Worse prognosis for IDH wild-type diffuse gliomas with larger residual biological tumor burden

The association of overall survival (OS) with tumor burden, including contrast enhanced (CE) volume on CE T1-weighted images, fluid-attenuated inversion recovery (FLAIR) hyperintense volume, and 3, 4-dihydroxy-6-[18F]-fluoro-L-phenylalanine (FDOPA) hypermetabolic volume, in isocitrate dehydrogenase (IDH) wild-type gliomas remains unclear. This study aimed to assess the association between biological tumor burden in pre- and post-operative status and OS in IDH wild-type gliomas, and evaluated which volume was the best predictor of OS. Thirty-four patients with treatment-naïve IDH wild-type gliomas (WHO grade II 6, III 15, IV 13) were retrospectively included. Three pre-operative tumor regions of interest (ROIs) were segmented based on the CE, FLAIR hyperintense, and FDOPA hypermetabolic regions. Resected ROIs were segmented from the post-operative images. Residual CE, FLAIR hyperintense, and FDOPA hypermetabolic ROIs were created by subtracting resected ROIs from pre-operative ROIs. Cox regression analysis was conducted to investigate the association of OS with the volume of each ROI, and Akaike information criterion was used to assess the fitness. Residual CE volume had a significant association with OS [hazard ratio (HR) = 1.26, p = 0.039], but this effect disappeared when controlling for tumor grade. Residual FDOPA hypermetabolic volume best fit the regression model and was significantly associated with OS (HR = 1.18, p = 0.008), even when controlling for tumor grade. FLAIR hyperintense volume showed no significant association with OS. Residual FDOPA hypermetabolic burden predicted OS for IDH wild-type gliomas, regardless of the tumor grade. Furthermore, removing hypermetabolic and CE regions may improve the prognosis.


Introduction
In 2016, the World Health Organization (WHO) classification of tumors of the central nervous system reclassified gliomas by integrating molecular status, such as isocitrate dehydrogenase (IDH) gene mutation and chromosomal 1p/19q co-deletion [1]. Overall, approximately 90% of glioblastomas are IDH wild type, whereas the remaining 10% are IDH mutant [2]. About 30% of grade II and III gliomas are IDH wild type, whereas the remaining 70% are IDH mutant [3,4]. Although treatment methods vary depending on patient prognostic factors, including histology, tumor grade, age at diagnosis, Karnofsky performance status, first presenting symptom, extent of resection, and tumor size and location, as well as the molecular status, the standard treatment for patients with newly diagnosed high-grade gliomas remains surgical resection followed by radiotherapy in combination with the DNAalkylating agent temozolomide [3].
Regardless of tumor grade, patients with IDH wildtype gliomas present with a median overall survival (OS) < 2 years, which is significantly shorter than that of IDH mutant gliomas [4]. The extent of surgical resection of contrast-enhanced (CE) regions on T1-weighted magnetic resonance imaging (MRI) has been associated with longer survival [5], and a more aggressive resection beyond CE regions was recently suggested for a better prognosis [6]. On the other hand, amino acid positron emission tomography (PET), including 3, 4-dihydroxy-6-[ 18 F]-fluoro-L-phenylalanine (FDOPA) and O-(2-[ 18 F]fluoroethyl)-L-tyrosine (FET), provides metabolic information to complement MRI-derived information. These PET examinations have improved distribution and efficiency because of the relatively longer half-lives of fluorinated tracers compared with carbon tracers. FDOPA PET was shown to be superior to FET PET when visualizing primary and recurrent gliomas in lesions outside the striatum. At the same time, the lower striatal uptake gives FET PET an advantage-especially in cases with gliomas localized to the basal ganglia. Several PET imaging metrics, including standard uptake value (SUV) and hypermetabolic volume [referred to as biological tumor volume (BTV)], were investigated to predict prognosis for gliomas. The BTV for glioblastomas with post-operative preradio-chemotherapy and for recurrent high-grade gliomas were reported to be a significant predictor of OS [7][8][9][10][11].
To date, no studies have evaluated the association of tumor burden, including CE volume, fluid-attenuated inversion recovery (FLAIR) hyperintense volume, and BTV, with OS focusing exclusively on IDH wild-type gliomas, although patients with IDH wild-type gliomas are known to have a poor prognosis, regardless of tumor grade. The purpose of the current study was to assess whether the tumor burden, including CE, FLAIR hyperintense, and FDOPA hypermetabolic regions, in pre-and post-operative exams are associated with OS in patients with IDH wild-type gliomas. We hypothesize that hypermetabolic tumor burden may be a strong predictor of OS compared with volume of CE or FLAIR hyperintense regions.

