Supporting Information Bloushtain-Qimron et al. 10.1073/pnas.0805206105 SI Text were performed on ontology terms (using data from CGAP, Cell Culture. For 2D culture, cells were plated at a concentration ftp://ftp1.nci.nih.gov/pub/CGAP/Hs࿝GeneData.dat) that are of 5,000–10,000 cells per well in 48-well plates or eight chamber present in the hypomethylated group versus those present in the glass slides coated with collagen or irradiated 3T3 feeder layer. background gene list, which are genes surrounding all uniquely Cells were plated in medium containing 2% Matrigel. Growth mapped AscI sites on the genome. SAGE library analysis was media used: Medium 1: MEBM plus supplements: B27, 20 ng/ml performed similarly but matching the tags to transcripts. Human EGF, 20 ng/ml bFGF, and 4 ␮g/ml heparin (bovine pituitary genome information (March 2006 hg18) were downloaded from extract was excluded); Medium 2: DMEM/F12 plus 1 mM UCSC site (http://hgdownload.cse.ucsc.edu/downloads. glutamine, 5 ␮g/ml insulin, 500 ng/ml hydrocortisone, 10 ng/ml html#human) and used to map all mRNAs immediately close to EGF, 20 ng/ml cholera toxin, and 0.1% BSA; Medium M3: equal AscI sites. This is further modified using a proximal promoter mix of MEBM supplemented with 5.0 ␮g/ml insulin, 70.0 ␮g/ml region of Ϫ5KbtoϪ80 Kb upstream of the transcriptional start bovine pituitary extract, 0.5 ␮g/ml hydrocortisone, 5.0 ng/ml site. When AscI falls inside a proximal promoter or inside a gene, EGF, 5.0 ␮g/ml transferrin, 10 nM isoproterenol, 2.0 mM any additional relations to other genes not meeting the same glutamine, and DMEM/F12 supplemented with 10 ␮g/ml insulin, criteria are rejected. This consistently brings down mapped gene 10 nM tri-iodothyronine, 1.0 nM 17b-estradiol, 0.1 ␮g/ml hy- number from 6,000 to 4,300 genes. A histogram of distances of drocortisone, 5 ng/ml EGF, 2 mM glutamine, 0.5% FCS and 4 AscI in relation the transcriptional start sites were made using ml of AlbuMAX (5% solution in water) with oxytocin added to the Ϫ20 Kb proximal promoter setting. Result is arbitrarily 0.1 nM final concentration. For immunofluorescence, cells were divided into 4 groups representing 3 peak AscI location groups fixed in 4% formaldehyde in PBS for 20 min at room temper- (blue, red, green) and a remaining distal AscI group (yellow). A ature (RT) and permeabilized in 0.1% Triton X-100 for 5 min. pie chart is drawn to show the percentage composition of each Nonspecific binding was blocked with 10% goat serum for 30 group. min., followed by incubation with primary antibody for1hand then with secondary antibody for 45 min at RT. To visualize Suz12 Target Enrichment. After merging and noise removal, SAGE nuclei, DAPI was included in secondary antibody incubation libraries (CD44, CD24, CD10, MUC1, in pairs) were subjected step at 0.2 ␮g/ml final concentration. to Poisson analysis (http://genome.dfci.harvard.edu/sager/). SAGE libraries are also normalized and ratio between CD44ϩ Immunofluorescence. For immunofluorescence, cells were fixed in cells and other group were calculated for each tag. Tags were 4% formaldehyde in PBS for 20 min at room temperature (RT) mapped to best genes using CGAP data file (ftp:// and permeabilized in 0.1% Triton X-100 for 5 min. Nonspecific ftp1.nci.nih.gov/pub/SAGE/HUMAN/) and gene list merged binding was blocked with 10% goat serum for 30 min, followed based on lower p value from the Poisson analysis. Differentially by incubation with primary antibody for 1 h and then with expressed genes were ordered evenly as a spectrum going from secondary antibody for 45 min at RT. To visualize nuclei, DAPI red (down-regulated genes) to black (no change genes) and to was included in secondary antibody incubation step at 0.2 ␮g/ml green (up-regulated genes). Yellow bars were used to mark final concentration. 2-fold threshold. Suz12 target genes were marked in black bars and their densities (sum of target gene numbers in a window of Analysis of SAGE and MSDK Data. To identify differentially meth- 2% total gene number) were appended. Fisher exact tests were ylated tags MSDK libraries were merged and differentially performed starting from two ends using a starting window with methylated tags were determined based on Poisson analysis genes above the threshold, then moving toward midpoint using (http://genome.dfci.harvard.edu/sager/, P Ͻ 0.05). MSDK tags the same size windows, testing the enrichment of Suz12 targets were mapped to genome (National Center for Biotechnology inside the windows. Information build 36) and genes nearest to tag associated AscI were collected. Ratios were drawn across the genome and where Pathway Map and Network Analyses Using METACORE. For enrich- two or more ratios were involved a line is drawn between these ment analysis of gene lists in functional gene ontology categories ratios to denote they