Illumina, Inc. · NASDAQ:ILMN
Illumina — Technical Presentation (May 2013)
64 slides from Illumina's technical presentation, May 30, 2013
Slide 1: Sequencing analysis of clinical tissue samples: Fresh Frozen vs. FFPE
Sequencing analysis of clinical tissue samples: Fresh Frozen vs. FFPE Gad Getz, Ph.D. Massachusetts General Hospital Cancer Center and Department of Pathology Broad Institute of Harvard and MIT Illumina Cancer Webinar Series May 30, 2013
Slide 2: Discovery: find cancer genes and pathways
(A) Discovery Find cancer genes and pathways cancer genome projects Characterization (Individual) Interpretation (Population) met somatic primary germline (1) Which genome alterations are statistically significant in the population? What is the full set of genome alterations (2) In which genes and
Slide 3: Discovery: find cancer genes and pathways
(A) Discovery Find cancer genes and pathways cancer genome projects Characterization (Individual) Interpretation (Population) met somatic primary germline The power What is thecomes full setfrom of genome The power comes from alterations depth of sequencing, within the reduced cancer (and number o
Slide 4: Clinical practice: dynamic clinical decision support
(B) Clinical practice Dynamic clinical decision support and follow-up FFPE CancerActionDB Interactive system •chr7:140099605 A>T Treatment bases on: •BRAF V600E • Patient data •Significant in melanoma • Tumor context •Activating mutation • Possible actions based •Sensitive to PLX4032 genetic and cli
Slide 5: Drop in sequencing costs and ability to select parts of the genome
Drop in sequencing costs and ability to select parts of the genome to sequence affordable to perform deep (100-150x) exome sequencing of many patients Gnirke et al. Nat Biotech. (2009)
Slide 6: Next-gen sequencing: base-level characterization
Next-gen sequencing: base-level characterization Meyerson, Gabriel, Getz. Nature Rev Genet. (2010)
Slide 7: Next-gen sequencing: base-level characterization
Next-gen sequencing: base-level characterization muTect Indelocator SegSeq dRanger & BreakPointer PathSeq Kristian Cibulskis Andrey Sivachenko Derek Chiang Mike Lawrence Alex Kostic Yotam Drier C. Pedamallu Cibulskis et al. Chiang, Getz et al. Drier et al. Kostic et al. Nat. Biotechnol (2013) Nat. M
Slide 8: Analysis pipeline @ Broad: Reads to BAMs
Analysis pipeline @ Broad: Reads to BAMs Picard pipeline Align to genome (BWA) Sequencing platform DNA Tim Fennell Mark duplicate reads Mark DePristo Base quality recalibration BAM file Alignment: Li and Homer. Review. Briefings in Bioinformatics (2010) Original Empirical Q – Reported Q Original Rec
Slide 9: Visualization of BAMs using IGV
Visualization of BAMs using IGV Picard pipeline Align to genome (BWA) Sequencing platform DNA Tim Fennell Mark duplicate reads Mark DePristo Base quality recalibration Jim Robinson Helga Thorvaldsdottir dbGaP BAM file Visualization (IGV) Integrative Genomic Viewer (IGV) at www.broadinstitute.org/igv
Slide 10: Analysis pipeline @ Broad: BAMs to annotated variants
Analysis pipeline @ Broad: BAMs to annotated variants Picard pipeline Align to genome (BWA) Sequencing platform DNA Tim Fennell Mark duplicate reads Mark DePristo Base quality recalibration Jim Robinson Helga Thorvaldsdottir dbGaP BAM file Visualization (IGV) Cancer Genome Analysis dbGaP QC ContEst
Slide 11: What's in a cancer sample? How did it get there?
