DNA methylation associations with cognitive function in early-stage hormone receptor-positive breast cancer patients

ABSTRACT

Background

Approximately one-third of breast cancer (BC) patients show poorer cognitive function (CF). Using DNA methylation (DNAm) data, here we aimed to identify genes and biological pathways associated with CF in postmenopausal women with early-stage hormone receptor-positive (HR+) BC.

Methods

Epigenome-wide association studies (EWAS) and differentially methylated region analyses were performed for each CF phenotype (seven objective domains and one subjective phenotype) using DNAm data from whole blood samples (n = 109) taken at the time of enrollment.

Results

When adjusting for age, verbal IQ scores, and global DNAm signature, cg10331779 near CTNND2 (p-value = $9.65\times {10}^{-9}$) and cg25906741 in MLIP (p-value = $2.01\times {10}^{-8}$) were associated with processing speed and subjective CF, respectively, while regions in/near SLC6A11, PRKG1/CSTF2T, and FAM3B for processing speed, and regions in/near PI4KB and SGCE/PEG10 for mental flexibility were differentially methylated. In addition, beta-estradiol was identified as a common upstream regulator for all the CF phenotypes, suggesting an essential role of estrogen in explaining variation in CF of HR+ BC patients.

Conclusions

In our EWAS of 8 CF phenotypes, we found two epigenome-wide significant signals, one for processing speed and the other for subjective CF. We also found three differentially methylated regions associated with processing speed and two associated with mental flexibility.

Clinical trial registration

www.clinicaltrials.gov identifier is NCT02793921.

Plain Language Summary

Many breast cancer patients experience problems remembering things, thinking clearly, or paying attention. Our study looked at how a chemical change called ‘DNA methylation’ – which controls how genes work without actually changing the genetic code itself – is related to these thought problems in 109 women with early-stage breast cancer. We found that methylation levels near the CTNND2 gene were related to how quickly the brain processes information. Methylation levels near the MLIP gene were related to patients’ reports of their thought problems. We also discovered that beta-estradiol, a type of estrogen hormone, seemed to be a key player in all the thought problems we looked at. Our results give us new clues about why some breast cancer patients have thinking problems and point to other opportunities for future research.

1. Introduction

Breast cancer (BC) is the most commonly diagnosed cancer and the second most common cause of death due to cancer for women in the United States [1,2]. One in eight women will be diagnosed with BC in their lifetime with the median age at diagnosis of 62 when most are postmenopausal [1]. With advancements in cancer treatment, the survival rate of BC has increased in recent years. However, some survivors experience long-term negative effects of cancer or treatments, among them is cancer-related cognitive decline (CRCD) [3–5]. CRCD particularly influences memory and decision-making, which leads to poorer quality of life [5]. Approximately 25% to 75% of BC patients develop symptoms of CRCD and 30%–35% of them show poorer cognitive function (CF) before the beginning of adjuvant therapy compared with age-matched healthy controls [4,6,7].

Objective and subjective CF in BC patients are complex phenotypes that can be affected by many components besides cancer-related factors, such as genetic predisposition, aging, physical inactivity, and social-demographic factors. Although several potential mechanisms contributing to cognitive impairment in BC patients have been proposed, including oxidative stress, chronic inflammation, and white matter damage, to date there has been no consensus. DNA methylation (DNAm), as a dynamic regulator of gene activity that is sensitive to internal and external environmental and behavioral factors, may provide new insights into the biological mechanisms of CF in BC patients.

In this study, we focus on a cohort of postmenopausal women with early-stage hormone receptor-positive (HR+) BC, a common subtype of BC comprising nearly 80% of the total invasive cases [1,8]. In this subtype, most patients receive adjuvant endocrine therapy (ET). Here, we aim to use epigenome-wide DNAm data to evaluate how the CF and the DNAm signature are related in this cohort and find possible genes and biological pathways associated with CF, which may provide insights into the poor CF observed in BC patients.

