Increased Sensitivity of Transforming Growth Factor (TGF) b 1 Null Cells to Alkylating Agents Reveals a Novel Link between TGF b Signaling and O 6 -Methylguanine Methyltransferase Promoter Hypermethylation*

Inactivation of the transforming growth factor b (TGF b )-signaling pathway and gene silencing through hypermethylation of promoter CpG islands are two frequent alterations in human and experimental cancers. Here we report that nonneoplastic TGF b 1 2 / 2 keratinocyte cell lines exhibit increased sensitivity to cell killing by alkylating agents, and this is due to lack of expression of the DNA repair enzyme O 6 -methylguanine DNA methyltransferase (MGMT). In TGF b 1 2 / 2 but not TGF b 1 1 / 2 cell lines, the CpG dinucleotides in the MGMT promoter are hypermethylated, as measured by restriction enzyme analysis and methylation specific polymerase chain reaction. In one unstable TGF b 1 1 / 2 cell line, loss of the wild type TGF b 1 allele correlates with the appearance of methylation in the MGMT promoter. Bisulfite sequencing shows that in the KO3 TGF b 1 2 / 2 cell line nearly all of the 28 CpG sites in the MGMT promoter 475 base pairs upstream of the start site of transcription are methylated, whereas most are unmethylated in the H1 TGF b 1 1 / 2 line. Treatment of the TGF b 1 2 / 2 cell lines with 5-azacytidine causes reexpression of MGMT mRNA and demethylation

Inactivation of tumor suppressor genes is a common feature of cancer development in humans and animal models. There is increasing evidence that methylation of normally unmethylated CpG islands in gene promoters is an important epigenetic mechanism for transcriptional inactivation of tumor suppressor and DNA repair genes (1)(2)(3). One DNA repair gene that is frequently hypermethylated in tumors is methylguanine methyltransferase (MGMT) 1 (4,5). MGMT removes alkyl adducts from O 6 -guanine residues by transferring the alkyl group to an active cysteine residue within its sequence in a reaction that inactivates further enzymatic activity (6). Since O 6 -alkylated guanine can mispair with thymine during replication to cause transversions as well as cross-link with cytosines on the opposite DNA strand (6), cells that are deficient in MGMT activity may be more susceptible to mutation and, hence, cancer development or progression. Supporting this role in cancer development, transgenic animals overexpressing MGMT are resistant to tumor formation induced by alkylating agents (7), whereas MGMT null animals exhibit an increased frequency of methylnitrosourea-induced tumors (8).
TGF␤1 is a member of a large family of multifunctional secreted polypeptides that are potent growth inhibitors of epithelial cells (9). In human cancers and animal models of multistage carcinogenesis, inactivation of TGF␤ signaling through mutations in the receptors (10 -12) or intracellular Smad proteins (13)(14)(15)(16) is associated with accelerated premalignant progression and malignant conversion. In the mouse epidermal carcinogenesis model, TGF␤1 acts as a tumor suppressor since progression of chemically induced benign tumors is associated with loss of TGF␤1 (17,18), and genetic inactivation of the signaling pathway in keratinocytes leads to rapid progression to squamous cell carcinoma (19,20). To understand the mechanism by which loss of autocrine TGF␤ signaling leads to accelerated tumor progression, we established a series of nonneoplastic, spontaneously immortal cell lines derived from newborn mouse TGF␤1ϩ/Ϫ and Ϫ/Ϫ keratinocytes. The TGF␤1Ϫ/Ϫ cell lines had a significantly higher level of gene amplification than controls (21). To explore further the role of TGF␤1 in genomic stability, we have examined the response of TGF␤1ϩ/Ϫ and TGF␤1Ϫ/Ϫ cell lines to different DNA-damaging agents. Our results show that the TGF␤1Ϫ/Ϫ cell lines are specifically more sensitive to cell killing by alkylating agents, and this is due to a lack of expression of MGMT mRNA and enzyme. Southern blot, MSP, and bisulfite sequencing of the MGMT promoter indicates that the lack of expression is due to hypermethylation of CpG islands in the MGMT promoter. This is the first demonstration of a link between TGF␤1 expression and aberrant promoter methylation, and it could have important implications for mechanisms of tumor progression caused by inactivation of TGF␤ signaling.

