Keywords

1 Introduction

The barley (Hordeum vulgare) genome comprises over 5 billion base pairs which equals about double the size of the maize (Zea mays) genome or twelve times the genome of rice (Oryza sativa). If one assumes an overall investment of 100 million dollars for the sequencing of the rice genome it is obvious that sequencing the barley genome was not fundable as long scenarios were based on classical Sanger sequencing (Sanger et al. 1977). Until recently the main sequence information of the barley genome has been obtained by sequencing expressed genes (Expressed Sequence Tags, ESTs, Zhang et al. 2004) or contiguous stretches of genomic DNA (Bacterial Artificial Chromosome (BAC) clone contigs) selected in frame of map-based cloning of important genes (reviewed by: Stein and Graner 2004; Krattinger et al. 2009; Eversole et al. 2009). Even at this limited access to the barley genome, important features of genome organization could be revealed and this has been reviewed before (Stein 2007). More recently, due to the availability of novel high-throughput sequencing technologies (Next Generation Sequencing, NGS, Mardis 2008; Holt and Jones 2008) costs for sequencing have tremendously decreased. This has stimulated and facilitated genome sequencing and genome analysis in many organisms including the major crop species. The present chapter reviews recent achievements in barley genome analysis that were exclusively triggered by the technical innovation provided by NGS.

2 Next Generation Sequencing: New Perspectives for Barley Genome Analysis

NGS originally referred to novel sequencing technology concepts that provided an alternative, cost efficient platform to previous “gold-standard” Sanger sequencing (Sanger et al. 1977), which relies on fluorescent-di-desoxy-terminator chemistry in combination with capillary electrophoresis and laser-detection devices. The first commercial systems entered the market as “short read” sequencing technologies providing sequence read lengths of between 30 and 100 bp (Service 2006; Holt and Jones 2008). Keeping in mind that these sequencing technologies were especially designed for “re-sequencing” of high-quality reference genome sequences (i.e. the human genome), it was unclear how useful such technologies would proof to be for de novo sequencing of large crop genomes like barley which used to lack any kind of reference sequence. This concern was based on the difficulties of the barley genome carrying 80 % repetitive DNA (Flavell et al. 1974) with the main constituent repetitive elements expanding each over several kilobases (i.e. the BARE1 element, Manninen and Schulman 1993). De novo sequencing and assembly of such structures would instantly lead to false assemblies and gaps. It was therefore important to critically assess the true potential of short read technologies and revisit regularly if their use would impact the strategy of barley genome sequencing.

2.1 Genome Survey and Genome Composition

Two early studies were designed to test the usefulness of 30 bp (Illumina Solexa) or 100 bp (Roche/454 GS20) short read technology, respectively, for skim sequencing the barley genome (Wicker et al. 2008; Wicker et al. 2009). Sequence depth was shallow in both cases reaching to between 1 % (100 bp GS20) and 10 % (30 bp Solexa) haploid genome equivalents. At this coverage any attempt of genome sequence assembly was rather meaningless. But the statistical properties of the datasets revealed interesting characteristics of the barley genome.

In the initial study (Wicker et al. 2008) the 30 bp short reads were utilized to generate an index of mathematically defined repeats (MDR index, Kurtz et al. 2008). For this purpose all high quality bases of the almost 270 Mio sequence reads were used to count the occurrence of all 20 mers (a sequence word of 20 consecutive nucleotides). 159 million discrete 20 mers (sequence differs at least at one nucleotide from all other 20 mers) could be determined and 88 % of the discrete 20 mers occurred only once in the dataset. Almost 99 % of the discrete 20 mers occurred only between 1 and 10 times in the barley genome. One percent of the discrete 20 mers occured eleven times or more often and thus constituted 30 % of the barley genome. One of these frequent 20 mers was present in almost 170,000 copies. Together this analysis provided a good reflection of the repetitiveness of the barley genome (Wicker et al. 2008).