2. Materials and methods

2.1. Participants and Samples

Participants in this study were a subsample of participants in the randomized controlled trial “Exercise Program in Cancer and Cognition” (EPICC) (NCT02793921) (Table S2). The EPICC study protocol and inclusion/exclusion criteria are detailed elsewhere [9]. Postmenopausal women with early-stage HR+ BC were enrolled from clinical sites in Southwestern Pennsylvania [9,10]. Postmenopausal women with stage I-IIIa HR+ BC who were younger than 80 years old, had ≥8 years of education, and had not initiated adjuvant therapy at the time of recruitment were eligible. The criteria were relaxed during the pandemic to include patients within 2 years of completing primary BC treatment (surgery ± chemotherapy). Some participants had started ET before their pre-randomization blood sample was collected. All participants provided written informed consent.

Blood samples and CF data were collected at pre-randomization and follow-up timepoints to study how ET and exercise intervention impact CF and the role of DNAm (Figure 1, Figure S1). Here, we study only the pre-randomization data because our aim was to determine genetic and biological pathways of variation in CRCD and not to determine the impact of the exercise intervention on CRCD (see Bender et al., 2024 [10]).

Figure 1.

Study workflow.

2.2. Phenotypes

2.2.1. Objective cognitive function

As previously described, a standardized neuropsychological test battery was administered to assess objective CF, yielding multiple measures [11,12]. These scores were z-transformed with respect to age- and education-matched controls without breast cancer and adjusted so that higher values represent better CF than the controls [10,11]. Seven CF domains were derived using exploratory factor analysis: executive function, attention, mental flexibility, learning and memory, verbal memory, working memory, and processing speed.

2.2.2. Subjective cognitive function

Subjective CF was self-assessed using the Patient’s Assessment of Own Functioning Inventory (PAOFI) [13]. The total score based on the sum of the values from 32 items was used in the following analyses, with higher scores indicating worse subjective CF (range 0–160).

2.3. DNA methylation sample collection

Whole blood samples were collected at pre-randomization and then DNA was extracted using QIAGEN DNA extraction kit. DNA samples were stored in 1X TE buffer at 4°C for future DNAm data collection. Genome-wide DNAm data were generated by the Infinium MethylationEPIC v1.0 BeadChip (Illumina, San Diego, CA, USA) at the Center for Inherited Disease Research, Johns Hopkins University.

2.4. Quality control

Quality control of DNAm data was performed by minfi, lumi, Enmix, and ewastools R packages [14–19]. After basic DNAm data processing, DNAm β values and M values were derived as follows:

$\mathit{\beta }=\frac{\mathrm{Meth}}{\mathrm{Meth}+\mathrm{Unmeth}+\mathrm{offset}}$ and $\mathrm{M}=\mathrm{lo}{\mathrm{g}}_2\frac{\mathrm{Meth}+\mathrm{offset}}{\mathrm{Meth}+\mathrm{Unmeth}+\mathrm{offset}}$ where Meth and Unmeth represent methylated and unmethylated signals, respectively, and the offset term was set as 100. For quality control details, see the Supplemental Material. Functional normalization with background and dye bias correction was applied on β values using funNorm [20]. Samples with a detection p-value >0.01 and/or marked as SNP outliers were removed. Technical replicates that had lower detection p-values were retained. After removal of probes with large detection p-values, probes with known SNPs nearby, cross-reactive probes, and Y chromosome probes, 700,779 probes were available (Table S1).

