouragedevolutionary biologists to re-examine this researchquestion by the genetic analysis of adaptive differencesbetween natural populations or species20.

Plant evolutionary biologists have embraced thischallenge and there are many examples of the use ofQTL analysis to determine the genetic basis of traits thatare thought to be of adaptive value. Several QTL havebeen identified for seed size, fruit size and seed numberin A. thaliana22. Seed size is a very important adaptivetrait in that large seeds often have higher fitness thansmall seeds23. Variation in seed size in A. thaliana wasfound to be determined by 11 QTL with relatively smalladditive effects. Seven of the seed size QTL localized tothe same position as seed weight QTL, indicating possi-ble pleiotropic actions of the underlying genes. So, QTLanalysis can provide information about the mode ofgene action.

QTL analysis has also been used to understand thegenetic correlations among traits in natural popula-tions. Floral traits, such as petal length and width,have long been used as an example of positively,genetically correlated traits24. In a study of floral mor-phology in A. thaliana, 18 QTL were found25. As inthe study of seed size, 11 floral trait QTL were associ-ated with more than one floral trait. So, the tightmorphological integration of the flower is possiblydue to pleiotropic gene action, or to tight linkage of loci25.

The timing of flowering is another trait of impor-tance to plants, as earlier reproduction often translatesinto higher fitness26. Three research groups have inde-pendently examined flowering time in A. thalianausing three unique crosses. A cross between the late-flowering strain, HM, and the early-flowering strain,WS, revealed that two unlinked QTL affect flowering

Genetics of evolutionQTL mapping is beginning to be embraced by evolu-tionary biologists as a means to answer various basicquestions. Heritable phenotypic variation is the rawmaterial of evolution, producing adaptations andorganismal diversity. A comprehensive understand-ing of these evolutionary mechanisms will beenhanced by elucidating the molecular genetic basisof quantitative traits.

Several important evolutionary models are basedon the key assumptions of quantitative genetics. Thesemodels assume that the complex phenotype of a traitis caused by the simultaneous segregation of a verylarge number of genes, each of which has a small,additive effect on the phenotype and interacts withother genes and with the environment18. QTL analysisallows us to test these assumptions. At one level, this is

a b c

d e f

g h i

j k l

Stigma(inserted)Corolla

(open)Anthers(inserted)

Nectarguides(brushy)

Nectarguides(flat)

Anthers(exserted)

Corolla(tubular)

Stigma(exserted)

Figure 2 | F2 design: genetic mapping in monkeyflowers. a | Mimulus lewisii andc | Mimulus cardinalis were crossed to produce b | a fertile F1 progeny. The F1 was self-pollinated to produce dl | various F2 progeny. The parent species differ in floral characteristics,including petal colour, COROLLA size and shape, presence or absence of NECTAR GUIDES, nectarvolume, and concentration and position of ANTHERS and STIGMA. These floral characteristics areimportant in maintaining pollinator-induced reproductive isolation between these two species inthe wild. (Adapted with permission from REF. 40 (1999) National Academy of Sciences, USA.)

COROLLA

Collectively, the petals ofa flower.

NECTAR GUIDES

Markings on the petals offlowers, often in contrastingcolours or visible only inultraviolet wavelengths,thought to act as directionalbeacons for pollinators,especially bees.

ANTHER

The pollen-bearing part of themale floral structure (stamen).



R E V I E W S

Genetics of speciation. Extending QTL analysis to otherplant systems is particularly important in addressing asecond fundamental question in evolutionary biology:Are the genes that are variable in a population the sameas those that cause divergence between populations andspecies?35 As in the case of adaptive traits, there has beensome controversy as to the role of the principal genes inspeciation36,37. Recent QTL analyses use crosses between(necessarily) closely related species. By creating suchhybrids, or by looking in natural HYBRID ZONES, the originof species can be investigated at the genetic level.

Toby Bradshaw, Douglas Schemske and their col-leagues pioneered this approach with their study of spe-ciation in monkeyflowers (Mimulus)3840. They crossedtwo species with contrasting floral traits. Mimuluslewisii is a bumble-bee-pollinated flower with pinkpetals, contrasting yellow nectar guides, a wide corollaopening, concentrated nectar, and inserted anthers andstigma (FIG. 2a). Mimulus cardinalis is pollinated by hum-mingbirds and has red petals, a narrow, tubular corolla,copious nectar and exserted anthers and stigma (FIG.2c)38. The two species grow and flower together, buthybrids are not commonly observed in nature.However, they are completely interfertile when artificial-ly mated (FIG. 2b). So, these two species are reproductive-ly isolated owing to different preferences of the pollina-

time on chromosome 5 (REF. 27). In a cross between theColumbia and Landsberg strains, two QTL, one onchromosome 1 and another on chromosome 2, werefound28. Finally, a cross between the early-floweringstrain, Li-5, and the late-flowering strain, Naantali,revealed at least seven QTL, with one QTL on chromo-some 4 accounting for over 50% of the variation inflowering time29. The other minor QTL were found oneach of the five chromosomes of A. thaliana29. As thisexample illustrates, the QTL found for any particularcharacter might be different depending on the parentsused in the original cross.

Although A. thaliana has been a popular plant forstudying the genetics of adaptation using QTL analysis,the power of the approach is best illustrated by consid-ering studies of plants that are decidedly not annualweeds. QTL analysis has been successfully applied tostudying the genetic basis of various important traits inforest trees, including pine, eucalyptus and poplar. QTLhave been mapped for seedling height, leaf area andfrost tolerance in the high altitude Eucalyptus nitens30,31

and for frost hardiness and timing of bud set in naturalpopulations of Scots pine (Pinus sylvestris)32. In studiesof poplars, cottonwoods and aspens (Populus spp.),researchers have found QTL for bud set, bud flush,growth, form, PHENOLOGY and leaf shape33,34.

STIGMA

The region, usually the apex,of the gynoecium that receivespollen grains and on which the pollen germinates. Thegynoecium is the seed-bearingorgan of flowering plants,consisting of the stigma, styleand ovary.

STYLAR TUBE

In the genus Iris, the styles(tubular columns of tissuearising from the top of theovary) look like flower petalsand are tightly appressed to thetop of the actual petals, formingthis tube between the petal andthe style.

PHENOLOGY

The timing of periodicbiological phenomena that are usually correlated withclimactic conditions.

HYBRID ZONE

A region of reproductionbetween individuals of differentspecies, usually occurring where the ranges of the speciescome together.

a b

c d

Figure 3 | Backcross design: genetic mapping in the Louisiana irises. a | Iris fulva (swamp iris) and d | Iris brevicaulis (blueiris) were crossed to produce an F1. An F1 backcrossed to Iris brevicaulis produced hybrids such as b |, and an F1 backcrossedto Iris fulva produced hybrids such as c |. The parental species differ in floral characteristics, including petal and sepal colour,presence or absence of nectar guides, degree of petal reflexivity (petals of I. fulva curl down and inward, whereas those of I.brevicaulis angle upward), and anther position (the anthers of I. fulva extend out of the STYLAR TUBE, those of I. brevicaulis areinside the stylar tube). Backcross genotypes show a wide range of variation in these traits. These floral characteristics, as well as others involved in the reproductive isolation of these species, are now being analysed using quantitative trait loci mappingmethods. (Courtesy of Amy Bouck, Department of Genetics, University of Georgia, USA.)



