Genome-wide high-density genetic linkage maps specific to a species are both useful and essential for a number of reasons. These include the identification of marker-trait associations via linkage analysis and association mapping, for isolation of genes by map-based cloning, for comparative mapping and for exploration of genome organisation [29, 30]. In our study, we have created the first high-density consensus genetic linkage map integrating DArT marker data from six triticale mapping populations.
Evaluation of method and test conditions
In this study, the integrated genetic linkage map was calculated by using the joint data approach and regression mapping algorithm of JoinMap® 4 . For effective linkage map pooling (and bridging) the individual mapping populations should share some common genetic bases (populations possessing a common parent for example) or common statistics (similar linkage information), but genetic map construction also relies on genome variation at loci detectable by molecular markers . Our study was based on segregation data from six partially connected mapping populations derived from nine triticale elite breeding lines (Table 1). In reference to the first two principal coordinates the parental lines of each cross were derived from distinct quadrants (Figure 1), generating a maximum amount of genetic diversity. On the other hand, genetic similarity values between individual populations and the amount of markers found to be in common between them indicate a certain connectedness between the populations (Table 2). These findings clearly show that the plant material underlying this study meets both requirements for linkage map construction and pooling. That is, sufficient variation for polymorphic segregation data as well as common genetic bases for a statistically firm integration.
Genetic linkage maps are established by heuristical algorithms and statistical approaches and thus, have an inherent statistical error. This holds true for the estimation of the recombination frequencies as well as for the integration of data from several populations. Sample size is crucial for genetic map construction as it affects the power of linkage detection and the accuracy of recombination fraction estimation . Random variation and potential biological variation can cause differences in estimated pairwise distances between individual populations, particularly if populations are of small size . We investigated six mapping populations incorporating 114 to 200 progenies, and a total sample size of 911 individuals (Table 1) which is similar to recent studies in rye , sorghum , red clover  and barley . We observed differences in the estimation of recombination frequency between identical pairs of markers in different populations  and found that for intervals up to 10 cM this heterogeneity increased successively and declined again for larger genetic distances (Additional file 2). The extent of heterogeneity between our individual mapping populations was expected due to both the effect of random sampling and biological variation which is occurring regardless of common genetic bases among populations.
In conclusion, the mapping populations used in our study were well suited for reliable consensus linkage map building. However, our results also highlight the fact that consensus maps always constitute a compromise which must be kept in mind.
Distribution of DArT markers
During the development and assessment of DArT markers in triticale, a larger number of polymorphic markers originated from the rye genome and more wheat markers from the B genome . With 57.8% of markers mapped to the R genome, 25.5% to the B genome and only 16.5% to the A genome (Table 3) we could verify these results in an applied mapping experiment with a large number of populations and individuals. Similar results have been found in studies of wheat for DArT [10–12] as well as other marker types [36–39]. We, therefore, conclude that the bias observed in our study is not attributable to the mapping individuals or the type of marker. Instead it may be due to the design of the triticale DArT array and the number of markers originating from the different genomes, but is also likely to reflect the different polymorphic nature of the A, B, and R genomes . In our study, 1.8% of the DArT markers mapped to two different loci on the consensus map, but never on the same chromosome (Additional file 1). In terms of the ratio of markers that occur in a multicopy manner, our results agree well with those reported for hexaploid wheat (2%), barley (1.4%) and sorghum (1.8%) [10, 33, 35]. This may be attributable to the polyploid nature of hexaploid triticale having an impact on the accuracy of DArT markers due to alternative binding sites on homeologous chromosomes or may be ascribed to paralogous sequences. Molecular markers are known for their tendency to cluster, caused either by an unbalanced distribution of recombination events along chromosomes or an unequal representation of chromosomal regions on the genotyping array . In accordance with this expectation we found that DArT markers clustered in several chromosomal regions (Figure 2). A possible explanation for regions with higher marker density on chromosomes could be that recombination occurs more frequently in gene-rich regions  which are present in clusters comprising physically small chromosomal regions and account for only 5-10% of the wheat genome [41, 42]. The observed gaps in the consensus map may, on the other hand, be caused by identity-by-descent of the parental genotypes in these genomic regions. Taken together, clustering of tightly linked loci and gaps with low marker density in the consensus map either reflect the genetic situation in triticale or are due to specific properties of the applied DArT markers (e.g. complexity reduction step, or redundant clones). Further research including alternative high-density marker systems, e.g. SNPs, will help to addess this question.
Consensus map features
Marker coverage and genetic map density are influenced by many criteria such as genome length, number of markers, distribution of markers and crossovers in the genome, mapping population size and mapping strategy . As a result of the integration of datasets from six mapping populations our final triticale consensus map incorporated 2602 loci (74.1% unique) covering 2309.9 cM (Table 3). The previously published triticale genetic linkage map  comprised 356 markers (AFLP, RAMP, RAPD, SSR) and spanned 2465.4 cM. Thus, despite different marker types being used, these results agree well in regard to computed map lengths for the triticale genome.
Likewise our results are in good accordance with the reported map length of 2383 cM based on 339 DArT markers for the related hexaploid bread wheat (AABBDD) . In our study the linkage groups of the A and B genome together covered 1646.6 cM and those of the R genome 663.3 cM. Contrary to these results, the published genetic maps of durum wheat (AABB)  and rye (RR)  based on DArT markers spanned 2022 and 3144.6 cM, respectively. As mapping functions were similar in the studies, the disparities may be explained by a different number of markers mapped and/or the different mapping algorithms applied. Referring to unique markers, many appeared redundant in component maps and were just 0.1 cM apart after integration. This raises the question whether these loci are really distinct from each other or are just a product of statistics and therefore should not be regarded as unique in further QTL mapping studies.