EXPERIMENTAL PROCEDURES
Cell Culture-The TGF␤1ϩ/Ϫ and TGF␤1Ϫ/Ϫ cell lines are spontaneously immortal, clonally derived non-tumorigenic keratinocyte cell lines isolated from primary epidermal cultures of newborn mice from the TGF␤1Ϫ/Ϫ strain (19,22). Cells were routinely cultured in Earle's minimum essential medium, 8% chelexed fetal calf serum, 0.05 mM CaCl 2 , and antibiotics. Unless indicated, all cell lines were used between passage 20 -35. Balb/c keratinocytes were isolated from newborn Balb/c mice using standard techniques (23) and cultured for 3-5 days before isolation of DNA. For 5-azacytidine treatment, cells were seeded at 1 ϫ 10 6 cells/175-cm 2 tissue culture flask, and exponentially growing cells were treated with 1 M 5-azacytidine (Sigma) for 48 h. After a 3-day recovery period, the treatment protocol was repeated, after which DNA and RNA were isolated. To isolate subclones of the KO3 passage 8 line, cells were seeded at low density in 150-cm 2 tissue culture dishes, and colonies that grew out were ring-cloned, expanded to a T-75 flask, and DNA-isolated. To generate continuous lines, primary TGF␤1Ϫ/Ϫ keratinocytes were cultured in medium containing 10 ng/ml keratinocyte growth factor for several weeks until immortal colonies grew out. These colonies were pooled and passaged twice before seeding at low density and ring-cloning. The TGF␤1 wild type keratinocyte cell line, NHK4, was clonally derived from newborn keratinocyte cultures isolated from p53 Ϫ/Ϫ mice (24). The B8 and M3 TGF␤1 wild type cell lines were derived from newborn epidermis from control mice of the c-fos Ϫ/Ϫ line (25).
Clonogenic Survival Assay-Approximately 500 cells of each cell line were seeded into 60-cm 2 culture dishes and allowed to attach for 24 h. Cells were treated with serial dilutions of the different drugs for 1 h in complete medium, then placed in fresh medium without drug and allowed to proliferate for 7-14 days. For ␥ and UV irradiation, the medium was removed, and irradiation done in a small volume of phosphate-buffered saline. Gamma irradiation was done with a MarkIV cesium source, and UV irradiation was done with a Stratalinker (Stratagene) set to deliver a preset joule/mm 2 . After irradiation, the medium was replaced. Colonies were stained with 0.5% crystal violet, 10% formaldehyde and counted with a dissecting microscope. Only colonies greater than 50 cells were counted. Each treatment was performed in triplicate, and the percent of colonies relative to the untreated cells was determined. For each cell line and treatment, the concentration producing 50% inhibition of colony formation was determined from the doseresponse curve.
MGMT Enzyme Activity Assay-MGMT activity in crude cellular extracts was determined by measuring the transfer of 3 H from [ 3 H]methylated Micrococcus luteus DNA to an acid-insoluble protein fraction (26). Briefly, cellular extracts were prepared by sonication of cell pellets in a buffer containing 50 mM Hepes, pH 7.6, 100 mM KCl, 1 mM EDTA, 5 mM dithiothreitol. 50 -200 g of protein was incubated with the 3 H substrate prepared from purified M. luteus DNA as described (27) at 37°C for 30 min and then precipitated with 1 M perchloric acid. Precipitated protein and DNA was then heated to 70°C for 1 h to hydrolyze precipitated DNA, and after washing, the amount of 3 H remaining in the insoluble fraction was determined by liquid scintillation counting (26). The reaction was linear between 25 and 500 g of cellular extract.
