330

Opinion

TRENDS in Plant Science

Vol.8 No.7 July 2003

Breeding by Design
Johan D. Peleman and Jeroen Rouppe van der Voort
Keygene N.V. Business Unit Genetics, PO Box 216, 6700 AE Wageningen, The Netherlands

Breeding by Designe is a concept that aims to control all allelic variation for all genes of agronomic importance. This concept can be achieved through a combination of precise genetic mapping, high-resolution chromosome haplotyping and extensive phenotyping. Thanks to marker technology, software tools and the know-how available today, this goal can now be achieved. Depending on the crop-specic generation time, controlled marker-assisted selection strategies could lead to the production of superior varieties within ve to ten years. The potential value of genetic markers, linkage maps and indirect selection in plant breeding has been known for more than 80 years. However, it was not until the advent of DNA marker technology in the 1980s, that a large enough number of environmentally insensitive genetic markers could be generated to follow the inheritance of important agronomic traits adequately. Since then, DNA marker technology has dramatically enhanced the efciency of plant breeding. In the past decade, many breeding companies and research institutes have, to varying degrees, started using molecular markers to increase the effectiveness of selection in breeding and to shorten the development time of varieties signicantly [1,2]. Currently, advances in automated technology have led to the development of a new approach in marker-assisted breeding. The advances in applied genomics and the possibility of generating large-scale marker data sets provide us with the tools to determine the genetic basis for all traits of agronomic importance. In addition, methods for assessing the allelic variation at these agronomically important loci are now available. This combined knowledge will eventually allow the breeder to combine the most favourable alleles at all these loci in a controlled manner, leading to superior varieties. The Breeding by Design strategy is an extension of the different levels of marker applications that are currently used in breeding. These levels will be discussed briey before addressing the Breeding by Design concept in more detail. Marker-assisted selection Marker-assisted selection (MAS) using DNA markers instead of phenotypic assays reduces the cost and increases the precision and efciency of subsequent selection steps applied in breeding [1 3]. A successful example of this strategy is marker-assisted backcrossing (MABC), which is applied in a large variety of crops. In MABC,
Corresponding author: Johan D. Peleman ( johan.peleman@keygene.com).

mapped markers are used to select backcross progeny with the highest percentages of recurrent parent genome together with a minimum number of donor segments. MABC programs reduce the number of generations needed for recovery of the recurrent genome from eight to three in maize [4]. Marker-assisted breeding Where breeding goals cannot be achieved using traditional approaches, there is now considerable scope for using molecular markers to develop new varieties. DNA markers can add signicantly more value than just improving the speed, cost or quality of existing breeding programs. Here, the limitation is not the available technology, but rather the challenge facing the (molecular) breeder to nd creative approaches for developing new products. By understanding the genetic basis of complex traits (or obstacles), it becomes possible to control them. Several examples of successful marker-assisted breeding (MAB) approaches are now emerging: for example, a method for PYRAMIDING (see Glossary) dominant resistance genes [5] and the breeding of an aphid (Nasonovia ribisnigri)resistant lettuce variety by introducing a recessive resistance gene surrounded by severe LINKAGE DRAG [6]. Breeding by Design The examples above show that applying markers in breeding not only improves existing selection processes but can aid in creating novel varieties bearing new characteristics of agronomic importance. Building on these capabilities, by understanding the genetic basis of all agronomically important characters and the allelic variation at those loci, the breeder would be able to design superior genotypes in silico. We have called this concept Breeding by Designe. This goal can be reached by following a three-step approach. (1) Mapping loci involved in all agronomically relevant traits To elucidate the genetic basis of agronomically important traits, mapping populations are needed in which those

Glossary
Heterosis: Hybrid vigour. Linkage drag: Co-inheritance of undesirable trait(s) with a gene of interest. Typically, this is a phenomenon that can be observed during backcross breeding using an exotic donor parent. Pyramiding: The accumulation of several desirable traits in the same genotype through backcross breeding.

