Background
Locally advanced breast cancer is a heterogeneous disease with varying outcomes and different responses to neoadjuvant chemotherapy (NACT), depending on breast cancer subtype. Neoadjuvant treatment has become a standard of care for locally advanced breast cancer, offering the benefit of downstaging the disease prior to surgery and the elimination of disseminated cancer cells to improve survival [
1,
2]. To improve personalized treatment and avoid late effects from unnecessary treatment, it is important to develop novel predictive and prognostic biomarkers for patient stratification based on response to NACT and patient survival.
DNA methylation is an epigenetic mechanism that regulates gene expression and chromatin structure. It influences gene expression in a complex manner; for example, promoter CpG island hypermethylation can repress gene transcription, while gene body hypomethylation can increase transcription [
3,
4]; however, promoter methylation could also be associated with increased transcription [
5,
6]. Aberrant methylation patterns can be detected early in cancer development [
7] and have been shown to be important for development and progression in breast cancer and other malignancies [
8‐
11].
Today, the molecular subgrouping of breast cancer [
12] as well as axillary lymph node status, tumor size, HER2 overexpression, histopathological grade and hormone receptor status are used to assess patients' risk and to decide treatment options according to national clinical guidelines [
13]. However, there is still substantial heterogeneity within the subgroups in respect to response to NACT and survival [
8,
9,
14‐
16]. The added informational value of DNA methylation patterns may contribute to the identification of patients who will respond to treatment and those who will have a more aggressive course of disease.
Several studies have explored how changes in gene expression and metabolite profiles during NACT correlate with treatment response and survival in breast cancer patients [
17,
18], while less is known about how DNA methylation changes in response to NACT. A recent study reported differential DNA methylation in whole blood following chemotherapy in breast cancer patients [
19], but the changes were not investigated in relation to treatment response or survival. Differential DNA methylation in tumor biopsies has been shown to predict treatment response in breast cancer biopsies [
20‐
22]. For instance, Klajic et al. [
20] showed that DNA methylation of cell cycle-related genes changed differently in responders and non-responders during NACT. A study by Luo et al. [
23] demonstrated changes in DNA methylation heterogeneity in response to NACT. DNA methylation patterns of pre-treatment biopsies have also been used to predict survival in a doxorubicin-treated cohort by Dejeux et al. [
24]. Changes in DNA methylation can influence both treatment response and survival and are therefore important for developing new therapeutic targets and prognostic markers.
Therapy resistance is a major challenge in cancer treatment, and increasing evidence suggests that exposure to chemotherapy may drive drug resistance through silencing and activation of genes caused by methylation [
25‐
28]. In this study, we assess the methylation patterns before and after NACT to determine how NACT affect tumor DNA methylation and investigate the predictive and prognostic potential of treatment-induced changes in DNA methylation patterns.
Discussion
In this study, we detected significant changes in DNA methylation in 5-year breast cancer survivors when comparing biopsies taken before and after NACT. We further developed a risk score based on methylation sites with significant change before versus after NACT associated with 5-year survival in our patient cohort that was validated in an independent cohort. While a few previous studies have described DNA methylation changes from before to after treatment in patients with breast cancer [
20,
23], this is the first study to correlate methylation changes to survival. In our study, the methylation changes were unique to survivors, as there were no significant changes in CpG sites in non-survivors from before to after treatment. Additionally, the top-ranked CpG sites by p value and effect size in survivors were low ranked in non-survivors, indicating that the difference in results between survivors and non-survivors is not due to differences in sample size.
The prognostic value of the risk score developed in this study was validated in a separate cohort of locally advanced breast cancer patients. This supports the idea that the changes in methylation during neoadjuvant treatment are important for breast cancer survival. The validation cohort had a similar treatment regimen as our discovery cohort, since epirubicin and doxorubicin are similar drugs. However, since a proportion of patients in both cohorts received other treatment regimens (paclitaxel or 5-FU and mitomycin) and the separation of the risk groups receiving different treatments in both cohorts were significant, this suggest that the risk score is not specific for a given NACT and may be valid independently of treatment regimen. Our study shows that changes in tumor DNA methylation are associated with survival in two separate cohorts. This highlights the importance of studying the molecular response of breast cancer tumors during NACT to be able to assist the prognosis of breast cancer patients. Although the risk score could not be used to predict survival before treatment, assessment of prognosis post-treatment would be useful to pinpoint the patients in need of closer follow-up and possibly extended treatment.
Given the complex association between DNA methylation and gene expression, it can be difficult to assess the exact function of single methylation sites. By investigation of the close by genes or the genes previously associated with the CpGs, we tried to interpret the biological relevance of the CpG sites in the methylation risk score. The first differentially methylated site (cg10298059) is annotated to the gene body of ZFHX3, which is a transcription factor and tumor suppressor gene, its expression associated with the prognosis of breast cancer patients [
42‐
44]. The second CpG site (cg27034150) is in the promoter region of SULT1A1, which is a sulfotransferase involved in the metabolism of drugs. Its involvement in tamoxifen metabolism has been reported previously [
45‐
47]. The third CpG site (cg01503450) is situated in the promoter region of LARP4B, a gene involved in RNA translation, and could be a tumor suppressor gene [
48‐
50]. The fourth CpG site (cg07959469) is situated downstream of the gene NR2F2, which is a transcription factor involved in ER-mediated transcriptional regulation and also involved in invasion and migration [
51‐
53]. In summary, the four identified CpGs are located around genes important for cancer cells and may explain why alterations in DNA methylation at these CpGs are associated with prognosis.
