Clinicopathologic analysis of microscopic tumor extension in glioma for external beam radiotherapy planning

In general, to meet the basic requirements of the linear regression model, the sample size (n) should be greater than 30, or n ≥ 3(k + 1), where k is the number of independent variables [11]. The preset number of independent variables in this study is 9 (k = 9), and the sample size n should be ≥ 30. Therefore, we sought an enrolled sample size of 30, and finally, 42 glioma patients treated at Shandong Cancer Hospital or the First Affiliated Hospital of Shandong First Medical University between February 2016 and July 2020 were recruited. The inclusion criteria were as follows: (a) patients aged 18 years or older, (b) preoperative Karnofsky Performance Status (KPS) ≥ 80, (c) tumor growth in non-functional areas, and (d) tumor location allowing supra-total resection and resection margins 2 cm and 3 cm from the tumor border (defined as the enhancing border of the tumor on T1-weighted MR imaging; for non-enhancing gliomas, tumor border was defined as the border of the hyperintensity signal seen on T2-weighted MR imaging). The exclusion criteria were as follows: (a) a history of a previous brain tumor, cranial surgery, radiotherapy, chemotherapy, or contraindication for MR imaging and (b) multifocal lesions. Intraoperative neuromonitoring and neuronavigation were performed to ensure safety and maximize the extent of intracranial tumor resection. The 2016 World Health Organization was used for postoperative grading [12]. This study was approved by two institutional review boards. Written informed consent was obtained from all participants.

Conventional gadolinium-enhanced T1-weighted (T1_ce), T2-weighted (T2_w), and T2-fluid attenuated inversion recovery (T2_FLAIR) images were acquired in all patients before surgery. Acquisitions were performed using a 3T whole-body system (Philips Achieva 3.0T) with an 8-channel head coil. The MR scanning images were obtained parallel to the orbitomeatal line (OML). Imaging sequences included (a) T1_ce (repetition time/echo time [TR/TE] = 495/10 ms, slice thickness/gap = 3 mm/0 mm, number of signal averaged [NSA] = 1, field of view [FOV] = 260 mm × 260 mm, matrix = 256 × 256); (b) T2_w (TR/TE = 13312/110 ms, slice thickness/gap = 3 mm/0 mm, NSA = 1, FOV = 260 mm × 260 mm, matrix = 416 × 416); and (c) T2_FLAIR (TR/TE = 11000/120 ms, slice thickness/gap=3 mm/0 mm, NSA = 1, FOV = 260 mm × 260 mm, matrix = 320 × 320). MR imagings were analyzed and judged by two senior radiologists (ZQ.C. and PP.S.). In enhancing gliomas, the tumor-containing zone was defined as the area of increased signal intensity on T1_ce image. In non-enhancing gliomas, the tumor-containing zone was defined as the area of the FLAIR hyperintensity signal seen on the T2_w image. For all patients, the radiological tumor size was determined as the longest axis of the tumor in all three directions, measured by a radiologist (PP.S.) blinded with respect to the macroscopic tumor sizes.

The process for large slice preparation was performed as described previously [13]. During surgery, the orientation of the excision specimen was marked with different alphabet tags (Fig. 1a). Meanwhile, the plane of the OML was marked on the specimen. Then, the specimen was fixed in 10% formalin for ≥ 24 h. The macroscopic tumor size, both before and after fixation, was recorded to determine the area retraction due to fixation (Fig. 1b, c and Additional file 1: Table S1). Next, referring to the preoperative MR imaging (Fig. 1d, e), the edges of the specimen were inked with different colors to indicate the original orientation of the specimen in the brain (Fig. 1f). We distinguished three directions: anterior-posterior, skull-falx, and cranio-caudal. Depending on the position of the tumor with respect to the brain, the skull-falx direction corresponds to the left-right direction in the patient. Subsequently, the plane of the OML was perpendicular to the table. The specimens were sectioned consecutively from the cranial side to the caudal side in 3 mm thick slices using a macrotome (Microm HM 450; GMI, Ramsey, Minnesota, USA) (Fig. 1g, h), which ensured that each specimen slice matched the MR imaging slice (Fig. 1i). Finally, sections were embedded in paraffin and cut into 5 μm thick sections, which were stained with hematoxylin and eosin (H&E) and subjected to immunohistochemistry to remark white matter (WM) fiber tracts (Fig. 1j). The following primary antibody was used: a mouse monoclonal antibody against S-100 (S1-61; Santa Cruz, California, USA; 1:100 dilution). An EnVision + System-HRP labeled polymer anti-mouse (Dako, California, USA) was used as the secondary antibody. Furthermore, molecular testing was performed in every patient. The details of the testing performed are included in the Additional file 2. Briefly, the methylation status of the MGMT promoter and the mutation of the IDH were identified using DNA pyro-sequencing. Deletion of chromosomal arms 1p and 19q was tested by use of fluorescence in- situ hybridization.

