We used two large segregating populations to unravel the genetic architecture underlying ten important agronomic and quality traits in rye. The results of the QTL mapping study opens for the first time comprehensive insights into the potential of marker-assisted selection (MAS) in rye.
Field testing resulted in high-quality phenotypic data
The precise estimation of phenotypic values is an important prerequisite for detecting QTL with a high power. We used F3:4 lines and consequently, only 1.5 times the additive genetic variance was exploited and not 2 times the additive genetic variance as by using for instance recombinant inbred line populations. Due to the high inbreeding depression in the outcrossing species rye, testcross progenies of two elite line populations were used in this study. This allows exploiting only half of the total additive genetic variation, but takes into account that line per se performance is only of limited predictive value for hybrid performance in yield-related traits in rye . Dominance is a serious obstacle when testcross performance is used in mapping biparental populations. When the tester contributes a strong dominant allele, the effect of the allele contributed by the other parent (an inbred line) is undetectable. The limitations on the use of strong elite testers were discussed by [29, 30]. However, despite these limitations we observed significant genetic variation (P < 0.01) for all measured traits. The significance of genotype-by-environment interaction for all traits illustrate that multi-environmental phenotyping is indispensible. Entry-mean heritabilites reached 0.9 for plant height, TKW, and test weight in both populations and even grain yield had a heritability of 0.7. Lower heritabilities of some traits were mostly associated with a small, although significant amount of genotypic variation. Consequently, a high power to map QTL should be possible.
The parents in the Pop-A did not differ much in their testcross performance, which underscores a typical situation in elite rye breeding, which relies on crosses of the “best × best” type. In contrast, in Pop-B a superior (Lo115) and a lower performing (Lo 117) parent were crossed. Despite this, genotypic variance in Pop-A was similar (Figure 1) pointing towards a high importance of transgressive segregation. Obviously, both parents contributed different positive alleles at several loci for each of the traits. This is in accordance with quantitative-genetic theory assuming a large number of segregating loci with mainly additive effects for complex traits . The contrast of the mean testcross performance of the parents and the average testcross performance of their segregating progenies is a test for net epistasis across the genome . Consequently, the lack of differences between the mean performance of the parents and the progenies suggests that epistasis was not important in these populations. This can be explained by prevalence of additive gene action but might also be due to the fact that negative and positive effects at individual loci may cancel each other out.
Plant height in elite rye populations is not controlled by major QTL
Plant height is regulated to a large extent by dwarfing genes in elite wheat and barley populations [33, 34], because they have positive effects on elite wheat grain yield . The genetic architecture of plant height in rye detected in this study, however, is in sharp contrast to the genetic architecture of plant height in elite wheat and barley. Five to nine QTL located on all seven chromosomes were identified for plant height in both elite rye populations. None of these QTL had a large genotypic effect and the recovery rates in permutation tests were low although the trait displayed a maximum heritability (0.9) and high proportions of explained genotypic variances (61 and 77% in Pop-A and Pop-B, respectively). This clearly illustrates that plant height in rye is a typical quantitative trait with a lot of segregating loci. Accordingly,  detected 11 genomic regions significantly contributing to plant height among testcrosses with two rye introgression line libraries. The main cause for the difference between wheat and barley on one hand, and rye on the other is that in rye the stem is used as reservoir for water and carbohydrates, especially when abiotic stress occurs. Very often grain yield is associated with tall plant stature as demonstrated by the significant correlation between plant height and grain yield in Pop-B (r = 0.30, P < 0.01). Therefore, no dwarfing gene has been successfully used in commercial rye cultivars until now although such a gene has been described and mapped . Interestingly, we observed in Pop-B a QTL for plant height in the telomeric region of chromosome 5RL (QTL #8) where the Ddw1 dwarfing gene from a Russian source was located (Figure 4, ). This QTL had in our population, however, a much smaller effect (R2 = 18.9) than expected from a major gene although the recovery rate was 96%, perhaps reflecting a multi-allelic series at this locus. In conclusion, quantitative inheritance of plant height in rye is an example for a highly crop-specific trait.
Genetic architecture of yield and yield-related traits is generally complex, but some major QTL occur
Segregation at major effect QTL underlying grain yield in rye is not expected as any large-effect QTL would have been long fixed in the course of breeding. In agreement with this expectation, we observed for Pop-A where two high-yielding elite parents were crossed only one QTL explaining 5% of the genotypic variation (p
). In Pop-B deducted from a high- and a lower-yielding parent, seven QTL for grain yield were detected with p
varying from 10 to 24% and low recovery rates. High effects in most of the tested individual environments reveal their environmental stability (Table 4). Accordingly, cross-validated variance explained by all QTL amounted to R2
CV = 51%. Here, linkage blocks due to limited resolution of QTL mapping in biparental populations may have lead to detection of clusters of linked QTL, thus underestimating the number of QTL involved in complex traits and overestimating their effects as shown by  in a comparison of a conventional F3 and an intermated F3 population.
