High-Definition DNA Methylation Profiles from Breast and Ovarian Carcinoma Cell Lines with Differing Doxorubicin Resistance

Acquired drug resistance represents a frequent obstacle which hampers efficient chemotherapy of cancers. The contribution of aberrant DNA methylation to the development of drug resistant tumor cells has gained increasing attention over the past decades. Hence, the objective of the presented study was to characterize DNA methylation changes which arise from treatment of tumor cells with the chemotherapeutic drug doxorubicin. DNA methylation levels from CpG islands (CGIs) linked to twenty-eight genes, whose expression levels had previously been shown to contribute to resistance against DNA double strand break inducing drugs or tumor progression in different cancer types were analyzed. High-definition DNA methylation profiles which consisted of methylation levels from 800 CpG sites mapping to CGIs around the transcription start sites of the selected genes were determined. In order to investigate the influence of CGI methylation on the expression of associated genes, their mRNA levels were investigated via qRT-PCR. It was shown that the employed method is suitable for providing highly accurate methylation profiles, comparable to those obtained via clone sequencing, the gold standard for high-definition DNA methylation studies. In breast carcinoma cells with acquired resistance against the double strand break inducing drug doxorubicin, changes in methylation of specific cytosines from CGIs linked to thirteen genes were detected. Moreover, similarities between methylation profiles obtained from breast and ovarian carcinoma cell lines with acquired doxorubicin resistance were found. The expression levels of a subset of analyzed genes were shown to be linked to the methylation levels of the analyzed CGIs. Our results provide detailed DNA methylation information from two separate model systems for acquired doxorubicin resistance and suggest the occurrence of similar methylation changes in both systems upon exposure to the drug.


Introduction
This study was designed to investigate epigenetic alterations which arise from treatment of tumor cells with the anthracycline antibiotic, doxorubicin [1].Early stage and metastatic breast cancer, as well as platinum-refractory/-resistant ovarian cancer is commonly treated by means of liposomal doxorubicin either as monotherapy or in combination with other chemotherapeutic drugs [2,3].Acquired resistance, however, frequently prevents successful doxorubicin treatment of those diseases.Epigenetic alterations are potential driving forces for acquired chemoresistance [4].A typical epigenetic modification, which is frequently observed in tumor cells, is aberrant methylation of cytosine bases (C) located 59 of a guanine base (G), so called CpG dinucleotides [5].Although CpG dinucleotides are generally underrepresented in mammalian genomes, they frequently cluster around the transcription start site (TSS) of genes, in genomic areas referred to as CpG islands (CGIs) [6].Genome-wide hypo-methylation, in combination with CGI specific hyper-methylation, is a common hallmark of cancer development [7].Hyper-methylation of CGIs located in the promoter region of a variety of genes implicated in cell cycle, invasion, apoptosis, DNA repair and drug transport has been linked to transcriptional silencing of the associated genes [4,8].Probably the most prominent gene involved in drug resistance being transcriptionally regulated via CGI methylation is ABCB1 (MDR1), encoding the drug efflux transporter Pglycoprotein.A number of studies have shown that overexpression of ABCB1 can render cell lines resistant to a wide range of chemotherapeutic drugs, including doxorubicin [9,10].However, not only drug transporters but also, genes involved in metabolizing drugs, repairing the cellular damage caused by them as well as inducing apoptosis in cells that have been irreparably damaged play a crucial role in the development of drug resistance, [11,12].
Based on these considerations, twenty-eight genes whose levels of expression were previously linked to resistance of different cancer types to DNA double strand break (DSB) inducing drugs were selected from literature.Furthermore, each of those genes was selected to contain one or more CGIs close to its TSS.By means of microarray hybridization, high-definition methylation profiles were recorded covering thirty-three CGIs associated with the twenty-eight selected genes.We used this method to determine methylation profiles from five carcinoma cell lines representing two cancer types commonly treated via the DSB inducing chemotherapeutic drug doxorubicin.
In order to study the effects of acquired doxorubicin resistance on DNA methylation in breast cancer, we examined methylation levels in the cell lines MCF-7_wt and MCF-7_ADR.Seeing as MCF-7_ADR was a doxorubicin selected sub-line of MCF-7_wt, both cell lines originally exhibited identical genetic as well as epigenetic backgrounds.Acquired changes in the epigenome of MCF-7_ADR thus were attributable to the effects of doxorubicin selection.We further examined CGI methylation levels in the ovarian carcinoma cell lines OVCAR-4, OVCAR-5 and NCI/ ADR-RES.Unlike MCF-7_wt/_ADR, each of those three cell lines originated from different patients and therefore exhibited dissimilar genetic and epigenetic backgrounds.Moreover, OV-CAR-4 and OVCAR-5 both represented non-doxorubicin selected cell lines, hence differences in their doxorubicin tolerance derived from inherent resistance.NCI/ADR-RES, on the other hand, constituted a doxorubicin selected sub-line originating from the ovarian carcinoma cell line OVCAR-8 [13] and served as a model for acquired doxorubicin resistance in ovarian carcinoma.
In detail, the questions addressed by this study were (1) how does doxorubicin treatment alter DNA methylation in relevant CGIs in-vitro, (2) how do these changes compare between different carcinoma cell lines and (3) are detected methylation changes linked to altered gene expression.