R E V I E W S

Box 2 | Quantitative trait loci mapping methods

Several techniques exist for mapping quantitative trait loci (QTL) in the population; I have used the example of trichomedensity in leaves described in the figure to illustrate these methods. a | In the regression technique, the phenotype iscorrelated with each marker genotype110 (the middle panel represents the differential migration of DNA on a gel). In thiscase, a single marker A is scored. Individuals that are homozygous for the A allele have high trichome density, individualsthat are homozygous for the a allele have low trichome density, and heterozygotes have intermediate trichome density. Alinear regression of trichome density on the number of A alleles shows a significant relationship between the marker andthe phenotype, which indicates that a QTL for trichome density is probably linked to that marker. The simple regressionmethod described is of limited use in localizing the chromosomal segment that contains a QTL. The methodunderestimates the effect of the QTL. The further the QTL is from the marker, the weaker the effect.

The interval mapping method uses a pair or two pairs of flanking markers at a time83,111. b | In this approach, the QTL is located within a chromosomal interval, defined by the flanking markers. The technique involves scoring a large numberof markers, as illustrated in the top panel, and then assessing the probability that an interval between two markers isassociated with a QTL that affects the trait of interest. The results of the analysis are plotted as a LIKELIHOOD-RATIO TESTSTATISTIC against the chromosomal map position, measured in recombination units (Morgans). The dotted line representsa significance threshold above which a likelihood-ratio test provides a statistically significant fit to a model of the data.The best estimate of the location of the QTL is given by the chromosomal location that corresponds to the highestsignificant likelihood ratio. Although the interval mapping method was an important advance, it too is statisticallybiased. In particular, QTL outside the interval under consideration can affect the ability to find a QTL within it112. Inaddition, false identification of a QTL can arise if other QTL are linked to the interval of interest (the false ghost peak onthe right)112. c | A third method, known as composite interval mapping, combines the interval mapping technique withmultiple regression analysis81,85,112. Composite interval mapping assesses the probability that an interval between twomarkers is associated with a QTL that affects the trait of interest, as well as controlling for the effects of other markers onthe trait (thus providing more accurate results). The results of the analysis are plotted as in b. In this method, the teststatistic is independent of QTL in other regions of the chromosomes (although it is still biased if the QTL is in the intervalimmediately adjacent to the interval of interest)85.

Zhao-Bang Zeng and his colleagues have extended this method to a multiple interval mapping technique. Multipleinterval mapping combines multiple QTL mapping analysis with the analysis of genetic architecture by using analgorithm to search for number, positions, effects and interactions of significant QTL113,114.

Several computer programs that carry out these mapping methods are freely available12,115. Unfortunately, mostof these programs lack an easy-to-use interface and are not of commercial quality, something desperately neededfor most users.

1

4

3

2 6

5

8

7 9

a

1 2 3 4 5 6 7 8 9

AA AA aa Aa Aa AA Aa aa aa

A

a

AA AaGenotype

aa

Tric

hom

e de

nsity

Z

1 2 3 4 5 6 7 8 9

E G

Test

sta

tistic

A B C D F

zAbc

DeFG

zABC

defG

zaBC

DEFG

zA/aBc

deFG

ZA/abc

DEfg

zABC

DEfg

DEFG

zA/abC

zaBC

dEFg

Zabc

defG

Z EMap position

G

Test

sta

tistic

A B C D F

b

c

Ghost peak

LIKELIHOOD-RATIO TEST

STATISTIC

A maximum-likelihoodmethod of hypothesis testing.The likelihood-ratio teststatistic is twice the naturallogarithm of the ratio of themaximum likelihood that thedata fit the alternativehypothesis to the maximumlikelihood that the data fit thenull hypothesis.



R E V I E W S

studying the common weed sunflower, Helianthusannuus, which is postulated to have colonized Texas byacquiring advantageous alleles from the locally adapted,Helianthus debilis4547. Michael Arnold and his col-leagues are studying hybrids of a cross of Iris brevicaulis(FIG. 3d), which prefers dry, sunny habitats, and Iris fulva(FIG. 3a), which prefers wet, shady habitats43,48. By map-ping QTL in the parents that control the ecological traitsthat are responsible for habitat preference and by exam-ining hybrids, it might be possible to show explicitly thatINTROGRESSIVE HYBRIDIZATION is an important evolutionarymechanism in plants.

Field of dreamsQTL mapping has enormous practical potential foragriculture. Although classical quantitative genetics andselective breeding have been enormously successful inagriculture, the process of breeding a crop plant with aparticular desirable trait is a combination of luck, hardwork, time and money. A detailed knowledge of thegenes that affect traits such as yield, DEHISCENCE andresistance, could facilitate and speed the developmentof improved crops49,50.

Gardeners accustomed to planting varieties oftomato know that they can have extraordinarily vari-able fruit sizes, ranging from a weight of a few gramsto as much as half a kilogram (FIG. 4a). Examples ofthese varieties are Burpees Big Boy, Big Girl andWatermelon Beefsteak at one extreme, and the manygrape, currant and cherry varieties at the otherextreme. Steven Tanksley and his colleagues used sevenwild species of tomato and seven different crossingdesigns to locate QTL for fruit size and weight51. Thiswork, starting in the 1980s has identified at least 28different QTL for fruit weight. Some of these QTLhave a significant effect on the phenotype, explainingover 20% of the phenotypic variance51.

Over the past ten years, finer-scale mapping tech-niques were used to localize one of these important QTL,fw2.2, to a narrow chromosomal region52,53. Recently, thestatistical fog was pierced by the cloning of the fw2.2QTL54. The fw2.2 QTL corresponds to a single open read-ing frame in a transformed cosmid (called ORFX) that isexpressed in the floral organs.When the wild-type alleleof the gene was introduced into a cultivated tomato, thetransformed plants produced small fruit of roughly theexpected weight (FIG. 4b). This transformation experimentmarks the movement of QTL mapping beyond a purelystatistical association between a gene and the phenotype.

Beyond the obvious importance of QTL mappingfor agriculture, some of the most striking examples ofthe use of QTL analysis to answer fundamental evolu-tionary questions come from the study of agriculturalspecies. An evolutionary approach to QTL mappinghas allowed new and powerful insights into the under-standing of plant domestication. The work of JohnDoebley and colleagues on the evolution of maize (Zeamays ssp. mays) from its probable wild ancestor,teosinte (Zea mays ssp. parviglumis), is a landmark inthe field of evolutionary QTL analysis5558. For the pastdecade, QTL that are responsible for key steps in the

tors. QTL for the floral traits that distinguish the mon-keyflower species (and, presumably, pollinator prefer-ence) are therefore candidates for speciation genes38.For each of eight floral traits studied, Bradshaw and col-leagues found at least one QTL accounting for morethan 25% of the phenotypic VARIANCE in floral morphol-ogy, and concluded that the evolution of reproductiveisolation might involve genes of major effect38. It is stilltoo early to conclude that speciation genes will be com-monly found in studies of reproductive isolation.Several similar studies are being conducted, includingwork on another pair of SYMPATRIC Mimulus species, M.guttatus and M. nasutus41, on the natural hybrids ofLouisiana irises (FIG. 3)42,43 and on the floral nectar spursof sympatric columbines44.

