Transcript Metastasis-Free Survival - Mediterranean School Of Oncology

Gene-expression signatures for breast cancer
prognosis, site of metastasis, and therapy resistance
John Foekens
Josephine Nefkens Institute
Dept. Medical Oncology
Mediterranean School of Oncology: Highlights in the Management of Breast Bancer
Rome, November 16, 2006
Breast cancer incidence
Worldwide ~1,000,000 new cases / year
1 out of 9 women will get breast cancer during life
~40% of the patients will die of breast cancer
Reason:
Development of resistance to therapy
in metastatic disease
What do we need?
Prognostic factors that accurately can predict
which patient will develop a metastasis and who
does not.

High-risk patients should receive adjuvant
therapy, while the low-risk patients could
be spared the burden of the often toxic
therapy or could be offered a less
aggressive treatment.
Metastasis-Free Survival (%)
MFS as a function of the number of involved lymph nodes
100
~35%
80
0
60
1
2-4
40
5-9
20
10
0
0
30
60
Time (months)
90
120
Metastasis-Free Survival (%)
MFS as a function of the number of involved lymph nodes
100
Absolute survival benefit: 5 - 15%
80
}
60
40
}
20
}
0
0
30
60
90
Time (months)
Adjuvant hormonal or chemotherapy
120
Metastasis-Free Survival (%)
MFS in lymph-node negative patients
100
~35%
80
60
~65% cured by local treatment:
40
surgery ± radiotherapy
20
0
0
30
60
90
Time (months)
Adjuvant therapy necessary ??
120
Consensus criteria for node-negative breast cancer
Age and menopausal status
Histological tumor grade
Tumor size
Steroid hormone-receptor and HER2 status

85 – 90% of node-negative patients should
receive adjuvant therapy

Over-treatment since only 5 – 10% of the
node-negative patients will benefit by cure
What do we need more?
Predictive factors that accurately can predict
which patient will respond favorably to a
certain type of treatment and who does not.
Final goal: Individualized targeted treatment
which is based on prognostic and predictive
factors, and new targets for treatment.
Steps in tumor progression
?
?
High-throughput methodologies
SNP arrays
Genetics
Epigenomics
CGH of BAC arrays
DNA-methylation profiling
mRNA
Genomics
Gene-expression profiling
Multiplex RT-PCR
TK profiling
Proteomics
Multiplex ELISA
Mass-spectrometry
High-throughput methodologies
SNP arrays
Genetics
Epigenomics
CGH of BAC arrays
DNA-methylation profiling
mRNA
Genomics
Gene-expression profiling
Multiplex RT-PCR
TK profiling
Proteomics
Multiplex ELISA
Mass-spectrometry
Gene expression analysis
<1995: Northern Blotting, RNAse protection etc
1 Week: Analyse several genes on 10s of samples
>1995: DNA Microarrays
1 Week: Analyse whole genome on 10s of samples
Chip design
Fluorescently labeled sample
Microarray
Add Sample
Silicon wafer
Glass microscope slide
Nitrocellulose
DNA Probes: 20 – 70 bases
Hybridization between sample and probe
Chip workflow
Sample
prep
Subtypes of breast cancer
“Molecular portraits of
human breast tumors”
496 “intrinsic” genes described by
Perou et al. (Nature 2000); array
with 8102 human genes
65 breast samples / 42 patients
78 breast carcinomas
3 fibroadenoma’s
4 normal breast tissues
Patients from Norway:
Very heterogeneous with respect
to nodal status, adjuvant and neoadjuvant therapy
Perou & Sorlie et al.
