The mediating effects of the triglyceride-glucose index in the association between circadian syndrome and breast cancer: Evidence based on NHANES and TCGA databases

Abstract

Objective

Breast cancer (BC) is a major global health issue for women. Although insulin resistance (IR) and circadian disruption are both associated with BC, their interaction remains unclear. This study examined the association between circadian syndrome (CircS) and BC, and explored the mediating role of the triglyceride-glucose (TyG) index and related molecular mechanisms.

Methods

This cross-sectional study analyzed 5,498 women from the NHANES 2005–2014. Multivariate logistic regression assessed the association between CircS and BC, with mediation analysis to evaluate TyG’s indirect effect. Bioinformatics validation used TCGA-BRCA data, including differential expression of TyG and circadian rhythm-related gene sets, survival analysis, and pathway enrichment.

Results

Among 5,498 participants, 158 had BC. After adjusting for sociodemographic, lifestyle, and cardiovascular factors, CircS was significantly associated with higher BC risk (OR = 2.07, 95% CI: 1.12–3.83). Each one-unit increase in the TyG index was also linked to increased BC risk (OR = 1.64, 95% CI: 1.20–2.23). Mediation analysis showed the TyG index mediated 10.84% of CircS’s effect on BC (Indirect Effect: 0.0042, 95% CI: 0.0011–0.0103). Bioinformatics analysis identified dysregulated prognostic circadian rhythm DEGs and insulin receptor DEGs in tumors, and their expression correlated with overall survival. These genes were co-enriched in oncogenic pathways such as lipid metabolism and MAPK signaling.

Conclusions

This nationally representative study indicates that CircS is strongly associated with a higher prevalence of BC, with IR—measured by the TyG index—partially mediating this association. Molecular evidence shows that circadian and metabolic disruptions converge on key signaling pathways, suggesting a synergistic role in BC development.

1. Introduction

Breast cancer (BC) remains the most commonly diagnosed cancer and a leading cause of cancer-related mortality among women worldwide, posing a substantial public health challenge. ¹ In 2023, approximately 2.3 million new cases were diagnosed globally. BC is the most prevalent cancer in 164 countries and the primary cause of cancer-related death in 134 of them. ² Therefore, it is necessary to identify modifiable risk factors and actionable biomarkers for early detection and primary prevention of BC, given the considerable socioeconomic impact of BC, including direct medical costs, loss of productivity, and long-term care of survivors. ³

Insulin resistance (IR) has been increasingly recognized as a key metabolic disorder, facilitating the development and progression of BC. ⁴ Especially among women after menopause, IR is associated with an elevated risk of BC. Hyperinsulinemia and heightened IGF-1 bioavailability are the crucial contributors, activating the PI3K/Akt and MAPK pathways while interacting with estrogen signaling. ⁵

The triglyceride-glucose (TyG) index has been validated as a reliable and accessible surrogate marker of IR. Meanwhile, it shows strong associations with gold-standard hyperinsulinemia-euglycemic clamp measurements. ⁶ Recent epidemiological studies have further shown that an elevated TyG index is associated with an increased risk of various cancers, including BC, suggesting its utility in risk stratification for cancer. ⁷

Beyond traditional metabolic dysregulation, circadian disruption has become a crucial contributor to the development of chronic diseases, including cancer. ⁸ Zimmet et al. introduce circadian syndrome (CircS) as an integrative clinical entity in 2019. This concept extends beyond conventional metabolic syndrome by including core features of circadian misalignment, such as sleep abnormalities and mood disorders. ⁹ CircS is defined as ≥ four of these components: abdominal obesity, raised FG, dyslipidemia, hypertension, shortened sleep duration, and depression. ⁹ Circadian disruption has been shown to alter the secretion of hormones, immune function, and cellular metabolism, potentially activating oncogenic pathways and impairing DNA repair mechanisms. ¹⁰ Consistent epidemiological and experimental evidence demonstrates that disrupted biological clocks are strongly associated with the progression of cancer. ¹¹

In recent years, transcriptomic studies have revealed that circadian disruption plays a crucial role in BC. Li et al. found that there were significant individual differences in the circadian clock function of patients with luminal A BC. ¹² Furthermore, higher circadian clock activity was associated with lower 5-year survival rates and more pronounced periodic activation of the Epithelial-Mesenchymal Transition (EMT) pathway. This finding suggests that tumor progression depends not on the mere presence or absence of the circadian clock, but on the strength of its function, which in turn affects the invasiveness of cancer cells. Ector et al. analyzed 14 BC cell lines and classified them into four rhythmic phenotypes: normal, weak, unstable, and disordered. ¹³ Different phenotypes showed significant differences in drug response, thereby refining the resolution of molecular typing. Accumulating evidence indicates that circadian disruption is a key mechanism for the occurrence and development of BC and has promising clinical utility. These findings have facilitated the integration of population-level circadian risk and multi-omics data.

Although both IR and circadian disruption are independently associated with BC, no population-based research has investigated whether IR, as assessed by the TyG index, mediates the association between CircS and BC. Furthermore, the shared molecular mechanisms underlying these associations are poorly defined. Therefore, exploiting data from the National Health and Nutrition Examination Survey (NHANES) and The Cancer Genome Atlas (TCGA), this research sought to evaluate associations between CircS and the prevalence of BC, estimate the mediating effects of the TyG index, and validate these findings and explore potential mechanisms through bioinformatics analyses of TyG-related and CircS-related gene sets. Elucidating this metabolism-circadian axis may provide novel insights into the pathogenesis of BC and inform integrated prevention strategies.

