Integrative analysis and visualization of clinical and molecular data for cancer precision medicine - Enzo Medico
1. Enzo Medico
University of Torino
Integrative analysis and visualization
of clinical and molecular data
for cancer precision medicine
Candiolo Cancer Institute
Laboratory of Oncogenomics
enzo.medico@ircc.it
7. Towards precision cancer medicine
Targeted
drug
Target
Response
Target
alterations
Sensitizing
alterations
De-sensitizing
alterations
8. Towards precision cancer medicine
Targeted
drug
Target
Response
Target
alterations
Tissue/context-
specific modifiers
Sensitizing
alterations
De-sensitizing
alterations
9. Further elements of complexity
• Intratumoral heterogeneity
De-sensitizing lesions only present in a fraction of the cancer cells
may lead to early recurrence
• Intracellular signaling is governed by networks
Dynamic adaptation to altered signaling.
• Tumor-host interactions
Tumor growth and response also depends on stroma, vasculature,
inflammation and immune response
10. • Analyzing inter-tumoral heterogeneity requires a
reference background
Focus on one specific tumour type
• Sensitizing/de-sensitizing lesions may be rare
Collect many cases
• Alterations may occur in different ways (mutations, CNA,
rearrangements, etc)
Multi-dimensional genomic exploration of high-quality tumour
material
Facing Challenges
11. International consortia for cancer genomics
TCGA
The Cancer Genome Atlas:
http://cancergenome.nih.gov
ICGC
International Cancer Genome Consortium:
www.icgc.org
12. Data available from TCGA (sept 2016)
TCGA data are hosted at the Genomics Data Commons: https://gdc.nci.nih.gov/
14. The TCGA pipeline
• Tissue samples along with clinical data are collected by Tissue
Source Sites (TSS) and sent to the Biospecimen Core
Resources (BCRs).
• The BCRs submit clinical data and metadata to the Data
Coordinating Center (DCC) and analytes to the Genome
Characterization Centers (GCCs) and Genome Sequencing
Centers (GSCs), where sequences and other molecular profiles are
generated and then submitted to the DCC.
• GSCs submit raw and processed data to the Cancer Genomics
Hub (CGHub) as well.
• Data submitted to the DCC and CGHub are made available to the
research community and Genome Data Analysis Centers
(GDACs).
• Analysis pipelines and data results produced by GDACs are served
to the research community via the DCC.
15. Multiple types of data
Clinical data
• Clinical diagnosis
• Treatment history
• Histologic diagnosis
• Pathologic status
• Tissue anatomic site
• Others…
Molecular data
• DNA sequence
• DNA copy number
• DNA methylation
• RNA expression
• Protein expression
• Others…
Clinomics:
“the study of -omics data along with its associated clinical data”
…and there is more…
16. “P0”
P1
P2
Biobank Archive
Nucleic Acid Extraction
“Xenotrial”P3
VECTOR
DRUG
More data: patient-derived xenografts (PDX):
"Tumorgrafts", "Xenopatients", "Avatars"
VECTOR
DRUG
(Engraftment)
(Expansion)
(Surgery)
17. Advantages of the PDX approach
• Possibility to conduct population-based studies
• Possibility of treating the same patient/tumor with
different drugs, alone and in combination
• Outcome is not confounded by cytotoxic activity of
conventional chemotherapeutics
• Treatment versatility: the system is amenable for
manipulation of treatment schedules
• Less stringent ethical issues: use of investigational
compounds awaiting approval for use in humans
• Virtually unlimited material available for genomic and
molecular characterization
18. Further data: cancer cell lines
The Cancer Cell Line Encyclopedia Consortium & The Genomics of Drug Sensitivity in Cancer Consortium
Nature 1-4 (2015) doi:10.1038/nature15736
19. Data integration,
analysis and
visualisation
Individual
patient
Patients
• Clinical data
• Histology
• Molecular profiles
Patient-derived models
(xenografts, cell cultures)
• Histology
• Molecular profiles
• Pharmacology
Public data
• Molecular datasets
• Pharmacogenomics
• Biomarker signatures
Bioinformatician
/ Translational
researcher
Data
mining
New biomarker /
stratification
hypotheses
T
C
G
A
I
C
G
C
Capture,Storage,
Standardisation
Integrative
visual reports
Diagnosis,
prognosis and
therapeutic
decision.
