The Changing Face of Sequencing Strategies for de novo sequencing of complex genomes
Quick Review: BACs Whole Genome Shotgun
First some history…. 2000: Arabidopsis 2005: Rice 2006: Poplar BAC 2000: Arabidopsis BAC & WGS 2005: Rice WGS 2006: Poplar WGS 2007: Grapevine BAC 2008: Maize WGS 2008: Papaya WGS 2009: Sorghum
BAC-based vs WGS BAC-by-BAC WGS Pros Cons Simpler more accurate assembly Localized sequence Easily distributed Can be targeted to regions Physical map not needed, but helps Logistically simple Low library costs Rapid Cons Requires physical map Labor intensive Expensive (more libraries) Slower Complex assembly Harder to localize sequence Requires centralized assembly Whole genome or nothing
What made WGS possible? Long, high quality Sanger reads (700-800bp) Paired-end libraries Range of insert sizes 3kb 8-10kb 40kb fosmids Assemblers tailored to these datatypes. Still not guaranteed… public maize project went BAC by BAC
NGS changes all the rules Quantity not quality is now the focus New platforms generate huge quantities of data Read length & PE’s initially limited de novo apps Rapid cycle of improvements No time for standard approaches to spread beyond genome centers before next cycle begins. Third party software sometimes slow to catch up Cost model has changed Library construction used to be minor component of cost Unit used to be 96 or 384 reads….. Choice is now more complex than BAC vs WGS
does not One size^fits all Every project has individual needs Monolithic reference genome is rarely needed now How bad are the repeat structures? Is it important to get them right? How important is it to anchor all the sequence to a genome location? What other genome data can be leveraged?
BACs and NGS – the problem Pre-NGS: To sequence a BAC: Make 1 sequencing library ~$50-100 Sequence two 384-well plates of clones ~$750 ~6x coverage With NGS: To sequence a BAC with 454: Make 1 sequencing library ~$300 Sequence 1/8 plate of 454: ~$1,000 ~600x coverage Too expensive, and too much coverage…..
New BAC-based approaches One library per BAC is cost-prohibitive Map-based BAC pooling Retain some of the assembly benefits of BACs Reduced library costs over BAC-by-BAC If contiguous, retains the genome localization benefits
Scaffolds from individual BAC pools BAC pooling strategy Chr3. shortarm Select FPC contigs on the shortarm FPC contigs Select overlapping BACs and bin them into 3Mb pools 3 Mb pools Selected BACs Pyrosequencing of BAC pools and assembly of raw sequences ~20x 454 Titanium Reads (~400bp each) Contigs from individual BAC pools 454 FLX PE’s (~250bp each) Contigs are organized into scaffolds using 454 paired end sequences Scaffolds from individual BAC pools Use BAC ends for very long scaffolds Generate superscaffolds using BAMBUS and BAC end sequences Superscaffolds spanning pool boundaries From Rounsley et al. (2009)
Results: Chr3S of Oryza barthii 6 x 3Mb BAC pools 1 Titanium Run 0.5 FLX Run ~$12k in reagents Contig N50: 14.3 kb Scaffold N50: 370.9 kb Scaffold N50: 3,165.1 kb (after BAC ends) Nt Accuracy: 2.2 errors per 10kb
2D pooling: An alternative to contiguous BAC pools Place ordered clones in plates 1 Library from each row 1 Library from each column Identify reads from each individual clone by sequence overlap. Then assemble each clone Assembly unit reduced to ~ single BAC Library cost drops with size of grid 10x10: 100 clones, 20 libraries 50x50: 2500 clones, 100 libraries 3D grid lowers cost even further 10x10x10: 1000 clones, 30 libraries 20x20x20: 4000 clones, 60 libraries Repeats may misbehave but can choose to ignore them
The ideal…. One library per BAC clone Barcoded Sequence all clones from BAC library in one combined, barcoded pool BUT: currently not cost-effective. Individual DNA preps for thousands of BAC clones is costly
Is WGS with NGS feasible yet? 400bp reads, + 4kb and 20kb insert PE protocols Success may be Species & Goal dependent: Arabidopsis small & low repeat content 21kb contig N50; 2.6Mb scaffold N50 Roche & Ecker Cassava 800Mb, lots of repeats 5.3kb contig N50; 180kb scaffold N50 Roche & JGI Missing half of the genome (repetitive half)
WGS with Solexa/Illumina Improved read-lengths, PE protocols Improved third party assemblers e.g. SOAPdenovo, Velvet Cucumber genome - BGI 300Mb genome 50x coverage with 50bp PE 5kb contigN50, 60kb scaffoldN50 Much better when mixed with 4x Sanger Missing half of genome (repeats) Panda Genome - BGI 3Gb genome 50x coverage with 75bp PE 300kb contigN50 (?) Big question: What is misassembly rate?
Building contigs from overlapping clones 5 overlapping BAC clones form small contig Cut with R.E. Overlapping BACs share common fragments
Building contigs from overlapping clones Measure lengths Overlapping BACs will share fragments of same size Make sequencing lib Sequence from each cut site Overlapping BACs will share sequence tags next to each cut site
A BAC-WGS hybrid? whole genome profiling by Keygene A: Solexa-based BAC map Construct BAC library; array into 2D pools Cut with restriction enzyme, and make 1 library per pool. Generate sequence from libraries Deconvolute pools to identify the Solexa reads from each BAC. Build a map from overlaps Map has short sequence tag every 1-2kb in genome B: WGS sequencing with Solexa Assemble short contigs (high stringency) Use above map to locate each contig in genome. Map can identify misassemblies C: Result: High quality map-based genome at fraction of cost
Simulation of Tag-based Map building Rice: 372Mb, 12 chromosomes Simulate a 10x BAC library 28,600 clones Cut the sequence for each clone with HindIII Simulate a short read sequence from each site 2.2 million sequence tags Build a map from these – overlapping clones share tags 33 contigs built (<3 contigs per chromosome) Only 1 misassembly!
So you want to sequence a genome? Lots of choices to make: BACs, WGS Which NGS technology? Single end, paired end? What size paired ends? What depth of coverage from each? How do you pick? Do lots of testing of strategies - $$$$$ Guess – Free Copy what someone else did - Free Educated Guess based on Simulation
How to decide on a strategy? Simulating Genome Sequencing “Plantagora” Plant Genome Assembly Simulation Platform Use existing genomes to simulate sequencing reads Combine reads in many combinations Assemble Score the results with meaningful metrics Report results on web site
Summary No longer BACs vs WGS Different ways of using BACs Linear pooling 2D pooling BACs for map, WGS for sequence WGS works on easy parts of genome Simulation is valuable in evaluating strategies