2. Method

2.1. Study population

NHANES is a nationally representative survey assessing the health and nutritional status of adults and children in the U.S. It consists of cross-sectional cycles beginning in 1999. Data were collected through interviews at home and physical assessments and laboratory tests within mobile centers. Data are organized into five categories: demographic, dietary, examination, laboratory, and questionnaires. Detailed information on survey design is attainable at https://wwwn.cdc.gov/nchs/nhanes/Default.aspx. NHANES was approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board, and written informed consent was obtained from all participants. This study was a secondary analysis of publicly available de-identified data and therefore did not require additional ethical approval from our institution. This study was conducted in compliance with the Declaration of Helsinki (1975) and its 2024 revision. The reporting of this study conforms to the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines. ¹⁴

This research employed a cross-sectional design and analyzed data from NHANES (in the U.S. from 2005 to 2014). ¹⁵ In this study, 22,504 participants under 20 years of age were excluded. Additionally, we excluded 7,844 male participants, 6,060 participants with missing data on CircS, 7 participants with missing data on BC, 3,082 participants without complete data on triglyceride, and 13 participants without complete data on fasting blood glucose. After these exclusions, 5,498 participants were included in the analysis (Figure 1).

Figure 1.

Flow diagram of the study cohort selection.

2.2. Assessment of BC

The status of BC was determined based on participants’ self-reported response to the question, “Has a doctor or other health professionals ever told you that you had BC?” Participants answering ‘yes’ were classified as having a history of BC.

2.3. Definition of CircS

Trained personnel carried out anthropometric and biochemical evaluations in participants’ homes as well as in mobile examination centers. CircS was defined as meeting ≥ four of the following components established by Shi et al. ¹⁶ : enlarged waist circumference (≥88 cm in women and ≥102 cm in men), raised FG (≥100 mg/dL or use of antihyperglycemic medication), increased triglycerides (≥150 mg/dL serum triglycerides or use of medication), diminished HDL-cholesterol (serum HDL-C <40 mg/dL in men and <50 mg/dL in women or use of medication), raised blood pressure (≥130 mmHg systolic or ≥85 mmHg diastolic blood pressure or use of antihypertensive medication), shortened sleep duration (self-reported <6 hours per day), and symptoms of depression (a score of ≥5 on the Patient Health Questionnaire-9). A comprehensive description of CircS has been provided previously. ⁹

2.4. Calculation of triglyceride-glucose (TyG) index

TyG index was calculated by the following formula ⁶ : TyG = Ln [fasting triglycerides (mg/dL) ×  FG (mg/dL)/2].

2.5. Covariates

We chose covariates according to published evidence and biological plausibility rather than statistical criteria alone. ¹⁷ Following covariates were included in multivariate models: age (≤60, >60), poverty income ratio (PIR) (≤1.3,>3.5,1.3-3.5), marital status (MS) (married/living with partner, never married, widowed/divorced/separated), race/ethnicity (Mexican American, Non-Hispanic Black, Non-Hispanic White, Other Hispanic, Others-encompassing Multi-Racial), education (< high school, > high school, high school diploma/equivalent), body mass index (BMI) (<25,≥25),waist (cm), history of cardiovascular conditions (heart failure, coronary heart disease, angina, heart attack, stroke), alcohol drinking (≥12 cups per year, No/<12 cups per year), use of hormones (no/yes), smoke (no/yes).

2.6. Statistical analysis

All analyses accounted for the multifaceted survey design of NHANES by incorporating sampling weights, strata, and clusters. All continuous variables were tested for normality, and all followed a normal distribution. Continuous variables were expressed as weighted mean ± standard error, and categorical variables were presented as weighted percentages. Intergroup differences were assessed using weighted t-tests for continuous variables and chi-square tests for categorical variables. The variance inflation factor (VIF) was used to assess potential multicollinearity among the covariates included in this analysis. The results demonstrated that all covariates had VIF values below 10, indicating the absence of significant multicollinearity within the current model.

Multivariable logistic regression (MLR) models were employed to evaluate associations between CircS, TyG index, and BC. Three models were constructed: Model1 was unadjusted; Model2 was adjusted for age, BMI, race, education, MS, and PIR; Model3 was further adjusted for smoking, alcohol use, hormone use, and cardiovascular comorbidities.

Subgroup analyses were conducted to assess effect modification by key demographic and clinical variables. Interaction terms were introduced and tested using likelihood ratio tests.

Using a product-of-coefficients approach with bootstrapped confidence intervals (CIs) (1000 repetitions), mediation analysis was conducted to quantify the direct effects (DEs) of CircS on BC and the indirect effects (IEs) mediated by the TyG index. The mediated proportion was computed by dividing the IE by the total effect.

Sensitivity analyses comprised both unweighted logistic regression and propensity score matching (PSM) with 1:2 nearest-neighbor matching (without replacement) to balance baseline covariates between women with and without BC.

All statistical tests were two-sided, and P <0.05 signified statistical significance. Analyses were implemented utilizing R (version 4.4.3) with the survey, mediation, and MatchIt packages.

2.7. Bioinformatics analysis

To explore the molecular mechanisms underlying the associations observed in the epidemiological analysis, this research performed comprehensive bioinformatics validation using the publicly accessible TCGA database.

2.7.1. Acquisition of gene sets

Based on the associations between CircS, TyG index, and BC identified in our NHANES population analysis, we further explored the potential molecular mechanisms underlying these relationships. We accessed the GSEA/MSigDB database (https://www.gsea-msigdb.org/gsea/index.jsp) and entered the keywords ‘circadian rhythm’ and ‘insulin resistance’ on the search page. Notably, the circadian-rhythm-related gene set curated in this study is broader than canonical core circadian clock genes (e.g., BMAL1, PER2, NR1D1, CLOCK) and includes diverse components involved in circadian-relevant biological pathways. Subsequently, we selected multiple gene sets from the search results page, downloaded them, and merged them for downstream analyses (Table S1).

2.7.2. Data sources for expression and clinical validation

RNA sequencing data and corresponding clinical metadata of 1,103 breast tumor samples and 113 adjacent normal tissue samples were downloaded from the official TCGA website (available at https://portal.gdc.cancer.gov). Extracted clinical variables included overall survival (OS) time, OS status, and PAM50 molecular subtypes (luminal A, luminal B, HER2-enriched, and basal-like).