Doxorubicin tolerance of investigated cell lines
For each carcinoma cell line investigated in this study, viability assays were performed in order to determine their tolerance against doxorubicin (Figure 1).As expected, IC 50 values differed significantly and documented the cells resistance status.

Detection of CpG methylation levels
We employed microarray technology for the detection of methylation levels from CpG sites.For that purpose, CGIs of interest were PCR amplified from sodium bisulfite converted genomic DNA (gDNA) of each cell line, changing unmethylated CpG dinucleotides into TpG while leaving methylated ones unchanged [14].The exact primer sequences and annealing temperatures used for PCR amplification are shown in Table S1.From each sample, pools of labeled PCR-fragments were hybridized to microarrays containing 25 nucleotide long probe sequences representing the formerly unmethylated TpG-as well as methylated CpG-containing sequences.The ratio (CpG/(CpG+ TpG))6100 calculated from both probe signal intensities provided a measure of the methylation level for each CpG site in percent [15].In order to determine the potential of each probe sequence to detect different levels of methylation from the selected CpG sites, we hybridized control pools of in-vitro methylated as well as unmethylated PCR fragments to individual microarrays.Only probe sequences exhibiting a methylation ratio above 75% for the fully methylated control, together with less than 25% for the unmethylated control were considered for subsequent analysis.The results from two replicates of independently labeled and hybridized control pools presented in Figure 2A illustrate the high reproducibility of the method (r 2 = 0.9879).Figure 2B further gives an overview of the methylation profiles obtained from three CGIs associated with the genes DNAJC15, ESR1 and GSTP1 respectively.Cytosine methylation levels from CpG sites within the three different CGIs presented in Figure 2B were additionally quantified via sequencing of nine or more sub-cloned PCR products from the two cell lines MCF-7_wt and MCF-7_ADR.In total, the methylation levels from 130 CpG sites were determined by means of microarray hybridization as well as sequencing and revealed a strong correlation with a coefficient of r 2 = 0.9258, indicating the high accuracy of the described method.Methylation profiles detected via hybridization compared to sequencing analysis are shown in Figure 2C.

