Introduction

Since quality demands of consumers and potato processing industry have become increasingly stringent, breeding efforts need to focus on quality traits in potato. Association mapping, also known as linkage disequilibrium mapping or gametic phase disequilibrium mapping—originally developed to study genetic disorders in humans—is a recently emerging tool in plant breeding research. Association mapping differs from classical Quantitative Trait Locus (QTL) mapping. An important distinction is that no segregating offspring have to be grown and phenotyped, since association mapping can deal with collections of existing cultivars. Another distinction is that parent choice is less of a dilemma, nor a limitation on the genetic diversity. A further difference is that phenotyping efforts may be reduced as existing phenotypic data from national lists, gene banks and breeding companies can be used in addition to or as a replacement of present day trials. Still, designed trials for the evaluation of selected germplasm will provide more reliable results. Lastly, some forms of association mapping, like the one presented in this paper and the one in Malosetti et al. (2007), provide a relatively simple approach to the identification of QTLs in tetraploids, comparable to the use of marker regression in segregating populations, but applied to a wider genetic background.

Association mapping has been and remains a popular research tool in human and animal genetics. Human disease genetics was the first area for which association mapping methodology was developed and where successes were achieved (Carlson et al. 2004; Jorde 2000; Lander and Schork 1994). In animal genetics, most concern was about the LD patterns in breeding populations to determine to what extent LD holds and the marker density required to fine map genes, e.g. in cattle (Farnir et al. 2000), pig (Nsengimana et al. 2004) and sheep (McRae et al. 2002). In plant genetics, however, fewer papers have been produced so far on association mapping, except for model plant systems like Arabidopsis thaliana (Hagenblad and Nordborg 2002; Nordborg et al. 2002), maize (Ching et al. 2002; Palaisa et al. 2003, 2004; Parisseaux and Bernardo 2004; Rafalski and Morgante 2004; Remington et al. 2001; Tenaillon et al. 2001) and rice (Garris et al. 2003; Lu et al. 2005; Semon et al. 2005), where association mapping is gaining importance due to the development of high-throughput marker systems and the availability of genome sequences. Other plant systems have been investigated for LD patterns and associations as well. In barley Kraakman et al. (2004, 2006) performed association analysis for agronomical, resistance related and morphological traits. Conifers were examined using the candidate gene approach to dissect complex traits (Neale and Savolainen 2004). Barnaud et al. (2006) reported on the LD pattern of grapevine. In Lolium perenne, Skøt (2005) found associations for heading date. LD patterns were examined deploying SNPs in soybean (Zhu et al. 2003), AFLPs in sugar beet (Kraft et al. 2000) and RFLPs in sugarcane (Jannoo et al. 1999). Kernel size and milling quality were fine mapped applying association mapping in wheat (Breseghello and Sorrells 2006). A complete overview on association mapping and the status of this approach in plants has been published recently (Gupta et al. 2005). In conclusion, although the amount of publications on association mapping in plant genetics is not as high as it is for animal or human genetics, there is a clear trend visible towards increasing number of plant species on which association mapping research has been performed with success.

Genetic and QTL studies in potato have predominantly been performed on diploid populations, e.g. leaf blight resistance QTLs have been detected in a di-haploid clone (Bradshaw et al. 2006), in a diploid S. phureja × S. stenotomum hybrid population (Costanzo et al. 2005; Simko et al. 2006), in diploid S. vernei (Sørensen et al. 2006), in diploid S. microdontum (Bisognin et al. 2005) and in diploid S. tuberosum (Visker et al. 2005). Cold sweetening QTLs have been mapped by Menendez et al. (2002) in diploid potato. Recently, tetraploid potato populations have been used as well. Simko and co-researchers mapped Verticillium resistance genes in tetraploid populations using various association mapping approaches, such as haplotype association—a whole genome approach—and candidate genes—a local approach (Simko et al. 2004a, b). With a limited number of targeted markers, Bormann et al. (2004) tagged QTLs for maturity corrected late blight resistance in tetraploid potato. However, the QTL studies in the tetraploid mapping population 12601ab1 × Stirling are unique in demonstrating the complexity of map construction and QTL analysis (Bradshaw et al. 2004; Bryan et al. 2004).

