» go back

Bereich 1 - Details (Abstract)

GABI 2

GABI SEED 2

Barley as a model and a crop: Gene expression networks determining seed traits

Participants
Ulrich Wobus 1, 2
Winfriede Weschke 2
Hans-Peter Mock 2
Marion Röder 2
Bioinformatics Center Gatersleben-Halle at IPK, Udo Seiffert, Falk Schreiber

1 - Institut für Pflanzengenetik und Kulturpflanzenforschung (IPK)

2 - Project-Coordinator: Prof. Dr. Ulrich Wobus
IPK Gatersleben
Tel.:     +49-39482-5 219
Fax:     +49-39482-5 500
Email:  wobusuipk-gatersleben.de

Overall goals of the project

The project aims at a comprehensive understanding of the gene expression networks controlling seed traits. Based on a genetical genomics approach, a set of about 100 introgression lines representing the whole wild barley (Hordeum spontaneum) genome introduced into a spring barley (H. vulgare, variety Brenda) background will be characterized by transcriptional, protein and metabolite profiling. The experimental set-up allows large-scale calculations of e(xpression)QTLs, based on a ca. 12,000 unigene array of exclusively seed-expressed genes, but also calculations of QTLs for proteins and metabolites. Coupled with large-scale gene mapping and available data of QTLs of agricultural traits, a unique data set will be created on trait-related genes and pathways for present and future molecular breeding projects. In addition, regulatory networks of seed development will be defined. All together the expected results will enable the development of cause-related molecular breeding strategies involving genetic engineering and/ or marker-assisted selection with markers derived from the project.

Scientific and technical goals

In phase I of the GABI program different genomics tools have been developed like the 12k macro array of seed-expressed genes (Hv12K seed array). This array and two sets of introgression lines containing wild barley genome segments in a spring barley background (Fig.1)  are instrumental in an experimental approach centered around Genetical Genomics and the QTL concept.

Fig. 1      Genome structure of 44 introgression lines covering the whole H. spontaneum genome. 16 of the lines are single insert lines. Horizontal numbers: lines. Vertical numbers: introgressions.

We aim at the following goals:    

Actual (June 2005) technical and scientific results

Mapping of transcription factors

From 173 putative transcription factors annotated by homology search using the specialized database Transfac, 193 primer pairs for SNP detection were developed and tested within eight different barley lines. The found SNPs are converted into markers for pyrosequencing for genotyping within mapping populations.

Expression analysis

Gene annotation. Genes on the 12k macro array filter were separated into functional categories. The functional categories are hand-curated and mapping bins derived according to the MAPMAN vocabulary (co-operation with M. Stitt/ MPI Golm). The MAPMAN tool is adapted now for investigation and visualization of expression analysis results of barley grain development generated by using the 12k filter.

Expression analysis of seed development in the parent line ‘Brenda’. A detailed expression analysis of data by a combination of K-mean and self-organizing map (SOM) clustering methods revealed six clusters of tissue- and development-specific expressed genes (Fig. 2). Analysis of regulatory complexes out of these clusters is under way.

Fig. 2      Tissue- and development-specific gene expression pattern of ‘Brenda’ caryopses. mRNA expression during seed development (0-26 DAF) in different tissues (maternal, endosperm, embryo) was analysed using the 12k cDNA macro array filter. The resulting signal intensities were grouped by combination of k mean and SOM clustering. Signal intensities are represented in different colours: blue - low, yellow - average and red - high signal intensities (see also the colour scale on top). Cluster groups and sub-clusters are numbered at the right-hand site.

Bioinformatics

Identification of clusters of gene expression profiles. In order to increase the reliability of the analytic results, alternative computational methods have been implemented and examined. It has turned out that standard analysis should be separated into two stages, rough profile pre-clustering as obtained by k-means or the Multi-Dimensional Scaling (MDS) clustering method, followed by a more sensitive fine clustering of the roughly pre-clustered subset. For sensitive clustering, the Self-Organizing Map (SOM), as an unsupervised trained artificial neuronal network, is a useful tool. Fig. 3 shows an example of fine-clustering of a set of 808 expression profiles pre-selected from the total number (11,786).

Fig. 3      Self-Organizing Map (SOM) clustering of expression data from a subset of 808 genes being up-regulated during barley endosperm development.
Left panel: U-Matrix of a SOM with 12x12 neurons arranged in hexagonal neighbourhood. A hexagon containing a black dot represents a single neuron while all others show the relation to its direct neighbours. The grey level of a hexagon indicates either the variance (the lighter the grey the smaller the variance) of all input vectors mapped to a particular neuron during the self-organizing training phase or the pairwise similarity to its neighbours (the lighter the grey the more similar the input data points mapped to a pair of neurons). During the training of the map all expression profiles are sorted onto the map according to their similarity. This way, clusters (light grey areas) A-F of profiles of co-expressed genes can be identified.
Right panel: Clustered profiles of co-expressed genes for regions A-F have been extracted and aligned for visualization. After alignment, the graphs (ordinate) reflect the overall profile characteristics rather than the expression level. Furthermore, the graphs have been vertically stretched to allow visual separation. Since, for instance, the profiles of cluster A and F are most dissimilar, these clusters are consequently located on the diagonal of the SOM. In contrast, very similar clusters, such as B and C, are adjacent. There is also a light warping to be observed between clusters D and E.

The selected profiles characterized by a strong increase of expression between 4 and 6 DAF, fall into six distinct SOM clusters (right panel). The SOM clusters represent candidate genes, which possibly belong to different regulatory pathways. Comparisons to functional annotation available for all genes on the filter is under way.

Identification of genes being responsible for the constitution of the clusters. Apart from clustering but based on its results, criteria can be derived to explain specific class memberships. One of the methods, the Supervised Relevance Neuronal Gas (SRNG) method, can be helpful to explain the importance of groups of genes causing the genetic difference between developmental stages during grain development (Strickert et al., 2005).

Strickert, M.; Sreenivasulu, N.; Weschke, W.; Seiffert, U.; Villmann, T.: Generalized Relevance LVQ with Correlation Measures for Biological Data. 13th European Symposium on Artificial Neural Networks, Bruges, April 2005, peer reviewed, accepted for oral presentation.

Proteom analysis

For the extraction of the water-soluble fraction from seeds of H. spontaneum cv HS213, H. vulgare cv Brenda as well as from the introgression lines, the method of low salt extraction after Finnie et al. (2002) was used. The spot pattern of water-soluble protein extracts were compared for both the parental lines Brenda and HS 213, and about 800 spots in both gels were detected (Fig. 4). In direct comparison with each other, 63 spots in Brenda and 80 spots in HS 213 were found showing a two-fold or higher difference in expression level.

Fig.5       Two-dimensional gel protein patterns of barley cv Brenda (A) and HS 213 (B) seeds. To demonstrate differences in the water-soluble protein fraction, close-up views from A and B are shown in C and D, respectively.

Up to now, the protein patterns of about 40 introgression lines were examined relative to the corresponding Brenda controls. The average differences between the introgression lines and Brenda are between 30 and 40 spots (Fig. 6).

Fig. 6      Comparison of the water-soluble protein fraction between Brenda (A) and line 8-2/5 shown as an example (B). Spots with distinct differences in normalised spot volume are marked.

Report prepared by Winfriede Weschke and Ulrich Wobus, June 6, 2005