Linking epigenetic signature and metabolic phenotype in IDH mutant and IDH wildtype diffuse glioma

Changes in metabolism are known to contribute to tumour phenotypes. If and how metabolic alterations in brain tumours contribute to patient outcome is still poorly understood. Epigenetics impact metabolism and mitochondrial function. The aim of this study is a characterisation of metabolic features in molecular subgroups of isocitrate dehydrogenase mutant (IDHmut) and isocitrate dehydrogenase wildtype (IDHwt) gliomas.


| INTRODUC TI ON
Diffuse gliomas include isocitrate dehydrogenase mutant (IDHmut) astrocytic and oligodendroglial tumours, as well as IDHmut and IDH wildtype (IDHwt) glioblastoma (GB). These tumours account for over 60% of primary brain tumours in the adult population in the United States [1]. The combined incidence is 4.2 per 100.000 capita.
This leads to DNA and histone hypermethylation, which ultimately produces a hypermethylation phenotype known as the 'glioma CpG-island methylator phenotype' (G-CIMP) [14,15]. Differences in DNAmethylation patterns allow for artificial intelligence-assisted tumour classification [16]. IDHwt GB has been further subdivided into three molecular subgroups: proneural (PN), classical (CL) and mesenchymal (MES). Initially, these subgroups were defined according to distinct gene expression profiling. In addition, DNA methylation analyses enable subdivision into these subgroups [16,17]. In the context of DNA methylation analyses, the subgroups are referred to as RTK I (PN), RTK II (CL) and mesenchymal.
Metabolic reprogramming is a characteristic hallmark of cancer [25]. Normalisation of deregulated cellular energy metabolism in tumours has been hypothesised to be a beneficial factor [25,26].
Metabolic changes in malignant tumours compared to non-neoplastic tissues were first described almost 100 years ago by the German biologist Otto Warburg [32]. He discovered that tumour cells metabolised glucose to lactate via glycolysis despite the presence of oxygen. This phenomenon of aerobic glycolysis is termed the Warburg effect [33,34]. Alterations in glucose metabolism and mitochondrial respiration in cancer have been intensively studied over recent decades. Their contribution to malignant transformation or progression across different cancer entities is undisputed. However, the extent to which metabolic alterations are necessary to sustain the malignant phenotype and the stage of tumour progression at which they occur are unclear. Examining these mechanisms will help to understand the unsuccessful therapeutic attempts of metabolic normalisation in gliomas.
To gain further insights into metabolic regulation of gliomas, we investigated the DNA methylome profiles of IDHmut and IDHwt gliomas focussing on glucose metabolism and mitochondrial respiration. Additionally, to assess mitochondrial biomass, we analysed mitochondrial DNA copy number profiles. We validated these results using an immunohistochemistry (IHC)-based metabolism panel of three glioma patient cohorts, including patients who had undergone anti-angiogenic treatment. To investigate the influence of the immune microenvironment on our results, we analysed the immune cell content and composition by epigenetic data deconvolution and IHC. The findings revealed that DNA methylation patterns of key metabolic genes are strongly associated with molecular subgroups of gliomas and that IDHmut gliomas are enriched for a respiratory mitochondrial signature. A mitochondrial signature in IDHwt glioblastomas was also associated with improved patient survival.
Finally, patients with a relatively increased mitochondrial signature displayed an improved response to anti-angiogenic treatment.

| Patient material and patient cohorts
DNA methylation analyses were performed using samples from 57 patients with glial brain tumours. All samples were collected as formalin-fixed and paraffin-embedded (FFPE) blocks from the University Cancer Center Frankfurt (UCT) Biobank from 2017 to 2019. The tissue was used for either DNA isolation as detailed in the corresponding segment or generation of tissue microarrays (TMAs).

Further information regarding this patient cohort is summarised in
File S1 (EPIC-Cohort). From all cohorts, we extracted a fourth cohort of 29 patients who received bevacizumab (Bev) during the course of their disease.
All samples were collected as paraffin blocks before Bev treatment.
Further clinical data regarding this cohort are summarised in File S4 (Bev-Cohort).

