Introduction

The agronomic performance of crop varieties is mainly influenced by complex quantitative traits, for example, yield and quality components. Since the development of molecular markers, it has become feasible to identify and genetically localize the underlying polygenes as quantitative trait loci (QTLs) and to utilize these QTLs for crop improvement. Subsequently, a still increasing number of QTL studies, involving all agronomic important crop species, have been conducted (e.g. Paterson et al. 1988).

Despite all efforts to map QTLs, their relevance for breeding new varieties is still low. Two reasons for this phenomenon might be important: (1) since most QTL projects have been conducted within the elite pool of a crop species, the identified favorable QTL alleles could not be utilized as a new genetic resource for elite improvement; (2) almost all QTL studies used early generations (F2, F3 and BC1) for mapping and QTL detection. Unfortunately, the favorable QTL alleles often lost their effects after they were purified into elite lines. This phenomenon might be explained with the relatively high level of epistatic interactions between QTLs and other donor genes in early generations. These interactions might be fixed in advanced generations possibly, leading to silencing of the measured QTL effects.

A potential solution to the problem was presented by Tanksley and Nelson (1996). The authors integrated the mapping of favorable QTL alleles and the introgression of these alleles into one process. In order to achieve this goal, they utilized exotic germplasm as the genetic donor for the improvement of quantitative agronomic traits and conducted the marker and phenotype analysis in advanced backcross generations like BC2 or BC3. It is expected that through the introgression of new exotic QTL alleles, the AB (advanced backcross – QTL) strategy will contribute to an increased level of genetic diversity in our modern crop varieties.

To date, several reports on the application of the AB-QTL strategy are available for tomato and rice. In all cases, favorable exotic QTL alleles for important agronomic traits have been identified. For instance, fruit yield could be improved in tomato through the introgression of wild-species alleles from Lycopersicon pimpinellifolium and L. peruvianum by 17% and 34%, respectively (Tanksley et al. 1996; Fulton et al. 1997). A further AB-QTL study, which used L. hirsutum as the donor species, revealed 25 favorable wild-species QTL alleles out of 121 detected QTLs (Bernacchi et al. 1998a). Again, the authors detected wild-species alleles which increased yield by 15%. The most recent AB-QTL study in tomato was published by Fulton et al. (2000). As in other tomato wild species, the authors could localize favorable exotic QTL alleles from L. parviflorum which, for instance, increased yield by 27%. Similar results could be extracted from AB-QTL studies in rice. Here, two wild-species QTL alleles have been associated with an increase of yield by 17% and 18% on rice chromosomes 1 and 11, respectively (Xiao et al. 1996, 1998). Subsequently, the yield QTL effect on chromosome 1 was validated in a second cross using the same Oryza rufipogon donor accession (Moncada et al. 2001).

The favorable wild-species QTL alleles are useful as a breeding resource after they have been fixed in nearly isogenic lines (QTL-NILs) and after the superior performance of a QTL-NIL has been confirmed in comparison to the recurrent elite line. Bernacchi et al. (1998b) have already validated the effects of exotic tomato QTLs in QTL-NILs. In field evaluations at five locations worldwide, 22 QTL-NILs out of 25 tested (88%) exhibited phenotypic improvement compared to the recurrent parent, as had been predicted in the previous AB-QTL analysis. For instance, a QTL-NIL possessing an exotic QTL allele for a 15% yield increase did, indeed, outperform the control line by 12%. These reports clearly illustrate that the AB-QTL strategy is a powerful tool for the improvement of quantitative agronomic traits in elite varieties.

In this paper, we report on the first AB-QTL project which utilizes spring barley as a model. Our goal was: (1) to localize QTLs for the expression of quantitative traits in spring barley and (2) to identify favorable QTL alleles from the wild barley donor which improve the respective traits.

Materials and methods

Plant material

For QTL analysis, a BC2F2 population originating from the cross of the German spring barley var. Apex and the Israeli Hordeum vulgare ssp. spontaneum (Hsp) wild barley accession ISR101-23 (A×101) was generated via two cycles of backcrossing with Apex followed by one cycle of selfing. The BC2F2 population consisted of 136 randomly chosen individuals.