What’s in a cancer sample? How did it get there? Last clonal Initiating ‘driver’ events ‘driver’ events ‘driver’ event Driver event ‘passenger’ events time Driver event – An event that increased the fitness of the cell when it occurred Cancer genes / pathways -- genes / pathways that harbor~1 mut/Mb
Slide 12: Effect of DNA from a mixture of cells
What’s the effect of having DNA come from a mixture of cells? Last clonal Initiating ‘driver’ events ‘driver’ events ‘driver’ event Driver event ‘passenger’ events time ~1 mut/Mb ~5-20 drivers Fitness Adapted from Campbell et al. Nature (2010)
Slide 13: Mutation allelic fraction depends on purity, copy number and multiplicity
Mutation allelic fraction depends on purity, local absolute copy number and multiplicity The mutation allelic fraction is the fraction of alleles (DNA molecules) from a locus that carry a mutation Also the expected fraction of supporting reads DNA from tumors is a mix of normal 33% N 67% T genomes
Slide 14: Allelic fraction and cancer cell fraction (CCF)
Mutation allelic fraction depends on purity, local absolute copy number and multiplicity and cancer cell fraction (CCF) but what if the mutation is only in a fraction of cancer cells? CCF = 1 CCF = 0.5 33% N 67% T 33% N 67% T Purity = 67% Purity = 67% Absolute copy number in tumor = 4 Absolute copy
Slide 15: MuTect: A Highly Sensitive Somatic Mutation Caller
Detecting Somatic Single Nucleotide Variations (SSNVs) MuTect: A Highly Sensitive Somatic Mutation Caller Cibulskis et al. Nat. Biotechnol. (Feb 10, 2013) Available at: www.broadinstitute.org/cancer/cga/mutect
Slide 16: Need a highly sensitive and specific mutation caller
(1) Need a highly sensitive and specific mutation caller. Benchmarking is critical to evaluate tools and monitor progression A mutation caller is a classifier asking at every genomic locus “Is there a mutation here?” Classifiers are evaluated using the Receiver Operators Characteristic (ROC) curve I
Slide 17: What is the signal that we are looking for?
What is the signal that we are looking for? How sensitive can we be? Frequency: Somatic single nucleotide variations occur ~ 1 mutation / Mb Number of alternate reads: Depends on sequencing coverage and allelic fraction 33% N 67% T Binomial distribution: 30 trials with p=0.1 Theoretical 30x coverage
Slide 18: What is the noise? Two types of false positives
What is the noise? two types of false positives 1. NO EVENT 2. GERMLINE EVENT (in T+N) TUMOR TUMOR NORMAL NORMAL At risk: Every base At risk: ~1000 germline / Mb (known) Source: Misread bases 10-20 rare germline / Mb (novel) Sequencing artifacts Source: Low coverage in normal Misaligned reads Need
Slide 19: Back of the envelope calculation of false positive rate
Back of the envelope calculation of false positive rate Assumptions: • 1 incorrect base called in ~3000 sequenced 1. NO EVENT bases (typical value) • Independent errors • Sequence depth of 30x TUMOR 99.7% of sites will have no errors 0.3% of sites will have 1 error 5 x 10-6 of sites will have 2 er
Slide 20: Virtual tumor approach to measure specificity and sensitivity
‘Virtual tumor’ approach to measure specificity and sensitivity (ROC curve) Control allele fraction
Slide 21: MuTect: higher sensitivity with similar false positive rate
MuTect: higher sensitivity with similar false positive rate Sensitivity calculations ROC curve at 30x, allele fraction=0.1 using real data MuTect can detecting low allele fraction mutations, hence suited for studying impure and heterogeneous tumors Cibulskis et al. Nat. Biotechnol (2013)
Slide 22: ABSOLUTE: Estimating purity, absolute copy number and CCF
ABSOLUTE: Estimating purity, absolute copy number and cancer cell fraction (CCF) Carter et al. Nat. Biotechnol. (2012) Available at: www.broadinstitute.org/cancer/cga/ABSOLUTE
Slide 23: ABSOLUTE: Inference of purity and average ploidy
ABSOLUTE: Inference of purity and average ploidy Carter et al. Nat. Biotechnol. (2012)
Slide 24: Cancer cell fractions may have large uncertainties
Cancer cell fractions may have large uncertainties Distribution of CCF of point mutations (purple) and copy-number alterations (cyan). Clonal mutations Assuming there are distinct sub-populations of cancer cells, subclonal mutations and copy- number alterations should cluster in a small unknown numb
Slide 25: Subclonal mutations travel in packs: example sample
Subclonal mutations travel in packs: example sample Before clustering After clustering
Slide 26: Subclonal mutations travel in packs: extended to 2D
Subclonal mutations travel in packs: extended to 2D The same methods work for longitudinal analysis Here – a CLL sampled a two timepoints Before clustering After clustering 1.0 ● ● ● 1.0 ● ● ● 0.9 0.9 0.8 0.8 Post CCF2 CCF 2 CCF 0.7 0.7 0.6 0.6 Post CCF Timepoint Timepoint 0.5 0.5 0.4 0.4 0.3 ● 0.3
Slide 27: Analysis of FFPE samples
Analysis of FFPE samples
Slide 28: Why use FFPE?
