Inflammatory Proteins Mediate the Causal Association between Sleep Traits and Breast Cancer: A Mendelian Randomization Study.

Introduction
Breast cancer (BC) is the most prevalent malignancy among women and ranks as a leading cause of cancer-related death [1]. According to GLOBOCAN 2022, over 2.29 million new female BC cases were recorded annually, making up nearly a quarter (23.8%) of all newly diagnosed cancers in women, while approximately 666,000 deaths (15.4% of female cancer mortality) were attributable to the disease [2]. The development of BC is the result of multiple interacting factors, including overweight and obesity, smoking, alcohol consumption, early menarche, late menopause, postmenopausal hormone therapy, benign breast disease, and genetic predisposition [3]. Because several of these risk factors are modifiable, clarifying the relationship between modifiable lifestyle behaviors and BC risk is essential for developing effective preventive strategies and reducing disease burden to relieve public health stress.
Sleep is a physiological process that fundamentally maintains metabolic homeostasis, immune competence, and circadian rhythm [4, 5]. Long-term sleep disorders can lead to chronic low-grade inflammation by activating the hypothalamic-pituitary-adrenal axis, enhancing pro-inflammatory pathways such as nuclear factor kappa B, and increasing inflammatory factors such as interleukin (IL)-1β, IL-6, and tumor necrosis factor-α [6]. Research indicates that short (<7 h) or long (>9 h) nightly sleep, insomnia, and use of sleep aid medications are associated with elevated concentrations of multiple inflammatory biomarkers in female breast tissue [7]. In addition, insufficient sleep is also causally associated with inflammatory diseases [8], further demonstrating the important role of sleep in inflammation regulation. The presence of chronic inflammation promotes angiogenesis, tumor invasion, metastasis, and progression [9]. A previous study demonstrated that heightened inflammation is in turn linked to increased BC risk [10].
The links between sleep disorders and inflammation, and between inflammation and BC, have been clearly studied. Nevertheless, findings on specific sleep traits and BC incidence have been inconsistent. A study reports significantly higher BC risk among women with insomnia, parasomnia, or obstructive sleep apnea [11]. Another study has indicated that, compared with women who sleep 6–7 h per night, longer sleepers face a higher risk of BC, whereas shorter sleep duration shows no significant association with BC risk [12]. A Spanish population-based case-control study further linked regular napping to higher BC risk yet found no significant links for sleep quality or total sleep duration [13]. These discrepancies may reflect differences in study populations, study design, residual confounding, or reverse causation. It is worth noting that the existing research focuses on the single effect of sleep or inflammation, and no research has systematically assessed the causal chain between sleep, inflammatory characteristics, and BC risk within the same analytical framework, and its potential biological mechanism lacks genetic evidence to support.
Mendelian randomization (MR) is an analytical method in epidemiology that employs genetic variants as instrumental variables (IVs) to enable causal inference, thereby reducing the potential for confounding and addressing reverse causation [14]. Against this background, the present study used a two-sample MR design to systematically assess the causal relationships between sleep traits and BC risk. The mediating analysis of inflammatory proteins was introduced for the first time to clarify the potential inflammatory mechanism of sleep affecting the occurrence of BC. This study is expected to provide a genetic basis for individualized sleep management and provide a reference for early prevention strategies of BC.

Methods

Study Design
A two-sample MR design was applied to estimate the causal relationships of 11 sleep-related phenotypes and inflammatory proteins on overall BC and its four main molecular subtypes (luminal A, luminal B, HER2-enriched, and triple-negative). Furthermore, the mediating roles of inflammatory proteins were evaluated. We also performed reverse MR analyses to exclude reverse causation. To minimize bias from population heterogeneity, the analysis utilized GWAS summary statistics derived exclusively from individuals of European descent, drawn from nonoverlapping cohorts and representing independence. Every MR estimate was derived under the three fundamental assumptions: (1) Relevance: IVs are strongly associated with the exposure; (2) independence: IVs are independent of confounders; and (3) exclusion: IVs do not affect the outcome directly but through the exposure [15, 16] (Fig. 1a, b). The study follows the STROBE-MR reporting guidelines [17].