Patient selection
A total of 37 patients with treatment naïve and histologically confirmed gliomas who underwent pre-operative FDOPA PET and MRI scans between 2007 and 2020 were retrospectively selected. MRI scans were performed within 2 months of the corresponding PET scans. All patients were diagnosed with IDH wild-type gliomas based on IDH1 mutational status. No patients underwent stereotactic biopsy prior to FDOPA PET or MRI scans. Among them, one patient was excluded from this study, because the period between their FDOPA PET and surgery exceeded half a year, and two patients were excluded, because they did not undergo a follow-up MRI examination after surgery. Finally, 34 patients with IDH wild-type glioma were included in this study. OS was measured from the time of the PET scan until death or the censored date (median time 617 days, range 29-2241). The median time between PET scan and surgery was 18 days (range 1-178 days), while the median time between surgery and the first post-operative MRI was 18 days (range 0-101 days). All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Written informed consent was obtained from all individual participants to have their advanced imaging, clinical, and molecular data included in our IRB-approved research database according to IRB#10-0,00,655. Some subjects were also included in a previous study [12], which evaluated voxelwise imaging correlations between FDOPA and MRI.

FDOPA PET image acquisition
A full-ring PET/computed tomography (CT) scanner (ECAT-HR, Siemens, Knoxville, TN, USA) was used to obtain PET images after the subjects fasted for more than 4 h. A CT was performed before the PET scan for attenuation correction. FDOPA was synthesized and injected intravenously, following previously reported procedures [13,14]. Three-dimensional FDOPA emission data were obtained for 30 min, and the data were integrated between 10 and 30 min following the injection to obtain 20 min static FDOPA images after reconstruction. FDOPA PET images were reconstructed using an ordered-subset expectation maximization iterative reconstruction algorithm with six iterations and eight subsets [15,16]. Then, a Gaussian filter with a full width at half maximum of 4 mm was applied. The resulting voxel sizes were 1.34 × 1.34 × 3 mm for FDOPA PET images. SUV maps of FDOPA were calculated based on the radioactive activity divided by the decay-corrected injected dose per body mass. Resulting SUV maps were subsequently normalized (nSUV) relative to the median value of the normal-appearing striatum [12,17,18].

Magnetic resonance image acquisition
The anatomical MRI scans consisted of standard T1-weighted pre-and post-contrast images (2D axial turbo spin echo with a 3 mm slice thickness and no interslice gap, or 3D inversion-prepared gradient echo images with 1-1.5 mm isotropic voxel size) [19]. The T2-weighted FLAIR images were acquired with a 3-mm slice thickness and no interslice gap using a 1.5-T or 3-T clinical MRI scanner. Anatomical images were obtained after surgery (closest date to the surgery), as well as before surgery.

