Multiple-response regression analysis links magnetic resonance imaging features to de-regulated protein expression and pathway activity in lower grade glioma

Michael Lehrer, Arvind Rao, Anindya Bhadra, Visweswaran Ravikumar, James Y. Chen, Max Wintermark, Scott N. Hwang, Chad A. Holder, Erich P. Huang, Brenda Fevrier-Sullivan, John FreymannOncoscience 2017 · DOI

TL;DR

This study pioneers a radioproteomic approach to link MRI features of lower-grade gliomas (LGGs) with their underlying protein expression and signaling pathway activity. By employing a novel multiple-response regression framework, the researchers identified specific imaging characteristics that correlate with distinct molecular profiles, offering a non-invasive window into tumor biology.

What Problem Does This Paper Address?

Lower-grade gliomas (LGGs), while less aggressive than glioblastomas, still pose significant treatment challenges and can progress to more malignant forms. Understanding the complex interplay between a tumor's physical appearance (as seen on MRI) and its molecular machinery is crucial for developing better diagnostics and predicting patient outcomes. Traditional approaches often analyze imaging and molecular data separately, missing potential correlations. This paper aims to bridge this gap by integrating radiomic features (derived from MRI) with proteomic data to uncover how macroscopic tumor characteristics relate to microscopic molecular events and signaling pathways.

Methods

The core of this study is a multiple-response regression analysis that accounts for the inherent correlations between protein expression levels. This is a key methodological innovation, as standard regression techniques often assume independence of responses, which can lead to inaccurate associations when dealing with complex biological data. The researchers used a Bayesian framework to handle the high dimensionality of the data (many imaging features and many proteins).

Here's a breakdown of the approach:

  • Image Feature Extraction: Radiologist-scored features from pre-operative MRI scans, known as the VASARI (Visually Accessible Rembrandt Images) feature set, were used. These features describe various aspects of the tumor's appearance, such as shape, size, enhancement patterns, and location.
  • Proteomic Data: Reverse-phase protein array (RPPA) data, which measures the expression levels of thousands of proteins and their phosphorylated forms, was obtained.
  • Multiple-Response Regression: This statistical model was employed to find associations between the VASARI imaging features (independent variables) and the RPPA protein expression levels (dependent variables). The model specifically accounts for the correlated nature of protein expression, providing a more robust analysis.
  • Pathway Analysis: Significantly associated proteins were then analyzed using Ingenuity Pathway Analysis (IPA) software to identify the biological pathways and cellular functions that were most strongly linked to specific imaging features.
  • Hierarchical Clustering: This technique was used to group VASARI features and pathways based on their correlation patterns, helping to visualize and understand the relationships between imaging phenotypes and molecular activities.

Datasets

  • TCGA (The Cancer Genome Atlas) LGG Dataset: This provided the matched proteomic (RPPA) and clinical data for 57 patients with lower-grade gliomas.
  • TCIA (The Cancer Imaging Archive) Dataset: This provided the MRI data for the same 57 patients.

The study leverages existing, well-established datasets, rather than introducing new benchmarks. The novelty lies in the integrated analysis of these datasets.

Key Insights and Ideas

  • Radiomics as a Window into Molecular Activity: The study demonstrates that macroscopic features visible on MRI are not just passive reflections of tumor presence but are actively linked to specific molecular processes and signaling pathways. This opens up possibilities for non-invasive assessment of tumor biology.
  • Correlated Protein Responses Matter: The use of a multiple-response regression framework that accounts for correlated protein expression was crucial. This allowed for the identification of more accurate and biologically relevant associations than traditional methods might have revealed.
  • Specific Imaging Features Link to Specific Pathways: Certain MRI characteristics were found to be strongly associated with distinct sets of signaling pathways. For example, edema was linked to increased NGF signaling, while the presence of cysts was associated with decreased PI3K/AKT and phospholipase C signaling.
  • Functional Implications of Imaging: The findings suggest that imaging features can reflect underlying biological events. For instance, hemorrhage, visible on MRI, was linked to stress response signaling pathways, highlighting how imaging can provide clues about tumor behavior and its response to its microenvironment.

Key Results

  • The multiple-response regression analysis identified multiple proteins significantly associated with each VASARI imaging feature.
  • VASARI features were found to be correlated with the expression of key signaling pathways, including IL8, PTEN, PI3K/Akt, Neuregulin, ERK/MAPK, p70S6K, and EGF signaling pathways.
  • Specific imaging features showed strong associations with unique patterns of pathway activation and suppression. For example, the T1/FLAIR ratio was associated with down-regulation of AMPK and acute myeloid leukemia signaling, while MRI necrosis was linked to up-regulation of PI3K/AKT/mTOR signaling and apoptosis.
  • Hierarchical clustering revealed distinct groups of VASARI features that were highly correlated with specific pathway alterations and biological dysfunctions.

Limitations and Future Work

  • Retrospective Study: The study is retrospective, relying on existing datasets. Future prospective studies would be beneficial to validate these findings.
  • Limited VASARI Features: While 27 VASARI features were used, the full set of 34 was not annotated for all patients.
  • Statistical Challenges: The paper acknowledges the inherent analytical and computational challenges in integrating diverse biomedical datasets and the need for robust methods to minimize false discovery rates.
  • Future Directions: The authors suggest that this approach could lead to novel diagnostics and prognostics, potentially even replacing traditional physical biopsies for determining signaling activity status non-invasively. Further development of these analytical techniques would be a major advancement. The study also highlights the potential for this framework to complement proteomics and genomics for a more comprehensive understanding of tumor biology and treatment response.

Summary updated

Abstract

Radioproteomics analysis might enable an insight into the phenotypic consequences of molecular aberrations in LGGs.
BibTeX
@inproceedings{lehrer2017multiple,
  author = {Michael Lehrer and Arvind Rao and Anindya Bhadra and Visweswaran Ravikumar and James Y. Chen and Max Wintermark and Scott N. Hwang and Chad A. Holder and Erich P. Huang and Brenda Fevrier-Sullivan and John Freymann},
  title = {{Multiple-response regression analysis links magnetic resonance imaging features to de-regulated protein expression and pathway activity in lower grade glioma}},
  booktitle = {Oncoscience},
  year = {2017},
  doi = {10.18632/oncoscience.353},
}
Figure