Simple sequence repeat markers

The following prefixes of SSR names indicate the published sources from which the primer sequences were taken: HVM, Liu et al. (1996); BMS, Russell et al. (1997); GMS, Struss and Plieske (1998); WMS, Röder et al. (1998); Bmac, Bmag, Ebmag and Ebmac, Ramsay et al. (2000); Hv, Becker and Heun (1995) and Pillen et al. (2000). A suffix with the chromosomal identifier in brackets was added to each SSR name as a simple reference. Linkage distances between SSR markers were inferred from Ramsay et al. (2000) and Pillen et al. (2000) or estimated after linkage analysis of 96 F2 individuals from the cross A×101 using the Mapmaker software (Lander et al. 1987).

Genotyping BC2F2 individuals

The genotypes of 136 BC2F2 individuals were determined by means of SSR analysis. For this, plant DNA was extracted, polymerase chain reaction (PCR) analyses conducted and SSRs surveyed after silver staining as reported in Pillen et al. (2000). At each informative SSR locus, homozygous H. vulgare ssp. vulgare (Hv), heterozygous and homozygous Hsp genotypes were distinguished. The genetic measure P[Hsp] as the percentage of Hsp alleles present in a single BC2F2 line was calculated from 40 co-dominantly scored SSRs according to the formula:

$$ {\rm P}[Hsp] = {{2[aa] + [Aa]} \over {2([AA] + [Aa] + [aa])}} $$
(1)

where [AA], [Aa] and [aa] correspond to the frequencies of the homozygous Hv, heterozygous and the homozygous Hsp genotypes, respectively, calculated from all investigated marker loci.

Phenotyping BC2F2-derived families

For field testing, each BC2F2 individual was bulk propagated until BC2F2:5 (1999) and BC2F2:6 (2000), respectively. During the seasons 1999 and 2000, the 136 backcross lines were grown at the three locations Dikopshof (University of Bonn), Gudow (Nordsaat Saatzucht) and Irlbach (Dr. J. Ackermann Saatzucht). These experimental stations are located in the west, north and south of Germany, respectively. At each location, a randomized complete block design with two replications was applied, except at the Dikopshof 1999, where three replications were taken. As a control, the recurrent parent was tested with ten replications per block. Plot size (3.75–5.5 m2), seeding rate (300 kernels/m2) and field management were in accordance with the local practice. At each plot, up to 13 quantitative traits were evaluated (Table 1).

Table 1. List of 13 quantitative traits measured in up to six environments
Full size table

QTL analysis

The QTL detection from BC2F2 genetic data and field data from multiple environments was conducted using the procedure GLM from the SAS software (SAS Institute 1999). In each environment, the field data of all replications were averaged by the least square means option in SAS. At each marker locus, only the homozygous genotypes (Hv or Hsp) were included in the calculation since the repeated selfing of heterozygous BC2F2 individuals lead to a mix of both homozygous genotypes in the derived BC2F2:5 and BC2F2:6 field plots, resulting in a false estimate of the performance of true heterozygots. The model used to detect QTLs included the effects marker genotype (M), environment (E) and M × E interaction. Under the assumption of a mixed model with the marker as a fixed effect and the environment as a random effect, the mean square of M × E was used as the error term for the computation of the F-statistic for a marker main effect. The mean square of the residual variance was used as the error term for the computation of the F-statistic for a M × E effect.

Following Stuber et al. (1992) and Xiao (1998), the presence of a stable QTL in the vicinity of a marker locus was accepted, if the marker main effect was significant at P < 0.01. In addition, the presence of an environment-dependent QTL was accepted, if the M × E interaction was significant at P < 0.01.

The explained phenotypic variances as a measure of the strength of a QTL main effect (R2 M) and a QTL × environment interaction (R2 I), respectively, were calculated as follows:

$$\hbox{R}_{\rm M}^2 = \hbox{SS}_{\rm M} /\hbox{SS}_{{\rm Total}} $$
(2)
$$\hbox{R}_{\rm I}^2 = \hbox{SS}_{{\rm M*E}} /\hbox{SS}_{{\rm Total}} $$
(3)

where SSM, SSM*E and SSTotal are the sum of squares of the factors marker and M × E interaction and the total sum of squares, respectively, which were obtained from the GLM procedure.

The relative performance of the homozygous Hsp genotype (RP[Hsp]) as a measure of the improvement of a trait by replacing both Hv elite alleles with the exotic Hsp alleles was calculated as follows:

$$\hbox{RP}[Hsp] = {{aa - AA} \over {AA}}*100$$
(4)

where, for each trait, aa and AA are the least square means of the homozygous Hv and the homozygous Hsp genotypes, respectively, calculated across all environments.

