Morphologic characteristics of gcs within GB, that is, their enlarged shape, together with literature review describing increased incidence of TP53 mutations and double strand breaks in this morphologic GB phenotype, led us to the investigation of aging/senescent programs within GB. As we concentrated on DNA methylation analysis, this question was majorly addressed by use of methylclocks. We found gcGB to have an accelerated DNAMethAgeAcc in most epigenetic clocks tested compared to non-gcGB and pediatric GB.

By definition and by construction, methylclocks intend to estimate chronological ages. Cancer-related methylomes are distinct from healthy tissues’ and imply tumor-specific alterations which raises the basic question whether age estimates can be drawn from tumor tissue at all. Perez et al. looked closer into DNA methylation changes associated with aging, cancer or shared between both conditions[30]. The roughly 2000 450 k Human Methylation datasets analyses showed higher variability on a global level between healthy and malignant states than older or younger age; however, aging was rather associated with a hypermethylation in a tissue-dependent manner, whereas the cancer methylome, including also glial neoplasms, displayed both hyper- and hypomethylated CpGs without a clear trend for one DNA modification. The occurrence of shared hypermethylated sites between aging and cancer was higher than expected by chance and mainly allocated to genes related to developmental processes, but CpG hypomethylation associated with both tumor and age was majorly not found[30]. The study included brain tumors also, showing a correlation coefficient of 0.97 between chronological and epigenetic (Horvath clock-based) ages for healthy tissue which dropped to 0.33 among malignant samples. Although cancer is supposed to reprogram the epigenetic clock as hypothesized by Horvath, it remains questionable whether the findings would have been the same, when brain tumor entities had not been mixed but considered separately[18]. This is especially important if globally hypomethylated, onco-histone mutant gliomas or hypermethylated IDH mutant gliomas are considered. Within the light of gcGB and non-gcGB being indistinguishable from each other on global DNA methylation analysis as well as reference-free tumor deconvolution approaches, the differences in DNAMethAgeAcc become even more striking.

We observed epigenetic ages to be lower in non-gcGB and higher in gcGB which leads to the shift of DNAMethAgeAcc between them. In order to understand whether either the non-gcGB epigenetic reprograming leads to younger tumor cells with respect to patient age and gcGB ages non aberrantly, or gcGB reprogramming lets tumor cells age quicker, we included methylomes of the pediatric GB subtype. Taken global DNA methylation patterns into account, adult and pediatric GBs were shown to cluster separately, indicating distinct changes of the methylome which are tumor-specific and thus allow for distinction into different methylation classes[4,5,6, 31]. Most intriguingly, when compared with other childhood CNS tumors, pediatric GBs rather cluster with formerly PNETs and medulloblastoma than adult GBs. This rather argues for an underlying tumor age-related factor inscribed into the methylome than an introduced noise due to healthy resident cells as the pediatric GB molecular subclass also allots to adult patients, represented in our study cohort by age ranges of 41 until 73 years. By comparison of DNAMethAgeAcc between those three subgroups of IDHwt GB, we saw, that the age acceleration gap became even bigger than in the comparison between gcGB and non-gcGB arguing for a boosted aging in gcGB. This is even more striking in light of pediatric GB showing some similarities to gcGB, as it is known to harbor TP53 alterations in up to 56% and an accumulation of DNA double-strand breaks[5, 32, 33].

