Glioblastoma mul,forme

Transcription

1 Glioblastoma mul,forme Glioblastoma mul,forme (GBM), or Glioblastoma, or Grade IV Astrocytoma is the most common and most aggressive malignant primary brain tumor. GBM is usually involving glial cells, and the incidence is about 2~3 cases per 100,000 person lifeyears.

2 Genomic Landscape of Primary GBM Cross-sectional studies (~200 patients) have identified many new driver genes: Science (2012), Nature Genetics (2013).

3 Treatment Stupp Therapy (Roger Stupp): Surgery radia,on with daily temozolomide (TMZ) six cycles of five-days-a-week TMZ. TMZ alkylates/methylates DNA, and triggers the death of tumor cells MGMT repairs this damage, diminishes the therapeu,c efficacy For now there is no standard treatment for recurrent tumor

4 Dynamic Landscape of GBM

5 Ques,ons How does GBM evolve under therapy? Can we reconstruct the main routes of GBM evolu,on? How spa,al informa,on (intratumor heterogeneity) informs evolu,on? Are there specific genes associated to GBM relapse?

6 The cohort Somatic Mutations" > Initial Common Recurrence Unknown 100 B! MUT" AMP" 0 R001" INCB" MD Anderson" R020" TCGA" R040" UCSF" KU" R060" R080" SMC" R100" Primary GBM" TMZ" MMR" Hyper Mut" TP53" ATRX" IDH1" EGFR" PTEN" PIK3CA" PIK3R1" PIK3CG" PDGFRA" RB1" NF1" PTPN11" LTBP4" CDK4" EGFR" MDM2" MDM4" PDGFRA" CDKN2A" TP53" RB1" PTEN" NF1" EGFRvIII" DEL" Only Initial" Common" Only Recurrence" * Common, VAF<5%" * Recurrent VAF<5%" * Initial VAF<5%" R114" Nature Genetics (2016).

7 How can we compare evolu,onary histories in different pa,ents?

8 Comparing evolu,onary histories If every tumor follows a different evolutionary path, how do we extend computational and statistical frameworks to study longitudinal data? Aims: 1) to develop a framework to compare the evolutionary history of different patients -> metric space that allow comparison of evolutionary histories. 2) To develop a statistical framework and machine learning approaches to longitudinal cohorts. 3) To develop a statistical/evolutionary framework to test different evolutionary models (e.g. multistage due to hard selective sweeps).

9 Cancer follows clonal Darwinian evolution

10 Patients represent forests of phylogenetic trees R001 R002 R003 R004 R005 R006 R007 R008 R009 R010 R011 R012 R013 R014 R015 R016 R017 R018 R019 R020 R021 R022 R023 R024 R025 R026 R027 R028 R029 R030 R031 R032 R033 R034 R035 R036 R037 R038 R039 R040 R041 R042 R043 R044 R045 R046 R047 R048 R049 R050 R051 R052 R053 R054 R055 R056 R057 R058 R059 R060 R061 R062 R063 R064 R065 R066 R067 R068 R069 R070 R071 R072 R073 R074 R075 R076 R077 R094 R095 R096 R097 R098 R099 R100 R101 R104 R106 R107 R108 R109

11 Moduli or classifying space of clonal evolu,on To compare two or more histories we would like to find a space that describe all possible evolu,onary histories. Moduli space: space whose points correspond to objects of some kind. Every tumor has a different history, so every tumor evolu,on corresponds to a par,cular point in that space.

12 Space of trees Theorem 1: The space of trees is a Polish space (one can define statistics). Theorem 2: The space of trees is a CAT(0) space (one can define means and variances in a meaningful way). Billera, Holmes, Vogtman (2003) Zairis, Khiabanian, Blumberg, Rabadan (2014)

13 GBM evolu,on shows dis,nct evolu,onary trajectories associated to specific molecular mechanisms Common" Initial" Recurrent"

14 GBM evolu,on shows dis,nct evolu,onary trajectories associated to specific molecular mechanisms Common" Initial" Recurrent"