CGI methylation changes linked to acquired doxorubicin resistance in breast carcinoma cells
We identified CGIs associated with thirteen genes to display changes in methylation levels between MCF-7_wt and MCF-7_ADR.Table 1 summarizes the tendencies of methylation changes between both breast carcinoma cell lines as well as the three investigated ovarian carcinoma cell lines, with plus (+) indicating hyper-methylation with acquired doxorubicin resistance and minus (2) indicating hypo-methylation.CGIs associated with the genes ABCG2, APAF1, ARHGEF2, AVEN, BAD, BIRC5, CDKN2A, FANCF, FOXO3A, MLH1, MSH2, PTEN and RALBP1 did not show detectable levels of methylation in any cell line and hence are not listed in Table 1.Table S2 summarizes the determined methylation levels from all analyzed CpG sites in each of the five cell lines.

Overlapping tendencies between breast and ovarian carcinoma cell lines
When comparing the alterations in methylation levels between the cell lines MCF-7_wt/_ADR to changes observed in the ovarian carcinoma cell lines, we found seven CGIs to display the same tendencies.Specifically, we identified hyper-methylation in CGIs associated with BRCA1, CDH1, DNAJC15 and SULF2 as well as hypo-methylation for ABCB1, APC and HIC1 with increased doxorubicin tolerance (Table 1).Interestingly, the same tendencies were observed in OVCAR-4 (inherent resistance) as well as in NCI/ADR-RES (acquired resistance) when compared to OV-CAR-5.Additionally, we detected hypo-methylation in a CGI linked to IGFBP3 with increased doxorubicin tolerance, which was not observed in breast carcinoma cell lines.

Gene expression levels of CGI associated genes
In order to determine the impact of CGI methylation on the expression of associated genes, we detected mRNA levels of a subset of genes via qRT-PCR.The normalized results relative to expression levels in MCF-7_wt are summarized in Table S3.
In the case of the breast carcinoma cell lines MCF-7_wt/_ADR, the methylation status of CGIs associated with the genes ABCB1, BRCA1, CDH1, DNAJC15, ESR1, GSTP1, PLAU, SULF2 and TGM2 was connected to the expression of the genes.This was assumed to be the case for concomitant mRNA up-regulation after CGI hypo-methylation as well as mRNA down-regulation after CGI hyper-methylation.In case of the ovarian carcinoma cell lines OVCAR-5, OVCAR-4 and NCI/ADR-RES, the methylation status of CGIs associated with the genes ABCB1, BRCA1, CDH1, DNAJC15, and SULF2 was connected to the expression of the genes.
In addition to genes linked to differentially methylated CGIs, we determined mRNA levels of the gene ABCG2, encoding a putative doxorubicin efflux transporter [16].The cell lines most sensitive to doxorubicin (MCF-7_wt, OVCAR-5) expressed much higher levels of ABCG2 than cell lines more resistant to the drug (MCF-7_ADR, OVCAR-4, NCI/ADR-RES).Consequently, ABCG2 expression is not likely to play a significant role in the mechanism causing resistance in the examined cell lines.We further determined TOP2A transcript levels; a primary target of doxorubicin [17,18], and found slightly decreased mRNA levels in doxorubicin resistant cell lines, which might contribute to their resistance (Table S3).

