A new study in contrasts: brain MRI for the depiction of tumor metabolism

See the article by Yao et al. in this issue, pp. 1184–1196.

Imaging of the brain began with coarse and indirect means like evaluating for brain shifts with catheter angiography and for ventricular size and position with pneumoencephalography. Sir Godfrey Hounsfield and head CT then revolutionized brain imaging in the early 1970s. Immediately on the heels of CT came the brilliance of brain MRI, developed through the Nobel-prize winning work of Damadian, Lauterbur, and Mansfield.¹ The ability to discriminate between types of tissues, through various image contrast-producing “weightings,” represented the quantum leap of MRI beyond CT. Depiction of anatomy became exquisite. Differentiation between gray and white matter was easy. Brain tumors were much more easily identified and characterized. Intravenous gadolinium contrast added even more discrimination.

This anatomical or structural imaging has become exquisite over the last 50 years. But we wanted more, to “see” physiology, and so physiologic/mechanistic/functional MRI techniques were developed. Restricted diffusion on diffusion weighted imaging (DWI) implies the energetic failure of cell membrane ion pumps in ischemic infarction but also increased cellularity in tumors. Perfusion imaging in its various iterations (dynamic susceptibility contrast [DSC], arterial spin labeling, dynamic contrast enhanced [DCE]) gives us metrics like cerebral blood volume, cerebral blood flow, and K_trans, important for evaluating tumor neovascularity. MR spectroscopy measures metabolites which are altered in disease states. The image contrast with all of these techniques is based on MR signal differences between tissues, which are derived from the differing electromagnetic characteristics of precessing protons existing within differing chemical environments. This tissue contrast is coaxed into visualization through the complexity and genius of MRI radiofrequency pulse sequences.

The power of physiologic imaging has been made apparent through reports like one in 2008 showing the ability of DSC perfusion to predict glioma time to progression better than histologic grade.² Although perfusion imaging has had its successes and limitations, it remains fairly established in advanced brain tumor evaluation. Many of us know some inherent challenges to the accuracy of DSC, such as leakage of gadolinium into tumors, difficulties in locations near the skull base, and the presence of intracranial hemorrhage. But as we press such physiologic MR techniques for reliable quantitative information, we find that many other points of variability exist in their implementation and analysis.³ The devil in these details remains an obstacle to their validation and widespread implementation.

As the supreme importance of genetic and molecular factors in tumor biology and the prediction of responses to therapy has become universally understood,⁴ we have immediately turned to imaging and the new field of radiogenomics to see if any of the currently existing image contrasts can help us to non-invasively predict tumor genetics and molecular characteristics. In gliomas, isocitrate dehydrogenase (IDH) mutation and 1p/19q codeletion were obvious characteristics to study. Contrasts within MR spectroscopy, DWI, perfusion, permeability, texture, and amide proton transfer imaging have all been used to predict glioma IDH mutation status.^5,6

Now there is the molecular imaging technique of amine chemical exchange saturation transfer–spin and gradient echo–echoplanar imaging (CEST-SAGE-EPI). In this issue of Neuro-Oncology, Jingwen Yao, Benjamin Ellingson, and coauthors present their latest work with this technique in a retrospective analysis of 90 glioma patients.⁷ CEST-SAGE-EPI is a novel and complex MR pulse sequence with an acquisition time of 7.5 minutes that gives us 2 new independent image contrasts which correlate with (that is, are weighted by or sensitive to) tissue pH and hypoxia. As described previously by this group and supported through phantom work,⁸ the CEST metric of magnetization transfer ratio asymmetry at 3 ppm correlates with tumor pH, and the SAGE-derived metric of R₂′ correlates with deoxyhemoglobin concentration and thus oxygen extraction fraction and hypoxia. With one MR pulse sequence, image contrasts for both acidity and hypoxia are created, allowing us to visualize these 2 aspects of tumor metabolism. For instance, the authors “see,” with MRI images, the Warburg effect in tumors: increased glycolysis (lower pH) despite adequate oxygenation (i.e., no increase in oxygen extraction fraction). And they see differences in metabolism between IDH-wildtype and IDH-mutated gliomas, the latter generally having less acidity and less hypoxia. In this study, pH-weighted and hypoxia-weighted MR parameters also correlate with tumor cell staining for hypoxia-inducible factor 1 alpha (HIF1α) and Ki67. In another recent publication, this group found an overall but locally varying correlation between tumor acidity and hypervascularity as measured by CEST and DSC perfusion.⁹ It is interesting to see the field of imaging contribute to the further understanding of tumor metabolism, as their results support the hypothesis that 2-hydroxyglutarate (an oncometabolite produced by IDH-mutant gliomas) activates the prolylyl-hydroxylase domain enzyme, which then leads to the degradation of HIF1α and prevents the metabolic shift from oxidative phosphorylation to glycolysis.

Yao et al find moderate accuracy with CEST-SAGE-EPI in differentiating gliomas by IDH mutation status (81% sensitivity and 81% specificity, area under the curve = 0.86).⁷ This is not exceptional performance compared with other MRI correlates of IDH mutation status.^5,6 But what is important is that two relatively new MRI contrasts have been introduced. It may be difficult to predict just how useful these particular ones may be, not just in diagnosis but also in guiding therapies and evaluating response. In our rapidly accelerating age of artificial intelligence and deep learning using just standard, anatomical MRI contrasts have already produced an accuracy of 94% in determining IDH1 mutation status.¹⁰ It is likely that if we give artificial intelligence new physiologic and metabolic image contrasts with which to work, it will perform even better.

The CEST-SAGE-EPI technique currently has its limitations and caveats, as the authors discuss well.⁷ Factors other than pH and oxygen extraction can confound the calculation of their pH- and hypoxia-sensitive MR parameters. And as with any quantitative imaging technique, there are many other points of unsettled variability all along the course of its performance and analysis, and the genetic, histologic, and metabolic heterogeneity of gliomas only adds to the complexity of analysis. But the point has been made. The brilliance of MRI, with its already spectacular image contrasts, leaps another quantum forward with the introduction of metabolic imaging contrasts.