Results and discussion

Traits

In order to assess the value of the backcross population, we calculated the least square means of all traits for the backcross population and for the recurrent parent Apex (Table 2). The comparison of the means revealed that the backcross significantly (P < 0.01) excelled the recurrent parent Apex in the two traits HEA and WA. On the other hand, the backcross was significantly (P < 0.01) inferior to Apex in traits HI, KER, MT, YLD (Table 2). The inferior performance of the backcross for most traits can be explained by the absence of any positive selection during the process of backcrossing. Our goal was to keep the available genetic contrast – for instance, with respect to yield – in order to maximize the chance of detecting QTLs. Nonetheless, we could always identify individual backcross lines which outperformed the recurrent parent (data not shown).

Table 2. Least square means of 13 traits calculated across all environments for the backcross population and the recurrent parent Apex
Full size table

A calculation of correlations between the 13 traits was conducted across all environments (Table 3). Significant correlations (P < 0.05) were found for 56 trait combinations. From these, LOF and LOH revealed the strongest correlation (r = 0.83), which was expected since both traits are related. HEI displayed medium positive correlations with HEA, LOF, LOH and MAS, where r values ranged from 0.64 to 0.67. YLD displayed medium negative correlations with WA and PRO (r = –0.59 and –0.64, respectively) and medium positive correlations with KER, MT and HI, where r values ranged from 0.47 to 0.49. Both the negative correlation of YLD with PRO and the positive correlations with the yield-related traits KER and HI have been frequently observed in other studies.

Table 3. Pearson's correlation coefficientsa (r) between 14 quantitative traitsb
Full size table

One could expect that the presence of large chromosomal segments from wild barley would exhibit negative effects on agronomic traits. In order to test this hypothesis, the percentage of Hsp alleles (P[Hsp]) present in a BC2F2 line was estimated from all co-dominantly scored SSRs. No strong correlations could be measured between P[Hsp] and other traits. However, P[Hsp] revealed significant negative correlations with YLD (r = –0.46), HI (r = –0.47) and MT (r = –0.41). The negative correlations, in accordance with the significantly reduced trait mean values of the backcross population, confirm that without selection the introgression of wild barley germplasm would not lead to an improvement of agronomic traits. The same tendency holds true for the traits HEA, HEI, LOF, LOH, MAS and PRO. These traits, which should be reduced in regard to the breeding goals, show significant but weak positive correlations with P[Hsp], indicating that there is an overall negative influence of Hsp alleles on most agronomic traits. Despite the aforementioned correlations, the goal of the AB-QTL analysis was to identify those rare positive effects which exotic alleles can exert on agronomically important quantitative traits.

Marker data

Altogether, more than 200 SSRs were tested for polymorphism between the parents Apex and ISR101-23. Sixty-seven SSRs turned out to be polymorphic. From these, 22 (33%) failed to exhibit the Hsp allele in the BC2F2 population. Thus, the 136 BC2F2 lines were successfully genotyped with 45 SSRs. They were distributed over all seven barley chromosomes (Fig. 1). The chromosomal location of the SSRs were inferred from Ramsay et al. (2000) and Pillen et al. (2000) or, if necessary, from linkage analysis in a reference F2-population from the A×101 cross (data not shown). The complete set of 67 mapped SSRs covers 852 cM of the barley genome; the mean SSR density is equal to 12.7 cM. The first almost complete SSR map for barley includes 299 SSRs and covers 1,173 cM (Ramsay et al. 2000). Thus, the backcross map represents 73% of the Ramsay map. The A×101 map possesses nine gaps with a marker distance of more than 30 cM. Large regions, which are not covered by any polymorphic SSR, can be found, for instance, on chromosomes 1H and 6H (Fig. 1). The low coverage of the A×101 map can be explained with the occurrence of genetic drift during the backcrossing process. Presumably because only four BC1 individuals were used for the production of BC2 families, several chromosomal Hsp segments were lost until the BC2F2 generation was reached. This is particularly true for chromosomes 3H and 6H where 11 and four polymorphic SSRs, respectively, failed to segregate in the BC2F2 population (Fig. 1). The occurrence of marker gaps was also observed in other QTL analyses. Backes et al. (1995), for instance, could map only a single marker to barley chromosome 1H using a set of 50 informative restriction fragment length polymorphism (RFLPs). Their map included nine gaps encompassing more than 30 cM.