Discussion
Comparison of CGI methylation profiles from the breast carcinoma cell line MCF-7_wt with profiles from its doxorubicin selected counterpart MCF-7_ADR, allowed us to attribute changes in methylation levels at specific CpG sites to acquired doxorubicin resistance.Between both cell lines, we found aberrant CGI methylation profiles linked to thirteen out of twenty-eight genes.The identified genes are involved in drug transport and detoxification (ABCB1 [9,19,20,21,22], DNAJC15 [23,24], GSTP1 [19,21,22], RAB6C [25,26]), DNA damage repair (BRCA1 [27,28]) as well as tumor cell proliferation/invasion (APC [29,30,31], CDH1 [29,32], ESR1 [12,33,34], HIC [29], PLAU [22,35], RASSF1 [29,30,31], SULF2 [36,37], TGM2 [12,38]).An overlapping set of seven genes (ABCB1, APC, BRCA1, CDH1, DNAJC15, HIC1 and SULF2) displayed the same methylation changes in the examined set of ovarian carcinoma cell lines.An overview of methylation tendencies with acquired doxorubicin resistance is given in Table 1.For clarity reasons, not all of the identified alterations are explicitly discussed in the following sections, but rather a selection of those which illustrate the most important findings of this study.The full data sets from each analyzed cell line, consisting of methylation levels from 800 CpG sites as well as the mRNA levels from a subset of analyzed genes are summarized in Tables S2 and S3 respectively.Hyper-and hypo-methylation events occur in distinct CGI sub-regions with increased doxorubicin resistance An important issue when analyzing DNA methylation levels is the heterogeneity displayed by many CGIs [39].While some of the analyzed islands exhibited almost identical levels of methylation at each CpG site, some others showed considerable variation in methylation levels between distinct CGI sub-regions, as was the case for DNAJC15 and ABCB1 (Figure 3).For that reason, it is essential to study changes in CGI methylation by means of a method that allows high-definition analysis.Methylation profiles from CGIs associated with DNAJC15, a putative inhibitor of ABCB1 transcription [40], displayed high levels of methylation between 200 nt upstream and 200 nt downstream from the TSS in all analyzed cell lines (Figure 3).In the region between 200 nt and 400 nt downstream from the TSS, however, methylation levels differed dramatically between sensitive cell lines and cell lines with acquired doxorubicin resistance.In MCF-7_wt as well as in OVCAR-5, methylation in the latter region was almost absent and DNAJC15 mRNA levels were high (Figure 3).In the more resistant cell lines, the complete CGI was found to be hypermethylated and, accordingly, mRNA levels were significantly lower.These findings were consistent with results from Strathdee et al. [23], who identified methylation of the same CGI within the first exon of DNAJC15 to be responsible for the gene's transcriptional regulation.Moreover, they linked loss of DNAJC15 expression to resistance to the chemotherapeutic drug cisplatin in an ovarian carcinoma cell line as well as in ovarian carcinoma patients [24].Here we provided additional evidence for the epigenetic regulation of DNAJC15 expression in breast and ovarian carcinoma cell lines and further demonstrated the correlation of the gene's loss-of-expression with resistance to doxorubicin.
Moreover, consistent with the proposed inhibitory function of DNAJC15 on the expression of ABCB1 [40], we found increased levels of ABCB1 mRNA in cell lines with decreased DNAJC15 levels (Figure 3).One exception, however, was the inherently resistant cell line OVCAR-4, which exhibited 10-fold decreased DNAJC15 mRNA levels compared to OVCAR-5, but almost identical levels of ABCB1.This finding might be explained by epigenetic silencing of ABCB1 in OVCAR-4, preventing expression despite reduced levels of its inhibitor DNAJC15.One possible explanation of the gene's transcriptional silencing in OVCAR-4 was given by its CGI methylation profile shown in Figure 3.While CpG sites upstream of 500 nt from the TSS showed no methylation, downstream CpG sites were highly methylated.Hence the identified region is of potential relevance for the epigenetic silencing of ABCB1 expression.
We further identified CGIs linked to the genes ESR1, HIC1, IGFBP3, SULF2, TGM2 and TP73 to exhibit distinct methylation sub-regions similar to the ones observed in DNAJC15 and ABCB1 (Table S2).Taken together, these findings highlight the importance of high-definition profiling for the precise mapping of drug resistance associated changes of DNA methylation.