A final example of the use of QTL analysis in study-ing MACROEVOLUTION is found in continuing work on sun-flowers4547 and the Louisiana irises (FIG. 3)42,43,48. In bothsystems, natural hybrid populations exist in which theparents are adapted to different habitats. It has beensuggested that hybrids can serve as a genetic bridgebetween isolated species, shuttling adaptive genes acrossthe hybrid zone. Loren Rieseberg and his colleagues are

VARIANCE

A statistic that quantifies thedispersion of data about themean. In quantitative genetics,the phenotypic variance (Vp) isthe observed variation of a trait ina population. Vp can bepartitioned into components,owing to genetic variance (Vg),environmental variance (Ve) andgene-by-environmentcorrelations and interactions.

SYMPATRIC

Occurring in the same areawithout loss of identity frominterbreeding.

MACROEVOLUTION

Evolution at or above the level ofspecies.

INTROGRESSIVE HYBRIDIZATION

Incorporation of genes from one species into the gene pool of another species.

DEHISCENCE

The splitting open of a fruit.

a

b

Figure 4 | Genetic basis of phenotypic variation in fruitsize in the tomato. a | Fruits from the wild tomato speciesLycopersicon pimpinellifolium (right) and the cultivatedtomato giant red (Lycopersicon esculentum). Reproducedwith permission from Science 289, 8588 (2000) TheAmerican association for the Advancement of Science. b | Phenotypic effect of a cosmid that contains the small-fruitallele of the fw2.2 transgene in the Mogeor cultivar of L.esculentum. The fruit on the left is a tomato from the Mogeorline. When the same line carries a partially recessive large-fruit allele of fw2.2 on a transgene, the weight of the fruit isreduced (right) on average by 15.5 g (REF. 54). (Courtesy ofSteven Tanksley, Department of Plant Pathology, CornellUniversity, USA.)



R E V I E W S

Finally, QTL analysis has been brought to bear on abiological problem that has stymied geneticists andevolutionary biologists for many years. Many flower-ing plant genomes are thought to be the product ofone or more polyploidization events64,65. As such,many of the worlds crop plants are polyploids, result-ing either from complete duplication of their owngenomes or from interbreeding with another specieswith a failure of complete disjunction at meiosis66. Oneof the most interesting applications of QTL mappingto evolutionary biology is the exploration of the join-ing of these genomes67.

Andrew Paterson and his colleagues have beenexploring such a polyploid event in species of domes-ticated cotton (Gossypium)68,69. Present-day varietiesof cultivated cotton are tetraploid, but are derivedfrom two distinct diploid parental species68. The mostimportant trait to a cotton farmer is the length andquality of the seed fibre. Interestingly, the QTL thatcontribute to fibre quality in domesticated cottonderive from the diploid parent species that possessesno spinnable fibre on its seeds. This indicates a possi-ble non-additive interaction between the two parentalgenomes with respect to seed fibre quality (BOX 3). Thisapplication of QTL analysis shows that the merger ofgenomes of divergent evolutionary histories can pro-duce unique avenues for selection68,69.

A step off the bandwagonThe use of QTL mapping represents a rebirth forquantitative genetics. Quantitative geneticists havebeen successful at both developing robust evolution-ary theory and providing practical benefits to agricul-ture, despite treating genes as a black box. QTL map-ping represents a promising link between thisstatistical approach and an explicit understanding ofthe molecular basis of variation in complex traits.Although I have highlighted examples of the powerand promise of QTL mapping for evolutionary biolo-gy, there are numerous and important caveats to keepin mind.

Maps and markers. Genetic maps are time consum-ing and expensive to construct. Despite this, thenumber of markers is rarely the limiting factor inQTL mapping experiments; the number of progenyexamined is, however, crucial70,71. Each progeny rep-resents an opportunity to identify a unique recombi-nation event between markers. Any QTL that existsbetween two completely linked markers will segregatewith both markers and will be indistinguishable fromeither. Because recombination rates vary across thegenome72, some QTL will be harder or easier to detectdepending on their genetic location. For example, thecentromeric regions, which are known to be sup-pressed for recombination72 and to be poor for genet-ic variation73, are a veritable black hole for QTL map-ping: even two markers that are physically far apartwill seem to be genetically close together near thecentromere. Localization of a gene in these regionscan be difficult.

evolution of maize have been identified59. Theseinclude QTL for the distinct change in branching fromthe bushy, multi-stemmed teosinte to the single-stemmed maize varieties used in agriculture today (FIG.5), and QTL for the change in fruit architecturebetween teosinte, with its encased kernels, to corn withits kernels exposed on the ear58,60. The actual gene thatunderlies the distinct morphological change inbranching pattern, called teosinte branched 1, has beenidentified (FIG. 5)61. Furthermore, teosinte branched 1has been subjected to a molecular population geneticanalysis to understand the evolutionary dynamics ofthe locus62.

An evolutionary approach to QTL mapping canextend beyond single species, yielding more insight intothe evolution of plant domestication. Diverse taxa incommon taxonomic groups often share gene order overlarge chromosomal segments.Comparative QTL map-ping is possible because chromosomes of these differenttaxa can be aligned on the basis of common referenceloci. The grass species sorghum, rice and maize, wereeach independently domesticated ~10,000 years ago.Each species has been selected to have large seeds,daylength-insensitive flowering and reduced fruit shat-tering. A small number of QTL were located for thesetraits63. Interestingly, the approximate location of theQTL for each trait mapped to roughly correspondinglocations in each of the three species, despite 65 millionyears of reproductive isolation between these species63.This conservation of QTL location indicates that, 10,000years ago, ancient farmers across three continents mighthave been independently selecting many of the samegenes to obtain convergent phenotypes in each of thethree species63.

a b c

Figure 5 | The evolution of apical dominance in maize (Zea mays). a | The maize cropsthat are cultivated today are probably a domesticated form of the wild Mexican grass, teosinte.Note the bushy form of the teosinte, Zea mays ssp. mexicana, shown here. b | The single-stalkbranching pattern of wild-type maize (Inbred A158). c | A maize plant that is mutant for theteosinte branched 1 (tb1) gene. The tb1 locus is likely to have had an important role in theevolution of maize plant architecture. (Reprinted with permission from REF. 61 (1997) Macmillan Magazines Ltd.)



