Background
DNA methylation (DNAm), a form of dynamic and reversible epigenetic regulation, is highly sensitive to environmental pressures on the body [
1,
2]. Hence, the cytotoxic nature of chemotherapy is expected to have a profound impact on a patient’s DNAm landscape [
3,
4].
In recent years, there is an increasing number of studies looking into the mechanistic, biomarker, and therapeutic roles of epigenetic pathways in response to chemotherapy [
3,
5]. Yao et al. characterized epigenetic changes in paired pre- and post-chemotherapy blood specimens from 93 breast cancer cases and found a marked impact on the leukocyte DNA methylome—4.2% of the CpG probes tested were significantly altered [
3]; significant changes in the abundance of CD4 + T cells, B cells, and monocytes were observed. The authors highlighted specific CpG sites in four genes,
VMP1/MIR21,
CORO1B,
SDK1, and
SUMF2. Notably, CpG cg16936953 in
VMP1/MIR21 was also the most significant locus in an independent study by Smith et al. comparing patients treated with chemotherapy and those untreated (
n = 61 breast cancer patients) [
4]. In a study by Sigin et al. using 62 breast biopsy samples, differential methylation in 10 genes (
SLC9A3,
C1QL2,
DPYS,
IRF4,
ADCY8,
KCNQ2,
TERT,
SYNDIG1,
SKOR2, and
GRIK1) was reported to help discriminate response to neoadjuvant chemotherapy [
6].
Using a larger cohort, this study aims to validate and extend the results obtained from previous studies that have investigated chemotherapy-induced epigenetic changes. We examined chemotherapy-induced changes in DNAm in 125 Asian breast cancer patients with paired blood specimens before and after treatment and validated the findings in a further 2145 patients treated (n = 1273) and not treated with chemotherapy (n = 872).
Discussion
Two datasets comprising a total of 2270 Asian breast cancer patients were used to study the effect of chemotherapy on the landscape of DNAm changes: (1) Patients with paired specimens collected pre- and post-chemotherapy, (2) single time point blood or saliva specimens from patients treated and not treated with chemotherapy. In addition, we found 141 differentially methylated CpGs and 11 promoters to be significantly associated with chemotherapy treatment in both the paired sample and single time point datasets after Bonferroni correction. Gene set enrichment analysis of promoters suggests an epigenetic basis by which chemotherapy treatment may affect the perception of smell. The effect size estimates obtained from the comparisons of treated vs non-treated samples generally decreased as the specimens were collected a longer time after the start of chemotherapy treatment, suggesting that chemotherapy-induced DNAm changes recover over time.
Results from our paired sample analysis validated Yao et al.’s work (93 paired pre- and post-treatment samples) which showed that the proportions of monocytes and CD4+ T cells estimated from DNAm data were significantly altered after treatment (3). However, no significant difference was found in the proportions of B cells in our study. In addition, we confirmed significant chemotherapy-associated DNAm changes for cg16936953 in TMEM49/VMP1/MIR21 (p = 6.08e−10), cg01252023 in CORO1B (p = 1.02e−16), and cg19956914 in SUMF2 (p = 3.30e−11). cg11859398 in SDK1 did not survive Bonferroni correction in our study.
Apart from CpG probes, we examined promoter regions as these may have distinct functional and regulatory roles. Eight of the eleven differentially methylated promoters that were significant across both datasets mapped to RNA genes. Although such genes have been suggested to play a role in the regulation of the olfactory system, their functions are unclear [
21].
KCTD11 plays a role as a marker and a regulator of neuronal differentiation.
AKNA acts as a transcription factor that specifically activates the expression of the
CD40 receptor and its ligand
CD40L/CD154, two cell surface molecules on lymphocytes that are critical for antigen-dependent B-cell development [
22].
Our promoter-level pathway analysis revealed that biological processes related to sensory perception and the olfactory transduction pathway are significantly altered by chemotherapy. The single time point dataset additionally revealed the KEGG pathway “Taste transduction” (hsa04742) to be suppressed in chemotherapy-treated patients. Although the taste-related pathway did not survive stringent Bonferroni correction, the p-values in the various models tested across blood and saliva specimens ranged from 0.0005 to 0.006, and the corresponding less-stringent false-discovery rate Q values from 0.002 to 0.133.