To avoid interobserver variations, the same pathologist (JJ.S.) assessed the specimens and identified microscopic evidence of ME. First, the H&E sections were scanned by a confocal microscope (LSM800 with Airyscan 2; Carl Zeiss, Oberkochen, Germany). Considering the shrinkage of the surgical specimen, each scanning slide was correct through its corresponding scaling parameters (Additional file 1: Table S1). Then, on histologic slides, the tumor border and PTBE border were assessed with the naked eye (without magnification) and marked on its digitized version. The histologic tumor size and PTBE size, defined as the longest axis of the tumor and PTBE on slides, was also measured and recorded. Subsequently, we used a magnification of 10× to 40×, to search for glioma cells outside the tumor border. Invasive glioma cells were identified by means of their nuclear atypia, heteropyknotic staining, and elongated and hyperchromatic tumor nuclei [13]. Finally, we measured the ME distance relative to the edge of the invasive tumor bulk at each section. The ME was defined as the maximum linear distance from the tumor border to the farthest extent of the invasive tumor cells. An example of ME measured in a primary tumor is shown in Fig. 1k. A single investigator (DB.M.) performed all measurements. Furthermore, the mode of tumor extension was assessed according to the terms described by Louis et al. [14], classified according to four main types: (a) direct extension; (b) perineural spread, i.e., glioma cells grow along WM tracts; (c) subpial spread, i.e., tumor cells spread along pia mater; and (d) perivascular spread, defined by the presence of free tumor cells migrating along the basement membranes of blood vessels (Fig. 2).

The difference between two groups was compared using the Student’s t-test, Fisher's exact test or Chi-squared test, as appropriate. Analysis of variance was used to investigate differences between groups numbering more than two. Post hoc analysis was used for pairwise comparisons. The relationship of radiologic size with macroscopic size and histology size was evaluated by Spearman’s rank correlation. Univariate and multivariate linear regression models were performed to determine independent factors associated with ME distance. To avoid multicollinearity, variance inflation factors for the variables were < 5. Statistical analyses were performed using SPSS Statistical software version 22.0 (IBM Armonk, New York, USA), and statistical significance was set at P < 0.05.

Thirty-eight patients who met the eligibility criteria for this study were analyzed. The other four patients were eliminated either because the specimen was not well-fixed (n = 1) or not intact (n = 3). No patients experienced surgery-related complications. The clinical and pathological characteristics of the patients are displayed in Table 1. The patients included 21 men and 17 women, with a mean age of 48.5 years (range, 20–63 years). Imaging analysis showed that 8 patients had non-enhancing tumors (all had low-grade gliomas [LGG]), and 30 patients had enhancing tumors (all had high-grade gliomas [HGG]). HGG had a significantly larger PTBE size than LGG (mean ± standard deviation [SD], 2.22 ± 0.54 cm vs. 1.00 ± 0.45 cm, P < 0.001). In total, 162 slides of LGG derived from 8 grade II gliomas and 652 slides of HGG derived from 17 grade III gliomas and 13 grade IV gliomas were examined.

The histologic specimens of glioma were almost identical to their radiologic images in morphology. Further analysis revealed that the radiologic tumor size was smaller than its macroscopic size (mean ± SD, 3.59 ± 1.37 cm vs. 3.79 ± 1.33 cm, P < 0.001), but slightly larger than its histologic size (3.59 ± 1.37 cm vs. 3.56 ± 1.29 cm, P = 0.655). On MR imaging, the radiologic size of LGG was 2.95 ± 1.01 cm and of HGG was 3.76 ± 1.42 cm. HGG was larger than LGG, but the difference was not statistically significant (P = 0.140). This difference was also found on macroscopic and histologic measurements (Table 2). However, the most important finding was that comparative analysis of these three measurements revealed a significant correlation between radiologic tumor size and macroscopic tumor size (r = 0.968, P < 0.001) and histology tumor size (r = 0.961, P < 0.001) (Fig. 3). This was also found for enhancing and non-enhancing gliomas (all P < 0.001) (Additional file 3: Fig. S1). These findings demonstrated that it was reasonable to delineate the GTV of the glioma in the MR images.

The mean ME distance was 1.17 cm (range, 0 to 2.80 cm). We observed a significant difference in ME between LGG and HGG (P < 0.001), with a mean of 0.69 cm (range, 0 to 1.80 cm) for LGG, and 1.29 cm (range: 0 to 2.80 cm) for HGG. Further analysis showed that grade IV gliomas had a significantly higher ME than grade III and grade II tumors (mean ± SD, 1.59 ± 0.67 cm vs.1.06 ± 0.51 cm vs. 0.69 ± 0.37 cm, all P < 0.001). Analysis of the ME distribution confirmed the difference between the three histologic types (Tables 3, 4, and 5). Figure 4 presents ME frequency tables classified by increments of 0.25 cm together. Furthermore, we found that MGMT unmethylated tumors had a significantly lower ME than their methylated counterparts (1.07 ± 0.66 cm vs. 1.24 ± 0.61 cm, P < 0.001). In contrast, IDH wild-type tumors had a higher ME than IDH mutated tumors (1.32 ± 0.65 cm vs. 0.93 ± 0.55 cm, P < 0.001). Tumors with 1p/19q non-co-deletion had a higher ME than those with 1p/19q co-deletion (1.24 ± 0.68 cm vs. 1.02 ± 0.50 cm, P < 0.001) (Additional file 4: Fig. S2). In paired analysis, tumors that had all three alterations (i.e., MGMT methylated plus IDH wild-type plus 1p/19q non-co-deleted [MGMT_m/IDH_wt/non-codel]) showed a significantly higher ME than the other counterparts (all P < 0.05) (Additional file 5: Fig. S3).

Figure 2 and Additional file 6: Table S2 show the invasion mode between different grades of gliomas. Direct extension (68%) was the most frequent pattern observed for all groups, followed by perineural spread (12%) and subpial growth (11%). Furthermore, compared with LGG, more HGG tumor cells infiltrated through perineural dissemination (13% vs. 4%, P = 0.001) and subpial growth (13% vs. 5%, P = 0.006).