R E V I E W S

Box 3 | Proposed formation of cultivated polyploid cotton (Gossypium)

Most of the worlds cotton crop is made up oftwo tetraploid (2n = 4x = 52) species, Gossypiumhirsutum (Upland cotton) and Gossypiumbarbadense (Pima, Sea Island or Egyptiancotton)68. These species have been cultivated toproduce long, spinnable fibres on their seeds: in eachpanel, the seed, with fibre removed, is shown on theleft and the fibre from that one seed, if any, is shownon the right. In the figure, a | both tetraploid speciesare thought to have arisen by the hybridization of twodiploid ancestors: a maternal Old World diploidspecies (denoted the A genome) and a paternal NewWorld diploid species (called the D genome)68. So,the resulting tetraploid has an AD genome.

The possible D genome parents of the tetraploid aretwo, extant, neotropical species, Gossypium raimondiior a sister species of Gossypium gossypioides (2n = 26).The possible A genome ancestors are two extant OldWorld species, Gossypium arboreum or Gossypiumherbaceum (2n = 26). Interestingly, when one looks atthe most distinctive and important trait to a cottonfarmer, fibre, in both wild and domesticated cottonspecies, only the A genome diploid and the ADtetraploid taxa produce seeds that are covered in long,spinnable fibres (a)68. In Asia, a domesticated Agenome diploid is still bred and cultivated for its fibre68.However, although wild D genome diploid speciesproduce hairy seeds, none produce spinnable fibre andnone has ever been successfully domesticated for fibreproduction68. Of course, humans have successfullydomesticated and selected AD tetraploids for highyield and quality of fibre. Therefore, the interaction ofthe A and D genomes in the tetraploid domesticatedspecies produces higher quality and higher quantityfibre than is found in either of the diploid ancestors,even the currently domesticated A genome diploids.

In b, an F2

mapping population was created that wasderived from a cross of the two species of cultivatedAD tetraploids for which they had developed detailedgenetic maps116. These two species have very differentseed fibres, and quantitative trait loci (QTL) thatdistinguished the parental types for various fibrecharacteristics were sought68. Remarkably, most of theQTL that influenced fibre quality and yield werelocated on the portion of the genome contributed bythe D genome (hypothetical data illustrated)68.Remember that the D portion of the genome isderived from an ancestor that has no spinnable fibre.So, most of the genetic variation available forimprovement of cultivated cotton fibres apparentlycomes from the parent without fibre. Perhapsthousands of years of selection on A genome diploidsexhausted genetic variation for fibre quality, but thatselection was absent from the D genome diploidsbecause their seed fibres were not desirable to ancientfarmers. (Images courtesy of Andrew Paterson,Applied Genetic Technology Center, Departments ofCrop and Soil Sciences, Botany, and Genetics,University of Georgia, USA, and Thea Wilkins,Department of Agronomy and Range Science,University of California, Davis, USA.)

Gossypium herbaceum

Gossypium hirsutum

Gossypium raimondii

2n = 26

'A' diploid genome 'D' diploid genome

2n = 26

2n = 4x = 52'AD' tetraploid

X

a

Gossypium hirsutum Gossypium barbadense

A

AD tetraploid

Parents

F1

D

X

A

AD tetraploid

D

F2 X

QTL for fibre length

QTL for fibre thickness

QTL for cotton mass

QTL for number of fruits

b



R E V I E W S

Experimental and statistical concerns. QTL mapping isa statistical approach and evolutionary biologists mustbe aware of the inherent limitations and biases of thestatistical procedures themselves11,77. For example, QTLanalyses assume that the distribution of trait values arenormally distributed. Important experimental consider-ations that are involved in implementing these statisticaltests include the heritability of the trait being mapped,the precision with which the trait can be measured andthe size of the mapping population11. Shortcomings inany of these areas can undermine the accuracy andpower of the QTL analysis.

The limits of QTL detection are determined by sev-eral factors, including recombination, the number ofprogeny in the mapping population and the number ofmarkers70,71. QTL mapping always underestimates thenumber of genes that are involved in controlling a traitbecause only genes of sufficiently large phenotypic effectwill be detectable as QTL70,71,78,79. Imagine staring out atthe African savanna with very poor binoculars youwill see the elephants (QTL of large effect), but perhapsnot the gazelles (QTL of small effect). With a progenypopulation of fewer than 500 individuals, regardless ofmarker density, there is little statistical power to identifyQTL of small effect71.

William Beavis summarized the results of severalQTL mapping experiments on yield and height ofmaize, including replicate studies of the samecrosses70,71. Although the same QTL were detectedacross studies, some of those detected were unique toeach cross. Even the replicate studies did not detect thesame QTL.

Another form of sampling bias, the use of only afew divergent parental lines in a minimum number ofenvironmental conditions, can lead to the underesti-mation of the number of genes and their effects. Forexample, Paterson and colleagues did a mappingexperiment for tomato in three environments. Of the29 QTL detected, 4 were detected in all three environ-ments, 10 in two environments and 15 in only oneenvironment80. Clearly, the environment will be shownto have an important role in QTL mapping. At onelevel, the environment might complicate our efforts tomap QTL, but understanding the interaction of QTLwith the environment will be crucial to our under-standing of gene function and evolution.

It is also possible that the QTL of large phenotypiceffect that we see are an artefact of the strong directionalselection often used to create the phenotypically diver-gent parental lines that are used for mapping19. Strongselection can fix alleles that normally segregate in thebase population. In addition, artificial selection mightcreate repeated bottlenecks through which only a sam-ple of segregating alleles can pass. Only segregating alle-les can be detected. So, fewer QTL will be detected andthose that are eventually detected might explain aninflated portion of the phenotypic variance.

In addition to the estimation of the number of QTL,the magnitude of QTL effects might also be biased bysmall sample sizes70,71. In his study on QTL experimentsin maize, Beavis showed that in all studies, one or a few

Moving from QTL to genes. A QTL is almost never anactual genetic locus. A QTL is a chromosomal segment,potentially encompassing many hundreds of individualloci, most of which have nothing to do with the pheno-typic trait of interest. An actual locus that contributes toa phenotype is a veritable needle in a haystack of QTL.Conversely, a QTL might contain many genes that con-tribute to the phenotype of interest.

Although there have been examples of QTL map-ping that yield an actual locus, these examples arerare54,61. Remember that it took more than ten yearsto winnow a single fruit size QTL to the actual gene.There are 28 known fruit size QTL in tomato51.Although it is doubtful that it will take 270 moreyears to find all the genes that influence fruit size intomato, finding those genes will be time consumingand expensive. In the near future, only in thoseorganisms for which genetic information is abundantwill we be able to find the actual genes that underliethe phenotypes of interest.

Even in model organisms, the ability to move fromQTL to gene will not be easy. In A. thaliana, the estimat-ed genetic map is 586 centiMorgans (cM) and the phys-ical size is ~125 megabases. On average, there are 213kilobases of DNA and ~50 genes per cM in A. thaliana1,72. Even in the best QTL studies, many QTLare defined by markers that are more than 10 cM apart.Sorting through 500 genes requires time and money.Even if a genome project has identified each of the genesin that interval, proving that any particular gene isresponsible for variation in a trait of interest is not atrivial exercise.