Association mapping methodology in potato till now is mainly restricted to standard two sample t-tests or Mann-Whitney U tests. Gebhardt et al. (2004) for example, deployed the non-parametric Mann-Whitney U test for an association assay on 415 cultivars, genotyped with 5 PCR markers closely linked to previously mapped QTL for late blight resistance and plant maturity, with evaluation data for late blight resistance and maturity. Simko et al. (2004a, b) detected significant association between Verticillium dahliae resistance and an SSR marker closely linked to a candidate gene locus for Verticillium resistance (StVe1) in tetraploid potato using a standard two sample t-test. Verticillium albo-atrum resistance appeared associated with a specific haplotype of the same candidate gene locus in tetraploid potato, detected using haplotype association with the Wilcoxon two sample test. Recently, Malosetti et al. (2007) presented a general mixed model approach to LD mapping, using potato data for illustration, giving attention to the incorporation of relationships between genotypes, whether induced by population substructure, pedigree or other, and the combination of data arising from multiple environments. Their approach is broadly applicable, without the need for specific software. The approach explained by Malosetti et al. (2007) is very similar to the mixed model approach of Yu et al. (2006). In the latter paper, the mixed methodology was illustrated by an example in maize.

In this paper, we will present preliminary results on the utility of a simple association mapping strategy for tetraploid potato using existing data from five potato breeding companies. With a limited number of markers and relatively simple regression models we detected marker-trait associations for a number of quality traits such as cold sweetening (Li et al. 2005; Menendez et al. 2002), after cooking darkening (ACD) (Friedman 1997; Wang-Pruski and Nowak 2004), enzymatic browning and/or blackspot bruising as catalysed by Polyphenol oxidase (PPO) (Lærke et al. 2002). These first results are very promising considering this was a preliminary experiment performed to sort out association mapping methodology before embarking on a more detailed and thorough follow up study that is presently undertaken.

Materials and methods

Plant material

With the aim to collect a representative subset of worldwide available potato germplasm, we selected 221 tetraploid potato cultivars and progenitor lines. The selection was based on (1) the acreage for seed potato production in the Netherlands, (2) the country of origin, (3) the year of commercial introduction, (4) the market niche (table, starch and processing), (5) phenotypic diversity for quality traits and (6) availability of the cultivars, especially for older genotypes. The material was kindly provided by Dutch breeding companies and several gene banks (see acknowledgements). Leaf material was harvested from greenhouse-grown and in-vitro-grown genotypes, was frozen with liquid nitrogen and stored at −80°C until DNA extraction.

Molecular marker analysis

DNA extraction was according to Van der Beek et al. (1992). DNA quality and concentration were visually examined using ethidiumbromide stained 1% agarose gels.

AFLP markers were generated according to Vos et al. (1995) using 5 well known Eco/Mse primer combinations: E + AAC/M + CAC, E + AAC/M + CAG, E + AAC/M + CCA, E + ACA/M + CAC, E + AGA/M + CAT. Fragments were separated using a capillary sequencer (MegaBACE 1000, Molecular Dynamics & Amersham, serial number 13757) according to van Eijk et al. (2004), each primer combination being labelled with either FAM, NED or JOE. The ROX channel was used for the MegaBACETM ET900-Rox size standard from GE Healthcare (Amersham Biosciences).

Pseudo gel images were scored at Keygene N.V. Marker nomenclature was based on primer combination and fragment mobility.

Normalisation of signal intensity variation between capillaries due to DNA loading effects was performed on log-transformed band intensity values using the ANOVA procedure in GenStat, release 8.11 (VSN International Ltd., Oxford, UK).

Position information of AFLP markers was retrieved from the ultra dense potato map, using the parental diploid genotypes SH83-92-488 and RH89-039-16 as internal reference (van Os et al. 2006; http://potatodbase.dpw.wau.nl/UHDdata.html).

Phenotypic data collection

An overview of the agro-morphological and quality traits of this study is shown in Table 1.

Table 1 Overview of phenotypic traits, their data types, scales and the numbers of genotypes for which data were provided by five breeding companies for these 11 phenotypic traits
Full size table

The agro-morphological traits tuber shape, flesh colour and maturity are scored by breeders on an ordinal scale. Trait values do not imply breeder or consumer preference. The flesh colour shades go from white to orange and can be interpreted as a ranking. Under water weight was assessed on a continuous scale and was based on a 5 kg sample size.