http://plants.trends.com 1360-1385/03/$ - see front matter q 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S1360-1385(03)00134-1

Opinion

TRENDS in Plant Science

Vol.8 No.7 July 2003

331

traits are segregating. To map all traits that are relevant in breeding a crop, we prefer to use introgression line (IL) libraries (Box 1). Considerable progress has been made in constructing IL libraries for several different crops over the past ve years [7]. IL libraries are extremely powerful tools in the quest to map the loci underlying all agronomically important traits. The key advantage of IL libraries is in reducing the complexity of polygenic traits by separating them into a set of monogenic loci (Box 1). To identify the allelic variation at each locus of interest by using chromosome haplotyping (see further), it is essential to determine a precise position for the loci of

interest. IL libraries also provide perfect starting material for this purpose: each line containing a locus of interest can be backcrossed to the recurrent parent (and, if necessary, selfed) to create a large segregating population. This population can be used to identify recombinants within the introgression segment using anking markers. Phenotyping these recombinants enables the locus to be mapped at high resolution. Another powerful ne-mapping approach is provided by the wealth of sequence information available from model plant species combined with the rapidly expanding understanding of gene function. This knowledge can be linked to

Box 1: Mapping advantages of introgression line libraries

An introgression line (IL) library consists of a series of lines harbouring a single homozygous donor segment introgressed into a uniform, cultivated background (Fig. I) (reviewed in [7]). The advantages of IL libraries compared with other mapping approaches are: IL libraries consist of homozygous immortal lines and therefore can be phenotyped repeatedly and used for the simultaneous mapping of many traits. IL libraries contain homogenous genetic backgrounds, only differing from one another by the introgressed donor segment. Thus, epistatic effects from the donor parent are eliminated. Quantitative trait loci (QTL) are dissected into separate monogenic components, which increases the reliability of measuring phenotypic traits ILs containing interesting QTL can be backcrossed to various lines to investigate interactive effects. Although dependent on the resolution of the IL library (i.e. the average introgression segment size), QTLs are typically mapped into smaller intervals than by classical QTL mapping. IL libraries provide optimal starting material for the ne-mapping of the mapped loci and for generating improved breeding lines. A secondary bonus of using IL libraries is that often new exotic alleles can be found that have a positive effect in the culture crop germplasm. To study interaction (epistasis) between non-allelic genes, whereby the expression of one gene interferes with the expression of the other gene, reciprocal IL libraries can be constructed. In such a case, IL libraries from line A into B and vice versa are constructed. By doing so, phenotypes that cannot be detected because they are mediated by interacting loci in the A B library will be measured as a knocked-out phenotype in the B A library. Subsequently, crosses between individual introgression lines each bearing one of the interacting alleles can be made to investigate the extent of the interaction [17]. To map loci contributing to heterosis, the IL library can be crossed to a tester parent. This will create an F1 IL library in which each introgression segment is present in the heterozygous state. This F1 IL library is then phenotyped to detect heterotic effects caused by specic introgression segments.

BC1

BC1 selection

BC2 selection

BC3 selection

BC3S1 selection

TRENDS in Plant Science

Fig. I. Example of an introgression line (IL) library construction process in tomato (12 chromosomes each separated by a grey bar) using Marker-Assisted BackCrossing. Each horizontal bar represents an individual [best visible in the BC1 (BackCross generation 1) selection]. Homozygous recurrent parent segments are shown in red. The homozygous donor segments are indicated in blue. Green bars represent heterozygous segments. The construction and use of IL libraries is extensively reviewed in [7].
http://plants.trends.com