2.5. Statistical analysis

2.5.1. Epigenome-wide association studies (EWAS) and differentially methylated region (DMR) analysis

EWAS were conducted for each phenotype separately using limma R package [21]. As recommended, both EWAS unadjusted for cell-type heterogeneity (CTH) (hereinafter referred to as CTH-unadjusted EWAS) and EWAS adjusted for CTH (CTH-adjusted EWAS) were performed [22]. To account for technical artifacts in the CTH-unadjusted analyses, batch, chip row position, median methylated intensities, and median unmethylated intensities were included in the model, in which DNAm M values were regressed on a CF phenotype, as well as age and verbal IQ scores. In CTH-adjusted EWAS, in addition to age and verbal IQ scores, surrogate variables generated by SmartSVA R package were included to control any unwanted variation such as CTH and technical artifacts [22–24]. The EWAS sample sizes varied since only samples with complete observations on CF phenotype and covariates were included (Figure S1). Cytosine – phosphate – guanine (CpG) sites with p-values <$9\times {10}^{-8}$ were regarded as significant differentially methylated CpGs (DMCs) and those with p-values <$1\times {10}^{-5}$ were referred to as suggestive DMCs [25].

DMR analyses were carried out by the dmrff R package using EWAS summary statistics [26]. DMRs that include multiple CpG sites whose DNAm levels were consistently associated with the trait are considered to have better biological meaning than a single CpG site. The dmrff program identified DMRs where the maximum distance between consecutive CpG sites was 500 bp, the EWAS nominal p-value of the CpG sites in the scanned genomic regions was <0.05, and the effect directions of CpG sites in the 500 bp window were consistent. DMRs with a Bonferroni-adjusted p-value <0.05 were considered significant.

2.5.2. Post-EWAS functional analysis

To explore which biological processes were associated with CF, we performed gene set enrichment analysis, Gene Ontology (GO) enrichment analysis, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis, using the missMethyl R package [27,28]. Limited by the small sample size, we used a less stringent threshold when selecting CpGs as input – CpGs whose nominal p-value was <0.01 in CTH-adjusted EWAS, where the confounding effect of CTH was better controlled, were included in the analysis. GO terms and pathways with a false discovery rate (FDR) <0.05 were considered significant. The same list of CpGs was used to conduct a core analysis in QIAGEN IPA (QIAGEN Inc., https://digitalinsights.qiagen.com/IPA, RRID:SCR_008653) software for pathway analysis and upstream regulator analysis [29].

In silico analyses were then performed on the significant DMCs from CTH-adjusted EWAS. First, a methylation quantitative trait loci (mQTL) lookup analysis was conducted using data from 2,358 blood samples of existing cohorts (TwinsUK and National Child Development Study 1946 and 1958 birth cohorts) based on Infinium MethylationEPIC BeadChip to explore whether the DMCs were under genetic control [30]. Associated SNPs within 1 MB of the selected CpGs were regarded as cis mQTLs; others were labeled as trans mQTLs. Next, we checked the correlation between DNAm in blood and brain tissues of the selected DMCs. Although blood is the most accessible tissue for generating DNAm data, the brain may be more ideal for studying CF. We queried two databases about the correlation between different brain regions and DNAm in blood generated by Infinium HumanMethylation450 BeadChips and one for DNAm generated by Infinium MethylationEPIC BeadChips [31–33].

2.5.3. Sensitivity analyses

Originally, the EPICC study aimed at recruiting eligible participants who had not started the adjuvant therapy at enrollment [9]. However, due to the COVID-19 pandemic, eligibility criteria were relaxed and some participants had already started ET when their blood samples were collected. Even though ET can improve overall survival for HR+ BC patients, it might be associated with poor CF [34]. To evaluate the relationship between DNAm and CF alone without ET, we performed sensitivity analyses for all the statistical analyses mentioned above on the subset of participants who began ET after blood sample collection.

3. Results

3.1. Sample characteristics

Table 1 shows the characteristics of participants included in the analyses. Participants were primarily diagnosed with stage 0 or stage I (77%) BC and self-reported as White (91%) with a mean age of 62 years, a mean estimated verbal IQ score of 112, and a mean educational attainment of 16 years. Compared with education- and age-matched controls without breast cancer, study participants showed poorer CF in many domains, including attention, working memory, verbal memory, and learning and memory. Regarding subjective CF, the mean total scores from self-assessed PAOFI were 20, similar to the average scores for participants without BC in an earlier study [35]. The study participants are similar to the larger EPICC study from which they were recruited (Table S2).