Fig. 1.
figure 1

Barley SSR map containing 86 putative QTLs with 29 favorable Hsp alleles detected from the BC2F2 cross Apex × 101-23. The SSR marker order is based on Ramsay et al. (2000). Twenty-two SSR markers that were polymorphic between the parents Apex and ISR101-23 but failed to show the ISR101-23 allele in the BC2F2 population are indicated by ⋄. The centromere positions, which were taken from the current Steptoe × Morex BIN map (http://barleygenomics.wsu.edu/databases/#databases.html), are indicated by open ovals. Putative QTLs which revealed either a significant (P < 0.01) marker main effect or M × E interaction are written to the left of the SSR locus. The abbreviations of the quantitative traits follows Table 1. A favorable effect of the Hsp allele at a putative QTL is indicated by *. Each chromosome name is followed in brackets by the number of detected putative QTLs and the number of favorable Hsp alleles

Full size image

From 45 informative SSRs genotyped, 40 (89%) could be scored co-dominantly. Three dominantly scored SSRs did not amplify a Hsp allele and two failed to amplify a Hv allele. The theoretical frequency of the three possible co-dominant marker genotypes is equal to 13:2:1 (AA:Aa:aa) in BC2F2. Thus, the expected mean portion of the Hsp genome (P[Hsp]) would be 0.125. Based on genotype data from the 40 co-dominantly scored SSRs, the P[Hsp] value was estimated for each BC2F2 line. The average P[Hsp] value of the BC2F2 lines was equal to 0.127 with a range from 0.093 to 0.296 and a standard deviation of 0.066. This finding confirms that the investigated A×101 population corresponds indeed to a BC2F2 generation. Furthermore, the good accordance of the expected and observed mean P[Hsp] values proved that during backcrossing no serious selection in favor of one parent was exercised in the population. Nevertheless, a distorted segregation was observed for single SSR loci. Altogether 16 SSRs (36%) revealed a significant distorted segregation at P < 0.01. However, no general prevalence of either Hv or Hsp alleles was detected. At eight loci, the population showed an excess of Hsp alleles and at the remaining eight loci the Hsp alleles were underrepresented.

QTL detection

A single-point marker analysis by means of a two-factorial ANOVA rather than an interval mapping was preferred for QTL analysis because of the low marker coverage and the presence of gaps in the A×101 map. It should be noted that QTL effects simultaneously detected at two or more adjacent marker loci by ANOVA are presumably due to the existence of a single, linked genetic factor which exerts the QTL effect. However, for two reasons we kept all putative QTLs in the records. First, the alternative hypothesis, that there are two or more QTL factors present on a single chromosomal segment, cannot be ruled out, and second, the ANOVA gives no clear indication which of the significant markers is located nearest to the true QTL factor.

Among 585 marker × trait combinations tested, 86 putative QTLs were detected. At 64 putative QTLs, the marker main effect and at 27 putative QTLs, the M × E interaction were significant at P < 0.01 (Fig. 1, Table 4). In five cases, both effects were significant. Altogether, 29 (34%) favorable QTL effects were detected. At these loci, the homozygous Hsp genotype was associated with an improvement of the trait compared to the homozygous Hv genotype (Fig. 1, Table 4). The putative QTLs were unevenly distributed over the chromosomes (Fig. 1). Whereas at least 20 QTLs were located on chromosomes 4H, 5H and 7H, only one or zero QTLs were detected on chromosomes 3H and 6H, respectively. However, on the latter chromosomes only four and two SSRs, respectively, were genotyped. Most of the favorable QTLs were located on chromosomes 1H, 2H and 4H (8, 6 and 7, respectively). No favorable QTLs were detected on chromosomes 3H and 6H. The distribution of putative QTLs among the 45 genotyped SSR markers was also uneven. Eight SSRs were not associated with any QTL effect, whereas marker EBmac0824[5H] showed putative QTL effects on nine traits, simultaneously. In the following, the detected putative QTLs are described for each trait (see also Table 4).

Table 4. List of 86 putative QTLs detected from the BC2F2 cross Apex × 101-23a
Full size table

Days until heading (HEA)

Twenty-two putative QTLs for HEA were located on five chromosomes. In all cases, the marker main effect was significant at P < 0.01. In addition, the loci HvB23D [4H] HvB23D4 [4H], and (EBmac0824[5H]) exhibited a significant M × E interaction. For eight QTLs, a favorable effect of the Hsp allele on HEA was observed. At these loci, which were located on chromosomes 1H, 2H and 5H, the presence of the Hsp allele was associated with a reduced heading time of up to 9.7% (GMS3[2H]). At the remaining QTLs, the Hsp allele was associated with an increased heading time of up to 7.4% (HvB23D [4H]). The maximum explained phenotypic variance was found at HvB23D4 [4H] with 6.8%.

