Genome-wide association mapping and genomic prediction of Fusarium head blight resistance, heading stage and plant height in winter rye ( Secale cereale )

Rye is a multi-purpose cereal crop grown in Central and Eastern Europe as well as in Western Canada. Fusarium head blight (FHB) is one of the diseases that have a severe negative impact on rye, but knowledge about FHB resistance at the genomic level is totally missing in rye. The objective of this study was to elucidate the genetic architecture of FHB resistance in winter rye using genome-wide association (GWA) mapping complemented by genomic prediction (GP) in comparison with marker-assisted selection (MAS). Additionally, plant height and heading stage were analysed. A panel of 465 S 1 -inbred lines of winter rye was phenotyped in three environments (location– year combinations) for FHB resistance by inoculation with Fusarium culmorum and genotyped with a 15k SNP array. Significant genotypic variation and high heritabili ties were found for FHB resistance, heading stage and plant height. FHB did not correlate with heading stage, but was moderately correlated with plant height ( r = −.52, p < .001) caused by some susceptible short inbred lines. The GWA scan identified 15 QTL for FHB resistance that jointly explained 74% of the genotypic variance. In addition, we detected 11 QTL for heading stage and 8 QTL for plant height, explaining 26% and 14% of the genotypic variance, respectively. A genome-wide prediction approach resulted in 44% higher prediction abilities than marker-assisted selection for FHB resistance. In conclusion, genomic approaches appear promising to improve and accelerate breeding for complex traits in winter rye.

therefore strictly regulated in the European Union. For rye and bread wheat, the same limits apply, being 1.25 and 0.1 mg/kg for DON and ZON, respectively, in unprocessed lots for human consumption (The Commission of the European Communities, 2006). In bread, the maximum allowed levels are 0.5 mg DON/kg and 0.05 mg ZON/ kg. For feed, different guidance values are recommended, of which the lowest is for pigs with 0.9 mg DON/kg. In naturally infected rye grains from Denmark, F. graminearum, F. culmorum, F. avenaceum and F. poae dominated among the Fusarium species (Nielsen et al., 2011).
Integration of resistant cultivars into other disease management practices such as crop rotation and good soil tillage is an efficient, cost-effective and ecologically safe method of reducing the impact of FHB in cereals. Hence, breeding for FHB resistance in rye is crucial given its use as bread cereal with 22% of the harvest and for feed with 64% of the harvest in Germany (BLE, 2018).
FHB resistance in rye is quantitatively inherited and mainly governed by additive gene action, similar to the other cereal species, with a large genotypic variation in breeding populations (Miedaner, Borchardt, & Geiger, 1993;Miedaner & Geiger, 1996). Highly resistant material, however, can rarely be found in existing nurseries.
Genotypic correlation coefficients between FHB symptoms and DON showed a tight association (r = .8-.9), allowing an indirect selection for a reduced DON content by selecting for high FHB resistance (Miedaner, Wortmann, & Geiger, 2003). Genotype-by-environment (G × E) interaction played a major role, illustrating the necessity of selecting in several environments (location × year combinations).
The genetic architecture of FHB resistance has been investigated in bread wheat, durum wheat and to some extent in triticale, but no information is available for rye. In bread wheat (2n = 6x = 42, genome composition AABBDD), about 550 QTL located on all chromosomes were reported for FHB resistance that could be reduced to 65 meta-QTL (Venske et al., 2019). Some major QTL, that is Fhb1, Fhb5 and Fhb6 from Chinese wheat, have higher effects on FHB resistance (Bai, Su, & Cai, 2018), and attempts are being made to introduce Fhb1 into European durum wheat breeding programmes (Prat et al., 2017). For triticale, several minor QTL for FHB resistance were reported on rye chromosomes (Dhariwal et al., 2018;Galiano-Carneiro, Boeven, Maurer, Würschum, & Miedaner, 2019;Kalih, Maurer, & Miedaner, 2015).
In rye, genomics is still lagging behind other small-grain cereals.
A few previous studies reported QTL for agronomic traits such as plant height (PH), flowering time, yield-related and quality traits (Falke, Wilde, Wortmann, Geiger, & Miedaner, 2009;Hackauf et al., 2017;Miedaner et al., 2012Miedaner et al., , 2018, frost tolerance (Li et al., 2011) as well as drought tolerance (Myşków et al., 2018). However, no QTL or genome-wide association study (GWAS) has been reported in rye for FHB resistance. Because FHB resistance is generally caused by many QTL with minor effects, genomic selection (GS) might be more appropriate for improving the trait. GS utilizes genome-wide marker data to predict the genotypic values of individuals to be selected, thus reducing phenotyping once the marker effects have been estimated (Edwards et al., 2019). GS methods were applied in rye for kernel weight and quality traits in two introgression libraries (Mahone et al., 2015) and two bi-parental populations (Schulthess et al., 2016;Wang et al., 2014) as well as diverse breeding material (Bernal-Vasquez et al., 2014). A comprehensive genetic map based on a large SNP array is meanwhile available for rye (Bauer et al., 2017). For FHB resistance, GS yielded cross-validated prediction accuracies of 0.59 to 0.95 in bread wheat (Mirdita et al., 2015;Rutkoski et al., 2012), durum wheat (Miedaner, Herter, Ebmeyer, Kollers, & Korzun, 2019;Miedaner et al., 2017) and triticale (Galiano-Carneiro et al., 2019). Therefore, it is worthwhile to assess the prospects of GS in winter rye as the only out-crossing small-grain cereal.
Our objectives were to (a) assess the genetic variation for FHB resistance and associated traits in rye, (b) identify QTL for FHB resistance by GWA mapping and estimate their effects, (c) investigate their co-localization with QTL for heading stage and plant height, and (iv) compare the potential of marker-assisted selection and genomic prediction (GP) to improve breeding for FHB resistance in winter rye. For this, a large population of 465 rye inbred lines was analysed by inoculation with F. culmorum.

