G Mustacchi 1, F Zanconati 2, D Bonifacio 2, L Morandi 3, MP Sormani 4, A. Gennari 5, P Bruzzi 4 1: Centro Oncolgico University of Trieste 2: Inst of Pathology, University of Trieste 3: USL Bologna, Alphagenics Trieste 4: National Cancer Institute Genova 5: Ospedali Galliera Genova A new 5-gene signature predictive of risk of relapse in early breast cancer. J Clin Oncol 31, 2013 (suppl; abstr 546) Presented as Poster

Background: The aim of this study was look for the "core-genes" of published Signatures in order to find a simpler one Methods: To select candidate genes we used data of NCBI Gene Expression Omnibus ( including 408 breast cancer cases. Raw intensity data of Affymetrix HU133A and HU133B arrays of the two datasets (GSE1456 and GSE3494) were preprocessed using R/Bioconductor and the supercomputer Michelangelo ( The candidate genes were selected from the “70-gene signature” (van 't Veer, Nature 2002), the “21-gene recurrence score” (Paik, NEJM 2004), the “two-gene–ratio model” (Ma, Cancer Cell 2004) and the “15-gene Insuline Resistence” signature (Gennari, JCO, Vol 25, No 18S, 2007: 10597), for a total of 98 genes.The 20 mRNA more significantly related to DFS were evaluated by quantitative reverse transcriptase PCR on 261 consecutive breast cancer cases, from paraffin embedded sections, split into a training (n 137) and a validation set (n 124). Results: The signature was developed on the training set and a multivariate stepwise Cox analysis selected 5 genes independently associated with DFS: FGF18 (HR=1.13, p=0.05), BCL2 (HR=0.57, p=0.001), PRC1 (HR=1.51, p=0.001), MMP9 (HR=1.11, p=0.08), SERF1a (HR=0.83, p=0.007). These 5 genes were combined into a linear score weighted according to the coefficients of the Cox model, (0.125 FGF BCL PRC MMP SERF1A). The linear score was highly associated with DFS (HR=2.7, 95%CI= , p<0.001). The signature was then evaluated on the validation set assessing the discrimination ability by a Kaplan Meier analysis, using the same cut offs classifying patients at low, medium or high risk of relapse as defined on the training set. The score resulted highly associated with DFS also in the validation set (p<0.001). Conclusions: Overexpression of BCL2 and SERF1A are related to a higher probability of relapse; overexpression of FGF18, PRC1 and MMP9 to a higher probability of survival without recurrence. The signature has a good discriminating ability and a further clinical validation is planned.

Creation of a new Signature SignatureAuthorNr of genes Mammaprint van 't Veer LJ, Dai H, van de Vijver MJ. Nature Jan 31;415(6871) Oncotype DX Paik S, Shak S, Tang G et al.. N Engl J Med ;351(27): gene-ratio model Ma XJ, Wang Z, Ryan PD, et al Cancer Cell 2004;5: Insuline resistance signature Gennari A, Sormani MP, Pronzato P et al ASCO meeting 2007, Abs TOTAL Nr of selected genes Excluding the 10 present in more of the above 98 Initial candidate gene selection: from NCBI Gene Expression Omnibus ( 408 breast cancer patients) we picked up all the genes included in the 4 following signatureshttp:// Mustacchi, Zanconati, Morandi, Sormani, Gennari, Bruzzi 48 genes with p < 0.01 to DFS 20 clusters with similar profile The most representative gene (Cox Regr. Model) 20 final genes

The final 20 selected genes SymbolGrouppHRlogPlogHR IGFBP7Ins Res0, , , ,41387 FGF18Amsterdam0, , , ,40553 IGFBP6Ins Res2,22E-060, , ,39806 TGFB3Amsterdam0,000860, , ,23141 IGF1Ins Res1,03E-060, , ,22086 TNFSF10Ins Res0, , , ,19505 IRS1Ins Res0, , , ,19238 IL6STIns Res2,77E-050,828164, ,18855 MS4A7Amsterdam0, , , ,16291 BCL2Paik0,003310, ,4802-0,1586 MMP9Amsterdam0, , , , PRC1Amsterdam7,17E-101, , , PITRM1Amsterdam0, , , , FBXO31Amsterdam0, , , , GPR180Amsterdam0, , , , ORC6LAmsterdam0, , , , SERF1AAmsterdam0, , , , AYTL2Amsterdam0, , , , OSOX2Amsterdam0, , , , IDEIns Res0, , , ,51816

Training and Validation sets 317 consecutive early breast cancer patients from Trieste ( ) with full information and at least 5 years of follow up (median 7.5 years) 262 eligible for PcR The biological samples dataset were split by a random procedure into a training and a validation set (ratio 1:1) Training Set n = 138 used for gene signature development Validation set n = 124 accessed once and only for estimating the prediction accuracy of the signature.

Selected Signatures 11 genes ( less “conservative” but more discriminant) 5 genes ( more “conservative” but less discriminant) 11 genes 5 genes FGF18Amsterdam BCL2Paik IDEIns Res OSOX2Amsterdam PITRM1Amsterdam PRC1Amsterdam MMP9Amsterdam IGFBP6Ins Res IRS1Ins Res TNFSF10Ins Res SERF1AAmsterdam FGF18Amsterdam BCL2Paik MMP9Amsterdam PRC1Amsterdam SERF1AAmsterdam

11 genes signature Probability of 5 years relapse Training set HR medium vs low risk = 2.7 (0.7-10) HR high vs low risk = 8.9 (2.7-30) Validation set p = : low risk 2: medium risk 3: high risk p <0.001 DFS (days)

5 genes signature Probability of 5 years relapse HR medium vs low risk = 4.9 (95% CI: ) HR high vs low risk = 12 (95% CI: ) Training setValidation set DFS (days)

Our Signature Overexpression of BCL2 and Serf1a : Poor Prognosis FGF18, PRC1 and MMP9 : Good Prognosis

Conclusions about Gene Signatures in Breast Cancer  Still Matter of Study and Research  “Natural” Risk of BC is “Low-Intermediate-High”  Paraffin embedded, formalin fixed tissue tests are more “user friendly”  Does Intermediate Risk of the Signature match with San Gallen “Gray Zone” ?  The price of a Test should be reasonnable  Ki67 and other biological patterns could assist in risk assessment