Transcript Document

Development and Validation
of a Radiosensitivity
Signature for Breast Cancer
A collaboration between the University of
Michigan (Drs. Pierce, Speers, and Feng) and
PFS Genomics
Department of Radiation Oncology
University of Michigan
Background
>6000 women with BCS and node-negative disease
EBCTCG, Lancet 2005;366:20872106
Can we identify these patient
populations currently?
Not well from current clinicopathologic data.
Additionally, there are conflicting findings from studies
assessing association between intrinsic molecular
subtypes and radiosensitivity.
There is a clear need to develop and validate molecular
signatures to predict who will benefit from
intensification or omission of radiotherapy.
Hypothesis:
•
Gene expression profiling data from breast
cancer cell lines coupled with intrinsic
radiosensitivity information can be used to
identify a radiosensitivity signature
•
This signature can be used to identify patients in
these two disparate groups and predict likelihood
of recurrence after adjuvant RT treatment in
early stage patients
Human Breast Cancer Cell Lines
Clonogenic survival assay performed on 21
BCC lines to determine intrinsic
radiosensitivity
10 basal, 8 luminal, 3 HER2/neu cell lines
Affymetrix Microarray profiling focused on RT related gene
Spearman’s correlation methodexpression
with RT sensitivity as a continuous variable
147 genes significantly associated with RT sensitivity (80 + correlated, 67 - correlated)
Gene enrichment analysis of positively and negatively
associated radiation resistant genes
Significant gene enrichment for genes involved in cell cycle arrest and DNA damage response
Expression validation (RNA and protein)
Functional validation
Training of Signature in Human Breast Cancer Datasets with Recurrence Data
343 pts with early stage, LN- IDC treated with BCS and RT (no adjuvant systemic
chemo) with LRF survival data-Random Forest Modeling
Clinical Validation in Human Breast Cancer Datasets with Recurrence Data
184 pts with early stage, LN- IDC treated with BCS and RT (no adjuvant
systemic chemo) with LRFS data
Resistant (SF >50% )
Basal - 4
Luminal- 2
HER2 - 1
Moderately Resistant (SF 49-39%
Basal - 3
Luminal -3
HER2 - 0
Sensitive (SF <39%)
Basal - 3
Luminal - 3
HER2 - 2
P-value: NS
• For each gene (~43,000 probe sets) calculate a correlation coefficient between
expression values and SF 2Gy value as a continuous variable
4.0
.80
Surviving Fraction after 2
Gy
RAD51A
P
.70
3.5
• Identify genes that
are positively or
negatively correlated
with clonogenic
survival with a Pvalue <0.05 and a
FDR of < 1%
3.0
.60
2.5
.50
.40
2.0
.30
1.5
1.0
.20
0
R: -0.92
P value:
<0.001
FDR: <0.001
1
2
3
Normalized
Gene
Expression
Expression
• Use unsupervised
hierarchical
clustering to evaluate
the identified gene
4 lists 5
Basal B
Basal A
Luminal
HER2
67 genes
increased in
radioresistant
cell lines
147 Genes
Correlated
with
Radiation
Sensitivity
80 Genes
decreased in
radioresistant
cell lines
-2.0
0
2.0
Training of the Signature in Clinical
Dataset
343 patients treated with BCS who received adjuvant radiation therapy:
Patient characteristics:
• 343 patients with mostly pT1 or pT2 tumors
• All patients managed surgically with BCS
• 215 patients with LN- disease, 128 patients with LN+ disease
• 77% ER + ; 23% ER• Median follow-up was 6.7 years (range, 0.05 to 18.3)
• 25% (119 patients) with locoregional recurrence events
• 141 patients received systemic therapy (110 received
chemotherapy, 8 received hormonal therapy, 23 received both)
Servant, Clin Can Res. 2012;18:17041715.
Training of the Signature in Clinical
Dataset
• Genes identified used to train a Random Forest Model
• Prognostic value of each gene calculated comparing
expression values from recurrent vs. non-recurrent patients
• Performance evaluated on each subset of genes using out
of bag (OOB) error rate
• Best performing gene signature was selected and locked
for cross-validation and external validation
Uni- and Multivariate Analysis in CrossValidation Clinical Dataset- Local
Recurrence
Validation of Signature in Clinical
Dataset
295 patients treated with BCS or mastectomy who received adjuvant
radiation therapy without neo- or adjuvant chemotherapy:
Patient characteristics:
• 295 patients with pT1 or pT2 tumors
• 55% (161 patients) with BCS and 45% (134 patients) with mastectomy, all
with axillary LN dissection
• LN negative (clinically)
• 51% (151 patients) LN-negative; 49% (144 patients) LN-positive
• Age < 53 yo
• 90 patients received chemotherapy, 20 patients received hormonal
therapy; 20 patients with both
• 1 patient treated with combined chemo +hormonal therapy
• 77% (226 patients) ER + ; 23% (69 patients) ER• Median follow-up was 6.7 years (minimum follow-up was 5 years)
van de Vijver, N Engl J Med. 2002 Dec
19;347(25):1999-2009.