The algorithms underlying methylclocks weigh CpG site methylation states with relation to age[18, 24, 29]. Nevertheless, the gcGB cohort in this study was significantly younger than the non-gcGB cohort, thus supposedly introducing a systematic bias when chronological ages were subtracted from epigenetic ones to compute DNAMethAgeAcc. Analyzing gliomas of different WHO grades and IDH mutation status by use of the Horvath clock and EpiTOC, a methylclock considering a tissue’s mitotic activity, Liao et al. found tumors of chronologically older patients to harbor higher epigenetic ages[34]; our results, however, were reproducible upon age-adjustment of the non-gcGB study subcohort to fit gcGB patient ages. Further arguing against an introduced age bias through cohort selection is the fact that epigenetic programs in gcGB must have accidently and significantly altered more CpG sites interrogated by the methylclock algorithms than the one in non-gc, which is rather less likely, especially as the CpG sites called in each methylclock do largely not overlap. Nevertheless, methylclock results on glioma dataset were shown to correlate and further be able to distinguish glioma subtypes from each other pointing at differential signatures of epigenetic aging[34]. GcGBs might therefore constitute a subgroup within GB with a distinct aging signature on an epigenetic level. In GB, giant cell enrichment was described to co-occur with defective DNA mismatch repair [10, 12]. Although the accumulation of somatic mutations in DNA mismatch repair deficiency was not strictly linked to giant cell accumulation in GB, and hypermutation with > 10 mutations/Mb found in up to 41% of gcGBs was independent of gc proportions within GB, this co-occurrence questions gcGb-specific genomic alterations potentially promoting epigenetic aging [12, 35]. In addition to mutations of genes related with mismatch repair deficiency gcGB was found to harbor higher mutational frequencies of RB1, NF1, TP53 and ATRX genes in comparison with conventional glioblastoma, whereas EGFR alteration occurred less frequent [12]. Beyond the Rb/p16inka and p53/p21 pathways hijacked in both neoplasia and senescence, the overlap between altered genes, typically associated with senescence and the gcGB-specific mutational landscape altogether is rather small [17, 23, 36]. Still, a one-to-one translation from cellular signaling patterns related to senescence to mechanisms fostering epigenetic aging is not feasible. With the methylclock’s ticking being related to physiological organ aging and cellular differentiation which per se do not succumb to patterns of genomic mutations but follow a trend of leveling DNA hyper- and hypomethylation, the association of epigenetic aging and genomic alterations is unassertive [18, 19]. There are, however, some genomic alterations which have been specifically linked to epigenetic aging. In breast carcinoma, steroid receptor mutations sped-up epigenetic aging as did TP53 mutations in many cancer entities but GB, where the latter were shown to have the contrary effect [18]. Along that line, a negative correlation between the mutational count and epigenetic age acceleration was described, which adds to the complexity of the gc-enriched GB phenotype [18]. Nevertheless, the lack of information on somatic mutational counts along with defects of DNA repair mechanisms especially in the gcGB subcohort are limitations of this study. In addition, a site-specific quantification of changes of methylation in gcGB versus non-gcGB with respect to CpG sites called by the methylclocks as well as a correlative analysis between cohorts targeting senescence-associated chromatin remodeling stretches beyond the scope of this study.

When on a cellular level aging is considered as gradual loss of function, it is tempting to speculate that in the case of gcGB this might translate into more inert, less invasive tumor cells[23]. In fact, many studies described gcGB to invade the surrounding brain parenchyma only minimally and although tumor size and location were not found to differ significantly from non-gcGB, gcGB patients received more extensive resections, which might indirectly also point toward a less invasive behavior[7, 10]. Corroborating this, upon reference-based deconvolution of methylomes, we found significantly lower amounts of neurons within the gcGB tumor bulks than that of the non-gcGB. There are some indications in literature that multinucleation blocks cell proliferation. Fujita et al. for example assign a more passive role to cells with a multinucleated than a mononucleated phenotype in GB[14]. Differential staining patterns of phosphorylation-specific antibodies to track tumor cell cycle stages showed that multinucleated cells got stuck in early mitotic phases[14, 37]. Another study about the effects of multinucleation on cell cycle progression showed a disruption before entering S phase. Hart et al. observed multinucleated cells to arrest in G1 phase even in a p53-compromised setting which was associated with p21 accumulation and absence of PCNA foci as indicators of S phase, thereby pointing toward the initiation of cellular stress programs which prevented the resumption of cell cycling[13]. Most interestingly, those cells were still viable and showed transcription in regions of the genome, not subjected to severe double strand breaks which is reminiscent of the senescent cell state with continued paracrine and transcriptional activity but exit from cell cycling[13, 38, 39]. Furthermore, we did not detect different staining frequencies of p21 when considering whole tumor sections in the study subcohorts. Besides the fact that cellular senescence and epigenetic aging must not be equaled, increased nuclear accumulation of p21 might not be a very reliable marker of senescent cells, as it might be difficult to detect by use of immunohistochemistry because of its heterogenous expression through compartments within the multinucleated cells and its verifiability also in quiescent cells[13, 17]. Stainings with the senescence marker beta-Galaktosidase would be an important add-on for future studies on gcGB to improve our understanding of quiescent versus senescence states in these cells.

