Cherbavaz, DB

Abstract BACKGROUND: Unique genes and pathways were identified for prognosis on tamoxifen (T, 5 yrs) and prediction on CAF-T vs T in S8814 using whole transcriptome RNA-Seq from archival FFPE tissue. (Albain, et al; Cherbavaz, et al; SABCS 2015) Discovery was robust for early DFS events but sparse for late. The aims of this new analysis were to 1) utilize a new endpoint BCSE for gene discovery of late events, prognosis and prediction and 2) add intronic counts to the previous exonic results to define whole genes impacting on late BCSE. METHODS: Charts of patients (pts) on CAF-T (212) vs T (142) were reviewed to define the BCSE endpoint (local/regional, contralateral, distant). Deaths without BC were treated as competing risks. BCSE models (including metagenes) of late prognosis and prediction used cumulative incidence functions. Consolidated intronic regions counts within genes were added to exonic regions counts. Using these “whole gene” (WG) counts, association of gene expression with time to BCSE was assessed by Cox regression. A multiple WG score (MWGS) for BCSE prognosis beyond 5 yrs (to 12.5 yrs) was constructed and evaluated for 1-3 and 4+ node (N) groups. False discovery rate was controlled at 10%. RESULTS: More exons and WG were discovered for prognosis on T alone over 12.5 yrs with the BCSE endpoint than DFS. For prognosis of late BCSE after 5 yrs, more genes were discovered using WG (n=111) than by exons (n=9). There were significantly fewer genes for late BCSE on CAF-T (8, WG; 0, exons). The functions of WG prognostic for late BCSE were: cell cycle/proliferation-26 genes, chromosome segregation/mitotic spindle-22, DNA repair/maintenance-10, transcriptional/translational control-5, cell adhesion/migration-4, immune-3, diverse/unknown-32 and growth factor/hormone receptor signaling-9 (this group was only found by WGs, not exons). Of these 111 WG, a MWGS prognostic for late BCSE on T used 57 previously discovered genes pre-specified for this analysis. Probability of BCSE beyond 5 yrs for low vs high MWGS was 8% vs 21% in N1-3+ and 17% vs 42% in N4+. Late prognosis on T differed by low vs high risk defined in a metagene model: cumulative BCSE at year 10 was 0% vs 47% (low vs high risk, p=0.001). Prediction of 10-yr incidence of BCSE varied by risk level by treatment in a metagene model: low risk- CAF-T=47%, T=0% (p=0.045); high risk- CAF-T=35%, T=45% (p=0.027). CONCLUSIONS: Gene discovery for prognosis of late BCSE is enhanced with a novel WG transcriptome expression approach. Use of chemotherapy (CT) before T significantly attenuated gene discovery, so that molecular tools for decisions on extending endocrine therapy (ET) may not be reliable in a setting of prior CT. Some pts on ET for 5 yrs may not require either longer ET or CT, given a N+ cohort was defined with no BCSE observed over 12.5 yrs. For prediction of CT benefit, CAF-T appeared to be inferior to T in a low risk metagene model for BCSE. In sum, these results add more evidence that ET alone may be sufficient (perhaps better) in select N+ settings. Validation in SWOG S1007 (RxPONDER) is planned. SUPPORT: NCI CA180888, 180819, 180821, 180820, 180863; in part, Genomic Health, Inc. Citation Format: Albain KS, Crager MR, Barlow WE, Baehner FL, Bergamaschi A, Rae JM, Ravdin PM, Tripathy D, Gralow JR, Livingston RB, Osborne CK, Ingle JN, Pritchard KI, Davidson NE, Carey LA, Cherbavaz DB, Sing AP, Shak S, Hortobagyi GN, Hayes DF. Discovery of molecular predictors of late breast cancer specific events (BCSE) in ER+, node+ breast cancer – new transcriptome expression whole gene analysis of the phase III adjuvant trial SWOG S8814 [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr PD7-07.