Table 1.

Sample characteristics.

A. Continuous variables
	N = 109
Characteristic	Missing	Min	Q1	Median	Mean	Q3	Max	Distribution
Age	0	24.00	58.00	64.00	62.35	67.00	78.00
Education in years	0	12.00	14.00	16.00	16.21	18.00	23.00
Verbal IQ scores	0	84.20	109.12	112.68	112.59	118.02	125.14
PAOFI total scores	6	0.00	7.00	15.00	20.11	27.44	83.00
Cognitive function domains (z scores)
Executive function	3	−1.96	−0.07	0.32	0.23	0.67	1.15
Attention	3	−2.17	−0.60	−0.11	−0.21	0.26	1.01
Mental flexibility	5	−3.03	−0.19	0.32	0.11	0.65	1.06
Learning and memory	3	−2.09	−0.60	−0.12	−0.13	0.44	1.26
Working memory	3	−1.63	−1.03	−0.52	−0.40	−0.03	1.66
Verbal memory	3	−2.04	−1.08	−0.52	−0.38	0.20	2.29
Processing speed	3	−2.29	−0.25	0.21	0.09	0.51	1.34
B. Categorical variables
	N = 109
Characteristic	n (%)
Race
Asian	1 (0.9%)
Black or African American	8 (7.3%)
Other	1 (0.9%)
White	99 (91%)
Ethnicity
Hispanic or Latino	1 (0.9%)
Non-Hispanic or non-Latino	107 (99%)
Missing	1
Tumor stage
DCIS	16 (15%)
Stage I	68 (62%)
Stage IIa	15 (14%)
Stage IIb	5 (4.6%)
Stage IIIa	5 (4.6%)
Had chemotherapy before
Yes	18 (17%)
ET status
Started ET after blood sample collected	65 (60%)
Started ET before blood sample collected	43 (40%)
Missing	1

3.2. EWAS and DMR analyses

In CTH-unadjusted analyses, with the genome-wide significance threshold of $9\times {10}^{-8}$, CpG cg08149859 in Pyruvate Dehydrogenase Phosphatase Catalytic Subunit 2 (PDP2) was associated with mental flexibility (p-value=$4.30\times {10}^{-8}$, coefficient = −0.031), although the association was not significant in the CTH-adjusted EWAS (p-value=$2.27\times {10}^{-6}$, coefficient = −0.024), suggesting that this signal was driven by cell-type composition (Table S5). CpG cg24022240 in Adducin 3 (ADD3) and cg03444675 near ER Membrane Protein Complex Subunit 2 (EMC2) were associated with PAOFI total scores (p-values of $5.23\times {10}^{-8}$ and $8.93\times {10}^{-8}$, respectively). Still, they were not significant after adjusting for CTH (p-values of $8.73\times {10}^{-7}$ and $3.40\times {10}^{-7}$, respectively). In the CTH-adjusted analyses, processing speed was significantly associated with cg10331779 in Catenin Delta 2 (CTNND2) on chromosome 5 with a p-value of $9.65\times {10}^{-9}$ (Table 2, Figure 2). CpG cg25906741 in Muscular LMNA Interacting Protein (MLIP) was negatively associated with PAOFI total scores with a p-value of $2.01\times {10}^{-8}$ (Table 2, Figure 3).

Table 2.

Significant differentially methylated CpG sites of CTH-adjusted EWAS.

Outcome	CpG name	Chromosome	Position*	Nearest gene	UCSC gene feature category	CpG annotation	Coefficient^	P-value	FDR Q value
Processing speed	cg10331779	5	11904811	CTNND2	TSS1500	South shore	−0.0369	9.65E–09	0.0067
PAOFI total scores	cg25906741	6	53976256	MLIP	5’UTR;1stExon;Body	Open sea	−0.0027	2.01E–08	0.0141

The annotation is based on hg19 genome build.

