Cancer is generally characterized by loss of CG dinucleotides methylation resulting in a global hypomethylation and the consequent genomic instability. The major contribution to the general decreased methylation levels seems to be due to demethylation of heterochromatin repetitive DNA sequences. In human immunodeficiency, centromeric instability and facial anomalies syndrome, demethylation of pericentromeric satellite 2 DNA sequences has been correlated to functional mutations of the de novo DNA methyltransferase 3b (DNMT3b), but the mechanism responsible for the hypomethylated status in tumors is poorly known. Here, we report that human glioblastoma is affected by strong hypomethylation of satellite 2 pericentromeric sequences that involves the stem cell compartment. Concomitantly with the integrity of the DNMTs coding sequences, we report aberrations in DNA methyltrasferases expression showing upregulation of the DNA methyltransferase 1 (DNMT1) and downregulation of the de novo DNA methyltransferase 3a (DNMT3a). Moreover, we show that DNMT3a is the major de novo methyltransferase expressed in normal neural progenitor cells (NPCs) and its forced re-expression is sufficient to partially recover the methylation levels of satellite 2 repeats in glioblastoma cell lines. Thus, we speculate that DNMT3a decreased expression may be involved in the early post-natal inheritance of an epigenetically altered NPC population that could be responsible for glioblastoma development later in adult life.
DNA methylation is highly regulated in normal cells, while a drastically modified methylation distribution is observed in transformed cells. The local hypermethylation observed in human cancer, involving tumor suppressor gene promoters, is often concomitant to a globally decreased cytosine methylation involving repeated pericentromeric DNA sequences. The Cytosine guanine (CpG) dinucleotides, highly represented in repeated regions named satellite 2, have been found to be hypomethylated first in ICF (immunodeficiency, centromeric instability and facial anomalies syndrome) patients and later on in different types of cancer contributing to genomic destabilization, chromosome defects and cellular transformation (Jeanpierre et al., 1993; Vilain et al., 1999; Ehrlich, 2002; Tsuda et al., 2002).
So far, three different DNA methyltransferases (DNMTs) have been identified and characterized as active enzymes in mammals: DNMT3a and DNMT3b are responsible mainly for de novo DNA methylation, while DNMT1 is preferentially involved in maintenance of DNA methylation established pattern (Bestor, 2000).
Grade IV astrocytoma (or glioblastoma multiforme – GBM) is the most aggressive and frequent subtype of glioma, and shows very little response to chemotherapy (Maher et al., 2001).
Several genetic alterations have been identified in the different grades of astrocytoma, demonstrating the high heterogeneity and genomic instability of this kind of tumor (Maher et al., 2001; Behin et al., 2003), but they are not specific markers of astrocytoma development, despite their accumulation can be considered peculiar of this neoplasia.
Moreover, gliomas offer several examples of CpG island hypermethylation within tumor-suppressor gene promoters (for example RASSF1A, MGMT, PTEN and p53) associated with decreased or abolished gene expression (Baeza et al., 2003; Blanc et al., 2004; Gao et al., 2004).
Recently, cell populations (named brain tumor stem cells – BTSCs) have been isolated from glioblastoma tissues showing features expected from stem cells and able to generate tumors in scid mice (Eramo et al., 2006), leading to the hypothesis that stem cell could be the target of neoplastic transformation and the source of glioblastoma development.
In this study, we first investigated the hypothesis that glioblastoma should be affected by altered methylation levels of pericentromeric DNA sequences (satellite 2) that could contribute to the massive genomic instability of this kind of tumor. Secondly, it was investigated that both the integrity and the possible changed expression profile of the DNMTs, at both messenger RNA (mRNA) and protein levels, as putative mechanism/s through which glioblastoma could acquire an altered pattern of methylation. DNA methylation and DNMTs expression studies were produced in different glioblastoma models, such as immortalized cell lines, primary cells and in three different BTSC populations.
Moreover, we investigated the reversibility of demethylated phenotype of glioblastoma cells, re-establishing the DNMT3a expression, and the peculiar expression of the de novo DNMTs in normal human neural progenitor cells (NPCs) in order to comprehend how methylation of satellite 2 DNA repeats can be lost.
Satellite 2 pericentromeric DNA sequence is hypomethylated in glioblastoma cells
Satellite 2 is a repetitive DNA sequence prevalently represented in pericentromeric regions of chromosome 1 and 16 and characterized by the (ATTCCATTCG)2 sequence (Prosser et al., 1986; Lee et al., 1997).