Northern and Southern Blot Analysis-mRNA was prepared from cell lines using the FastTrack mRNA isolation kit (Invitrogen). 2 g of poly(A) RNA was electrophoresed through a 1% formaldehyde-agarose gel and transferred to Nytran filters (Schleicher & Schuel). Northern filters were hybridized to [ 32 P]dCTP-labeled DNA probes in 50% formamide at 42°C. The methylpurine glycosylase cDNA was obtained from ATCC. The human MGMT cDNA was isolated from pKT100 (28). Genomic DNA was isolated from cultured cells according to standard methods (29), and 40 g was restricted with either MspI or HpaII, electrophoresed through a 2% agarose gel, and transferred to Nytran filters. The filters were hybridized to a 1.9-kilobase pair PCR fragment of the mouse MGMT promoter. Primers used for amplification were obtained from the published sequence (30), 5Ј-GGATCCCAGTTCTAAT-TGGGCCTT, downstream 5Ј-CACCAAGATCTGGCACTGAGAAAT.
PCR reactions were carried out in a 20-l volume using 100 ng of modified DNA, 1ϫ PCR buffer (PerkinElmer Life Sciences) 0.2 units of Amplitaq Gold (PerkinElmer Life Sciences) under the following amplification conditions: 94°C for 90 s, 60°C for 90 s; 72°C for 2 min for 35 cycles. With these conditions, no amplification was observed with the wild type primers on modified DNA or with unmethylated/methylated primers on unmodified DNA. Controls without DNA were performed with each set of PCR. PCR products were electrophoresed through a 2% agarose gel and visualized by ethidium bromide fluorescence.
Bisulfite Sequencing-Genomic DNA was treated with bisulfite as above, and a 560-base pair fragment between 1401 and 1960 of the published mouse MGMT promoter sequence (30) was amplified using primers that are independent of methylation state and recognize bisulfite-modified DNA. The PCR primers were 5Ј-TTTAGTTGGGTAGT-GATTGGATTTTTAGTG and 5Ј-CCCCAAAACTCACCAACTTA-CAAACTACAA. The PCR fragment was cloned into pCR2.1 using the TA-cloning method (Invitrogen), and selected clones were sequenced with M13 primers using the dye terminator DNA-sequencing kit (PE Applied Biosystems) with a PerkinElmer ABI Prism 377 DNA sequencer.

Sensitivity of TGF␤1Ϫ/Ϫ Keratinocytes to Alkylating
Agents-TGF␤1ϩ/Ϫ and Ϫ/Ϫ keratinocyte cell lines were treated with different DNA-damaging agents, and the ability of the treated cells to form viable colonies was used as a measure of relative DNA repair capacity. Fig. 1 shows that for UV and ␥ irradiation or cisplatin there was no consistent difference in the IC 50 for inhibition of colony formation between the TGF␤1Ϫ/Ϫ and TGF␤1ϩ/Ϫ genotypes. Similar dose-response curves were obtained with the topoisomerase inhibitors camptothecin or etoposide (data not shown). However, the TGF␤1Ϫ/Ϫ cell lines were 5-fold more sensitive to cell killing by the alkylating agent MNNG than the TGF␤1ϩ/Ϫ lines. Similar results were obtained with methylnitrosourea (data not shown). TGF␤1ϩ/ϩ cell lines derived independently from other transgenic lines (B8, M3, NHK4) had sensitivities to MNNG that were similar to the TGF␤1ϩ/Ϫ cell lines. Since the NHK4 cell line is p53Ϫ/Ϫ (24), the increased sensitivity to alkylating damage is specific to the TGF␤1Ϫ/Ϫ genotype.