2.7.3. Identification of prognostic circadian rhythm DEGs (CR-DEGs) and insulin resistance DEGs (IR-DEGs)

Gene expression values of TCGA-BRCA (breast invasive carcinoma) were normalized from raw count data to log2-transformed transcripts per million (log2 (TPM)). Based on the screening criteria of |logFC| > 0.585 and an adjusted P-value (adj.p) ≤ 0.05, differentially expressed genes (DEGs) between tumor and normal tissues were identified using the limma R package. After we excluded samples with OS < 30 days, a survival analysis was performed in the TCGA-BRCA cohort using the survival and survminer packages to identify prognostic genes. We then intersected the sets of DEGs, prognostic genes, circadian rhythm genes, and insulin resistance genes, for obtaing the prognostic CR-DEGs and the prognostic IR-DEGs.

2.7.4. The comprehensive assessment of prognostic CR-DEGs and IR-DEGs

The diagnostic potential of these prognostic CR-DEGs and IR-DEGs in distinguishing breast tumor tissues from normal tissues was estimated using receiver operating characteristic (ROC) curve analysis. Additionally, pairwise correlations among prognostic CR-DEGs and IR-DEGs were further analyzed using the ComplexHeatmap package. Meanwhile, pathway enrichment analysis was conducted for each DEG using the Gene Set Variation Analysis (GSVA) package.

3. Results

3.1. Baseline features of study participants

This research included 5,498 participants from the NHANES database (2005–2014), of whom 5,340 were BC-free and 158 had BC. Significant differences were observed between the two groups across multiple baseline characteristics (Table 1).

Table 1.

Baseline features of individuals with or without breast cancer.

Characteristic	Overall	BC-free participants	BC patients	p-value
Age (years) group	​	​	​	<0.001
≤60	3,802 (74.61%)	3,764 (75.91%)	38 (27.60%)	​
>60	1,696 (25.39%)	1,576 (24.09%)	120 (72.40%)	​
PIR	​	​	​	0.11
≤1.3	1,830 (22.43%)	1,787 (22.47%)	43 (20.84%)	​
>3.5	1,579 (40.21%)	1,532 (40.45%)	47 (31.85%)	​
1.3-3.5	2,089 (37.36%)	2,021 (37.08%)	68 (47.31%)	​
Marital status	​	​	​	<0.001
Married/Living with partner	3,053 (61.98%)	2,963 (61.91%)	90 (64.56%)	​
Never married	929 (15.63%)	921 (15.99%)	8 (2.70%)	​
Widowed/Divorced/Separated	1,516 (22.39%)	1,456 (22.10%)	60 (32.74%)	​
Race/ethnicity	​	​	​	0.003
Mexican American	868 (7.42%)	854 (7.55%)	14 (2.53%)	​
Non-Hispanic Black	1,100 (11.57%)	1,070 (11.61%)	30 (10.09%)	​
Non-Hispanic White	2,548 (69.71%)	2,449 (69.37%)	99 (81.75%)	​
Other Hispanic	534 (4.97%)	525 (5.04%)	9 (2.47%)	​
Other Race - Including Multi-Racial	448 (6.33%)	442 (6.42%)	6 (3.16%)	​
Education	​	​	​	0.052
<High school diploma	1,346 (16.18%)	1,310 (16.18%)	36 (16.08%)	​
>high school diploma	2,951 (61.80%)	2,870 (62.07%)	81 (52.03%)	​
High school diploma/equivalent	1,201 (22.02%)	1,160 (21.75%)	41 (31.90%)	​
BMI(kg/m2)	​	​	​	0.9
<25	1,717 (34.60%)	1,668 (34.62%)	49 (33.78%)	​
≥25	3,781 (65.40%)	3,672 (65.38%)	109 (66.22%)	​
TG (mg/dL)	120.44 (102.84)	119.66 (102.15)	148.47 (122.49)	<0.001
Waist (cm)	96.18 (16.64)	96.14 (16.65)	97.67 (16.14)	0.2
Glucose (mg/dL)	102.32 (28.30)	102.07 (28.15)	111.56 (31.80)	<0.001
Daily sleep time(h)	6.94 (1.37)	6.94 (1.37)	6.95 (1.34)	0.8
Heart failure	​	​	​	0.001
No	5,348 (97.73%)	5,198 (97.84%)	150 (93.88%)	​
Yes	150 (2.27%)	142 (2.16%)	8 (6.12%)	​
CHD	​	​	​	0.081
No	5,367 (98.04%)	5,216 (98.10%)	151 (95.91%)	​
Yes	131 (1.96%)	124 (1.90%)	7 (4.09%)	​
Angina	​	​	​	0.7
No	5,382 (98.11%)	5,228 (98.10%)	154 (98.54%)	​
Yes	116 (1.89%)	112 (1.90%)	4 (1.46%)	​
Heart attack	​	​	​	0.15
No	5,341 (97.35%)	5,190 (97.41%)	151 (95.18%)	​
Yes	157 (2.65%)	150 (2.59%)	7 (4.82%)	​
Stroke	​	​	​	0.009
No	5,295 (96.68%)	5,150 (96.80%)	145 (92.29%)	​
Yes	203 (3.32%)	190 (3.20%)	13 (7.71%)	​
Alcohol drinking	​	​	​	0.084
At least 12 cups per year	3,374 (68.94%)	3,287 (69.17%)	87 (60.94%)	​
No/<12 cups per year	2,124 (31.06%)	2,053 (30.83%)	71 (39.06%)	​
Use of hormones	​	​	​	<0.001
No	4,374 (77.34%)	4,269 (77.69%)	105 (64.82%)	​
Yes	1,124 (22.66%)	1,071 (22.31%)	53 (35.18%)	​
Smoke status	​	​	​	0.7
At least 100 cigarettes in life	2,055 (39.31%)	1,993 (39.25%)	62 (41.48%)	​
No	3,443 (60.69%)	3,347 (60.75%)	96 (58.52%)	​
Hypertension	​	​	​	<0.001
No	3,117 (60.90%)	3,075 (61.69%)	42 (32.41%)	​
Yes	2,381 (39.10%)	2,265 (38.31%)	116 (67.59%)	​
Diabetes	​	​	​	<0.001
No	3,098 (60.20%)	3,039 (60.82%)	59 (37.88%)	​
Yes	2,400 (39.80%)	2,301 (39.18%)	99 (62.12%)	​
Depression	​	​	​	0.076
No	4,916 (91.28%)	4,777 (91.43%)	139 (85.82%)	​
Yes	582 (8.72%)	563 (8.57%)	19 (14.18%)	​
CircS	​	​	​	<0.001
No	3,444 (66.64%)	3,381 (67.42%)	63 (38.59%)	​
Yes	2,054 (33.36%)	1,959 (32.58%)	95 (61.41%)	​
TyG index	8.54 (0.64)	8.53 (0.64)	8.82 (0.63)	<0.001