Association studies hold some promise in assess-ing the correlations between specific genetic variants(usually SNPs) and trait differences on a populationlevel74,75. The most commonly used approach search-es for differences in allele frequency between individ-uals with a particular phenotype and unrelated con-trol individuals. However, many statistical caveatsaccompany such studies and they have, to date,been plagued by numerous spurious (false-positive)correlations74,76.

Along similar lines, a candidate gene approachmight help in linking QTL with particular genes. In thisapproach, a gene known to be in a particular pathway,or have a predicted function, can be related to genesalready known to have specific phenotypic effects, andwill be considered to be a gene correlated to a QTL. Asmany genes will probably be included within the chro-mosomal boundaries of the QTL, it will still be difficultto provide convincing molecular evidence, such as com-plementation, that a candidate gene is the locus thatcontributes to the trait under study.

Those biologists working with less geneticallyendowed organisms might be able to lever the geneticinformation from model organisms by taking advan-tage of homology. In this way, a reverse quantitativegenetics approach could be fruitful in that one couldask how much phenotypic variation in a non-modelorganism is explained by the homologue of a gene witha similar phenotype in a model organism.



R E V I E W S

LOD SCORE

(Base 10 logarithm of the odds,or log-odds.) A method ofhypothesis testing. Thelogarithm of the ratio betweenlikelihoods under the null andalternative hypotheses.

MONTE CARLO SIMULATION

The use of randomly generatedor sampled data and computersimulations to obtainapproximate solutions tocomplex mathematical andstatistical problems.

PERMUTATION TEST

A method of hypothesis testing.In these tests, an empiricaldistribution of a test statistic isobtained by permuting theoriginal sample many times.Each permuted sample isconsidered to be a sample of thepopulation under the nullhypothesis.

BAYESIAN APPROACH

An alternative statisticalmethod that allows the use ofprior information to evaluatethe posterior probabilities ofdifferent hypotheses.

SIMPLEX SEGREGATION

Segregation in polyploids.Segregation with no crossoversof the simplex genotype Aaaawould result in a gametic ratioof 1/2 Aa and 1/2 aa.

understanding of quantitative traits. Genomics willgreatly assist in providing numerous markers and morecomplete maps. Genomic techniques might also make itpossible to create larger progeny arrays as the cost ofgenotyping will probably decrease markedly.

Theoretically, this QTL mapping approach is asapplicable to animals, including humans, as it is toplants. However, there are some significant advan-tages to studying complex traits in plants. Plants areeasy to replicate and one can generate severalparental lines and large progeny arrays. The variablethat is often limiting in QTL studies is the cost ofgenerating and maintaining large numbers of proge-ny. For humans, the number of progeny is necessarilysmall. Generating inbred lines in plants is generallypossible. Because animals tend to be outcrossing,inbreeding can be a problem.

QTL mapping in populations or species of out-breeding organisms faces additional statistical andbiological challenges. In these cases, the parents usedin the mapping cross, either from controlled crossesor natural populations, are not necessarily fixed foralternate alleles, as is the case with inbred parentallines. The parents might be polymorphic for segre-gating alleles. Several experimental and statisticalsolutions to these challenges have been suggested foroutbred species8992. Statistical approaches usingpedigrees are being developed93,94 that should beapplicable to humans. In highly heterozygous organ-isms, QTL mapping can be done in the F

1generation

itself, on the basis of SIMPLEX SEGREGATION of polymor-phic markers95,96. In addition, new approaches haverecently been presented for QTL mapping in poly-ploids and even in hybrid zones97,98.

Finally, in a plant system, one can easily assessQTL in several realistic ecological conditions. Parentscan be taken out of the field and offspring can begrown back in the field. If our ultimate goal is tounderstand how genes form complex phenotypes, wemust come to realize that the environment has a cru-cial role. Replicated microarray studies that cansimultaneously assess genome-wide gene expressionand can be used on field experiments might eventual-ly be a tool that geneticists and ecologists find invalu-able. Understanding how the environment interactswith genes to yield phenotypes might be the mostsignificant challenge to all geneticists.

QTL of large effect were identified, along with severalQTL of small effect. This distribution was more skewedin experiments that used small numbers of progeny.The fewer the progeny, the higher were the estimatedeffects of the largest QTL identified.

Finally, the issue of significance testing is still incom-pletely resolved. The statistical tests for assessing if aQTL actually exists are many and not independent. So,QTL mapping will yield a significant number of falseQTL (ghost peaks). One commonly used solution tothis problem is to use a conservative threshold value toreduce the probability of false positives81,82. For humandata, it has been estimated that the threshold valueshould be a LOD SCORE of 3.3 (REF. 82). This value indicatesthat the probability that a QTL occurs in a particularinterval is over 1,000 times more likely than the nullhypothesis that no QTL exists in the interval. MONTECARLO SIMULATIONS8385 and PERMUTATION TESTS86,87 are twoother approaches that are used to explicitly determine ifa QTL is significant. BAYESIAN APPROACHES to QTL mappingare being introduced32,88.

Future challenges Given the caveats described above, knocking down thestraw man of quantitative genetics (many genes ofsmall, additive effect) might be more difficult thaninitial efforts have led us to believe. After all, howmuch phenotypic variation does a QTL have toexplain before we call it a QTL of large effect? Doesthe quantitative genetic model really predict that wewill find no single QTL that explains any significantamount of the total phenotypic variation? Given thenumerous caveats that apply to QTL mapping studies,the number of genes estimated by QTL mappingshould be viewed as a hypothesis of genetic architec-ture. Furthermore, it is crucial that evolutionary biol-ogists define their questions with the caveat that theymight never actually find genes. In agriculture, havinga QTL might be enough to serve in a marker-assistedselection programme. In how many evolutionarystudies will knowing only a relatively large chromoso-mal region be informative? The challenge for evolu-tionary biologists will be to think carefully about howunderstanding the genetic basis of complex traits willinform their studies, especially if those conclusionsrest on knowing the actual genes underlying the traitof interest. When is identifying a large chromosomalsegment interesting enough to justify an expensiveand time-consuming hunt for QTL?

Children of the cornAt some level, all geneticists, from molecular geneticiststo population geneticists, are interested in finding theconnection between the gene and the phenotype.Understanding this connection is most difficult for thecases of complex traits, such as most human diseasesand many examples of adaptive evolution. QTL map-ping holds some promise in helping us to make thisconnection. With the advent of genomics, geneticists seea path through the fog and there is growing awarenessthat an understanding of human disease will require an

Links

FURTHER INFORMATION Arabidopsis thaliana | rice |Mendelism | R. A. Fisher | Francis Galton | KarlPearson | Arabidopsis Biological Resource Centre(ABRC) | Cereon Genomics Arabidopsis SNPcollection | Nottingham Arabidopsis Stock Centre:Columbia Landsberg RI lines | RockefellerUniversitys collection of genetic analysis software |The Arabidopsis Information Resource (TAIR) | The Institute for Genomic Research | RodneyMauricios labENCYCLOPEDIA OF LIFE SCIENCES Adaptationgenetics | Quantitative genetics



R E V I E W S

1. Kaul, S. et al. Analysis of the genome sequence of theflowering plant Arabidopsis thaliana. Nature 408, 796815(2000).

2. Theologis, A. et al. Sequence and analysis of chromosome1 of the plant Arabidopsis thaliana. Nature 408, 816820(2000).