"Precision Oncology"
20. Colorectal Cancer: progression and
hallmarks
Uncontrolled proliferation
Resistance to death signals
Invasion and metastasis
21. Colorectal cancer molecular heterogeneity
85-90%
10-15%
MSS
MSI
MSS
MSI
Normal
epithelium
Hyperproliferative
epithelium
Early Intermediate Late
adenoma
Carcinoma
Invasion and
metastasis
Loss of
APC
DNA
hypomethylation
KRAS
activation Loss of 18q
PRL3
amplification
TGFβRII, PIK3CA mutations
Loss of p53
Normal
epithelium
Hyperproliferative
epithelium
Early Intermediate Late
adenoma
Carcinoma
Invasion and
metastasis
MMR mutation
MLH1 hypermethylation
BRAF
activation
PIK3CA mutations
Loss of p53
TGFβRII, IGF2R, BAX, E2F4,
MRE11A, hRAD50
frameshift mutations
Mutator phenotype
23. Colorectal cancer transcriptional subtyping:
the class discovery-class prediction strategy
• Class discovery:
Group samples based on their gene expression
profile and find the optimal number of groups
("subtypes")
• Class prediction:
Use subtype-specific genes to classify
independent CRC samples
• Explore correlations between subtypes and molecular,
biological and clinical features
37. Comments
• The MMRA pipeline combines supervised statistics with
unsupervised network analysis to detect microRNAs
potentially driving CRC subtypes
• This approach allowed detection of four microRNAs
antagonizing the poor-prognosis SSM subtype in tumor
samples and cell lines
• This functional role was validated in vitro, by
downregulating each microRNA in CRC cell lines
38. Why WT cell lines are resistant to
therapy?
Back to CRC treatment: possible alternative options
to treat WT tumors resistant to cetuximab?
40. What is an outlier?
"…rara avis in terris nigroque simillima cygno"
Juvenale, Saturae, VI, 165.
“…a rare bird in the lands and very much like a black swan"
When the phrase was coined,
black swans were presumed not to exist.
Indeed, they do exist.
42. Outlier kinase genes are aberrantly expressed
in cell lines
CRC cell lines
(n=151)
43. Outlier kinase genes identified in cell lines are
aberrantly expressed also in CRC tumors
CRC cell lines
(n=151)
CRC tumors
(TCGA; n=352)
44. Gene outlier Cell line
CTX
sensitivity
Drugs available
ALK CRC-01 RES Crizotinib
NTRK1 CRC-71 RES
Imatinib, Nilotinib,
CEP107, AR523
NTRK2 CRC-122 RES
Imatinib, Nilotinib,
CEP107, AR523
FGFR2 CRC-97 RES AZD4547
KIT CRC-43 RES Imatinib, Nilotinib
PDGFRA CRC-12 RES
Sorafenib, Sunitinib
Imatinib, Nilotinib
RET CRC-97 RES Sunitinib
Outliers: 7/8 are druggable kinase
50. Comments
• The compendium of 151 CRC cell lines properly
recapitulates:
– genetic heterogeneity of CRC
– transcriptional subtypes and mRNA/microRNA interactions
– Genotype- and subtype-drug correlations
• Transcriptional outlier analysis identified a subset of
KRAS/BRAF wild type cells, intrinsically resistant to
EGFR inhibition, which are functionally and
pharmacologically addicted to kinase genes
ALK <1%
RET <1%
KIT <1%
FGFR2 <1%
NTRK1 <1%
NTRK2 <1%
51. CRC PDXs
@IRCC
n = 180
n = 110
n = 515
Bertotti et al, Cancer Discovery 2011
52. Response of colorectal cancer PDXs to cetuximab
Genetic status significantly affects CRC response rate
Cancer Discovery 1:508-523
KRAS cod 12
KRAS cod 13
53. Genetic selection increases the response rate
Other genetic biomarkers of resistance?
Cancer Discovery 1:508-523
55. Analysis of gene expression outliers
HER2 amplification, in cetuximab-resistant CRC
Cancer Discovery 1:508-523
56. Efficacy of combinatorial anti-EGFR/HER2
treatment in HER2-amplified CRC PDXs
Pertuzumab
Vehicle
Cetuximab+Pmab
Lapatinib
Cmab+Lapatinib
Pmab+Lapatinib
Cancer Discovery 1:508-523
57. Comments
• Dataset size matters
• Once a targetable genetic lesion is identified, not any
drug targeting that lesion will be effective
• Rational combinations are more likely to be effective,
and preclinical testing may help choose the most
promising one
HERACLES trial:
Targeting HER2 & EGFR
in liver-metastatic CRC
with amplified HER2
58.