A methylation-sensitive endonuclease-based assay revealed, in all glioblastoma model systems considered in this study, a decreased satellite 2 methylation, if compared with the corresponding not neoplastic samples (NHA – normal human astrocytes and NPC), at levels comparable with those shown by human cells treated with demethylating agent (5-aza-2′-deoxycytidine; Figure 1).
DNMT genes are not mutated in glioblastoma cells
Satellite 2 hypomethylation has been correlated, at least in 60–70% of ICF patients, to genetic alterations affecting the DNMT3b gene (Hansen et al., 1999; Okano et al., 1999; Xu et al., 1999). To verify the presence of some genetic alterations involving the DNMT genes that could justify the observed satellite 2 loss of methylation, the integrity of DNMTs genes was investigated. No mutation in the coding sequence of DNMT1, DNMT3a and DNMT3b was detected in all three BTSC populations (BTSC no. 1, BTSC no. 23, and BTSC no. 30 – data not shown).
DNMTs expression pattern in glioblastoma cells
In order to establish differences in the expression of DNMTs between normal and tumor cells, we analysed the mRNA levels of DNMT1, DNMT3a and all five isoforms of DNMT3b, by real-time quantitative reverse transcription–PCR (RT–PCR), in two immortalized glioblastoma cell lines (U-373MG and T98G), two primary glioblastoma cell populations (FCN-9 and MZC-12) and three BTSC populations (BTSC no. 1, BTSC no. 23 and BTSC no. 30).
Comparing the neoplastic cells (U-373MG, T98G, MZC-12 and FCN-9) with the normal counterpart (NHA), we observed an increased expression of DNMT1 (two- to fourfold) and decreased levels of DNMT3a expression (three- to fivefold) (Table 1). Comparable results were obtained with the three BTSC populations that, respect to the NPC, showed an increased expression of DNMT1 (>2 folds) and decreased levels of DNMT3a expression (two to threefolds) (Table 2). Despite some fluctuation monitored, probably due to the very low abundance of transcripts, it was not possible to detect relevant changes in the expression of DNMT3b between normal and neoplastic cells.
Biochemical analysis was considered to investigate if the altered DNMTs mRNA levels could be monitored also at the protein level. Immunoblot analysis showed, in all model systems, an increased DNMT1 expression in glioblastoma samples and decreased levels of DNMT3a protein. However, it was not possible to detect the DNMT3b expression (Figure 2).
To increase western blot sensitivity and to control the specificity of the antibody used (see Materials and methods), two additional polyclonal anti-DNMT3a antibodies were considered to immunoprecipitate the endogenous DNMT3a protein. Immunoblotting the immunoprecipitate samples, by using the anti-DNMT3a/3b antibody (ab2851), it was possible to detect a protein (about 110 kDa) expressed at lower level in neoplastic U-373MG samples (Figure 3, black arrow).
Moreover, a comparable immunoprecipitation experiment was performed against the DNMT3b by using two different polyclonal antibodies that were able to immunoselect a protein with the expected molecular weight (<97 kDa) expressed at comparable levels in neoplastic U-373MG and normal NHA cell lysates (Figure 3, white arrow).
Finally, we forced the expression of the DNMT3a in glioblastoma cells by using a variant third-generation lentiviral vector pRRL-CMV-PGK-GFP-WPRE (called Tween) (Ricci-Vitiani et al., 2004). After infection, glioblastoma cells were sorted for GFP expression and both DNMT3a protein and satellite 2 methylation levels were monitored.
As reported in Figure 4a, DNMT3a expression was detected upon infection in all the three cell lines tested (U-373MG, U-87MG and FCN-9). Interestingly, the methylation-sensitive assay revealed a slight increase of satellite 2 methylation in U-87MG and FCN-9 but not in U-373MG glioblastoma DNMT3a transduced cells (Figure 4b).
DNMTs 3a and 3b expression profile in NPCs compartment
The de novo methyltransferases expression in NPC was monitored by real-time quantitative RT–PCR and DNMT3a was found to be the most expressed de novo DNMT (9–10 fold more expressed than DNMT3b), while DNMT3b3 was the most abundant DNMT3b splice variant (Figures 5b and c). On the contrary, human testis shows a different pattern of DNMT3b expression in which DNMT3b1/2 and DNMT3b4 can also be detected (Figure 5a).
Among human astrocytic gliomas, GBM is the most common and aggressive kind of tumor characterized by high invasivity and refractariety to several therapeutic approaches. GBM is a useful model system to study cancer etiology at least for two reasons. First, GBM can be considered the prototype of the genomic instability that is thought to be the main feature for the accumulation of several genetic defects during cell life and the subsequent cellular transformation. Second, GBM is a tumor from which it was possible to identify and isolate a transformed cellular population with maintained stem cell potentiality (BTSCs) and considered the starting point of tumor development.