Coefficients in DNAm β-value scale (Δβ) were calculated using the formula here [36].

Figure 2.

CTH-adjusted EWAS results for processing speed.

A. Manhattan plot. The red line represents the epigenome-wide significance threshold with a p-value of $9\times {10}^{-8}$ and the blue line represents the suggestive threshold with a p-value of $1\times {10}^{-5}$. B. Quantile–Quantile (QQ) plot with 95% confidence interval shaded in gray. The red line represents the reference line where the points should fall under the null hypothesis of no association. C. CoMET plot of a 30 kb region on chromosome 5 centered on cg10331779. The upper panel shows the regional plot of the DMC and its adjacent CpGs. CpGs positively associated with processing speed are displayed in red, while CpGs negatively associated with processing speed are displayed in aqua blue. The middle panel shows the gene annotation track (yellow) and CpG island track (green). The lower panel shows the correlation pattern of DNAm beta values for CpGs in this region.

Figure 3.

CTH-adjusted EWAS results for the Patient’s assessment of own functioning inventory (PAOFI) total scores.

A. Manhattan plot. The red line represents the epigenome-wide significance threshold with a p-value of $9\times {10}^{-8}$ and the blue line represents the suggestive threshold with a p-value of $1\times {10}^{-5}$. B. QQ plot with 95% confidence interval shaded in gray. The red line represents the reference line where the points should fall under the null hypothesis of no association. C. CoMET plot of a 60 kb region on chromosome 6 centered on cg25906741. The upper panel shows the regional plot of the DMC and its adjacent CpGs. CpGs positively associated with PAOFI total scores are displayed in red, while CpGs negatively associated with PAOFI total scores are displayed in aqua blue. The middle panel shows the gene annotation track and the lower panel shows the correlation pattern of DNAm beta values for CpGs in this region.

Two sets of DMR analyses were carried out – (1) using the summary statistics from the CTH-unadjusted EWAS and (2) using the summary statistics from the CTH-adjusted EWAS. One significant DMR near RNF26 with a Bonferroni-adjusted p-value less than 0.05 was found for learning and memory without the adjustment of CTH (Table S6). When adjusting for CTH, three significant DMRs were identified for processing speed, respectively located in/near SLC6A11, PRKG1/CSTF2T, and FAM3B (Table 3). And significant DMRs in/near PI4KB and SGCE/PEG10 were detected for mental flexibility. Details of the sensitivity analysis results are described in the Supplemental Material and Supplemental Tables (Table S23 – Table S24).

Table 3.

CTH-adjusted differentially methylated regions with adjusted p-value less than 0.05.

Chromosome	Start position	End position	Number of CpGs	Nearest genes	Adjusted p-value
Processing speed
3	10857258	10857717	8	SLC6A11	3.68E–02
10	53459369	53459604	6	PRKG1; CSTF2T	2.59E–02
21	42688191	42688675	3	FAM3B	3.19E–03
Mental flexibility
1	151300283	151300723	9	PI4KB	5.84E–03
7	94286304	94286834	18	SGCE; PEG10	2.82E–04

3.3. Post-EWAS functional analysis

Although no significant GO terms or KEGG pathways were identified for any CF phenotype, several canonical pathways and upstream regulators were found from the IPA core analysis (Table S7 – Table S22). For example, the axonal guidance signaling pathway, which is involved in receiving and processing cues to guide the axon to its final destination, was identified as the top one for processing speed (p-value = $1.2\times {10}^{-7}$) and mental flexibility (p-value = $8.9\times {10}^{-8}$). For PAOFI total scores, the synaptogenesis signaling pathway, which is essential in brain development, was identified as the most significant pathway. Several upstream regulators were shared for most CF phenotypes, including the chemical decitabine, iron channel molecule KCNJ2, growth factor TGFB1, and estrogen receptor ESR1. In addition, beta-estradiol was identified as a significant common upstream regulator for all the CF phenotypes; for four of the CF phenotypes (attention, mental flexibility, working memory, and PAOFI total score), it was the most significant upstream regulator (rank 1) and for processing speed, it was the second most significant (Table 4).