59. CRC transcriptional subtypes and PDXs
Key questions:
• How reliably can the transcriptional subtypes, and their
correlates, be explored in CRC PDXs?
– Are the subtypes maintained in PDXs?
– What is the role of the tumor stroma?
• How reliably could information obtained in PDXs be
applied to CRC patients?
60. Tumor vs PDX transcriptome
Total
RNA
Total
RNA
Expression in TumorExpressioninPDX
Human-specific Array
Isella et al., Nature Genetics 47:312, 2015
PDX Sample
Cancer Cells
(human)
Stromal Cells
(human)
+
Human Tumor
Cancer Cells
(human)
Stromal Cells
(mouse)
+
?
61. Tumor vs PDX transcriptome
Total
RNA
Total
RNA
Expression in TumorExpressioninPDX
Human-specific Array
Genes "Lost in PDX"
Isella et al., Nature Genetics 47:312, 2015
PDX Sample
Cancer Cells
(human)
Stromal Cells
(human)
+
Human Tumor
Cancer Cells
(human)
Stromal Cells
(mouse)
+
62. Infl – CMS1 Goblet – CMS3 Ent – CMS2
TA – CMS2 Stem – CMS4
Expression in Tumor
ExpressioninPDX
Expression of subtype genes in tumor vs
PDX
63. Classification "reshuffling" in PDX
Inflammatory
Goblet
Enterocyte
TA
Stem
CMS1
MSI IMMUNE
CMS3
METABOLIC
CMS2
CANONICAL
CMS4
MESENCHYMAL
64. Hunting for "lost" genes by
RNAseq analysis of PDX samples
RNAseq
Reads mapped
only on Hs
Genome
Reads mapped
only on Mm
Genome
Cancer Cell
Transcriptome
Stromal Cell
Transcriptome
PDX Sample
Cancer Cells
(human)
Stromal Cells
(mouse)
+
Total
RNA
Isella et al., Nature Genetics 47:312, 2015
65. CMS4/SSM genes are expressed
as mouse transcripts in PDXs
Mousetranscriptlevel
(RPM)
Human transcript level (RPM)
INFL-GOBL
(CMS 1-3)
TA-ENT
(CMS2)
SSM
(CMS4)
67. Definition of stromal cell-specific signatures
Differential
Gene
Expression
PDX human arrays
Genes
never expressed
by cancer cells
Stromal cell-
specific
signatures
Isella et al., Nature Genetics 47:312, 2015
74. A CAF-specific score predicts CRC prognosis
and treatment response
All cases
No Adjuvant
Treatment
Adjuvant
Treatment
Isella et al., Nature Genetics 47:312, 2015
75. A compound stromal score predicts response of
rectal cancer to preoperative radiotherapy
Isella et al., Nature Genetics 47:312, 2015
76. Comments
• Transcriptional subtypes hold reasonably well in PDXs,
with the exception of SSM
• SSM genes are expressed by stromal rather than
epithelial cancer cells
• Most SSM genes are readily detected in PDX samples
as mouse rather than human transcripts, confirming their
stromal origin
• Stromal transcriptomes reflect the composition and
functional state of stromal cells, with prognostic and
therapeutic implications.
77. Class discovery in CRC PDXs
• Expression dataset (Illumina human arrays) on 515
PDXs from 250 tumors
• Class discovery by NMF-consensus
• Construction of a classifier excluding genes also
expressed by the stroma
• Assessment of classification performance on
independent human CRC datasets and analysis of
molecular, biological and clinical correlates
79. TARGET
Pevonedistat blocks the NEDD8
conjugation pathway
• Shah et al., CCR 2016: Phase I
Study on Relapsed/Refractory
Multiple Myeloma or Lymphoma.
• Sarantopoulos et al., CCR 2015:
Phase I Study on Advanced Solid
Tumors.
N8
Cul
N8
E1 E2
N8
E3NEDD8-Activating
Enzyme
Pevonedistat (MLN4924)
N8
80. Matched
PDXs
CRC cell lines
(n=122)
In vitro
response
CRC liver MTS
(n=87)
Molecular
profiles
A two-arm preclinical platform to study CRC
response to NEDD8 pathway inhibition by
pevonedistat.