Increased Sensitivity to MNNG Due to Absence of MGMT-MNNG and methylnitrosourea produce a high proportion of O 6 -methylguanine adducts that are specifically repaired by the enzyme O 6 -methylguanine DNA methyltransferase (MGMT) (6). Cells that lack this enzyme exhibit increased sensitivity to cell killing by alkylating agents (4). The level of MGMT enzyme activity in the TGF␤1ϩ/Ϫ and TGF␤1Ϫ/Ϫ keratinocyte cell lines was determined using a well characterized assay that measures the ability of crude cellular extracts to transfer a [ 3 H]methyl group from M. luteus DNA to protein (26). Table I shows that all TGF␤1ϩ/cell lines had levels of MGMT activity ranging from 0.102-0.255 pmol/mg of protein. Treatment of these cells with O 6 -benzylguanine, a specific inhibitor of MGMT, eliminated enzyme activity. In contrast, 4/5 TGF␤1Ϫ/Ϫ keratinocyte cell lines had no detectable MGMT activity, whereas in KO6 the levels were 0.005 pmol/mg of protein, barely above the background. Northern blot analysis showed that the lack of MGMT enzyme in the TGF␤1Ϫ/Ϫ cell lines was due to the absence of the MGMT transcript (Fig. 2). There was no difference between the two genotypes in expression of another repair enzyme, methylpurine glycosylase, which removes N-methylpurines and other damaged purines in DNA (33) (Fig. 2). No difference in hybridization pattern was seen when restriction-digested genomic DNA from the TGF␤1ϩ/Ϫ and TGF␤1Ϫ/Ϫ cell lines was hybridized to a MGMT cDNA probe (data not shown). These results indicate that the lack of MGMT expression in the TGF␤1Ϫ/Ϫ cell lines was not due to deletion or rearrangement of the MGMT gene.
Hypermethylation of MGMT Promoter in TGF␤1Ϫ/Ϫ Keratinocytes-A frequent mechanism for inactivation of MGMT expression in human tumors and tumor cell lines is hypermethylation of CpG islands in the gene promoter (3,34). Initial Southern blot analysis of methylation in the MGMT promoter using the methylation-sensitive restriction enzyme pair MspI and HpaII revealed specific methylation of the Ϫ328 HpaII site in two TGF␤1Ϫ/Ϫ cell lines (not shown). To assay the methylation status of the MGMT promoter in all cell lines, we developed a methylation-specific PCR assay for the mouse MGMT promoter (31). Treatment of DNA with bisulfate converts unmethylated cytosines to uracil, whereas methylated cytosines are resistant to this chemical modification (31). Specific primers were generated to distinguish methylated (M) from un-methylated (U) DNA in the mouse MGMT promoter based on sequence alterations following bisulfite modification. Fig. 3 shows that in all of the TGF␤1ϩ/Ϫ lines as well as Balb/c keratinocytes, a PCR product was obtained after bisulfite modification only with the unmethylated PCR primers and not with the methylated primers. Thus some or all of the CpG sites within these primer sequences are unmethylated in the control cells. In contrast, the methylated primer pair yielded a strong PCR product with bisulfite-modified DNA from all of the TGF␤1Ϫ/Ϫ lines, whereas variable levels of PCR product was generated with the unmethylated primer pair. To examine the extent of methylation in the MGMT promoter, DNA from the KO3 and H1 cell lines at passage 35 was subjected to bisulfite modification followed by sequencing of PCR products amplified with non-methylation-specific primers. Of 28 CpG dinucleotides between Ϫ475 and the start of transcription, virtually all were unmethylated in different PCR clones from the H1 cell line. In contrast, between 57 and 82% of the CpG dinucleotides . Cells were plated at clonal density as described under "Experimental Procedures," allowed to recover 24 h, and then treated with the indicated concentration of the drug or irradiated. Colonies, which formed 7-14 days after treatment, were stained and counted, and the number of colonies formed at each dose was determined as a percentage of the untreated control. Each dose was done in triplicate. MNNG treatment was repeated 4 times with identical results. TGF␤1 wild type lines were generated from the epidermis of p53Ϫ/Ϫ mice (NHK4) and from the wild type genotype of c-fosϪ/Ϫ mice (B8, M3).

TABLE I Absence of MGMT enzyme activity in TGF␤1
Ϫ/Ϫ keratinocyte cell lines MGMT enzyme activity (pmol of 3 H removed/mg of protein) was measured in crude cellular extracts of the TGF␤1 Ϫ/Ϫ and ϩ/Ϫ keratinocyte cell lines as described under "Experimental Procedures." Values represent the average of 2-3 independent determinations. MGMT activity was inhibited by O 6 -benzylguanine (BZG), indicating specificity of the enzyme activity measured for MGMT. were methylated in different PCR products sequenced from the KO3 cell line (Fig. 4). Apart from a region just upstream from Ϫ25, which was free of methylation in all clones, virtually the entire promoter was methylated.

