Download Breast cancer classification: beyond the`intrinsic` molecular subtypes

Breast cancer classification:
beyond the‘intrinsic’ molecular subtypes
Britta Weigelt, PhD
Signal Transduction Laboratory
CRUK London Research Institute
Summary
•  Breast cancer heterogeneity
•  Molecular classification of breast cancer
•  Prognostic gene signatures
•  Outlook
Breast cancer patient management
Size
Grade
Type
Lymph node metastasis
Vascular invasion
HER2
HER2
ER, PR and HER2
Breast cancer
‘Individualised’
breast patient
cancer therapy
patient therapy
‘Intrinsic’ molecular subtypes of breast cancer
Normal
Breast Luminal B
Basal-like
HER2
ER-negative!
Perou et al, Nature, 2000; Sorlie et al, PNAS 2003
Luminal A
ER-positive!
“Intrinsic” gene set
‘Intrinsic’subtypes are associated with outcome
Normal!
Breast! Luminal B!
HER2+!
Luminal A!
Basal-like!
Perou et al, Nature 2000; Sorlie et al, PNAS 2001; Hu et al, BMC Genomics 2006
Cancer Invest 2008
Identification of intrinsic molecular subtypes
Hierarchical clustering
‘Intrinsic’gene lists
Normal
Breast
Basal-like
HER2
Centroids
Single sample predictors
Luminal B
Luminal A
-  Large number of samples
-  Retrospective assignment
-  Centroid: mean expression profile for
each of the five subtypes
-  Classification of individual samples
-  Prospective assignment
‘Intrinsic’ molecular subtype evolution
‘Intrinsic’ genes
Single sample
predictor genes
496
Perou CM et al, Nature 2000
456
Sørlie T et al, PNAS 2001
534
500
1300
306
1906
50
Sørlie T et al, PNAS 2003
Hu Z et al, BMC Genomics 2006
Parker JS et al, J Clin Oncol 2009
Hierarchical cluster analysis
Limitations hierarchical clustering
•  Clustering algorithms always detect clusters, also in random data
•  Stability of clusters identified by hierarchical clustering analysis
•  Number of clusters is unknown
Aim
• 
To determine the objectivity and inter-observer
reproducibility of the assignment of molecular subtype
classes by hierarchical cluster analysis
3
2
1
3
2
2
1
2
Material and Methods 1
•  3 publicly available datasets
–  NKI-295 dataset (n=295)
–  Wang dataset (n=286)
–  TransBig dataset (n=198)
•  5 intrinsic gene lists
– 
– 
– 
– 
– 
Perou et al, 2000
Sorlie et al, 2001
Sorlie et al, 2003
Hu et al, 2006
Parker et al, 2009
•  5 observers
Material and Methods 2
1. 
Inter-observer agreement (%)
2. 
Free-marginal Kappa scores
•  for the whole classification
•  for each molecular subtype separately
Kappa scores
Slight: 0.01-0.20
Fair: 0.21-0.40
Moderate: 0.41-0.60
Substantial: 0.61-0.80
Almost perfect: 0.81-0.99
Molecular subtype assignment based on
dendrogram analysis is subjective
Mackay A*, Weigelt B* et al, JNCI 2011
Basal-like and HER2 ‘intrinsic’ subtypes are
reproducibly identified
Mackay A*, Weigelt B* et al, JNCI 2011
Molecular subtype evolution
‘Intrinsic’ genes
Single sample
predictor genes
496
Perou CM et al, Nature 2000
456
Sørlie T et al, PNAS 2001
534
500
1300
306
1906
50
Sørlie T et al, PNAS 2003
Hu Z et al, BMC Genomics 2006
Parker JS et al, J Clin Oncol 2009
Do different SSPs consistently
classify the same patients into the
molecular subtypes?
Agreement between different SSPs
performed by Sorlie and Perou
NKI-295 cohort
Sorlie’s SSP
Chang et al
(Sorlie’s group)
Hu’s SSP
Fan et al
(Perou’s group)
Agreement: moderate
Kappa score: 0.527 (95% CI 0.456-0.597)
Kappa scores
Slight: 0.01-0.20
Fair: 0.21-0.40
Moderate: 0.41-0.60
Substantial: 0.61-0.80
Almost perfect: 0.81-0.99
Material and Methods
•  3 publicly available datasets
–  NKI-295 dataset (n=295)
–  Wang dataset (n=286)
–  TransBig dataset (n=198)
•  1 in-house dataset
–  Grade III invasive ductal carcinomas, microdissected (n=53)
•  3 SSPs
–  Sorlie et al, 2003
–  Hu et al, 2006
–  Parker et al, 2009
•  Agreement between molecular subtype assignment
-  Kappa scores
Weigelt et al, Lancet Oncol 2010
Reproducibility of ‘intrinsic’ molecular subtypes
NKI-295 dataset
295 cases
Wang dataset
286 cases
TransBig dataset
198 cases
Sorlie SSP, 2003
Hu SSP, 2006
Parker SSP, 2009
Sorlie SSP, 2003
Hu SSP, 2006
Parker SSP, 2009
Sorlie SSP, 2003
Hu SSP, 2006
Parker SSP, 2009
GIII IDC dataset
53 cases
Sorlie SSP, 2003
Hu SSP, 2006
Parker SSP, 2009
-  Agreement – moderate to substantial (κ=0.40 - 0.79)
-  Classification of each patient is dependent on the SSP
-  Only basal-like form a stable group
Outcome prediction using distinct SSPs
Weigelt et al, Lancet Oncol 2010
Outcome prediction using distinct SSPs
Weigelt et al, Lancet Oncol 2010
5715 breast tumours
Haibe-Kains B et al, JNCI 2012
Assignment of luminal subtypes 1
Hierarchical clustering
Single sample predictors
Sorlie SSP, 2003
Hu SSP, 2006
Parker SSP, 2009
Weigelt et al, Lancet Oncol 2010; Mackay A*, Weigelt B* et al, JNCI 2011
Assignment of luminal subtypes 2
Luminal A: ER
Proliferation
Luminal B: ER
Proliferation
???