Individuals with BC were significantly older compared to those without BC, and a higher proportion of these patients were aged over 60 years (72.40% vs. 24.09%, p<0.001). Significant differences were also observed in MS and race/ethnicity (p<0.01). Meanwhile, the BC cohort demonstrated higher proportions of widowed/divorced/separated individuals and non-Hispanic White patients.

Metabolically, individuals with BC had significantly higher triglyceride-glucose index (p<0.001). The prevalence of hypertension, diabetes, and CircS was also significantly heightened among individuals with BC (all p<0.001). Additionally, heart failure and stroke history were more common among individuals with BC (p<0.01).

No significant differences were observed in BMI, waist circumference, sleep duration, coronary heart disease, smoking, or alcohol use between the groups.

3.2. CircS was positively associated with BC

The results of MLR analyses showing associations between CircS and the prevalence of BC are demonstrated in Figure 2.

Figure 2.

Association between circadian syndrome (CircS) and breast cancer. Forest plot showing odds ratios (ORs) and 95% confidence intervals (CIs) for the association between CircS and breast cancer across three sequential adjustment models. Participants without CircS serve as the reference group. Model 1: unadjusted. Model 2: adjusted for age, BMI, race, education, marital status, and poverty income ratio (PIR). Model 3: further adjusted for smoking status, alcohol drinking, hormone use, heart failure, coronary heart disease, angina, heart attack, and stroke. The vertical dashed line indicates OR = 1 (no association).

In the unadjusted Model 1, populations with CircS had a significantly elevated prevalence of BC (OR = 3.29, 95%CI: 2.10–5.16). After adjusting for age, BMI, race, education, MS, and PIR in Model 2, CircS remained significantly associated with increased risks of BC (OR = 1.98, 95%CI: 1.08–3.61). In the extensively adjusted Model3, which further controlled for smoking status, alcohol drinking, hormone use, heart failure, CHD, angina, heart attack, and stroke history, CircS continued to show a marked positive association with BC (OR = 2.07, 95%CI: 1.12–3.83).

These results indicated that CircS was significantly associated with an elevated prevalence of BC, and this association remained robust after comprehensive adjustments for potential confounders.

3.3. Higher levels of TyG were associated with a higher prevalence of BC

The results of MLR analyses showing associations between the TyG index and the prevalence of BC are demonstrated in Table 2.

Table 2.

Associations between TyG and breast cancer.

Participants	Model 1		Model 2		Model 3
	OR (95% CI)	p-value	OR (95% CI)	p-value	OR (95% CI)	p-value
TyG
Continuous	1.86 (1.49, 2.33)	<0.001	1.60 (1.17, 2.18)	0.004	1.64 (1.20, 2.23)	0.002
T1	ref.	​	ref.	​	ref.	​
T2	2.12 (1.20, 3.75)	0.01	1.75 (0.96, 3.22)	0.069	1.77 (0.97, 3.21)	0.062
T3	3.30 (1.77, 6.16)	<0.001	2.16 (1.03, 4.52)	0.041	2.22 (1.07, 4.62)	0.034

OR = Odds Ratio, CI = Confidence Interval.

Model 1: Unadjusted.

Model 2: Adjusted for age, BMI, race, education, marital status and PIR.

Model 3: Adjusted for age, BMI, race, education, marital status, PIR, smoke status, alcohol drinking, use of hormones, heart failure, CHD, angina, heart attack and stroke.

When analyzed as a continuous variable, the TyG index was significantly associated with risks of BC in the unadjusted Model 1 (OR = 1.86, 95%CI: 1.49–2.33, p < 0.001). This relationship remained significant after age, BMI, race, education, MS, and PIR were adjusted for in Model 2 (OR = 1.60, 95%CI: 1.17–2.18, p = 0.004). In the extensively adjusted Model 3, which controlled for additional covariates including lifestyle and cardiovascular factors, each unit elevation in the TyG index was associated with 64% higher probability of BC (OR = 1.64, 95%CI: 1.20–2.23, p = 0.002).

3.4. Significant interactions of age and education level in the association between CircS and BC

Due to small event counts within certain subgroups that could yield unstable effect estimates and low statistical power, marital-status categories of “never married” and “widowed/divorced/separated” were combined into one group. For race/ethnicity, Mexican American, Non-Hispanic White, Other Hispanic, and Other Race-Including Multi-Racial were pooled into the “Other Race” category, while Non-Hispanic White remained as a separate reference group. For the same reason of insufficient event numbers, the heart failure subgroup was excluded from subgroup analyses. Subgroup analysis demonstrated pronounced heterogeneity in the association between CircS and risks of BC across different population characteristics (Table 3).

Table 3.

Subgroup analysis of associations between CircS and breast cancer based upon different variables.