GWAS Summary Data
We incorporated GWAS data of 11 sleep-related phenotypes from the Common Metabolic Disease Knowledge Portal (CMDKP) database (https://md.hugeamp.org/downloads.html#SD). Participants from the UK Biobank study were aged 40–69 years, including both males and females, with information adjusted for age and sex factors. Short sleep duration was defined as ≤6 h per night and long sleep duration as ≥9 h per night [18]. Chronotype information was derived from a meta-analysis of 449,734 individuals who self-reported their chronotype and encoded as definitely morning (2), more morning than evening (1), more evening than morning (−1), or definitely evening (−2) [19]. The number of sleep episodes and sleep efficiency were derived from accelerometer records of 85,670 UK Biobank participants. Efficiency was calculated as sleep duration divided by the sleep period time window [20]. The data on sleep apnea syndrome were obtained from five cohorts in the UK, Canada, Australia, the USA, and Finland, involving 362,638 samples and 25,008 patients [21].
GWAS data of inflammatory proteins originated from 14,824 European-ancestry adults across 11 cohorts in whom 91 plasma proteins were quantified with the Olink Target inflammation panel. The data were adjusted based on age and gender factors, and proteomic data were generated at Olink Laboratory in Uppsala, Sweden [22]. These protein GWAS summary datasets are publicly available through the GWAS Catalog (https://www.ebi.ac.uk/gwas/). Detailed characteristics of 91 inflammatory proteins are listed in online supplementary Table S1 (for all online suppl. material, see https://doi.org/10.1159/000550315).
Data of BC and its subtypes were retrieved from 82 studies by the Breast Cancer Association Consortium (BCAC), comprising 133,384 BC cases and 113,789 controls, all women of European descent. Subtype-specific counts were 14,900 triple-negative, 57,400 luminal A, 13,800 luminal B and 6,400 HER2-enriched cases, with 91,477 controls [23]. These data can be downloaded from https://www.ccge.medschl.cam.ac.uk/breast-cancer-association-consortium-bcac. Table 1 summarizes all GWAS datasets used in the present investigation.

Selection of IVs
IV selection criteria were as follows: (1) Genome-wide significant single-nucleotide polymorphisms (SNPs) (p < 5 × 10−8) were used as IVs. If the number of SNPs available under strict thresholds was insufficient to support robust causal inference, the screening criteria should be appropriately relaxed. We adjusted the significance level to p < 5 × 10−6 in analyses of short sleep duration and sleep apnea syndrome and to p < 1 × 10−6 for inflammatory proteins. In the reverse-MR analysis, p < 5 × 10−6 was applied to HER2-enriched BC. (2) SNPs were clumped (r2 = 0.001, window = 10,000 kb) to remove linkage-disequilibrium effects and guarantee mutual independence of selected IVs. (3) To exclude SNPs associated with known confounding factors, candidate SNPs were screened and matched against known traits (https://www.ebi.ac.uk/gwas/) to exclude SNPs associated with established confounders, including body mass index (BMI), smoking, socio-economic position, age at menopause, or any BC-related outcomes [24]. (4) Harmonization of exposure and outcome datasets was performed to align allele directions, and ambiguous palindromic and incompatible SNPs were removed. (5) Steiger filtering was applied to discard SNPs whose explained variance in the outcome exceeded that in the exposure. (6) To test weak IVs, the proportion of variance explained (R2) and F statistic were calculated for each SNP. All SNPs included had F statistics >10 to minimize the bias of weak IVs. (7) Finally, the MR-PRESSO outlier test was executed to detect outlying SNPs to eliminate bias and correct for horizontal pleiotropy. The number of IVs and F statistic used in forward and reverse MR analyses are shown in online supplementary Figure S1.

MR Analysis
The random-effects inverse variance-weighted (IVW) method was the primary analysis. Robustness was evaluated with four additional regression models: MR Egger, weighted median, weighted mode, and simple mode. For exposures with a single IV, causal inference relied on the Wald ratio estimator. For exposures with multiple IVs, the causal effect was derived principally from the IVW method.
To dissect mediating effects, we implemented two steps for the MR analysis. First, we identified sleep-related traits and inflammatory proteins that exhibited significant causal effects on BC and its subtypes and with evidence of no horizontal pleiotropy in their genetic associations with the BC outcome. Second, guided by these causal estimates and applying identical analytical criteria, we quantified the causal influence of each sleep-related phenotype on inflammatory proteins. The causal association between sleep-related phenotypes and overall or subtype-specific BC is represented by the total effect (β). We retained potentially mediating factors that were logically consistent, based on the direction of the effect. The causal association between sleep-related phenotypes and inflammatory proteins is denoted as β1, while the causal effect of inflammatory proteins on BC is β2. Mediation was quantified as the product of the two path coefficients (β1 × β2), with the proportion mediated obtained by dividing this term by the overall effect. The 95% confidence interval of the mediation effect was estimated with the delta method [25].

