DNA methylation is a heritable epigenetic marker in the genome that does not impact the genetic makeup through the addition of a methyl group to the DNA molecule. Most DNA methylation in humans occur at the 5′ carbon of the cytosine base which is followed by a guanine nucleotide. Addition of the methyl group changes the chromatin structure, making it more condensed, which results in DNA being less accessible for transcription [1
]. DNA methylation patterns change during development [3
] and with age [4
]. DNA methylation in the genome also exhibits tissue- and cell-type differences, for example, highly methylated content is observed in brain tissue [5
]. Other than these differences, which occur naturally during the human life cycle, methylation changes are also related to diseased cell states. Many studies have revealed two common alterations in cancer: DNA hypermethylation and hypomethylation [6
]. Hypermethylation involves DNA methylation events at the Cytosine-phosphate-Guanine (CpG) site within the promoter of tumor suppressor gene while hypomethylation removes the methyl group at the promoter site of oncogenes and can be widespread around the genome to promote tumor progression [9
Although the genome is the same in essentially all cells of the human body, the epigenome differs from cell to cell and is dynamic, changing with time and exposure to the environment. Epigenetic mechanisms affect all steps of the gene expression process, from chromatin state to transcription, post-transcriptional RNA processing, and translation. Epigenetic mechanisms for regulating gene expression include DNA methylation and histone modification, arguably the best-established epigenetic processes that regulate gene expression at the level of transcription. Histones are highly conserved basic proteins around which DNA is tightly wrapped to form nucleosomes, or the “beads on a string structure” that make up chromatin. Histone tails protrude from the nucleosome and are subject to a variety of post-translational modifications (PTMs), primarily methylation, acetylation, and phosphorylation. These PTMs affect gene expression by controlling accessibility of the chromatin structure to expose DNA-binding sites or by closing DNA-binding sites to facilitate transcription [11
]. Non-coding RNAs also have a role in epigenetic modification by regulating gene expression and chromosomes to control cell differentiation [12
]. In addition, gene expression can be regulated post-transcriptionally through dynamic and reversible RNA modifications. Extra methyl group modifications induce either duplex stability or protein–RNA affinity and positively correlate with translation [13
]. Epigenetic mechanisms for regulating gene expression clearly are complex and diverse to create a dynamic epigenome and epitranscriptome.
Glioma is a type of cancer that occurs in the brain and it is the most prevalent brain tumor observed in the population. Gliomas are comprised of four subtypes which are classified based on the World Health Organization of Central Nervous System Tumor Classification guidelines updated in 2016, with the introduction of molecular biomarkers to further refine the diagnostic criteria [14
]. Grade II and Grade III gliomas include astrocytoma, oligodendroglioma, and oligoastrocytoma which are frequently considered to have significantly better patient survival than Grade IV glioma [15
]. Glioblastoma multiforme (GBM), a Grade IV glioma subtype, is an aggressive brain tumor and one of the most fatal due to its lack of curable treatment. The median survival time for patients is between 15 to 16 months [16
]. One commonly observed biological event that affects patient survival time is DNA methylation [17
]. Studies confirmed a phenomenon called CpG Island Methylator Phenotype (CIMP) after analyzing DNA methylation data of GBM and LGG (lower glioma grade) cohorts from The Cancer Genome Atlas (TCGA) database. It noted hypermethylation on the CpG island of a subset of genes, including ANKRD43
, and FAS-1
, and classified tumor samples as CIMP-positive or CIMP-negative based on the methylation level detected on those genes. Glioma patients with CIMP-positive tumors correlate with better survival. Another well-studied methylation biomarker in GBM patients is in the DNA repair gene called O6-methylguanine–DNA methyltransferase (MGMT
). Methylation on the promoter site of MGMT
reduces gene expression and protein activity to prevent it from rescuing tumor cells with alkylating agent-induced damage caused by chemotherapy [18
]. As a result, GBM patients with loss of or low MGMT activity have higher sensitivity to temozolomide, a common chemotherapy used to treat GBM. Although there is growing attention around the classification of glioma patients through molecular profiling such as genetic and methylation signatures [19
], studying the connection between DNA methylation pattern changes and tumor grades, which is classified through histological characteristics as a standard protocol by the World Health Organization, will improve knowledge about glioma tumor progression.