Population size used was about 220 lines per population. This is similar to the US-nested association mapping (NAM) population  in maize, but an even greater size might be valuable for quantitative traits. For grain yield,  detected two QTL by analyzing 244 testcross progenies of maize, but up to seven QTL when regarding 976 progenies. We can, therefore, expect that the number of QTL estimated here might represent the lower limit of QTL segregating.
In conclusion, grain yield follows an infinitesimal model as already proposed by  and MAS for individual QTL seems not to be a realistic option. Yield components might identify better candidates. Indeed in Pop-A, where grain yield had only a minimum cross-validated R2 of 3% only, TKW resulted in 64%. In Pop-B this value was similar high for both traits (>50%). Interestingly, in both populations individual QTL for TKW with large effects were found: QTL on chromosomes 6 and 7 in Pop-A as well as on chromosomes 5 and 6 in Pop-B.
The most prominent QTL for TKW on chromosome 7R had a remarkably high p
value of 48% and a recovery rate of 87% (Additional file 4). Interestingly, a major QTL on this chromosome was already reported in a Petkus population associated with the marker SCM40 . This marker was located in the centromeric region of the chromosome  and might correspond to our QTL #5 for TKW in Pop-A. A second locus for this trait  was associated with SSR marker WMS5 located in the telomeric region of chromosome 5RL (renamed as GWM5, ) and might correspond to QTL #2 in Pop-B. The two other loci with high p
values and recovery rates >90% were located in a similar region on chromosome 6 in both populations but contributed by different parents. These results illustrate that some QTL for TKW (#4 and #5 in Pop-A and #3 in Pop-B) have such high effects that they might be caused by single genes. Further fine mapping of multiple alleles per QTL in this region will be needed to test this hypothesis.
A similar localization of QTL was observed for some yield-related traits. Two QTL for grain yield on chromosome 1R (Pop-A) and 4R (Pop-B) had similar positions like each of one QTL for TKW and test weight reflecting the significant correlations between the traits. The corresponding QTL were contributed by the same parent in both instances. Thus, an indirect improvement of grain yield by MAS of individual QTL for yield components that have a higher recovery rate might be feasible.
For yield and yield-related traits we tested our mapping populations at ten environments including locations from North, East and South Germany as well as Poland. Compared to other QTL studies, this is a high number and broad range of environments and we report here only QTL with significant effects across all environments. This surely restricts the number of QTL, but enables us to detect only environmentally stable QTL that should be valuable also in environments not tested here. QTL for TKW and grain yield detected in the combined analysis had similar effects in most of the individual environments.
In summary, QTL effects for grain yield were mainly small as expected from theory, three QTL with high effects, however, were detected for TKW. They were highly stable across environments and had a high recovery rate in the cross validation and, thus, should be investigated further.
Quality traits are regulated by a complex genetic network
Two to nine QTL were detected for each of the quality traits. Their p
values were mostly >60%, starch content amounted even to 84% in Pop-B, although testing intensity was lower than for the yield-related traits. It should be noted that until now selection in practical breeding programs where the elite parents were derived from was solely based on falling number as a measure for pre-harvest sprouting. Despite its high heritability reported in earlier studies  it is a complex inherited trait [10, 11]. In this study, we found each of two QTL for falling number in both populations and the QTL on chromosome 4 RL might correspond to a QTL already described .
Generally, quality traits are regulated by a complex genetic network resulting in phenotypic and pleiotropic interactions among the traits as shown for the maize nested association mapping (NAM) population . Also in hybrid rye, significant correlations were detected among traits ranging from −0.2 to +0.6 (Table 1). In particular, QTL for starch content overlapped with QTL for grain yield on chromosomes 3R, 4R and 7R in Pop-B. Accordingly, both traits showed a significant (P < 0.01) phenotypic correlation (r = 0.32). On chromosomes 1R, 3R and 5R a QTL for test weight had a similar position than a QTL for starch content indicating that plumber kernels might have higher starch content. Significant (P < 0.01) correlations were indeed found between starch content and test weight. The negative correlation between starch and protein content in both populations (r = −0.6 and r = −0.7) is well known from other studies in cereals (Jansen, pers. commun.), but could not be explained by co-localization of QTL in our populations. A negative correlation between starch and TKW was also observed in durum wheat . The authors state that end-use of the kernel is clearly influencing the physio-chemical kernel characteristics implying properties such as milling quality, kernel hardness, and kernel protein content. In rye, a high selection pressure is on large, plump kernels, because rye as a crop tends to low kernel size limiting flour yield.
In conclusion, each of one QTL with large effects (p
> 20%) and high recovery rates (>90%) was found for test weight (QTL #2), falling number (#2) and starch content (#2) in Pop-A and starch content (#1) in Pop-B that should be considered for MAS.