Height (HEI)

Seventeen putative QTLs for HEI were located on four chromosomes. All loci exhibited significant marker main effects. For five QTLs, a favorable effect of the Hsp allele on HEI was observed. At these loci, the presence of the Hsp allele led to a reduction in plant height of up to 10.4% (GMS3[2H]). Most QTL alleles from Hsp resulted, however, in an increased plant height of up to 13.4% (EBmac0824[5H]). The phenotypic variance explained by a QTL reached its maximum with 14.1% at Bmag0120[7H].

Lodging at flowering (LOF)

Only two QTLs for LOF were located on chromosome 5H. The unlinked SSRs HVM6[5H] and EBmac0824[5H] exhibited a significant marker main effect and a significant M × E interaction, respectively. The two QTLs showed effects of opposite direction. Whereas the Hsp allele was associated with a 33.5% reduction of lodging at HVM6[5H], it increased the tendency to lodge by 106.2% at EBmac0824[5H]. The maximum explained phenotypic variance was reached with 2.6% at EBmac0824[5H].

Lodging at harvest (LOH)

As in the case of LOF, two QTLs were located for LOH. However, these were mapped in close proximity on chromosome 1H. Both QTLs exhibited significant marker main effects with favorable Hsp alleles. The stronger effect was measured at locus Bmag0211[1H]. Here, the Hsp allele resulted in a lodging tendency at harvest which was reduced by 27.2%. The explained phenotypic variance for Bmag0211[1H] amounted to 1.1%.

Ears per square meter (EAR)

No QTLs were detected for EAR in this study.

Kernels per ear (KER)

Only one QTL was detected for KER. Marker GMS21[1H] exhibited a significant main effect. The negative effect of the Hsp allele resulted in a 6.5% reduction of kernels per ear. The explained phenotypic variance for GMS21[1H] amounted to 3.1%.

Thousand grain weight (TGW)

A total of 12 putative QTLs were located for TGW on four chromosomes. While six loci exhibited a significant marker main effect, the other six loci showed a significant M × E interaction. Favorable effects of the Hsp alleles on TGW were observed for five QTLs on chromosomes 2H, 5H and 7H. At these loci, the presence of the Hsp allele resulted in a TGW increase of up to 4.6% (HvUDPGPP [5H]). Contrasting negative effects of the Hsp allele were detected on chromosomes 4H, 5H and 7H, which resulted in reduced TGW of up to 4.7% (GMS1[5H]). The highest portion of the phenotypic variance was explained by HvB23D4 [4H] (2.6%).

Yield (YLD)

Altogether, 13 putative QTLs for YLD were located on all barley chromosomes except for 6H. Whereas 11 loci exhibited a significant marker main effect, four loci showed a significant M × E interaction. Both, the main effect as well as the interaction were significant at the linked loci EBmac0824[5H] and GMS27[5H]. Most QTL alleles from Hsp resulted in yield reductions with a maximum of 21.0% (EBmac0824[5H]). However, in two cases a favorable wild species effect on plot yield could be detected. The Hsp alleles of the two loosely linked SSRs, HVM40[4H] and GMS89[4H], were associated with a yield increase of 5.6% and 7.7%, respectively. The yield increasing main effect of both Hsp alleles was detected in all six environments measured. A maximum explained phenotypic variance was found at EBmac0824[5H] with 6.9%.

Above ground biomass (MAS)

Only one QTL was detected for MAS. Marker HvA22S [7H] exhibited a significant main effect. The negative effect of the Hsp allele resulted in a 5.1% reduction of the above ground biomass. The explained phenotypic variance for HvA22S [7H] amounted to 0.6%.

Harvest index (HI)

Five putative QTLs were located for HI on chromosomes 4H, 5H and 7H. While four loci exhibited a significant M × E interaction, HVM4[7H] showed a significant marker main effect. No favorable effects of the Hsp alleles on harvest index were observed. The presence of the Hsp allele resulted in a HI decrease of up to 15.6% (EBmac0824[5H]). As in case of YLD, the highest portion of the phenotypic variance was explained by (EBmac0824[5H]) with 14.2%.

Protein content (PRO)

Three putative QTLs were located for the quality-related trait PRO on chromosomes 4H and 5H. All QTLs were detected as significant M × E interactions and exhibited, with respect to brewing, negative effects of the Hsp alleles. The highest increase of the protein content (5.5%) was associated with the Hsp allele from the EBmac0824[5H] locus. At this locus, the explained phenotypic variance reached a maximum with 1.2%.