Despite the current understanding around how alterations in DNA methylation induce tumor generation and progression [20
] or predict patient outcome independently [23
], it is unclear how DNA methylation changes progress between tumor grades or how they influence patient survival. We hypothesized there is a correlation between methylation pattern and glioma grade and such correlation could be used for tumor progression monitoring as well as patient prognosis. We evaluated the changes in DNA methylation patterns in LGG and GBM patients and identified changes in methylation level for CpG sites that have significant impact on patient survival.
With differential variability analysis, we identified CpG sites that behave similarly within the same glioma grade but differently from the other glioma grades. We compared Grade II vs. Grade III and Grade III vs. Grade IV and identified 2241 and 59,234 significantly differentially variable (DV) CpG sites, respectively. The epigenetic profile difference between Grade III and Grade IV was much greater than the differences observed between Grade II and Grade III. It was not surprising since previous methylation profiling studies have shown a similar pattern where Grade II and Grade III samples are more likely to be clustered together and separate from Grade IV [19
]. We noticed a progressive demethylation condition in the top 10% of the DV CpG sites via the methylation profiling heatmap and this observation aligns with the observations from other studies [22
We applied the network construction function from WGCNA to the subset of CpG sites and found three comethylation modules highly correlated with the progression of glioma grade, the clinical trait of interest in our study. Most of the CpG sites displayed a negative correlation with the increase of tumor grade indicating that more CpG were demethylated as the tumor progresses. The demethylation pattern in the data set indicates the up-regulated gene expression in higher glioma grade samples. As hyper- and hypomethylation in cancer normally applies to extensive methylation or demethylation around the promoter site, we assessed genes that have five or more significant CpG sites associated with them. Out of the 150 candidate genes, four of them, SMOC1, KCNA4, SLC25A21, and UPP1 were most outstanding due to the distinct gene expression pattern changes between the three glioma grades. Gene expression of SMOC1, KCNA4, and SLC2521 declined as patient glioma grade increased while UPP1 gene expression showed a positive relationship with tumor grade. Ten of the CpG sites associated with these four genes displayed a strong inverse relationship between gene expression and methylation level implying a high probability that these CpG sites impact gene expression.
gene encodes a matricellular protein called secreted modular calcium-binding protein 1 [29
] and this protein was shown to regulate growth factors [30
]. Boon et al. showed SMOC1
is a Grade II and III astrocytoma-associated gene [33
] and this conclusion aligns with the gene expression data we analyzed; the greatest median gene expression was observed in Grade II glioma samples and subsequently dropped particularly low for Grade IV glioma samples. Previous studies revealed several functions of SMOC1 including the promotion of angiogenesis through regulation of transforming growth factor β signaling pathway in cultured endothelial cells [32
] and inhibition of cell migration induced by tenascin-C, an extracellular protein that is overexpressed in many human cancer types, in glioma cell lines [34
]. Our analyzed data showed a dramatic reduction in SMOC1
expression and we predict the main function of SMOC1
in Grade IV glioma is to promote tumor invasion and migration since rapid spreading is one of the signatures of glioblastoma. Our survival analysis of the SMOC1
targeted CpG sites reflected a poor prognosis in highly methylated samples which corresponded to down-regulation of SMOC1
KCNA4 (Potassium voltage-gated channel subfamily A member 4, aka Kv1.4) is a member of the potassium voltage-gated channel family and one of its major functions is the cardiac transient outward K(+) currents [35
]. However, Zheng et al. observed hypermethylation at KCNA4
promoter site in serum as well as tissue samples of gastric cancer patients, and it was one of the markers which showed good sensitivity and specificity for detection [36
]. Coma et al. found a global reduction in voltage-gated potassium channel expression, including KCNA4
, in the brain of tumor-bearing animals suffering from cancer cachexia [37
]. Although there is limited information around how KCNA4
is associated with tumors, other members from the potassium voltage-gated channel family such as Kv1.3 and Kv1.5 are well-studied and have shown correlations with several human cancers including gliomas [38
]. Further investigation on how KCNA4
expression and methylation impact glioma is needed. The KCNA4
CpG sites we identified indicate worse patient survival in the hypermethylated group correlating with decreased gene expression as glioma grade progresses.