Sensitivity for recurrence: 85%
Negative Predictive Value: 97%
Log-rank P-value <0.001
Hazard Ratio: 6.1 (95% CI 4.48-
Uni- and Multivariate Analysis in CrossValidation Clinical Dataset- Local
Recurrence
Rate of distant recurrence as a continuous function of the Recurrence
Score®. The continuous function was generated using a piecewise log hazard ratio
model. The dashed curves indicate the 95% CI and the rug plot (x-axis) shows the
Recurrence Score for individual patients in the study. from Paik et al NEJM 2004
Rate of local-recurrence as a continuous function of the Radiation Signature
Score from the random forest model prediction. The continuous function was
generated using a Cox stepwise logistic regression model. The dashed curves
indicate the 95% CI
Conclusions
• Genes associated with intrinsic radiation resistance or
sensitivity can be identified by combining gene expression
data and clonogenic survival data from human breast cancer
cell lines
• Intrinsic radiation sensitivity is independent of breast cancer
subtype
• Radiation signature development identifies genes with novel
association to radiation resistance
• This signature predicts likelihood of response to adjuvant
radiotherapy and may be useful in identifying patients who
may require treatment intensification
Selecting a platform
• Easiest options include
• qPCR array (like Oncotype)
• Nanostring platform (like Prosigna)
• focused microarray (like Mammaprint)
• However, these options don’t allow for
• Assessment of multiple signatures (particularly
relevant for tissues from valuable phase III studies)
• Flexibility in signature refinement
• Discovery
• Thus, we decided to go with a clinical-grade highdensity array (one of the highest-throughput assays
that can be run on formalin-fixed tissue)
Precision genomic technology
Human Exon Arrays as a Discovery and Validation
Platform
ARCHIVED
FFPE TISSUE
GENETIC
MATERIAL
GENECHIP
TECHNOLOGY
GENOME
ANALYSIS
Long term followup available
Measuring activity
of genes
Genome-wide
analysis
Cancer progression
gene signature
• Uses archived FFPE tissues (success with up to 25 year old samples)
• Clinical-grade expression assay – CLIA certified lab
• Robust technology and comprehensive and in-depth data analysis
Abdueva et al., Journal of Molecular Diagnostics 2010, Vergara et al., Frontiers in Genetics 2011, Erho et al., Journal of Oncology 2012
22
Human Exon Array:
Derived from ENCODE RNA expression data
• 5 million features on array
• 1.4 million RNA transcripts
• 0.2 million mRNA exons
• 0.2 million intronic/anti-sense transcripts
• ~ 1 million non-coding RNA transcripts!
Publications using this array technology in prostate cancer
Initial reports of the Decipher signature in different cohorts
• Erho N et al. Discovery and validation of a prostate cancer genomic
Mayo
classifier that predicts early metastasis following radical prostatectomy.
PLoS One. 2013 Jun 24;8(6):e66855.
• Karnes RJ et al. Validation of a genomic classifier that predicts
Mayo
metastasis following radical prostatectomy in an at risk patient
population. J Urol. 2013 Dec;190(6):2047-53.
• Klein EA et al. A genomic classifier improves prediction of metastatic
disease within 5 years after surgery in node-negative high-risk prostate
Cleveland
cancer patients managed by radical prostatectomy without adjuvant
Clinic
therapy. Eur Urol. 2014. In press.
• Den RB et al. Genomic prostate cancer classifier predicts biochemical
failure and metastases in patients after postoperative radiation therapy.
TJU
Int J Radiat Oncol Biol Phys. 2014 Aug 1;89(5):1038-46
• Additional cohorts being assessed from the University of Michigan,
Johns Hopkins, NYU, Moffitt, and the Radiation Therapy Oncology
Group (RTOG)
GenomeDx Biosciences Confidential
17/07/2015
24
Publications using this array technology in prostate cancer
Secondary analyses of the datasets
• Prensner JR et al. RNA biomarkers associated with metastatic
Michigan progression in prostate cancer: A multi-institutional high-throughput
analysis of SChLAP1. Lancet Oncology 2014. Accepted and in press.
• Den RB et al. A genomic classifier identifies men with adverse
pathology after radical prostatectomy who benefit from adjuvant
TJU
radiation therapy. Journal of Clinical Oncology 2014. Accepted and in
press.
• Cooperberg MR et al. Combined Value of Validated Clinical and
Genomic Risk Stratification Tools for Predicting Prostate Cancer
UCSF
Mortality in a High-risk Prostatectomy Cohort. Eur Urol. 2014. Accepted
and in press.
• Ross AE et al. A genomic classifier predicting metastatic disease
progression in men with biochemical recurrence after prostatectomy.
