It is becoming increasingly common to include data on molecular alterations from patient tumor samples into routine clinical practice as a means of improving prognosis and evaluating the predictive power of alterations of interest. As technology improves and population-based studies and clinical trials are conducted, medicine is being ushered into a new era of molecular characterization of disease. Tumor-node-metastasis (TNM) stage is the current prognostic indicator for breast cancer, though several clinical trials are currently under way to investigate the utility of molecular markers 
, and as more patients elect neoadjuvant therapy (specifically pre-operative chemotherapy), improved clinical staging and additional staging tools are poised to have great impact. Most current studies and one commercially available tool (Oncotype DX) are focused on gene expression markers, though the inherent instability of mRNA may make implementation of these strategies challenging outside of major surgical centers or centralized commercial laboratories. In contrast, DNA methylation is a stable mechanism of control of transcription, and the stability of DNA makes it an attractive target for accurate and reproducible assessment. Here we reported that tumor size, a cornerstone of breast cancer prognosis, is associated with tumor DNA methylation profile. In addition, we found that alcohol and folate intake, exposures related to disease risk, are independently associated with tumor DNA methylation profiles. This work sheds light on the relationship between important etiologic exposures and molecular subclasses of disease, extends the evidence for the utility of molecular characterization in tumor staging, and can be accomplished with minimal tissue in a pre-operative context.
The recently updated American Joint Committee on Cancer (AJCC) staging manual for breast cancer does not include additional molecular markers, though the committee acknowledged their consideration of markers such as hormone receptor status and stated that TNM staging “may play increasingly less important roles than understanding the biology of the cancer” 
. Examining TNM variables we found that overall DNA methylation profile and methylation alterations in dozens of individual CpG loci were significantly associated with tumor size (all increased methylation). In contrast, methylation alterations of only five CpG loci (two in COL1A2
, and one each in FAS
, and P2RX7
) were significantly associated with disease-positive lymph nodes. However, methylation of four of five lymph-node-positive associated CpGs (excepting FAS
) were also significantly associated with tumor size, suggesting that these phenotypes are mechanistically related, and at least in part manifest via epigenetic alterations. As FAS
encodes a TNF-receptor involved in regulating apoptosis it is not surprising that methylation-induced silencing of this receptor is associated with disease-positive lymph node status. In addition, hypermethylation of COL1A2
(collagen type I, alpha 2) has been associated with both proliferation and migration activity in bladder cancer 
is involved in the control of normal collagen deposition 
, and P2RX7
loss has been linked to morphologic changes in stroma related to altered collagen fibril alignment 
. Collectively these data suggest that perturbations in collagen and collagen-related genes promote tumor growth and invasion, perhaps by altering the architecture of connective tissues in the tumor microenvironment. In support of this hypothesis, recent work in a mouse model has shown that altered mammary stromal tissue collagen expression significantly increases tumor formation and invasiveness potential 
. Additionally, Chernov et al. showned that epigenetic alterations in collagen and collagen-related genes allows the deposition of an invasion-promoting collagen matrix in both breast and brain tumor cell lines 
The primary objective of TNM staging is to provide a standard prognosis nomenclature for patient care 
, and our results suggest that methylation markers may be a robust proxy for tumor size. Importantly, broader application of neoadjuvant therapy complicates breast cancer staging since chemotherapy can considerably decrease tumor size prior to surgical treatment, and it is still unclear whether clinical or pathologic stage best informs prognosis and treatment decisions 
. The AJCC has added methodology (yc or ypTNM) for differentiating clinical and pathologic staging; in part, this is from recognition of the increasing use of neoadjuvant therapy for patients with operable, early stage disease 
. Our data illustrate the promise of tumor DNA methylation for augmenting tumor staging. However, additional study of the relationship between tumor methylation and size in both pretreatment and postoperative samples is necessary. Specifically, the value of methylation to act as an additional marker of size in the neoadjuvant setting should be evaluated in future studies that compare both imaging and pathologically based size determination.