Solute Carrier Family 25 Member 21 is encoded by the SLC25A21
gene and it catalyzes the transportation of 2-oxoadipate and 2-oxoglutarate across the mitochondrial membranes [40
]. Rochette et al. reviewed several abnormal SLC25
activities that are linked to cancer including the overexpression of SLC25A1
in lung cancer, SLC25A43
gene deletion as well as elevated SLC25A33
expression in breast cancer, and SLC25A10 that regulates the redox homeostasis was also increased in multiple cancer types [41
]. More importantly, SLC25A12
expression in hepatocellular carcinoma cell line increased through the modification of histone acetylation [42
]. Although the impact of SLC25A21
on cancer has not been evaluated, the vast number of studies done on other members of the SLC25 family implies that SLC25A21
can also be a promising tumor-associated marker. This is proved with the survival analysis of our study where methylation changes observed in different glioma grade serves as a good prognosis marker; hypermethylation on the CpG site targeted by cg25051529 is associated with poorer survival.
gene encodes the Uridine Phosphorylase 1 that catalyzes the reversible phosphorylation of uridine to uracil [43
] and maintains uridine homeostasis [44
]. Overexpression of the UPP1
is associated with cancer. Guan et al. analyzed the TCGA cohort and found elevated UPP1
in thyroid cancer patients compared to normal tissue samples [45
]. The up-regulated UPP1
expression increased lymph node metastasis risk and promoted tumor growth. Noushmehr et al. identified hypermethylation of UPP1
gene in patients belong to the proneural subgroup that were diagnosed with low grade gliomas and many of the patients were classified to have CIMP-positive tumors [17
]. Wang et al. found UPP1
gene expression was up-regulated as glioma grade increased using more than 900 samples from the TCGA database [46
]. The Gene Ontology analysis revealed that UPP1
is likely associated with immune and inflammatory response and the increase of expression negatively impacted patient survival. The extensive data have proven UPP1
is an efficient prognosis marker for cancer. Our analysis aligns with the previous studies where UPP1
gene expression was greatest in glioma Grade IV samples with demethylation at the CpG site captured by cg16270885 and correlates with lower patient survival.
All ten prognostic methylation CpG candidates showed good prognostic capability individually but we wanted to investigate their collective efficiency in the prediction of patient survival. We took the poor prognosis conditions obtained from the ten candidates (Table 2
and Figure 6
) and assigned patients from the training sample set that met all ten poor survival criteria to the poor prognosis group. The survival probability of the poor prognosis group dropped remarkedly; 50% survival probability at less than 14 months. The good prognosis group had a significantly better survival where the median survival time was more than 81 months. This set of prognostic markers was validated collectively in the validation sample set and a similar trend was observed: median survival time was approximately 18 and 74 months for poor and good prognosis group, respectively.
We plotted the identified prognostic CpG sites and observed a consistent methylation change as glioma grade increases. CpG site of UPP1
demethylate as glioma grade increases while the remaining genes, SMOC1, KCNA4,
, methylate along with higher tumor grade. This indicates the methylation pattern changes at these specific CpG sites change collectively as the tumor progresses. We compared the changes at those CpG sites with other molecular alterations in glioma and as well as molecular subtypes assigned by the Ceccarelli et al. We found that the methylation profile of these CpG sites align with specific molecular subtypes. The greatest methylation signal correlation for our probes was observed in samples with IDH mutation and 1p/19 codeletion, with a slight decrease in methylation for samples containing only IDH mutation, and the lowest methylation for samples with IDH wild type genotype and no 1p/19q codeletion. IDH mutation can promote CIMP in gliomas [47
] which explains the increased methylation in UPP1
-associated probes, but our observation suggests that UPP1
methylation can also be associated with 1p/19q codeletion.