3. Lin, X. Y. et al. Sequence and analysis of chromosome 2 ofthe plant Arabidopsis thaliana. Nature 402, 761768(1999).

4. Salanoubat, M. et al. Sequence and analysis ofchromosome 3 of the plant Arabidopsis thaliana. Nature408, 820822 (2000).

5. Mayer, K. et al. Sequence and analysis of chromosome 4of the plant Arabidopsis thaliana. Nature 402, 769777(1999).

6. Tabata, S. et al. Sequence and analysis of chromosome 5of the plant Arabidopsis thaliana. Nature 408, 823826(2000).

7. Adam, D. Now for the hard ones. Nature 408, 792793(2000).

8. Bulmer, M. G. The Mathematical Theory of QuantitativeGenetics (Clarendon, Oxford, 1985).

9. Falconer, D. S. & Mackay, T. F. C. Introduction toQuantitative Genetics (AddisonWesleyLongman,Harlow, 1996).

10. Lynch, M. & Walsh, B. Genetics and the Analysis ofQuantitative Traits (Sinauer Associates, Sunderland,Massachusetts, 1997).

11. Tanksley, S. D. Mapping polygenes. Annu. Rev. Genet. 27,205233 (1993).

12. Liu, B.-H. Statistical Genomics: Linkage, Mapping andQTL Analysis (Boca Raton, Florida, USA, 1998).An excellent overview of QTL mapping techniques.

13. Burr, B. & Burr, F. A. Recombinant inbreds for molecularmapping in maize: theoretical and practical considerations.Trends Genet. 7, 5560 (1991).

14. Moreno-Gonzalez, J. Efficiency of generations forestimating marker-associated QTL effects by multipleregression. Genetics 135, 223231 (1993).

15. Sax, K. The association of size differences with seed-coatpattern and pigmentation in Phaseolus vulgaris. Genetics8, 552560 (1923).The first mapping of a QTL by association with aMendelian marker.

16. Lister, C. & Dean, C. Recombinant inbred lines formapping RFLP and phenotypic markers in Arabidopsisthaliana. Plant J. 4, 745750 (1993).

17. Cho, R. J. et al. Genome-wide mapping with biallelicmarkers in Arabidopsis thaliana. Nature Genet. 23,203207 (1999).

18. Barton, N. H. & Turelli, M. Evolutionary quantitativegenetics: how little do we know? Annu. Rev. Genet. 23,337370 (1989).

19. Lande, R. The response to selection on major and minormutations affecting a metrical trait. Heredity 50, 4765(1983).

20. Orr, H. A. & Coyne, J. A. The genetics of adaptation: areassessment. Am. Nat. 140, 725742 (1992).The landmark paper that exhumed the argument thatgenes of major effect could be important inadaptation.

21. Fisher, R. A. The Genetical Theory of Natural Selection(Dover, New York, 1958).

22. Alonso-Blanco, C., Blankestijn-de Vries, H., Hanhart, C. J.& Koornneef, M. Natural allelic variation at seed size loci inrelation to other life history traits of Arabidopsis thaliana.Proc. Natl Acad. Sci. USA 96, 47104717 (1999).

23. Stanton, M. L. Seed variation in wild radish: effect of seedsize on components of seedling and adult fitness. Ecology65, 11051112 (1984).

24. Fisher, R. A. The use of multiple measurements intaxonomic problems. Ann. Eugen. 7, 179188 (1936).

25. Juenger, T., Purugganan, M. D. & Mackay, T. F. C.Quantitative trait loci for floral morphology in Arabidopsisthaliana. Genetics 156, 13791392 (2000).

26. Kelly, M. G. & Levin, D. A. Directional selection on initialflowering date in Phlox drummondii (Polemoniaceae). Am.J. Bot. 87, 382391 (2000).

27. Kowalski, S. P., Lan, T. H., Feldmann, K. A. & Paterson, A.H. QTL mapping of naturally-occurring variation inflowering time of Arabidopsis thaliana. Mol. Gen. Genet.245, 548555 (1994).

28. Mitchell-Olds, T. Genetic constraints on life historyevolution quantitative trait loci influencing growth andflowering in Arabidopsis thaliana. Evolution 50, 140145(1996).

29. Kuittinen, H., Sillanp, M. J. & Savolainen, O. Geneticbasis of adaptation: flowering time in Arabidopsis thaliana.Theor. Appl. Genet. 95, 573583 (1997).

30. Byrne, M. et al. Identification and mode of action ofquantitative trait loci affecting seedling height and leaf areain Eucalyptus nitens. Theor. Appl. Genet. 94, 674681(1997).

31. Byrne, M., Murrell, J. C., Owen, J. V., Williams, E. R. &

Moran, G. F. Mapping of quantitative trait loci influencingfrost tolerance in Eucalyptus nitens. Theor. Appl. Genet.95, 975979 (1997).

32. Hurme, P., Silanp, M. J., Arjas, E., Repo, T. &Savolainen, O. Genetic basis of climatic adaptation inScots pine by Bayesian quantitative trait locus analysis.Genetics 156, 13091322 (2000).

33. Bradshaw, H. D. Jr & Stettler, R. Molecular genetics ofgrowth and development in Populus. IV. Mapping QTLswith large effects on growth, form, and phenology traits ina forest tree. Genetics 139, 963973 (1995).

34. Frewen, B. E. et al. Quantitative trait loci and candidategene mapping of bud set and bud flush in Populus.Genetics 154, 837845 (2000).

35. Lewontin, R. C. The Genetic Basis of Evolutionary Change(Columbia University Press, New York, 1974).

36. Gottlieb, L. D. Genetics and morphological evolution inplants. Am. Nat. 123, 681709 (1984).An early, but underappreciated, paper that raised theissue that genes of major effect could be importantin speciation.

37. Coyne, J. A. & Lande, R. The genetic basis of speciesdifferences in plants. Am. Nat. 126, 141145 (1985).

38. Bradshaw, H. D. Jr, Wilbert, S. M., Otto, K. G. &Schemske, D. W. Genetic mapping of floral traitsassociated with reproductive isolation in monkeyflowers(Mimulus). Nature 376, 762765 (1995).One of the first attempts in natural populations toidentify genes that are important in speciation byQTL mapping.

39. Bradshaw, H. D. Jr, Otto, K. G., Frewen, B. E., McKay, J.K. & Schemske, D. W. Quantitative trait loci affectingdifferences in floral morphology between two species ofmonkeyflower (Mimulus). Genetics 149, 367382 (1998).

40. Schemske, D. W. & Bradshaw, H. D. Jr. Pollinatorpreference and the evolution of floral traits inmonkeyflowers (Mimulus). Proc. Natl Acad. Sci. USA 96,1191011915 (1999).