Common side effects of chemotherapy include unwanted changes in taste and smell, which have repercussions on nutritional status, dietary intake, appetite, body mass index, and quality of life [
23,
24]. It has been reported that chemosensory alterations occur in as many as 86% of chemotherapy-treated patients [
25,
26]. The biological mechanisms driving such taste and smell disturbances are unclear [
27].
The prevailing explanations include cytotoxic damage to proliferative olfactory and gustatory receptor cells, changes to oral microbiota, mucositis, nutritional imbalance, and alterations of salivary quantity and composition [
27,
28]. However, previous studies have shown that sensory variability in taste and smell may be driven by epigenetic markers in non-cancer cohorts [
29,
30]. As sensory changes are typically transient and mostly recover to baseline levels after cessation of treatment, a modifiable epigenetic mechanism such as DNAm is highly possible [
31,
32]. In agreement, our results support the hypothesis that DNAm influences sensory factors to cause dysfunctions in olfactory measures, in chemotherapy-treated breast cancer patients. The attenuation of effect sizes of differentially methylated probes in our results supports clinical observations of recovery of the phenotype over time.
One notable caveat of the striking pathway analysis results is that the olfactory gene family is large, and genes related to olfactory receptors tend to be clustered in the genome, which can lead to spurious enrichment in gene ontology tests [
33,
34]. However, Lerm et al. revealed that long-lasting alterations to DNAm patterns induced by SARS-CoV-2 infection are associated with negative impacts on odor perception, corroborating our finding that smell may be epigenetically rewired [
35].
This is the first epigenome-wide study describing associations between chemotherapy treatment and genomic DNAm in a sizable cohort of Asian breast cancer patients. The convergence of findings using two separate study designs (paired sample pre- and post-chemotherapy and data obtained from a large cohort of chemotherapy-treated and non-treated patients) increases the rigor and validity of the study. The genome-wide differential methylation screening approach allows an unbiased selection of DNAm markers, which constitutes the main strength of our study.
It is important to acknowledge that the process of prioritizing samples for inclusion in DNAm experiments based on the completeness of clinical and follow-up data could potentially introduce a selection bias. In particular, included patients were more recently diagnosis, of younger age at diagnosis, and had a shorter time between sample collection and the start of chemotherapy (Additional file
17). Efforts should be made to minimize missing data and ensure comprehensive data collection to reduce the potential for selection bias in future studies. In addition, surgery is a major stress event for patients and has been shown to cause changes in DNAm [
36]. However, the number of neoadjuvant breast cancer patients who received chemotherapy before surgery comprises only a quarter of our patient population. The low sample size limits our ability to examine surgery as a potential confounder. Further studies with a larger sample size will be required to study the effects by specific chemotherapy regimens. Due to the retrospective nature of the study, DNA samples were stored for different periods before processing for the DNAm experiments. Variation in DNAm stability over time in different genomic loci is likely [
37]. DNAm is also tissue- and cell-specific, limiting the generalizability of the results to the sample types studied [
38,
39]. The time between blood draw and the start of chemotherapy was thus taken into account in the analyses. The underlying distributions of DNAm data do not always satisfy the assumptions of linear regression models used for the analyses. However, this statistical method is valid for exploratory studies [
40]. As data on body composition and behavioral risk factors such as dietary intake, physical activity, smoking, and drinking after breast cancer diagnosis were not collected, mediation analysis was not conducted.
Acknowledgements
This work was supported by the A*STAR Computational Resource Centre through the use of its high-performance computing facilities. The study team thanks all Singapore Breast Cancer Cohort patients for their participation. We would also like to thank the programme manager Jenny Liu and clinical research coordinators/research assistants Nur Khaliesah Binte Mohamed Ri, Siok Hoon Yeo, Kimberley Chua, Ting-Ting Koh, Amanda Ong, Jin-Yee Lee, Michelle Mok, Jing-Jing Hong, Hui-Min Lau, Siew-Li Tan, Ganga Devi d/o Chandrasegran, and Yen Shing Yeoh for their contributions in recruitment and data collection.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.