Sensitivity Analyses
Outlier SNPs were identified with the MR-PRESSO test and removed before the final MR analysis. Cochran’s Q test was used to assess heterogeneity across the IVs. p < 0.05 was taken as evidence of heterogeneity. Since the IVW estimator uses a random-effects model by default and allows certain levels of heterogeneity, modest heterogeneity does not compromise the interpretation of the results. To assess violation of the standard IV assumptions, we evaluated potential horizontal pleiotropy with MR-Egger regression and the MR-PRESSO global test. MR-Egger performs weighted linear regression without an unconstrained intercept. A statistically significant intercept (p < 0.05) indicates horizontal pleiotropy. Analyses showing evidence of horizontal pleiotropy were excluded, ensuring the robustness of the MR findings. The direction of causality was verified with Steiger’s filtering test to rule out reverse causation [26]. Finally, to verify that the overall MR estimate was not unduly swayed by any one variant, we carried out a leave-one-out analysis, by sequentially removing each exposure-related SNP and recalculating the causal effect.

Co-Localization Analysis
In order to further evaluate the shared genetic signal between inflammatory proteins and BC at the gene level, we performed co-localization analysis on significant inflammatory proteins [27]. Co-localization analysis aims to determine whether two phenotypes are driven by the same genetic variation within the same gene region. Specifically, we first extracted all SNPs within the 50 kb upstream and downstream regions of each inflammatory protein-coding gene to ensure coverage of potential regulatory sites near the gene. Subsequently, the “coloc” package was used for analysis to calculate five posterior probabilities (PPH0-PPH4), where PP4 represented the probability of two phenotypes sharing the same genetic signal. Co-localization evidence was defined as PPH4>0.75.

Statistical Analysis
All analyses were performed in R (v4.4.1) using the “TwoSampleMR” package (v0.6.13) and the “MR-PRESSO” package (v1.0). Statistical significance was determined at p < 0.05.

Results

Genetic Associations between Sleep-Related Phenotypes and Overall or Subtype-Specific BC
IVs for the 11 sleep-related phenotypes are listed in online supplementary Table S2. The number of IVs was between 14 and 130, and the F-statistic range was 21–221, avoiding potential weak IV bias. IVW estimates showed that genetically proxied morning chronotype was inversely associated with overall BC risk (OR = 0.910, 95% CI: 0.852–0.973, p = 0.006). Morning chronotype decreased risks for both overall BC (OR = 0.936, 95% CI: 0.893–0.980, p = 0.005) and the luminal A subtype (OR = 0.944, 95% CI: 0.894–0.996, p = 0.037). In contrast, total sleep duration displayed a positive association with overall BC (OR = 1.146, 95% CI: 1.007–1.304, p = 0.039) and with luminal A BC (OR = 1.358, 95% CI: 1.151–1.602, p < 0.001). Long sleep specifically increased the risk of triple-negative BC (OR = 9.433, 95% CI: 2.419–36.775, p = 0.001) and luminal A BC (OR = 2.186, 95% CI: 1.111–4.302, p = 0.023). Short sleep was associated with a lower risk of overall BC (OR = 0.482, 95% CI: 0.284–0.818, p = 0.007) and of luminal A BC (OR = 0.385, 95% CI: 0.194–0.766, p = 0.007). Sleep apnea syndrome also emerged as a risk factor for overall BC (OR = 1.106, 95% CI: 1.005–1.218, p = 0.040) (Fig. 2a; online suppl. Table S3).
In the sensitivity analyses, we excluded the causal estimates for sleep duration on overall BC and luminal A BC, as well as the effects of chronotype (morning person), chronotype, and sleep apnea syndrome on BC, because all of these showed evidence of pleiotropy. All remaining estimates passed directional testing and leave-one-out validation, indicating robustness (Table 2; online suppl. Fig. S2). There was no evidence that the observed relationships were driven by reverse causation. After discarding reverse-MR estimates that exhibited horizontal pleiotropy, we noted that overall BC and luminal A BC significantly increased the risk of insomnia, while luminal B BC showed a positive association with short sleep duration (Fig. 2b, Table 2; online suppl. Table S4).
Considering the association between sleep quality and BMI [28], to evaluate the possible mediating effect pathway of BMI, we repeated the MR analysis while retaining the BMI-related SNP. The results showed that the daytime nap was related to the increased risk of HER2-positive BC, indicating that the daytime nap might indirectly affect the risk of HER2-positive BC through BMI (online suppl. Fig. S3; online suppl. Table S5). Other results were consistent with the results of the principal analysis, indicating that the MR results were robust.