Subgroup	N	OR (95% CI)		p-value	p for interaction
		No	Yes
Age (years) group	​	​	​	​	0.003
≤60	38/3802	ref.	6.50 (1.45, 29.06)	0.015	​
>60	120/1696	ref.	1.29 (0.80, 2.08)	0.289	​
Race/ethnicity	​	​	​	​	0.126
Non-Hispanic White	99/2548	ref.	1.80 (0.88, 3.65)	0.100	​
Other Race	59/2950	ref.	4.46 (1.48, 13.5)	0.009	​
Education	​	​	​	​	0.047
<High school diploma	36/1346	ref.	0.72 (0.29, 1.77)	0.469	​
>High school diploma	81/2951	ref.	1.72 (0.85 3.47)	0.127	​
High school diploma/equivalent	41/1201	ref.	4.82 (1.30, 17.80)	0.019	​
Marital status	​	​	​	​	0.104
Married/Living with partner	90/3053	ref.	3.16 (1.28, 7.81)	0.014	​
Never married/Widowed/Divorced/Separated	68/2445	ref.	1.09 (0.56,2.14)	0.800	​
PIR	​	ref.	​	​	0.795
≤1.3	43/1830	ref.	1.90 (0.77, 4.73)	0.162	​
>3.5	47/1579	ref.	1.38 (0.51, 3.73)	0.519	​
1.3-3.5	68/2089	ref.	2.87 (1.09, 7.60)	0.034	​
BMI	​	ref.	​	​	0.130
<25	49/1717	ref.	3.70 (1.07, 12.83)	0.039	​
≥25	109/3781	ref.	1.48 (0.82, 2.68)	0.191	​
Smoke status	​	ref.	​	​	0.239
At least 100 cigarettes in life	62/2055	ref.	2.85 (0.93, 8.76)	0.066	​
No	96/3443	ref.	1.63 (0.79, 3.36)	0.184	​
Alcohol drinking	​	ref.	​	​	0.377
At least 12 cups per year	87/3374	ref.	1.78 (0.73, 4.36)	0.203	​
No/<12 cups per year	71/2124	ref.	2.85 (1.37, 5.95)	0.006	​
Use of hormones	​	ref.	​	​	0.090
No	105/4374	ref.	2.88 (1.34, 6.23)	0.008	​
Yes	53/1124	ref.	1.07 (0.47, 2.43)	0.875	​
Heart failure	​	ref.	​	​	0.726
No	150/5348	ref.	2.04 (1.09, 3.81)	0.027	​
Yes	8/150	ref.	6.40 (0.39, 106.14)	0.169	​
Heart attack	​	ref.	​	​	0.315
No	151/5341	ref.	2.06 (1.09, 3.87)	0.026	​
Yes	7/157	ref.	0.28 (0.01, 13.98)	0.481	​
Stroke	​	ref.	​	​	0.205
No	145/5295	ref.	2.30 (1.18, 4.51)	0.016	​
Yes	13/203	ref.	0.41 (0.07, 2.37)	0.298	​

Note. Adjusted for age, race, education, marital status, PIR, BMI, smoke status, alcohol drinking, use of hormones, heart attack, stroke for subgroup variables. PIR:Ratio of family income to poverty; BMI, body mass index; CI, confidence interval; OR, odds ratio. The bold values mean significant p < 0.05.

In age-stratified subgroups, the association between CircS and BC was more pronounced among participants ≤60 years (OR = 6.50, 95%CI: 1.45–29.06, p = 0.015) than in those >60 years (p = 0.289), with a significant interaction effect (p for interaction = 0.003).

Additionally, education level demonstrated a significant interaction effect (p = 0.047). A marked association between CircS and BC was solely detected in the high school diploma/equivalent group (OR = 4.82, 95%CI: 1.30–17.80, p = 0.019). Stratified analysis by race showed that CircS was significantly associated with breast cancer risk in all racial groups except Non-Hispanic White (p=0.009). No significant interaction effect was detected (p=0.126). In other subgroups, the relationship between CircS and BC remained pronounced among the “Other Race” group (excluding Non-Hispanic White) (OR = 4.46, 95% CI: 1.48–13.5, p = 0.009), the Married/Living with partner group (OR = 3.16, 95% CI: 1.28–7.81, p = 0.014), non-drinkers (OR = 2.85, 95%CI: 1.37–5.95, p = 0.006), non-hormone users (OR = 2.88, 95%CI: 1.34–6.23, p = 0.008), and those without heart failure, heart attack, or stroke history (p < 0.05).

These results indicated that associations between CircS and risks of BC were significantly modified by age and education level, suggesting that these demographic characteristics might exert pivotal effects in the association between CircS and BC.

3.5. The association between TyG and BC was consistent across different subgroups

The positive association between TyG and BC was notably significant in subgroups aged ≤ 60 years, Non-Hispanic Whites, other races, those with high school education, non-drinkers, the Married/Living with partner group, the Never married/Widowed/Divorced/Separated group, and non-hormone users (all p < 0.05). Most interaction test results were not pronounced (p> 0.05), indicating a consistent association across populations (Table S2).

3.6. Partial mediation of the TyG index in the association between CircS and BC

Mediation analysis was performed to quantify the extent to which the TyG index mediated the association between CircS and BC. The total effect represents the overall association between CircS and BC. The DE (DE = 0.0341, 95% CI: 0.0160–0.0410, p < 0.001) represents the portion of the association not mediated through the TyG index, whereas the IE (IE = 0.0042, 95% CI: 0.0011–0.0103, p < 0.05) represents the portion of the association mediated through the TyG index. The calculation formula for the mediation proportion is IE/TE × 100%, which reflects the extent to which the mediating variable explains the total effect. The results of this study showed that the TyG index mediated 10.84% of the association between CircS and BC (Figure 3).

Figure 3.

Analysis of TyG-mediated associations between CircS and breast cancer. The direct effect (DE) represents the impact of CircS on BC that is not mediated by the TyG index, while the indirect effect (IE) refers to the impact transmitted through the TyG index. The calculation formula for the mediating proportion is IE/TE × 100%, which reflects the extent to which the mediating variable explains the total effect. The results of this study indicate that the TyG index mediates 10.84% of the association between CircS and BC.

3.7. The positive association between CircS and BC remained robust after PSM

To evaluate the robustness of associations between CircS and BC and to control for potential confounding bias, we conducted two sensitivity analyses (Tables S3 and S4). First, an unweighted logistic regression analysis showed that CircS remained significantly associated with risks of BC (extensively adjusted OR = 1.66, 95% CI: 1.15–2.43, p = 0.008), aligning with the primary weighted analysis (Table S3).

