We found that a substantial proportion (46%) of the assessed histopathological and immunohistochemical features were significantly associated with the amount of viable tumor material on HE slides. All significantly associated features were found to be less present or of a lesser degree in cases with a smaller amount of tissue. Several of the significant features are relevant for the grading of diffuse astrocytic tumors, i.e., small necroses, palisades, microvascular proliferation, atypia, mitotic count, and Ki-67/MIB-1 PI. We also found that “sparse” tissue amount was strongly associated with a smaller tissue volume sent for neuropathology, indicating that neurosurgical sampling impacts the histology. Interestingly, “sparse” tissue amount was commonly obtained from surgical resections, where it presumably would be possible to provide larger or more tumor samples. Our results show that several of the histopathological features in GBMs are heterogeneously distributed, which limits the histological representativeness of small tissue samples. These findings underline the importance of adequate tissue collection to increase diagnostic accuracy and quality of histological research.
Previous studies have demonstrated the risk of histological undergrading of GBMs on small tissue samples [
4,
8,
11,
12,
16,
19,
25,
26,
29,
36,
45]. In contrast to our study, these studies were focused on grading, whereas our study assessed the representativeness of individual histological features. These previous studies also only focused on biopsied cases [
4,
8,
11,
16,
25,
26,
29,
36,
45] or on the volume of the pathological specimen sent for analysis [
12,
19]. Hence, the role of sampling errors in resected tumors is less studied. Moreover, none of the previous studies accounted for the presence of necrosis in the material, which is likely to cause a further decrease of the histological representativeness. As most of the histological features assessed in this study are only found in the viable tumor tissue, their representativeness is more precisely estimated by quantifying the area of the
viable tumor as the tissue amount.
Grading features
Several of the hallmark features of GBMs, small necroses, palisades, and microvascular proliferation, were significantly less present in cases with “sparse” tumor material. The significant associations suggest that these features are heterogeneously distributed, which limit their representativeness in small tissue samples. The only hallmark feature that was not significantly dependent on tissue amount was large, ischemic necrosis. This feature was found in a high proportion in all tissue categories, suggesting that it is a very frequent and homogenously distributed feature in GBMs. The fact that 90% of the cases (44 cases) with “sparse” material had large necrosis indicates that most of these cases were never at risk of being undergraded despite the scant amount of viable tissue. However, the diagnosis of the five remaining cases without large necrosis relied solely on the presence of the other hallmark features shown to be at risk of underrepresentation. Still, all five had visible necrosis on the preoperative MRI scan (data not shown) and would therefore have been treated as GBMs by many institutions, because it has been shown that lower grade astrocytomas with radiological necrosis exhibit comparable survival to GBMs [
22]. Hence, our study highlights the importance of considering clinical and neuroradiological information in glioma diagnostics due to the risk of histological undergrading of small tissue samples.
In addition to the hallmark features, other features relevant for grading of diffuse astrocytic tumors, mitotic count, atypia, and Ki-67/MIB-1 PI, were also significantly associated with tissue amount. Cellular density was not significantly associated; however, there was a near-significant trend that cases with “sparse” tissue were more often categorized as “mild” and less often as “high.” Both mitotic count and Ki-67/MIB-1 PI were significantly higher in the “medium” versus “sparse” tissue categories, but neither was significantly different between the “medium” and the “substantial” categories. These findings are in accordance with the known regional heterogeneity of proliferative cells [
9,
32,
33] and highlight the limitation of sampling errors in proliferative quantifications of GBMs. Interestingly, both atypia and cellular density were only categorized as “low” or “mild” in cases with “sparse” tissue amount, which suggests that these “sparse” samples might have been taken from infiltration zones of the tumor. Our findings are in line with the previous studies showing that GBMs can be histologically undergraded on small tissue samples [
4,
8,
11,
16,
19,
25,
26,
29,
36,
45]. Moreover, it is also likely that some IDH wt grade II and III tumors with molecular features of GBM represent undergraded IDH wt GBMs [
7,
37], as it has been shown that these tumors follow the same clinical course as GBMs [
2,
37,
43]. However, undergrading is a less probable cause when radiology is in accordance with low-grade glioma [
14,
43], and it has been suggested that such tumors may represent early stage GBMs [
14]. Nevertheless, our study is in line with studies indicating that some of the IDH wt diffuse astrocytic gliomas with molecular features of GBM are undergraded IDH wt GBMs.
