Lung Cancer SPORE

Cores

BIOINFORMATICS/BIOSTATISTICS CORE

Core Co-Leaders: Stephanie Land, PhD and Vanathi Gopalakrishnan, PhD

Core members have participated in the regular meetings of the Pittsburgh Lung Cancer SPORE. Dr. Land serves on the SPORE Tissue Utilization and Prioritization Committee evaluating requests for sample use. Another major contribution of the Core in Year 2 is the development of an Access database for data from the Carinal Registry Study. This database includes patient characteristics and clinical follow-up for patients whose tissues have been banked. Those samples are a rich repository for the experimental studies related to Projects 1 and 2 of the SPORE. The use of the Core by SPORE projects was as follows:

Project 1 — Intersection of Estrogen Receptor Signaling and Epidermal Growth Factor Receptor Signaling in Lung Cancer.

Several studies have been conducted that provide additional pilot information for Project 1. In one study, tumor tissue from Carinal Registry patients was analyzed for ER-alpha, ER-beta, PR, aromatase, and EGFR markers. Statisticians Land and Shuai analyzed the associations between these markers and patient and tumor characteristics. In addition, overall survival and progression free survival were modeled based on marker expression, age at tissue collection, smoking, gender, histology, and disease stage. Analytic techniques included tree-based methods, Cox regression and Fisher’s exact tests. A manuscript is in preparation. Dr. Land also provided sample size calculations for a xenograft experiment for Project 1. Dr. Land is providing support for the dissertation project of Ji Young Song, student of Dr. Weissfeld. That study will be a comparison of ER expression in Carinal Registry Study lung tissue with tissue from non-cancer controls, and an examination of the relevance of ER expression in normal lung tissue to lung cancer outcomes in Carinal patients. Dr. Land also provided a statistical analysis plan and sample size calculations for proposed work by Dr. H. Srinivas. In that project, we will test newly standardized ER-alpha and ER-beta antibodies on paraffin-embedded lung tumor sections of patients, which we have obtained from Lung SPORE tissue bank. We will perform statistical analysis of the associations between ER immunostaining and clinical and histopathological variables.

Project 2 — Cyclin B1 in Immunotherapy, Diagnosis and Prognosis of Lung Cancer

Statisticians Land and Shuai examined laboratory assays of cyclin B1 antibody levels. We estimated the effect of each plate (assay) using linear regression in order to develop a normalization procedure, which is performed by subtracting the relevant plate effect from the value for each test sample. We also compared healthy controls with cancer patients (Carinal Registry), and tested the effect of age on antibody levels. The cyclin-B1 antibody levels were significantly lower among cancer patients (estimated difference in medians 0.053; p=0.0017). There was no significant difference by age (p=0.81). In a third analysis, we analyzed the association of survival time and progression free survival time with cyclin B1 antibody levels in tissue from non-small cell lung cancer patients from the Carinal Registry study. These analyses were performed with Cox proportional hazards regression.

Project 3 — Serum-Based Proteomics for Lung Cancer Detection and Prognosis

Bioinformaticians Gopalakrishnan and Hauskrecht have been integral to the work of Project 3. Our goals for this year were to perform extensive analyses of the preliminary data generated in Project 3 by proteomic profiling of lung cancer in order to be able to (a) validate previous analyses of the same data; and (b) suggest protein identities for putative disease-specific biomarkers. To achieve these goals, we performed multiple analyses on the UPCI and Vanderbilt data that include and extend the machine learning analysis and validation conducted previously. We measured sensitivity, specificity and the achieved classification error across multiple techniques. The putative biomarker discovery phase utilized a rule learning algorithm, RL and two of the datasets used in the analyses.