15 100%! 90%! P<0.01! P<0.01! 80%! %! 60%! 50%! 40%! 30%! 20%! 10%! 0%! non-tmz Initial! non-tmz TMZ non-hm Recurrence! Recurrence! TMZ HM Recurrence! C>T/G>A! C>G/G>C! C>A/G>T! A>T/T>A! A>G/T>C! A>C/T>G! n=93" Silent/Missense Ratio Untreated_NM Recurrent_NM Recurrent_HM Untreated! Recurrent! Non-HM! n=87" Recurrent! HM! 0.8 n=87" 0.6 P<0.01! P<0.01! 0.4 m=160" HM Score! Recurrence Initial Mean Expression expression! Number of mutations" HM gene" (9,838)" M gene" (3,278)" NM gene" (10,375)"

16 Considering subclonal muta,ons MRCA of initial sample" C" Recurrent" C" S" X" MRCA of recurrent sample" Type I! Initial" S" X" Initial sample" Recurrent sample" Time"

17 Can we reconstruct the main routes of GBM evolu,on under therapy?

18 Type I model of tumor evolu,on Birth Tumor common ancestor Time Diagnosis & Treatment Relapse Primary sample Relapse sample

19 A simple model of tumor evolu,on Mutation rate before treatment (events per year) Length of shared branch (years) Mutation rate after treatment (events per year) By observing the number and clonality pa6erns of muta,ons......we can es,mate branch lengths and muta8on rates

20 Tumor substitution rates: " u 1 u 2 t Grow! t U! Untreated! t S! t R1" t Grow! t R2" Recurrent! Age at tumor lineage divergence " Age at Age at diagnosis " recurrence " Post-treatment substitution rate (/Mb-Year) Time between tumor divergence and diagnosis (years) Pre-treatment mutation rate (/Mb-Year) 0 0% 20% 40% 60% 80% 100% Percentile

21 Ordering genes Common! IDH1 TP53 ATRX PTEN EGFR PIK3CG MT ND4 PIK3CA PTPN11 PIK3R1 RB1 NSD1 NF1 EGFRvIII PDGFRA GLI3 MSH6 LTBP4 PREX1 Untreated Only! Recurrent Only! Nature Genetics (2016).

22 GBM evolu,on shows dis,nct evolu,onary trajectories associated to specific molecular mechanisms Early Late

23 How spa,al informa,on (intratumor heterogeneity) informs evolu,on?

24 Expression-based subtyping Proneural R012 R022 R027 R029 R076 R077 R084 R114 R078 R103 R105 R004 R005 R008 R009 R023 R025 R030 R033 R034 R038 R042 R061 R067 R068 R070 R082 R090 R095 R104 R106 R110 R001 R002 R003 R006 R007 R010 R019 R020 R021 R024 R026 R035 R056 R059 R060 R063 R064 R065 R066 R071 R072 R074 R080 R081 R083 R085 R086 R087 R088 R089 R091 R092 R093 R099 R100 R101 R102 R107 R108 R109 R111 R112 R113 R017 R018 R028 R031 R032 R036 R037 R039 R040 R041 R096 Neural Classical Mesenchymal Initial" Recurrence" Proneural" Mesenchymal" Neural" Classical" Switched" Not Available" Not changed" * Max ES" Switched" I" R" I" R" I" R" I" R" Primary GBM" EGFRvIII" I" R" Secondary GBM"

25 Subtype switching in primary GBM ITH01 1 ITH01 1 ITH01 2 ITH01 2 Number of patients" ITH03 1 ITH %! 80%! 60%! 40%! ITH %! ITH04 2 ITH04 2 ITH05 1 0%! ITH05 1 Proneural Neural Primary! Secondary! Not Changed! Switched! ITH05.3 ITH ITH06 ITH ITH06 1 ITH01 2 ITH06 2 ITH01 2 ITH07 1 ITH03 1 ITH08 1 ITH03 1 ITH08 1 ITH04 1 ITH09 1 ITH04 2 ITH09 1 ITH04 2 ITH10 ITH ITH10 ITH ITH11 ITH ITH11 2 ITH06 1 ITH12 ITH ITH12 1 ITH06 2 ITH13 ITH ITH13 2 ITH08 1 ITH ITH08 1 ITH ITH09 1 ITH ITH09 1 ITH14 1 ITH10 1 ITH14 1 ITH10 1 ITH ITH11 2 ITH ITH11 2 ITH ITH12 1 ITH15 ITH ITH15 1 ITH13 2 ITH01 ITH02 ITH03 ITH04 ITH05 ITH06 ITH07 ITH08 ITH09 ITH10 ITH11! ITH13 2 ITH ITH ITH ITH14 1 ITH14 1 ITH ITH ITH ITH15 1 ITH % Classical 12.5% 25% 13.0% 5.6% 29% 5.6% 50% Proneural Mesenchymal 47.8% 21.7% Classical 38.9% 22% 17.4% Mesenchymal Switched! Expression subtype change Recurrence! Recurrence! Recurrence! 22% Neural 44.4% Multi-Section! Longitudinal! 11% 67%! 33%! Switched! Not Switched! 33%! 67%! Switched! Not Switched!

