Clinical grade next-generation sequencing of UM

Elisha Roberson, Ph.D.Depts. of Internal Medicine and Genetics

Washington University in St. Louis

All talk content can be tweeted and blogged

@thatdnaguy

Sanger ≠ next-gen sequencingSanger

• Consensus of a population of molecules• ~600-800bp sequence• Very low error rate• Low-throughput• Targeted• Expensive* (256 bp/$)

NGS• Single-molecules• 35 bp – 3+ kb•High error rates (1-15+%)• High-throughput• Can shotgun• Cheap* (11.5 Mbp/$)

*Our lab’s current Sanger & HiSeq2500 costs

UM Clinical sequencing• CLIA/CAP– Detect actionable somatic variants• Insurance pre-approval• Submit DNA• Wait a long time• Pay $7000+ / sample

• Research (discovery, epidemiology, etc)– IRB approved DNA collection– You sequence and interpret

My sequencing wishlist• Tissue

– Fresh tissue– Large amount of high-quality DNA

• No PCR amplification

• Sequencing (whole genome)– Low-error, long reads– Paired-end– 30X or greater germline coverage– 60X or greater tumor coverage

• Bioinformatics– De novo assembly of both germline and tumor, compare to

reference & each other– Genotyping algorithm that is pair-aware

NGS technologies

• ABI – SOLiD– Emulsion PCR by di-base ligation

• Illumina – Solexa– Single-molecule fluor sequencing

• Life Tech - Ion Torrent– Single-molecule semiconductor sequencing

• PacBio – SMRT– Zero-mode wave guide with single polymerase in-well

Where should I get DNA???

• Germline–NO blood• Sequencing single molecules!!!

– Spit kits, washed skin biopsies

• Tumor tissue– Fresh primary tumor– Fresh metastatic tumor– FFPE?

Calculating sequencing depth

• NGS technology dependent• How deeply do you want to see mosaicism?

• General guidelines– 30X germline– 60X or higher tumor

• BUT sequencing follows Poisson distribution– i.e. 30X average coverage != all targets 30X

% Target covered ≥30X at different average coverage

Coverage also varies by tissue*

*Plot available on Figshare

FASTQ Preprocessing• Demultiplex samples

– Discard no index

• Convert to PHRED quality scale– -10 x log10( probability of base error )

• Remove adapter contamination– cutadapt

• Trim low-quality trailing bases and 3’ Ns– No 5’ trimming!!!

• Run FASTQC!

Alignment to reference

• Current human is GRCh37

• Repeat masking– Hard mask repeats are N– Soft mask repeats are lowercase– Prefer soft masking

• Ref aligners have generally low memory use– Mostly use Burrows-Wheeler transform– bowtie 1 & 2– bwa-aln & bwa-mem– Novoalign (high memory)

De novo assembly• Most De Bruijn graphs with kmers of

sequence

• Mostly very high memory usage– Depends on depth and number of kmers– Try running diginorm first (C. Titus Brown)

• Aligners– ABySS– MIRA– SOAPdenovo– Velvet

Post-processing• Picard tools– Convert to BAM format– Add read-group tags–Mark duplicates (Picard tools)

• Genome Analysis ToolKit (GATK)– Local realignment of indels– Base quality score recalibration

Genotyping• General– Genome analysis toolkit• Unified Genotyper

– Samtools• Mpileup

• Somatic specific–MuTect– Somatic Sniper– VirMiD

Variant filtering strategies –sequential evolution

Initiating event

Mosaic primary

Mutation with metastatic advantageleaves the eye

Metastasis has primary mutations &Metastatic mutation (maybe in primary?) &New mutations

Variant filtering strategies –parallel evolution

Initiating event

Mosaic primary

Metastasis has few primary mutations &Metastatic mutation (not in primary) &New mutations

Very little overlapin mutations between

primary and met!

Variant confirmation

• Sanger sequencing

• Fluidigm

• TaqMan– castPCR

• Sequenom

• Illumina Golden Gate

Discover in a focused set with sequencingType with these technologies in everything

Many thanks to