Why use FFPE? • Very large numbers of samples in tissue banks and Biorepositories worldwide • Samples often very well-characterized with histological, pathological and follow-up clinical data Can fill the accrual gap in cancer genome projects “We need to get to 10,000 patients per tumor type” --
Slide 29: Challenges with FFPE?
Challenges with FFPE? • Difficulty of extracting samples • Deparrafinization & de-cross-linking of protein-DNA. • Degradation of RNA • Physical size of the samples can be small • Yield • Poor quality of extracted material due to: • Warm-ischemic time in operating room • Type of formalin used, how fi
Slide 30: FFPE samples vary in size (TCGA samples)
FFPE samples vary in size (TCGA samples) • ADD TCGA FIGURE Nationwide Children’s Hospital Biospecimen Core Resource
Slide 31: Samples from clinical study of drug resistance (Broad)
Samples from clinical study of drug resistance (Broad) Small Very small Tiny ? Where is it? ? Kristin Ardlie
Slide 32: General protocol
General protocol • Extraction of DNA: FFPE DNA Extraction was done using a slightly modified version of the Qiagen QIAamp DNA FFPE protocol. Extraction from either 2mm cores or slides. • Library construction: We used a slightly modified version of the protocol described in Fisher et al. Genome biolo
Slide 33: Questions regarding the use of FFPE samples
Questions regarding the use of FFPE samples 1. Can we generate high quality data from FFPE samples? 2. Can we detect the genomic events in FFPE data? Are they real? 3. What is the correlation/overlap of genomic events between FFPE and fresh frozen (FF)? 4. Can we perform cancer genome projects with
Slide 34: FFPE sample sets analyzed
FFPE sample sets analyzed • TCGA Prostate – “trios” Frozen FFPE – 4 FFPE Tumor samples + 4 Fresh Frozen Tumor/Normal pairs Tumor Tumor – Sequencing Coverage: • FFPE samples: 200x Normal • Fresh Frozen pairs: 100x blood • TCGA Colon – “trios” – 3 FFPE Tumor samples + 3 Fresh Frozen Tumor/Normal pairs
Slide 35: Exome seq: Data quality
Exome seq: (1) Data quality similar coverage Frozen FFPE 3 Colon, 4 prostate FFPE 200x, FF 100x % of Target Bases Covered Estimated Library Size Estimation of unique molecules in exome sequencing library 100% 3.E+08 98% % of Target Bases Covered at X Depth 96% # of Unique Molecules -Estimated 2.E+
Slide 36: Exome-seq: Can we find mutations?