| Plant materials and field experiments
A panel of 465 rye (Secale cereale L.) S 1 lines from the company HYBRO Saatzucht GmbH & Co. KG were used for the study. The lines descended from the 'Carsten' heterotic pool that is used as pollinator pool and were made up of 372 lines that were selected for FHB resistance in a recurrent selection (RS) programme across five cycles. The RS procedure was a typical S 1 -line testing (Lynch & Walsh, 1998) with a three-year cycle: (a) selfing and testcrossing of non-inbred materials, (b) multi-environment selection of FHB rating for inbred line and testcross performance, respectively, with a weighted index of 3:1 and (c) recombination of the superior lines.
To widen genetic variation, 93 lines unselected for FHB resistance were added to the last RS cycle that was analysed here. All lines were near Wriedel, Lower Saxony in 2017. Entries were mechanically sown in single-row observation plots 0.8-1.2 m long at a sowing density of 270 kernels/m 2 . In HOH17 and WUL17, the experimental design used was α-lattice design with two replicates. Each replicate consisted of 54 incomplete blocks and 10 genotypes per block. To fill up the field design, standard lines were used. For the field trial in HOH18, a row-column partially replicated design was used because less seeds were available for some genotypes. The number of rows and columns was 40 and 23, respectively. Eighty-five per cent (85%) of the genotypes were replicated. All genotypes were treated with standard agronomic practices as described by Gaikpa et al. (2019). Genotypes were inoculated with one Fusarium culmorum isolate (FC46) at a concentration of 7.5 × 10 5 spores/ml using a tractor-driven sprayer. The inoculation begun at the onset of flowering of early genotypes and was repeated for 4-5 times at 2-3 days intervals to ensure that all entries were inoculated at least once at mid-anthesis.
The traits recorded included FHB severity, heading stage (HS) and plant height (PH). On plot basis, FHB severity was visually rated using a scale of 0%-100% of infected spikelets per genotype, starting from the onset of FHB symptoms differentiation (Miedaner et al., 2001). Two successive ratings were taken in WUL17, five ratings in HOH17 and four ratings in HOH18. HS was rated on 1-9 scale, where 1 = the ear/head of the crop still remain in the leaf sheath and 9 = ear stalk is at least 10 cm long under the ear or above the leaf sheath. Plant height (cm) was measured from the ground level to the tip of the heads after full flowering using a metre rule.

