6/9/2017 Using a polygenic risk score and breast density to predict interval cancers Disclosures Yiwey Shieh, MD UCSF Division of General Internal Medicine I have no potential conflicts of interest to disclose. I am a general internist. International Breast Densitometry & Cancer Risk Assessment Workshop June 9, 2017 Interval cancers Intrinsic subtypes • Interval cancers arise symptomatically between screening Subtype Receptor status Prevalence rounds, following a normal ER+ or PR+ low-grade ER+ Luminal A 30-70% mammogram HER2- ER+ or PR+ high-grade ER+ Luminal B 10-20% • Present at more advanced stage, HER2+ worse survival in some studies ER-/PR- Basal 15-20% ER- HER2- (high-grade) • Associated with dense breasts ER-/PR- HER2 5-10% (masking) HER2+ Normal-like • “High-risk” biology: ER-negative, proliferative ER: estrogen receptor PR: progesterone receptor Kirsch JNCI 2011 HER2: human epidermal growth factor receptor 2 1
6/9/2017 Genetic determinants of Risk stratification with a 77-SNP PRS cancer • Single nucleotide polymorphisms (SNPs) → genetic variants responsible for differences in phenotype • Individual SNP odds ratio = 0.8 to 1.3 • 157 SNPs associated with breast cancer (p < 5 x 10 -8 ), though ~90 published Polygenic risk scores (PRS) represent cumulative effects of multiple SNPs AUROC 0.62 Mavaddat JNCI 2015 Our work: 83-SNP PRS improves Why an interval cancer PRS? on existing risk models • Many SNPs have differential associations with ER+ and ER- cancers - also survival, age of onset - suggests some genetic determination of cancer phenotype • Clinical setting: p = 0.01 • unfitted PRS → when to begin screening, consider prevention Top vs bottom quartile: • next steps (?) OR = 1.7 (95% CI 1.2-2.5) for • “interval cancer PRS” → how often to screen, modality of BCSC model OR = 3.2 (95% CI 2.2-4.7) for screening BCSC-PRS model • risk-stratify women with dense breasts Shieh BCRT 2016 2
6/9/2017 Can PRS predict interval cancers? PRS and tumor characteristics High PRS protective against ER-negativity, high grade, and possibly larger tumor size and lymph node involvement • 77-SNP PRS from Mavaddat et al tested in Swedish cohort • High PRS → lower risk of: • interval cancers, OR 0.91 (95% CI 0.63-1.01) per SD • poor prognostic features: ER-negativity, high grade Holm JCO 2015 Li Annals of Oncology 2015 Holm JCO 2015 Alternate approach Pick SNPs associated with “high risk” disease Methods Use high-risk SNPs to Fit currently known SNPs predict interval breast to high-risk features of cancers in independent breast cancer dataset 3
6/9/2017 Overview of methods The Cancer Genome Atlas Test set: nested • The Cancer Genome Atlas (TCGA): publicly accessible Development set: TCGA study in screening dataset with comprehensive molecular portraits of tumors cohort • 1,094 breast cancers with available data: • mRNA expression (Agilent) • array-based SNP genotypes (Affymetrix 6.0) • whole-exome sequencing • DNA methylation Fit currently known SNPs Use high-risk SNPs to • miRNA sequencing to high-risk features of predict interval breast • clinical outcomes breast cancer cancers Identifying PC’s associated with Risk of recurrence (ROR-P) score “high-risk” features • Based on PAM50, gene 1. Perform principal components (PC) analysis of expression array that classifies gene expression data in TCGA breast cancer into 4 intrinsic subtypes a. PC1-4, orthogonal (rotated) • ROR-P score = model for risk 2. Identify PCs associated with high-risk features like of relapse based on tumor proliferation or ER-negativity subtype correlations + expression of a subset of 11 a. confirm ER-negative PCs using ESR1 genes correlated with proliferation expression b. confirm proliferation PCs using risk of ROR-P = -0.001*Basal + 0.7*Her2 - 0.95*LumA + recurrence-proliferation (ROR-P) score 0.49*LumB + 0.34*Proliferation Nielsen Clinical Cancer Research 2010 4