Reis-Filho & Pusztai, Lancet 2011
12 years of molecular subtyping
•  ER-positive and ER-negative tumours
–  Fundamentally different diseases
•  Breast cancer molecular subtypes
– 
– 
– 
– 
Not stable
Only basal-like is robust
Limited clinical application
No validated/ standardised methodology
•  PAM50?
Additional molecular subtypes"
•  Interferon-rich
•  Molecular apocrine
•  Claudin-low
•  Molecular subtypes of triple negative breast cancer
•  METABRIC subtypes
Hu et al, BMC Genomics 2006; Farmer et al, Oncogene 2005; Doane et al, Oncogene 2006; Prat et al, Breast Cancer Res 2010;"
Lehmann et al, JCI 2011; Curtis et al, Nature 2012 "
Prognostic gene signatures
Mammaprint
Oncotype DX (21-gene signature)
ER+/ LN-/ Tamoxifen-treated patients
Proliferation
Ki67
STK15
Survivin
CCNB1
MYBL2
HER2
GRB7
HER2
GSTM1
Oestrogen
ER
PGR
BCL2
SCUBE2
CD68
Invasion
MMP11
CTSL2
BAG1
Reference
ACTB
GAPDH
RPLPO
GUS
TFRC
Recurrence score
Low risk – RS≤18
Intermediate risk – 18>RS<31
High risk – RS≥31
A signature to rule them all?
A signature to rule them all
Fan et al. NEJM 2006; Sotiriou et al. JNCI 2006
Proliferation
Proliferation
Meta-analysis – gene signatures
Blue dots: good prognosis
Red dots: poor prognosis
Wirapati et al. Breast Cancer Res 2008;10:R65
What do prognostic signatures offer?
•  ER-positive disease: good discriminatory power
•  Limited value for ER-negative disease
•  Correlate with proliferation (and grade!)
–  Ki-67?
Immune response related signatures are
prognostic in TNBC
…but the good prognosis group still has a
high number of events
Rody et al. Breast Cancer Res 2010; Karns et al. PLoS One 2011
What do prognostic signatures offer?
•  ER positive disease - good discriminatory power
•  Limited value for ER negative disease
•  Correlate with proliferation (and grade!)
–  Ki-67?
•  Immune response-related signatures
–  Prognostic in ER-/HER2- and HER2+ disease
–  Potential predictive of response to chemotherapy
Take home messages
•  Molecular classification
–  not ready yet for use in clinical practice
–  standardised methods/ definitions required (PAM50?)
•  First generation prognostic signatures
–  complementary to histopathology
–  determined by proliferation
–  discriminatory power only in ER-positive disease
• 
Second generation immune-related prognostic signatures
–  discriminatory power in ER-negative disease
–  not yet sufficient for clinical decision-making
Outlook
survival
Good
Current classification: descriptive and prognostic
Poor = treat
time
Weigelt et al, Nat Rev Clin Oncol 2011
Outlook
survival
Good
Current classification: descriptive and prognostic
Poor = treat
time
Future: predictive sub-classification
Mutation X?
Amplification Y?
Weigelt et al, Nat Rev Clin Oncol 2011
Sensitive drug A
Resistant drug B
Acknowledgements
Julian Downward
Alan Mackay
Rachael Natrajan
Maryou Lambros
Anita Grigoriadis
Alan Ashworth
Jorge Reis-Filho
Roger A‘Hern
Mitch Dowsett
Bas Kreike
Immune response may predict pathological
complete response following neoadjuvant chemo
All breast cancers
statistically significant
Ignatiadis et al. J Clin Oncol 2012
ER-/HER2-
not statistically significant
HER2+
ER+/HER2-
PAM50 vs Ki67
103 ER+/HER2- breast cancers profiled with PAM50 and Ki67 IHC
Ki67
Luminal A (n=76)
Luminal B (n=27)
<13.25%
64 (84%)
10 (37%)
≥13.25%
12 (16%)
17 (63%)
Kappa score = 0.4607 (0.2609 to 0.6605)
Kelly et al. Oncologist 2012
OncotypeDx vs PAM50
Is low RS synonymous with luminal A?
108 ER+/ HER2- breast cancers profiled with GHI OncotypeDx and ARUP labs PAM50
Low (n=59)
90%
Intermediate (n=39)
59%
33%
90%
High (n=10)
0%
Lum A
Kelly et al. Oncologist 2012
8% 2%
10%
20%
Lum B
30%
40%
50%
HER2-enriched
8%
10%
60%
70%
80%
Basal-like
90%
100%
PAM50 vs OncotypeDx
Is luminal A synonymous with low RS?
108 ER+/ HER2- breast cancers profiled with GHI OncotypeDx and ARUP labs PAM50
70
Lum A (n=76)
Lum B (n=27)
19
HER2-enriched (n=4)
30
48
25
75
Basal-like (n=1)
100
0%
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Low
Kelly et al. Oncologist 2012
33
Intermediate
High