In the second study the genome was sequenced to 1 % haploid genome coverage with 100 bp reads of the early Roche/454 GS20 technology platform (Wicker et al. 2009). The resulting 570,000 sequence reads were systematically compared against databases in order to determine the fraction of sequences that could either be related to genes or known repetitive DNA elements of Triticeae genomes. Interestingly but not unexpectedly, similarity to genes was detected only in less than 1 % of the entire dataset. 50 % of the sequence data could be assigned to only 14 families of transposable elements (TE) with the BARE1 family alone representing about 13 % of the barley genome. This is a 5–10 times higher frequency compared to its original copy number estimates (Manninen and Schulman 1993; Vicient et al. 1999). Sequences that could not be immediately related to any known class of DNA (TE, genes, SSR, organellar genomes) were assembled. Based on the shallow sequence depth resulting sequence clusters could per se be classified as repetitive sequence. Altogether 70 % of this “snapshot” sequence dataset comprised repetitive DNA. Based on the assumption that the dataset was representative for the overall composition of the barley genome it was compared to the sequence composition of individual BAC clone sequences from public databases. Such “gene-containing” BAC clones exhibited significantly different sequence compositions. Caspar transposons were over- and BAGY2 retrotransposons were underrepresented on the analysed set of BACs. The genome-wide distribution of these two TE classes was monitored by fluorescent in-situ hybridization (FISH) to barley metaphase spreads. The Caspar elements were predominantly clustered at the telomeric ends of chromosomes whereas BAGY2 elements produced almost a mirror image labeling of chromosomes covering all the pericentromeric regions but avoiding subtelomeric parts. This analysis of sequence element frequencies in comparison to a genomic index thus confirmed the subtelomeric origin of the analysed BAC clones (Wicker et al. 2009).

2.2 Survey Sequencing of Sorted Chromosomes—A Gene Index of Barley

Barley genome size is a major disadvantage for genome sequencing. On the other hand, the size of its genome is an advantage for cytogenetic applications. This feature could be exploited for PCR-based detection of markers from micro-dissected chromosomes (Sorokin et al. 1994). The first “physical” map of all barley chromosomes was produced by PCR amplification of genetic markers from micro-dissected chromosome arms or deletion chromosomes (Künzel et al. 2000). The usefulness of the size of Triticeae chromosome was further demonstrated by the feasibility of purifying mitotic chromosomes. Chromosome suspensions obtained from synchronized root tip meristems can be utilized for flow-cytometric sorting of over 90 % pure fractions of individual chromosomes (reviewed by Doležel et al. 2007). The Roche/454 GSFLX system was utilized to test whether such purified chromosomal DNA could be used for direct shotgun sequencing (Fig. 16.1). Chromosome 1H of barley was sequenced to about 1-fold coverage and the dataset provided partial sequence access to up to 80 % of all genes located on this respective chromosome (Mayer et al. 2009). By exploiting the extent of conserved synteny between barley and the sequenced genomes of rice and sorghum (International Rice Genome Sequencing Project 2005; Paterson et al. 2009), a potential linear order model of almost 2,000 barley genes detected in the shotgun sequences could be proposed (Fig. 16.1). These results stimulated a genome-wide analysis and all remaining barley chromosomes were sequenced to 1-fold coverage by using Roche/454 GSFLX Titanium (Mayer et al. 2011). The strategy previously applied to chromosome 1H was further reinforced by including into the analysis the information of a third sequenced model grass genome, Brachypodium distachion (The International Brachypodium Initiative 2010). A linear gene order map of more than 21,000 genes—perhaps two thirds of all barley genes—could be constructed with sequence tag access to all of these genes (Mayer et al. 2011). The strong enabling potential of survey sequencing of flow-sorted barley chromosomes was very convincing and stimulated similar studies on wheat chromosomes 1A, 1B, 1D (Wicker et al. 2011), 4A (Hernandez et al. 2011), 5A (Vitulo et al. 2011), 7BS and 7DS (Berkman et al. 2011a; Berkman et al. 2011b). The International Wheat Genome Sequencing Consortium has furthermore adopted the concept for wheat genome survey sequencing (www.wheatgenome.org).

Fig. 16.1
figure 1

Chromosomal genomics in barley. Barley chromosomes differ in size. This size difference can be exploited for the purification of individual chromosomes which provide, after enzymatic amplification of the minute amounts of DNA, an excellent template for Next Generation Sequencing. Sequencing to around 1-fold chromosomal coverage is providing access to about 70–80 % of all genes present on the respective chromosome. This data can be integrated on the basis of a dense gene-based marker map of barley and gene order information from syntenic regions of sequenced model grass genomes into linear gene order maps of individual barley chromosomes—so-called “genome zippers” (Mayer et al. 2011; Mayer et al. 2009)