CGI associated with GSTP1 becomes partially hypomethylated with acquired doxorubicin resistance
The enzyme, glutathione S-transferase P1, encoded by the gene GSTP1, has long been known to conjugate drugs, including doxorubicin, with glutathione, resulting in their detoxification [41].Its expression has previously been linked to doxorubicin resistance in ovarian carcinoma cell lines and patients [42,43].We found the CGI located within the gene's first exon to display continuously high levels of methylation in the doxorubicin sensitive cell line MCF-7_wt, decreasing to about 50 percent in MCF-7_ADR (Figure 2B).These findings were confirmed via clone sequencing (Figure 2C) and thus further illustrate the presented method's potential to quantify levels of DNA methylation.
Seeing as the cell line MCF-7_ADR derived from one single clone, the existence of two sub-populations, each carrying exclusively methylated or unmethylated copies of GSTP1 associated CGIs, is unlikely.A more plausible explanation would be hypo-methylation of the CGI associated with one of two copies of GSTP1 during the doxorubicin selection process, while the second one remained methylated.Concomitantly, GSTP1 mRNA levels in MCF-7_ADR were strongly increased when compared to MCF-7_wt, similar to the levels observed in each of the three completely unmethylated ovarian carcinoma cell lines (Figure 4).These findings suggest DNA methylation in the CGI of the gene's first exon to be involved in the transcriptional regulation of GSTP1.While GSTP1 mRNA levels are highly different between MCF-7_ADR and MCF-7_wt; they are largely the same between the analyzed ovarian carcinoma cell lines.These results argue against GSTP1 expression levels being a major determinant of doxorubicin tolerance in OVCAR-5, OVCAR-4 and NCI/ADR-RES.Table 1.Tendency of methylation level alterations linked to doxorubicin resistance.

Acquired doxorubicin resistance decreased methylation of hyper-methylated tumor markers
CGIs associated with three putative tumor suppressor genes APC, HIC1 and RASSF1 have previously been shown to be frequently hyper-methylated in breast and ovarian tumors when compared to healthy tissue [29,30,31].Furthermore, promoter hyper-methylation of APC and RASSF1 has been demonstrated to be a prognostic factor for the survival of breast cancer patients [30].Accordingly, we detected high levels of CGI methylation for APC, HIC1 and RASSF1 in the breast carcinoma cell line MCF-7_wt and for APC and HIC1 in the doxorubicin sensitive cell line OVCAR-5.Surprisingly, in carcinoma cell lines of acquired doxorubicin resistance (MCF-7_ADR, NCI/ADR-RES), methylation of CGIs from APC, HIC1 and RASSF1 was strongly reduced or completely absent (Table S2).Interestingly, none of the observed changes in CGI methylation levels were linked to changes in mRNA levels of the associated genes (Figure 4).These findings suggest that hypo-methylation of certain CGIs that typically become hyper-methylated during carcinogen-esis might be a common event in breast and ovarian carcinoma cell lines with acquired doxorubicin resistance, but does not lead to re-expression of associated genes.

Interplay between DNA methylation and chromatin remodeling
In addition to genes such as APC, HIC1 and RASSF1 which show altered DNA methylation, but no concomitant change in gene expression levels we found a number of genes to display altered gene expression levels but no concomitant change in DNA methylation.This issue is best illustrated by means of the examined CGI associated with the gene ABCG2.While no DNA methylation could be detected in the analyzed region in any of the cell lines (Table S2), gene expression levels differed significantly between them (Figure 4).In this respect it is important to keep in mind that DNA methylation changes on their own do not regulate gene expression levels, but that this process is tightly coupled to chromatin modifications such as histone acetylation, methylation or phosphorylation [44,45].There is evidence that histone Methylation profiles from CGIs linked to all analyzed twenty-eight genes are shown in Table S2 and expression levels of a subset of those genes in Table S3.doi:10.1371/journal.pone.0011002.g003S3. doi:10.1371/journal.pone.0011002.g004modification and subsequently altered gene expression can precede DNA methylation changes [5,46].Hence, in the example of ABCG2 it is possible that histone modifications influence its expression independently from DNA methylation in the examined region.