are linked to the same AscI site. Ratio lists of differentially methylated or expressed genes were ana- points for genes of interest were circled and genes labeled. To lyzed for relative enrichment in certain functional ontology calculate an arbitrary hypomethylation score for each MSDK categories in Metacore, including GO and GeneGo cellular library, every AscI site on genome (National Center for Bio- processes, and canonical pathways maps. P values were calcu- technology Information build 36) was accessed for its methyl- lated using a basic formula for hypergeometric distribution ation status based on MSDK tag counts. One AscI site can give where the P value essentially represents the probability of raise to up to two MSDK tags. When either one of these tags has particular mapping arising by chance, given the number of genes a count of greater than or equal to 2, the site is marked as in the set of all genes on maps/networks/processes, genes on a hypomethylated. A total of 6,079 MSDK tags were mapped to particular map/network/process and genes in the experiment. 3,517 AscI sites within all MSDK libraries generated. The For network visualization and analysis sets of genes differentially hypomethylation score for each library is then calculated as the methylated or expressed in the four normal mammary epithelial sum of all hypomethylated AscI sites in one library. MSDK that cell types were uploaded into the Metacore analytical suite are significantly hypomethylated in one MSDK library compared version 4.2 (GeneGo, Inc. St. Joseph, MI). with another were collected based on Poisson analysis (http:// genome.dfci.harvard.edu/sager/, P Ͻ 0.05) and normalized ratios Immunohistochemistry. For double staining formalin-fixed, par- (cutoff at 2-fold difference). These tags were mapped to unique affin-embedded 5-␮M sections were mounted on glass slides, AscI sites on genome. Hypomethylated genes were then inferred deparaffinized, and rehydrated through graded alcohols. For as genes surrounding the hypomethylated AscI sites. To find antigen retrieval, sections were immersed in 10 mM citrate genes enriched in hypomethylated subgroup, Fisher exact tests buffer (pH 6.0) and heated in a pressure cooker (120°C for 5 min, Bloushtain-Qimron et al. www.pnas.org/cgi/content/short/0805206105 1of31 followed by 90°C for 10 seconds). Slides were blocked with and Data Set 2 (from http://www.rii.com/publications/2002/ hydrogen peroxide, avidin, biotin, and serum, and incubated nejm.html and http://microarray-pubs.stanford.edu/wound࿝NKI/ with anti-HoxA11 antibody at 1:200 overnight at 4°C. Slides were explore.html). The clinical data table for Data Set 1 listed incubated with biotinylated secondary antibody followed by sample 625 as having a relapse, but this was changed to relapse- streptavidiperoxidase complex (Biogenex, San Ramon, CA) and free because all information on the National Center for Bio- 3,3Ј-diaminobenzidine as a substrate. Blocking steps were re- technology Information database indicates that this patient is peated and slides were incubated with anti-CD44v6 antibody relapse-free. Only lymph node-negative tumors of patients who (clone VFF-18, Millipore) at 1:400 for1hatroom temperature. had not undergone chemotherapy or hormonal treatment were Detection with secondary antibody was repeated and slides were included in our analyses since Data Set 1 only contained patients incubated with VIP substrate (Vector Labs, Burlingame, CA). of this type. Information about the microarray probes down- Counterstaining was performed with Methyl green. loaded with the data was used to map them to official National Center for Biotechnology Information gene symbols. Data were Statistical Analyses. To determine association of methylated sta- log2-transformed, any flagged expression values were removed, tus of the three genes (PACAP, SLC3A9R1, and FOXC1) and and remaining values were median-centered by array. tumor ER/PR/HER2 and BRCA1/2 characteristics and to ana- Microarray probes in the two datasets corresponding to genes lyze association of methylated status of two additional genes identified as differentially expressed and methylated between (LHX1 and HOXA10) and tumor BRCA1/2 characteristics. CD44ϩ cells and other normal cell types in this study (CD24ϩ, Kruskal-Wallis tests were used to evaluate whether distributions MUC1ϩ, and CD10ϩ cells) were identified. Data for these of gene MSP methylation values were different among different probes were median-centered by gene. Then, only genes with tumor ER/PR/HER2 categories. Wilcoxon rank sum tests were probes with expression values Ͼ(mean ϩ standard deviation) or used to evaluate if distributions of gene MSP methylation values Ͻ(mean – standard deviation) in at least 10% of each dataset were different between BRCA1 and BRCA2. Each of the three were kept for further analyses.