| IHC, microscopy, scoring and semi-automated digital quantification of immune cell infiltration
All TMA cores were punched from representative non-necrotic tumour areas. Three-micrometre thick sections of the TMAs were made with a microtome. The sections were placed on SuperFrost-Plus slides (Thermo Fisher Scientific, Waltham, MA, USA) and stored overnight in a 37°C incubator. IHC was performed using antibodies against the following antigens: semi-purified mitochondrial prepa- For semi-quantitative evaluation, the H-Score was applied as described previously [35]. The relationships between the metabolic surrogate markers were investigated using quartile ratios of each tumour and the corresponding marker (1. quartile 0-25%, 2. quar-

| DNA isolation for mtDNA copy number analyses
Tissue specimens from FFPE samples were examined using haematoxylin and eosin staining to determine non-necrotic regions of tumour tissue. One or two cores 2 mm in diameter and 3 to 5-mm thick were punched out from the selected regions of the paraffin blocks. DNA was extracted using the Stratec Genomic DNA Kit (Stratec, Berlin, Germany) according to the manufacturer's protocols.

| EPIC DNA methylation analysis
DNA from 57 patient samples was isolated as described above. CpG sites illustrated in the text were analysed with the EPIC 850 K Array (Illumina, San Diego, CA, USA). After standardised DNA processing, bisulphite treatment and further processing and hybridisation were performed as recommended by the manufacturer and previously described [16]. The resulting data were processed using Illumina

| DNA methylation-based immune cell deconvolution
The deconvolution was performed using RnBeads 2.0, which includes the leukocytes unmethylation for purity (LUMP) algorithm.
This algorithm estimates the immune cell content of tumours based on 44 non-methylated immune-specific CpG sites [37,38].

| Data visualisation and statistical analyses
All statistical analyses were performed using either JMP11/14 (SAS Institute) or R (R Core Team, 2019). Epigenomic and IHC data visualisation were performed with R utilising a ComplexHeatmappackage [39] to generate heat maps and hierarchical clustering, and JMP11/14. JMP add-in with the R-package Rtsne was used to generate t-distributed stochastic neighbour embedding (t-SNE) plots.

| DNA methylation profiling defines distinct metabolic clusters of diffuse gliomas.
In the first approach, analysis of 57 patients in the EPIC-Cohort for the 10,000 most variable CpG sites revealed strong clustering into IDHmut and IDHwt tumours, as well as the strong association of clusters with established molecular subclasses (File S7).
To determine the epigenetic regulation of cellular bioenergetics, Although IDHmut gliomas displayed a CpG site hypermethylation phenotype as compared to IDHwt gliomas, this did not entirely refer to the investigated 'Metabol' CpG site gene cluster. In contrast, methylation profiles appeared similar to a certain extent. Only a small subset of CpG sites seemed to determine the different clusters. Interestingly, clustering can not only separate IDHmut from IDHwt tumours, but also suggests subclustering into known epigenetic subgroups [16,17].
Next, we assessed the most relevant shared CpGs between the groups of most variable CpGs and 'Metabol' CpGs. Only 17 CpG sites were shared from both the 'Metabol' CpG site cluster and from the 10,000 most variable CpGs in the EPIC-Cohort (see the Venn diagram presented in Figure 1C). The genes with the most represented shared CpGs were the glycolytic gene PFKP, encoding the enzyme phosphofructokinase and the ATP5G2 gene, which encodes a subunit of the mitochondrial ATP synthase ( Figure 1D). These 17 CpG sites were able to discriminate IDHmut from IDHwt tumours (File S7).

| LDHA promoter-associated CpGs are highly regulated in different glioma subtypes
LDHA is upregulated in IDHwt glioma as compared to IDHmut glioma [40]. Furthermore, a recent study reported the increased hypomethylation of LDHA promoters during malignant progression in IDHmut gliomas [41]. To further investigate potential functionally relevant methylation patterns in gliomas, we investigated only annotated promoter-associated CpGs of the group of shared CpGs ( Figure 1D), including LDHA promoter-associated CpGs (Figure 2A).
The analysis led to the investigation of 65 CpG sites in the EPIC-Cohort. The resulting clusters were again almost exclusively linked to the IDH mutation status of the samples. Many of these promoterassociated CpGs were hypomethylated. Only promoter-associated CpGs of ATP5G2 and LDHA showed obvious differential methylation in the molecular subgroups ( Figure 2A). Interestingly, LDHA promoter-associated CpG sites were hypermethylated in almost all IDHmut tumours, while some showed hypermethylation only in a subset of CpG sites. Interestingly, none of the 1p/19q co-deleted tumours showed a fully hypomethylated LDHA promoter, whereas some IDHmut astrocytomas and high-grade astrocytomas presented with LDHA promoter hypomethylation. AllIDHwt gliomas showed a hypomethylated LDHA promoter. These collective observations implicated LDHA promoter-associated CpG site hypomethylation as a feature associated with glioma aggressiveness and potentially associated with malignant progression of high-grade IDHmut astrocytomas. The ATP5G2 promoter-associated CpG site methylation patterns were not entirely associated with IDH mutation status. Nevertheless, IDH mutant tumours showed a hypermethylated phenotype, which was also observed in some IDHwt tumours ( Figure 2A). In contrast, the promoter-associated CpGs of NDUFV2A, which encodes the NADH-ubiquinone oxidoreductase complex (complex I) of the mitochondrial respiratory chain, was unmethylated in any sample from the EPIC-Cohort, suggesting that these genes are essential for cellular energy homoeostasis.
To assess whether the epigenetic switch of LDHA promoter methylation and demethylation has direct effects on the protein level, we investigated LDHA expression via IHC ( Figure 2B). LDHA expression at the immunohistochemical level was significantly correlated with LDHA promoter methylation status ( Figure 2C,D). The findings directly linked epigenetic regulation to protein expression.

