Region-by-region analysis of PET, MRI, and histology in en bloc-resected oligodendrogliomas reveals intra-tumoral heterogeneity

In our recently published co-registration study, we included five patients with suspected diffuse low-grade glioma (DLGG) that were suitable for en bloc tumor resection [22]. Four of these five patients had histopathological diagnosis of oligodendrogliomas (Figs. 1, 2, 3, and 4) and were included in the present study. The institutional review board approved the protocol for the present analysis, and written informed consent was obtained prior to patient participation. A summary of the clinical characteristics of these patients and histopathological diagnoses is shown in Table 1.

MRI was performed according to our original study protocol [23], using a 3-Tesla scanner (Achieva, Philips Healthcare, Best, The Netherlands), including perfusion sequences. Morphological imaging included axial and coronal T2-weighted fluid attenuated inversion recovery (FLAIR) scans (0.45 × 0.45 × 6 mm³ voxel size). Dynamic susceptibility contrast (DSC) perfusion imaging was performed as previously described [24], with a T2*W single-shot gradient echo-EPI sequence (1.7 × 2.3 × 5.0 mm³). Whole-brain coverage was obtained using these parameters.

PET scanning was performed on a Discovery ST PET/CT (General Electric Medical Systems) in a 3D acquisition mode. This scanner has 24 detector rings with 420 detectors resulting in 47 image planes with a plane separation of 3.27 mm and a total axial field of view of 157 mm. Images were reconstructed using ordered-subsets expectation maximation (OSEM; 2 iterations, 15 subsets, 128 × 128 matrix) with a 2.14 mm FWHM post-filter applying appropriate corrections such as scatter, randoms, and attenuation correction based on a low-dose CT scan. Voxel size of the resulting images was 1.95 × 1.95 × 3.27 mm and the spatial resolution 6 mm. The tracer ¹¹C-Methionine (4.0 MBq/kg) was injected 25 min before start. Scan time was 10 min, 1 frame.

PET and MRI were performed at a mean of 34 days (range, 20-52 days) respectively 28 days (range, 5-53 days) before surgery.

Figure 5 illustrates the design of the study with the different steps that have been performed. Preparation of the en bloc tumor tissues was performed as previously described [22]. After resection, the tumor surfaces (anterior, posterior, medial, and lateral) were color-marked for spatial orientation. The en bloc tissue sample was fixed in formalin, subsequently cut into consecutive tissue slices, again fixed in formalin, and embedded in paraffin wax. Serial microsections were rehydrated prior to immunohistochemical staining with antibodies against the tumor cell marker IDH1-mutated protein (IDH1-R132H), the proliferation marker Ki67 and the hematopoietic progenitor cell antigen CD34 (Fig. 5, steps 1–2).

Software was developed for anatomical landmark-based co-registration of consecutive histological images of various immunostainings. In summary, a landmark-based multi-step elastic image registration was designed using an in-house Matlab application (The MathWorks, Inc., Natick, MA, USA) and implemented allowing a region-by-region comparison of histological assessment [25]. A set of 5–10 anatomical landmarks was found accurate enough for consecutive co-registration of serial histological images (Fig. 5, step 3). Each histological image was divided into sub-images (500 × 500 pixels) for region-by-region quantification of protein expression of histological cell markers, using an additional in-house Matlab application (see Supplementary file for software description).

Automated region-by-region analysis of protein expression was performed using an in-house built pipeline in the CellProfiler software [26] (see Supplementary file). We created a pipeline of individual modules that were able to estimate localization, intensity, shape, and size of protein expression for the three histological cell markers (IDH1-mutated protein as a marker for tumor cells, Ki67 as a marker for proliferating cells and CD34 as a marker for blood vessels). Each separate sub-image (500 × 500 pixels) traveled through the pipeline and was processed by each module to quantify the absolute count per square of protein expression. In the final analysis, mean values of protein expression were calculated in each region of interest defined on MET PET (see under PET analysis). This allowed region-by-region comparison between MET uptake, tumor perfusion, and protein expression (Fig. 5, step 4).