Quality traits were all scored on an ordinal scale (Table 1). Cold sweetening was not measured as such, but through the highly correlated characteristics frying and chipping colour. All tubers were stored at 8°C before frying analysis. Frying colour was determined after 3 min of frying at 180°C, chipping colour after frying at 180°C until no more air bubbles appear. After baking darkening was determined following 2 min of pre-frying at 160°C, merely scoring the amount of greyness which has a different causal background than frying colour. After cooking darkening was phenotyped after steam cooking at 100°C, as the severity of greyness. To determine the bruising score, a sample of 50 tubers representative for the field plot was taken—when available—and shaken on a mechanical shaker. One week later, the tubers were peeled and subdivided into classes of severity of bruising (mild, medium, severe) according to size and frequency of blue spots. To determine the bruising score, tuber amounts per severity class were counted and weighed applying the following formula: (50/number of tubers used) * ((number of mild + 2*number of medium + 3*number of severe)/3). For enzymatic browning, tubers were rasped and exposed to air and at a fixed time point enzymatic browning severity was scored. Cooking type was scored following steam cooking.

Phenotypic data came from field trials performed by five Dutch breeding companies (see acknowledgements). For a number of traits, the trait values for individual genotypes represented summary statistics across years and locations, calculated following company specific procedures that were not accessible for us. The phenotypic dataset was unbalanced because the breeding companies analysed different sets of cultivars and traits (Table 1).

Statistical procedures

Linkage disequilibrium (LD) between loci was quantified by the squared correlation coefficient, r 2, between the (logarithm of) band intensities (Flint-Garcia et al. 2003; Remington et al. 2001; Zhao et al. 2005). LD decay was investigated by plotting r 2 versus map distance.

We chose band intensities for the estimation of LD and marker-trait associations and not band presence/absence, because in a specially designed pilot experiment we found band intensity to be related to allele dosage (unpublished). The main conclusion of the pilot was a clear intensity difference between no allele and at least one allele, but not a very strong trend between intensity and allele dosage in the range of one to four alleles. We have experimented with different functions representing the relation between band intensity and allele dosage, but in the end decided that at the moment of writing the most robust approach was to use (the logarithm of) band intensity as such. The use of band intensity produced results that were very closely the same to results using band presence/absence. In genetic terms this means that we worked with a model in which band presence is fully dominant, i.e., it does not matter whether a band corresponds to an allele dosage of 1, 2, 3 or 4, the effect on the phenotype will be the same for any dosage.

The problem of possible population structure in the potato cultivar collection was approached in two ways. Firstly, as an a posteriori approach, using MEGA 3.1 (Kumar et al. 2004), a neighbour joining dendrogram was calculated based on Euclidian distances of the AFLP band intensities. This dendrogram was subsequently inspected for possible clusters. These clusters would then indicate subpopulations. A contrasting, a priori approach consisted in assuming that the breeding companies that developed the cultivars represented subpopulations. In the association analysis, correction for population structure can then be achieved by inclusion of a factor whose levels are defined by the breeding companies.

A two step approach to association mapping was used. First, Best Linear Unbiased Estimators (BLUEs) were calculated per trait for the genotypes, using GenStat, release 8.11 (VSN International Ltd., Oxford, UK). For each trait an additive analysis of variance model was fitted, response = breeding company + genotype + error, where the breeding company factor referred to the company that provided the data for the particular trait (see Table 1). As the breeding companies differed in the set of genotypes they evaluated, the data were unbalanced and therefore the BLUEs for the genotypes were adjusted means.

The traits were all analysed as quantitative traits, assuming the error variation to be normally distributed with constant variance, although the majority of the traits was measured on an ordinal scale. Inspection of diagnostic plots for residuals indicated no strong violations of these assumptions.

After calculation of genotypic BLUEs, markers were screened for association with traits. For each trait and for each of the 149 markers, two models were fitted;

  1. 1)

    Response = marker + error

  2. 2)

    Response = breeder + marker + error

The marker information that entered the models consisted of the logarithm of the band intensity, so that the markers were introduced as continuous variables for which a slope parameter was estimated by standard least squares. The breeder term refers to the breeding company that developed the particular cultivar. This term is supposed to correct for possible population structure.

