Presentation on theme: "Translating NGS Data into a Clinically Actionable Assay Elaine R. Mardis, Ph.D. Professor of Genetics Co-director, The Genome Institute NCI Workshop: NGS."— Presentation transcript:
Translating NGS Data into a Clinically Actionable Assay Elaine R. Mardis, Ph.D. Professor of Genetics Co-director, The Genome Institute NCI Workshop: NGS in Clinical Decision Making
Why is cancer WGS analysis “easy”? The comparison of a patient’s tumor to their normal genome Provides an individualized comparison of what is truly somatic vs. what is truly inherited (germline) Existence of online information about frequently mutated genes in cancer samples (COSMIC) Large-scale efforts using NGS methods to catalogue mutated genes (e.g. TCGA)
Why is cancer genome analysis challenging? In solid tumors, there are normal cells present to differing degrees. Certain tumor types are quite diffuse (prostate, pancreas) and may require specific tumor cell isolation by LCM or flow sorting Conventional pathology may require the majority of the tumor block, leaving little for genomics (melanoma) FFPE preparation from pathology (DNA/RNA degradation) Genomic aneuploidy and amplification of chromosomal segments impacts the coverage model Cellular heterogeneity is a reality (not all cells contain all mutations) In blood or “liquid” tumors, a skin biopsy is taken for the normal but may contain high circulating tumor cell counts at diagnosis
Whole Genome Sequencing Process Human reference alignment SNP Typing of Tumor and Normal gDNA Shotgun library construction Sequence data generation Computational detection of somatic changes The human genome reference sequence is the keystone for cancer genome sequence analysis. Tumor and normal genomes are compared separately to the human reference sequence, then to one another, to identify somatic variation of all types. Mis-aligning sequences identify structural alterations.
+Normal sequence alignments 1.Strand bias (via binomial test) 2.Distance to effective 3’ end of read (via K-S test) 3.Paralog filter (via sum of mismatch base qualities) 4.Homopolymer filter (number of consecutive bases preceding or following the variant) Somatic Point Mutation Discovery SNVs indentified in Tumor Predicted Tumor-unique SNVs Candidate Patient Tumor-unique SNVs Tier 1: Coding NS SNVs Splice site SNVs Coding SS SNVs SNVs in RNA genes Tier 1: Coding NS SNVs Splice site SNVs Coding SS SNVs SNVs in RNA genes Tier 2: SNVs in highly conserved blocks SNVs in regulatory regions Tier 2: SNVs in highly conserved blocks SNVs in regulatory regions Tier 3: SNVs in Non-repetitive regions Tier 4: The rest Somatic SNiPer Subtract dbSNP
WUGC Somatic SV/CNV Pipeline Validation data go through parts of this pipeline
Custom Capture for Validation and Read Depth gDNA Illumina library Custom capture probes (target each variant site) Hybridization Bind to Streptavidin Magnetic Beads Sequence Variant Sites and SV assemblies at ~1000-fold Depth
Cancer Genomics R.K.Wilson 2011 “AML1”: Cancer Genomics by Whole Genome Sequencing Caucasian female, mid-50s at diagnosis De novo M1 AML Family history of AML and lymphoma Informed consent for whole genome sequencing Solexa sequencer, 32 bp unpaired reads 10 somatic mutations detected Ley et al., Nature 2008
Tumor Sequencing is Driving Discovery Total WGS samples: 1351 Pediatric and adult tumors with comprehensive clinical data to address clinically relevant questions
Every cancer is different…
Conclusion: “…whole genome characterization will become a routine part of cancer pathology.”
Cancer Genomics in the Clinic Therapeutic Options via NGS
tAMLCase Presentation 37 y.o. female presented with T2N1 Breast CA ER/PR/Her2+. BRCA1/2-normal. At age 39-Stage III-C ovarian CA diagnosed. At age 43-locally recurrent ovarian CA. 2 months after completing chemotherapy, presented with t-AML/respiratory failure. Expired 9 days after presentation. Detailed family history did not suggest inherited cancer susceptibility. Patient has three minor children. Link et al., JAMA 2011; 305(15):
tAML: TP53 germline deletion Whole genome analysis indicates the patient has Lei-Fraumeni syndrome. Previously undetected by clinical assay due to nature of the 3 exon deletion. Genomic data are supported by RNA analysis.
Clinical case: atypical APL 37 y.o. female with de novo AML; M3 morphology 37 y.o. female with de novo AML; M3 morphology Complex cytogenetics, persistent leukemia Complex cytogenetics, persistent leukemia First remission, referred to WU for SCT. rBM: normal morphology, cytogenetics; negative for PML/RARA. First remission, referred to WU for SCT. rBM: normal morphology, cytogenetics; negative for PML/RARA. Allogeneic SCT Consolidation + ATRA Chemo + ATRA Chemo only ???