The average density in our consensus map was one unique marker every 1.2 cM, and 98% of all intervals between adjacent loci were smaller than 10 cM. This is sufficient for QTL mapping and many modern genomics approaches . It may, however, be worthwhile amending larger gaps by target-oriented employment of additional markers. Thus, we conclude that through integration of datasets from six mapping populations we were able to improve both the density and the quality of the component maps up to the final high-density DArT marker triticale consensus map.
One possible method to assess the quality of a consensus map is to compare the locus arrangement of the consensus map, which was optimised at the multi-population level, with the arrangement of loci in the component maps (each one optimised independently) . A consistent order is hypothesised if the markers identify identical chromosomal locations and if there are no incorrect or missing scores . In addition, this colinearity comparison can identify chromosome rearrangements in individual populations. Our tests for colinearity resulted in an overall good consistency affirming the high quality of our consensus map (Figure 3). We also found, however, regions where groups of neighbouring loci showed identical order of loci but inversion within a linkage group or even positioning of a region with conserved order at the opposite end of the chromosome. Such inconsistencies were reported for other species including red clover  and sorghum . In triticale, chromosomal rearrangements are known to occur . Thus, these inversions of marker order or positioning could reflect real genetic events such as small chromosome rearrangements or, as they occur mostly after gaps, they could also be caused by statistical uncertainty due to many weak linkages contributing to the adjustment. Furthermore, marginal shifts in locus order were found in regions with highly dense markers. Similar results were reported before in several mapping experiments [32–35, 46]. Despite a certain heterogeneity of recombination frequencies between mapping populations this must mainly be attributed to the dependency of estimated gene orders on sample size . Especially for the high-density regions, extremely large mapping populations would be required to resolve the correct order of markers.
The colinearity plots revealed that respective linkage groups were generally longer in the component maps than in the consensus map and this effect was even larger in denser linkage groups. The application of different algorithms (maximum likelihood or regression approach) during component and consensus map construction has been reported to affect map lengths, despite the same mapping function [32, 34, 46]. Another explanation could be that the condensed map length may be the intended outcome of the addition of more markers during the integration process .
Segregation distortion is known to strongly impact genetic map construction and QTL mapping [26, 27] but distorted markers may also be beneficial for QTL mapping if handled properly . Whereas highly deviating markers cannot be placed in the respective component maps, they can be included in the consensus map through integration of unbiased information available from other populations without segregation deviation. Our experimental design with multiple segregating populations thus offered an excellent basis for the evaluation of segregation distortion and the mapping of segregation distortion QTL.
The component maps of two populations were completely lacking certain linkage groups (2A and 1R in EAW74, and 2R in EAW78) (Table 3). As the three linkage groups were well covered in other populations and the same markers were also positively scored in the populations with the lacking chromosomes we can exclude a scarcity of markers. The majority of the markers on these chromosomes (79 and 100% of the markers mapped on chromosomes 2A and 1R in the consensus map, respectively), however, showed significant segregation distortion in population EAW74. This illustrates the possible consequences of segregation distortion which not only affects genetic map distances and ordering of loci, but can even result in complete chromosomes being absent from genetic maps.
The populations underlying our study were five DH and one F2 population. The DH populations were produced by different methods, either by maize pollination of the oocytes (DH06, DH07, DH_LxA) or by microspore culture (EAW74, EAW78) and were not pre-selected for any trait. We observed segregation distorted regions caused by biological factors (regions of distorted markers which had the same skew direction, distinguishable from deviating loci scattered apart along the chromosomes which are likely due to genotyping errors or non-functional markers) only in DH populations and mainly in microspore culture derived populations (Figure 4). The patterns observed in segregation distortion regions can be explained by both the distance between the markers and the segregation distortion loci linked to them and the effects of those loci . Our test for segregation distortion in the microspore derived populations resulted in clusters of distorted markers on chromosomes 2B, 3B, 1R, 4R, and 7R. In studies of wheat DH populations, chromosome 2B was reported to harbour QTL responsible for green spot initiation and plant regeneration  and a different type of in vitro culture response in anther culture . Furthermore, QTLs located on chromosomes 1R, 4R and 7R were previously reported to have an effect on the yield of green plantlets from anthers in culture and embryo induction (7R) in triticale . Segregation distortion on chromosome 7R was, besides EAW74 and EAW78, also observed in DH06. This chromosome has been implicated in the selection of zygotes or female gametes in rye . To our knowledge no QTL affecting segregation distortion have been described for chromosomes 3B (EAW74 and EAW78) and 2R (DH_LxA) yet. Due to the consistent occurrence in both microspore derived DH populations we assume that this region located on chromosome 3B harbours a novel QTL responsible for in vitro or androgenetic response in triticale.
Epistasis refers to interactions between two or several loci [51, 52] and has recently been shown to contribute to segregation distortion [53, 54]. In accordance with this we observed epistatic interactions involved in segregation distortion (Figure 5). These epistatic interactions point towards selection for specific allele combinations for in vitro or androgenetic response in triticale.
The DH technology has become an indispensable part of both research and breeding of triticale and many other agronomically important crops. Depending on the effect of the segregation distortion locus, it can influence the allele frequencies on an entire chromosome, including all genes involved in the expression of important agronomic traits. If the segregation distortion locus and the QTL for the agronomic trait are linked in repulsion, the agronomic QTL will be underrepresented or in the most extreme case absent from the population. The same holds true for the introgression of traits, if the QTL are by chance located on chromosomes harbouring segregation distortion loci. The relatively high number of segregation distortion loci identified in our study highlights this problem both for research and applied breeding. Further characterisation or even identification of the nature of segregation distortion loci may facilitate solving these issues.