41. Kelly, A. J. & Willis, J. H. Polymorphic microsatellite loci inMimulus guttatus and related species. Mol. Ecol. 7,769774 (1998).

42. Cruzan, M. B. & Arnold, M. L. Ecological and geneticassociations in an Iris hybrid zone. Evolution 47,14321445 (1993).

43. Arnold, M. L. Andersons paradigm: Louisiana irises andthe study of evolutionary phenomena. Mol. Ecol. 9,16871698 (2000).

44. Hodges, S. A. & Arnold, M. L. Floral and ecologicalisolation between Aquilegia formosa and Aquilegiapubescens. Proc. Natl Acad. Sci. USA 91, 24932496(1994).

45. Rieseberg, L. H., Sinervo, B., Linder, C. R., Ungerer, M. C.& Arias, D. M. Role of gene interactions in hydridspeciation: evidence from ancient and experimentalhybrids. Science 272, 741745 (1996).

46. Kim, S.-C. & Rieseberg, L. H. Genetic architecture ofspecies differences in annual sunflowers: implications foradaptive trait introgression. Genetics 153, 965977(1999).

47. Rieseberg, L. H., Whitton, J. & Gardner, K. Hybrid zonesand the genetic architecture of a barrier to gene flowbetween two sunflower species. Genetics 152, 713977(1999).

48. Hodges, S. A., Burke, J. M. & Arnold, M. L. Naturalformation of Iris hybrids: experimental evidence on theestablishment of hybrid zones. Evolution 50, 25042509(1996).

49. Stuber, C. W. Mapping and manipulating quantitative traitsin maize. Trends Genet. 11, 477481 (1995).

50. Young, N. D. A cautiously optimistic vision for marker-assisted breeding. Mol. Breed. 5, 505510 (1999).

51. Grandillo, S., Ku, H. M. & Tanksley, S. D. Identifying the lociresponsible for natural variation in fruit size and shape intomato. Theor. Appl. Genet. 99, 978987 (1999).

52. Alpert, K. B., Grandillo, S. & Tanksley, S. D. fw-2.2 amajor QTL controlling fruit weight is common to both red-fruited and green-fruited tomato species. Theor. Appl.Genet. 91, 9941000 (1995).

53. Alpert, K. B. & Tanksley, S. D. High-resolution mappingand isolation of a yeast artificial chromosome contigcontaining fw2.2: a major fruit weight quantitative traitlocus in tomato. Proc. Natl Acad. Sci. USA 93,1550315507 (1996).

54. Frary, A. et al. fw2.2: A quantitative trait locus key to theevolution of tomato fruit size. Science 289, 8588 (2000).A landmark in QTL analysis: the first molecularcharacterization of a locus that was originallyidentified entirely by QTL mapping. References5153 are the prologue to this milestone.

55. Doebley, J. & Stec, A. Genetic analysis of themorphological differences between maize and teosinte.Genetics 129, 285295 (1991).

56. Doebley, J. Mapping the genes that made maize. TrendsGenet. 8, 302307 (1992).

57. Doebley, J. & Stec, A. Inheritance of the morphologicaldifferences between maize and teosinte: comparison ofresults for two F2 populations. Genetics 134, 559570(1993).

58. Doebley, J., Stec, A. & Gustus, C. Teosinte branched 1and the origin of maize: evidence for epistasis and theevolution of dominance. Genetics 141, 333346 (1995).

59. White, S. & Doebley, J. Of genes and genomes and theorigin of maize. Trends Genet. 14, 327332 (1998).

60. Dorweiler, J., Stec, A., Kermicle, J. & Doebley, J. Teosinteglume architecture 1: a genetic locus controlling a key stepin maize evolution. Science 262, 233235 (1993).

61. Doebley, J., Stec, A. & Hubbard, L. The evolution of apicaldominance in maize. Nature 386, 485488 (1997).

62. Wang, R. L., Stec, A., Hey, J., Lukens, L. & Doebley, J. Thelimits of selection during maize domestication. Nature 398,236239 (1999).References 5562 chronicle a decade of work onunderstanding the genetics of the domestication ofmodern maize.

63. Paterson, A. H. et al. Convergent domestication of cerealcrops by independent mutations at corresponding geneticloci. Science 269, 17141718 (1995).An excellent example of the power of comparativeQTL mapping.

64. Ramsey, J. & Schemske, D. W. Pathways, mechanisms,and rates of polyploid formation in flowering plants. Annu.Rev. Ecol. Syst. 29, 467501 (1998).

65. Soltis, P. S. & Soltis, D. E. The role of genetic and genomicattributes in the success of polyploids. Proc. Natl Acad.Sci. USA 97, 70517057 (2000).

66. Hilu, K. W. Polyploidy and the evolution of domesticatedplants. Am. J. Bot. 80, 14941499 (1993).

67. Paterson, A. H. et al. Comparative genomics of plantchromosomes. Plant Cell 12, 15231539 (2000).

68. Jiang, C. X., Wright, R. J., El-Zik, K. M. & Paterson, A. H.Polyploid formation created unique avenues for responseto selection in Gossypium (cotton). Proc. Natl Acad. Sci.USA 95, 44194424 (1998).

69. Wright, R. J., Thaxton, P. M., El-Zik, K. M. & Paterson, A.H. D-subgenome bias of Xcm resistance genes intetraploid Gossypium (cotton) suggests that polyploidformation has created novel avenues for evolution.Genetics 149, 19871996 (1998).

70. Beavis, W. D. The power and deceit of QTL experiments:lessons from comparative QTL studies. Proc. Corn andSorghum Industry Res. Conf., Am. Seed Trade Assoc.,Washington DC, 255266 (1994).

71. Beavis, W. D. in Molecular Dissection of Complex Traits(ed. Paterson, A. H.) 145162 (CRC, Boca Raton, Florida,1998).References 70 and 71 provided the first and mostinfluential caveats for the use of QTL analysis in bothagriculture and evolutionary biology.

72. Copenhaver, G. P., Browne, W. E. & Preuss, D. Assayinggenome-wide recombination and centromere functionswith Arabidopsis tetrads. Proc. Natl Acad. Sci. USA 95,247252 (1998).

73. Begun, D. J. & Aquadro, C. F. Levels of naturally occurringDNA polymorphism correlate with recombination rates inD. melanogaster. Nature 356, 519520 (1992).

74. Risch, N. J. Searching for genetic determinants in the newmillennium. Nature 405, 847856 (2000).

75. Weiss, K. M. & Terwilliger, J. D. How many diseases does ittake to map a gene with SNPs? Nature Genet. 26,151157 (2000).

76. Cardon, L. R. & Bell, J. I. Association study designs forcomplex diseases. Nature Rev. Genet. 2, 9199 (2001).

77. Doerge, R. W., Zeng, Z. B. & Weir, B. S. Statistical issuesin the search for genes affecting quantitative traits inexperimental populations. Stat. Sci. 12, 195219 (1997).An excellent overview of the statistical issuesinvolved in QTL analysis.