Full size image

2.3 Next Generation Sequencing of BAC Clones

Shotgun sequencing of larger stretches of genomic DNA (i.e. the insert of an average sized BAC clone) of barley can be challenging even if Sanger sequencing is applied. Seldom would an entire insert of a BAC immediately assemble into one single contig. This most often is hampered due to the presence of multiple copies of highly conserved members of the same class of repetitive elements. Therefore, it was uncertain how sequencing and assembly of barley BAC clones would perform on the basis of NGS. It could be shown by re-sequencing a few BACs (previously sequenced by Sanger technology) that Roche/454 GS20 sequencing could provide more even sequencing coverage due to the lack of cloning bias (Wicker et al. 2006). Assemblies were more fragmented but sequence accuracy in the assembled regions was comparable to Sanger sequencing. The even read coverage helped to detect mis-assemblies in the originally Sanger sequenced and assembled BACs. Importantly, the genes present on the sequenced BACs almost exclusively assembled into single contigs comprising the entire gene structure (Wicker et al. 2006). This feature was exploited in a recent study of sequencing 400 gene-containing BAC clones of chromosome 3H from barley cultivar Haruna Nijo. Pools of 10 or 20 BACs each were sequenced and the resulting sequences were combined in a mixed assembly. This produced 7,512 contigs larger than 500 bp. In contrast to the 444 ESTs originally used for the identification of BAC clones overall 1,239 open reading frames (ORF) could be detected in the assembled sequence and thus assigned to chromosome 3H (Sato et al. 2011).

The large sequencing capacity provided by NGS platforms per individual sequencing run requires for multiplexing of samples if the advantage of cost efficiency should not be compromised. Several procedures were established all aiming at the incorporation of unique sequence identifiers (multiplex identifier, MID; barcode tags) into the fragments obtained from different samples while ligating specific sequencing adaptors (i.e. Meyer et al. 2008). Ninety-six barley BAC clones were sequenced in parallel to average 20-fold coverage in a single Roche/454 GSFLX run. Eighty percent of the analysed clones assembled into less than 10 contigs at N50 of 50 kb for all 96 clones. Superior assembly results may be expected if BAC clones would not just be shotgun sequenced but if paired-end or mate-pair sequencing strategies would be applied which provide two sequences per DNA fragment and their physical linkage information. This strategy has been introduced in the early times of genome sequencing (Roach et al. 1995). It can be implemented today also on all NGS platforms, however, together with individual sample bar-coding it comes at the disadvantage of substantially more labor. The effect of implementing information obtained from mate-pair sequencing has been simulated for barley by combining the de novo shotgun 454 sequence assemblies of the above mentioned 96 BACs with additional 2 × 36 bp sequence reads from a non-barcoded 2.5 kb mate-pair library of the same 96 BACs which allowed to scaffold 80 % of the assembled contig length (Taudien et al. 2011). Based on the outcome of Roche/454 barcoded BAC pool sequencing two projects for sequencing chromosome 3H in barley (http://barleygenome.org) and 3B (http://urgi.versailles.inra.fr/Projects/3BSeq) in wheat have been initiated.

2.4 Genotyping by Sequencing

Access to high-throughput sequencing has not only stimulated research towards whole genome sequencing. It can be predicted that many of today’s marker technologies will be replaced by one or the other kind of NGS application in the near/midterm future. Whole genome skim sequencing was used in rice for high-density SNP mapping in a population of 150 recombinant inbred lines (RILs). Each individual was sequenced to average 0.02-fold genome coverage which allowed scoring of 1.2 million SNPs at a density of 3.2 SNP/kb (Huang et al. 2009). To apply this approach economically to large genome species it is appropriate to implement steps of reducing genome complexity. In restriction-associated DNA (RAD) sequencing a fraction of restriction fragments produced with specific endonucleases is size selected and sequences are generated adjacent to these restriction sites (Baird et al. 2008). A high-density haplotype map was developed by such approach for maize (Gore et al. 2009) and other species (reviewed by Rowe et al. 2011). Two studies showed the applicability to barley. 10,000 RAD fragments were generated between the parental genotypes of the Oregon Wolfe Barley (OWB) mapping population. This included 530 fragments with codominant polymorphism between both genotypes of which 436 could be genetically mapped (Chutimanitsakun et al. 2011). This approach was extended further by including presence/absence polymorphisms to the linkage analysis. For the same OWB population ~ 24,000 sequence tags could be mapped into the existing framework map comprising already 2,382 markers (Elshire et al. 2011).