Nature 2000; PNAS 2001
Subtypes of breast cancer
HER2
ER
EGFR
Rotterdam data set:
Affymetrix U133A chip
luminal B
luminal A
HER2
norm
basal
344 untreated lymph
node-negative patients
The Amsterdam prognostic profile
Training set: 78 patients
Study design
gyui
78 breast tumors
Patients < 55 years
Tumor size <5 cm
Lymph node negative (LN0)
No adjuvant therapy
Prognosis reporter genes
Distant metastasis
< 5 years (n=34)
NO distant
metastasis
in 5 years (n=44)
van ‘t Veer et al, Nature 2002
70-gene signature
 Validation
MFS in 151 LNN patients
Testing set: 295 patients,
including 151 lymph-node
negative patients
van de Vijver et al, NEJM 2002
The Rotterdam – Veridex study
Aim:
To develop a prognostic profile that can be
used for all lymph-node negative breast
cancer patients, irrespective of age, tumor
size, and steroid hormone-receptor status.
Lancet 365:671-679 (2005)
Patients & Methods
Patients
Total: 286 primary breast cancer patients
No (neo-)adjuvant systemic therapy (
pure prognosis)
Median follow-up 101 months
Clinical endpoint: metastasis-free survival (MFS)
Methods
Quality check of RNA by Agilent BioAnalyzer
Affymetrix oligonucleotide microarray U133A GeneChip
(22,000 transcripts)
RNA isolation
frozen primary breast cancer tissue
30  sections
>70% tumor area
30  sections
check
RNA isolation
check
RNA isolation
combine
Agilent BioAnalyzer
Clear distinct 18S and 28S peaks
RNA quality check
No minor peaks present
Area under 18S and 28S peaks >15% of total RNA area
28S/18S ratio should be between 1.2 and 2.0
Analysis of metastasis-free survival
Affymetrix
oligonucleotide
microarray
time
primary
tumor
metastasis-free survival
surgery
NO adjuvant systemic therapy
metastasis
Gene-expression profiling
Steps to follow in the clinical development of
expression profiles
Training set to generate profile
Independent testing set for validation of the profile
Multi-center (retrospective) study
Prospective clinical trial
Gene-expression profiling
Steps to follow in the clinical development of
expression profiles
Training set to generate profile
Independent testing set for validation of the profile
Multi-center (retrospective) study
Prospective clinical trial
Genes
Unsupervised clustering analysis
ER-
ER+
Tumors
Determining the signature for ER+ and ER- patients
286 LNN patients
ER status
ER-positive
ER-negative
209 patients
77 patients
supervised classification
80 patients
(training)
35 patients
(training)
gene selection
(Cox model, bootstrapping)
76 gene set
171 patients
(testing)
validation
ER negative
0.95
16 genes
0.90
ER positive
0.85
AUCs of ROC
1.00
Determining the 76-gene signature
60Score
genes
=AI+
Relapse
60
Iw x
0.80
i =1
i
16
i
+ B  (1 - I) +  (1 - I ) w j x j
j=1
where
~
0
Wang et al, Lancet 2005
1 if ER level > 10
=
I 
115 training set patients

0
if
ER
level
10

A and B are constants
50
100
150
200
w i is the standardized Cox regression coefficient
Number of genes
x i is the expression value in log2 scale
Gene-expression profiling
Steps to follow in the clinical development of
expression profiles
Training set to generate profile
Independent testing set for validation of the profile
Multi-center (retrospective) study
Prospective clinical trial
Comparison of the 76-gene signature and the current
conventional consensus on treatment of LNN breast cancer
Patients guided to receive adjuvant therapy
Metastatic disease at 5 years
Metastatic disease free at 5 years
St. Gallen 2003
52/55 (95%)
104/115 (90%)
NIH 2000
52/55 (95%)
101/114 (89%)
76-gene signature
52/65 (93%)
60/115 (52%)
MFS in patients with T1 tumors
0.8
0.6
0.4
0.2
poor signature (n = 47)
Sensitivity 96% (24/25)
Specificity 57% (31/54)
0.0
Metastasis-Free Survival
1.0
good signature (n = 32)
0
20
HR: 14.1 (95% CI: 3.34–59.2), P = 1.6x10-4
40
Months
60
80
Gene-expression profiling
Steps to follow in the clinical development of
expression profiles
Training set to generate profile
