Abstract 3228: Using paired tissue and serum samples to characterize human lung cancer metabolomics with 1H HRMAS MRS

• Published in: Cancer research (Chicago, Ill.) Ill.), 2013-04, Vol.73 (8_Supplement), p.3228-3228
• Main Authors: Huang, Hailiang, Decelle, Emily A., Berker, Yannick, Schuler, Andreas, Dittman, Isabel, Su, Li, Mark, Eugene J., Daly, Mark J., Christiani, David C., Cheng, Leo L.
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2013-3228

Purpose: We used high resolution magic angle spinning proton MRS to identify metabolomic profiles of lung cancer tissue and serum samples. Previous studies have reported that profiles for lung cancer tissue may be predictive of the profiles of matched serum samples. We further searched correlations of metabolomic data with traditional histopathology from the same tissue samples, and identified serum lung cancer metabolomic markers based on matched tissue analysis. Methods: Paired tissue and serum samples from 107 patients of adenocarcinoma (AC) and squamous cell carcinoma (SCC), and 29 serum samples from control subjects without lung disease, were analyzed. MR experiments were carried out on a Bruker AVANCE spectrometer operating at 600 MHz (14.1T) and pre-cooled to 4°C. A 4mm zirconia rotor was used with inserts to create a 10μl sample space, and D2O was added for 2H field locking. Rotor spinning rate was regulated and verified by measuring the inter-SSB distances from spectra with an accuracy of 1.0Hz. A repetition time of 5s and 128 transients were used to acquire each spectrum. Spectra were collected with a spinning rate of 3600Hz, with a rotor synchronized CPMG filter to reduce broad resonances. Spectra were analyzed by an in-house MatLab based program. After spectroscopy, tissue samples were fixed in formalin, embedded in paraffin, cut into sets of 5μm sections at 100μm intervals, and stained with hematoxylin and eosin. Volume percentages of histological features (cancer, stroma, necrosis, lymphatic structures, and cartilage) were quantified by a pathologist. Results: We used Lasso, a feature selection method using L1 regularized linear regression, to select a subset of peaks from the spectral results to build a linear model to predict AC/SCC readings. A 2-peak model (peaks at 2.14-2.10 and 0.89-0.89 ppm) has been identified as the most optimal model, and the predicted probability from this model agrees well with the AC/SCC readings. We used the same procedure to build predictive models for quantitative histopathology readings from serum and tissue samples. Out of the four histopathology readings (%Cancer, %Fibrosis/ Inflammation, %Necrosis and %Cartilage), we were able to build a model for %Fibrosis/Inflammation (p =1E-5) using 3 peaks from serum samples (peaks at 2.09-2.03, 1.33-1.32 and 3.27-3.24), and a model for %cancer cells (p =0.002) using one peak from tissue samples (peak at 3.73-3.71). Further data analyses are still underway. Conclusions: