Imaging glioma biology: spatial comparison of amino acid PET, amide proton transfer, and perfusion-weighted MRI in newly diagnosed gliomas

Using FET as a tracer, PET can visualize the amino acid uptake in gliomas and thus metabolically active tumor cells [1]. Several studies have demonstrated the clinical utility of FET-PET for preoperative grading [2] and biopsy planning [3] as well as for differentiating tumor progression from therapy-associated changes [4]. MR-based DSC perfusion provides evidence of neoangiogenesis [5]—a hallmark of malignant gliomas—and thereby also helps to distinguish between gliomas of different WHO grade and malignancy [6]. In addition, DSC has been shown to reflect differences in angiogenic pathways between isocitrate dehydrogenase (IDH) mutant and wild type gliomas [7] and to help predict patient survival [8] and response to anti-angiogenic therapy (bevacizumab) [9].

Amide proton transfer–weighted (APTw) imaging is a relatively novel MRI technique that relies on the constant dissociation and transfer of amide-bound hydrogen atoms into the surrounding water. By first saturating the amide-bound hydrogen, and then measuring the decrease in free water signal (due to the transfer of saturated hydrogen atoms into the surrounding free water pool), APTw imaging semi-quantitatively reflects the concentration of endogenous proteins and peptides. Two studies have found a good correlation between APT-weighted signal values and protein profiles [10] and tissue proliferative index (Ki-67) [11], respectively. APTw imaging has been shown to readily differentiate between gliomas of different WHO grades [12] and has shown promise to aid in the differentiation of tumor progression from therapy-associated changes [13, 14].

Integrating information from different techniques and modalities has helped to further decipher the complex interplay of biological processes in gliomas [15]. A few studies have already investigated the added value of APTw to diffusion- and perfusion-weighted imaging [16] or methionine PET [17].

In gliomas, a recent study surprisingly reported no correlation between APTw and FET [18]. Given the partially overlapping biological processes captured by these imaging modalities, these findings motivated us to investigate the synergism of APTw, FET, and CBV and how well they capture cellularity and vascularisation in a series of patients with newly diagnosed gliomas, both in IDH mutant and wild type tumors.

In newly diagnosed gliomas, we found evidence for a relevant overlap of tumor areas defined by established cutoffs for APTw and FET, both in CET and FHT. This overlap was strongest in the contrast-enhancing areas of GB, but remained observable even in LGG, where APTw was generally low. Further speaking to the similar biological processes captured by APTw and FET, we found a correlation between cellularity for both imaging modalities.

Our findings contradict a recent report by da Silva et al. [18], who reported no overlap between APTw MRI and FET-PET in simultaneous PET-MRI of eight patients. APTw values cannot be directly compared, because different RF saturation (B_1,rms = 1.0μT, T_sat = 0.1 s) and image acquisition techniques (steady-state saturation transfer with 3D gradient echo EPI readout) were used in that study. The steady-state technique operates at a lower RF saturation duty cycle (74%) and part of the data (about 10%) is acquired with short saturation, before a steady-state is reached. Thus a lower APTw tumor to normal tissue contrast is expected. A low contrast-to-noise ratio (CNR) is further implied by using gradient echo readout with a low flip angle (11°) and spending only less than 10% of the acquisition time (70 offsets ± 14 ppm) near the nominal RF offset for APTw (+ 3.5 ppm). Higher CNR in shorter scan time is achieved with the fast spin echo sequence used in our study, spending 56% (5 out of 9 volumes) of the acquisition at + 3.5 ppm ± 0.8 ppm.

In contrast to that, our results agree with and expand several prior studies on APTw in LGG and GB [11, 12, 16]. In our study, the relative volumes of APT-hyperintense areas were significantly lower in the FHT in LGG compared with GB. This is in good agreement with the literature, where higher mean APTw values were found to be characteristic of more malignant gliomas. Nonetheless, by analyzing not just averaged voxel values, but integrating spatial information, we find that even in LGG, there are areas of APTw hyperintensity above the threshold defined by Jiang et al. [14]. More importantly though, even in LGG, we observe some overlap between APTw and FET. We speculate that areas defined as hotspots both by APTw and FET may represent the most malignant parts of tumors. This needs to be clarified in future prospective studies collecting targeted biopsies, guided both by FET and APTw.