Postprocessing and ROI analysis
Pre-operative MRI and PET images were registered to the pre-operative post-contrast T1-weighted images for each patient using a 6 degree of freedom rigid transformation and a mutual information cost function using FSL software (flirt, FMRIB, Oxford, UK http:// www. fmrib. ox. ac. uk/ fsl/). Post-operative MRI scans were also registered to the post-operative post-contrast T1-weighted images, and then registered to the pre-operative post-contrast T1-weighted images using a 6 degree of freedom rigid transformation and a mutual information cost function. Figure 1 illustrates the process of region-of-interest (ROI) analysis. Using pre-operative images, three tumor ROIs were segmented based on (1)   post-operative images, which have been registered to the pre-operative images. Residual contrast-enhanced ROIs, residual FLAIR ROIs (blue), and residual FDOPA hypermetabolic ROIs (green) are created by subtracting resected ROIs from pre-surgery ROIs. In this slice, the contrast-enhanced ROI is completely resected nih. gov). For the segmentation of CE and FLAIR hyperintense ROIs, a semi-automatic method was employed. As detailed in a previous study [20], the semiautomated procedure consisted of (1) manually defining relatively larger CE regions on CE-T1 weighted subtraction maps, which were created by the voxelwise subtraction of pre-CE T1-weighted images from post-CE T1-weighted images, (2) determining the threshold (0 for most cases) for removing non-CE regions on the subtraction maps, (3) finally, extracting only CE regions as CE ROIs. Similarly, relatively larger FLAIR hyperintense regions were manually segmented, and intensity thresholds were chosen by visual inspection for each patient to remove normal intensity areas. As a result, only regions with FLAIR hyperintensity were extracted as FLAIR hyperintense ROIs. For FLAIR hyperintense ROIs, necrotic areas were not excluded if they showed hyperintensity on FLAIR images. FDOPA hypermetabolic ROIs within FLAIR hyperintense ROIs were extracted with a higher threshold than the median uptake value of the striatum.
Using post-operative images, which were registered to pre-operative images, a single ROI corresponding to the resected tumor was segmented based on the cavity of removed lesions. For gliomas with a large necrotic region or mass effect, the segmentation of resected ROIs was complicated, since the cavity may have moved from the pre-surgical regions. The resected ROIs were, therefore, determined using all available pre-and post-operative images, including CE-T1WI, non-CE-T1WI, T2WI, and FLAIR images, to differentiate the resected ROIs from adjacent brain regions by referring to normal brain structures as well as resected cavities. Then, the ROIs were validated by another boardcertificated neuroradiologist (H.U. with 14 years of clinical experience) by consensus.
Finally, the registered resected ROIs were subtracted from the CE ROIs, FLAIR ROIs, and FDOPA hypermetabolic ROIs, resulting in residual CE ROIs, residual FLAIR ROIs, and residual FDOPA hypermetabolic ROIs, respectively. All volumes of tumor ROIs are reported in milliliters (mL). The maximum nSUV (nSUV max ) within the FLAIR ROIs and residual FLAIR ROIs was also calculated.

Statistical analyses
Cox univariate proportional hazards regression analyses were conducted to investigate the association between OS and predictor variables including clinical information, such as age and tumor grade, and imaging metrics, such as the CE volume, residual CE volume, FLAIR volume, residual FLAIR volume, FDOPA hypermetabolic volume, residual FDOPA hypermetabolic volume, nSUV max , and residual nSUV max . All variables, including tumor grade, were treated as continuous variables. When the CE or FDOPA hypermetabolic regions were unidentifiable or completely resected, the volumes represented 0 mL. For imaging metrics showing a significant association in the Cox univariate regression, Cox multivariate regression analyses controlling for age, tumor grade, or both age and tumor grade were performed separately. To compare the fitness of the regression model, the Akaike information criterion (AIC) was calculated. The most fitted model is the one with the minimum AIC among all models. Statistical analysis was performed using R software (version 3.5.2 http:// www.r-proje ct. org/), and statistical significance was defined at p < 0.05.

Results
The current study included 34 treatment-naïve IDH wildtype glioma patients (13 females) with a mean age ± standard deviation of 61.0 ± 10.8 years at the time of the PET examination. Nine patients underwent partial resection/biopsy, 19 underwent subtotal resection, and six underwent gross total resection. According to the 2007/2016 WHO criteria, six gliomas were grade II, 15 were grade III, and 13 were grade IV. Detailed patient demographics are shown in Tables 1, 2 [1]. Seventeen patients did not have CE regions on preoperative images (grade II 5, grade III 10, grade IV 2), while six patients did not have FDOPA hypermetabolic regions (grade II 2, grade III 3, grade IV 1).

