As shown in the flow diagram of Figure 1 , patients were screened from a retrospective cohort of 1299 continuous patients with suspected gliomas admitted into Beijing Tiantan Hospital between January 2013 and January 2017, according to the following criteria: (a) without surgical treatment before; (b) single lesion; (c) pathologically confirmed diffuse gliomas; (d) supratentorial locations and not confined within rare locations such as the optic pathway; (e) MRI within 7 days before surgery; (f) adult patients (age ≥ 18).

This study was approved by Beijing Tiantan Hospital institutional review board. Written informed consent was waived for the retrospective nature.

To describe the locations of lesions, we divided firstly, based on the international classification of diseases (ICD-11, https://www.who.int/standards/classifications/classification-of-diseases), the supratentorial brain matter into eight anatomic sites of the “lobular” level (referred to as the Regions), including the frontal, the temporal, the parietal, the occipital and the insulae, as well as those nuclei or white matter areas of the basal ganglia, the corpus callosum, and the thalami, which could not be assigned to the spectral of any single lobe. Besides, four eloquent areas of clear margins (the pre- and postcentral gyrus, the gyrus cinguli, and the hippocampus, referred to as the Subregions) were also considered. Each Region or Subregion was also divided into the left and right ones (or the half for the corpus callosum), taking the anatomic midline of each axial slice as a reference.

An original atlas was manually segmented into those anatomic sites based on MNI-152 (17), the widely recognized standard brain template, in 3D-slicer 4.13.0 (www.slicer.org). The Desikan-Killiany Atlas (18) was taken as a reference for the division of the cortical areas. The white matter was segmented and adjusted in every slice of the axial, sagittal, and coronal views. The volume of each anatomic site was then measured based on this atlas.

Two neurosurgeons (> 10 years of experience, ZH and ZL) reviewed the MRIs of each patient separately to specify the tumors’ locations by anatomic sites. The senior neurosurgeon (>30 years of experience, JX) made the final decision when there were disagreements. For lesions without evident enhancement, the margins were determined by axial T2/FLAIR images. Meanwhile, post-contrast T1 images (in sagittal and coronal views) were also referred to. And for lesions with evident enhancement, post-contrast T1 images in the three views determined the margins together. A lesion can be defined as invading more than one site: taking the natural anatomic margins as references, topologic areas outside the core of the bulk were identified as affected only when the scopes of lesions invaded across those borders exactly. Otherwise, they were not affected when the anatomic margins shifted due to the mass effect without clear lesional signs within the spectra of target sites. For gliomas with Subregions affected, corresponding Regions were also identified as affected.

Furthermore, the number of Regions invaded was also counted (neglecting the lateral, for instance, lesions in bilateral frontal lobes and corpus callosum invaded two Regions: the frontal and the corpus callosum) for each patient. For tumors invading more than one Region (referred to as Multi-regional tumors, and Localized ones for their counterpart), they were regarded as lesions in different locations separately unless otherwise specified. And to evaluate the trends from anterior to posterior, the frontal, motor region, sensory region, parietal, and occipital were indexed as 1 to 5.

Pathological data were extracted from each patient’s medical records, including histological grades and molecular subtypes (if any) of isocitrate dehydrogenase (IDH) mutation, 1p/19q-codeletion, O(6)-methylguanine DNA methyltransferase (MGMT) methylation, as well as telomerase reverse transcriptase promoter (pTERT) mutation. And for histological grades, lesions were further divided into subtypes of glioblastomas (GBMs, WHO IV) and lower-grade gliomas (LGGs, WHO II or III). Moreover, molecular subtypes were divided as mutant type (-mt) or wildtype (-wt) for IDH and pTERT, co-deleted (-cd) or non-co-deleted (-ncd) for 1p/19q, and methylated (-mt) or unmethylated (-wt) MGMT, respectively. And subtypes of 1p/19q-cd were only considered in those patients of IDH-mt.

The volume varies widely among locations. Larger areas like the frontal should be more vulnerable even if there were a uniform distribution. Therefore, we calculated the frequency density of each site by adjusting the invaded ratios with its volume. Moreover, to reduce the volume measurement error, it was corrected to two decimal places at the unit of 10^5 mm3. The frequency density was compared among locations to reveal the actual sites of predilection.

All the statistical tests were performed using Scipy 1.8.0 (https://scipy.org/) in Python 3.10.2. The invaded ratios and constitutions of pathological subtypes were compared among sites by chi-square tests, while one-tailed t-tests were used to distinguish the age gaps among sites and between pathological subtypes. Mann-Whitney U test and Spearman’s rank correlation analysis were also applied in this study and are noted otherwise. P < 0.05 was considered statistically significant. The region-based lesion maps illustrating the probability of adult supratentorial diffused gliomas were generated using Nibabel 4.0.2 (https://nipy.org/packages/nibabel/index.html). And tree-maps of pathological constitutions were drawn using Plotly 5.6.0 (https://github.com/plotly).