In order to evaluate the predictive power of DNA methylation profiles and individual loci for disease prognosis and recurrence, these patients continue to be followed for these events. Associations between DNA methylation and patient survival have been reported for individual genes such as GSTP1
and PITX 
, though overall DNA methylation profiles, or patterns of methylation at selected CpG loci or genes, may improve predictive power. Well recognized molecular subtypes of breast cancer such as hormone receptor negative and ERBB2
over-expressing tumors are known to be associated with reduced survival 
, and it will be necessary to extensively examine methylation markers stratified by commonly used molecular tumor markers. However, we did not find significant associations between ERBB2
status and CpG methylation in our analysis. Nonetheless, other well recognized molecular subtype markers; estrogen receptor, progesterone receptor, and triple negative status were among the covariates with the highest number of significant CpGs from array-wide locus-by-locus analysis. However, hormone receptor status and triple negativity were not associated with methylation profile when modeling all cases. Premenopausal patients' tumors in our surgical cohort had a higher prevalence of hormone receptor positivity compared to the overall population of premenopausal patients diagnosed with breast cancer. In order to address the potential bias this introduced we modeled the methylation profiles of postmenopausal patients' tumors separately and found a significant association between estrogen receptor status and methylation class. Additional study will be needed to better understand the role of hormone receptor and growth factor receptor expression in these tumors as they relate to methylation profile in the context of a patient's menopausal status.
We found significant, independent associations between both alcohol and folate intake and overall tumor DNA methylation profiles when controlling for potential confounders. Folate is a B vitamin that donates its methyl group for homocysteine remethylation to methionine as part of one-carbon metabolism. In turn, methionine is the methyl donor for DNA methylation via S-adenosyl methionine. However, alcohol is known to interfere with folate absorption in the intestine and hepatic release of folate, and hence, supply to tissues 
. In fact, strong evidence of an etiologic role for alcohol in breast cancer has been reported in multiple meta-analyses of prospective and case-control studies with an excess risk for each alcoholic drink per day of about 10% 
. In contrast, meta-analysis of prospective studies has not provided clear support for an overall protective association between folate intake and breast cancer risk 
. Yet, meta-analysis of case control studies of dietary folate, including results from the Shanghai Breast Cancer Study (whose participants are not regular alcohol drinkers) generally support a protective role for folate 
While there have been numerous studies of alcohol and folate in relation to risk of breast cancer, investigations of the relationship between these exposures and epigenetic alterations in tumors themselves are scarce. Tao et al.
reported that the prevalence of breast tumor methylation at CDKN2A
, and RARB
did not differ by folate intake or lifetime alcohol consumption in genotype strata of one-carbon metabolism enzymes methylenetetrahydrofolate reductase (MTHFR
) and methionine synthase (MTR
. Consistent with these findings (and perhaps the lack of similar null results in the literature), we too did not find associations between alcohol or folate and methylation of CpG loci in CDKN2A
, and RARB
. Further, after correcting for multiple comparisons, no CpG loci had significant alcohol-related methylation, and only one CpG locus (in the IL17RB
promoter) was associated with folate intake. Alone, these results suggested that folate and alcohol intake do not influence tumor DNA methylation. However, plots of regression coefficients indicated strong independent trends for increased folate and reduced alcohol intake associations with increased CpG methylation. Since global, low-level effects of alcohol and folate intake on CpG methylation may not be detectable at individual CpGs in a genome-wide context, we examined the global relationships between alcohol or folate intake and DNA methylation using RPMM methylation classes. Modeling both exposures together revealed highly significant, independent associations between alcohol and folate and DNA methylation profile. Another human cancer for which alcohol is an important etiologic factor is head and neck squamous cell carcinoma, and previous work from our group demonstrated a similar relationship between DNA methylation profiles of these tumors and alcohol consumption 
. Taken together with the weak mutagenic potential of alcohol 
, these results suggest that a major carcinogenic mechanism of action of alcohol is interference with epigenetic regulation through disruption of one-carbon metabolism.
In summary, we found tumor DNA methylation associated with tumor characteristics predictive of prognosis, and DNA methylation and patient exposures known to be related to disease risk. Additional study is needed to determine the prognostic value of DNA methylation markers. However, the potential clinical utility of tumor-size-related DNA methylation is apparent.