Welch et al., JAMA 2011: 305(15):
“Genome-Guided Medicine”: An early example 37 y.o. female with de novo AML, M3 morphology, CTG, no PML- RARA. Referred to WUSM for SCT. 37 y.o. female with de novo AML, M3 morphology, CTG, no PML- RARA. Referred to WUSM for SCT. Detection of PML-RARA by WGS, Confirmed by FISH, RT-PCR (CLIA/CAP) Detection of PML-RARA by WGS, Confirmed by FISH, RT-PCR (CLIA/CAP) Consolidation: Chemo + ATRA Sustained remission
Cancer Genomics in the Clinic Therapeutic Options via “Gx,Ex,Tx”
NGS “Diagnostic Trials: An N of 1” Cancer patients consented for genomic sequencing and return of information Cancer biopsies studied by WGS, exome and transcriptome integrated analysis WGS drives discovery Exome contributes read depth for heterogeneity/clonality analysis Transcriptome monitors aberrant gene expression and validates fusions Interpretive analysis should accurately identify actionable targets and available clinical trials. All possibly actionable mutations/alterations are verified in CLIA lab with pathology sign-off. A Tumor Board model for education, decision-making, and patient monitoring is critical. Sharing results to the community is desired/critical!
Tumor Immunoediting : Somatic mutations as vaccine targets Combined exome capture and in silico epitope prediction in a chemically- induced mouse sarcoma model We identified a highly immunogenic tumor-specific mutated protein antigen that targets tumor cells for elimination in an immune-capable host. First demonstration using genomics to identify a tumor antigen from an unedited tumor, and to demonstrate that T-cell-dependent immunoselection is a mechanism underlying the outgrowth of tumor cells that lack a strong rejection antigen(s).
Examples of Diagnostic Sequencing Metastatic breast cancer
HG1 Patient History Female patient, mid-50’s with history of DCIS and Paget’s disease of the left nipple 2007 Widespread metastatic breast cancer to bone 2009, biopsy shows ER- HER2+ disease (FISH amplified), highly responsive to paclitaxel + trastuzumab Brain metastasis in posterior fossa diagnosed May 2010, treated with surgery (sample for sequencing obtained) radiosurgery and lapatinib Progressive disease in March 2011: treated with further surgery and whole brain irradiation July 2011: systemic disease still under control with trastuzumab in combination with lapatinib
Somatic mutation frequencies hint at heterogeneity Read coverage (X) Tumor variant allele frequency Proportion Tumor variant allele frequency Metastatic breast cancer (to brain) 92 point mutations are identified in genes
Somatic copy number variants – genome wide Metastatic breast cancer (to brain)
Somatic copy number variants – chromosome 17 Metastatic breast cancer (to brain) HER2 / ERBB2 is heavily amplified in this tumor HER2
RNA-seq confirms the HER2, PR, & ER status Metastatic breast cancer (to brain) vs. four primary HER2 –ve breast cancers HER2 +ive PR- ER- Gene expression values from RNA-seq Used to confirm HER2, PR, & ER status of each patient Tumor is HER2+, PR-, ER-
Somatic copy number variants – chromosome 6 Metastatic breast cancer (to brain) HDAC2 (histone deacetylase 2) is amplified to almost the same degree as HER2
RNA expression – HDAC2 Metastatic breast cancer (to brain) vs. four primary HER2 –ve breast cancers HDAC2 genomic amplification is accompanied by high RNA expression RNA expression pattern confirms HDAC2 over- expression. Patient predicted to respond to the HDAC2 inhibitor Vorinostat [DrugBank].
PNC-2 Tumor: Pancreatic Neuroendocrine Metastatic Disease Initial diagnosis: Pancreatic Neuroendocrine tumor First metastatic tumor (liver) banked in 2005 (FFPE), no adjuvant chemotherapy Second metastatic tumor to liver banked in 2011(FFPE), following neoadjuvant chemotherapy, including everolimus + Bevacizumab Patient consented for return of results from whole genome sequencing We produced WGS and exome capture data from the two metastatic tumors and a blood normal. RNA-seq from both metastatic tumors.