In this study, we did not assess other molecular parameters than IDH mutation status. However, as mentioned, extensive molecular analyses such as next-generation sequencing and methylation profiling have been shown to be useful tools in glioma diagnostics [
6,
38]. Especially methylation profiling in combination with standard histopathology has shown promising results [
6,
7,
18]. Two prospective studies showed that the use of methylation profiling led to a change in diagnosis in 12% of cases [
6] and in 84% of diagnostically challenging cases [
18]. The latter study also found a substantial clinical benefit of the change in diagnosis [
18]. Unfortunately, intratumoral heterogeneity is also a limitation of the molecular analyses, as studies have found that different molecular GBM subtypes can exist within the same tumor [
31,
47]. However, despite the finding of varying methylation subtypes, all spatially collected biopsies from the same tumor were consistently classified as GBM IDH wt or mt [
47]. Still, methylation profiling is limited when tumor material is scant, illustrated by a large study in which 4% of the patients could not be profiled due to a low tumor cell content [
6]. Other limitations of methylation profiling are the long turnaround time (a median of 25 days in one trial) [
18], and that it is not available to most centers [
2,
37]. Therefore, despite the promising introduction of extensive molecular analyses in glioma grading, the limitation of reduced histological representativeness of small tissue samples is still highly relevant.
Other features
In addition to the abovementioned grading features, hemorrhages, pseudorosettes, subpial clustering, lymphocytic infiltration, small cells, and CD105-MVD were also significantly associated with tissue material. All the features except CD105-MVD were significantly less present in cases with “sparse” material, suggesting that these are heterogeneously distributed features. Regarding CD105-MVD, it was only significantly higher in the “substantial” tissue category than in the two lower categories, which suggests a large degree of heterogeneity in the distribution of vascular hotspots. Despite the well-known observed heterogeneity in the vascular structures on GBMs [
35,
46], the degree of the heterogeneity has been sparsely studied. However, in accordance with our findings, Di Ieva et al. [
10] found a large degree of heterogeneity of the vascularity of GBMs measured by digital pathology.
Thrombosis, perineuronal satellitosis, angiocentric structures, desmoplasia, macrophages, and all the cellular differentiation patterns despite small cells were not significantly associated with tissue amount. The findings indicate that these features are homogenously distributed and less prone to sampling errors. Consequently, these features have potential clinical utility in that their presence could suggest a grade IV diagnosis, given that the features have been found to strongly associate with a GBM diagnosis. Thrombosis is of particular interest, as it has been shown to associate with aggressiveness in diffuse astrocytic tumors [
1,
44] and it has been suggested as a diagnostic criterion of GBM [
34,
42,
44]. One study also found that the presence of thrombosis independently predicted wildtype IDH status, and they therefore suggested screening for thromboses in IDH1-R132H-negative lesions to help decide if additional sequencing of IDH1/2 is worthwhile when resources are limited [
44]. Like thromboses, macrophages have been associated with aggressiveness in gliomas, and the number of macrophages has been found to increase with higher astrocytoma grades [
13,
21]. However, we only recorded distinct macrophages in HE sections (i.e., not immunostained), which predominately were foamy macrophages found at the edge of necroses. Hence, the high frequency of macrophages is probably explained by the widespread presence of necrosis, and the clinical utility of macrophages is therefore limited. Moreover, the clinical utility of the secondary structures of Scherer is limited by their frequent presence in lower grade diffuse astrocytic tumors [
23]. Regarding desmoplasia and the cellular differentiation patters, these are epiphenomena of the aggressive GBM biology. However, these features can also be found in other lower grade gliomas that are relevant differential diagnoses [
24]. In summary, of the non-significant features, only thromboses have promising clinical utility in that their presence in a histologically lower grade IDH wt tumors could indicate that it is an undersampled IDH wt GBM.
Clinical features
Perhaps to no surprise, the tissue amount was significantly associated with the type of surgical procedure, the number of HE slides, and tissue volume. However, the tissue amount was not associated with either total tumor volumes or volumes of the contrast-enhancing compartment on the preoperative T1wGd MRI scans, which suggest that larger tumors and more contrast enhancement did not impact histology. On the other hand, the strong association between tissue amount and tissue volume indicates that neurosurgical sampling affects the histopathology. The same association was also found when biopsied cases were excluded (
p < 0.001, data not shown). Put together with the finding that most of the cases with “sparse” tissue had undergone resections, our data suggest that more tissue could have been retrieved from the resected tumors. Our findings are in agreement with the study by Lasocki et al. [
22], which showed that undergrading also occurred in patients who had resections. Extensive necrosis can also cause a smaller amount of viable tissue, and it is likely the explanation for the relatively large tissue volumes found in the upper range in “sparse” and “medium” tissue categories. Nevertheless, our findings indicate that neurosurgeons should be encouraged to send larger tumor samples to the pathologist to avoid potential histological undergrading.