Results

LD decay and population structure

Markers which could be scored in an unambiguous manner were retained. In total 551 AFLP fragments were distinguished on 227 genotypes, including a three-fold repetition of the diploid mapping parents SH83-92-488 and RH89-039-16 (van Os et al. 2006), which were included as internal reference genotypes. From these 551 AFLP markers 149 markers could be assigned to mapped AFLP bands in the ultra dense potato map. We assumed that the position of these 149 markers in the diploid UHD map is essentially not different from the position of these bands in the collection of tetraploid genotypes. Besides position information, the AFLP bands represent alleles on two homologous chromosomes from the female and male mapping parents. Although male and female cM positions may differ due to differences in recombination in male and female meiosis, we did not attempt to assign sex-averaged recombination distances between the markers. Markers from the same homology group were tested for LD while ignoring from which homologous chromosome they were derived.

Linkage disequilibrium (LD) decay across the potato genome was calculated deploying r 2, the squared correlation between the (log-) band intensities of AFLP marker loci. A graphical representation, plotting the resulting r 2-values in relation to the UHD map distance in centiMorgan (cM) between the AFLP markers, was constructed. Linkage disequilibrium between markers mapped on different non-homologous chromosomes is embedded in the same graph by using ∞ cM as symbolic distance between these unlinked markers (Fig. 1). In this AFLP marker dataset r 2 seems to drop below 0.1 at about 3 cM, which is promising for LD mapping in potato and its applications in marker assisted breeding. For the current sample size of 221 and a significance level of 0.01, the critical value for r 2 is 0.03. Applying this value as reference, LD seems to disappear after about 8 cM.

Fig. 1
figure 1

Linkage disequilibrium (LD) decay plot. The squared correlation between paired marker intensities on the y-axis is plotted against the distance between pairs of markers in centiMorgan (cM) on the x-axis. LD between pairs of markers on different non-homologous chromosomes is represented by the symbol ∞. The commonly accepted reference value for r 2 of 0.1, indicates LD to decay at about 3 cM. The critical test value for r 2 is 0.03 for a test level of 0.01. Using this critical value as threshold, LD decays at about 8 cM

Full size image

Population structure of our cultivar and progenitor line collection was examined, using the neighbour joining facilities of the package MEGA 3.1 for clustering (Kumar et al. 2004) and defining distance between genotypes by the Euclidian distance on the basis of the AFLP (log-) band intensities. Although there is a suggestion of some clusters of cultivars sharing country of origin or market niche, the neighbour joining analysis revealed no clear structural relationships, as demonstrated by the very short branches separating clusters from the central point of the astral projection of the dendrogram (Fig. 2).

Fig. 2
figure 2

Neighbour joining dendrogram visualised with MEGA3.1 (Kumar et al. 2004). Euclidian distances between band intensities for 149 AFLP markers defined distance between genotypes. The population of genotypes comprised 221 potato cultivars and progenitor lines

Full size image

Marker-trait associations

Numbers of marker loci associated with traits at a test level of 0.01 are given for the models with and without correction for breeding origin of the material in Table 2. The breeder term was significant for all traits. As expected, the number of marker-trait associations typically decreased after inclusion of the correction for breeder. Since all 149 AFLP markers in the genotypic dataset had a known map location, the loci associated with particular traits could be allocated to a potato chromosome. Table 2 offers an overview per trait of the detected loci together with their chromosome allocations.

Table 2 This table presents on the left side the number of detected marker-trait associations deploying the model without (Simple model) and deploying the model with correction for population structure (Structure model) together with their chromosomal location, based on the UHD map, for the 11 traits studied
Full size table

Table 2 also presents a comparison between results described in literature and our association mapping results. In most instances, detected QTLs coincide with chromosomes for which associations with the respective trait were previously reported. For cooking type and tuber shape, our QTL chromosomes differed from literature reports. For maturity, after cooking darkening, chipping colour and enzymatic browning more loci were detected than described in literature. For the other traits like flesh colour, under water weight and frying colour less loci were detected than reported elsewhere. This paper presents the first QTLs reported for after baking darkening and black spot bruising.

Illustration for two marker trait associations

Choosing a significance level of 0.01, which involves a very modest multiple testing correction, eleven AFLP marker loci were detected for maturity, scattered over six chromosomes of the potato genome, namely 1, 4, 5, 6, 9 and 10 (Table 2). The observed distribution of maturity is depicted in Fig. 3. The histogram in Fig. 3 illustrates the spectrum of earliness as determined throughout the collection of 221 potato cultivars and progenitor lines. Unfortunately, there was no maturity data available for all 221 genotypes, only 155 genotypes were represented in the phenotypic dataset.