| Copy number of mtDNA is linked to IDH mutational status
The foregoing data showed that several promoters of genes that might be needed for cellular energy homoeostasis were not differentially methylated between IDHwt and IDHmut tumours, including NDUFV2.
Thus, we were interested in determining whether the total mitochondrial biomass varied between these glioma subgroups. To gain insights into mitochondrial biomass, we used mtDNA copy number as a surrogate in our mtDNA-Cohort. The mtDNA copy number was higher in IDHmut tumours than in IDHwt tumours ( Figure 3A). Furthermore, when comparing primary with matched pair recurrent tumours, mtDNA copy number was not statistically significantly different, although some patients showed differences in mtDNA copy number between primary and recurrent tumours ( Figure 3B). To extend these results to the proteomic level, we employed two immunohistochemical biomarkers. One was the antibody to MTC02, which recognises a 60-kDa non-glycosylated protein component of mitochondria. We used this as a surrogate for mitochondrial biomass. We also used MT-CO1, which is essential for oxidative phosphorylation as an integral part of the electron transport chain (Complex IV). MT-CO1 serves as a surrogate for mitochondrial respiratory capacity ( Figure 3C).
Correlation of the relative expression of these proteins with mtDNA copy number revealed a positive correlation between mtDNA copy number and MTC02. No correlation was evident between mtDNA copy number and MT-CO1, or between MTC02 and MT-CO1 ( Figure 3D). The findings implicated MTC02 as a useful surrogate biomarker for mitochondrial biomass but also suggested that pure mitochondrial biomass might not be directly linked to functionality. Finally, a higher relative mtDNA copy number was associated with better patient overall survival ( Figure 3E).

| IHC of metabolic surrogate biomarkers reveals strong association with IDH mutation status and mesenchymal signature in IDHwt gliomas
Next, we were interested in further comparing our epigenetic findings of the EPIC-Cohort with protein expression of metabolic surrogate biomarkers of the mtDNA-Cohort. As metabolic surrogates, In the first approach, we were interested in the expression of these metabolic surrogates in epigenetically defined gliomas. IDHmut astrocytomas ( Figure 4C) suggested that IDHwt gliomas with a mesenchymal signature tend to be associated with a more oxidative phosphorylation phenotype.
We combined the expression profiles of the three metabolic surrogate markers to an IHC panel and performed hierarchical cluster analysis. Strong enrichment of oxidative phosphorylating tumours and glycolytic tumours was evident ( Figure 4D).
To gain insights into whether the different glioma subclasses in our cohort favoured anaerobic glycolysis over oxidative phosphorylation, as described by the Warburg effect, we analysed LDHA expression in relation to MT-CO1 expression. We generated quartile ratios for each specimen and metabolic surrogate marker ( Figure 4E). The quartile ratio showed higher values in the IDHmut subclasses than in the IDHwt subclasses. These results confirmed our epigenetic and genetic results of LDHA hypermethylation and higher mitochondrial respiratory capacity in IDHmut tumours.
To investigate whether this ratio was defined by an epigenetic signature of metabolic genes in IDHwt glioma subclasses, we performed hierarchical clustering based on the 2542 'Metabol' CpGs and annotated dichotomised MT-CO1/LDHA quartile ratios (high versus low; Figure 4F). Clustering revealed four subgroups linked to the IDHwt epigenetic subclasses. We observed five tumours with high MT-CO1/LDHA ratios. Interestingly, these tumours almost exclusively belonged to the mesenchymal subclass of gliomas.

