Benedikt Zimmermann

Discovery and Validation of a Prostate Cancer Genomic Classifier that Predicts Early Metastasis Following Radical Prostatectomy

Purpose Clinicopathologic features and biochemical recurrence are sensitive, but not specific, predictors of metastatic disease and lethal prostate cancer. We hypothesize that a genomic expression signature detected in the primary tumor represents true biological potential of aggressive disease and provides improved prediction of early prostate cancer metastasis. Methods A nested case-control design was used to select 639 patients from the Mayo Clinic tumor registry who underwent radical prostatectomy between 1987 and 2001. A genomic classifier (GC) was developed by modeling differential RNA expression using 1.4 million feature high-density expression arrays of men enriched for rising PSA after prostatectomy, including 213 who experienced early clinical metastasis after biochemical recurrence. A training set was used to develop a random forest classifier of 22 markers to predict for cases - men with early clinical metastasis after rising PSA. Performance of GC was compared to prognostic factors such as Gleason score and previous gene expression signatures in a withheld validation set. Results Expression profiles were generated from 545 unique patient samples, with median follow-up of 16.9 years. GC achieved an area under the receiver operating characteristic curve of 0.75 (0.67–0.83) in validation, outperforming clinical variables and gene signatures. GC was the only significant prognostic factor in multivariable analyses. Within Gleason score groups, cases with high GC scores experienced earlier death from prostate cancer and reduced overall survival. The markers in the classifier were found to be associated with a number of key biological processes in prostate cancer metastatic disease progression. Conclusion A genomic classifier was developed and validated in a large patient cohort enriched with prostate cancer metastasis patients and a rising PSA that went on to experience metastatic disease. This early metastasis prediction model based on genomic expression in the primary tumor may be useful for identification of aggressive prostate cancer.

Validation of a Genomic Classifier that Predicts Metastasis Following Radical Prostatectomy in an At Risk Patient Population

PURPOSE Patients with locally advanced prostate cancer after radical prostatectomy are candidates for secondary therapy. However, this higher risk population is heterogeneous. Many cases do not metastasize even when conservatively managed. Given the limited specificity of pathological features to predict metastasis, newer risk prediction models are needed. We report a validation study of a genomic classifier that predicts metastasis after radical prostatectomy in a high risk population. MATERIALS AND METHODS A case-cohort design was used to sample 1,010 patients after radical prostatectomy at high risk for recurrence who were treated from 2000 to 2006. Patients had preoperative prostate specific antigen greater than 20 ng/ml, Gleason 8 or greater, pT3b or a Mayo Clinic nomogram score of 10 or greater. Patients with metastasis at diagnosis or any prior treatment for prostate cancer were excluded from analysis. A 20% random sampling created a subcohort that included all patients with metastasis. We generated 22-marker genomic classifier scores for 219 patients with available genomic data. ROC and decision curves, competing risk and weighted regression models were used to assess genomic classifier performance. RESULTS The genomic classifier AUC was 0.79 for predicting 5-year metastasis after radical prostatectomy. Decision curves showed that the genomic classifier net benefit exceeded that of clinical only models. The genomic classifier was the predominant predictor of metastasis on multivariable analysis. The cumulative incidence of metastasis 5 years after radical prostatectomy was 2.4%, 6.0% and 22.5% in patients with low (60%), intermediate (21%) and high (19%) genomic classifier scores, respectively (p<0.001). CONCLUSIONS Results indicate that genomic information from the primary tumor can identify patients with adverse pathological features who are most at risk for metastasis and potentially lethal prostate cancer.

Whole-transcriptome profiling of thyroid nodules identifies expression-based signatures for accurate thyroid cancer diagnosis.

PURPOSE Due to the limitations of fine-needle aspiration biopsy (FNAB) cytopathology, many individuals who present with thyroid nodules eventually undergo thyroid surgery to diagnose thyroid cancer. The objective of this study was to use whole-transcriptome profiling to develop and validate a genomic classifier that significantly improves the accuracy of preoperative thyroid cancer diagnosis. MATERIALS AND METHODS Nucleic acids were extracted and amplified for microarray expression analysis on the Affymetrix Human Exon 1.0 ST GeneChips from 1-mm-diameter formalin-fixed and paraffin-embedded thyroid tumor tissue cores. A training group of 60 thyroidectomy specimens (30 cancers and 30 benign lesions) were used to assess differential expression and for subsequent generation of a genomic classifier. The classifier was validated in a blinded fashion on a group of 31 formalin-fixed and paraffin-embedded thyroid FNAB specimens. RESULTS Expression profiles of the 57 thyroidectomy training and 31 FNAB validation specimens that passed a series of quality control steps were analyzed. A genomic classifier composed of 249 markers that corresponded to 154 genes, had an overall validated accuracy of 90.0% in the 31 patient FNAB specimens and had positive and negative predictive values of 100% and 85.7%, respectively. The majority of the identified markers that made up the classifier represented non-protein-encoding RNAs. CONCLUSIONS Whole-transcriptome profiling of thyroid nodule surgical specimens allowed for the development of a genomic classifier that improved the accuracy of preoperative thyroid cancer FNAB diagnosis.