332

Opinion

TRENDS in Plant Science

Vol.8 No.7 July 2003

commercial species by exploiting synteny between both species, thereby opening the possibility for developing candidate-gene mapping approaches and causal SNP markers [8,9]. In our view, the candidate gene approach can provide a shortcut in the ne mapping or cloning of loci that are roughly mapped using an IL library. Linkage Disequilibrium Mapping (LD Mapping) provides an alternative but more risky approach to ne-map loci of interest. LD mapping relies on associations between the phenotype and neutral polymorphisms located near the trait locus [10]. Because of the complexity of this approach and the associated risks, our preference is to use targeted LD Mapping. Once the approximate position of a locus is known, LD mapping can be applied using markers (Box 2) in the area of that locus, leading to the identication of markers or haplotypes that show a strong association with the phenotype under investigation, thereby ne mapping the locus of interest [11]. (2) Assessment of the allelic variation at those loci Although several strategies based on different population structures exist to unravel the genetic basis of complex traits, it is still impossible to predict the phenotypes generated by those genes in the germplasm. This is because of the allelic variation that exists in the germplasm at those loci. Typically, in a segregating population (F2, BackCross, recombinant inbred lines, double haploids, IL library) only two alleles are segregating per locus. Mapping the genes involved in a desired phenotype only enables the prediction of the phenotype within the same population or within populations in which the same alleles are segregating at that locus. To obtain broad predictive power, we need to be able to distinguish all alleles at the locus of interest and to assign phenotypic values to the different alleles. As mentioned above, a potential short cut to mapping genes and allelic variation simultaneously could be achieved using association mapping with single markers [10]. However, a better method for assessing allelic variation at loci of interest is by using marker haplotypes. Using a set of tightly linked markers at a locus, the combined scores of all markers in principle can distinguish all the different alleles that occur at that locus in a set of accessions (Box 2). For example, a high correlation was found between resistance phenotypes and marker haplotypes derived for the largest resistance gene cluster in lettuce [12]. By extending this technique to the whole genome, complete chromosome haplotypes can be generated (Fig. 1). Assuming a high level of saturation with markers, these chromosome haplotypes enable us to determine the allelic variation at any position in the genome. After ne mapping the genes of agronomic importance, for example, by using IL libraries combined with targeted LD mapping, these chromosome haplotypes can be used to assess the allelic variation at those loci. Once the allelic variation at the loci of interest is identied, it is essential to attribute phenotypic values to the different alleles. For this purpose, the inbred lines bearing different alleles at the loci of interest need to be
http://plants.trends.com

Box 2: Predicting trait values by use of haplotypes

The predictive value of a marker for a trait of interest in a broad germplasm is determined by the mode of inheritance of the trait, the recombination distance between the marker and the trait locus, and the mutation rate of the marker. Instead of single-marker-based tests, combinations of linked marker alleles at the same chromosome (haplotypes) can be used to predict phenotypic trait values in a panel of lines. Haplotypes of multiple markers have a higher information content compared with single markers. Therefore, an increased predictive power is achieved by using haplotypes. We have successfully applied this approach in different crop species by using AFLP markers to identify and predict different alleles of agronomically important loci. Figure I shows an example of three allelic variants of a hypothetical gene. Three tightly linked single nucleotide polymorphism (SNP) markers were identied in a segregating population derived from parental lines G1 and G2. However, once the linked markers are used in a germplasm panel (G1 G6), typically the association between (most of) the markers is lost because of the independent origin of the SNPs. In this example, the haplotypes composed of the three SNPs are required to discriminate the three different alleles. Haplotype-based analyses of SNPs are the focus of intensive research because of the effectiveness of using haplotypes in (ne) mapping of complex traits and scanning for allelic variants (e.g. [18,19]).

SNP1 G1 G2 +

SNP2 +

SNP3 +

G3

G4

G5

G6

+
TRENDS in Plant Science

Fig. I. Predictive value of haplotypes. G1 and G2 represent parents of a segregating population, G3 G6 represent germplasm lines. Four gene alleles are distinguished: orange, white, green and blue. For clarity, the genotypes of three SNP (single nucleotide polymorphism) markers are indicated as /2 scores. The white allele is indicative of two different haplotypes whereas the orange, green and blue alleles correspond to unique haplotypes. In this example, lines G3 and G4 have different haplotypes while bearing the same gene allele owing to a point mutation at the SNP2 locus.