3 Whole Genome Sequencing of Barley—the Challenge Shifts from Sequencing to Assembly

Whole genome sequencing in barley is feasible now because of the possibilities provided by NGS. The pure sequencing costs for the entire barley genome likely amount to less than 3 million EURO for a multiplex barcoded BACpool sequencing project with Roche/454 GS FLX + (60,000 BACs, average insert 100 kb, 50 EURO/BAC) or to as little as 50,000 EURO for 100-fold whole genome shotgun sequencing with Illumina Hiseq2000 technology. Costs of labor and bioinformatics required for assembly and annotation have not been considered and would add up. However, having these different options in mind it is important to consider the quality of sequence that is aimed for. Recently published genome sequences produced by NGS often suffer from limited quality, hence they do allow only for limited genome wide studies and conclusions (Chain et al. 2009; Feuillet et al. 2011). The International Barley Genome Sequencing Consortium (IBSC, http://barleygenome.org) has proposed a multistep genome sequencing strategy which implements the advantages and quality provided by whole genome shotgun and hierarchical BAC-by-BAC sequencing utilizing a densely anchored physical map as a template (Schulte et al. 2009).

Since generating sequence data is principally no longer a limitation, the main challenge in genome sequencing shifted to sequence assembly. In any shotgun sequencing approach (WGS and hierarchical clone-by-clone shotgun sequencing) a genome needs to be reconstructed from all sequence reads by connecting overlapping reads (Miller et al. 2010). An assembly can be divided into three phases (Fig. 16.2): The first phase is the construction of contigs where a contig is a set of reads that are inter-related by overlap of their sequences (Staden 1980). All reads belong to one and only one contig. Each contig contains at least one read. In the second phase these unordered contigs are then ordered and directed. Results of the second phase are called scaffolds where a scaffold is a set of directed and ordered contigs, but gaps between such contigs are allowed. In a third phase, gaps between contigs within scaffolds are closed and the number of scaffolds is minimized. The result of phase three is a reference sequence ideally reflecting the sequence of the real genome.

Fig. 16.2
figure 2

Schematic figure of a reads, b contigs, c a scaffold and d a finished sequence. Sequence reads are provided as an output of a sequencing machine. They are used in the first phase of an assembly to produce unordered contigs. The second phase of the assembly results in scaffolds. The order and mutual direction of contigs within one scaffold is known, but gaps between the contigs are allowed. Ideally the distance of contigs is also known. In a third phase the gaps between contigs are closed

Full size image

The main challenges of whole genome sequence assembly in barley are imposed by genome size and content of repetitive DNA. Different algorithms for assembly may not necessarily cope equally well with both issues. A collection of current assembly tools for NGS data is listed in Table 16.1. Assembly softwares usually run a graph-based data structure internally to compute contigs. The traditional graph structures are overlap-graphs: a node in a graph represents a sequence read and an edge between two nodes represents an overlap between the two reads (Fig. 16.3a). The overlap-graph is read-centric, containing as many nodes as reads are used. The assembled consensus contig sequence is then extracted from the graph. This may be achieved by applying weight to edges according to overlap length and include edges subsequently starting with the largest overlap and discarding every edge that would result in a circular sub-graph (Myers et al. 2000). Graph-based assembly tools like MIRA or NEWBLER proved to be efficient for highly accurate assembly of sequenced barley BAC clones (Steuernagel et al. 2009; Taudien et al. 2011; Wicker et al. 2006). At a size of 5 Gbp, barley whole genome sequencing to 50-fold coverage with 100 bp reads would produce 2.5 billion reads. Since the overlap-graph algorithms require every-read to every-other-read comparisons this results in a quadratic problem of necessary computing steps. Additionally all reasonable read-pairings have to be stored. This quickly exceeds the memory capacity of even a super-computer (~ 1 Terabyte RAM) and basically excludes overlap-graph based assembly algorithms for whole genome assembly in barley. This bottleneck may not occur if the construction algorithm and storage data-structure is optimized dramatically. An example for such an optimized solution is maybe provided by the optimized SGA-assembler (Simpson et al. 2009). Another alternative is provided by algorithms using a de Bruijn graph data structure (Fig. 16.3b). Here a reduction of sequence dataset complexity is achieved by extracting k-mers (a short sequence of length k, where k is a positive integer). Every node in a de Bruijn graph is a unique k-mer that occurs at least once in the input sequence dataset. An edge is always drawn between two nodes if the first node’s suffix of length k-1 is equal to the second node’s prefix of length k-1. Since each k-mer must only occur once in the graph, the number of nodes depends on the number of different k-mers in the genome but not on the number of input reads. This diminishes the dilemma of NGS to produce short reads and high coverage. However, the problem to extract consensus contigs from a de Bruijn graph structure is far more complex than from an overlap graph. Tracking the number of occurrences of each k-mer in the input reads helps to resolve loops in the graph, since it is used to infer the number of occurrences of each k-mer in the consensus contigs.