Breast and ovarian carcinoma cell lines with acquired doxorubicin resistance display similar methylation profiles
When we compared methylation levels from all 800 CpG sites between all breast and ovarian carcinoma cell lines analyzed in this study, we found methylation profiles from the breast carcinoma cell line MCF-7_ADR and the ovarian carcinoma cell line NCI/ADR-RES to be the most similar (Table 2).Given that both cell lines originated from different cancer types but were selected for doxorubicin resistance, this finding strongly suggested similar changes in methylation patterns during formation of resistance in breast and ovarian carcinoma cells.In comparison, we found no correlation between profiles from the breast carcinoma cell line MCF-7_ADR and its parental counterpart MCF-7_wt (Table 2), further illustrating the profound alterations of methylation patterns during the acquisition of doxorubicin resistance.

Conclusions
The detected profiles represent highly accurate comprehensive pictures of CGI methylation from sets of selected putative drug resistance genes.Pre-selection of genes allowed the detailed analysis of methylation profiles from genes of potential relevance to doxorubicin resistance.It was found that several CGIs exhibited doxorubicin-related hyper-as well as hypo-methylation only at specific CpG positions.These findings illustrate the importance of high-definition profiling as compared to the analysis of only individual CpG sites.

Selection of MCF-7_ADR cells from MCF-7_wt
MCF-7_wt cells were exposed to increasing concentrations of doxorubicin, initially to 2-fold the IC 50 for 24 hours, followed by washing and incubation in drug-free culture medium until new colonies had formed.This procedure was repeated several times, each time doubling the original IC 50 up until 64-fold the original IC 50 was reached.Surviving cells were subjected to a doxorubicin dilution series ranging from 16-to 512-fold the original IC 50 .Cells which proliferated at the highest drug concentration within one week were considered chemotherapy refractory.Resistant colonies were picked from cells treated with 128-fold the original IC 50 and expanded in continuous presence of 10-fold the original IC 50 .The cell line MCF-7_wt and the doxorubicin selected subline MCF-7_ADR was kindly supplied by the laboratory of Dr. Ralf A. Hilger.Before the final cell viability assays, cells were cultured in doxorubicin-free medium for two weeks.

Cell culture conditions and viability assays
MCF-7_wt and MCF-7_ADR cells were cultured in DMEM (10% FCS, 1% penicillin/streptomycin 10,000 U) and OV-CAR-5, OVCAR-4 and NCI/ADR-RES cells were obtained directly from the National Cancer Institute and cultured in RPMI (10% FCS, 1% penicillin/streptomycin 10,000 U) under standard cell culture conditions (37uC, 5% CO 2 ).For viability assays, cells were seeded in triplicate into 96 well microplates at 1,000 as well as 2,000 cells per well.Twenty-four hours later, cells were treated with different concentrations of doxorubicin.Seventy-two hours post treatment, cells were incubated with 50 ml medium containing resazurine (20 mg/ml) and incubated for one to four hours before detection of fluorescence (Ex: 544 nm/Em: 590 nm).After background subtraction, cell viability from every drug concentration was normalized to the untreated control.