thoroughly phenotyped. In the case of polygenic traits, it is possible to pre-select those lines that bear different combinations of alleles at the different loci for phenotyping. Currently, we are developing algorithms that can combine those data with the phenotypic data on the lines to determine the contributing value of each allele to the phenotype. In our view, the combination of mapping loci using simple mapping populations, in combination with allele

Opinion

TRENDS in Plant Science

Vol.8 No.7 July 2003

333

TRENDS in Plant Science

Fig. 1. Chromosome haplotypes. Chromosome haplotypes of a set of maize lines for one chromosome. Horizontal lines represent the same chromosome for a panel of lines. Identical stretches of marker scores (haplotypes) are indicated by the same colour. Box 1 indicates a region with depleted diversity whereas box 2 shows a highly variable region.

assessment using chromosome haplotyping of lines, in combination with phenotyping, provides the most powerful route for optimally exploiting the germplasm of a species.

(3) Breeding by Design Eventually, knowledge of the map positions of all loci of agronomic interest, the allelic variation at those loci, and their contribution to the phenotype should enable

A
1 2 3 4 5

2 x 3 F2 selection Intermediate line I x 1 BC selection Intermediate line II x 5

BC selection Ideal line R Sugars virus R insect Fruit mass I Fruit shape Plant height Fruit color Cold tolerance Flowering time I Drought resistance Fruit mass II

TRENDS in Plant Science

Fig. 2. The principle of Breeding by Design. Subsequent selngs (F2) and BackCross (BC) selections using markers lead to the desired superior elite line genotype. Three chromosomes, A, B and C, of ve parental lines, 1 5 are shown side by side. Specic recombination points are selected on chromosomes A and B whereas chromosome C is selected from parental line 1. Dotted lines indicate marker positions used to select for the desired recombinants. Below the desired genome composition of the ideal line, hypothetical resistance (R) and quality traits are mentioned.
http://plants.trends.com

334

Opinion

TRENDS in Plant Science

Vol.8 No.7 July 2003

the breeder to design superior genotypes comprising a combination of favourable alleles at all loci. Because the positions of all loci of importance are mapped precisely, recombination events can be accurately selected using anking markers to collate the different favourable alleles next to each other (Fig. 2). Software tools should enable us to determine the optimal route for generating those mosaic genotypes by crossing lines and using markers to select for the specic recombinants that will eventually combine all those alleles. Because this is a precisely dened process, selection by phenotyping can be omitted. Only the eventually obtained superior varieties will be evaluated for eld performance. Discussion The above-described approach assumes several prerequisites that require more discussion. (1) Extremely saturated marker maps must be available to enable the generation of high-resolution chromosome haplotypes. Preferentially, a few markers are needed per window of LD to ensure reliable high-resolution chromosome haplotypes. Obviously, the extent of LD is strongly dependent on several factors: among others, the crop species, the germplasm selection and the genome region of interest [10]. Taking the rice genome as an example: assuming that the average size of an LD window in rice is 100 kb and the rice genome comprises 450 megabases, this means that 10 000 20 000 mapped markers need to be scored on a panel of inbred lines. This is feasible using the currently available automation tools. Using a multiplex PCR marker system such as AFLPw [13,14], generating 20 100 markers per PCR, such a task can be performed within a year with a few well-trained people and the appropriate equipment. High resolution mapping of AFLP markers can be performed using physical mapping strategies such as Happy Mapping and Radiation Hybrid Mapping [15] or applying BAC pooling strategies [16]. (2) The described approach requires extensive phenotyping of all agronomic traits, both the mapping populations and the inbred lines that are used for chromosome haplotyping and allele assessment. Moreover, to map loci involved in HETEROSIS , hybrids derived from selected combinations of haplotyped lines must also be phenotyped. The most important factor determining success is the precision of the phenotyping. Dissecting the phenotypes into components helps in objectively measuring phenotypes: for example, a complex trait such as taste can be dissected into sugar, acid and volatile avour molecules, which can be measured separately using biochemical analysis equipment such as gas chomatographers. Extending this approach to all traits of agronomic importance will require great organizational skills and will be a major effort. In our view, the phenotyping is the most crucial and challenging factor in achieving the goals set by Breeding by Design. Breeding by Design involves the integrative, complementary application of technological tools and the materials currently available to develop superior varieties.