To further address confounding bias introduced by covariates, this research performed a sensitivity analysis utilizing PSM. Nearest neighbor matching was applied at a 1:2 ratio to balance the distributions of age, BMI, race, education, MS, PIR, and other clinical characteristics between the BC and non-BC groups (Figure S1). After matching, CircS continued to exhibit a significant association with risks of BC across all models (extensively adjusted OR = 1.84, 95% CI: 1.05–3.22, p = 0.035) (Table S4). These sensitivity analyses indicated the robustness of positive associations between CircS and risks of BC. These associations were consistent across different statistical approaches, even after potential confounders were balanced through PSM.

3.8. Identification of prognostic circadian rhythm- and insulin resistance-related genes in BC

Based on the GSEA/MSigDB database, we obtained 297 circadian rhythm genes and 170 IR genes. The details are presented in Table S1.

3.8.1. Differential expression and prognostic significance of CR-DEGs and IR-DEGs

To identify prognostic CR-DEGs and IR-DEGs, we performed an intersection analysis of DEGs, prognostic genes, circadian rhythm genes, and insulin resistance genes. Finally, we obtained 2 prognostic IR-DEGs and 5 prognostic CR-DEGs. Among them, ACSL1 and SOCS3 were more highly expressed in the normal group (Figure 4(A) and (B)). Survival analysis further indicated that low expression of ACSL1 was associated with a better prognosis, whereas the opposite trend was observed for SOCS3 (Figures 5(A) and (B); 6(A) and (B)). Meanwhile, SUV39H2 was upregulated in tumors (Figure 4(G)) and related to a poorer prognosis (Figures 5(G) and 6(G)). In contrast, EGR3, KLF15, NR1H3 and NTRK3 were expressed at lower levels across all tumor subtypes than in normal tissues (Figures 4(C)–(F)), and the high expression correlated with a more favorable prognosis (Figures 5(C)–(F) and 6(C)–(F)).

Figure 4.

Expression profiles of DEGs across breast cancer molecular subtypes and normal tissues. This Figure displays the expression levels of 7 DEGs (including IR-DEGs and CR-DEGs) in breast cancer molecular subtypes (Basal, Her2, LumA, LumB) and adjacent normal tissues, visualized by violin plots. Different colors represent distinct groups (purple: Basal; pink: Her2; yellow: LumA; blue: LumB; teal: Normal). Statistical significance (assessed by one-way ANOVA with post-hoc tests) is marked as ns (not significant), P < 0.001, indicating significant differential expression between tumor subtypes (especially LumB) and normal tissues for most genes.

Figure 5.

Kaplan-Meier overall survival curves stratified by median gene expression. Kaplan-Meier curves showing overall survival (OS) in breast cancer patients, stratified by high (above median) and low (below median) expression of 7 DEGs (IR-DEGs and CR-DEGs) stratified by high (orange) and low (green) expression levels. The x-axis represents follow-up time in years; the y-axis represents survival probability. Log-rank test p-values are provided for each gene to indicate statistical significance of the survival difference between expression groups.

Figure 6.

Kaplan-Meier overall survival curves stratified by optimal gene expression cut-off value. Kaplan-Meier curves showing overall survival (OS) in breast cancer patients, stratified by high (red) and low (green) expression of 7 DEGs (IR-DEGs and CR-DEGs) based on optimal expression cut-off values. The x-axis represents follow-up time in years; the y-axis represents survival probability. Log-rank test p-values are provided for each gene to indicate statistical significance of the survival difference between high and low expression groups defined by optimal cut-offs.

3.8.2. The comprehensive assessment of prognostic CR-DEGs and IR-DEGs

When we further analyzed the diagnostic potential and correlation of the DEGs, ACSL1 (AUC = 0.890) and KLF15 (AUC = 0.969) demonstrated favorable discriminatory ability, and a significant positive association was observed between them (Figure 7(A) and (B)). The GSVA results indicated that these genes primarily encompassed the activation of tumor progression (cancer pathways, cell cycle), metabolic regulation (lipid metabolism, type II diabetes pathways), signal transduction (MAPK, Wnt pathways), and immune-related pathways. Meanwhile, all genes uniformly inhibited genetic information processing (RNA degradation, DNA repair) and basic metabolic pathways (Figure 8(A)–(G)).

Figure 7.

Diagnostic value and inter-gene correlation of DEGs in breast cancer. (A) ROC curve analysis: ROC curves (y-axis: Sensitivity; x-axis: 1 – Specificity) evaluate the diagnostic ability of 7 DEGs to distinguish breast cancer tissues from normal tissues. (B) Correlation matrix: Scatter plots show the relationship between each pair of genes, with red fitted lines indicating correlation direction and strength. Diagonal panels display gene expression distributions as histograms. Correlation coefficients and significance levels (p < 0.05, **p < 0.001) are shown in the upper panels.

Figure 8.

GSVA pathway enrichment analysis of DEGs in breast cancer. This Figure displays GSVA pathway enrichment profiles for 7 DEGs (IR-DEGs and CR-DEGs) in breast cancer. Each subplot (A–G) corresponds to one gene, with the left panel listing enriched KEGG pathways and the right panel showing differential pathway activity (t value of GSVA score) between high-expression (Up, pink/orange) and low-expression (Down, blue/green) groups. Positive t values indicate pathway enrichment in the high-expression group, whereas negative values indicate enrichment in the low-expression group. Key oncogenic pathways (e.g., PI3K-Akt, MAPK signaling) are predominantly enriched in the high-expression groups of the genes, revealing their functional roles in facilitating the progression of breast cancer.

4. Discussion

Utilizing nationally representative data from NHANES 2005-2014, this large cross-sectional study provides important insights into complicated associations between CircS, TyG index, and risks of BC. Our main findings demonstrate that CircS is markedly linked to elevated prevalence of BC, even after comprehensive adjustments for potential confounders. Also, as a surrogate marker of IR, the TyG index shows a consistent positive association with the prevalence of BC. Furthermore, the TyG index mediates roughly 10.84% of the total effects of CircS on risks of BC, indicating a partial mediating role through IR pathways. Moreover, further bioinformatics analysis confirmed and deepened this discovery at the molecular level, revealing its potential gene expression and pathway mechanisms.