As a first step, we imported original co-registrations of MRI and histological images (denoted as co-registration 1) created in our previous study [22] into Carimas software (kindly provided by the Turku PET center, Finland). We then imported the T2-FLAIR images that were co-registered with the co-registration 1 image using the Normalized Mutual Information (NMI) tool by Carimas [27, 28]. Following this step, we imported both MET PET scans and MRI perfusion maps (cerebral blood volume, CBV) into the project and co-registered these to the FLAIR images using the same NMI tool (denoted as co-registration 2). This way, all co-registered images had similar orientation (Fig. 5, step 5).

Analysis of MET PET was performed using Carimas 2.9 software (Turku PET center, Finland). MET uptake in regions of interest (ROIs), defined as described below, was calculated as tumor-to-normal tissue (T/N) ratio, i.e., the ratio between the maximum standardized uptake value (SUV), reflecting the value of the pixel with highest radioactivity in the ROIs and the mean uptake in the contralateral normal brain. The normal tissue reference region was drawn as a 1-cm-thick cortical ROI at the level of the basal ganglia, ranging from the occipital to the frontal cortex in the contralateral hemisphere. Additionally, we defined a deep white matter reference region in the contralateral hemisphere above and lateral to the ventricle.

Serial ROIs were systematically selected on all MET PET scans. We defined three different types of ROIs based on MET uptake in the entire tumor volume: ROI₁ = areas with highest MET uptake (hotspot), ROI₂ = areas with medium MET uptake, and ROI₃ = areas with lower MET uptake located in the infiltrating tumor border (Fig. 5, step 6).

DSC perfusion was analyzed as previously described [24], with calculation of CBV maps using nordicICE software (NordicNeuroLab AS, Bergen, Norway). Relative CBV (rCBV) was calculated as tumor-to-normal tissue (T/N) ratio by dividing the mean CBV in ROIs defined on PET, with the mean CBV within a reference area located in normal appearing white matter of the contralateral hemisphere above and lateral to the ventricle.

To compare quantified protein expression of histological cell markers within the ROIs defined on PET, we applied a grid using Adobe Photoshop CC (Adobe Systems Software). The grid was placed on top of co-registration 2, enabling a detailed and direct comparison between radiological and histological images. All histological sub-images (500 × 500 pixels) within one specific ROI were selected using the grid. This procedure was applied for each histological image of all four tumors. In the final analysis, we used the mean count of protein expression per ROI for each of the three histological cell markers (Fig. 5, step 7).

A 33-year-old man had a non-enhancing heterogeneous tumor in the right frontal pole on MRI. En bloc resection was performed as described previously [22] and histopathological examination showed an oligodendroglioma WHO grade II with IDH1 mutation, 1p/19q codeletion, and Ki67 < 5% (Table 1). A total of 29 ROIs were selected on PET; 16 areas with highest MET uptake (ROI₁), seven areas with medium uptake (ROI₂), and six areas with lower uptake in the tumor periphery (ROI₃) (Fig. 1).

The MET uptake and rCBV values and the quantified protein counts (mean number of counts per ROI) in all 29 ROIs, as well as the statistical correlations between the various parameters are presented in Table 2. As shown, there was a strong correlation between MET uptake and tumor cell density (MET-IDH1: r = 0.91; p < 0.0001), MET uptake and vessel density (MET-CD34: r = 0.73; p < 0.0001), and MET uptake and proliferation (MET-Ki67: r = 0.71; p = 0.0465) in this tumor. No significant correlation was found between MET uptake and tumor perfusion (MET-rCBV: r = 0.06; p = 0.74). In addition, rCBV showed no significant correlations with histological cell markers.

A 50-year-old woman was diagnosed with a tumor in the left frontal pole with heterogeneous signal characteristics on MRI (Fig. 2). The tumor was removed en bloc and histopathological examination showed oligodendroglioma WHO grade II with densely packed IDH1-labeled tumor cells located mainly in the grey matter [22]. Molecular analysis showed IDH1 mutation, Ki67 < 5% and 1p/19q codeletion (Table 1). A total of 17 ROIs were selected of which eight representing regions with highest MET uptake (ROI₁), four with medium uptake (ROI₂), and five with lower uptake located in the tumor periphery (ROI₃). The MET uptake, rCBV value, and protein expression of histological markers in these ROIs are presented in Table 2. There was a statistically significant correlation between MET uptake and IDH1 count (MET-IDH1: r = 0.51; p = 0.0345). As shown in Table 2, there were no significant correlations between MET uptake with tumor perfusion and with expression of Ki67 and CD34. We observed that several of the ROI₃ in this tumor were located adjacent to or partially overlapping with the cortex, with inherent higher perfusion (Table 2).