Table 4.

IPA upstream regulator results for beta-estradiol.

Trait	p-value of overlap	B-H corrected p-value	Rank	Table
Processing speed	1.74E–10	9.94E–07	2	S8
Attention	1.48E–11	1.52E–07	1	S10
Executive function	1.82E–07	2.55E–04	8	S12
Mental flexibility	8.92E–14	1.08E–09	1	S14
Verbal memory	2.24E–07	2.82E–04	9	S16
Learning and memory	5.32E–06	1.10E–02	4	S18
Working memory	2.77E–10	3.19E–06	1	S20
PAOFI total score	2.06E–10	2.36E–06	1	S22

No SNPs were found for cg10331779 in the mQTL lookup analyses. One cis mQTL rs684940 (chr6:53961513 in MLIP) was identified for cg25906741 (p-value=$1.65\times {10}^{-22}$, beta = −0.328). However, no associations between this SNP and CF phenotypes were found in the GWAS Catalog (http://www.ebi.ac.uk/gwas/, RRID:SCR_012745) [37].

Three databases were applied comparing blood and brain DNAm levels. First, the blood brain DNA methylation comparison tool (https://epigenetics.essex.ac.uk/bloodbrain/) showed that cg10331779 in CTNND2 was significantly correlated with DNAm in the superior temporal gyrus (p-value = 0.033, r = 0.25), a region involved in language processing, and the mean blood DNAm was higher than mean DNAm in all the four brain regions (prefrontal cortex, entorhinal cortex, superior temporal gyrus, and cerebellum, Figure S24A). Secondly, BECon (https://redgar598.shinyapps.io/BECon/) identified a moderate correlation between blood DNAm and Brodmann area 10, which is linked to higher CFs like decision-making and planning, for cg10331779 (r = 0.34, Figure S24B). The third comparison tool, IMAGE-CpG (http://han-lab.org/methylation/default/imageCpG), showed no evidence of correlations between blood and brain DNAm for either cg10331779 or cg25906741.

4. Discussion

In this study, we evaluated the relationships between seven objectively measured CF domains from a comprehensive neuropsychological test battery and one subjective CF phenotype and DNAm in early-stage HR+ postmenopausal BC patients. We carried out two types of EWAS with and without the adjustment for CTH, which aided in understanding the role of cell-type compositions, followed by DMR analyses. Among the eight CF phenotypes, we identified two DMCs in the CTH-adjusted EWAS, cg10331779 in CTNND2 with processing speed and cg25906741 in MLIP associated with PAOFI total scores, and five DMRs including regions in/near SLC6A11, PRKG1/CSTF2T, and FAM3B for processing speed, and regions in/near PI4KB and SGCE/PEG10 for mental flexibility. Post-EWAS analyses were performed to identify the biological processes and pathways associated with CF.