PNC2: Comparing metastatic tumor presentations Met1 Clonality Met2 Clonality Although 33 mutations were identified in the tumor genome, none were considered druggable…
PNC2: RNA-seq analysis PNC2: Met1 Event typegene_nameEffect (FPKM)drug_name RNAseq CufflinksCCL Mimosine RNAseq CufflinksF Suramin RNAseq CufflinksFKBP1A Sirolimus RNAseq CufflinksPLA2G2A Suramin RNAseq CufflinksPSMD Bortezomib RNAseq CufflinksSLC25A Clodronate RNAseq CufflinksTUBA1A Vinblastine RNAseq CufflinksVEGFA Bevacizumab PNC2: Met2 Event typegene_nameEffect (FPKM)drug_name RNAseq CufflinksCCND Arsenic trioxide RNAseq CufflinksHDAC Vorinostat RNAseq CufflinksPLA2G2A Suramin RNAseq CufflinksPSMD Bortezomib RNAseq CufflinksSLC25A Clodronate RNAseq CufflinksSLC25A Clodronate RNAseq CufflinksTUBA1A Vinblastine RNAseq CufflinksVEGFA Bevacizumab Based on our RNA-seq analysis, VEGFA is increasing in its expression levels from the initial metastatic lesion sampled in 2005, to the present lesion sampled in The DrugBank prediction for VEGFA overexpression is treatment with Bevacizumab/Avastin.
PNC2: “Post-mortem” Diagnostics Baseline 3 weeks on Everolimus6 weeks after adding Bevacizumab As happens in sequencing advanced metastatic patients, this patient died before being treated based on the genomic predictions. However, post-mortem consultation with the patient’s oncologist indicated that perfusion CT during bevacizumab treatment showed response was evident. However, Bevacizumab had been withdrawn due to side effects.
Example case Acute lymphocytic leukemia
Case study: 2 nd relapse B-ALL Age 25: Initial presentation of classic pre B-ALL Standard induction, consolidation, and 2 years of maintenance therapy Marrow banked Age 30: 1 st relapse CR obtained with salvage chemo consolidation with a matched sibling allo transplant very mild GvHD Age 33: 2 nd relapse, CNS involvement (July 2011) During induction chemotherapy, we sequenced T/N genomes using banked blasts from initial presentation, exomes (T/N) and RNA-seq of blasts
ALL-1: Somatic single nucleotide variations 91 somatic coding SNVs 42 with evidence for expression in RNA-seq GeneRef.Var.AATypeWGSExomeRNA-seq UBXN4TCDDDDsilent51.25%40.56%53.40% OGTGTCFCFmissense39.22%38.7%35.79% KIAA1033CTTITImissense38.89%48.1%50.75% C15orf39CGAAAAsilent47.37%37.5%48.74% SPTAN1TCLPLPmissense42.39%49.03%50.17% DDX6AGLPLPmissense21.78%14.29%18.14% CCDC47CTATATmissense22.34%21.9%23.36% NF1CTR*R*nonsense64.58%64.04%47.06% TNRC6BGASNSNmissense13.25%12.8%19.46% KIAA1462GCPAPAmissense70%45.05%41.14% Although 91 mutations were identified in the tumor genome, none were considered druggable…
ALL-1: Tumor heterogeneity Read coverage (X) Tumor variant allele frequency Proportion Tumor variant allele frequency 2,074 tier1-3 somatic variants. 91 are tier1 (coding exons) NF1 Acute lymphocytic leukemia
ALL1: RNA-seq analysis Naive B- cells Pre-B-ALLALL1 Patient’s activated FLT3 gene was targeted with sunitinib, complete clinical remission was achieved in 12 days, enabling MUD SCT. Identification of discrete chromosomal deletions in tumor cells provides a means for ongoing tumor assessment with interphase FISH (presence of MRD) Four months post-SCT, the patient is back at work. CD135 (FLT3) added to the flow panel for all B-ALL patients at Barnes-Jewish Hospital.
Summary We can produce comprehensive whole genome analysis of cancer patients now, and the data can provide very important input for clinical, therapeutic decision making. Not all patients will benefit, however, because of our current knowledge gaps and because targeted therapies are not yet available for many important cancer genes. Many regulatory issues must be resolved before these tools can be used widely. Discussions are ongoing at NIST, FDA, CAP etc. Each case represents a focused effort involving genomicists, oncologists, pharmacologists and pathologists (at least). Physician education in genomic data interpretation is a tangential benefit. Off label use of therapies may become common. Sharing results is a critical exercise at sites implementing this approach.
Name Acknowledgements The Genome Institute Li Ding, Ph.D. Malachi Griffith, Ph.D. David Dooling, Ph.D. David Larson, Ph.D. Nathan Dees, Ph.D. Vincent Magrini, Ph.D. Sean McGrath Jason Walker Amy Ly Daniel Koboldt Lucinda Fulton Robert Fulton Lisa Cook Ryan Demeter Todd Wylie Kim Delehaunty Michael McLellan Rick Wilson WUSM/Siteman Cancer Center Timothy Ley, M.D. Matthew Ellis, M.B., Ph.D. Benjamin Tan, M.D. John DiPersio, M.D., Ph.D. Timothy Graubert, M.D. Matthew Walter, M.D. John Welch, M.D., Ph.D. Jackie Payton, M.D., Ph.D. Peter Westervelt, M.D., Ph.D. Lukas Wartman, M.D. Our patients NHGRI NCI WUCGI