We identified patients diagnosed between January 1997 and May 2018 with non-contrast-enhancing gliomas at the University Medical Center Mannheim. From this patient cohort, we identified 24 patients with partial resections or lesions from stereotactic biopsies that were non-contrast-enhancing, T2-FLAIR (T2-weighted fluid-attenuated inversion recovery) hyperintensive and that progressed to a secondary contrast-enhancing tumor (Figures 1A,B; Supplementary Figure S1). We did not find patients with oligodendrogliomas and decided to exclude them from this study, since tumor microtubes are rarely found in this entity.

Because of the long time span of our study, molecular sample analysis was only partly available for the patients. All tumor entities included in the study are known to extend long membrane tubes and form a functional communicating multicellular network. The diagnosis of astrocytoma was confirmed in all cases, using histological criteria, and immunohistochemical staining for IDH1-R132H, with IDH mutation found in 11/24 cases. ATRX nuclear staining was performed in all patients, too, and an ATRX loss was confirmed in 21/24 patients. Loss of ATRX expression in the presence of an IDH mutation generally excludes 1p/19q loss of heterozygosity and therefore a diagnosis of oligodendroglioma (which are generally IDH mutated) (20, 21).

The study protocol conformed to the ethical guidelines of the Declaration of Helsinki and was approved by the institutional ethical review board of Heidelberg University (2018-843R-MA and 2018-614N-MA).

All patients included in the present analysis received early postoperative MRI within 72 h after surgery to determine the extent of resection. Remaining postoperative T2-FLAIR hyperintensity was considered residual tumor. Evaluation of the extent of resection was performed by comparing segmented pre- and postoperative T2-FLAIR hyperintensity images side by side (Brainlab Elements, Brainlab AG, Munich, Germany). We defined partial resection with a cut-off level of ≤95% based on the T2-FLAIR images.

Malignization was defined as new contrast enhancement during follow-up MRIs. We stratified new contrast enhancement into defined recurrence patterns based on their absolute distances from the resection cavity: regional (in the wall of the resection cavity), marginal (≤20 mm of the resection cavity) and distant (>20 mm from the resection cavity); The absolute distance was measured from the border of the contrast enhancement to the resection cavity in either the axial, coronal or sagittal plane, in whichever plane the distance was shortest. A “multiple” category was also generated to designate the incidence of more than one focus of contrast enhancements—in such cases, the contrast enhancement nearest to the resection cavity was measured (Table 1). To account for the respective variations in residual tumor volume among the 21 patients that developed a new contrast enhancement within the residual tumor, we also described the distance between contrast enhancement and the resection cavity in relationship to the size of the residual tumor (Table 1). Therefore, contrast enhancement distances are described relatively, as a ratio of their distance from the resection cavity to the length of the residual tumor within the same MRI plane.

Histological reassessment was performed if and when patients underwent another surgery after development of a contrast enhancement. The primary endpoint within this study was onset of a new contrast enhancement; the secondary endpoint was overall survival.

The Department of Neuropathology at Heidelberg University provided immunohistochemical staining of paraffin-embedded tissue sections, using DAB to reveal the respective primary antibodies to IDH1-R132H (DIA-H09, Dianova, RRID: AB_2335716), ATRX (HPA 001906, Sigma-Aldrich, RRID: AB_1078249) and Ki67 (ab15580, Abcam, RRID: AB_443209) and using hematoxylin as counterstain. To analyze stained sections, these were first digitized and converted into RGB TIFF files with an Axio Scan.ZI slide scanner controlled by ZEN software (Zeiss).

Quantification of IDH1-R132H-positive staining was performed computationally with a Fiji image-analysis software (22) macro using three representative 250 µm-square regions per tumor sample. Specifically, IDH1-R132H staining was separated from hematoxylin staining with the color deconvolution plugin using the built-in “H DAB” vector settings. This separated DAB-only layer represented overall IDH1-R132H staining. Then an overlay of quantifiable false-colored pixels (binary—stained vs. unstained pixels) was created by applying the default iterative threshold algorithm using a minimum-threshold value set to zero and a maximum-threshold value set to an 8-bit number (0–255) in which visually stained tissue was optimally segmented from unstained tissue; the sum of the pixels from this overlay quantified total staining. A second overlay from the DAB-only layer was created in which cell body staining was segmented from cytoplasmic tumor microtube density, again using the default threshold algorithm and zero minimum threshold, but adjusting the maximum threshold value to an 8-bit number in which only cell bodies were visually segmented; the sum of the pixels from this overlay quantified cell body staining. The difference between the sums of the two overlays gave the value for net tumor microtube network density. To compensate for variations in the number of tumor cells per region, cell bodies were counted and then the average net tumor microtube network density per tumor cell was calculated by dividing net tumor microtube network density by the total number of cell bodies. The results of the three representative regions were averaged to compensate for sample variation.

This entire process of quantifying net tumor microtube density was performed and evaluated independently by two individuals (SB, MR) to offset inter-rater variability. See Supplementary Figure S2 for an illustration of the quantification process.

Ki67 is a proliferation marker that we used to compare the malignization of tumors between sequential resections.