Genetic Associations between Inflammatory Proteins and Overall or Subtype-Specific BC
The final sets of IVs for the inflammatory proteins are listed in online supplementary Table S6. The number of IVs was between 1 and 14, and the F-statistic range was 24–1,216. Figure 3 summarizes the causal relationships between these inflammatory proteins and overall or subtype-specific BC. Overall, BC showed significant inverse associations with 8 proteins, including IL-22RA1, CASP8, and IL-17C, and positive associations with 6 proteins, notably CD40L, ADA, and CCL19. LIF, CD5, and MMP1 significantly reduced the risk of HER2-enriched BC, while IFN-γ, IL-5, and IL-20RA were associated with increased risk of HER2-enriched BC. CXCL11, IL-6, LIFR, and FLT3L were inversely associated with luminal A BC, whereas CD40L, ADA, and CCL19 were positively associated with luminal A BC. IL-24 alone showed a negative association with luminal B BC. Six inflammatory proteins, CASP8, AXIN1, and CD244, acted as protective factors against triple-negative BC, whereas TNFSF12, SCF, and β-NGF emerged as risk factors. Notably, IFN-γ, LIFR, and MMP1 displayed concordant associations across overall BC and multiple subtypes (Fig. 3a; online suppl. Table S7).
Sensitivity analyses found no sign of horizontal pleiotropy in any of the forward MR estimates linking inflammatory proteins to BC or its subtypes, and all associations passed the Steiger directionality test for causal inference (online suppl. Table S8). Reverse MR analyses indicated that the observed forward associations were not affected by reverse causation. After exclusion of outliers with horizontal pleiotropy, we observed positive associations between overall BC and CCL28, HER2-enriched BC and both CXCL9 and IL-17C, and luminal A BC and DNER. Luminal B BC was inversely associated with TNFSF12 but positively associated with IL-10, whereas triple-negative BC was inversely associated with S100a12 and positively associated with both IL-10 and CCL28 (Fig. 3b; online suppl. Tables S8–S9). In addition, we also selected cis loci encoding gene regions ±1 Mb for these 30 significant inflammatory proteins and performed MR according to the same standards. The results were consistent with the above analysis, indicating the robustness of the results (online suppl. Fig. S4; online suppl. Table S10). In the co-localization analysis, there was evidence of co-localization between inflammation-related proteins, such as ADA and CASP8 and luminal A BC (PPH4 >0.75) (online suppl. Table S11). For the results without co-localization evidence, this may reflect that the driving variation of MR results mainly comes from trans-action sites rather than local regulatory signals of protein loci.

Mediation Analysis
On the basis of the above findings, we selected 30 inflammatory proteins as potential mediators of the causal pathways linking sleep-related phenotypes to the risk of overall or subtype-specific BC. We then estimated the causal effects of the three sleep-related phenotypes, short sleep, long sleep, and morning chronotype, on each of these proteins. MR revealed a positive causal effect of short sleep duration on elevated CXCL11 levels (OR = 3.443, 95% CI: 1.284–9.233, p = 0.014) and a positive effect of morning chronotype on S100a12 (OR = 1.103, 95% CI: 1.015–1.200, p = 0.021). Neither association was influenced by reverse causation (Fig. 4a, b; online suppl. Table S12). Sensitivity analyses confirmed the absence of pleiotropy or heterogeneity, indicating robustness of the results (online suppl. Table S13). Further mediation analysis demonstrated that CXCL11 mediated 22.4% of the protective effect of short sleep duration on luminal A BC risk (Table 3).