The top DMC cg10331779 in the CTH-adjusted processing speed EWAS, with p-value=$9.65\times {10}^{-9}$, maps to the promoter region in CTNND2. Although in the sensitivity analysis using only participants who had not received ET this CpG did not reach genome-wide significance (p-value=$1.14\times {10}^{-4}$), their coefficients were similar with −0.0369 in the main analysis and −0.0398 in the sensitivity analysis (Figure S14). This potentially indicates that ET may not be driving the variation in methylation at cg10331779 associated with processing speed. Methylation of cg10331779 was also associated with poorer mental flexibility at a nominal significance level (p-value = 0.018, Table S3). Further searching found that DNAm levels of cg10331779’s of blood correlated with those in the superior temporal gyrus, an area involved in language processing, and Brodmann area 10, which contributes to high-level CF including decision-making and prospective memory [38–40]. CTNND2, which encodes an adhesive junction-associated protein delta catenin and is mainly expressed in the brain, has been widely studied in neurocognitive function [41,42]. Deletion of chromosome 5p where CTNND2 is located can cause Cri-du-chat syndrome which is characterized by a high-pitched cat-like cry, intellectual disability, and poor development [43,44]. Earlier studies showed that polymorphisms in CTNND2 may be associated with cognitive ability [45–47]. For instance, rs61749834 was a suggestive SNP of processing speed, the CF domain where differential methylation in CTNND2 was detected in the current study, for teenagers between 10 and 17 years of age [45]. SNPs rs62337555 and rs12519314 were associated with general cognitive performance in a multivariate GWAS when jointly analyzed with education-related variables and personality phenotype, respectively [46,47]. Furthermore, genetic variations in CTNND2 were shown to be linked with a variety of neurodevelopmental disorders, including autism, attention deficit hyperactivity disorder (ADHD), depression, Alzheimer’s disease (AD), and myopia [48–55]. CTNND2 is also known as a modulator of the Wnt/β-catenin signaling pathway which is associated with several diseases and disorders, including cancer and neurodegenerative diseases [56–58]. Our results from the IPA core analysis showed that a transcription regulator SOX2, which also regulates the Wnt/β-catenin signaling pathway, was identified as an upstream regulator for processing speed, playing an important role in nervous system development, neurodevelopmental disorders, and adult brain function [59,60].

DNAm variation at cg25906741 in MLIP was significantly associated with subjective PAOFI total scores in the main CTH-adjusted EWAS (p-value = $2.01\times {10}^{-8}$) and it was a suggestive DMC in the sensitivity analysis (p-value = $5.18\times {10}^{-6}$). In previous research, MLIP was found to be related to AD development and ADHD [61,62]. While MLIP is known to interact with lamin A/C (LMNA) associated with muscle-related disorders, studies on MLIP in CF and neurodevelopment are limited [63–67]. Cluett et al. showed that polymorphism in the LMNA gene was associated with CF, as measured by the Mini-Mental State Examination, in older adults [68]. In addition, mRNA levels of LMNA were increased in the hippocampus for patients with late-stage AD [69]. It is possible that the biological mechanisms of CF impairment in BC patients can partially overlap with AD pathology with the interaction between MLIP and LMNA.

We compared the associations of these two significant DMCs identified in our current study with a recently published EWAS study in the Generation Scotland cohort. McCartney and colleagues evaluated the relationship between DNAm and several CF measurements, including digit symbol tests, logical memory, verbal fluency, vocabulary, general fluid CF, and general CF [70]. Neither DMC showed significant associations with any CF phenotypes measured in the McCartney et al. study (Table S4, Figure S25).

In our study, we also noticed the potential impact of estrogen on CF for HR+ BC patients when identifying the upstream regulators using IPA. Beta-estradiol, a steroid estrogen hormone, was observed as an upstream regulator for all eight cognitive function phenotypes in both the main and the sensitivity analyses. It is converted from testosterone by aromatase and binds to the estrogen receptors (ER) to regulate ER target genes, and it plays an important part in HR+ BC by increasing the risks of cancer cell growth and migration [71,72]. Beta-estradiol is associated with CF and may be involved in the promotion of memory [73]. Therefore, based on previous evidence and our results, we hypothesize that ET, as a popular therapy to treat HR+ BC, may disrupt the synthesis or the activity of beta-estradiol that impacts CF by regulating downstream CF-related genes. Even in patients free of ET, levels of beta-estradiol also regulate CF (results not shown). However, future studies will be required to demonstrate this relationship.