Analyses were conducted using SigmaPlot 14.0 (SYSTAT). Two-by-two tables were constructed and analyzed using Fisher’s exact test to test if there is statistical significance that patients without adjuvant therapy mostly showed new contrast uptake regional or marginal. Using the two-tailed Mann-Whitney U test, p-values were determined from changes in Ki67 relative count and IDH1-R132H net tumor microtube network density during the progression of the disease. Statistical significance was set to p < 0.05.

As a result of our inclusion and exclusion criteria, out of 127 patients with WHO grade 1–3 gliomas, ultimately 24 patients were included (Figure 1A), of which 17 patients underwent partial resections and seven patients received stereotactic biopsies only of a non-contrast-enhancing astrocytoma (Table 1). Twenty-one patients were previously diagnosed as lower-grade astrocytoma (WHO grade 2–3). One additional patient was initially diagnosed under guidance of 2007 WHO classification as “oligoastrocytoma”, but has since been revised to astrocytoma (MA17; Figure 4A). Two patients initially did not reveal a clear pathological diagnosis after stereotactic biopsy of a non-contrast-enhancing T2-FLAIR hyperintensive lesion.

Patients that underwent open surgery with planned partial tumor resection received a median 42 ± 27 (s.d.) percent tumor resection. The median follow-up period for the patients was 68.5 ± 56.9 (s.d.) months (range: 12–193 months) (Table 1).

A novel contrast enhancement was seen during follow-up at a median contrast enhancement onset of 35 ± 30 (s.d.) months (range: 1–137 months) (Table 1). Patients developing an early, new contrast enhancement did not undergo immediate re-resection until a follow-up MRI confirmed suspected contrast enhancement progression. Seventeen patients underwent a re-resection 29 ± 24 (s.d.) months after their first surgery (range: 1–81 months). Pathologists concluded that samples from all seventeen re-resections were derived from solid tumors that had increased at least one WHO classification grade (Figure 4A); pseudoprogression was not diagnosed in any of the cases.

We reasoned that if the remaining tumor load solely accounts for recurrence and malignization, the distribution of new contrast enhancement within the residual tumor regions of patients would be stochastic (Figure 3A). Alternatively, if surgical lesions instigate secondary malignization, contrast enhancement would develop predominantly within the direct vicinity of the resection cavity (Figure 3B).

The follow-up MRIs indicated that, with only a few exceptions, new contrast enhancement emerged from within residual tumor regions (21/24 patients, 87.5%) (Figures 1B, 2; Supplementary Figure S1). The distribution of new contrast enhancement within residual tumor regions was not stochastic, with the majority of new contrast enhancement clearly developing at the resection cavity (Figure 2; right x-axis, Figure 3C). Our measurements revealed that new contrast enhancement occurred directly adjacent to the resection cavity in 12 patients (57.1%) and within 1 cm of the resection cavity in six patients (28.6%) (Figures 2, 3C; Table 1).

In sum, 11 new contrast enhancements were regional, seven marginal, three distant and three were multiple. Among the multiple contrast enhancement recurrence pattern cases, patient MA02 developed two contrast enhancement foci within the residual tumor and both fell into the marginal class. Patient MA15 developed four contrast enhancement foci: three were marginal and within the RT; one was distant. Patient MA19 developed two contrast enhancement foci: one CE was regional and within the residual tumor; one was distant, likely due to subependymal spread. (Supplementary Figure S1; in cases of multiple contrast enhancements, only MRI data of the nearest contrast enhancement are shown).

Considering the size of the remaining tumor at index surgery the measurement of relative distances to the resection cavity (Table 1) demonstrated, that 13 patients developed contrast enhancement at a relative distance of close to zero, seven patients developed contrast enhancement within half the distance of the residual tumor, while only one developed contrast enhancement within more than half the distance of the residual tumor (Figure 2; right x-axis).

In only three cases from our patient cohort (12.5%) contrast enhancement developed elsewhere in the brain, outside of the previously T2-FLAIR hyperintensive region. The contrast enhancement in these cases was located 10–30 mm from the resection cavity (Figure 2; left x-axis, Table 1).

In the context of tumor recurrence, within the group of patients with distant and multiple contrast enhancements, four out of those six received additional adjuvant therapy, which has been shown as a trigger of tumor self-repair in preclinical studies (13, 14). Patients without adjuvant therapy mostly developed new contrast uptake regionally or marginally (9 of 10 cases); however, this observation did not reach statistical significance (p = 0.36).

We detected an increase in Ki67 expression from initial to final resections in 14/17 patient tumors (Figure 4B), which was in line with their histological and radiological malignization (Figure 4A). Between first and second resections, fold change in Ki67 positivity was a median of 1.6 (range: 0.2–3.8); between first and third resections, fold change was a median of 5.0 (range: 0.4–11.0, p = 0.008) (Figure 4D).

Eleven patient tumors harbored a IDH1-R132H mutation (Figure 4A), which allowed for a tumor cell-specific staining of tumor microtubes and their cell bodies (14, 16, 23) (Figure 4F). Nine of these 11 patients had more than one resection; we found that in all nine patient tumors, total tumor microtube area increased between sequential resections (Figures 4C, F). In relation to first resections, total tumor microtube area increased 1.36-fold in second resections (0.95–1.96) (p = 0.001) and 1.11-fold in third resections (1.08–2.16) (p ≤ 0.001), respectively (Figure 4E).