Discussion
Through a two-sample MR analysis, we systematically evaluated the causal contribution of multiple sleep traits to BC risk and quantified the extent to which inflammatory proteins mediate these relationships. The study found that a morning chronotype exerted a protective effect against BC. Short sleep duration was associated with a reduced risk of overall and luminal A BC, whereas long sleep duration significantly increased the risk of triple-negative BC and luminal A BC. Furthermore, this study identified that CXCL11 played a partial mediating role in the causal relationship between short sleep duration and luminal A BC. These findings provided robust genetic evidence supporting a causal link between sleep traits and BC.
Circadian disruption has long been suspected to drive the tumorigenesis of BC. The circadian rhythm is regulated by the master biological clock located in the suprachiasmatic nucleus of the hypothalamus, which maintains physiological homeostasis by modulating melatonin secretion, body temperature fluctuations, and various hormone levels [29]. Epidemiological cohorts have repeatedly shown that prolonged night-shift work suppresses melatonin secretion, accelerates telomere shortening, and increases BC risk [30–32]. Evening-type populations often exhibit a significant misalignment between the timing of melatonin secretion and sleep onset, leading to circadian disruption. Individuals with morning chronotype, in contrast, generally demonstrate more regular sleep-wake schedules, more stable circadian rhythms, and more pleasant awakening experiences, with minimal discrepancy between melatonin secretion and sleep initiation [33, 34]. A stable circadian rhythm helps maintain the normal function of DNA repair mechanisms, prevents the accumulation of genetic mutations, suppresses excessive inflammatory activation, and modulates immune responses, thereby reducing the risk of BC [29].
Patients with obstructive sleep apnea have been found to have a significantly increased risk of BC [11]. However, in this study, the causal relationship between sleep apnea and BC has a level of pleiotropy. It may reflect that the sleep apnea is genetically related to a variety of confounding factors (such as obesity, metabolic disorders, and inflammation-related characteristics) and indirectly affects the risk of BC. The impact of sleep apnea on BC is not a direct impact made by itself. As for sleep efficiency, a previous study has suggested that its improvement can reduce the overall mortality of BC patients [35]. However, in our causal analysis, we did not find that sleep efficiency is significantly associated with the risk of BC. This may be due to the small impact of sleep efficiency on the incidence of BC or the limitation of sample phenotype measurement.
Findings in the present study indicated that long sleep duration markedly raised the risk of triple-negative BC and luminal A BC, which is largely consistent with previous epidemiological findings. Prior research similarly demonstrated that compared to women with 6–9 h of sleep per night, women with longer sleep duration (≥9 h per day) had a 59% increase in the risk of BC [36]. Prolonged sleep may promote tumorigenesis in BC through several mechanisms. First, long sleepers exhibit markedly elevated levels of inflammatory factors such as IL-6 and CRP [37]. IL-6, a key inflammatory mediator in BC development, stimulates tumor-cell proliferation, migration, and invasion and induces angiogenesis, thereby driving deterioration of the tumor microenvironment and tumor progression [38]. Second, excessive sleep duration is associated with higher risks of obesity and metabolic syndrome, both recognized risk factors for BC [39, 40].
Short sleep duration was associated with a lower risk of overall BC and the luminal A BC in the present study, a finding that diverges from earlier reports. Most previous observational studies have either linked short sleep with a significantly increased BC risk or found no significant association [12, 36]. This discrepancy likely reflects the unavoidable reverse causation, measurement error in self-reported sleep duration, and residual confounding. In this study, the data on sleep duration came from participants’ self-reported data, and there may still be some measurement errors. Although using genetic IVs for MR analysis can alleviate reverse causality and confounding bias to some extent, there may still be potential horizontal pleiotropy, longitudinal pleiotropy, or other unmeasured confounding factors that can lead to residual bias. Therefore, future research can further verify the causal relationship between sleep and BC and improve the certainty of conclusions through objective monitoring (such as wearable sleep tracking equipment), large-scale multicenter cohort, and more sophisticated phenotypic hierarchical studies.
