The Genom of Homo sapiens.pdf

190 GIBBS AND WEINSTOCKBACATLASWGSFigure 3. Formation of eBACs by BAC-Fishing. Sequencingreads from a lightly sequenced (~1x coverage) BAC are used toidentify reads from WGS sequencing (5–6x coverage) that overlapand thus come from the genomic region of the BAC. The relevantWGS reads are co-assembled with the BAC reads usingPhrap to produce an eBAC, an assembly with much greater coverage(6–7x) than the original skimmed BAC sequence.quence (Myers et al. 2002; Waterston et al. 2002a, 2003).Even apart from these debates, the shortcomings of eachapproach are apparent. On the one hand, the hierarchicalapproach is laborious and requires the development of anintermediate product (the clone map) that is ultimatelyabandoned. On the other hand, WGS sequencing requiresextensive work to piece together the segments ofgenomes that are repeated over distances longer than onekilobase.To improve the sequencing of large genomes, wetherefore developed a new “Combined Strategy” and appliedit to the Rat Genome Sequencing Project (RGSP)(Rat Genome Sequencing Project Consortium 2004). Theunderlying principle is to combine the precision and orientationafforded by BAC clones with the ease and scalabilityof a WGS approach (Fig. 2). As a consequence, theRGSP was designed to include low-coverage (1.5x) sequencefrom a full set of BAC clones that covered the entirerat genome, as well as an abundance of WGS readsthat would provide deep overall sequence coverage. Therationale was that specific WGS reads could be recruitedto join the sequences that were derived from within BACclones. When brought together, these mixed reads couldbe assembled using our familiar software for handlingBAC-sized sequencing projects.This was the first, real-time combined use of BACs andWGS data to generate a complete genome assembly,since, although the previous Drosophila and concurrentmurine genome projects both used WGS and BACs, thesewere employed in sequential and separate phases. Wepredicted the BAC component of the Combined Strategywould be particularly important in the rat genome sequencingprogram, as it aimed to generate a “draft” ofabout 7x sequence coverage, but not finished sequence.In this case, the BACs were likely to confer an overallhigher quality on the eventual assembly.Figure 3 illustrates the process of BAC Fishing, the firststep and a key element of the Combined Strategy. At thisstage, the small numbers of “bait” sequence reads fromindividual BACs are used to identify larger numbers ofmatching “catch” DNA sequence reads from WGS libraries.Typically 500–800 BAC reads “fish” 2,500–3,000 reads from a 6x WGS pool. We developed BACfisher(Havlak et al. 2004) software that enabled this fishingprocess, and PHRAP was used for the subsequentstringent step of local assembly to form the “enrichedBACs” (eBACs) containing the two kinds of reads.The high quality of the data contained within eacheBAC validated the basic principle of the CombinedStrategy (Fig. 4). The eBACs were made publicly availableas a useful resource while the project progressed, andformed the foundation for the remainder of the rat assembly.To generate a complete rat genome draft, ~19,000eBACs were assembled into strings of BACtigs on the basisof their sequence overlaps. The BACtigs were nextjoined into super-BACtigs by large clone insert mate-pairinformation and ultimately aligned to genome-mappingdata to form the complete assembly (Fig. 5). More detailsof this assembly procedure are discussed elsewhere(Havlak et al. 2004).The overall quality of the rat assembly was very high,and the project therefore managed to capture both theeconomic and genome coverage advantages of WGSwhile retaining the ease and precision of assembly affordedby the BAC components. The additional advantagesof the BACs were illustrated by a comparison withWGSBACCombined AssemblyFigure 2. The Combined Strategy for sequencing largegenomes.Figure 4. High quality of eBACs. An example of a base-by-basecomparison between an eBAC (ordinate) and the same region offinished sequence (abscissa). The colinearity and a few gaps areevident. No repeat filtering was performed, resulting in the largeamount of signal off the diagonal throughout the graph.