We identified several CF-associated DMRs mapping to genes that maintain normal cognition and brain functions. Among these, SLC6A11, which encodes a GABA transporter protein, is associated with many neurological disorders including epilepsy and intellectual disability [74,75]. Additionally, PRKG1 and CSTF2T may be associated with learning ability in previous reports [76,77], while SGCE and PEG10, both imprinted genes, are connected to psychiatric disorders and tumor progression, respectively [78,79]. Our results provide additional evidence that variations in DNAm across these genes may be associated with variation in CF in BC patients.

Our study has several potential limitations. First, due to the small sample size, this study may miss true DMCs associated with CF and could produce false positives. Second, our discoveries have not been validated in independent cohorts. Larger validation cohorts can help determine whether the results are robust. Finally, the heterogeneity of ET status should not be neglected. Because of the pandemic, the original enrollment criteria were changed and so some participants had already started ET when their blood sample was collected. We addressed this by conducting a sensitivity analysis where only participants who were free of ET at blood sample collection were included. By comparing the main and sensitivity analyses, the significant DMCs were different. The observed differences were not surprising given the smaller sample size used in the sensitivity analyses, taking the potential influence of the winner’s curse into account [80]. Although the p-values differed, the effect sizes of the significant and suggestive DMCs were similar between the two analyses (Figure S12-S19), suggesting no clear evidence of confounding effects. Moreover, we would like to note that even if the observed DNAm was influenced by such effects, this would not preclude the potential use of DNAm as a predictive biomarker [22]. In addition, since EPICC participants were treated with ET after enrollment (if not before) and followed up for 6 months, we plan to conduct subsequent studies to enhance the understanding of how ET impacts DNAm signatures and CF.

5. Conclusion

In conclusion, our study evaluated the relationship between DNAm variations and CF phenotypes in early-stage HR+ BC patients and identified several CpGs and DMRs that may enhance the understanding of the biological mechanisms behind CF performance in this population. Further validation in independent cohorts will be necessary to verify our discoveries.

Funding Statement

This manuscript was funded by the National Cancer Institute of the National Institutes of Health under award numbers [R01CA221882] (Conley, Bender, Erickson) and [R01CA196762] (Bender, Erickson). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Article highlights

• Blood genome-wide DNAm data were assessed in early-stage HR+ breast cancer patients to explore the biological foundations of cognitive function changes.

• EWAS revealed DNAm at cg10331779 near CTNND2 was associated with processing speed and cg25906741 in MLIP was associated with subjective cognitive function PAOFI total scores.

• Beta-estradiol was the common upstream regulator for all the cognitive function phenotypes, suggesting a potential role of hormone regulation in cognitive function variation.

Author contributions

Shuwei Liu: Data curation, formal analysis, investigation, methodology, visualization, writing – original draft preparation, writing – review and editing

Dongjing Liu: Investigation, methodology, writing – review and editing

Catherine Bender: Conceptualization, funding acquisition, writing – review and editing

Kirk Erickson: Conceptualization, funding acquisition, writing – review and editing

Susan Sereika: Data curation, methodology, writing – review and editing

John Shaffer: Conceptualization, methodology, funding acquisition, supervision, writing – review and editing

Daniel Weeks: Conceptualization, investigation, formal analysis, methodology, resources, funding acquisition, supervision, writing – review and editing

Yvette Conley: Conceptualization, methodology, resources, funding acquisition, supervision, project administration, writing – review and editing

Disclosure statement

Dongjing Liu contributed to this work when she was a PhD student at the University of Pittsburgh. At the time of manuscript submission, she is an employee of GlaxoSmithKline. The company did not sponsor or fund this research.

The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Ethical Declaration

This study was reviewed and approved by the Institutional Review Board (IRB) of the University of Pittsburgh (STUDY19080223 and STUDY19010174). All participants provided written consent forms at enrollment.

Data availability statement

The data that support the findings of this study are openly available in the database of Genotypes and Phenotypes (dbGaP) at https://www.ncbi.nlm.nih.gov/gap/under the accession number phs003959.v1.p1 (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs003959.v1.p1).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/17501911.2025.2542116