| Phenotypic data analysis
Two, four and five FHB ratings from WUL17, HOH18 and HOH17, respectively, were averaged to get mean FHB severity for each genotype per environment and used for all analyses. Adjusted means and variance components of each trait were calculated based on best linear unbiased estimation (BLUE) and best linear unbiased prediction (BLUP), respectively. ASReml package (Butler, 2009) within the statistical software R (R Core Team, 2018) was used for all phenotypic analyses. Because of the different field designs used in 2017 and 2018, a two-step analysis was done to get the adjusted means and variance components. At the first step, means for each trait in WUL17 and HOH17 were estimated separately using the model: where Y ijk = the observed phenotypic mean for genotype i in replicate j and block k, μ = general mean, G i = effect of the ith genotype, R j = effect of the jth replicate, B jk = effect of the kth block in the jth replicate and e ijk = residual error.
For HOH18, means were estimated using the model: where Y ijk = the observed phenotypic mean for genotype i in replicate j, row k and column l, μ = general mean, G i = effect of the ith genotype, R j = effect of the jth replicate, W k = effect of the kth row and C l = effect of the lth column and e ijkl = residual error.
Row, column and genotype were treated as fixed effects and blocks and replicates considered as random effects. Adjusted entry means and corresponding standard errors of genotypes from each environment were analysed in the second step to obtain genotypic means across environments by using the following mixed model: where Y ij = the observed phenotypic mean for genotype i in environment j, μ = general mean, G i = effect of the ith genotype, E j = effect of the jth environment, GE ij = effect of genotype-environment interaction and e ij = residual error.
A weighting factor of one divided by the squared standard error of each mean from the first step was used, so the residual variance was set to one, according to method 3 proposed by Möhring and Piepho (2009). BLUEs were estimated across environments assuming fixed effects for the genotypes and environments. Variance components were determined by the restricted maximum likelihood (REML) method. Genotype, environment and genotype-environment interaction were treated as random..
Significance of variance components was determined using the likelihood ratio test.
Broad sense (entry-mean) heritability (H 2 ) was estimated based on the generalized method proposed by Cullis, Smith, and Coombes (2006) as follows: where vBLUP is the squared average standard error of difference of the BLUPs and 2 g = genotypic variance. Phenotypic association between FHB and heading ratings as well as FHB and plant height were estimated by Pearson correlation tests using the "cor.test" function in the R statistical software (R Core Team, 2018).