Water absorption (WA)

Six putative QTLs were located for the quality-related trait WA on chromosomes 4H, 5H and 7H. All QTLs were detected as significant M × E interactions and exhibited, with respect to brewing, favorable effects of the Hsp alleles. The highest increase in water absorption (1.8%) was associated with the Hsp allele from the EBmac0824[5H] locus. The highest portion of the phenotypic variance was explained by HvB23D4 [4H] (6.1%).

Malt tenderness (MT)

Two putative QTLs were located for the quality related trait MT on chromosomes 5H and 7H. Both QTLs were detected as significant marker main effects. The effects of both Hsp alleles were negative in regard to brewing. The highest reduction in malt tenderness (8.2%) was associated with the Hsp allele from the EBmac0824[5H] locus. At this QTL the explained phenotypic variance reached its maximum with 1.7%.

Comparison with other QTL analyses in barley

In the following, the AB-QTL analysis will be compared with other barley QTL analyses – first, with respect to the differences in methods and second, with respect to QTLs mapped in common.

To date, more than 60 publications are available which deal with yield- and quality-related QTL analyses in barley – for instance, Hayes et al. (1993), Han et al. (1995), Backes et al. (1995), Thomas et al. (1995, 1996), Tinker et al. (1996), Mather et al. (1997), Bezant et al. (1997a, 1997b), Kircherer et al. (2000), Marquez-Cedillo et al. (2000, 2001). For comparison with the AB-QTL analysis, the aforementioned studies will be termed "classical" QTL analyses. While classical QTL analyses were conducted in early, balanced generations like doubled haploid (DH) and F2, our AB-QTL analysis was based on a BC2F2 population. This change was necessary since we used an exotic cross with the barley progenitor Hsp as the donor of potential favorable QTL alleles. By means of backcrossing, the genome portion of Hsp was reduced to a mean P[Hsp] value of 0.127. It is assumed that masking negative side effects of linked or unlinked Hsp alleles on quantitative traits are reduced by this strategy. The high frequency of 33.7% of trait-improving Hsp alleles supports this assumption. However, it is still open if the identified favorable QTL alleles from Hsp are indeed unmatched in the elite gene pool of barley. Studies in Oryza and Lycopersicon have shown that the overwhelming majority of the genetic diversity in those taxa is present in the wild-species gene pool (Tanksley and McCouch 1997). Similar results for Hordeum have been reported by Powell and Russell (2000). Based on this findings, it is likely that at least a portion of the identified favorable QTL alleles from Hsp are new alleles, so far not present in the barley elite gene pool.

The number of genotyped DNA markers in classical QTL analyses range from 50 to 129, with the majority of markers being RFLPs and, to some extent, amplified AFLPs. In future, it can be expected that the majority of QTL analyses will be conducted with highly automatable markers like SSRs or single nucleotide polymorphism (SNPs). For barley, there are already more than 300 high quality SSRs available in the public domain (Becker and Heun 1995; Liu et al. 1996; Russell et al. 1997; Struss and Plieske 1998; Pillen et al. 2000; Ramsay et al. 2000; Kota et al. 2001). To our knowledge, our study represents the first example of a QTL study in barley which is exclusively based on SSR markers. However, the number of 45 SSRs still segregating in the BC2F2 population is relatively low. Our attempts to fill in gaps, for instance on chromosome 3H and 6H, failed since either no SSRs were available for that particular chromosomal region or an informative SSR did not segregate within the BC2F2. The latter phenomenon is most likely caused by the loss of Hsp alleles during the backcrossing process.

The number of 86 putative QTLs corresponds to classical QTL studies in barley which have detected between 11 and 159 QTLs. A variety of factors may affect the outcome of a QTL analysis – For example, the selection of the cross, population structure and size, number of measured replications and environments and type, number and density of markers. In addition, the selection of the statistical method exerts a major impact on the results of a QTL experiment. Examples of refined methods applied to classical QTL analyses are simple interval mapping (SIM, Haley and Knott 1992), composite interval mapping (CIM, Jansen and Stam 1994; Zeng 1994) and simplified CIM (sCIM, Tinker and Mather 1995). Several software programs which are based on these methods have been written for detection of QTLs, e.g. mapmarker/qtl (Lander and Botstein 1989), qtl-cartographer (Basten et al. 1994), mqtl (Tinker and Mather 1995) and plab-qtl (Utz and Melchinger 1996). Unfortunately, these programs are focused on the analysis of balanced populations which are used in classical QTL analyses. For unbalanced populations, which are used in AB-QTL studies, the program qgene was written (Nelson 1997). qgene operates with single marker regression as well as simple interval mapping for QTL detection. However, qgene cannot handle multiple environments simultaneously. Since our AB-QTL study was conducted in six separate environments and since we wanted to include the M × E interaction as a measure of the environment stability of a QTL effect, we preferred to use a two-factorial ANOVA with the marker genotype and the environment as factors. By including the environment in the statistical model, we expected to reduce the residual variance of the experiment and, thus, to increase the probability of detecting a QTL effect. Simultaneously, a two-factorial model allowed us to differentiate between a QTL significant as a marker main effect, which is considered to be stable across the tested environments, and a QTL significant as a M × E interaction where the effect is considered to depend on a particular environment.