The robust association between CircS and BC (extensively adjusted OR = 2.07) strengthens the accumulating evidence linking circadian disruption to carcinogenesis. ⁸ Our findings are particularly significant as they demonstrate that the composite measure of CircS (which integrates metabolic, sleep, and mood components) serves as a stronger indicator of BC prevalence than its individual components alone. Although this study analyzed CircS as a syndrome, the impacts of its individual components remain unclear. Further studies are needed to determine the effects of specific components within it, such as central obesity, IR, sleep disorders, or depressive symptoms. Circadian disruption is mechanistically linked to the progression of cancer through multiple pathways, including alterations in hormonal secretion patterns (particularly melatonin and cortisol), which may subsequently affect estrogen signaling and promote tumorigenesis.^10,18 The impaired secretion of nocturnal melatonin observed in circadian disruption not only loses its direct oncostatic effects but may also enhance estrogen receptor-mediated signaling, thereby creating a permissive environment for the progression of BC. ¹⁹ Furthermore, core clock genes like CLOCK, BMAL1, and PER families have been shown to regulate pivotal cellular processes encompassing cell cycle progression, DNA damage repair, and apoptosis-all crucial mechanisms in the prevention of cancer. ²⁰

A maintained positive association between the TyG index and the prevalence of BC is observed in this research. After comprehensive adjustment, for every one-unit elevation in the TyG index, the prevalence of BC increases by 64%. This finding significantly underscores that IR plays a crucial role in the pathogenesis of BC. Our findings align with prior studies, which demonstrate the role of hyperinsulinemia in breast carcinogenesis.^4,21 The biological plausibility of this association is supported by several mechanisms. Hyperinsulinemia resulting from IR activates the PI3K/Akt and MAPK signaling pathways, promoting cell proliferation and inhibiting apoptosis in breast tissue.^21,22 Additionally, IR is linked to diminished synthesis of sex hormone-binding globulin (SHBG), inducing increased bioavailable estradiol and potentially impacting hormone receptor-positive BCs. ²³ As a readily accessible and cost-effective marker of IR, the TyG index might function as an effective tool for determining women with higher odds of BC in clinical practice, especially in resource-limited settings where more sophisticated measures of IR are unavailable. Although the TyG index is widely used as a practical surrogate marker for assessing IR, it does not fully capture all aspects of IR, particularly hyperinsulinemia. Moreover, recent studies have shown that triglyceride metabolism is closely associated with circadian rhythm robustness, ²⁴ chronotype, and environmental light exposure. ²⁵ Given that both circadian rhythm disruption and light-related behaviors are linked to the risk of BC,^26,27 the mediating effect observed with the TyG index may reflect more complex interactions among metabolic dysfunction, circadian regulation, and environmental factors.

The mediation analysis reveals that the TyG index mediates roughly 10.84% of the effect of CircS on risks of BC, providing novel insights into the potential mechanisms linking circadian disruption and BC. This finding suggests that although IR is a mechanism linking the two conditions, it is not the sole one. The modest mediation proportion indicates that multiple mechanisms are likely to contribute to the association between CircS and BC. Other potential pathways may include inflammatory processes (elevated pro-inflammatory cytokines), oxidative stress, altered gut microbiota composition, and DEGs of clock gene dysregulation on oncogene expression.^8,11,28 Notably, circadian disruption has been shown to activate the NF-κB pathway and facilitate production of pro-inflammatory cytokines like IL-6 and TNF-α, which can create a tumor-promoting microenvironment. ²⁹ Additionally, reactive oxygen species generated due to circadian disruption may cause DNA damage and genomic instability, further increasing risks of cancer. ³⁰

Subgroup analyses indicated notable heterogeneity in associations between CircS and BC, providing insights for targeted prevention strategies. The stronger association observed in younger women (≤60 years) with CircS (OR = 6.50) may reflect more pronounced circadian disruption in premenopausal and perimenopausal women, whose hormonal fluctuations might interact with circadian rhythms. This finding is particularly relevant given that circadian disruption exerts a more pronounced effect on menstrual cycle regularity and ovulatory function in younger females. ³¹ These contributing factors need to be further explored. ³² The significant association observed specifically in the high school diploma/equivalent group (but not in other education categories) might suggest complex socioeconomic interactions or differential access to healthcare resources. To explore the molecular basis of the above epidemiological associations, this research used the TCGA-BRCA data. Our bioinformatics analysis revealed that not all genes associated with TyG and CircS were uniformly upregulated. Instead, they exhibited a ‘bidirectional dysregulation’ pattern: certain genes (e.g.SUV39H2) were highly expressed in tumors and served as potential oncogenic drivers. Notably, these up-regulated genes showed preferential overexpression in the luminal B and basal-like subtypes, which are more aggressive BC phenotypes characterized by higher proliferation rates and poorer prognosis. These findings suggest that metabolic and circadian disturbances are particularly pronounced in these high-risk subtypes. Nevertheless, others (e.g., KLF15, SOCS3, NTRK3) were significantly downregulated, and their high expression was correlated with a favorable prognosis, indicating that these genes might function as endogenous protective factors. For example, SOCS3 is a crucial negative regulator of insulin/inflammatory signaling, and its loss of function can enhance the activation of pro - cancer pathways such as JAK - STAT. ³³ KLF15 is a core integrator of the biological clock and metabolism, ³⁴ and its downregulation may result in the disruption of cellular metabolic homeostasis. The widespread inactivation of these protective factors may be one of the core molecular events through which CircS-related metabolic disorders promote the progression of tumors. Recent transcriptomic studies have provided complementary evidence for this complex regulatory landscape. Li et al. have demonstrated that circadian clock strength in luminal A breast tumors is widely heterogeneous among patients. Furthermore, high clock strength is associated with reduced 5-year survival and enhanced cycling of the epithelial-mesenchymal transition pathway. ¹² This finding suggests that it is the functional status of the circadian clock, not its mere amplitude, that may determine its oncogenic or tumor-suppressive role. Additionally, Ector et al. have performed deep circadian phenotyping across 14 BC cell lines and identified four distinct circadian-based phenotypes: functional, weak, unstable, and dysfunctional clocks. ¹³ The bidirectional expression patterns we observed among circadian genes may thus reflect the heterogeneous clock phenotypes described by these authors. This observation further supports the notion that not the mere presence of the clock, but its functional state and metabolic interactions, are critical for the pathogenesis of BC. Functional enrichment analysis provides mechanistic insight into the biological role of the TyG index. These genes, whether upregulated or downregulated, primarily exert their effects by regulating lipid metabolism and the MAPK signaling pathway, thereby influencing the development of BC. This discovery carries important implications. It suggests that the circadian disruption associated with CircS and the IR represented by the TyG index may collaboratively establish a microenvironment favorable to tumor growth. Mechanistically, this collaboration involves the synergistic promotion of lipid metabolic reprogramming, which provides cancer cells with energy and membrane components, along with the sustained activation of pro-proliferative signals such as MAPK. Meanwhile, the consistent suppression of these genes impairs DNA repair and other fundamental maintenance pathways, further undermining genomic stability. From a systems biology perspective, this finding provides a rational molecular model: TyG index mediates the association between CircS and BC. Notably, the evidence that circadian clock features shape drug sensitivity ¹³ suggests that circadian-metabolic profiling may guide both risk stratification and therapeutic optimization in patients with BC.