| Molecular data analysis
Fresh leaves were collected from the 465 rye lines at four-leaf stage, and genotyping was performed by a commercial laboratory by Illumina Technology (Illumina, San Diego) with a 15 k in-house single nucleotide polymorphism (SNP) chip yielding 8,942 polymorphic SNPs. We checked the quality of the markers using the "check.
marker ( )" function in the R package GenABEL (Aulchenko, Ripke, Isaacs, & van Duijn, 2018) and removed SNPs which showed more than 20% missing genotypes or had a minor allele frequency <5% from further analyses. In the end, 7,728 SNPs were available for the genome-wide association mapping across the 465 genotypes. Only 2,719 SNPs in our marker data overlapped with the markers in an already published linkage map (Bauer et al., 2017). To increase the number of mapped markers for our study, we established a con- A genome-wide association scan was performed to analyse marker-trait associations for FHB resistance, heading stage and plant height using the R package GenABEL (Aulchenko et al., 2018).
Principal component analysis based on the distance matrix of genomic kinship showed two major clusters in the association panel ( Figure 1). Therefore, both the genomic kinship matrix (K) and the first principal component were included in the linear mixed model of the "polygenic( )" function to correct for the confounding effects of family and population structure in the data set (Price et al., 2006;Würschum, 2012;Yu et al., 2006). The "ibs( )" function of GenABEL package was used to estimate the kinship matrix based on the SNPs.
We conducted the GWA mapping assuming additive effects of markers using the Q + K mixed linear model (Yu et al., 2006): where y = a vector of observed phenotypic means, Xβ = the fixed effects other than the SNP under testing and the population structure, β = a vector of fixed effects other than SNP or population group effects, α = a vector of SNP effects, v = a vector of population effects, u ∼ N 0,A 2 g = a vector of random polygenic background effects with A being the genomic relationship matrix of the lines and 2 g the additive genetic variance, e = a vector of residual effects, Q = a matrix from the structure relating y to v, and S, X, Z = incidence matrices of 1s and 0s relating y to β, α and u, respectively.
To control for multiple testing, significant SNP-trait associations were determined using a Bonferroni-corrected threshold of p < .05 (0.05/number of hypothesis tested) and in addition by an exploratory significance threshold of p < .0001. To identify the likely chromosomal position of unmapped significantly associated SNPs, we assessed the linkage disequilibrium (LD) between these markers and all mapped SNPs. We estimated the LD (r 2 ) values between the SNPs by applying the function "r2fast( )" which was based on a slightly modified code of Hao, Di, and Cawley (2007). A pair of SNPs having r 2 values >.60 were considered as being in LD. In addition, we used the LD values to correct all significantly associated SNPs for collinearity, that is to determine which of the significant markers likely identify the same putative QTL. The total proportion of genotypic variance (ρ G ) explained by the identified QTL was estimated as: where H 2 is the heritability of the trait, and R 2 adj is the adjusted R 2 (Utz, Melchinger, & Schön, 2000). The adjusted R 2 was obtained by fitting all significant SNPs simultaneously in a linear model in a decreasing order of the strength of their association with the trait, that is they were fitted beginning with the SNP that had the lowest P-value (Würschum, Langer, & Longin, 2015). The linear model can be represented as: where y is the calculated phenotypic mean, m i , m j and m k are the marker effects where the P-value of m i < m j < m k < m … (i.e. in a decreasing order of the strength of their association with y ijk ) and e is the residual error.
The ρ G of individual QTL was estimated by using the sums of squares obtained from the analysis of variance of the linear model including the significant SNPs (Würschum et al., 2015), that is: where SS m refers to the sums of squares of the individual SNP and SS total refers to the total sums of squares. In addition, we calculated the additive effect (α-effect) of each significant SNP by fitting only one SNP at a time in a linear model.
Furthermore, a genomic prediction (GP) approach was applied to exploit the additive effects of small-effect QTL which cannot be identified in the GWA mapping. The GP was conducted by ridge regression-BLUP (RR-BLUP) with the R package "rrBLUP" (Endelman, 2011;Endelman & Jannink, 2012) using imputed SNPs, including both mapped and unmapped markers. A weighted ridge regression-BLUP (wRR-BLUP) was also performed, where the significant SNPs from the GWA mapping, explaining more than 5% of the genotypic variance, were treated as fixed effect in the GP model (Spindel et al., 2016;Zhao, Mette, Gowda, Longin, & Reif, 2014).
Additionally, we compared the predictive ability of MAS and GP.
For each trait, the significant SNPs explaining >5% of the genotypic variance in the GWA mapping were used for MAS and all genome-wide SNPs were used for GP .