Exome-seq: (2) Can we find mutations? Total Count of mutations is similar Frozen FFPE 4 prostate 135 137 Total territory: 130.49 MB Total territory: 130.66 MB Frozen FFPE 17 lung Total territory: 499.2 Mb Total territory: 512.2 Mb Broad GSC
Slide 37: Are the FFPE mutations artifacts? Similar mutation spectra
Exome-seq: Are the FFPE mutations artifacts? No – similar mutation spectra 4 prostate 17 lung Broad GSC
Slide 38: Can we detect copy number changes? Example #1
Exome-seq: (2) Can we detect copy number changes? (3) Are they the same? Example #1 Frozen FFPE Using CapSeg, Aaron McKenna, Scott Carter Broad GSC
Slide 39: Can we detect copy number changes? Example #2
Exome-seq: (2) Can we detect copy number changes? (3) Are they the same? Example #2 Frozen FFPE Using CapSeg, Aaron McKenna, Scott Carter Broad GSC
Slide 40: Are we finding the same mutations in FFPE and frozen?
Exome-seq: (3) Are we finding the same mutations in FFPE and frozen? 19 lung FFPE Frozen 1909 3230 2225 44% 4 prostate FFPE Frozen 56 77 26.5% 79 Broad GSC
Slide 41: Comparing frozen to FFPE changes two variables at once
A fundamental observation: When comparing frozen to FFPE we are changing TWO variables at once Tumor sample Frozen sample FFPE sample (1)Frozen vs FFPE (2)Two different pieces of the tumor -- Different in terms of tumor purity -- Different with respect to sub-clonal composition THIS AFFECTS ALL COMP
Slide 42: Sensitivity depends on coverage and allelic fraction
Sensitivity to detect (and even observe) a mutation – depends on coverage and allelic fraction Sensitivity calculations 33% N 67% T Purity = 67% Absolute copy number in tumor = 4 Mutation multiplicity = 1 Allelic fraction = 2/10 = 0.2 ABSOLUTE Carter et al. Nat. Biotech. (2012) ABSOLUTE: SNP array
Slide 43: Allelic fraction differs due to differences in purities
Allelic fraction in frozen and FFPE are different due to differences in purities 17 lung Broad GSC
Slide 44: Instead of independent calling, validate
Instead of independent calling validate FFPE Frozen 1909 3230 2225 AF in Frozen 44% • The missing mutations actually do exist • Instead of calling validate Validation • Convert allelic-fraction from sample that it was detected to the other • Consider only mutations that we had statistical power
Slide 45: Distinguishing clonal and sub-clonal mutations
But not all mutations are clonal … We can distinguish between clonal & sub-clonal mutations A single ovarian cancer Clonal Sub-clonal mutations Sub-clonal copy-number events ABSOLUTE can identify clonal and sub-clonal mutations Carter et al. Nat. Biotech. (2012)
Slide 46: Clonal mutations validation rate
Clonal mutations, validate = 2+ reads, corrected allelic fraction, power ≥ 95% ~90% validation rate, even better at high allele fractions Petar Stojanov, Eli Van Allen
Slide 47: Interpretation
Interpretation How can we distinguishing “driver” from “passenger” alterations? Model background mutational processes Identify genes/regions/pathways with more mutations than predicted by the background model candidate ‘driver’ events or inaccurate background mode systematic artifacts in mutation ca
Slide 48: Tools for detecting significantly altered cancer genes
Tools for detecting significantly altered cancer genes / regions / pathways GISTIC 1.0 NetSig (in development) ALL MODALITIES Zou et al., in development COPY NUMBER Beroukhim et al. PNAS (2007) GISTIC 2.0 Uses: all types of Mermel et al. Genome Biol. (2011) alterations to identify clusters of mutate
Slide 49: MutSig: scoring genes by number and types of mutations
MutSig: Approach -- Score genes/genesets according to number and types of mutations Gene-specific models, expression levels using intron mutations (as in InVEx), and more Mike Lawrence, Petar Stojanov, et al
Slide 50: Long tailed distribution of cancer genes
(4) Long tailed distribution of cancer genes GBM (n=84) TCGA network, Nature (2008) Head & Neck (n=74) Stransky, Egloff, Tward et al, Science (2011) Ovarian (n=316) Lung Adenocarcinoma (n=188) TCGA network, Nature (2011). 300 Ding, Getz, Wheeler. Nature (2008) 250 200 150 100 50 0 2% + BRAF, PIK3CA,
Slide 51: Can we perform cancer genome projects using FFPE samples?