During this process, an enormous resource of knowledge is generated that should enable breeders to deploy more rational and rened breeding strategies in the near future. The recent developments in technology and statistical methodology have now brought this strategy within reach. The optimal exploitation of the naturally available genetic resources should create unsurpassed possibilities to generate new traits and crop performance. In our view, Breeding by Design has as much crop improvement potential as GMO strategies while requiring less investment and without challenging public acceptance.
Acknowledgements
We gratefully acknowledge Maarten Koornneef (Wageningen University and Research Center), Jonathan Crouch (ICRISAT), Tom Gerats (University of Nijmegen) and the comments of two anonymous reviewers for their contributions and discussions during the development of this paper. The AFLPw technology is covered by patents and/or patent applications of Keygene N.V.; AFLPw is a registered trademark of Keygene N.V.; A trademark registration for Breeding by Design has been led by Keygene N.V.

References
1 Ribaut, J.M. and Hoisington, D.A. (1998) Marker-assisted selection: new tools and strategies. Trends Plant Sci. 3, 236 239 2 Young, N.D. (1999) A cautiously optimistic vision for marker assisted breeding. Mol. Breed. 5, 505 510 3 Tanksley, S.D. et al. (1989) RFLP mapping in plant breeding: new tools for an old science. Biotechnology 7, 257 264 4 Frisch, M. et al. (1999) Comparison of selection strategies for markerassisted backcrossing of a gene. Crop Sci. 39, 1295 1301 5 Lambalk, J.J.M. et al. (1999) Method for obtaining a plant with a lasting resistance to a pathogen. International application published under the patent cooperation treaty (PCT) No. WO 00/63432 6 Jansen, J.P.A. (1996) Aphid resistance in composites. International application published under the patent cooperation treaty (PCT) No. WO 97/46080 7 Zamir, D. (2001) Improving plant breeding with exotic genetic libraries. Nat. Rev. Genet. 2, 983 989 8 Fulton, T.M. et al. (2002) Identication, analysis, and utilization of conserved ortholog set markers for comparative genomics in higher plants. Plant Cell 14, 1457 1467 9 Oh, K. et al. (2002) Fine mapping in tomato using microsynteny with Arabidopsis genome: the Diageotropica (Dgt) locus. Genome Biol. 3, 0049.I 0049.II 10 Weir, B.S. (1996) Genetic Data Analysis II, Sinauer 11 Thornsberry, J.M. et al. (2001) Dwarf8 polymorphisms associate with variation in owering. Nat. Genet. 28, 286 289 12 Sicard, D. et al. (1999) Molecular diversity at the major cluster of disease resistance genes in cultivated and wild Lactuca spp. Theor. Appl. Genet. 99, 405 418 13 Vos, P. et al. (1995) AFLP, a new technique for DNA ngerprinting. Nucleic Acids Res. 23, 4407 4414 14 Zabeau, M. and Vos, P. (1993), Selective restriction fragment amplication: a general method for DNA ngerprinting, European Patent 0 534 858 15 Waugh, R. et al. (2002) Physical education new technologies for mapping plant genomes. Trends Plant Sci. 7, 521 523 16 Klein, P.E. et al. (2000) A high-throughput AFLP-based method for constructing integrated genetic and physical maps: progress toward a sorghum genome map. Genome Res. 10, 789 807 17 Eshed, Y. and Zamir, D. (1996) Less than additive epistatic interactions of QTL in tomato. Genetics 143, 1807 18 Johnson et al. (2001) Haplotype tagging for the identication of common disease genes. Nat. Genet. 29, 233 237 19 Cardon, L.R. and Abecasis, G.R. (2003) Using haplotype blocks to map human complex trait loci. Trends Genet. 19, 135 140

http://plants.trends.com