A few studies have investigated how APTw and CBV can be used synergistically for tumor grading [12, 16] or differentiating tumor progression from pseudoprogression [33]. Across both entities and regions studied, we find that the overlap between APTw/CBV is relevantly lower than between APTw/FET, indicating a potential synergistic value of combining APTw and CBV information. This is further corroborated by investigating the stereotactic biopsies, where we find a more pronounced association of CBV and vascularity (compared with APTw) on the one side, and a stronger correlation of APTw with cellularity on the other side. Given the biological processes captured by both modalities, this disparity makes sense. As previously noted, we observe an association for FET with both cellularity and vascularity [34]. Our findings also underline the potential of multi-modal imaging to make oncogenic processes visible and thus support decision-making in clinically challenging situations such as glioma grading or differentiating tumor progression from radiation necrosis. Along this line, such information-rich datasets serve as an ideal basis for training machine learning classifiers which are able to (non-linearly) integrate the multimodal input data, as we and others have previously demonstrated [15].

While the cutoffs for FET [28] and CBV [29] are rather well-established, three studies have reported different thresholds for APT: for differentiating grade II from grade III/IV gliomas, Togao et al. [16] reported an APTw cutoff of 2.56 in 34 gliomas without intense contrast-enhancement. In contrast, Choi et al. [12] found a cutoff of 1.53 to optimize sensitivity and specificity in their setting. More recently, Jiang et al. reported their results on using APTw for distinguishing between vital tumor and radiation necrosis [14]: In a cohort of 21 patients with a suspected early tumor progression vs. pseudoprogression, several stereotactic biopsies were taken to account for intratumoral heterogeneity and a region-wise analysis of APTw intensities as performed. To maximize the discrimination between vital tumor and treatment effects, these authors report a cutoff of 1.79. For our analysis, we decided to apply this cutoff, as it best reflected active, vital tumor and was similar to the cutoff by Choi et al. However, we also investigated the increase of APTw values in PET-positive subregions of CET and FHT, eliminating the need for selecting an APTw cutoff. Given the almost perfect overlap of APTw and FET hot-spots in CET, we observed no significant differences in CET. Importantly though, we found significantly higher APTw values in PET-positive areas in FHT in GB and a strong trend towards higher APTw values in FET hotspots even in the FHT in LGG, where APTw was generally low.

In this study, we included PET acquisitions from PET-CT and PET-MRI. To minimize the potential bias when comparing PET data from both sources, we followed the approach from Koesters et al. [21], which demonstrated a good agreement between PET data from MRI and CT. We specifically investigated whether there were differences between volume and overlap data between patients scanned at different machines, but found no evidence for this (all comparisons p > 0.2). While scanning all patients in the same machine makes for a more homogeneous cohort, results obtained from images acquired on multiple scanners tend to be more robust and generalizable.

The main limitation of our study is the non-simultaneous acquisition of PET and APTw/CBV. While we aimed to minimize the delay, some cases had 4–5 days between both scans. Given the rapid growth and genomic instability of tumors, this may have an impact on the spatial overlap of imaging sequences. On the other hand, for biological imaging to be clinically useful (e.g., for biopsy or radiotherapy planning), the information contained in the images must be considered stable over a few days. In addition, the sample size investigated here is rather small. However, we took care to only include genomically well-characterized, accordingly grouped samples [19] to avoid leveling differences between biologically distinct groups of tumors. Nonetheless, validation of our findings in independent cohorts is necessary. Lastly, the scoring system for neovascularization used in our study is semiquantitative, and all biopsy samples were analyzed by a single neuropathologist to avoid interobserver variance, as described previously [34].

Both in IDH wild-type glioblastomas and IDH mutant lower-grade gliomas, we observed relevant overlap between tumor areas defined by different imaging modalities, strongest for APTw and FET in contrast-enhancing parts of glioblastomas, but also in the FLAIR-hyperintense region of lower-grade gliomas. Further, we find that both APTw and FET are correlated with cellularity, as opposed to CBV. Given that APTw and FET reflect partially overlapping biological information like cellular density [11, 28], this is biologically plausible. Future work to better understand the complementary and synergistic value of biological imaging modalities is warranted.