Fig. 3
figure 3

Histogram for maturity. Number of observations was 155

Full size image

Because literature has reported several times on a major QTL influencing maturity residing on chromosome 5, we were interested in the effect of a chromosome 5 AFLP marker associated with maturity in our collection of genotypes. Figure 4 illustrates the effect of this chromosome 5 marker in the context of the observed data when portrayed by boxplots; marker presence will result in an earlier ripening phenotype. Observed difference between the medians of genotypes with and without band was 0.83 units. The estimate for the effect of the band on maturity in the association model was 0.97 units with an estimated standard error of 0.20. The F-test for association of this marker with maturity had a P-value of 10e−6.

Fig. 4
figure 4

Boxplots for maturity using presence and absence of an associated AFLP marker on chromosome 5 to define the groups. The presence and absence of the specific marker was determined using peak detection on the pseudo gel image. The median values are written in the boxes, the difference in median between genotypes with and without band was 0.83 points on the maturity scale, meaning that a genotype having this marker band will in general be earlier ripening in the field

Full size image

Frying colour, an indicator for cold sweetening, appeared associated with three AFLP marker loci at the significance level of 0.01, spread over three chromosomes of the potato genome, namely 1, 2 and 7 (Table 2). The distribution of this trait is depicted in Fig. 5, where the histogram reflects the spectrum of frying colour as determined throughout the collection of 221 potato cultivars and progenitor lines. Unluckily, there was again no frying colour data available on all 221 genotypes, only 145 genotypes were represented in the phenotypic dataset.

Fig. 5
figure 5

Histogram for frying colour. Number of observations was 145

Full size image

Literature has reported on the location of cold sweetening influencing loci throughout the whole potato genome (Menendez et al. 2002). Therefore we decided to look at the effect of the molecular marker showing the strongest significant association with frying colour in our collection of genotypes. Figure 6 illustrates the effect of this chromosome 7 marker. As can be deduced from the boxplot, marker presence will result in a slightly better frying quality, more specific a less dark frying colour, meaning a less severe cold sweetening phenotype. Observed difference in median frying colour between genotypes with and without the band at this marker was 0.38 units. The effect of band presence for this marker was estimated to be 0.37 units with a standard error of 0.13. The F-test for association for this marker had a P-value of 0.004.

Fig. 6
figure 6

Boxplots for cold sweetening using presence and absence of an associated AFLP marker on chromosome 7 to define the groups. The presence and absence of the specific marker was determined using peak detection on the pseudo gel image. The median values are written in the boxes. The difference in median between genotypes with and without the band was 0.38 points on the frying colour scale, meaning that a genotype having this marker band will in general have a less dark frying colour, i.e. a higher frying quality, than a genotype not expressing this marker band

Full size image

Discussion

Association mapping approach

When comparing LD mapping in tetraploid potato with QTL analysis on segregating populations, LD mapping is less cumbersome than standard QTL mapping: it is less time-consuming as no segregating offspring need to be grown and phenotyped and no parent choice needs to be made. For LD mapping, the relationships between marker and trait loci are not merely explained by genetic distance, but also by the number of meiotic generations separating genotypes from each other. On the one hand, this renders the genetics more complex, because population admixture, selection or genetic drift, can bias detected associations. On the other hand, due to recombination events occurring each meiotic generation, stretches of linkage disequilibrium will tend to be somewhat shorter, which can be advantageous to pinpoint genetic markers closer to a trait locus of interest. In LD studies in potato, the role of recombination may be restricted. Gebhardt et al. (2004) showed for a subset of German potato germplasm that there are only 4–5 meiotic generations separating heirloom potatoes from their recent descendants. In contrast, our collection contains exceptional cultivars, like for example Biogold being separated by 10 meiotic generations from his ancestors (van Berloo et al. 2007).

LD decay

We estimated the linkage disequilibrium pattern of the potato genome deploying r 2 between (log-) band intensities, thereby following an approach similar to Remington et al. (2001) and Flint-Garcia et al. (2003). Applying the commonly used reference value of 0.1 for r 2, we found that LD seems to drop at about 3 cM. LD decay in potato has already been addressed by Gebhardt et al. (2004), where they examined four DNA markers within 1 cM on chromosome V. For two markers residing within 0.3 cM LD was maintained, whereas for markers being separated by 0.6 cM LD had decreased and for markers being 0.9 cM apart linkage equilibrium had been reached. Simko et al. (2006) found indications of a fast decaying linkage disequilibrium on short range (r 2 = 0.208 at 1 kb) but then afterwards a slower decreasing LD (r 2 = 0.137 at 70 kb) was found. They conclude from their data that LD decayed below 0.10 at about 10 cM genetic distance.