| Immune cell content is associated with epigenetic subclass but not with MT-CO1/LDHA ratio
To evaluate the impact of immune cell composition on the detected signatures, we employed tumour deconvolution analyses based on the EPIC Array data (LUMP algorithm) as well as immune cell quantification using IHC and semi-automated analysis of CD8, CD68 and Iba1. Immune cell infiltration was highest in IDHwt compared to IDHmut specimens with the highest immune cell levels in the mesenchymal subclass ( Figure 5A). The predominant immune cell type was microglia ( Figure 5B). CD8-positive T-lymphocytes were scarce in the analysed specimen, with mean frequencies ranging from 0.08% in 1p/19q co-deleted oligodendroglioma to 0.39% in mesenchymal subtype glioblastomas ( Figure 5B). Immune cell content detected by epigenetic deconvolution strongly correlated with immune cell content detected by IHC and semi-automated analyses ( Figure 5C). From all analysed metabolic markers, only LDHA expression was associated with immune cell content ( Figure 5D). We did not detect statistically significant differences in cell type-specific immune cell content between MT-CO1/LDHA high versus low tumours ( Figure 5E).

| Metabolic surrogate biomarkers are associated with patient prognosis in a large population-based glioma cohort
To test whether our results are applicable in a routine histopathological setting of non-epigenetically classified gliomas, we performed  Figure 6C). To exclude bias induced by IDH mutation status, we performed the same analyses focussing only on IDHwt GB. Survival was worse in Cluster 4 patients than in Cluster 1 patients ( Figure 6D).

| MT-CO1/LDHA quartile ratio is a biomarker for therapy response in Bev-treated gliomas
Finally, we hypothesised that the distinct metabolic status of gliomas would lead to different responses to a metabolically effective treatment strategy, such as anti-angiogenic treatment with Bev. We