6/9/2017 Using PC’s to choose SNPs for Calculating the polygenic risk polygenic risk score score 1. Candidate SNPs = genome-wide significant • Step 1: for each SNP, calculate probability of association vs breast cancer or phenotype (ER genotype given disease status, survival, etc) 2. Regress PC vs candidate SNPs • Step 2: calculate LR for SNP a. adjusted for ancestry (principal components of • Step 3: multiply LRs for each SNPs) of x SNPs to obtain final 3. Select SNPs based on “direction” of association with PRS PC (pos or neg beta) & significance (p<0.2) • Step 4: use Bayes theorem 4. Use SNPs to modify existing 83-SNP polygenic risk to modify pretest probability score as predicted by risk model Testing the PRS versus interval Statistical methods cancers • Case-control • California Pacific Medical Center Research Institute (CPMCRI) cohort • Outcome: interval cancer vs controls • ~19,000 women undergoing screening 2004-2011 who • Predictors: PRS, density gave blood samples for research • density adjusted for age, BMI, race/ethnicity • questionnaire data (SFMR), cancer outcomes • Case-case • density measured using BI-RADS • Outcome: interval vs screen-detected cancers • Subset were genotyped (OncoArray): • Same predictors, adjustment as above • 481 cases: 102 interval, 369 screen-detected • Evaluated discrimination with receiver operating characteristic (ROC) curve analysis • 496 controls matched by age, race/ethnicity 5
6/9/2017 Principal components analysis Results SNP selection Principal components PC1 → ER status • 18,321 available transcripts - dropped 5,662 genes with missing data - 12,659 remaining genes positive PC1 → ER neg 6
6/9/2017 PC1 and ESR1 expression PC3 → proliferation/grade ρ = -0.80 negative PC3 → positive PC1 → proliferative ER neg SNP selection PC3 and ROR-P 83 SNPs in naive (unfitted) PRS + 4 hits vs ER- or high-grade (add high-risk SNPs) ρ = -0.64 - 13 hits vs ER+ or low-grade (subtract low-risk SNPs) (retain “neutral” SNPs) 74 SNP PRS fitted to “high risk” cancers 7
6/9/2017 Demographics Controls Interval cancers n = 496 n = 102 Age, median (IQR) 55 (47-64) 53 (45-60) Race/Ethnicity, n(%) Interval cancer prediction White 396 (80.1) 84 (82.3) Black 10 (2) 1 (1) Asian 53 (10.1) 10 (9.8) Hispanic 24 (4.9) 3 (2.9) Mixed 11 (2.2) 4 (3.9) BMI, median (IQR) 23.4 (21.2-26.4) 22.5 (20.9-25.8) Prior biopsy, n(%) 95 (19.2) 33 (32.4) Positive family 89 (17.9) 27 (26.5) history, n(%) BIRADS density ROC curve, BIRADS density interval cancers vs controls interval cancer vs controls OR 4.4 Density* (2.1-9.2) OR 3.0 AUC = 0.66 (1.7-5.4) (95% CI 0.60-0.71) referent OR 0.3 (0.1-1.7) *adjusted for age, BMI, race/ethnicity Adjusted for age, BMI, race/ethnicity 8
6/9/2017 Histogram of 74-SNP PRS PRS comparison interval cancer vs controls Interval cancer vs Case-control control OR per SD OR per SD (95% CI) (95% CI) PRS83 1.39 1.32 (unfitted) (1.20-1.62) (1.06-1.64) PRS74 1.43 1.53 (“high-risk”) (1.23-1.69) (1.21-1.92) ROC curve, density + PRS74 Quartiles of Density/PRS model interval cancer vs controls Interval cancers vs controls Density OR 7.3 AUC 0.66 (3.5-15.4) (95% CI 0.60-0.71) OR 3.8 PRS+density (1.8-8.0) AUC 0.68 OR 1.9 (95% CI 0.62-0.74) (0.9-4.4) p = 0.14 9
6/9/2017 What about interval vs screen- Histogram of 74-SNP PRS detected cancers? interval vs screen-detected Density* AUC = 0.64 (95% CI 0.58-0.70) *adjusted for age, BMI, race/ethnicity PRS comparison Interval cancer Interval cancer Case-control vs screen- vs control (OR, 95% CI) detected (OR, 95% CI) (OR, 95% CI) Discussion PRS83 1.39 1.32 0.96 (unfitted) (1.20-1.62) (1.06-1.64) (0.78-1.19) PRS74 1.43 1.53 1.07 (“high-risk”) (1.23-1.69) (1.21-1.92) (0.88-1.31) 10
6/9/2017 Key findings Explanations (limitations) • Breast density is strongly associated with interval • TCGA and/or CPMC datasets may be cancers underpowered • Able to identify SNPS associated with ER status & • ER-negativity and grade only modestly associated proliferation (per gene expression) in TCGA with interval cancer status in CPMC dataset • Modifying existing PRS according to these SNPs → • TCGA hits for ER-negative don’t replicate in CPMC minimal improvement over density in interval cancer prediction Next steps Acknowledgements • Incorporation of newly discovered ER-negative Donglei Hu Jeff Tice SNPs from OncoArray in Breast Cancer Association Scott Huntsman Steve Cummings Consortium Lin Ma Celine Vachon Charlotte Gard Karla Kerlikowske • Repeat SNP discovery in larger dataset Jessica Leung Elad Ziv • Expand validation dataset Celine Vachon Christopher Scott 11
6/9/2017 Questions? 12
Recommend
More recommend