Trying to link the already reported genomic instability of GBM with a possible epigenetic perturbation of this biological model, we started to monitor the methylation levels of satellite 2 pericentromeric DNA sequence known to be highly methylated in normal cells.
The hypothesized epigenetic perturbation was monitored in all immortalized cell lines (U-373-MG, T98G, U-87MG) and primary cells tested (FCN-9 and MZC-12) showing a decreased methylation of pericentromeric satellite 2 sequences. Most importantly, the epigenetic alteration, involving pericentromeric DNA, was also monitored in all three BTSC populations.
Despite evidence already suggesting the existence of epigenetically modified stem cell population, from which tumor arise (Feinberg et al., 2006), the phenomenon of loss of DNA methylation in BTSCs shown in this study is the first experimental evidence of the existence of an epigenetic perturbated tumor stem cell population.
In order to understand the mechanism responsible for satellite 2 hypomethylation, all the three DNMT coding sequences (DNMT1, DNMT3a and DNMT3b) were analysed and no mutations were found. These data were not unexpected, since, as anticipated, DNMT3b mutations were not detected in 30–40% of the ICF cases, and also in all kind of cancer so far tested (Saito et al., 2002), suggesting a more complex mechanism based on which the global DNA demethylation can emerge.
The expression of DNMT1, DNMT3a and DNMT3b was investigated to monitor possible changes in mRNA or protein expression levels. Quantification of the DNMT transcripts showed an increased expression of DNMT1 and decreased level of DNMT3a.
Some experimental troubles had to be resolved, regarding antibody specificity and the low expression level of de novo DNMTs 3a and 3b, to biochemically confirm the same results obtained at mRNA level. In fact, DNMT1 protein was monitored upregulated, while DNMT3a protein expression was monitored only in the normal samples.
A possible role of altered DNMTs expression pattern in the decreased satellite 2 DNA methylation levels in GBM, especially considering the downregulation of the de novo DNMT3a expression, can be only hypothesized at this stage of knowledge.
However, there are some aspects that have to be underlined to propose our point of view. The brain can be considered as a peculiar tissue characterized by some interesting features considering both the DNMTs expression in the adult organ and the temporal regulation of the de novo DNMTs 3a and 3b from fetal to adult life.
Regarding DNMTs expression in the adult brain, it is known in human that DNMT1 is the most expressed DNMT, while DNMT3a and DNMT3b are expressed at very low level and that the major DNMT3b isoform expressed is DNMT3b3 (Robertson et al., 1999).
The described pattern of DNMTs expression in the adult human entire brain was experimentally confirmed here and, respect to the previous knowledge, we also reported that DNMT3a is the most expressed de novo DNMT in NPCs.
In addition, the very low level of DNMT3b protein is represented by only the DNMT3b3 splice variant that is not still proved to have an efficient catalytic activity in vivo and to have a questionable catalytic activity in vitro (Aoki et al., 2001; Chen et al., 2003). Thus, considering the DNMT3a as the only de novo active DNMT expressed in NPCs, its decreased expression could play a role in decreased methylation of satellite 2, as supported by the experiment in which satellite 2 methylation is partially re-acquired by transducing exogenous DNMT3a in glioblastoma cells. However, forced DNMT3a expression is not sufficient to revert completely the demethylation status, or not able to change it at all (as reported in U-373MG DNMT3a infected cells), probably due to unknown mechanisms that make the methylation phenotype irreversible.
Anyhow, a reduced expression of DNMT3a should not be sufficient to justify a decrease of DNA methylation satellite 2 sequences. In fact, there are not genetic evidence of pericentromeric DNA methylation loss in DNMT3a−/− mice (Okano et al., 1999). The hypothesis of a passive demethylation in our opinion have to be discarded, since the increased expression of DNMT1 in GBM should be sufficient to assure enough enzymatic activity to maintain the inherited DNA methylation pattern.
To monitor the possibility of an increased demethylase expression, hypothesizing an active demethylation, we quantified the expression of the MBD2 demethylase gene by real-time quantitative RT–PCR in both NPC and BTSC model systems but we could not detect changes (data not shown).
At this point, it remains to be clarified if the ‘stem cell’, from which glioblastoma arises, has been correctly imprinted at its origin.
