High-dimensional regression analysis links magnetic resonance imaging features and protein expression and signaling pathway alterations in breast invasive carcinoma

Michael Lehrer1, Anindya Bhadra2, Sathvik Aithala1, Visweswaran Ravikumar 1, Youyun Zheng 3, Basak Dogan4, Emerlinda Bonaccio5, Elizabeth S. Burnside 6, Elizabeth Morris7, Elizabeth Sutton7, Gary J. Whitman8, Jose Net9, Kathy Brandt10, Marie Ganott11, Margarita Zuley11, Arvind Rao1, TCGA Breast Phenotype Research Group12

1 Department of Bioinformatics and Computational Biology, University of Texas MD Anderson Cancer Center, Houston, TX, USA

2 Department of Statistics, Purdue University, West Lafayette, IN, USA

3 Department of Biostatistics, Emory University, Atlanta, GA, USA

4 Department of Radiology, UT Southwestern, Dallas, TX, USA

5Department of Diagnostic Radiology, Roswell Park Cancer Institute, Buffalo, NY, USA

6Department of Radiology, University of Wisconsin—Madison, Madison, WI, USA

7Department of Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, USA

8Department of Radiology, MD Anderson Cancer Center, Houston, TX, USA

9Department of Radiology, University of Miami Health System, Miami, FL, USA

10Department of Radiology, Mayo Clinic, Rochester, MN, USA

11Department of Radiology, University of Pittsburgh, Pittsburgh, PA, USA


Correspondence to:

Arvind Rao, email: aruppore@mdanderson.org

Keywords: breast invasive carcinoma; MRI; protein expression; signaling pathway analysis; TCGA

Received: October 16, 2017 Accepted: December 15, 2017 Published: February 26, 2018


Background: Imaging features derived from MRI scans can be used for not only breast cancer detection and measuring disease extent, but can also determine gene expression and patient outcomes. The relationships between imaging features, gene/protein expression, and response to therapy hold potential to guide personalized medicine. We aim to characterize the relationship between radiologist-annotated tumor phenotypic features (based on MRI) and the underlying biological processes (based on proteomic profiling) in the tumor. Methods: Multiple-response regression of the image-derived, radiologist-scored features with reverse-phase protein array expression levels generated association coefficients for each combination of image-feature and protein in the RPPA dataset. Significantly-associated proteins for features were analyzed with Ingenuity Pathway Analysis software. Hierarchical clustering of the results of the pathway analysis determined which features were most strongly correlated with pathway activity and cellular functions. Results: Each of the twenty-nine imaging features was found to have a set of significantly correlated molecules, associated biological functions, and pathways. Conclusions: We interrogated the pathway alterations represented by the protein expression associated with each imaging feature. Our study demonstrates the relationships between biological processes (via proteomic measurements) and MRI features within breast tumors.

PII: 397