| Phenotypic variation among rye genotypes
The F. culmorum isolate FC46 caused FHB infection in all three environments with a slightly higher infection level occurring in HOH18 ( found that was mainly triggered by some very short susceptible lines ( Figure 1). For the selected lines, the correlation was considerably lower, although significant (r = −.22, p ≤ .001). They were, on average, 13.52 cm taller than the non-selected lines, but also 11.9% more FHB resistant (Table S2). Heading stage, by contrast, showed no substantial difference between the selected and the unselected subpopulations (5.37 vs. 4.95, Table S2).

| Genome-wide association mapping and genomics-assisted selection
Principal coordinate (PC) analysis showed two major population substructures reflecting the genetic background of the lines used in this study ( Figure 1). The first and second PC explained 60.6% and 15.2% of the variation, respectively. The larger group comprised of 372 S 1 lines selected for FHB resistance in a recurrent selection breeding programme across five cycles, and the smaller group comprised of 93 S 1 lines not previously selected for FHB resistance. As a result, we used the first PC and the K matrix to correct for population substructure and familial relatedness, respectively.
The GWA scan revealed ten SNP-trait associations for FHB severity on chromosomes 1R, 3R, 5R and 6R that exceeded the Bonferroni-corrected significance threshold with a P-value of 8.13E-06 (Figure 2). At the exploratory threshold (p < .0001), significant associations for FHB severity were found on all chromosomes except for chromosome 7R (Table 2, Figure 2). In total, 15 putative QTL were identified with this threshold for FHB severity which jointly explained 74% of the genotypic variance. Each QTL explained between 0.22% and 33.12% of the genotypic variance, five explaining more than 5% ρ G (  (Figure 3).
Three SNP-trait associations were identified for heading stage (1R, 2R, 5R) at the Bonferroni-corrected significance threshold (pvalue = 8.13E-06, Figure S1). At the exploratory significant threshold, we found significantly associated SNPs on all chromosomes except chromosome 6R (Table 2, Figure S1). Overall, 11 QTL were detected for heading stage and jointly explained 26% of the genotypic variances, with the ρ G of single QTL ranging between 0.01% and 12.02%. Two SNPs explained more than 5% ρ G for this trait.
Additive effects of the QTL ranged from −0.62 to 0.44 for heading stage. SNPs isotig12834 and isotig32608 showed an additive and a dominance effect, respectively, for heading time.
For plant height, one significant SNP was found at the Bonferronicorrected threshold on chromosome 2R. Three significant associations exceeding the exploratory threshold were identified on chromosomes 2R, 3R and 7R, with several significantly associated markers located on chromosome 3R (Table 2, Figure S1). In total, 8 putative QTL were identified for plant height with the exploratory significance threshold. These QTL jointly explained 14% of the genotypic variance and individually between 0.06% and 5.42% (Table 2).
Only two SNPs explained slightly more than 5% ρ G . Additive effects of the QTL for plant height ranged from −2.40 to 3.01. Both loci, iso-tig24773 and isotig23589, showed dominant allelic effects for mean plant height (Figure 3).
No common QTL were found for FHB severity and heading stage. The significantly associated SNP isotig18865 on chromosome 3R was common to heading stage and plant height. There was a high LD between SNP isotig 15,081 (FHB QTL) and SNP isotig24773 (plant height QTL) on chromosome 3R (r 2 = 0.84).
Because of the apparent presence of additively inherited minor-effect QTL contributing to the total genotypic variance of FHB severity, heading stage and plant height, we compared the potential of MAS and GS in a fivefold cross-validation procedure ( Figure 4). GS was clearly superior over MAS for all three traits. For FHB severity, the prediction ability of MAS approach was 44% less than the prediction ability of the two genomic prediction approaches. Similarly, genome-wide predictions were 42% and 63% higher than MAS for heading stage and plant height, respectively (Figure 4). A weighted GS approach, incorporating the identified medium-to large-effect QTL as fixed effects, did not yield a higher mean prediction ability than the non-weighted GS approach. Cross-validated prediction ability of the RR-BLUP procedure was 0.86 illustrated by a narrow correlation between observed and predicted FHB severities ( Figure S2).

