We have previously identified the promoter regions of miRNA genes expressed in normal breast cells represented by HMEC and HMF 
, and in this study we determined which of these promoters are targets for aberrant DNA methylation in breast cancer. DNA methylation of miRNA promoters was analyzed in seven breast cancer cell lines and twelve breast tumor tissue specimens as well as in three cultured normal finite lifespan HMEC strains and five breast non-tumor tissue specimens, using 5-methylcytosine immunoprecipitation coupled to custom tiling microarray covering miRNA gene regions.
Multidimensional scaling analysis of DNA methylation microarray data from all miRNA promoters revealed a link between DNA methylation status of miRNA gene promoters and the phenotype of individual specimens (). In this analysis, the normal state was represented by cultured pure populations of finite lifespan HMEC and HMF, as well as non-tumor breast tissue from surgical specimens. Tumor samples were represented by pure populations of breast cancer cell lines as well as heterogeneous tumor tissues, containing a complex mixture of cell types. shows that normal in vitro and in vivo samples form two clusters on the left side of the multidimensional scaling plot, while the tumor samples are more dispersed. The most separated are the pure cancer cell lines and the pure normal HMEC and HMF strains. Probably due to their heterogeneous nature, the distance between tumor and non-tumor tissues is smaller, nevertheless the direction of the change is the same, indicating similar DNA methylation differences in both cancer cell lines and tumor tissue. To discern possible effects of in vitro culture versus heterogeneity of in vivo tissues we also analyzed the in vitro and in vivo samples separately (). Again there is overall larger difference between cultured cancer cell lines and HMEC or HMF cells than between tumor tissue and non-tumor tissue. Analysis of the in vitro samples alone also shows the physiological DNA methylation differences between HMEC and HMF (). Overall, these observations indicate that the DNA methylation status of a large portion of miRNA promoters is changed in cancer samples.
Microarray analysis of DNA methylation status of miRNA promoters.
We next sought to determine which miRNA promoters are affected by changes in DNA methylation. First, we analyzed all promoters for the presence of DMRs in seven breast cancer cell lines relative to three genotypes of normal pre-stasis finite lifespan HMEC. Almost one third (55 of 167 or 33%) of miRNA promoters active in normal mammary cells became hypermethylated in cancer cells (), while only eleven promoters (6.6%) became hypomethylated. The full list of miRNA promoters with DMRs is included in and Table S1
miRNA promoters with hypermethylated DMRs in breast cancer cell lines.
We have previously shown that normal CTS miRNA expression is often regulated by epigenetic mechanisms including DNA methylation 
. To determine whether the promoters of these CTS miRNAs are more likely to become methylated in cancer, we split the miRNA genes into cell-type specific and those without distinct cell-type specificity to determine the proportions of promoters with DMRs in individual groups of miRNA genes in cancer cells. Most of the promoters (143) belong to the non CTS miRNA genes; they contain hypermethylated DNA in 42 promoters (29%) (, ). CTS miRNA promoters, although lower in number (24), have significantly higher (p-val
0.017) proportion of hypermethylated promoters (13 out of 24 or 54%) (). When these 24 promoters are further split into HMEC and HMF specific, we found hypermethylation in 9 out of 15 (60%) HMEC specific and in 4 out of 9 (44%) HMF specific miRNA promoters, respectively (, ). Overall, these data show that miRNA gene promoters are frequent targets of DNA hypermethylation in cancer. The hypermethylation is common for both CTS and non CTS miRNA genes, although the CTS miRNA genes are hypermethylated more frequently.
We further sought to determine whether the changes in DNA methylation observed in the in vitro cultured cancer cell lines also occur in vivo in breast tumors and therefore may have clinical relevancy. To avoid possible confounding effects from different proportions of differentially methylated normal cell types (HMEC and HMF) in individual tissue samples, we also restricted this analysis to non-CTS miRNA promoters. Hypergeometric test revealed highly significant overlap between non-CTS miRNA promoters containing hypermethylated DMRs in cancer cell lines and those hypermethylated in a group of twelve tumor tissues (). For non-CTS miRNA, 10 out of 42 promoters hypermethylated in cancer cell lines also contained hypermethylated DMRs in tumor tissues group (). Furthermore, the tumor tissue did not show any significant DMRs not represented in the cancer cell lines group (), indicating that cancer cell lines represent most (if not all) of the DNA methylation changes observed in in vivo tumor tissues. Finally, 19 of 32 miRNA promoters that do have significant hypermethylated DMRs in the cancer cell lines group, but are not significantly hypermethylated in the tumor tissue group ( left part), contain hypermethylated DMRs in at least one individual tumor sample (). The remaining 13 promoters that were found to be hypermethylated in cancer cell lines, but were not detected as DMR in the tumor tissue samples, were either not present in the analyzed cohort or were undetectable due to the tissue heterogeneity. Overall these data show that a majority of the miRNA promoters aberrantly methylated in breast cancer cell lines are also aberrantly methylated in in vivo breast cancer samples and may therefore be clinically relevant and not an artifact of tissue culture.