Independent testing set for validation of the profile
Multi-center (retrospective) study
Prospective clinical trial
2nd validation: EORTC - RBG
Participating institutions:
- University Medical Center Nijmegen, The Netherlands
- Technische Universität München, Germany
- National Cancer Institue, Bari, Italy
- Institute of Oncology, Ljubljana, Slovenia
Methods EORTC – PBG validation study
Patients
Total: 180 node-negative primary breast cancer patients
No (neo-)adjuvant systemic therapy
Median follow-up: 100 months
Clinical endpoint: metastasis-free survival (MFS)
Methods
Tissues sent to Rotterdam for RNA isolation
Quality check of RNA by Agilent BioAnalyzer
Affymetrix dedicated VDX2 oligonucleotide microarray
(76 genes + 221 control genes) analysis at Veridex
43% of the tumors have a ‘good’ signature
2nd validation: MFS in 180 patients
0.8
0.6
0.2
0.4
poor signature (n = 102)
HR: 7.41 (95% CI: 2.63–20.9), P = 8.5x10-6
0.0
Metastasis-Free Survival
1.0
good signature (n = 78)
0
5
Years
Foekens et al, JCO 2006
10
Multivariate analysis in multi-center validation
Metastasis-Free Survival
HR
(95% CI)
P-value
Age (per 10 yr increment)
0.70
(0.44-1.11)
0.13
Menopausal status (post vs. pre)
1.26
(0.43-3.70)
0.67
Tumor size (>20 mm vs. ≤20 mm)
1.71
(0.84-3.49)
0.14
Grade (moderate/good vs. poor)
1.24
(0.61-2.52)
0.56
ER (per 100 increment)
1.00
(0.99-1.01)
0.13
11.36
(2.67-48.4)
0.001
76-gene signature (poor vs. good)
MFS in post-menopausal patients
0.8
0.6
0.2
0.4
poor signature (n = 69)
HR: 9.84 (95% CI: 2.31–42.0), P = 0.0001
0.0
Metastasis-Free Survival
1.0
good signature (n = 57)
0
5
Years
10
MFS in St. Gallen average risk group
0.8
0.6
0.2
0.4
poor signature (n = 97)
HR: 6.08 (95% CI: 2.15–17.2), P = 0.0001
0.0
Metastasis-Free Survival
1.0
good signature (n = 64)
0
5
Years
10
Site of metastasis
76 gene high risk profile is NOT able to distinguish
between site of relapse
genes
Site of metastasis was defined as follows:
AIM: Identify genes associated with a relapse to the
be
present in the primary breast tumor.
- Non-bone: women with a relapse, excluding the bone as a site
- Bone: women with a bone relapse, including those with an
bone
since
biological
features (e.g. homing) may
additional
relapse
elsewhere
of relapse
Bone metastasis
The bone is the most abundant site of distant relapse in breast,
prostate, thyroid, kidney and lung cancer patients.
Bone micro-environment may facilitate circulating cancer cells
to home and proliferate.
Bisphosphonate therapy available.
Profile for bone metastasis
286 patients, 107 relapses (Lancet, 2005)
Training
Validation
72 patients:
- 46 x bone
- 26 x non-bone
SAM and PAM analysis
31 - gene set
35 patients:
- 23 x bone
- 12 x non-bone
Performance of the 31-gene predictor
Validation set of 35 patients
Probe-id
205009_at
204623_at
209173_at
214440_at
205081_at
214774_x_at
214858_at
219197_s_at
215108_x_at
206754_s_at
210056_at
205186_at
203130_s_at
Sensitivity:
100% (23/23)
Specificity:
50% (6/12)
Gene Symbol
TFF1
TFF3
AGR2
NAT1
CRIP1
TNRC9
--SCUBE2
TNRC9
CYP2B6
RND1
DNALI1
KIF5C
Smid et al, JCO 2006
SAM
Score
-4,92
-4,23
-4,06
-4,04
-3,80
-3,72
-3,60
-3,59
-3,57
-3,57
-3,48
-3,45
-3,42
Fold
Change
3,1
2,6
1,9
2,5
1,9
1,9
2,0
2,1
1,9
2,1
1,7
2,0
2,0
Genes higher
expressed in
bone
FDR (%)
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
*
1,9
PAM
yes
yes
yes
yes
yes
yes
yes
yes
Gene Title
trefoil factor 1
trefoil factor 3 (intestinal)
anterior gradient 2 homolog
N-acetyltransferase 1
cysteine-rich protein 1 (intestinal)
trinucleotide repeat containing 9
Pp14571
signal peptide, CUB domain, EGF-like 2
trinucleotide repeat containing 9
cytochrome P450, family 2, subfamily B, polypeptide 6
Rho family GTPase 1
dynein, axonemal, light intermediate polypeptide 1
kinesin family member 5C
Pathway analysis
There is criticism and non-understanding about
the minimal overlap of individual genes between
various multigene prognostic signatures.