26 Are there specific genes associated to relapse?

27 Novel MGMT fusions in Recurrent GBM BTRC MGMT Exon 1 Exon 2 Exon 1 TRC SAR1A Chr1 Exon 1 Ex2 Exon 1 5 Exon 1 Ex2 Exon 1 Exon 2 Exon 3 Exon 4 Exon 5 3 (bp) Primer 1 Primer 3 Fusion gene BTRC-MGMT Primer 2 Set1 Set2 Actin M I R I R I R Primer 4 I : Initial tumor R: Recurrent tumor NFYC Exon 1 Exon 2 Exon 1 MGMT Fusion gene Exon 1 Ex2 Exon 1 Exon2 Set1 Set2 Actin Primer 1 Primer 2 (bp) M P R P R P R Primer 3 5 Ex2 Exon 1 Exon 2 Exon 3 Exon 4 Exon 5 3 NFYC-MGMT Primer I : Initial tumor R: Recurrent tumor 27

28 Initial OS METHYLATION Time (month) MGMT Low MGMT High Censored p-value=1e-4 MGMT EXP" p-value=5.9e MGMT UNMET" MGMT MET" OS n=29 EXPRESSION Initial p-value=0 n=32 MGMT Low MGMT High Censored Time (month) Recurrence OS METHYLATION Recurrent Time (month) MGMT Low MGMT High Censored p-value= p-value=1e MGMT UNMET" MGMT MET" OS n=25 EXPRESSION Recurrence p-value=4e-4 n=46 MGMT Low MGMT High Censored Time (month)

29 Conclusions - How does GBM evolve under therapy? - Type I: fits 70% clonal histories. The clone at diagnosis is not direct ancestor of clone at relapse. - Type II cases (22%) have shorter time between diagnosis and relapse (0.7 vs 1.5 years). - Type III: enriched in RB1 deletions. - Most patients the evolutionary rate ~ 0.03 /Mb.year, - Subsets of patients (17%) undergo hypermutation. Hypermutations associated to dinucleotide signatures (CpC/GpG) in genes with higher expression. - Common ancestor present years before diagnosis. - Expression typing is not stable in primary GBM (63% change). - Subtype does not reflect clonal history but changing mixture of cell populations. - Unsupervised dimensionality reduction does. - Can we reconstruct main evolutionary routes? - Type I model allows (topological) ordering events. - IDH1 -> TP53, PTEN ->EGFR, MSH6, LTBP4 - Are there specific alterations associated to progression? - Preserved in evolution: IDH1, FGFR3-TACC3, - Lost in evolution: EGFRviii lost at relapse. - Gained in evolution: - New relapse specific alterations (> 10%): LTBP4 (TGF-β ), MSH6, PREX1, etc. - Novel relapse specific MGMT gene fusions, MGMT expression at relapse associated to prognosis.

30 Columbia University: Departments of Systems Biology and Biomedical Informatics: Jiguang Wang, Oliver Elliott, Pablo Camara, Albert Lee, Francesco Abate, Sakellarios Zairis, Kevin Emmett, Daniel Rosenbloom, Chioma Madubata, Erik Ladewig, Zhaoqi Liu, Mykola Bordyuh, Patrick van Nieuwenhuizen, Udi Rubin, Tim Chu. Institute for Cancer Genetics and Irving Cancer Center: Veronica Fratini, A. Iavarone, A. Lasorella. Texas (Austin): A. Blumberg. Stanford: G. Carlsson. Istituto Neurologico Carlo Besta (Milan, Italy): Emanuela Cazzato, G. Finnochiaro. Samsung Medical Center (Seoul, Korea): Jason Sa, Jin-Ku Lee, Do-Hyun Nam.