5-Azacytidine Causes Demethylation and
Reexpression of MGMT mRNA-To further strengthen the link between methylation of the MGMT promoter and lack of mRNA expression in the TGF␤1Ϫ/Ϫ lines, these cells were treated with 5-azacytidine to block methylation. Fig. 5A shows that after treatment of the TGF␤1Ϫ/Ϫ cell lines with 5-azacytidine MGMT mRNA was reexpressed (Fig. 5A). This correlated with demethylation of the MGMT promoter as measured by an increase in the unmethylated-relative to the methylated-specific PCR product after bisulfite modification and MSP in KO1 KO3 and KO6 DNA (Fig. 5B).
Relationship of TGF␤1 to Methylation Is Indirect and Linked to Immortalization-To examine the correlation between the lack of autocrine TGF␤1 expression and aberrant methylation of the MGMT promoter, we analyzed the MGMT methylation status in the unstable TGF␤1ϩ/Ϫ cell line H7. Previous studies show that the TGF␤1 wild type allele in this cell line is lost with increasing passage (21). Methylation-specific PCR analysis of the MGMT promoter in H7 showed that at passage 9 a PCR product was generated only with the unmethylated primer pair, but by passage 12, a methylated band is evident, and this increases relative to the unmethylated band by passage 24 (Fig. 6). These results suggest that there is a close association between autocrine TGF␤1 expression and hypermethylation of the MGMT promoter. However, short or long term treatment of cells with TGF␤1 did not alter MGMT expression. Table II shows that treatment of Balb/c primary keratinocytes with 0.5 ng/ml TGF␤1 for 48 h did not alter MGMT enzyme activity, nor did a similar treatment induce expression of MGMT in the KO1 and KO3 cell lines. Additionally, treatment of the KO3 cells with exogenous TGF␤1 at 50 pg/ml for up to 1 month did not reduce methylation of the MGMT promoter. These results suggest that the effect of a TGF␤1 null genotype on methylation of the MGMT promoter is indirect or that once established, the methylation pattern cannot be reversed by addition of TGF␤1. To test whether hypermethylation is an inherent property of the TGF␤1 null cells, we examined MGMT promoter methylation by MSP in primary keratinocytes of all genotypes as well as TGF␤1Ϫ/Ϫ cells at different times of culture. Table II and Fig. 7A show that regardless of TGF␤1 genotype, all primary keratinocytes express similar levels of MGMT enzyme and have an unmethylated MGMT promoter. However, with continued passage of the null cells in culture, a faint band was reproducibly detected with the methylated PCR primers in passage 8 TGF␤1Ϫ/Ϫ cells, and this increased in proportion such that by passage 35 the predominant band was with the methylated-specific primers. These results suggest that the methylation state of the MGMT promoter is unstable in the KO keratinocytes. To test this more directly, we subcloned the null keratinocytes at pas-sage 2 and 8 and examined methylation status of these clones by MSP. Fig. 7, B and C, shows that in subclones from both, there is considerable variability in the methylation state of the MGMT promoter. DNA from some clones was completely methylated (clones 1, 3,4,6,8), whereas in others it was unmethylated (clones 2, 9, 10), and the remainder had amplification with both unmethylated-and methylated-specific PCR primers. Thus the methylation state of the MGMT promoter in the TGF␤1Ϫ/Ϫ cells exhibits clonal-and time-dependent variability.

