DNA methylation plays an important role in normal cells, being involved in X-chromosome inactivation, imprinting and repression of repetitive elements such as retrotransposons and endogenous retroviruses 
. At the same time, CpG islands in the promoter region of single-copy genes are methylation-free, which is important to allow transcription. In cancer, a reverse scenario is found, with single-copy CpG island hypermethylation and genome-wide hypomethylation. The aim of the present work was to determine the relationship between these two abnormal events, using cancer cell lines from several tissue-types and primary colorectal tumors as a model.
While genome-wide demethylation and single-copy CpG island hypermethylation occur in cancer, it is poorly understood whether these two alterations are linked. Our data based on methylation levels of repetitive elements, using both a specific assay for LINE-1 methylation analysis and a microarray platform comprising almost 800 repetitive DNA sequences from different classes, show that those tumors with the highest levels of aberrant hypermethylation (CIMP+/MSI+), also showed the lowest levels of genome-wide hypomethylation, compared to normal adjacent mucosa. Indeed, these tumors showed frequent increase in methylation of repetitive elements, as revealed by both LINE-1 and microarray analysis. Interestingly, CIMP+/MSI-and CIMP-/MSI-showed higher levels of LINE-1 hypomethylation, reinforcing the uniqueness of CIMP+/MSI+tumors. Although, CIMP+/MSI-and CIMP-/MSI-groups were equally hypomethylated for LINE-1, suggesting that microsatellite instability is the main molecular alteration associated with lack of LINE-1 hypomethylation. These findings are concordant with previous reports of lack of global hypomethylation in microsatellite unstable tumors 
. The consensus interpretation of these data is that colorectal tumors arise from two distinct progression pathways: global hypermethylation with microsatellite instability and global hypomethylation with chromosome instability. However, it is necessary to note that the CIMP+/MSI+group is not only characterized by microsatellite instability, but also for a higher frequency of hypermethylated CpG islands. Indeed, the causative factor of the observed microsatellite instability is the exclusive hypermethylation of MHL1 in these tumors, and other genes like p16 and THBS1 are also found more frequently methylated in CIMP+/MSI+compared to CIMP+/MSI-
. In addition, the microarray analysis of repetitive DNA supported the existence of a group of tumors (CIMP+/MSI+) under strong pressure to de novo methylation of both CpG island promoters and repetitive elements and re-classified colorectal tumors into their known CIMP/MSI groups. Notably, CIMP+/MSI-and CIMP-/MSI-were mostly clustered apart, suggesting that the microarray analysis revealed some special features of each group not seen by LINE-1 bisulfite-pyrosequecing. For example, SINE repeats show a gradual change in hypermethylation/hypomethylation according the CIMP/MSI, with CIMP+/MSI-being an intermediate group. Also, satellite repeats (mainly represented by centromeric and pericentromeric repeats) showed a more stable pattern of methylation and it maybe related to their functional role in chromosome segregation during cell division. By contrast, Ehrlich et al. 
found that hypomethylation and hypermethylation are independent in ovarian cancers, based in the capacity of these alterations to predict the degree of malignancy in ovarian tumors. However, direct comparisons of hypomethylation in cancers with and without high levels of methylation (i.e. CIMP) were not studied. More studies are necessary to answer why difference repetitive elements classes have different susceptibility to DNA hypomethylation. In general, our microarray data suggest that some classes of repetitive elements can be subject to the same methylation pressures exerted on CpG islands on CIMP+cancers.
Using LINE-1 methylation as a surrogate for global demethylation, we found a large variation in methylation levels between different cancer cell lines, with tissue specificity. Some tissues like breast, CNS and lung undergo marked LINE-1 demethylation in cancers in a fairly homogeneous fashion. In other tested tissues, like colon and leukemia, some cell lines had methylation levels similar to those exhibited by normal colon and blood tissues, while others were profoundly demethylated. Although a follow-up study including normal samples from the same studied tissues is required to confirm this observation, an analysis performed by Chalichagorn et al. 
did not showed a marked difference in methylation between normal samples from various tissues. Similarly to our results, a previous study by Florl et al. 
had found a marked difference between bladder and renal carcinomas, with only the first exhibiting LINE-1 demethylation. The large variation in global hypomethylation observed implicates non-stochastic mechanisms for this defect, and also suggests a selective advantage for tumors with severe hypomethylation. Indeed, recent experiments show that tumor formation is induced in mice after global genomic hypomethylation 
. Using conditional transgene technology to reduce expression of DNMT1, Gaudet et al. 
observed spontaneous formation of T-cell lymphomas with acquisition of additional genomic changes. Holm et al. 
generated mice that mimic loss of imprinting (LOI), presumed to be due to hypomethylation, and in these animals tumor formation was also observed. The causes of such differential demethylation among cancer of different tissues are unknown, and both genetic and exposure factors may play roles in this. Profound hypomethylation, as observed in the cell line K562 and others could be related to specific loss of function of genes that control methylation of repetitive elements. Candidate genes are those coding for proteins that have been described to exert function as “heterochromatin guardians”. For example, the LSH protein, a member of the SNF2/helicase family proteins, is required for genome-wide methylation. Knockout mice for the Lsh gene displayed perinatal mortality and showed marked demethylation of repetitive elements that is independent from alterations in RNA levels of DNMT1 
In summary, our results show that genome-wide hypomethylation is highly variable in cancer cells, as is single-copy CpG island hypermethylation. Both alterations can be found in the tumors and each one can promote tumorigenesis by independent processes. Our study also provides evidence for a strong inverse link between global hypomethylation and microsatellite instability in cancer.