Hopkins
Prostate Cancer Prostatic Dis. 2014 Mar;17(1):64-9
• Additional paper from the University of Michigan on age-related
biological changes in tumors
GenomeDx Biosciences Confidential
17/07/2015
25
Publications using this array technology in prostate cancer
Clinical utility studies
• Badani K et al. Impact of a genomic classifier of metastatic risk on
Columbia postoperative treatment recommendations for prostate cancer patients:
a report from the DECIDE study group. Oncotarget. 2013 Apr;4(4):600-9
• Badani KK et al. Effect of a genomic classifier test on clinical practice
Columbia decisions for patients with high-risk prostate cancer after surgery. BJU
Int. 2014. In press.
• Nguyen PL et al. Impact of a genomic classifier of metastatic risk on
post-prostatectomy treatment recommendations by radiation oncologists
and urologists. Urology 2014. In press.
Harvard/
Michigan
GenomeDx Biosciences Confidential
17/07/2015
26
Publications using this array technology in prostate cancer
Validation of laboratory biology studies
• Prensner JR et al. The long noncoding RNA SChLAP1 promotes
Michigan aggressive prostate cancer and antagonizes the SWI/SNF complex.
Nature Genetics 2013 Nov;45(11):1392-8.
• Prensner JR et al. The IncRNAs PCGEM1 and PRNCR1 are not
Michigan implicated in castration resistant prostate cancer. Oncotarget. 2014 Mar
30;5(6):1434-8.
• Hurley PJ et al. Secreted protein, acidic and rich in cysteine-like 1
(SPARCL1) is down regulated in aggressive prostate cancers and is
Hopkins
prognostic for poor clinical outcome. Proc Natl Acad Sci U S A. 2012
Sep 11;109(37):14977-82.
• Additional studies submitted to JNCI (Hopkins), European Urology
(Michigan), IJROBP (Michigan)
GenomeDx Biosciences Confidential
17/07/2015
27
Discovery of SChLAP1 (a long noncoding RNA) as the top
gene associated with metastatic progression in prostate cancer
Prensner et al, Nature Genetics, 2013; Prensner et al, Lancet Oncology (accepted), 2014
28
FFPE samples profiled using Exon arrays
Tumor Type
Prostate
Bladder
Sarcoma
Thyroid
Breast
Pancreas
Kidney
n
3,200
300
240
120
72
58
20
Erho, N., et al. Discovery and validation of a prostate cancer genomic
classifier that predicts early metastasis following radical prostatectomy.
PLoS One. 2013 Jun 24;8(6):e66855. doi:
10.1371/journal.pone.0066855. Print 2013.
Erho, N. et al. Transcriptome-wide detection of differentially expressed
coding and non-coding transcripts and their clinical significance in
prostate cancer. J Oncol. 2012;2012:541353. Epub 2012 Aug 16.
Abdueva D, et al. Quantitative expression profiling in formalin-fixed
paraffin-embedded samples by affymetrix microarrays. J Mol Diag
2010;12:409-17.
Karnes, R.J. et al. Validation of a genomic classifier that predicts
metastasis following radical prostatectomy in an at risk patient
.
population. J Urol. 2013 Dec;190(6):2047-53.
Mitra, A.P., et al. Discovery and validation of a novel expression
signature for recurrence in high-risk bladder cancer post-cystectomy. J
NCI accepted April 2014
Presner, J., et al. The long noncoding RNA SChLAP1 promotes
aggressive prostate cancer and antagonizes the SWI/SNF complex.
Nat Genetics 2013 45(11):1392-8.
Wiseman, S.M. et al. Whole-transcriptome profiling of thyroid nodules
identifies expression-based signatures for accurate thyroid cancer
diagnosis. J Clin Endocrinol Metab 2013 98(10):4072-9
Knudsen, E.S. Progression of ductal carcinoma in situ to invasive
breast cancer is associated with gene expression programs of EMT
and myoepithelia.2012 Breast Cancer Res Treat. 133(3):1009-24.
29
Exon arrays used to examine laser capture microdissected*
stromal and epithelial cells from DCIS and IBC
Knudsen, E.S. Progression of ductal carcinoma in situ to invasive breast
cancer is associated with gene expression programs of EMT and
myoepithelia.2012 Breast Cancer Res Treat. 133(3):1009-24.
*LCM peformed on FFPE specimens
Exon arrays profiled using an input of 50 ng of RNA
30
RNA Extraction using the GenomeDx protocol
• RNA extraction from formalin-fixed, paraffin-embedded
specimens follows a procedure over 3 days to first digest,
then extract, then isolate and purify RNA for expression
analysis.
• All conducted in a CLIA-certified laboratory
• Optimized for both blocks or slides
• Is now semi-automated
RTOG 96-01 RNA Yields/Purity
(~20 year old blocks)
The RNA extraction/exon array approach is now being
used for samples from the following RTOG trials:
• 96-01
• 92-02
• 94-08
• 94-13
• 99-02
• 99-10
• 01-26
Conclusions
• A reliable molecular tool is needed to personalize
radiotherapy for early stage breast cancer patients
• We have developed a signature for radiation
intensification
• We have a platform that allows for validation of existing
signatures and development of new ones
• Thanks to Ian Kunkler, David Cameron, and John Bartlett,
we have established a collaboration that aims to apply this
platform to randomized clinical trial samples