Fig. 16.3
figure 3

Examples for an overlap graph a and a de Bruijn graph b. In an overlap graph every node represents one read from a sequencing machine. An edge between two nodes is drawn if the reads overlap. The edge can be weighted according to the degree of overlap. In a de Bruijn graph each occurring k-sized substring of any read is represented as a node but the substring of each node is unique. Only the number of occurrences of the substring can be tracked for each node. Thus the complexity of this graph does not increase if sequencing with a larger depth. Numbers in the graphs help to extract consensus contigs. In a the weight of a node represents the degree of overlap between reads. Circular structures in the graph are resolved by deleting edges with the least overlap. In b the numbers below each node track the number of occurrence in the input sequence. In this simple example the sequence can be constructed from the graph by following the directed graph until each node is passed as many times as its occurrence is denoted

Full size image
Table 16.1 Selection of current state-of-the-art NGS assembly softwares for de novo genome assembly
Full size table

Still the number of assemblers that can compute barley genome sized datasets is small. The commercial software CLC assembly cell (www.clcbio.com) is able to process such a dataset with less than 250 Gb of Memory (A standard personal computer is equipped with 4 Gb). Such an assembly of the barley genome has been made recently available providing direct access to most of the barley genes and for over 20,000 genes with transcript evidence also a genetic and physical position in the context of the barley genome physical map (The International Barley Genome Sequencing Consortium 2012). SOAP Denovo (v. 1.05) is an alternative assembler that can process barley data on a comparable computer up to a sequencing depth of 30 fold coverage (own unpublished data).

The repetitive nature of the barley genome causes the second challenge to sequence assembly since highly conserved multiple copies of repetitive DNA are a major cause of mis-assemblies. A mis-assembled sequence differs from the original DNA and mainly two classes of mis-assemblies occur. In the first case two pieces of sequence are tied together on the basis of partial sequence identity although they do not truly overlap physically. This may apply to two low copy sequences situated adjacently to highly conserved copies of the same class of TE. The second case results from collapsing two or more (almost) identical sub-sequences producing a shorter consensus sequence than in the original DNA. This situation may occur at highly conserved tandemly repeated sequences (i.e TE or genes). The main strategy of preventing repetitive DNA based mis-assemblies builds on the use of paired end sequencing. All advanced assemblers have implemented the inclusion of paired-end information for correct contig construction and most times also its incorporation for immediate scaffolding. Additionally paired-end data can be exploited to validate finished assemblies and detect mis-assemblies (Phillippy et al. 2008). These strategies could successfully be implemented in mammalian whole genome shotgun sequencing projects. The genome of giant panda (size: 2.6 Gb) was sequenced using Illumina including paired-end (and mate pair) libraries of varying insert sizes from 150 bp to 10 kb. Half of the assembled sequence could be covered with scaffolds (N50) larger than 1.3 Mb (Li et al. 2010). The assembly software ALLPATH-LG was introduced and tested on human genome shotgun data sequenced on Illumina (Gnerre et al. 2011). The software requires a specific variety of paired-end libraries reaching from overlapping read-pairs to large insert mate pairs. The N50 scaffold even reached a size of more than 11 Mb which is comparable to the quality previously obtained by Sanger-based shotgun sequencing of the human genome.

4 Outlook

Progress in barley genome analysis was accelerated by the availability of the different Next Generation Sequencing technology platforms. It could be demonstrated that sequencing the barley genome is feasible now due to the availability of NGS and the decrease in sequencing costs coming along with. Access to a relatively good quality draft sequence of barley has been provided recently. How much further such draft sequences can be developed into high-quality reference sequences will be depending on, however, the introduction of a further generation of improved sequencing technology (Metzker 2010). Such platforms will allow single molecule, real time, very long read sequencing thus helping to overcome the limitations of sequence assembly due to high repetitive DNA content. Until then, the new knowledge base obtained recently for the barley genome by high-throughput NGS is providing novel opportunities and tools to the research community for addressing questions of Triticeae biology and performance much more efficiently.