DISCUSSION
Inactivation of the TGF␤ signaling pathway is a common event in the progression of human and experimental cancers. In the multistage skin carcinogenesis model v-ras Ha retrovirus transduced TGF␤1Ϫ/Ϫ primary keratinocytes undergo rapid aneuploidy in culture (35) and progress to squamous cell carcinoma, whereas control genotypes show limited instability and progression in vivo (19). Nontumorigenic cell lines derived from primary cultures of TGF␤1Ϫ/Ϫ keratinocytes also have a high frequency of gene amplification N-(phosphonacetyl)-L-aspartate (21), suggesting that loss of autocrine TGF␤1 signaling results in decreased genomic stability and rapid malignant progression. Here we show that these TGF␤1Ϫ/Ϫ cell lines also exhibit a specific defect in the DNA repair enzyme MGMT, which is crucial to repair of adducts caused by alkylating agents. Relative to the TGF␤1ϩ/Ϫ cell lines, all of the TGF␤1Ϫ/Ϫ cell lines had a 5-fold increase in sensitivity to cell killing by MMNG, which produces O 6 -methylguanine adducts with high frequency. In addition, the TGF␤1Ϫ/Ϫ cell lines did not have measurable MGMT enzyme activity or express MGMT mRNA, the enzyme responsible for repair of this DNA lesion (6). In contrast there was no significant difference in sensitivity to ␥ or UV irradiation, cisplatin, or topoisomerase inhibitors. All of these agents produce lesions that utilize repair pathways distinct from MGMT. There was also no difference in mRNA expression of another repair enzyme, methylpurine glycosylase, between the two TGF␤1 genotypes, pointing to a specific defect in MGMT expression. Using a combination of Southern blot analysis with methylation-sensitive restriction enzymes, methylation-specific PCR, and bisulfite sequencing, we found that the promoter region of the mouse MGMT gene was also heavily methylated in the MGMT-deficient TGF␤1Ϫ/Ϫ cell lines but not in the MGMT proficient TGF␤1ϩ/Ϫ cell lines. Furthermore, treatment of the TGF␤1Ϫ/Ϫ lines with 5-azacytidine caused reexpression of MGMT mRNA expression and demethylation of the promoter. These results are in agreement with many studies of MGMT expression in human cancers and tumor cell lines that show a strong correlation between promoter methylation and silencing of MGMT expression (4,5) and support the idea that hypermethylation of the MGMT promoter specifically in the TGF␤1Ϫ/Ϫ cell lines is responsible for lack of expression.
Bisulfite sequencing revealed that although the MGMT promoter was hypermethylated in the KO3 cell line, there was heterogeneity between individual DNA clones sequenced and regions of both frequent and rare methylation. Similar heterogeneity of methylation at specific CpG sites between individual DNA clones was also found by bisulfite sequencing of the human MGMT promoter from the Mer Ϫ BE colon tumor cell line (34). Whether this reflects cellular or allelic variability is not clear, but it must reflect an inherent variability in the process of methylation itself. The regions from Ϫ450 to Ϫ100 and Ϫ25 to ϩ25 were highly methylated in all KO3 DNA copies sequenced, whereas there was infrequent methylation in the region from Ϫ80 to Ϫ30. It is remarkable that a similar regional pattern of CpG methylation is found in the MGMT

FIG. 3. MSP analysis of MGMT promoter CpG methylation.
A, genomic DNA from the indicated TGF␤1ϩ/Ϫ and TGF␤1Ϫ/Ϫ cell lines, and Balb/c keratinocytes (B) was modified with bisulfite as described under "Experimental Procedures" and analyzed for methylated CpG sites using PCR primers, which distinguish unmethylated (U) and methylated (M) sequences. Amplification did not occur with either primer set using unmodified DNA. PCR products were analyzed on a 2% agarose gel and visualized by ethidium bromide fluorescence.
promoter of the BE and HeLa S3 tumor cell lines, with a region of rare methylation between Ϫ100 to Ϫ30 surrounded by highly methylated regions on either side (34) even though there is no sequence homology between the mouse (30) and human (36) MGMT promoters. 6/6 TGF␤1Ϫ/Ϫ cell lines and none of the TGF␤1ϩ/Ϫ cell lines lacked MGMT expression and had a hypermethylated MGMT promoter. We found no increase in sensitivity to MNNG in a p53Ϫ/Ϫ keratinocyte cell line derived in a similar manner as the TGF␤1Ϫ/Ϫ cell lines, pointing to a specific defect of MGMT that is related to TGF␤1 genotype and not to a nonspecific effect of knockout production or inherent instability related to loss of a tumor suppressor gene. However, primary TGF␤1Ϫ/Ϫ null keratinocytes were MGMT-proficient and did not have apparent methylation as judged by the MSP assay, but rather this analysis showed that methylation increased specifically during passaging of the TGF␤1Ϫ/Ϫ cells.   (5), and passage 50 of KO3 cell line (6). B, variable methylation in clones of passage 2 TGF␤1Ϫ/Ϫ keratinocytes. Colonies of keratinocytes were ring-cloned and expanded, and the isolated DNA was subjected to bisulfite modification and MSP. C, variable methylation in clones of passage 8 KO3 cell line. Colonies of KO3 were ring-cloned and expanded, and the isolated DNA was subjected to bisulfite modification and MSP. U, unmethylated-specific primer; M, methylated-specific primer.