Abstract BACKGROUND: In SWOG-8814A, pts with ER+ node+ breast cancer and low 21 gene recurrence scores (RS) had good prognosis and no CAF benefit, but high RS predicted longer survival from CAF followed by T (CAF-T) vs T (Albain, Lancet Oncol 2010). The aims of SWOG-8814B were to identify novel genes and networks for 1) prognosis of early and late relapse and 2) prediction of CAF benefit, using whole transcriptome expression analysis with next generation RNA sequencing (NGS). METHODS: Stored RNA previously extracted for SWOG-8814A (T, CAF-T arms; T, 5 yrs) was analyzed for RNA/library yield (see companion abstract Cherbavaz et al. for methods). Genes were sequenced and expression of mRNA species was related to disease-free survival (DFS) using Cox proportional hazards. Discovery analyses controlled false discovery rate (FDR) at 10%. Genes were identified for prognosis on T and prediction on CAF-T vs T. Networks of genes/pathways were explored. Early (0-5 yrs) and late (5-13+ yrs) time periods were studied. Gene Ontology, Cytoscape, pathway and hierarchical clustering were used for functional gene and metagene analyses. RESULTS: Of 367 samples, 354 (96%; 142 T, 212 CAF-T; 141 DFS events) had sufficient RNA/library yield, with 20,101 genes sequenced. For prognosis on T, there were 2327 and 568 genes discovered in early and all-yrs follow-up, with only 9 genes prognostic after 5 yrs. Prognosis analyses for residual risk after CAF-T were uninformative. Functional mapping found that genes prognostic for worse DFS were enriched for proliferation (G2M, M-phase), cellular metabolism, DNA repair, stress response and EMT; whereas, those with better DFS involved transcription regulation/repression via zinc finger proteins. Hierarchical clustering (T arm) found significant DFS prognostic metagene signatures for ER-related genes, immune response, ECM/stroma, chromatin remodeling-transcription factor activity and TGFb pathway. All varied for early vs late DFS events. For example, low ER/high stroma expression signatures correlated with high proliferation gene expression and were strongly associated with early events (standardized [st] HR 2.94, p<0.001). Late recurrence was associated with high proliferation, both individually (stHR 1.51, p=.035) and in combination with higher ER expression (stHR 1.51, p=0.09). Fifteen genes predicted CAF benefit (9 better DFS, 6 worse), or 129 genes if FDR relaxed to 20%. Cluster analysis for CAF prediction is ongoing. CONCLUSIONS: Unique genes, clusters and pathways were identified by NGS of archival material in ER+ N+ breast cancer, including previously unreported signatures. While ER, stroma and proliferation-related signatures were associated with early prognosis, proliferation best predicted worse DFS after 5 yrs. NGS of the primary tumor is most informative for early events in pts with only 5 years of T, with few genes selecting only for late relapse. If validated, these signatures may identify pts with excellent DFS despite positive nodes for endocrine therapy alone as well as others for whom chemotherapy and/or biologics are also required . SUPPORT: NCI CA 180888, 180819, 180821, 180820, 180863; in part, Genomic Health, Inc. Citation Format: Albain KS, Crager MR, Barlow WE, Baehner FL, Bergamaschi A, Rae JM, Ravdin PM, Tripathy D, Gralow JR, Livingston RB, Osborne CK, Ingle JN, Pritchard KI, Davidson NE, Carey LA, Cherbavaz DB, Sing AP, Shak S, Hortobagyi GN, Hayes DF. Molecular predictors of outcome on adjuvant CAF plus tamoxifen (T) vs T in postmenopausal patients (pts) with ER+, node+ breast cancer – Transcriptome expression analysis of the phase III trial SWOG-8814. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr S3-02.

Abstract BACKGROUND: We previously reported that low 21 gene Recurrence Scores (RS) identify patients with ER-positive, lymph node-positive breast cancer who may not benefit from anthracycline-based adjuvant chemotherapy added to tamoxifen (SWOG-8814A NCI correlative science study; Albain et al. Lancet Oncol 2010). New exploratory and comprehensive quantitative analyses now permit whole transcriptome NGS on residual RNA extracted from FFPE blocks 12-18 years post-fixation. Herein, we report methodology details and feasibility results (see companion abstract, Albain et al., for clinical outcomes correlations). METHODS and RESULTS: Sequencing was carried out in Illumina HiSeq 2000 instruments, yielding 4.2 trillion data points. Messenger RNA expression was quantified using 3rd quartile normalization. Both Library (RNA) and Sequencing Standards showed high quality coverage as measured by median uniquely mapped reads over a 13 month window (168M and 182M, respectively, including duplicate reads). The median absolute deviation (MAD) of the relative log expression (RLE) of mapped reads for the Library and Sequencing Standards was 0.22 and 0.05, respectively. The Library Standard variation was greater than the Sequencing Standard, as library preparation was manual. Of 360 patient samples with sufficient RNA (≥ 100 ng total RNA), 354 (98.3%) were successfully sequenced and included in the final analysis data set. Average library yield was 39 ng/μL. Only 5 libraries failed yield requirements and one library failed expression quality metrics. The median insert length was 120 bp with the first and third quartiles 93 and 152 bp, respectively. After removal of duplicate reads, 82% of reads were uniquely mapped, and the median library size was 8.95M (number of unique mapped reads). Sequences with counts <10 for all 354 patients were excluded. The medians of the 1st, 2nd and 3rd quartiles for exons mapped to the RefSeq database were 20, 40 and 78 counts, and for introns, 72, 134 and 244 counts, respectively. The majority of exonic (86.7%) and intronic (95%) sequences were mapped. There were 20,101 RefSeq mapped exons with counts ≥10. Of these exons, 988 passed additional filtering criteria and were subjected to hierarchical clustering, with Gene Ontology and pathway analysis performed on selected gene expression patterns (for results, see companion abstract of Albain et al.). CONCLUSIONS: High quality whole transcriptome NGS is feasible from decades-old clinical trial FFPE specimens that have not been stored in any special fashion. Controlled laboratory, bioinformatics and biostatistics methods, with inclusion of appropriate process controls, ensure robustness and reliability of the NGS process. This in turn results in the discovery and validation of biologically and clinically relevant variations from prior landmark clinical trials. SUPPORT: NCI CA 180888, CA180819, CA180821, CA180820, CA180863; in part, Genomic Health, Inc. Citation Format: Cherbavaz DB, Hayes DF, Qu K, Crager MR, Barlow WR, Goddard AD, Beasley EM, Jeong J, Collin F, Liu M-L, Rae JM, Ravdin PM, Tripathy D, Gralow JR, Livingston RB, Osborne CK, Ingle JN, Pritchard KI, Davidson NE, Carey LA, Sing AP, Baehner FL, Hortobagyi GN, Shak S, Albain KS. Successful whole transcriptome analysis of 25-year-old breast tumor samples from the phase III trial SWOG-8814 by next generation sequencing (NGS): Standardized analytical methods for exploratory and validation studies. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P5-07-01.