| D ISCUSS I ON
Knowledge about the genetic architecture of FHB resistance is vital for genomics-assisted resistance breeding, but to date nothing ), residual error variance (σ 2 ) and entry-mean heritability (H 2 ) of these traits across three environments F I G U R E 2 Manhattan plot of the genome-wide association scan for Fusarium head blight (FHB) severity (%). Bon. = Bonferroni-corrected significance threshold at p < .05 and Expl. = Exploratory significance threshold at p < .0001 is known about QTL that confer this resistance in rye. The aim of this study was therefore to (a) perform the first GWA mapping to discover QTL that control FHB resistance in rye and (b) evaluate the potential of genomics-assisted selection for FHB resistance breeding.

| Construction of the GWA population
For European wheat, many QTL with small effects were reported to govern FHB resistance, as known for other quantitative traits (Lynch & Walsh, 1998). In the cross-pollinating rye, we expect many alleles TA B L E 2 Significant SNP marker detected for Fusarium head blight (FHB) severity, heading stage (HS) and plant height (PH) projected on the map of Bauer et al. (2017) and proportion of explained genotypic variance (ρ G ) and additive (α) effect per locus (Newell & Butler, 2013). Therefore, a bi-parental mapping population exploring only the effects of the two parental alleles of the population is not adequate to identify QTL for FHB. Moreover, performing a GWAS in unselected populations may not be able to identify effectively QTL present at low frequency. Therefore, we followed here the strategy to enrich our GWA population for FHB resistance by several cycles of recurrent selection. To demonstrate the genetic progress, we added 93 unselected S 1 lines to the 372 selected lines. On average, the selected lines showed an 11.9% higher resistance against FHB than the unselected lines (Table S2). Thus, this strategy can be expected to improve the chances of identifying QTL and superior combinations of resistance QTL alleles. Because all tested lines belonged to the pollinator gene pool, our population can serve as a training population for this heterotic group. Whether it can also be used for the opposite pool requires further research.

| Phenotypic variation for FHB resistance and agronomic traits
The rye lines analysed in this study showed a high variation for FHB resistance, heading stage and plant height ( Table 1) Obviously, in the long-strawed rye the correlation between FHB severity and plant height occurs only when short entries are included.
Heading stage varied among the genotypes but did not correlate with FHB severity. This result corresponds to a previous report in triticale (Miedaner, Kalih, Großmann, & Maurer, 2016) and can partly attributed to the synchronization of the plant developmental stage with the date of inoculation, which ensured that each genotype was inoculated at the optimal growth stage (mid flowering), thus reducing the confounding effects of heading stage on disease severity.
The finding here indicated that selection for FHB resistance did not affect the maturity period of the rye genotypes, hence simultaneous selection for high FHB resistance and earliness should be possible in this breeding material.
Entry-mean heritabilities were high for all traits, which is in part attributable to the high genotypic variation present in this panel, and similar to the values reported in previous studies in rye (Gaikpa et al., 2019;Hackauf et al., 2017;Miedaner et al., 2012;Wang et al., 2014).

| GWA mapping identified QTL for FHB resistance in rye
Generally, cross-pollinating, highly heterozygous rye cultivars are more resistant to FHB than self-pollinating homogeneous triticale, durum wheat and bread wheat cultivars (Arseniuk et al., 1999;Gaikpa et al., 2019;Langevin, Eudes, & Comeau, 2004). Therefore, analysis of the genetic mechanisms underlying the increased resistance in rye is of vital importance. For the first time, we performed GWAS to elucidate the genomic basis of FHB resistance in winter rye and identified 15 genomic regions that are associated with FHB resistance (Table 2). These 15 QTL jointly explained a rather high proportion of the genotypic variance (74%), which may in part be due to the accumulation of these QTL in the recurrent selection programme. Interestingly, we found two major-and several me-  (Kalih et al., 2015).