We investigated the sample distribution in both good prognostic and poor prognostic groups to understand the significance of these genes. All samples in the poor prognostic group have IDH wild type genotype with no 1p/19q codeletion, which aligns with current knowledge that IDH mutation and 1p/19q codeletion usually result in more favorable overall survival compared IDH wild type and non-1p/19q codeletion, independently [15
]. The poor prognostic group consists of samples with molecular subtypes of classic-like, mesenchymal-like, and one case from LGm6-GBM. As summarized by Ceccarelli et al., samples from G-CIMP-low and the subtypes mentioned above have poorer patient survival compared to the samples classified as codeletion, G-CIMP-High, or Pilocytic Astrocytomas (PA) subtypes. Our results suggest the genes we identified are generally associated with molecular changes observed in glioma since they identify samples with tumors that display classic gene expression signature (classic-like) and mesenchymal-like instead of the codel (1p/19q codeletion) and CIMP subtypes. Mair et al. have reviewed the patient survival data of gliomas and suggest oligodendroglioma has the most favorable survival in Grade II and III gliomas with a median overall survival for oligodendroglioma of at least above 11 years [15
]. However, our prognostic model allows us to identify oligodendrogliomas that exhibit similar methylation profile at these CpG sites, as the Grade III astrocytoma, Grade III oligodendrogliomas, and Grade IV GBM, thus refining the prognosis for these patients. We found that most of the poor prognostic samples contain chromosome 7 amplification and chromosome 10 deletion alteration. Chromosome 7 amplification is linked to increased mesenchymal gene expression which supports our findings of these group of CpG sites correctly diagnosing poor prognosis.
We identified the median survival time of the patients from the poor prognosis group to be between 14 to 18 months which is equivalent to the median survival of high glioma grade, glioblastoma multiforme. Interestingly, the patient composition of the poor prognosis group was a mixture of mainly glioma Grade III and glioma Grade IV patients suggesting the devastating effect is observed across tumor grades.
Glioblastoma multiforme is known to be an aggressive brain tumor with a median survival of 15 to 16 months while Grade II and Grade III gliomas are less destructive and leave patients with longer survival time. However, lower-grade gliomas will ultimately progress to glioblastoma. We need improved systematic approaches to monitor tumor progression and predict patient survival based on the real-time molecular changes that result from tumorigenesis and progression. The increased attention around tumor detection through somatic mutations found in cellular tumor DNA permits an early and noninvasive way for diagnosis [58
]. Methylation changes identified via methylation sequencing provide more context to the diagnostic process and patient care optimization by pinpointing the origin of cancerous sites as well as providing progression monitoring [59
In conclusion, we identified ten methylation markers affecting four genes (SMOC1, KCNA4, SLC25A21, and UPP1) that are associated with glioma grade progression, and demonstrate a strong prognostic probability for patient prognosis which is also able to identify patients usually considered to be good survivors (oligodendroglioma). We found the gene expression level of SMOC1, KCNA4, SLC25A21, and UPP1 to be closely correlated with the methylation level of specific CpG sites for each gene. The methylation signal of these ten CpG sites changed progressively with glioma grade and they showed good prognostic capability collectively. In addition, the identified CpG sites show high correlation with molecular subtype, IDH alteration, and chromosome 1p/19q alteration, strengthening the validity of our model.
Our ten-methylation-marker model predicts survival for patients with oligodendrogliomas that exhibit similar epigenetic profiles as patients with higher grade glioma and poorer prognosis. As there is some evidence suggesting that UPP1 and SMOC1 are markers for glioma, our finding of KCNA4 and SLC25A21 add to the previously identified gene list to fortify patient outcome prediction and guide future investigations on the impact of these genes and their pathways involved in glioma progression. The ten methylation markers evaluated in this study will contribute to the continuous improvement of patient prognosis; patient prognosis should imitate the model of precision medicine where patients are treated based on their unique circumstances and given precise diagnosis to receive the appropriate medical care.