(5) Can we perform cancer genome projects using FFPE samples? MutSig: Significant genes (17 samples) Frozen FFPE Orange background if within top 30 of other list Yes! Very similar MutSig lists Old MutSig version Broad GSC
Slide 52: Can we sequence clinical FFPE samples for clinical decisions?
(6) Can we sequence clinical FFPE samples for clinical decision making? Yes! 17 lung samples Gene # in Frozen # in FFPE TP53 7 7 KRAS 4 5 EGFR 2 1 STK11 2 4 KEAP1 0 1 ATM 1 2 NF1 4 4 Broad GSC
Slide 53: Subclonal mutations in clinical sequencing: CLL example
The importance of subclonal mutations in clinical sequencing: Example from Chronic Lymphocytic Leukemia Landau, Carter, Stojanov et al. Cell (Feb 14, 2013) Collaboration with Catherine Wu’s lab from DFCI
Slide 54: Study design
Study design 149 CLL samples, 18 sampled at 2 timepoints WES / SNP array Allelic fractions / copy ratios ABSOLUTE (Carter et al., 2012) Cancer Cell Fractions (CCF) Clonal mutations: Subclonal mutations: Affecting all cells; Affecting a subpopulation; - Founder and earlier events - Later events
Slide 55: Recurrent drivers in CLL
25 recurrent drivers in CLL MutSig * 9 novel putative CLL drivers identified 16 previously reported drivers in: Wang et al., NEJM, 2011 Quesada et al., Nat Gen, 2011 Fabbri et al., JEM, 2011 GISTIC Brown et al., Clin Can Res, 2011 Edelmann et al., Blood 2012
Slide 56: Patterns of clonal evolution in longitudinal samples
Patterns of clonal evolution in longitudinal samples (n=18) No evolution Evolution
Slide 57: Stable subpopulation sizes without intervening therapy
Stable relative subpopulation sizes in CLLs without intervening therapy (n=6)
Slide 58: Frequent clonal evolution with intervening therapy
Frequent clonal evolution in CLLs with intervening therapy (n=12) • Expanding mutations are highly enriched in drivers • Patients receiving therapy have frequent clonal evolution
Slide 59: CLLs with and without subclonal drivers
CLLs with and without subclonal drivers in single timepoint samples Subclonal driver present (n=68) Subclonal driver absent (n=81) Driver mutations Sequenced CLL samples (n=149)
Slide 60: Subclonal drivers impact clinical outcome
Subclonal drivers impact clinical outcome in single time-point analysis
Slide 61: Emergence of fitter subclones following treatment
Emergence of fitter subclones following treatment
Slide 62: Summary 1/2
Summary 1/2 • Exome Sequencing of FFPE samples is robust – we can extract DNA, capture and sequence • We need to carefully calculate overlap between FFPE and frozen samples controlling for relative coverage and adjust for different allelic fractions • Mutation rates and categories are very similar •
Slide 63: Summary 2/2
Summary 2 / 2 Tasks of cancer genome projects: characterization and interpretation Characterization is also needed for clinical use. Characterization 1. Systematic benchmarking of analysis methods is needed: ROC curves 2. MuTect can detect low-allele fraction mutations. 3. Need estimates of purity a
Slide 64: Acknowledgements
Acknowledgements Analysis Team Broad Institute Mike Lawrence Genomics Platform Kristian Cibulskis Tim Fennell Petar Stojanov Sara Chauvin Scott Carter NHGRI NCI Lauren Ambrogio Andrey Sivachenko Sheila Fisher Mara Rosenberg Stacey Gabriel Matthew Meyerson Joshua Levin Eli Van Allen Xian Adiconis Lev