All gene signatures for separating patients into different
risk groups, so far, were derived based on the performance
of individual genes, regardless of its biological processes
or functions.
It might be more appropriate to study biological themes,
rather than individual genes.
Predictive signatures
Diagnosis / Surgery
Relapse
? Predictive profile
Response
Systemic therapy
No response
Analysis of type of response
Microarray
time
primary
tumor
CR / PR
PD
metastasis-free survival
surgery
metastasis
tamoxifen
Tamoxifen profile in ER+ tumors
112 patients (60 progressive disease, PD, 52 objective response, OR)
cDNA array
analysis
QC arrays
46 patients (25 PD, 21 OR)
Training
66 patients (35 PD, 31 OR)
Validation
BRB, duplicate arrays
P<0.05, QC spots
81 - gene set
Discriminatory genes
44 - gene set
Predictive signature
Molecular classification: 1st line tamoxifen
112 ER+ primary breast tumors
from patients with recurrent
disease and treated with first-line
tamoxifen
Training set:
21 OR v 25 PD
81 genes differentially expressed
44-gene predictive signature
Validation:
31 OR v 35 PD
Response :
OR = 3.16
(P=0.03)
HR = 0.48
(P=0.03)
PFS:
Jansen et al, JCO 2005
What do we need more?
Predictive factors that accurately can predict which patient
will respond favorably to a certain type of treatment and who
does not.
Approach:
Microarray analysis of primary tumor RNA to assess the type
of response (objective measure) in the metastatic setting;
- 1st line tamoxifen therapy
- 1st line chemotherapy
Analysis of type of response
Affymetrix U133plus2 array: 54,000 probe IDs
time
primary
tumor
CR / PR
metastasis-free survival
surgery
metastasis
chemotherapy
PD
Summary gene expression signatures
- 76-gene prognostic signature
- Bone metastasis signature
- Chemotherapy resistance signature
- Tamoxifen resistance signature
- Liver metastasis signature (in progress)
- Pathway-derived signatures
- Others ……
+ a growing number of published signatures for various clinical questions
Contributors gene-expression profiling
Erasmus MC
Anieta Sieuwerts, Mieke Timmermans, Marion Meijer-van Gelder, Maxime Look,
Anita Trapman, Miranda Arnold, Anneke Goedheer, Roberto Rodriguez-Garcia,
Els Berns, Marcel Smid, John Martens, Jan Klijn & John Foekens
Veridex LLC (Johnson & Johnson), La Jolla, USA
Yixin Wang, Yi Zhang, Dimitri Talantov, Jack Yu, Tim Jatkoe & David Atkins
EORTC – RBG members (1st multi-center validation)
-Nijmegen:
P. Span, V. Tjan-Heijnen, L.V.A.M. Beex, C.G.J. Sweep
-Munich:
N. Harbeck, K. Specht, H. Höfler, M. Schmitt
-Bari:
A. Paradiso, A. Mangia, A.F. Zito, F. Schittulli
-Ljubljana:
R. Golouh, T. Cufer
Third multi-center validation, institutions above +
+Basel
S. Eppenberger et al.
+Dresden
M. Kotzsch et al.
+Innsbruck
G. Daxenbichler et al.
TransBig group:
second multicenter
validation study