990 adult patients with newly-diagnosed supratentorial diffuse gliomas were finally enrolled in this study. They all had routinely presurgical T2/Flair and post-contrast MRI images acquired within 7 days before the surgery. Histological data were available for every patient, while some had molecular data and others not, as shown in Table 1 .

Age gaps were there between different pathological subtypes: patients with glioblastoma were older than those with LGGs (p < 0.0001), meanwhile, patients with IDH -wt are older (p < 0.0001), as well as those of pTERT-mt (p < 0.0001). No significant sex gap was found between all the pathological indicators above.

For these patients, 531 left hemispheres were affected, while 512 neoplasms invaded the counterpart, among which 53 were bilateral lesions and 56.77% were Localized tumors.

As shown in Figure 2 and detailed in Supplementary Table S1 , the frequencies of invasion were in descending order from the frontal, temporal, and insula to other sites ( Figure 2A ). But, isolating the effect of anatomic volume, the insulae, hippocampi, and corpus callosum became the top 3 locations of predilection ( Figure 2B ), which were stable in almost every sub-groups of Localized ( Figure 2C ) or Multiregional lesions ( Figure 2D ) and LGGs ( Figure 2E ) or GBMs ( Figure 2F ). Taking the motor areas and sensory areas as references, a descending trend of the occurrence dense from the anterior to the posterior (frontal → motor area → sensory area → parietal → occipital) stood out. The temporal, parietal, occipital, insulae, corpus callosum, and basal ganglia were all Regions that tended to harbor more Multi-regional lesions with p-values less than 0.0001, similarly Gyri cinguli (p < 0.0001), and the left hippocampus (p = 0.0136) as Subregions. And Mann-Whitney U tests revealed that among Multi-regional lesions, those with the frontal (p = 0.0020), parietal (p = 0.0287), or occipital (p = 0.0384) in left hemispheres as well as the temporal (p < 0.0001), insulae (p < 0.0001), basal ganglia (p < 0.0001) or thalami (p = 0.0003) in either hemisphere tended to invade more Regions.

Besides those regarding pathological subtypes, age gaps were also found among sites. Patients with affected thalamus in the left (p = 0.0021) or affected frontal (p < 0.0001) were younger, while those with affected temporal (p = 0.0001), parietal (p < 0.0001), or occipital (p = 0.0008) were older. In addition, patients who harbored Localized lesions were younger than those Multi-regional victims (p = 0.0003). Though there were not any significant differences in age found for those with invaded insula, the left (p = 0.0186) and the right (p = 0.0054) affected ones were younger in subgroups of Multi-regional and Localized tumors, respectively. Spearman’s correlation also showed an ascending trend of age from anterior to posterior ( Table 2 , p < 0.0001), but no significant correlation between age and the number of invaded regions (p = 0.8199).

Thanks to the large base, the frontal accounted for most in nearly every pathological subtype considered in this study. And as shown in Figure 3 , pathological factors varied in many aspects. Histologically, for frontal (p < 0.0001) and insulae (p < 0.0001), the proportions of LGGs were larger, which were smaller for the thalamus (p = 0.0009), temporal (p = 0.0141), and basal ganglion (p = 0.0064) in the right, and the parietal (p < 0.0001), occipital (p < 0.0001) lobes ( Figure 3A ). Spearman’s rank correlation analysis revealed an ascending trend of the grade from the anterior to the posterior ( Table 2 , frontal - motor region - sensory region - parietal – occipital, p < 0.0001).

Regarding the molecular status, differences were also there. Firstly, in the frontal (p < 0.0001), IDH-mt lesions were more than IDH-wt tumors, while in the parietal (p < 0.0001) and occipital (p = 0.0002), opposite results were there ( Figure 3B ). And in the insulae, we found a nearly significant p-value of 0.0501 (p = 0.0126, neglecting the factor of lateralization) for the right one, which suggested more lesions were IDH-mt, while insignificant for the left one (p = 0.2856). Furthermore, Spearman’s rank correlation analysis also revealed an ascending trend of the proportion of IDH-wt from the anterior to the posterior (frontal - motor region - sensory region - parietal – occipital, p < 0.0001). Secondly, for 1p/19q, we found nearly significant p-values for tumors invading the frontal (p = 0.0793) or the motor region (p = 0.0624) in the left, which suggested that there might be more 1p/19q-cd tumors and became significant neglecting the factor of lateralization (p = 0.0082 and p = 0.0186 respectively, Figure 3C ). Next, more of the lesions with the right basal ganglion invaded are MGMT-wt (p = 0.0007), and for GBMs, more of the ones invaded the right temporal were also MGMT-wt (p = 0.0181). No significant differences between MGMT subtypes were found for the left temporal. Finally, left frontal gliomas were more likely to be pTERT-wt (p = 0.0259), while those right frontal (p = 0.0185) and right temporal ones (p = 0.0214) were more of pTERT-mt in LGGs and HGGs respectively.

In addition, none of the pathological subtypes was found to invade significantly more Regions than their counterpart. But concerning these subgroups of Localized lesions only, different from the entirety, the left temporal lobes (p = 0.0180) and the hippocampus (neglecting the lateral factors, p = 0.0058) harbored more HGGs than LGGs.