Next, we compared our AB-QTL analysis with classical QTL analyses with respect to common QTL effects. Unfortunately, a direct comparison of QTL effects is difficult. The classical barley studies have been conducted with RFLPs and, to a lesser extent, with AFLPs, whereas our AB-QTL study is exclusively based on SSR markers. Nonetheless, an indirect comparison of QTL effects is possible by means of the current Steptoe × Morex map, published by A. Kleinhofs (http://barleygenomics.wsu.edu/databases/databases.html ). The map integrates RFLP, AFLP and SSR markers, which are mapped in independent linkage studies, by allocating them to 99 evenly spaced BIN groups. Based on the Steptoe × Morex BIN map, eight SSRs of our AB-QTL analysis – which are also placed in BIN groups and which have been associated with QTL effects – can be compared with classical barley QTL analyses. These markers are: HvBKASI [2H] HVM54[2H], HVM40[4H], WMS6[4H], HVM67[4H], HVM6[5H], HVM4[7H] and HvPRP1B [7H] The QTL comparison was carried out by means of an up-to-date compilation of mapped barley QTLs from P. Hayes (http://www.css.orst.edu/#barley/nabgmp/qtlsum.htm), which is based on the Steptoe × Morex BIN classification. A possible common QTL was assumed, if the BIN groups of two independently detected QTLs were identical or, at least, overlapped.

The eight selected SSRs could be associated with 11 putative QTLs in the A×101 backcross. By comparison of BIN groups, four QTL effects (36%) could be verified in at least one classical barley QTL analysis. The first common QTL is the putative QTL for HEA associated with HVM67[4H] on BIN 13. A classical QTL for HEA was also detected in the same or in overlapping BIN groups at locus Bmy1 (Hackett et al. 1992) and in marker intervals ABG397-ksuH11 and ABG397-Bmy1 (Hayes et al. 1993, 1996). The second common QTL was again detected for HEA but associated with HvPRP1B [7H] on BIN 12. This QTL was recovered at marker BCD512A (Laurie et al. 1995) and in marker interval MWG539-MWG929 (Backes et al. 1995). The third common QTL was found for LAF associated with HVM6[5H] on BIN 15. This QTL was also detectable in marker interval MWG650-MWG002 (Backes et al. 1995). The fourth common QTL was found for TGW associated with HvBKASI [2H] on BIN 8. In this case, a classical QTL for kernel plumpness was also detected in the overlapping BIN groups for marker interval HvBKASI -vrs1 (Marquez-Cedillo et al. 2001).

No classical QTL matched the favorable yield QTL which was associated with HVM40[4H] on BIN 2. However, corresponding classical QTLs for the favorable yield QTL at marker locus GMS89[4H] could be located, although a direct allocation of GMS89[4H] onto the BIN map has not been made yet. Because GMS89[4H] was placed 5 cM south of HVM68 (Ramsay et al. 2000), we propose to allocate GMS89[4H] to BIN group 7 or 8 on chromosome 4H. In that case, classical QTLs for grain yield have been reported in overlapping BIN groups at marker Xpsb37(L) (Bezant et al. 1997a) and in marker intervals ABG472-ABG366 (Tinker et al. 1996) and ABG472-ABG397 (Hayes et al. 1993). In addition, Ellis et al. (2002) also reported a QTL for grain yield in the region between HVM68 and HVM67 on chromosome 4H where GMS89[4H] is placed.