A cell can enter the senescence state upon any oncogenic stress in order to prevent malignant degeneration. The consequent growth arrest is a point of no return unless its gatekeepers, and the p53/p21 and Rb/CDKN2A pathways are enabled. An activated senescence program elicits changes on chromatin level, known as senescence-associated heterochromatin foci, but stretched also beyond the single cell level[40]. Senescent cells with impaired genomes or epigenomes have been shown to secret cytokines, growth factors, proteases, and chemokines, summarized under the term SASP which acts on the microenvironmental level and is supposed to also hold back cancer establishment[23]. In our reference- and methylation-based deconvolution of gcGB versus non-gcGB tumor bulk we found lower proportions of CD4+ T cells in the gcGB microenvironment. This is counterintuitive on one hand, because SASP can promote leukocyte infiltration thereby leading to an apathogenic, low-level chronic inflammation, but on the other hand might reflect an age-dependent drop in the overall CD4+ T cell population in addition to increasing levels of IL-6 associated with aging and their negative effect on CD4-mediated antitumor response within an assumptive aged/senescent gcGB microenvironment[41,42,43]. Surprisingly, the cellular composition of the gcGB microenvironment did not overlap with the one in silico deconvolution approaches computed for tumors with increased DNAMethAgeAcc. This might be associated with the samples’ allocation to DNA methylation subclasses as these were repeatedly found to show different levels of immune cell infiltration, especially for T cells[44]. In line with Jeanmougin et al. delineating a beneficial prognostic impact when higher proportions of immune cells are present within the GB microenvironment, our tumor deconvolution analyses point toward more infiltration of B cells, NK cells and T regulatory cells into the microenvironment of the more favorable tumors with increased DNAMethAgeAcc[45]; Lower proportions of CD8-positive T cells in the latter are nevertheless unintelligible, but might argue for the importance of functional status of T cell subsets to exert their tumor-suppressive role thereby questioning the accuracy of reference-based deconvolution in terms of immune cell subtype distinction[46]. Adding to the need of approaches with strong discriminatory power between cell types, the accuracy and reliability of in silico-deconvolution by use of DNA methylation were shown to largely depend on concordance between the samples and the reference matrix used for deconvolution both in terms of tissue type and analysis technique as well as bioinformatical pipelines to assure overall high site-specific and genome-wide assay quality [47]. Future studies with multimodal approaches to investigate the microenvironment of epigenetically more or less aged GBs are therefore required for validation and represent a limitation of this study.

As dysfunctional mitochondria count for a hallmark of aged/senescent cells, we assessed DNA methylation patterns of nuclear coding mitochondrial genes together with genes associated with metabolism in general[17]. Our data contradict differential metabolic, epigenetically controlled programs between gcGB and non-gcGB. On one hand, our findings on protein level were not consistent as we admittedly observed increased staining frequency and intensity against MT-CO2 but could not corroborate a supposed accumulation of mitochondria indicative of a senescent phenotype by use of other antibodies. On the other hand, increased MT-C02 scores were characteristic of gcGBs only and not found in prognostically more beneficial tumors with increased DNAMethAgeAcc which might subtly purport the involvement of mitochondrial processes in the gcGB phenotype. Other techniques than immunohistochemistry should be considered to answer these questions in gcGB, because it neither captures structural changes of senescent mitochondria, nor their functionality or accurate number[17, 48, 49].

The gcGB phenotype consolidates nevertheless some cellular and molecular characteristics which support the hypothesis of gcGB being the “older” variant among IDHwt GB. Besides its tumor cells’ morphology, it was shown to accumulate DNA double strand breaks, involves lower invasion capability and harbors increased DNAMethAgeAcc metrics which all renders an activation of senescence programs in gcGB conceivable[10]. Yet, this hypothesis still awaits its final acceptance or rejection which can only be achieved by careful implementation of multilayered data from the ex vivo analysis of senescence-associated beta-Galactosidase staining patterns, cell cycling and proliferation parameters and structural as well as molecular properties of the nucleus bearing in mind that quiescent cells might cause false-positive results[17, 48].

Together with the fact that we saw boosted DNAMethAgeAcc in gcGB, increased DNAMethAgeAcc was significantly and independently associated with better patient survival. In addition to serving as prognostic biomarker, an increased DNAMethAgeAcc might point at distinct cellular programs worth enhancing in malignant gliomas. This is in line with the comprehensive analysis of IDH wild-type GB datasets by Bady et al., who proved the prognostic importance of a sped-up DNAMethAgeAcc by use of HumanMethylation 450 k data and the Horvath methylclock[50]. In this study, increased age acceleration was associated with the RTK II molecular subtype of GB rather than with RTK I or mesenchymal subclasses; while it was additionally shown to be independent of copy number variations and tumor purity methylclock-associated CpG sites overlapped significantly with those called for assignment of molecular subclasses[50]. These findings highlight the additive value of tumor-intrinsic epigenetic aging as clinical biomarker and argue for further studies to explore these mechanisms’ overlap with cellular senescence pathways. Whether those pathways may be targeted therapeutically to guide malignant glioma toward a biological dead end and giant cells might foster that development in GB warrants future examination.