MicroRNA promoters hypermethylated in cancer cell lines are hypermethylated also in tumor tissue samples.
To verify and expand the microarray data, twelve of the differentially methylated miRNA promoters representing those with significant hypermethylation in the tumor tissue group (mir-130a, mir-30e/mir-30c-1, mir-137, and mir-140) as well as those hypermethylated in individual tumor tissue samples only (mir-31, let-7a-3/let-7b, mir-155, mir-148a, mir-34b/mir-34c, mir-99a/let-7c, mir-125b-1, and mir-125b-2) were analyzed by Sequenom MassARRAY (). In addition to the samples analyzed by microarray, six of these promoters (mir-130a, mir-137, mir-140, mir-31, let-7a-3/let-7b, mir-155) were also analyzed on a larger sample set of 7 cancer cell lines and 26 breast tumor specimens ( (top part)). The MassARRAY data in general confirmed the observations from DNA methylation microarray analysis (). Similar increases in DNA methylation were also observed in the additional tumor tissue samples that were not part of the original microarray analysis ( (top part), (top part)). The level of DNA methylation observed in the heterogeneous tissue samples was frequently lower than in the cancer cell lines (), likely due to the cancer cells in tumor tissue being diluted by normal cells lacking the hypermethylation. This situation also makes the detection by microarray of hypermethylated regions in the tumor tissue more difficult and may help to explain the higher number of significant DMRs detected in pure cancer cell lines. Overall, these data confirm the findings from the microarray analysis, and show that these twelve hypermethylated DMRs observed at non-CTS miRNA promoters in in vitro grown cancer cell lines could also be found in in vivo tumor tissue samples, including additional samples independent of the original tumor cohort.
MassARRAY analysis of DNA methylation status of twelve selected miRNA promoters.
Overview of DNA hypermethylation status for 12 miRNA promoters analyzed by MassARRAY.
Polycomb targeted genes are frequent targets of aberrant DNA methylation in cancer 
, and we have previously found that the proportion of polycomb targeted miRNA genes is about three fold greater than protein coding genes 
. Therefore, we examined whether polycomb targeted miRNAs are more likely to be targets of aberrant DNA methylation in breast cancer. To this end we integrated our H3K27me3 data from HMEC 
with publicly available H3K27me3 domain data from embryonic stem cells (ESC) from two independent studies 
. ESC represent here universal ancestors of all cell types including HMEC and HMEC represent ancestors of breast carcinoma cells. We found a significant overlap between miRNA promoters occupied by the polycomb specific H3K27me3 mark in either ESC or HMEC and those aberrantly methylated in cancer cells. The comparisons show that miRNA promoters hypermethylated in cancer cell lines are significantly enriched in regions targeted by polycomb in both ESC and HMEC (). Similar results were obtained when the analysis was restricted to non-CTS miRNA promoters and the hypermethylation occurring in tumor tissue samples (Figure S1
). These observations provide additional evidence for the link between polycomb repression and aberrant DNA methylation in cancer. The predisposition of polycomb target genes to aberrant DNA methylation in cancer, together with the higher frequency of polycomb targets among miRNA genes, might explain the higher than expected proportion of miRNA genes with aberrant DNA methylation observed in breast cancer cells.
MicroRNA promoters occupied by polycomb specific H3K27me3 in normal cells are frequently hypermethylated in cancer.
Aberrant DNA methylation in cancer cells is often linked to gene silencing. We therefore analyzed the expression levels of the aberrantly methylated miRNA genes using quantitative real-time PCR. In general, negative or no correlation between miRNA expression and DNA methylation was observed (), no positive correlations were found. Six out of twelve analyzed miRNA genes show a negative correlation between the level of DNA methylation determined by Sequenom MassARRAY and miRNA expression determined by real-time PCR. The negative correlation is significantly strong for four miRNAs (miR-31, miR-130a, let-7b, miR-155) and moderate or weak for miR-137 and miR-34c respectively (). The other six miRNA genes did not display a significant relation between DNA methylation and miRNA expression level. This may be due to the tissue heterogeneity, where the presence of normal cells could obscure the detected miRNA level in tumor samples. In addition to DNA methylation, other factors like variable level of transcription factors in individual cancers may also be involved in variation of miRNA levels in these cases. Overall, these data confirm that the aberrant DNA methylation of miRNA gene promoters is linked to silencing of miRNA genes in cancer.
Aberrant DNA methylation of miRNA gene promoters in cancer is linked to silencing of miRNA expression.