CpG islands in normal cells are in general unmethylated, but during aging and cancer development, there is an overall gain of methylation in CpG islands, while the rest of the genome loses methylation [
10,
54]. Gain of methylation in CpG islands can repress tumor suppressor, apoptosis, cell adhesion and DNA reparation genes, while loss of methylation outside CpG islands is associated with activation of oncogenes, reactivation of fetal genes and loss of repression of transposable elements, leading to chromosomal instability [
55,
56]. In our study, there was a predominant loss of methylation in CpG islands in 5-year survivors after treatment and an overall gain in methylation outside of CpG islands. In addition, 90% of the differentially methylated sites had a change in methylation toward normal breast tissue. The combined methylation changes we observed for the survivors in our study thus suggest a reverse cancer progression or fewer aggressive cells in the tumor after treatment.
When exploring the biological implications of the differentially methylated sites in functional regions of the genome, we found reduced methylation of CpG islands in genes involved in transcription factor activity and cell adhesion. Differential methylation associated with these biological processes has also been found to discriminate breast cancer from normal tissue [
57] and to be associated with response to NACT in triple-negative breast cancer [
58]. DNA methylation is known to regulate transcription factors in human cancers, which in turn regulates oncogenes and signaling pathways important for prognosis [
9,
16,
59]. Cell adhesion genes are less methylated in noninvasive breast cancer cell lines compared to invasive breast cancer cell lines [
60]; thus, a loss of methylation as observed in this study could be a sign of less invasive cancer cells in the post-treatment samples. Since the relation between DNA methylation and gene expression is complex, it is important to validate these findings by gene expression and protein analysis before developing therapeutics targeting these processes.
Methylation of immune response genes before treatment in breast cancer has previously been associated with prognosis [
61] as well as the infiltration of lymphocytes in the tumor [
62]. Since we found immune system genes overrepresented in the pathway analysis, we explored this further by immune cell deconvolution by MethylResolver [
38] and found a reduced estimated T cell fraction in the samples after treatment.
The function of regulatory T cells is to regulate and moderate immune reactions. In the tumor microenvironment, they are pro-tumorigenic and protect the tumor from immune destruction [
63]. Thus, fewer regulatory T cells make the tumor more exposed to the anti-tumor immune response. Low regulatory T cell infiltration identified by immunohistochemistry has been connected to a better prognosis (complete response) in locally advanced and local breast cancer [
64‐
67]. In this study, 5-year survivors and non-survivors had no significant differences in immune cell composition before treatment. However, the survivors’ immune cell composition changed toward a more anti-tumorigenic response, highlighting that the dynamics of the immune infiltration could be important for patient survival, especially as a response to chemotherapy. Although the effect of treatment-induced immune response observed in our study is interesting, the immune response is a complex system and additional studies and methods are needed to investigate this in more details.
In previous studies, both survival and DNA methylation patterns have been related to the molecular subgrouping and receptor status of breast cancer [
14], and breast cancer may even be further sub-grouped using DNA methylation and clustering methods [
12,
68,
69]. It is therefore important to establish whether prognostic biomarkers are valid in one, several or across all subgroups. As there were no significant changes in methylation in hormone receptor positive, HER2 positive or within specific intrinsic subgroups from before to after NACT, we conclude that the change in methylation is either consistent across subtypes or that there are too few patients in each subgroup to detect subtype-specific differential methylation.
Studies using DNA methylation signatures to predict NACT response have previously been conducted in pre-treatment tumor biopsies and blood samples of breast cancer patients [
21,
22,
58]. However, we did not find DNA methylation changes that were related to treatment response. A plausible reason could be because our study cohort does not include patients with complete response where no residual tumor was left after treatment, nor patients with progressive disease. In addition, the evaluation of treatment response was performed based on caliper measurement of the tumor before and after treatment, which could heighten the risk of inaccurate measurements, as compared to radiological evaluations. Also, the treatment response criteria used in our study were according to the UICC recommendation applicable at the time of patient recruitment, and some of the patients with stable disease would have been classified as partial responders according to the RECIST criteria used today. In the current study, 5-year non-survivors had a lower percentage of partial response compared to 5-year survivors, 45.8% versus 69.6%, respectively. In our previous study of the same patient cohort, we detected changes in tumor metabolism after treatment. The changes were related to survival but not to treatment response [
18], similar to what we observed for methylation patterns in the current study. Many patients in the non-survival group had a relatively good response to treatment, but still experienced a rapid progression. This implies the importance of studying the molecular tumor response to treatment and its effect on survival in addition to shrinkage of tumor size.
The patients included in this study were treated with monotherapy regimens consisting of either epirubicin or paclitaxel. In case of non-satisfactory response, they were assigned to the other chemotherapy. Although standard clinical guidelines today advocate a combination of different chemotherapies, our study investigates the effect of these two drugs both separately and in sequence. Our results show no significant differences in DNA methylation when comparing the different treatment regimens, which suggest that differential DNA methylation observed in survivors is not dictated by either of these two chemotherapies initiated as the first regimen or when given in sequence. Interestingly, the risk score could separate patients having received monotherapy by either epirubicin or paclitaxel into low- and high-risk groups when the treatment groups were examined separately.
This study has some limitations. The cohort contains a mix of breast cancer subtypes, which introduced challenges due to heterogeneity. The patients were treated with monotherapy, which is an older treatment regimen compared to nowadays. The study is constrained by the lack of available tumor tissues for further analyses, especially in case of validation of estimated immune cell fractions, for example, by immune histochemistry or quantitative polymerase chain reaction (qPCR). However, the signature developed has been validated in an independent cohort, which demonstrates its clinical potential with regard to survival. Further studies are, however, needed to fully understand the biological implication of these methylation sites and how they are associated with breast cancer prognosis.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.