Amplification and labeling of sample targets
In order to analyze methylation levels from the cell lines of interest, total genomic DNA (gDNA) was isolated from each cell line using the QIAamp DNA Micro Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions.Following purification, 2 mg of gDNA from each cell line were subjected to sodium bisulfite treatment via the EpiTect Bisulfite Kit (Qiagen) according to the manufacturer's instructions.Each CGI of interest was PCR amplified using the primer pair given in Table S1.Primer pairs were designed using the web-based software MethPrimer [47] and their optimal annealing temperature was determined via gradient PCR.The exact sizes of each PCR product (ranging from 198 nt up to 777 nt in size) as well as Ensembl transcript IDs and exon IDs from the examined sequences are summarized in Table S1.Genomic regions of interest were amplified by means of PCR using 20 ng of purified sodium bisulfite treated template DNA, 0.4 mM forward and 0.4 mM reverse primer, 250 mM of each dNTP (Fermentas, St. Leon-Rot, Germany), 16 HotStart Buffer (Qiagen), 16 Q-Solution (Qiagen), 1.5 mM MgCl 2 , 2.5 units HotStart polymerase (Qiagen) and in a total volume of 50 ml.Thermal cycler PCR conditions were 95uC for 15 min followed by seven cycles of 95uC for 1:00 min, 64uC for 2:00 min, 72uC for 2:00 min with a decreased annealing temperature of 1uC in each cycle before an additional 39 cycles of 95uC for 0:40 min, optimized annealing temperature (see Table S1) for 2:00 min, 72uC for 1:30 min and Table 2. Correlation coefficients (r 2 ) display strongest similarity between methylation profiles from cell lines with acquired doxorubicin resistance.finally 72uC for 10:00 min.PCR products were purified using Millipore MultiScreen PCR m96 filter plates according to the manufacturer's instructions.For every individual cell line of interest, equimolar amounts of all purified PCR fragments were pooled.For labeling, 300 ng from each of those PCR product pools were incubated together with 30 ng/ml random primer oligonucleotides (Invitrogen, Karlsruhe, Germany) in a total volume of 28 ml at 99uC for 5 min.After the denaturation step 16 reaction buffer (1 M Hepes pH 6.6, 250 mM Tris-HCl pH 8.0, 25 mM MgCl 2 , 50 mM 2-mercaptoethanol), 2 mM of each dATP, dCTP, dGTP and 1.3 mM dTTP, (Fermentas) together with 0.7 mM biotinylated-dUTP (Roche, Mannheim, Germany), 0.4 mg/ml BSA (Sigma, Hamburg, Germany) and 7.5 units Klenow fragment (New England Biolabs, Frankfurt, Germany) was added to a total volume of 40 ml.After incubation at 37uC for 3 h and 75uC for 10 min, 4 ml of 3 M sodium acetate (pH 5.6) and 100 ml ethanol were added and the DNA was precipitated at 280uC for 2 h.After centrifugation at 18,3206g for 20 min the supernatant was aspirated, the pellet was dried and resuspended in 15 ml 16 hybridization mix (100 mM 2-[Nmorpholino]ethanesulfonic acid (MES), 0.9 M NaCl, 20 mM Na 2 EDTA, 0.01% (v/v) Tween-20, 0.5% BSA 0.1 mg/ml herring sperm DNA (Febit Biomed, Heidelberg, Germany).

Preparation of control targets
From total human gDNA (Invitrogen), fragments containing each CGI of interest, including the primer binding sites shown in Table S1, were amplified via PCR.From each fragment, 1 mg was incubated at 37uC for 3 h together with 8 units of SssI methyltransferase (NEB, Frankfurt, Germany).To confirm complete in-vitro methylation 200 ng DNA were digested, using 20 units of the methylation sensitive restriction enzyme BstUI (NEB) for two hours at 60uC.From each fully methylated as well as unmethylated target, 1 ng was used for subsequent sodium bisulfite treatment via the EpiTect Bisulfite Kit (Qiagen) followed by a PCR amplification step of each CGI.After purification via QIAquick PCR purification columns (Qiagen), an equimolar pool of fully methylated as well as one of unmethylated control targets were prepared.Both pools were labeled and hybridized the same way as sample target pools.

