1. 4. DPKK Workshop in Bonn/Königswinter 1.-2.12.2006 Quantitative multi-gene expression analyses on paired prostate tissue samples from radical prostatectomies and on artificial prostate biopsies Susanne Füssel & Susanne Unversucht Axel Meye, Michael Haase, Andrea Lohse, Silke Tomasetti, Michael Fröhner, Uta Schmidt, Rainer Koch, Gustavo Baretton, Manfred. P. Wirth Dept. of Urology & Institute of Medical Informatics and Biometry & Institute of Pathology Technical University of Dresden

2. main problem: early identification of aggressive PCa for therapeutic decisionsmain problem: early identification of aggressive PCa for therapeutic decisions need for new additional PCa-markers to improve diagnostic and prognostic powerneed for new additional PCa-markers to improve diagnostic and prognostic power quantification of transcript markers as promising toolquantification of transcript markers as promising tool expression signatures more reliable than single markersexpression signatures more reliable than single markers Objective

3. establishment of standardized QPCR-assaysestablishment of standardized QPCR-assays 1. study: 9 PCa-related genes + 4 housekeeping genes1. study: 9 PCa-related genes + 4 housekeeping genes 2. study: 4 new PCa-related genes, TBP as reference gene2. study: 4 new PCa-related genes, TBP as reference gene 169 paired tissue samples (Tu + Tf) from RPE explants169 paired tissue samples (Tu + Tf) from RPE explants evaluation of single & combined markers (ROC-analyses)evaluation of single & combined markers (ROC-analyses) mathematical models for PCa-specific transcript signaturesmathematical models for PCa-specific transcript signatures aim: prediction of PCa-presence and tumor extension using minimal tissue specimens (prostate biopsies)aim: prediction of PCa-presence and tumor extension using minimal tissue specimens (prostate biopsies) Material & Methods

4. Evaluation of single markers: overexpression in PCa? PCA3 (=DD3), AMACR, PSGR, hepsin, TRPM8 & PSMA  most promising PCa transcript markers

5. 1. Study: optimized 4-gene-model for PCa-prediction: EZH2 + PCA3 + prostein + TRPM8 1- Specificity AUC = 0.893 (95% CI 0.756... 1.000) ROC-analysis of the 4-gene-combination probability (p) of PCa presence in the analyzed tissue samples (Tf and Tu) median p Tu 0.81 Tf 0.21 predicted probability of tumor classification of relative expression levels of these 4 genes according optimized cut-offs  logit-value for each tissue sample (Tu and Tf)classification of relative expression levels of these 4 genes according optimized cut-offs  logit-value for each tissue sample (Tu and Tf) logit-model 1: p = exp(logit)/[1+exp(logit)]logit-model 1: p = exp(logit)/[1+exp(logit)] correctly predicted: with p  0.7 for Tu : 70 % of Tu-sampleswith p  0.7 for Tu : 70 % of Tu-samples with p  0.3 for Tf : 73 % of Tf-sampleswith p  0.3 for Tf : 73 % of Tf-samples sensitivity 79.3% & specificity 84.0%sensitivity 79.3% & specificity 84.0%

6. 2. Study: optimized 8-gene-model for PCa-prediction: AMACR + AR + EZH2 + hepsin + PCA3 + PDEF + prostein + TRPM8 using log-transformed relative expression levelsof these 8 genes as continuous values  logit-value for each tissue sample (Tu and Tf)using log-transformed relative expression levelsof these 8 genes as continuous values  logit-value for each tissue sample (Tu and Tf) 2. logit-model: p = exp(logit)/[1+exp(logit)]2. logit-model: p = exp(logit)/[1+exp(logit)] 1- Specificity AUC = 0.94 (95% CI 0.79... 1.00) predicted probability of tumor ROC-analysis of the 8-gene-combination probability (p) of PCa presence in the analyzed tissue samples (Tf and Tu) median p Tu 0.93 Tf 0.07 correctly predicted: with p  0.7 for Tu : 78 % of Tu-sampleswith p  0.7 for Tu : 78 % of Tu-samples with p  0.3 for Tf : 78 % of Tf-sampleswith p  0.3 for Tf : 78 % of Tf-samples sensitivity 89.3% & specificity 86.4%sensitivity 89.3% & specificity 86.4%

7. Dependence of marker expression on tumor stage: Discrimination between of organ-confined disease (OCD) and non- organ-confined disease (NOCD) for therapeutic decision? comparison only of Tu-samples of OCD vs. NOCD orcomparison only of Tu-samples of OCD vs. NOCD or comparison of TF-samples vs. Tu-samples of OCD vs. Tu-samples NOCDcomparison of TF-samples vs. Tu-samples of OCD vs. Tu-samples NOCD  mathematical models for OCD-prediction in process