A 39-year-old man was examined showing a non-enhancing, slightly heterogeneous tumor and en bloc tumor resection was performed. Histopathological examination showed a WHO grade II glial tumor with exclusively oligodendrocytic differentiation. Molecular analysis showed IDH1 mutation but no 1p19q codeletion. In spite of the intact 1p19q chromosomes, the tumor was morphologically diagnosed as an oligodendroglioma based on its characteristic oligodendroglial phenotype throughout the entire resection (Table 1). A total of 23 ROIs were selected, of which 14 in hot spot regions (ROI₁), five in areas with medium uptake (ROI₂), and four in areas with lower uptake in the tumor periphery (ROI₃) (Fig. 3) (Table 2). There was a significant correlation between MET uptake and tumor cell count (MET-IDH1: r = 0.44; p = 0.0371), proliferation count (MET-Ki67: r = 0.69; p = 0.0095), and vessel count (MET-CD34: r = 0.67; p = 0.0005). No significant correlations were present between tumor perfusion and histological cell markers.

A 53-year-old man was diagnosed with a left frontal tumor showing minimal contrast enhancement on MRI (Fig. 4). En bloc resection was performed, with some loss of white matter tissue on the medial/inferior side of the tumor. Histopathological examination showed IDH1-mutated codeleted oligodendroglioma WHO grade III, Ki67 proliferation rate was 25% (Table 1) [22]. A total of 15 ROIs were identified on PET, of which nine in the hot spot (ROI₁), five with medium uptake (ROI₂), and due to loss of white matter tissue during en bloc resection only one representative ROI₃ with lower MET uptake located in the tumor periphery. Statistical analysis showed a strong correlation between MET uptake and IDH1 (MET-IDH1: r = 0.85; p < 0.0001) (Table 2). No significant correlation was found between MET uptake and tumor perfusion or the expression of other histological markers. Similar to patient 2, we observed that the single ROI₃ in this tumor was located adjacent to the cortex, resulting in inherent increased perfusion values.

MET uptake was consistently correlated with tumor cell density throughout the entire tumor volume in all four oligodendrogliomas. Similar results were obtained using white matter as reference area for MET uptake (data not shown). Tumor perfusion in the ROIs defined by MET uptake did not correlate with MET uptake or with any of the histological cell markers.

In two tumors (patients 1 and 3), MET uptake also correlated with the density of proliferating cells and blood vessels. These two tumors showed consistently decreasing rCBV values at longer distance from the tumor core. In the other tumors (patients 2 and 4), we noticed an unexpected increase of rCBV values in ROI₃, which was due to partial volume effects with adjacent cortex and large vessels in the peritumoral region.

One limitation of this study is the small number of patients, which is however inherent to the methodological design of the study. Only a selection of gliomas can be surgically removed by en bloc tumor resection. It should also be noted that we applied four separate analyses of intra-tumoral heterogeneity, which means that comprehensive data were not pooled for inter-tumoral analysis. Instead, we analyzed multiple ROIs in each tumor and aimed to identify common patterns between tumors. Future studies with larger tumor samples and including astrocytic tumor subtypes are needed to confirm our data. In addition, there are some technical limitations to be considered. Both MET uptake and rCBV were measured in vivo before surgery and compared with ex vivo markers. Therefore, anesthesia or surgery may have affected microvessel contractility [48]. We encountered no significant difficulties during fusion of conventional MRI and PET images using the Carimas software. The Normalized Mutual Information (NMI) tool used in this study has previously proven to be highly accurate, with only a 1.2% failure [28]. In our previous work, we provided a detailed description of the limitations occurring during the manual process of histology-to-radiology co-registration [22]. Deformations and shrinkage that occur when tissue is processed for histology (i.e., formalin fixation, sectioning, and staining) make the procedure of co-registration challenging. Potential mismatch between PET/MRI and histological parameters has been minimized by applying a grid to the en bloc tumor resections to select specific areas, and by calculating mean values within relatively large ROIs to account for inaccuracies. However, the selection of ROIs located adjacent to or partially overlapping the cortex has probably influenced our results. Finally, a fixed pipeline was performed in CellProfiler to prevent bias, but not all histological stainings were done simultaneously, which could cause batch-related variations with regard to relative immunohistochemical staining intensities.