Microarray analysis
The photo-controlled in-situ synthesis technology Geniom One (Febit Biomed) was used for synthesis, hybridization and detection of microarrays [48].The Geniom One microarray is divided into eight individually accessible subarrays allowing the analysis of eight samples in parallel.Probe sequences the size of 25 nucleotides were synthesized, resembling the sodium bisulfite converted sequence of each CGI of interest.Every probe sequence was designed to feature at its central position a particular CpG site of interest either as a CpG or a TpG dinucleotide, hence being complementary to the methylated (M) or unmethylated (U) cytosine after sodium bisulfite treatment.Consequently, the microarray layout included 1600 different probe sequences in four replicates.Before hybridization the biotinylated target pools in 16 hybridization mix described above, were heated to 95uC for 3 min then placed on ice for 1 min.The denatured targets were then applied to individual subarrays of the Geniom One microarray and incubated at 45uC for 16 h.After washing routines according to the Febit protocol, each subarray was incubated with 5 mg/ml streptavidin phycoerythrin (Invitrogen) in 66 SSPE (0.9 M NaCl, 60 mM NaH 2 PO 4 , pH 7.4 and 6 mM Na 2 EDTA).Signal intensity detection was performed using the inbuilt CCD camera of the system and local backgrounds were subtracted by means of internal Geniom One software routines.Finally, signal intensities from probe sequences were used to determine methylation levels for each CpG site by calculating the CpG/(CpG+TpG) probe signal intensity ratio, representing the methylation level M/(M+U).Consequently, a fully methylated CpG should result in a signal intensity ratio of one whereas a fully unmethylated CpG should return a value of zero.However, owing to unspecific cross-hybridization mostly caused by low sequence complexity of sodium bisulfite treated sequences, not all probe sequences were suitable for further analysis.In order to identify probe sequences that could be used to accurately detect methylation levels from specific CpG sites, in-vitro methylated as well as unmethylated control pools were used for initial calibration.The two control pools were hybridized to individual sub-arrays followed by calculation of M/(M+U) probe signal intensity ratios for every represented CpG site.In the final microarray layout, probe sequence pairs were only included when the hybridization of the fully methylated control pool returned a probe signal intensity ratio above 0.75 and the hybridization of the fully unmethylated control pool returned a probe signal intensity ratio below 0.25.

Clone sequencing
PCR products selected for validation were cloned into pCR4-TOPO via the TOPO TA cloning kit (Invitrogen) and the constructs were transformed into TOP10 cells.From each cloned PCR fragment, twelve different clones were picked and sequenced (GATC Biotech, Constance, Germany).Final methylation levels were determined from nine or more sequences obtained from each PCR fragment.

Quantitative RT-PCR
Total RNA was isolated from each cell line using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instructions.The one-step QuantiFast SYBR Green RT-PCR Kit (Qiagen) was used in combination with primers from the QuantiTect Primer Assay (Qiagen).Reactions were performed in 386 well format in triplicate with 25 ng total RNA per well in a LightCycler 480 (Roche).The endogenous controls ACTB1, GAPDH and TUBA3C were used for normalization.
Table S3 Gene expression levels relative to MCF-7_wt.Gene expression levels from indicated genes were determined from the cell lines MCF-7_wt, MCF-7_ADR, OVCAR-5, OVCAR-4 and NCI/ADR-RES.To allow comparison between cell lines, expression levels were normalized to a set of house-keeping genes within each cell line and are presented as fold-changes from MCF-7_wt.Columns SD show standard deviations from triplicates.

Figure 2 .
Figure 2. Methylation profiling -Reproducibility and validation.A: Reproducibility, illustrated via two replicate hybridizations, of fully methylated and fully unmethylated control fragment pools.B: Methylation profiles determined from CGIs around the TSS of the genes DNAJC15, ESR1 and GSTP1.Shown are hybridizations of fully methylated and unmethylated control fragments as well as fragments amplified from the cell lines MCF-7_wt and MCF-7_ADR.C: Validation of methylation profiles via clone sequencing of the PCR fragments used for microarray hybridizations shown in Figure 2B.doi:10.1371/journal.pone.0011002.g002

Figure 3 .
Figure 3. Hyper-and hypo-methylation events occur in distinct CGI sub-regions with increased doxorubicin resistance.A: A subregion of a CGI associated with the gene DNAJC15 becomes hyper-methylated and the gene's mRNA levels decrease with increased resistance.B: A sub-region of a CGI associated with the gene ABCB1 becomes hypo-methylated and the gene's mRNA levels increase with increased resistance.