78. Melchinger, A. E., Utz, H. F. & Schon, C. C. Quantitativetrait locus (QTL) mapping using different testers andindependent population samples in maize reveals lowpower of QTL detection and large bias in estimates of QTLeffects. Genetics 149, 383403 (1998).

79. Otto, S. P. & Jones, C. D. Detecting the undetected:estimating the total number of loci underlying a quantitativetrait. Genetics 156, 20932107 (2000).

80. Paterson, A. H. et al. Mendelian factors underlyingquantitative traits in tomato: comparison across species,generations, and environments. Genetics 127, 181197(1991).

81. Jansen, R. C. Interval mapping of multiple quantitative traitloci. Genetics 135, 205211 (1993).

82. Lander, E. & Kruglyak, L. Genetic dissection of complextraits: guidelines for interpreting and reporting linkageresults. Nature Genet. 11, 241247 (1995).

83. Lander, E. S. & Botstein, D. Mapping Mendelian factorsunderlying quantitative traits using RFLP linkage maps.Genetics 121, 185199 (1989).In this paper, interval mapping with molecular



R E V I E W S

markers to map QTL was first proposed for speciesin which many morphological markers wereunavailable. A maximum-likelihood statisticalapproach for QTL mapping was also developed.

84. Knott, S. A. & Haley, C. S. Maximum likelihood mapping ofquantitative trait loci using full-sib families. Genetics 132,12111222 (1992).

85. Zeng, Z.-B. Precision mapping of quantitative trait loci.Genetics 136, 14571468 (1994).

86. Churchill, G. A. & Doerge, R. W. Empirical threshold valuesfor quantitative trait mapping. Genetics 138, 963971(1994).

87. Doerge, R. W. & Churchill, G. A. Permutation tests formultiple loci affecting a quantitative character. Genetics142, 285294 (1996).

88. Sillanp, M. J. & Arjas, E. Bayesian mapping of multiplequantitative trait loci from incomplete inbred line crossdata. Genetics 148, 13731388 (1998).

89. Haley, C. S., Knott, S. A. & Elsen, J.-M. Mappingquantitative trait loci in crosses between outbred linesusing least squares. Genetics 136, 11951207 (1994).

90. Hoeschele, I., Uimari, P., Grignola, F. E., Zhang, Q. & Gage,K. M. Advances in statistical methods to map quantitativetrait loci in outbred populations. Genetics 147, 14451457(1997).

91. Sillanp, M. J. & Arjas, E. Bayesian mapping of multiplequantitative trait loci from incomplete outbred offspringdata. Genetics 151, 16051619 (1999).

92. Prez-Enciso, M. & Varona, L. Quantitative trait locimapping in F2 crosses between outbred lines. Genetics155, 391405 (2000).

93. George, A. W., Visscher, P. M. & Haley, C. S. Mappingquantitative trait loci in complex pedigrees: a two-stepvariance component approach. Genetics 156, 20812092(2000).

94. Hoeschele, I. in Handbook of Statistical Genetics (edsBalding, D., Bishop, M. & Cannings, C.) 599644 (John

Wiley & Sons Ltd, London, 2001).95. Liu, S.-C., Lin, Y.-R., Irvine, J. E. & Paterson, A. H. in

Molecular Dissection of Complex Traits (ed. Paterson, A.H.) 95101 (CRC, Boca Raton, Florida, 1998).

96. Ming, R. et al. Detailed alignment of Saccharum andSorghum chromosomes: comparative organization ofclosely related diploid and polyploid genomes. Genetics150, 16631682 (1998).

97. Doerge, R. W. & Craig, B. A. Model selection forquantitative trait locus analysis in polyploids. Proc. NatlAcad. Sci. USA 97, 79517956 (2000).

98. Rieseberg, L. H. & Buerkle, C. A. Genetic mapping inhybrid zones. Am. Nat. (in the press).

99. Provine, W. B. The Origins of Theoretical PopulationGenetics (Chicago Univ. Press, Illinois, 1971).A superb history of the early conflict between theMendelians and the biometricians, and itsresolution.

100. East, E. M. A Mendelian interpretation of variation that isapparently continuous. Am. Nat. 44, 6582 (1910).

101. Fisher, R. A. The correlation between relatives on thesupposition of Mendelian inheritance. Trans. R. Soc.Edinb. 52, 399433 (1918).

102. Lande, R. The maintenance of genetic variability bymutation in a polygenic character with linked loci. Genet.Res. 26, 221235 (1976).

103. Lande, R. Quantitative genetic analysis of multivariateevolution, applied to brain:body size allometry. Evolution33, 402416 (1979).

104. Lande, R. & Arnold, S. J. The measurement of selection oncorrelated characters. Evolution 37, 12101226 (1983).

105. Via, S. & Lande, R. Genotypeenvironment interaction andthe evolution of phenotypic plasticity. Evolution 39,505522 (1985).

106. Turelli, M. & Barton, N. H. Dynamics of polygeniccharacters under selection. Theor. Popul. Biol. 38, 157 (1990).

107. Barton, N. H. & Turelli, M. Natural and sexual selection onmany loci. Genetics 127, 229255 (1991).

108. Turelli, M. & Barton, N. H. Genetic and statistical analysesof strong selection on polygenic traits: what, me normal?Genetics 138, 913941 (1994).

109. Robertson, A. in Heritage from Mendel (ed. Brink, A.)265280 (Wisconsin Univ. Press, Madison, Wisconsin,1967).

110. Haley, C. S. & Knott, S. A. A simple regression method formapping quantitative trait loci in line crosses using flankingmarkers. Heredity 69, 315324 (1992).

111. Thoday, J. M. Location of polygenes. Nature 191,368370 (1961).

112. Zeng, Z.-B. Theoretical basis for separation of multiplelinked gene effects in mapping quantitative trait loci. Proc.Natl Acad. Sci. USA 90, 1097210976 (1993).

113. Kao, C. H., Zeng, Z.-B. & Teasdale, R. D. Multiple intervalmapping for quantitative trait loci. Genetics 152,12031216 (1999).

114. Zeng, Z.-B., Kao, C. H. & Basten, C. J. Estimating thegenetic architecture of quantitative traits. Genet. Res. 74,279289 (1999).

115. Basten, C. J., Weir, B. S. & Zeng, Z.-B. QTL Cartographer:A Reference Manual and Tutorial for QTL Mapping(Department of Statistics, North Carolina Univ. Press,Raleigh, North Carolina, 1995).

116. Reinisch, A. J. et al. A detailed RFLP map of cotton,Gossypium hirsutum x Gossypium barbadense:chromosome organization and evolution in a disomicpolyploid genome. Genetics 138, 829847 (1994).

AcknowledgementsI thank M. Arnold, R. Baucom, A. Bouck, C. Goodwillie, A. Johnson,A. Paterson, L. Rieseberg and J. Willis for unpublished material,helpful discussions and constructive comments on the manuscript.


Mauricio NRG 2001

Documents

Transcript of Mauricio NRG 2001