Several methodological strengths enhance the validity and generalizability of these findings. The utilization of NHANES data with its complex sampling design ensures national representativeness of the non-institutionalized U.S. population. This research performed comprehensive adjustment for potential confounders across multiple domains (sociodemographic, behavioral, metabolic, and cardiovascular). Furthermore, rigorous sensitivity analyses were conducted, including both unweighted models and PSM, to minimize residual confounding. The mediation analysis provides mechanistic insights beyond simple associations, which represents an important strength compared to previous studies. Furthermore, the standardized protocols for anthropometric and laboratory measurements in NHANES reduce measurement bias and enhance the reliability of our exposure assessments. Finally, this research combined multi-level evidence from population epidemiology and bioinformatics to strengthen the robustness of our conclusions. These findings also have potential clinical implications. CircS components are routinely assessed in primary care. Consequently, CircS can be easily identified without additional cost. The TyG index offers a simple, reproducible marker for IR, and this marker could be integrated into risk assessment tools. Our bioinformatics analysis identified both risk-associated genes (ACSL1, SUV39H2) and protective genes (SOCS3, NTRK3) converging on lipid metabolism and MAPK signaling. These genes may serve as potential therapeutic targets or prognostic biomarkers. Most importantly, both CircS components and IR are modifiable. Lifestyle interventions such as improving sleep quality, optimizing light exposure, increasing physical activity, and adopting healthier dietary patterns could offer low-cost, accessible strategies for reducing the risk of BC. These strategies are particularly relevant for shift workers and individuals with irregular lifestyles. However, these clinical implications remain hypothesis-generating and need to be validated in prospective studies.

However, this study also has several limitations. First, because of the cross-sectional design, temporal relationships and causal inferences cannot be established, and residual confounding may still exist despite the use of PSM and extensive covariate adjustment. Second, several key variables were subject to measurement limitations. The status of BC was self-reported and may have been misclassified. Additionally, the TyG index is an indirect measure that may not fully capture IR, and the PHQ-9 is a proxy measure that may not fully capture the psychological dimension of CircS. Third, although the overall sample size was adequate, the relatively small number of BC cases and the limited number of events in certain subgroups may have reduced the precision of subgroup estimates and statistical power. Fourth, because NHANES primarily represents the U.S. population, the generalizability of our findings to other ethnic populations remains uncertain. Finally, the bioinformatics analyses were based on observational associations, and the underlying biological mechanisms need to be further validated through experimental studies. Therefore, prospective studies in more diverse populations are required to confirm these findings and explore the underlying mechanisms.

5. Conclusion

This study suggests that CircS may be associated with an elevated risk of BC. Population-based analysis indicates that CircS may correlate with a higher BC prevalence. This relationship may be partially mediated by IR, as measured by the TyG index. At the molecular level, bidirectional expression patterns of metabolic and circadian-clock-related genes in tumors tissues, along with high expression of risk-related genes such as ACSL1 and SUV39H2, may correlate with poor prognosis. These genes may collectively activate lipid metabolism and oncogenic pathways such as MAPK signaling. Our findings may suggest a potential mechanism by which metabolic processes and circadian clock function synergistically contribute to BC development.

Appendix.

Abbreviations

BC

Breast cancer

IR

Insulin resistance

FG

Fasting glucose

TyG

Triglyceride-glucose

CircS

Circadian syndrome

TCGA

The Cancer Genome Atlas

BMI

Body mass index

VIF

Variance inflation factor

MLR

Multivariable logistic regression

CIs

Confidence intervals

IEs

Indirect effects

PSM

Propensity score matching

DEs

Direct effects

GSEA

Gene Set Enrichment Analysis

OS

Overall survival

K-M

Kaplan-Meier

HRs

Hazard ratios

ROC

Receiver operating characteristic

AUC

Area under the ROC curve

PIR

Poverty Income Ratio

DEGs

Differentially expressed genes

CR-DEGs

Circadian rhythm DEGs

IR-DEGs

Insulin resistance DEGs

ORCID iD

Xun Peng https://orcid.org/0009-0007-8288-9403

Ethical considerations

Ethical review and approval were waived for this study, as it was conducted using publicly available, de-identified data from the National Health and Nutrition Examination Survey (NHANES). Because NHANES data are anonymized and collected with informed consent by the National Center for Health Statistics, additional ethical approval was not required for this secondary data analysis.

Data Availability Statement

The data presented in this study are openly available in the National Health and Nutrition Examination Survey (NHANES) database (https://www.cdc.gov/nchs/nhanes/index.htm).