Discovery of a biomarker signature that predicts upgrading or upstaging in patients with low-risk prostate cancer.

37 Background: More than 90% of patients diagnosed with organ-confined prostate cancer (PCa) choose upfront definitive treatment (e.g., radical surgery) even though many are excellent candidates for delayed therapy (i.e., active surveillance [AS]). Therefore, patients may suffer from the adverse effects of treatment without gaining any benefit. Biomarker signatures that predict tumour aggressiveness are promising tools for identification of patients suited for AS. In this study, we use a transcriptome-wide assay to develop a biomarker signature for patients assessed as low risk at diagnosis who are upgraded or upstaged following radical prostatectomy (RP). METHODS Gene expression data of 56 RP samples from the Memorial Sloan Kettering Oncogenome Project (GSE21034) which met the low risk criteria (i.e., biopsy Gleason score (GS) ≤ 6, clinical stage T1 or T2A, and pre-operative PSA (pre-op PSA) ≤ 10 ng/ml) were used to develop the signature. Of these tumors, 31 underwent upgrading or upstaging (defined by pathological GS ≥ 7 or a pathological tumor stage > T3A). In the training set (n = 29) a median fold difference filter (MFD > 1.4) was applied to select features. The top 16 t-test ranked features were modelled with a K-nearest-neighbor (KNN) classifier (k = 3) which predicts upgrading/upstaging events. RESULTS The KNN was applied to the test set (n = 27) and achieved an area under the receiver operating characteristic curve (AUC) of 0.93, significantly better discrimination than pre-op PSA (AUC = 0.52) or tumor stage (AUC = 0.63). Compared to the null models accuracy of 56%, the KNN correctly predicts 81% (p-value < 0.005) of the upgrading/upstaging events. In multivariable analysis with pre-op PSA, tumor stage, and age at diagnosis, the KNN remained the only significant (p < 0.05) factor with an odds ratio of 2.7. CONCLUSIONS A 16 marker signature was identified from RP specimens and shown to accurately segregate true low risk patients from those which transitioned to higher risk. Validation studies of this signature in prospectively designed cohorts of active surveillance candidates are underway to determine if the molecular signature can improve treatment and management decisions for low risk PCa patients.

Validation of a genomic-clinical classifier model for predicting clinical recurrence of patients with localized prostate cancer in a high-risk population.

175 Background: The efficient delivery of adjuvant and salvage therapy after radical prostatectomy in patients with prostate cancer is hampered by a lack of biomarkers to assess the risk of clinically significant recurrence and progression. METHODS Mayo Clinic Radical Prostatectomy Registry (RP) patient specimens were selected from a case-control cohort with 14 years median follow-up for training and initial validation of an expression biomarker genomic classifier (GC). An independent, blinded case-cohort study of high-risk RP subjects was used to validate GC, comparing the performance of GC to a multivariate logistic regression clinical model (CM) and GC combined with clinical variables (genomic-clinical classifier, GCC) for predicting clinical recurrence (defined as positive bone or CT scan within 5 years after biochemical recurrence). The concordance index (c-index) and Cox model were used to evaluate discrimination and estimate the risk of clinical recurrence. RESULTS In the training subset (n=359), both GC and GCC had a c-index of 0.90 whereas CM had a c-index of 0.76. In the internal validation set (n=186), GC and GCC had a c-index of 0.76 and 0.75, while CM had a c-index of 0.69. In an independent high-risk study (n=219), GC and GCC had a c-index of 0.77 and 0.76, while CM had a c-index of 0.68. In subset analysis of Gleason score 7 patients within the high-risk group, GC and GCC showed improved discrimination with c-index of 0.78 and 0.76, respectively compared to 0.70 for CM. In the high-risk group, the risk of recurrence by GC model score quartiles at 5 years after RP was estimated at 1%, 5%, 5% and 18%. CONCLUSIONS The GC model shows improved performance over CM in the prediction of clinical recurrence in a high-risk cohort and in subset analysis of Gleason score 7 patients. The addition of clinical variables to the GC model did not significantly contribute to classifier performance in patients with high-risk features. We are further testing the performance of the GC and GCC models and their usefulness in guiding decision-making (e.g., for the adjuvant therapy setting) in additional studies of prostate cancer clinical risk groups.

Development and validation of a digital Gleason score biomarker signature for risk stratification of patients with prostate cancer.