The overall moderate conformity between the QTLs identified in our AB-QTL analysis and in classical QTL analyses can be regarded as a confirmation that most QTL effects from the exotic donor Hsp are unique. Thus, these QTLs can be exploited for improving and broadening the genetic basis of the barely elite gene pool. Nevertheless, it should be noted that there is also little conformity present between classical QTL studies. For example, Thomas et al. (1995) reported considerable differences in QTL identification between the Scottish cross Blenheim × E224/3 and the North American crosses Steptoe × Morex and Harrington × TR306. Likewise, Mather et al. (1997) reported that, when comparing the two aforementioned North American crosses, they found more differences than confirmations of QTL positions.

Comparison with AB-QTL analyses in tomato and rice

This report represents the first application of the AB-QTL strategy in barley. The goals of the AB-QTL analysis are the identification and simultaneous transfer of those exotic QTL alleles which have the potential to improve yield-related and quality-related agronomic traits. Within the A×101 population, a total of 29 favorable Hsp alleles (34%) were identified among 86 localized QTLs. These favorable Hsp alleles were detected for seven of the 13 traits investigated. The portion of favorable exotic alleles identified in A×101 is lower than, for instance, in rice where Xiao et al. (1998) reported a frequency of 51% favorable exotic alleles among 68 QTLs identified from an elite backcross with the wild rice O. rufipogon. On the other hand, Bernacchi et al. (1998a) found only 20% favorable alleles from the wild tomato species L. hirsutum among 121 QTLs localized.

The strength of the trait improvement can be taken as a further measure of the efficiency of the QTL detection. In all AB-QTL analyses published so far for tomato and rice, the total yield could be raised due to the presence of at least one favorable exotic QTL allele. The yield increases amounted to maximal values of 18% in rice (Xiao et al. 1996; Xiao et al. 1998) and of 17%, 34%, 15% and 27%, respectively, in four tomato studies (Tanksley et al. 1996; Fulton et al. 1997, 2000; Bernacchi et al. 1998a). In barley, the maximum yield increase was associated with the exotic Hsp allele at locus GMS89[4H]. The increase of yield at locus GMS89[4H] could be detected in all six environments and ranged from 2.7% to 10.9%, with an average of 7.7% across all environments. The failure to reach the level of yield increases observed in tomato and rice could be due to the wrong selection of the Hsp donor accession. However, the accession ISR101-23 was selected in a pre-test among ten candidate donor accessions due to its relatively high frequency of BC1 and BC2 lines revealing transgressive segregation for single plant grain yield (data not shown). Under this light, a stronger effect of other Hsp donor accessions on yield might be questionable.

An alternative explanation for the lower positive effect of Hsp alleles on yield compared with exotic alleles from tomato and rice could be the different breeding system. Whereas we have compared the performance of homozygous backcross lines derived from barley BC2F2 individuals, hybrids derived from BC2 individuals were compared in tomato and rice. Thus, it cannot be ruled out that the stronger effects on yield detected in tomato and rice are due to heterosis effects occurring in hybrids between elite and exotic germplasm. The test of this hypothesis would be difficult to conduct in barley since an effective hybrid system, which can produce sufficient numbers of test cross seeds for field testing is, to date, not available. It would, thus, be more appropriate to compare the performance of hybrids and homozygous lines derived from an AB-QTL population in tomato or rice where both populations can be produced simultaneously.

Although the favorable QTL effects of the Hsp donor accession ISR101-23 are less pronounced than the effects from exotic donors in previous AB-QTL analyses in tomato and rice, the first application of the method to barley can be regarded as a success. In order to validate the significant effects presented in this study, we are currently producing NILs. By means of marker-assisted backcrossing and selfing, we generate, for instance, NILs from selected BC2F2-lines which harbor the yield increasing Hsp alleles around the SSR loci HVM40[4H] and GMS89[4H]. Due to recombination events, the resulting NILs should possess only small Hsp segments in the vicinity of the selected SSR locus. The NILs will be exploited for the validation of the original favorable Hsp effect and, as pure introgression lines, can be utilized for further breeding cycles. Simultaneously, the NILs can be utilized for high-resolution mapping of the region of interest, ultimately leading to a map-based cloning of the QTL factor. Both strategies have already been carried out in tomato. For example, Bernacchi et al. (1998b), Monforte and Tanksley (2000) and Monforte et al. (2001) produced detailed high-resolution maps of introgressed exotic tomato segments based on older AB-QTL analyses and validated the detected exotic effects in refined QTL-NILs. Furthermore, the production of a high-resolution map has already led to the first cloning of a QTL factor (Alpert and Tanksley 1996; Frary et al. 2000). In this case, it was shown that the QTL fw2.2, which controls the fruit weight of tomato, presumably acts as a negative regulator of cell division in the tomato carpel tissue.