Taking in consideration the expression of de novo DNMTs in the brain, during the fetal and the first period of post-natal life, it is known that in mice DNMT3b can be transiently detected early during neurogenesis, whereas DNMT3a can be monitored in both embryonic and post-natal central nervous system (CNS) (Feng et al., 2005).
In particular, DNMT3a expression in CNS increases during the first 3 weeks of post-natal development suggesting that DNMT3a activity may be crucial for CNS maturation.
Accepting the same behavior also in human, we hypothesize that a neural stem cell could be affected somehow by a low abundant expression of DNMT3a, during the period in which an increased expression is expected, that can contribute to generate a not completely methylated pericentromeric DNA and, at the end, creating an unstable stem cell genomic DNA. The genomic instability inherited through the several generations of an NPC will contribute in our opinion to acquire further genetic alterations during adult life generating a BTSC, considered the origin of human glioblastoma.
The mechanism leading to a global hypomethylated and unstable NPC, and more in general in cancer, is still obscure. However, in our opinion, a missed de novo DNA methylation, instead of a passive or active demethylation, has to be considered to explain the demethylated glioblastoma phenotype: the decreased DNMT3a expression observed could be at least a part of the phenomenon. An effective role of de novo DNMT downregulation in cancer has been recently proved by other authors (Dodge et al., 2005) that report genomic instability and cellular immortalization in DNMT3b-deficient MEF.
Furthermore, early epigenetic changes, as those reported in NPC, are in agreement with the recent point of view that considers the stem cell as the target of initiating events of cellular transformation and that could explain the heterogeneous properties of glioblastoma.
Materials and methods
Human glioblastoma U-373MG, U-87MG and T98G cells were obtained from the American Type Culture Collection (ATTC, Rochville, MD, USA). Cells were cultured in Dulbecco's modified Eagle medium (Cambrex, Walkersville, MD, USA) supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin and 1% glutamine.
Astrocytoma primary FCN-9 and MZC-12 cell lines (WHO grade IV) were established from tumor specimens of patients and cultured as described (Calogero et al., 2004).
NHA were purchased from Cambrex and were cultured in astrocytes basal medium (ABM) enriched with growth supplements, according to the manufacture's instructions.
RNA extraction and DNMT quantitative RT–PCR
RNA isolation, real-time RT–PCR assay and analysis were performed as described (Caprodossi et al., 2005). Primers (Primers listed in Table 3 available as Supplementary material) were designed using Primer Express software (PE Applied Biosystem, Foster City, CA, USA). To specifically amplify DNMT3b splice variants, primers were selected spanning isoform-specific exon–exon boundary. Each sample was analysed using ribosomal protein RPLP0 as endogenous reference gene for mRNA normalization and valuated as fold-induction compared with the normal (not neoplastic) sample.
Qualitative RT–PCR of DNMT3b splice variants was conducted as reported (Saito et al., 2002).
The DNMT1, DNMT3a and DNMT3b sequences were determined by direct sequencing of PCR products, obtained by using different sets of primer (Primers listed in Table 4 available as Supplementary material) performed by the dideoxy chain-termination method with the ABI PRISM BigDye Terminator Cycle Sequencing kit (PE Applied Biosystem), according to the manufacture's instructions.
Satellite 2 DNA methylation analysis
Pericentromeric satellite 2 CpG methylation levels were monitored by Southern blotting analysis (Hassan et al., 2001) by using a satellite 2 probe obtained by PCR in the presence of DIG-11-dUTP as tracer (Roche, Mannheim, Germany).
Protein immunoprecipitation, transient transfection, infection and western blotting
Immunoprecipitations were conducted as described (Fanelli et al., 2004) with two different antibodies able to specifically immunoprecipitate DNMT3a and DNMT3b proteins. Anti-DNMT1 (ab11890), anti-DNMT3a (ab2850), anti-DNMT3b (ab16049 and ab2851) antibodies were purchased from ABCAM (ABCAM, Cambridge, UK). Anti-DNMT3a (NB 100–265) was purchased from Novus (Novus Biologicals, Littleton, CO, USA). However, the anti-DNMT3b (ab2851) showed, in denaturing condition, crossreactivity with DNMT3a protein and was named anti-DNMT3a/3b (Figure 3b).
Stable DNMT3a expression in glioblastoma cell lines was obtained by using a new lentiviral vector called Tween following the experimental procedure described (Ricci-Vitiani et al., 2004).
We thank Dr Jae Hoon Chung for the Dnmt3b expressing vector and all friends of PGP's group of research. This study was supported by a grant of the Ministero della Salute – Progetto di Ricerca Finalizzata (prot. N. DGRST/CRS/RF-2003/1920).