Whether this represents a specific growth advantage of rare cells with MGMT methylation and TGF␤-signaling defects or ongoing methylation of the promoter in the absence of TGF␤ signaling will require bisulfite sequencing of cells at different passage number for clarification. Although methylation was undetectable in passage 2 and slightly detectable in passage 8 TGF␤1Ϫ/Ϫ cells, subclones of these passages exhibited variable methylation ranging from none to complete. Since the MSP assay is reported to detect methylated alleles at a frequency of 0.1% (31), it seems unlikely that the appearance of methylated alleles with such high frequency in the subclones represents enrichment of rare cells with methylation. Previous studies with fibroblasts show that genes unmethylated in adult tissues or primary cell cultures become methylated in immortal cell lines that grow out after crisis (37). A plausible hypothesis is that TGF␤1Ϫ/Ϫ cells are more susceptible to hypermethylation, and this occurs randomly at the MGMT locus with every cell division. Such a model would allow for recovery of clones containing MGMT alleles that were unmethylated, completely methylated, or both.
An important question left unresolved is how loss of autocrine TGF␤1 could be linked to hypermethylation. TGF␤1 could directly regulate expression of one of the DNA methyltransferases (DNMT1-3), a demethylase (38), or proteins that bind to methylated DNA (39,40). Since DNMT1 is both regulated by Rb (41) and forms a complex with Rb, E2F1, and HDAC1 (42), it is possible that altered TGF␤1 signaling by modulating the phosphorylation state of Rb could indirectly affect DNMT1 levels or activity. However, the inability of short or long term TGF␤1 treatment to induce MGMT or reduce MGMT methylation in the TGF␤1Ϫ/Ϫ lines suggests that MGMT regulation by TGF␤1 is indirect and that hypermethylation, once achieved, is stable. Both the human and mouse MGMT promoters contain regions of high GC content and multiple SP1 sites (30,36). A recent model for the evolution of hypermethylation in promoters suggests that SP1 sites protect GC-rich regions from spreading of methylation (43). It is intriguing that Smad3 and Smad4, intracellular mediators of TGF␤1 signaling, can activate transcription through interaction with SP1 proteins and SP1 DNA binding sites (44,45). It is tempting to speculate that alterations in Smad levels or activity due to inhibition of TGF␤1 signaling could influence occupancy of SP1 sites and effect increased accessibility to de novo methylation.
In conclusion we have demonstrated distinct patterns of CpG methylation at the MGMT promoter in immortal keratinocyte lines, which differ by ability to produce autocrine TGF␤1. Hypermethylation of the MGMT promoter specifically in the TGF␤1Ϫ/Ϫ keratinocytes occurs during immortalization and subsequent passaging, suggesting that these cells, because of the absence of TGF␤1 signaling, are more susceptible to hypermethylation. Analysis of CpG islands of a number of genes will be required to determine if this is specific for the MGMT locus or represents a general methylator phenotype. However, our results also show that the MGMT promoter also becomes methylated during tumor development in the mouse multistage skin carcinogenesis model, 2 suggesting that this in vitro observation is relevant to neoplastic transformation of mouse keratinocytes in vivo. It will be important to determine if aberrant MGMT methylation contributes to either spontaneous or alkylationinduced mutation and transformation in the TGF␤1Ϫ/Ϫ cells. Given that loss of expression of TGF␤1 is associated with malignant progression in the mouse skin tumor model (17) and that defects in TGF␤ signaling occurs frequently in both human tumors and tumor cell lines (10,11,32), it is tempting to speculate that inactivation of the TGF␤ pathway may play a causal role in the widespread hypermethylation of genes that occurs during cancer development and progression.