Molecular
predictor
Predicted
sensitive
Predicted
resistant
In vivo
response
In vivo
response
81. Data integration,
analysis and
visualisation
Individual
patient
Patients
• Clinical data
• Histology
• Molecular profiles
Patient-derived models
(xenografts, cell cultures)
• Histology
• Molecular profiles
• Pharmacology
Public data
• Molecular datasets
• Pharmacogenomics
• Biomarker signatures
Bioinformatician
/ Translational
researcher
Data
mining
New biomarker /
stratification
hypotheses
T
C
G
A
I
C
G
C
Capture,Storage,
Standardisation
Integrative
visual reports
Diagnosis,
prognosis and
therapeutic
decision.
"Precision Oncology"
Data integration,
analysis and
visualisation
82.
83. Pathology
XENOPATIENTS
PRECLINICAL
STUDIES
DATA
INTEGRATION
MODULE 3:
MOLECULAR DATA
MODULE 1:
CLINICAL DATA
Laboratory Imaging
Medical
Records
Interface
Interfaces Interfaces Interfaces Interfaces
BIOREPOSITORY
DNA profiling
Interfaces
RNA profiling
Interfaces
Microscopy
Interfaces
Protein profiling
Interfaces
Interface
MODULE 2:
BANK/XENO DATA
Interface
MODULE 4:
in vitro DATA
Interface
TISSUE
SAMPLES
84.
85. MULTI-DIMENSIONAL MOLECULAR PROFILING
(primary samples, xenopatients, cells)
microRNA
profiling
Sequencing
Genotyping &
Array-CGH
Epigenomics Proteomics
mRNA
profiling
Sequence/expression
databases
Gene sets (MSigDB)
Functional databases miRNA targets
Promoters
protein interactionPublished signatures
Genome and
transcriptome
DATA INTEGRATION
STANDARDIZATION – STORAGE
PROCESSING – ANNOTATION
ANALYSIS – VISUALIZATION
CLINICAL AND
PATHOLOGICAL
DATA
PRECISION MEDICINE
Predictions of individual treatment
response/resistance, risk stratification,
definition of clinical decision trees
Treatments and
responses in
Xenopatients
CANDIDATE PRIORITIZATION
Coding/non-coding sequences whose
gain/loss-of-function is likely to affect
response to treatments
DATA MINING
Follow-up
Anamnestic data
Clinical history
Imaging
Pathology
Treatment(s)
EXPERIMENTAL
DATA
Treatments and
responses in cells
Functional/drug
screenings in
cells
99. a tri-dimensional environment in which different
types of information, such as gene expression,
dosage, methylation and clinical data can be
concomitantly visualized and analyzed.
:
http://genomecruzer.com/
101. Summary
• Multiple levels of molecular alteration are functionally
involved in cancer initiation, progression, and response to
treatment.
• Tumor cells interact with stromal and inflammatory cells,
which influence cancer progression and therapy response.
• Pathological, radiological, clinical and preclinical data
contribute important prognostic and predictive information
that should be further incorporated
• Reliable prediction of tumor aggressiveness and therapy
response requires integrative analysis of all data.
• Particular attention should be dedicated to interactive visual
environments, where end-users could easily navigate the
integrated information, at the genome, gene or patient level.
102. Oncogenomics
Claudio Isella
Gabriele Picco
Consalvo Petti
Sara Bellomo
Andrea Terrasi
Daniela Cantarella
Roberta Porporato
Molecular Oncology &
Cancer Epigenetics
Carlotta Cancelliere
Mariangela Russo
Michela Buscarino
Federica Di Nicolantonio
Alberto Bardelli
Surgery &
Gastroenterology
Alfredo Mellano
Michele De Simone
Andrea Muratore
Giovanni Galatola
Pathology , Torino University
Paola Cassoni
Translational Cancer
Medicine
Giorgia Migliardi
Davide Torti
Francesco Galimi
Francesco Sassi
Eugenia Zanella
Stefania Gastaldi
Andrea Bertotti
Livio Trusolino
Candiolo Cancer Institute
UZ Brussel
Mark De Ridder
Guy Storme
Acknowledgments
Millennium Pharmaceuticals
Allison Berger
enzo.medico@ircc.it
Luca Vezzadini
Riccardo Corsi
www.kairos3d.it