| DISCUSS ION
Since 2008, the increased understanding of tumour biology of diffuse gliomas with the discovery of IDH mutations has included the definition of new subclasses of gliomas [7]. Basic and translational studies have investigated the role of mutant IDH and discovered that IDH mutations result in a DNA and histone hypermethylation phenotype [10][11][12][13][14][15]. DNA methylation analyses permit precise tumour classification [16]. However, there is still a lack of understanding of the relationship between epigenetic changes and metabolism.
The aim of our study was to characterise the metabolic differences in diffuse gliomas with a special focus on epigenetic (DNA methylation) and proteomic (metabolic panel IHC) alterations, as well as differences in mitochondrial biomass and genetics (metabolic panel IHC and D-Loop qPCR). We were able to confirm that IDHmut tumours showed a hypermethylated phenotype. The in IDHmut gliomas [41].
ATP5G2 was identified as one of the most extensively regulated genes at the DNA methylation level. ATP5G2 encodes a subunit of the ATP Synthase complex V of the electron transport chain. We observed that most of the ATP5G2 promoter-associated CpG-islands were hypermethylated in the IDHmut gliomas. Only a small subset of IDHwt samples also showed increased methylation, but in fewer CpG sites. ATP5G2 has been previously described as being epigenetically regulated by DNA methylation in non-muscle invasive bladder cancer and renal cell carcinoma [42,43]. In these studies, ATP5G2 was identified as a hypermethylated potential tumour suppressor gene that was increasingly inactivated High and low values were grouped after median split. Dots display single patients (black dots IDH1R132H mut, grey dots IDH1R132H non-mut tumours). Investigated tissue corresponds to last resection/biopsy before Bev treatment was initiated with malignancy. Interestingly, in the EPIC-Cohort in patients with IDHwt tumours promoter-associated CpG sites were mostly hypomethylated. Another study found that ATP5G2 was upregulated in an ischaemic rat brain model. The authors described upregulated ATP5G2 gene expression 24 h after artery occlusion [44], suggesting that ATP5G2 gene expression is an early hypoxia/ ischaemia response gene. The role of the expression of genes like ATP5G2 in gliomas needs further investigation.
Cellular metabolism and the provision of bioenergetics are not only restricted to glycolysis. Thus, we were interested in whether the mitochondrial biomass varied between IDHmut and IDHwt tumours. We investigated our mtDNA-Cohort for mitochondrial DNA copy number variations. Interestingly, mitochondrial DNA copy number was higher in IDHmut tumours than in IDHwt tumours, suggesting a relatively increased mitochondrial signature in IDHmut tumours. Additionally, we showed that mitochondrial copy number correlated with a surrogate parameter of mitochondrial biomass (antibody MTC02) but not with protein expression of MT-CO1, a major contributor to oxidative phosphorylation. These findings underline that mitochondrial biomass should not be directly associated with mitochondrial functionality. In line with our findings, Navis and co-workers described higher amounts of mitochondria in an IDHmut glioma model [45]. Furthermore, in non-glial tumours, Farshidfar and co-workers identified an mRNA signature of IDHmut cholangiocarcinoma (CCA) enriched for genes involved in mitochondrial structure and function. The authors also described a higher mitochondrial DNA copy number in IDHmut CCA [46]. Our clinical data suggest that higher amounts of mitochondrial biomass are associated with a beneficial clinical course of diffuse gliomas. In this context, it is important to mention that mtDNA copy number did not significantly change between primary and recurrent tumours, indicating that, in contrast to the potential epigenetic adaptation of LDHA promoter hypomethylation, the mitochondrial biomass does not change throughout the clinical course. The reason for the elevated mitochondrial biomass in IDHmut as compared to IDHwt tumours remains speculative. In Saccharomyces cerevisiae, analogous to glioma-associated mutations of the NADP + isocitrate dehydrogenase, mutations resulted in increased levels of 2-hydroxyglutarate and extensive mtDNA loss, leading to a loss of respiratory capacity [47]. The gain of mitochondrial biomass is probably a compensatory mechanism of the cell to neutralise the toxic effects of 2-hydroxyglutarate. While it is plausible that the assumed reduced aerobic glycolysis in IDHmut gliomas could be associated with an increase in turnover in the citric acid cycle, a recent publication demonstrated the opposite [48].
We were able to further corroborate our findings at the pro- Not all mesenchymal IDHwt gliomas harboured this signature. Some did, indicating that mesenchymal IDHwt gliomas have the potential to have a mitochondrial signature.
When analysing DNA methylation signatures of tumour tissue, it should be considered that these signatures also reflect the non-tumour cell compartment. Cellular deconvolution revealed significant differences in immune cell infiltration in the distinct epigenetic glioma subclasses. In particular, mesenchymal GB showed high cellular heterogeneity with increased microglia/ macrophage and lymphocyte infiltration, as previously reported [49,50]. Although there is a strong interaction between tumours and especially microglia, our data do not support the hypothesis that the metabolic signature is a pure immune cell signature. Our data indicate that key metabolic markers are not associated with immune cell content.
When we investigated metabolic aspects in our population-based historical Glioma Cohort, we detected two distinct subclasses of mainly IDHwt gliomas, while IDHmut gliomas were enriched in one cluster. Survival analyses revealed that the cluster of IDHwt tumours with high LDHA and low MT-CO1/MTC02 expression showed worse patient prognosis. This again emphasised that, in case of IDHwt status, a mitochondrial signature might be associated with better patient survival.
To further explore our hypothesis that metabolic surrogate biomarkers are prognosticators in gliomas, we analysed glioma tissue from patients treated with Bev. Strikingly, patients who showed a mitochondrial signature in their tumour tissue had significantly better overall survival following Bev treatment. These findings are consistent with experimental data demonstrating that glioma cells without functional mitochondria are resistant to Bev [51].Although this finding needs to be validated in larger cohorts of patients, these pilot data are compatible with the hypothesis that tumours with a mitochondrial signature are susceptible to anti-angiogenic therapy because they are less capable of adapting their metabolism to conditions of Bev-induced hypoxia.
In summary, the metabolism of IDHmut and IDHwt tumours differs substantially and relevant metabolic genes are regulated by DNA methylation. Furthermore, mitochondrial biomass is increased in IDHmut tumours and is stable during the clinical course. Importantly, even within IDHwt gliomas, a mitochondrial signature was associated with a better clinical outcome. At the epigenetic level, a mitochondrial signature was associated with the mesenchymal subclass in GB.
Additionally, respiratory tumours may respond better to anti-angiogenic treatment than glycolytic tumours. The collective data indicate that for therapeutic regimens that either directly or indirectly affect tumour metabolism, the interrogation of multidimensional metabolic biomarkers is a promising strategy to define predictive signatures that can guide individualised treatment decisions.

ACK N OWLED G EM ENT
The authors thank Tatjana

E TH I C A L A PPROVA L
The study was approved by the local ethics committee (GS 4/09, SNO-11-2017).

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.