40 Background: Gleason score (GS) is the most widely used grading system of cell differentiation in prostate cancer and one of the best pathological predictors of disease progression. Patients with high GS (> 7) are the most likely to experience metastasis, whereas patients with low GS (< 7) are expected to have favorable long-term outcomes. In addition to the subjective nature of GS assessment, patients with GS 7 represent a heterogeneous group in terms of patient outcomes. In this study a biomarker signature is developed and validated which could improve the prediction of high risk disease among GS 7 radical prostatectomy (RP) patients. METHODS Patient specimens from the Mayo Clinic RP Registry (n = 764) were used for feature selection and training of a 392-feature K-nearest neighbor (KNN, k = 11) classifier. These 392 genomic features were identified by assessing differential expression between patients with GS < 7 and those with GS > 7, using a Bonferroni adjusted T-Test with a p-value threshold of 0.05. The classifier was subsequently validated in two independent patient cohorts (Memorial Sloan Kettering [MSKCC] and German Cancer Research Center [DKFZ]) and compared to a state of the art biomarker signature (Penney et al. 2011). RESULTS In the MSKCC dataset (GSE21034) our model segregated GS < 7 from GS > 7 patients (n = 56) with an area under the receiver operating characteristic curve (AUC) of 0.97, comparable to the AUC of 0.94 obtained by Penney et al. in their independent validation set (n = 45). Strong performance was observed by our model when discriminating between primary Gleason grade (pGG) 3 and pGG 4 & 5 patients in the MSKCC (n = 130) and DKFZ prostate cancer (GSE29079, n = 47) datasets, achieving AUCs of 0.75 and 0.87, respectively. In the challenging GS 7 subset, our model segregated GS 3 + 4 and 4 + 3 patients in the DKFZ dataset (n = 64) with an AUC of 0.81 outperforming the Penney et al. signature (AUC = 0.60). CONCLUSIONS A biomarker signature was developed which discriminates between low and high GS patients, outperforming a previously reported signature. Further validation of this biomarker signature in additional post RP patients, as well as, pretreatment biopsy specimens is warranted.

Abstract B46: Identification of biomarkers for stratification of Gleason score 7 patients using whole-transcriptome expression profiling

Abstract Motivations: Gleason score is the most prognostic variable in prostate cancer pathology. Patients with high Gleason (8+) are the most likely to experience biochemical recurrence (BCR), clinical progression (CP) and prostate-cancer specific mortality (PCSM), whereas patients with Gleason ≤6 are expected to have excellent long-term outcomes. However, the majority of prostate cancer patients are diagnosed with Gleason score 7, a heterogeneous group in terms of patient outcomes. In the absence of adverse pathological factors (APF, such as preoperative PSA, SM+, ECE and SVI) most Gleason 7 patients will have excellent outcomes. In contrast, Gleason 7 patients with APF are considered at high-risk for recurrence and therefore are candidates for adjuvant therapy. In this study, we have investigated whether a classifier based on differentially expressed genomic features could be used to improve prognostication of Gleason 7 patients, independent of APF. Methods: Two whole-transcriptome (1.4 million feature) oligonucleotide microarray expression datasets from radical prostatectomy specimens were analyzed. The first one (GEO accession: GSE21032) consisted of 125 fresh frozen (FF) samples from the MSKCC Oncogenome Project and the second dataset had 545 formalin-fixed-paraffin- embedded (FFPE) samples from the Mayo Clinic tumor registry. For genomic feature selection, differential expression in the FF dataset between Gleason ≤6 (n=40) and ≥8 (n=15) was assessed with false discovery rate adjusted t-tests (p Results: The KNN classifier segregated the Gleason 7 samples into high and low risk CP groups. The KNN odds ratios were 2.9 (p Conclusions: A KNN classifier developed using differentially expressed genomic features significantly stratified Gleason score 7 patients for several clinically relevant endpoints. Future validation studies will address the utility of this classifier in predicting pathological Gleason score from biopsy specimens, which may be particularly useful for evaluation of biopsy Gleason 6 cases. In these patients, deferring prostatectomy in favour of active surveillance is an appealing treatment option unless the ‘true’ Gleason of the index lesion is in fact a Gleason 7 or higher, where immediate surgery is likely a life-saving procedure. Citation Format: Nicholas Erho, Thomas Sierocinski, Elai Davicioni, Robert B. Jenkins, Mercedeh Gadhessi, Christine Buerki, Ismael A. Vergara, Anamaria Crisan, Zaid Haddad, Benedikt Zimmermann, Anirban Mitra, Timothy J. Triche. Identification of biomarkers for stratification of Gleason score 7 patients using whole-transcriptome expression profiling [abstract]. In: Proceedings of the AACR Special Conference on Advances in Prostate Cancer Research; 2012 Feb 6-9; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2012;72(4 Suppl):Abstract nr B46.