8. translation of the techniques to prostate biopsiestranslation of the techniques to prostate biopsies  additional diagnostic tools for better PCa-prediction? correct prediction of tumor stage & aggressivenesscorrect prediction of tumor stage & aggressiveness  RPE or not, adjuvant hormone therapy or CT or not correlation of transcript signatures with outcome?correlation of transcript signatures with outcome?  follow-up needed for prognostic purposes detection of PCa-specific transcripts in urine samplesdetection of PCa-specific transcripts in urine samples  non-invasive tumor detection? Outlook

9. Aim: transfer of techniques/ statistical models to artificial prostate biopsies from RPE explantstransfer of techniques/ statistical models to artificial prostate biopsies from RPE explants  additional diagnostic tools on minimal prostate tissue samples tissue samples  11 selected PCa-related genes and TBP (reference)  11 selected PCa-related genes and TBP (reference)  first results of application and validation of two multi-gene-models for PCa prediction multi-gene-models for PCa prediction Quantitative multi-gene expression analyses on artificial prostate needle core biopsies from radical prostatectomies

10. Artificial prostate needle core biopsies from radical prostatectomies Material & methods: artificial biopsies (11 patients): Tf/Tu from one RPE explantartificial biopsies (11 patients): Tf/Tu from one RPE explant snap-frozen in liquid nitrogensnap-frozen in liquid nitrogen  cryo-cuttings for RNA-isolation / pathological survey  cryo-cuttings for RNA-isolation / pathological survey H&E-stained cuttings (PCa-patient: pT2a, pN0, pMx Gleason Score: 7 [3+4]) Tu-prostate tissue sample Tf-prostate tissue sample

11. Artificial prostate needle core biopsies from radical prostatectomies Patient`s cohort: 11 patients with a primary PCa age: 51 to 71 years (median 66 years) serum PSA levels: 1.29 to 24.32 ng/ml (median 6.9 ng/ml) Histopathological examinations: (according to the UICC system) 7 patients (64%) with organ-confined disease (OCD; pT2) 4 patients (36%) with non organ-confined disease (NOCD; pT3/ pT4) Tumor grading: 2 patients with low grade (GS 2 to 6) 8 patients with intermediate grade (GS 7) and 1 patient with high grade (GS 8 to 10)

12. Artificial prostate needle core biopsies from radical prostatectomies relative expression levels [zmol gene/ zmol TBP] (n = 38 samples) Transcript marker name malignant (Tu) n=26 median non-malignant (Tf) n=12 median P-values (unpaired t-test) median over expression (Tu vs. Tf) LNCaP (control) mean AMACR PCA3 PSMA 2,104 (25.4 to 4,800) 36.45 (5.4 to 166.3) 25.87 (1.7 to 221.5) 91.65 (5.4 to 640.2) 1.67 (0.1 to 34.4) 2.49 (0 to 72.6) <0.001 0.001 0.062 22.96 21.83 10.39 25.83 0.19 24.83 PSGR TRPM8 EZH2 hepsin PDEF * PSA 47.67 (2.2 to 222.9) 31.71 (6.8 to 218.1) 0.80 (0.1 to 1.807) 0.38 (0.2 to 1.080) 34.13 (1.8 to 136.1) 174.36 (26.8 to 1,395) 8.80 (0.2 to 313.4) 6.95 (0.1 to 58.7) 0.17 (0 to 1.222) 0.12 (0 to 0.80) 14.34 (0.2 to 63.2) 78.18 (0.2 to 737.0) 0.611 0.017 0.004 0.002 0.076 0.207 5.41 4.56 4.71 3.16 2.38 2.23 0.02 1.99 4.48 0.05 3.39 5.38 prostein AndrRec * 8.74 (0.9 to 47.0) 14.52 (4.3 to 31.8) 6.99 (0 to 45.3) 11.77 (0.5 to 18.7) 0.502 0.030 1.25 1.23 1.54 14.23 * n (Tu-specimens) = 25; n (Tf-specimens) = 10

13. Artificial prostate needle core biopsies from radical prostatectomies Validation of two multi-gene models: 4-gene model (EZH2, TRPM8, PCA3, prostein) 8-gene model (PDEF, AndrRec, EZH2, PCA3, hepsin, AMACR, prostein, TRPM8) Tu-prostate biopsies (n = 26) PCa-prediction: 77 % (20 biopsies) PCa-prediction: 100 % (26 biopsies) Tf-prostate biopsies (n = 14) „false postive“: 43 % (6 biopsies) „false postive“: 57 % (8 biopsies)