Department of Radiology and Imaging Sciences Division of Neuroradiology University School of Medicine

Associations Between MR Imaging and Genomic Features of Glioblastomas The Cancer Genome Atlas (TCGA) Glioma Research Group MRI Characterization Study

Associations Between MR Imaging and Genomic Features of Glioblastomas Emory University In Silico Brain Tumor Research Center (NCI) –Department of Radiology and Imaging Sciences –Department of Pathology –Department of Biomedical Informatics Thomas Jefferson University, Department of Radiology University of Virginia, Department of Radiology and Medical Imaging NCI / NIH SAIC-Frederick, Inc.

Disclosures None

Background – TCGA Joint effort of the National Cancer Institute (NCI) and the National Human Genome Research Institute (NHGRI) Publicly available database for multiple types of tumors from different anatomic sites, including glioblastomas –Pathology specimens –Imaging Purpose: to advance the understanding of the molecular basis of cancer through the application of genome analysis technologies, histopathology, and imaging

Purpose To identify imaging features of primary glioblastomas that may predict genomic features, including mutation status, gene expression, and DNA copy number variations

Methods Multireader assessment of pre-op MR images –75 glioblastoma patients –VASARI imaging feature set At least 3 CAQ neuroradiologists (from a panel of 6) independently reviewed each MRI data set

Methods VASARI feature set – –26 imaging features –For the current analysis, we focused on the following parameters: Tumor dimensions (axial plane) Percent enhancing tumor Percent nonenhancing tumor Percent necrosis Enhancing pial involvement Definition of nonenhancing margin

Other VASARI Features Tumor location Side Eloquent brain Avidity of enhancement Cysts Thickness of enhancing margin Definition of enhancing margin Hemorrhage Diffusion Pial invasion Midline crossing –Enhancing tumor –Nonenhancing tumor Distribution –Multifocal –Multicentric –Gliomatosis T1/FLAIR ratio –Expansive –Mixed –Infiltrative Ependymal invasion Cortical Involvement Deep WM invasion Satellites Calvarial remodeling

Lesion Dimensions Maximum and perpendicular dimensions on the basis of the T2w/FLAIR images

Tumor Proportions Assume the tumor consists of: –Enhancing tumor –Nonenhancing tumor –Necrosis –Edema Percent volume of each component rated on an ordinal scale –0, 1-5%, 6-33%, 34-67%, 68-95%, 95-99%, 100%

Proportion Enhancing Tumor 1-5% 68-95%

Proportion Nonenhancing Tumor 68-95% 6-33%

Proportion Edema 0% 34-67%

Proportion Necrosis 34-67% 1-5%

Methods Each parameter reduced to a single score –Majority rule –Random selection in case of a tie Statistical tests –Linear correlation –Fisher exact test –Student’s t-test

Methods - Genomic Features Genomic data from TCGA’s publicly available information (TCGA, Nature 455:1061, 2008) Mutation status (presence versus absence of a gene mutation) –TP53, PTEN, EGFR, NF1, and IDH1 Quantitative measurements of gene expression –TP53, PTEN, EGFR, NF1, IDH1, and TGFB2 DNA copy number variations –high-level EGFR gene amplification –high-level PDGFRA amplification –homozygous deletions of CDKN2A –deletions of NF1

Results TP53 mutant tumors had a smaller mean tumor size (p=0.002), measured as the maximum tumor dimension on the T2-weighted and/or FLAIR images EGFR mutant tumors were significantly larger than TP53 mutant tumors (p=0.0005)

Results Association between enhancing tumor (>5% proportion) and high-level EGFR amplification (p=0.01) Enhancing pial involvement was associated with high-level EGFR amplification (p=0.02) Enhancing pial involvement was also associated with the presence of a CDKN2A homozygous deletion (p=0.03) Association between the presence of CDKN2A homozygous deletion and an ill-defined nonenhancing tumor margin (p=0.007) Having >5% proportion of necrosis may be associated with high-level EGFR amplification (p=0.06) No significant associations with PDGRFA amplification or NF1 deletions There were trends NF1 deletion was associated with the presence of edema (>5% proportion) (p=0.06) and with the presence of nonenhancing tumor (>5% proportion) (p=0.06)

Results Positive correlation between tumor size and EGFR expression (r=0.30, p=0.02) Negative correlation between tumor size and IDH1 expression (r=0.33, p=0.006) Mean IDH1 expression in tumors without a significant proportion of enhancement (≤5% of the tumor volume) was less than the mean in tumors with >5% enhancement (p=0.049)

Conclusions Multiple statistically significant associations between imaging and genomic features in glioblastomas were identified In particular, EGFR mutant tumors were significantly larger than TP53 mutant tumors, and were more likely to demonstrate pial involvement CDKN2A homozygous deletion was associated with an ill-defined, nonenhancing tumor margin and enhancing pial involvement

Future Work Analysis of additional images and genomic data from TCGA Quantitative/volumetric image analysis Perfusion and DTI data Upcoming RTOG study using 5-ALA for resection guidance plans to enroll 600 glioblastomas from centers

THANK YOU!

TCGA Glioma Research Group Holder, C.A. 1 Hwang, S.N. 1 Clifford, R.J. 2 Huang, E. 2 Raghavan, P. 4 Nicolasjilwan, M. 4 Wintermark, M. 4 Hammoud, D. 3 Gutman, D.A 1 Cooper, L. 1 Moreno, C. 1 Freymann, J. 5 Kirby, J. 5 Krishnan, A. 1 Dehkharghani, S. 1 Jaffe, C.C. 3 Saltz, J.H. 1 Brat, D.J. 1 Flanders, A.E. 6 Buetow, K.H. 2 1 Emory University 2 National Cancer Institute 3 National Institutes of Health 4 University of Virginia 5 SAIC-Frederick, Inc. 6 Thomas Jefferson University