Population structure

We examined population structure with the neighbour joining algorithm using Euclidian distance on the AFLP marker band intensities. This revealed no clear clusters, although some geographically isolated cultivars (USA and UK) and some market niche cultivars (starch processing) tended to group together (Fig. 2). Cultivars known to be mutants of one another, for example Duke of York (Eersteling) and Red Duke of York (Rode Eersteling) or cultivars known to be closely related to each other, such as Craig’s Defiance and Craig’s Alliance or Arran Banner and Arran Victory, clustered nicely together. Certainly, the clustering of cultivars could change if more marker data had been included in the cluster analysis, but still, this dataset, although small, sets a trend and more or less confirms what could be expected, namely that other continents and different market niches use to some extent a different gene pool.

Our correction for origin of germplasm in the form of inclusion of breeding company in the association analyses was successful in that this term was judged to be significant in the models for all phenotypic traits. Thus, it made sense to include this correction for population structure in the models for association.

Relations between traits

As illustrated in Table 2, some different types of discolouration have loci on the same chromosomes, namely chromosomes 2 and 7. For the traits chipping colour and frying colour this confirmed our expectations, since both traits are used as indirect indicators of cold sweetening. Moreover, both traits were correlated in our dataset (r = 0.49) and the detected loci on chromosomes 2 and 7, at 3.7 cM and 63.0 cM respectively, seemed pleiotropic because they were associated with the same AFLP markers. Contrarily, we did not expect after cooking darkening to share loci with cold sweetening nor with enzymatic browning, since these traits are caused by different biochemical processes. Indeed, no associated markers were found for those traits that showed association with more than one trait, thereby providing more evidence for close linkage of different QTLs than pleiotropic effects of a single QTL. Nevertheless, with the current precision for marker positions, it is not possible to draw firm conclusions on pleiotropy versus linkage as long as the markers are within 5–10 cM from each other.

The traits after cooking darkening and after baking darkening shared loci on chromosomes 1, 4 and 7 (see Table 2). Except for one marker tagging chromosome 1, the detected loci were not associated with the same AFLP markers. Nevertheless, the markers of these associated loci map closely together on the UHD map which could indicate a common QTL. This confirms our expectations, since both traits are assumed to be caused by the same process (see M&M), the only difference between both traits lies in the temperature to which the samples have been exposed before darkening took place.

Prospects

We used AFLP markers for an association study in potato using a compilation of existing data belonging to different breeding companies. We found in total 69 associated AFLP marker loci for 11 phenotypic traits at the significance level of 0.01, spread throughout the entire potato genome (Table 2), showing that even this simply constructed phenotypic historic data set in combination with a modest number of markers of the widely available and easy to use AFLP marker system and a relatively straightforward statistical analysis was able to produce interesting results for applied breeding. Our association approach identified markers that could be useful in the development of marker assisted breeding strategies. These initial results obtained with association mapping should be compared with the power to detect loci involved in quality traits with conventional QTL mapping approaches (both at the diploid and at the tetraploid level). At this moment we regret that in spite of the number of QTL studies on pathogen resistance, the number of QTL studies on quality traits is limited. We anticipate that association mapping will appear a valid method to analyse complex traits in tetraploid potato and will be used along with diploid QTL mapping studies. We can, however, compare the input required by either approach. Our association approach and QTL mapping in diploid mapping populations are both relatively simple in methodology and modestly demanding with respect to genotyping. This is in contrast to the genotyping and methodological demands for QTL mapping using tetraploids (Luo et al. 2006). A similar conclusion was obtained by Malosetti et al. (2007). Although these promising results were based on existing phenotypic data, we will invest in an improved phenotypic dataset. This will be achieved through a field trial with replicates including all cultivars and progenitor lines of the present study. Furthermore, we aim to expand the current marker dataset with additional AFLP markers, but also candidate genes and microsatellites. To further improve on the methodological part of the association approach, research is going on with more advanced mixed models that are able to correct for pedigree relationships and can introduce different marker types into the model. Finally, we aim at obtaining allele dosage observations and estimates for inclusion in our genetic models.