Discussion
This study evaluated whether the tumor burden, including pre or post-operative CE, FLAIR hyperintensity, or FDOPA hypermetabolic volume, was associated with OS in IDH wild-type gliomas. Residual FDOPA hypermetabolic volume after surgery was the strongest predictor of OS when controlling for age and tumor grade. Previous studies using FET PET showed that BTV before radio-chemotherapy was a significant predictor of OS for glioblastoma, but did not evaluate whether this applied to other tumor grades [7,8]. Another study using 11 C-methionine PET also described BTV as a significant predictor of OS for recurrent gliomas of grade III/IV [9]. No studies, to the best our knowledge, have yet evaluated the association of OS with BTV specifically within IDH wild-type gliomas. This study revealed that residual BTV after surgery was significantly associated with OS for IDH wild-type gliomas regardless of tumor grade. Since most newly diagnosed glioblastomas are biologically IDH wild type [2], our results are consistent with the findings of previous studies evaluating higher grade gliomas using amino acid PET [7,8]. Conversely, due to the biological similarities to glioblastomas [21], patients with lower grade IDH wild-type gliomas also showed shorter survival than patients with IDH mutant gliomas. As such, residual BTV could be a crucial biomarker to predict prognosis for IDH wild-type gliomas regardless of tumor grade.
This study suggests that CE volume may not be an appropriate biomarker to predict prognosis especially for lower grade gliomas with IDH wild type due to the relatively low rate of contrast enhancement. Several previous studies have documented an association of OS with residual CE volume on post-operative images for glioblastomas [5,22]. However, the residual CE volume, which showed a significant association with OS in Cox univariate regression analysis, was not significantly associated with OS when controlling for tumor grade. A small population size may have influenced the statistical power. Meanwhile, 71% (15/21) of grade II/III IDH wild-type gliomas showed no contrast enhancement, and 15% (2/13) of grade IV glioblastomas showed no contrast enhancement in this study. Less than half of grade II/III IDH wild-type gliomas are reported to show contrast enhancement [4,23], and not all grade IV glioblastomas showed contrast enhancement. Therefore, CE volume should be paid particular attention to best predict prognosis in lower-grade IDH wild-type gliomas.
The gross macroscopic resection of the CE component of glioblastomas was associated with longer survival [22,24]. A recent large cohort study supported this association [2]. MRI is the first choice for treatment planning, and the maximal resection of CE regions represents the neurosurgical standard, when safely feasible. A more aggressive resection of FLAIR hyperintense regions was also suggested [6]. However, contrast enhancement is seen in areas with disrupted blood-brain barrier; in addition, FLAIR hyperintense regions can reflect tissue edema, and thereby do not necessarily correspond to the extent of tumor cells. Furthermore, due to the inherent trade-off between the extent of the resection of the tumor region and postoperative functional status, aggressive regression requires sufficient attention. Meanwhile, amino acid tracer accumulates in tumor cells and could define tumor extension more accurately, especially in metabolically active non-CE and FLAIR hyperintense regions [25][26][27]. Our results suggest that maximal surgical reduction of the BTV may contribute to the better prognosis of patients with resectable gliomas, and that residual hypermetabolic volume is a more useful biomarker to predict prognosis than CE volume. Therefore, the maximal removal of both hypermetabolic and CE regions may improve the prognosis of patients with IDH wild-type gliomas with relatively small functional defects.
There are a number of limitations to our study. First, for gliomas with large necrotic regions or mass effects, the resected cavity may shrink or move from pre-surgical locations after surgery, incurring difficulties in segmentation. To mitigate this problem, two neuroradiologists validated the resected ROIs by consensus using all available images. Alternatively, we calculated all imaging metrics from preoperative images; thus, the strength of our method lies in the fact that imaging metrics do not suffer from post-surgical effects, such as blood-brain barrier breakdown or a postoperative flare phenomenon, which may incur additional contrast enhancement or amino acid tracer uptake [28]. Second, due to the retrospective nature of the study, the parameters of MRI sequences and the term between PET examination and surgery or between surgery and post-operative imaging examination were not consistent. In particular, the time period between surgery and post-operative imaging is associated with post-surgical contrast enhancement or edema, which may affect the segmentation of resected ROIs [29,30]. The third limitation is the small population size, which made it difficult to control for multiple covariates. A larger study with an external data set to validate the association between the extent of resection of hypermetabolic regions is warranted. Fourth, additional clinical information, including Karnofsky performance status and clinical symptoms, and tumor markers, including O 6-methylguanine-DNA methyltransferase (MGMT) methylation, telomerase reverse transcriptase (TERT) promotor mutations, and epidermal growth factor receptor (EGFR) mutation/amplification, are important to evaluate survival; however, they were not available for all subjects, and were not analyzed in this study. Furthermore, this study did not take post-operative treatment methods nor tumor location into account, although these factors may affect prognosis. Lastly, differences in patient age and sex may have affected FDOPA background uptake, influencing BTV.
In conclusion, residual FDOPA hypermetabolic tumor burden after surgery was significantly associated with OS for IDH wild-type gliomas, even when controlling for age and tumor grade. In the context of surgical planning, the maximal removal of hypermetabolic regions as well as CE regions may improve the prognosis of patients with IDH wild-type gliomas.