| QTL for heading stage and plant height in rye
Heading stage and plant height are important agronomic traits which might confer passive resistance to FHB in small-grain cereals (Mesterházy, 1995). Therefore, we took these traits into account in our GWAS to investigate their possible co-localization with FHB resistance QTL.
For heading stage, we found 11 QTL, but none of these QTL co-localized with the QTL for FHB severity, which is in line with the lack of phenotypic correlation between these traits. Here, only two of the significantly associated SNPs explained more than 5% of the  (Kalih et al., 2015).  (Table 2). Generally, major genes controlling plant height are not routinely used in rye breeding to date and the height seems to be controlled by a plethora of minor QTL as reported previously in rye (Miedaner et al., 2018;Miedaner, Müller, Piepho, & Falke, 2011) and triticale (Galiano-Carneiro et al., 2019;Kalih et al., 2015).

| The potential of marker-assisted and genomic prediction in winter rye
For all traits analysed, prediction abilities of both RR-BLUP and wRR-BLUP were by far higher than predictions based on markerassisted selection (MAS, Figure 4) that considers only QTL with medium to major effects. Overall, the mean prediction abilities of MAS ranged from 29% to 48%, while the mean prediction abilities of the genome-wide approach ranged from 72% to 86% for the three traits ( Figure 4). This implies that improvement of FHB resistance, heading stage and plant height by MAS will be slower compared to genomic prediction approaches. This result is in accordance with previous studies reporting higher prediction abilities for genomic prediction than for MAS in triticale (Galiano-Carneiro et al., 2019), bread wheat (Mirdita et al., 2015;Rutkoski et al., 2012) and durum wheat . However, in the present study, we observed a higher prediction ability for FHB resistance than reported in the earlier studies and even slightly higher than for heading stage and plant height. This is likely due to the continuous selection for FHB resistance, resulting in increased resistance allele frequencies for the QTL underlying this trait ( Figure 2). Thus, recurrent selection breeding schemes assisted by genomic prediction appear promising to improve rye resistance against FHB. It is worth to note that our prediction accuracies might be overestimated to some extent, because the training (80% of the lines) and prediction (20%) set were from the same population and have been tested in the same environments.

| CON CLUS IONS
Resistance towards FHB might become a trait of increasing importance in hybrid rye breeding. The observed phenotypic variation in elite germplasm is an important and promising prerequisite to gain breeding progress with respect to FHB resistance in rye breeding programmes. There is great potential to improve FHB resistance by genome-based approaches. For the first time, GWA mapping identified several significant marker-trait associations for FHB severity in winter rye of which two can be classified as major QTL. These are candidates for further analyses of FHB resistance to increase our understanding in resistance mechanisms in rye. No co-localization of QTL for FHB and plant height or rather heading stage was observed, which mirrors the moderate correlation between FHB and plant height. Genomic prediction yielded similar high prediction abilities with and without weighted data. Genomics-assisted recurrent selection appears as a promising tool to accelerate breeding for complex disease resistances in rye. These results encourage further research to study FHB resistance in rye hybrids. One opportunity in rye breeding is the reduced level of mycotoxins in the harvest compared to wheat that should facilitate the increase of rye productivity and consumer protection.

ACK N OWLED G EM ENTS
This study was partly financed by the German Academic Exchange Service (DAAD, Bonn, Germany) as a personal grant to David S.

Gaikpa (grant number 91650671), by the company HYBRO Saatzucht
GmbH & Co. KG and by a grant of the University of Hohenheim (TG77). We thank Ana Carneiro-Galiano and Paul Gruner of the State Plant Breeding Institute, University of Hohenheim, Stuttgart for sharing the R script for the genomic prediction and helping to construct a consensus map, respectively.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.