In mediation analyses, we further showed that CXCL11 mediated part of the protective effect of short sleep on luminal A BC, accounting for 22.4% of the total causal pathway. CXCL11, a key chemokine secreted by multiple cells, including tumor cells, monocytes, and fibroblasts, recruits activated T cells and NK cells to infection or tumor sites, thereby promoting anti-tumor immunity [41]. High CXCL11 levels have been associated with enhanced immune activation, greater infiltration of anti-tumor immune cells, and improved sensitivity to both immunotherapy and chemotherapy [42]. At present, research on the direct impact of sleep duration on CXCL11 expression is relatively limited, but there is evidence to suggest that CXCL11 is regulated by circadian rhythms [43]. Thus, short sleep may reinforce anti-tumor immune surveillance and consequently lower the risk of luminal A BC by upregulating CXCL11 expression. It is worth noting that the immune tolerance of luminal A BC is lower than that of triple-negative BC, which may make it more vulnerable to CXCL11-mediated immune mechanism regulation [44]. It should be emphasized that this study reveals the causal relationship between the genetic prediction of short sleep tendency and the risk of BC, rather than the behavioral effect of actively shortening sleep. Genetic evidence does not equate to clinical or public health recommendations. Numerous studies have confirmed that long-term sleep deprivation leads to cardiovascular disease, metabolic syndrome, and increased all-cause mortality [45–47]. Even if there is a genetic association between short sleep and reduced risk of certain cancer subtypes, the negative effects of sleep deprivation on overall health far outweigh the possible partial benefits.
This study identified causal associations between multiple inflammatory proteins and overall or subtype-specific BC. IFN-γ exerted the dual functions of promoting tumor progression and inhibiting tumor growth. In our data, it acted as a risk factor for overall, luminal A, and HER2-enriched BC. There are several possible mechanisms. First, IFN-γ modulates the metabolism or microenvironment of tumor cells, which helps malignant cells withstand stress and evade immune attack. Second, inflammation is restrained by eliciting a homeostatic response, which promotes tumor cells to produce immunosuppressive molecules or recruit immunosuppressive cells, thereby establishing an immunosuppressive tumor microenvironment [48]. LIFR was protective for BC overall, luminal A, and triple-negative BC. Previous studies have shown that LIFR can induce a dormant subtype in BC cells, thereby inhibiting proliferation and metastasis [49]. Further research has demonstrated that histone deacetylase inhibitors can significantly suppress primary tumor growth by upregulating LIFR expression and activating intrinsic dormancy-regulating programs in BC cells [50]. These findings suggested that LIFR is not only a key protective factor against BC progression but also a potential therapeutic target for novel treatment strategies.
Reverse MR analyses revealed influences of BC on sleep traits. BC overall and its luminal A subtype in particular significantly increased the risk of insomnia, whereas the luminal B subtype was positively associated with short sleep duration. In a prospective observational cohort, the prevalence of insomnia symptoms and insomnia disorder rose from 24.9% before diagnosis to 46.1% at diagnosis and remained at approximately 50% thereafter. Chemotherapy and baseline insomnia severity index were the strongest predictors of persistent insomnia disorder [51]. Additional drivers include anxiety, reduced paraspinal muscle strength and tumor-related pain [52]. Luminal B subtype typically displays a high proliferation index, which may elicit heightened physiological stress and consequently poorer sleep quality and shorter sleep duration [53].
Our findings carried important clinical and public health implications. The protective effect of morning chronotype suggested that cultivating healthy sleep habits and maintaining a stable circadian rhythm may be an effective, low-cost strategy for BC prevention. Clinicians should therefore integrate sleep-health management into BC prevention and treatment, especially for high-risk populations. The observation that short sleep duration was associated with lower risks of overall and luminal A BC must, however, be interpreted cautiously. It should not be taken as an endorsement of chronic sleep restriction, which is linked to multiple adverse health outcomes. In fact, women who routinely slept ≤6 h before diagnosis experienced worse BC survival [54]. The sleep-BC link appears appreciably more intricate than earlier recognized and must be appraised against a backdrop of individual heterogeneity, sleep quality, and global lifestyle factors. When formulating sleep health guidelines, more attention should be paid to sleep quality rather than just the duration of sleep.
Limitations also deserve mention. First, the GWAS data used were derived predominantly from European-ancestry cohorts, potentially limiting generalizability to other ethnicities. Second, genetic susceptibility is likely to interact with environmental exposures such as lifestyle, psychosocial stress, and dietary patterns, but these interaction effects were not explicitly captured in our MR analyses. Third, some sleep traits were self-reported and hence subject to measurement error. Validation in multiethnic cohorts and the molecular mechanism underlying sleep-BC associations are therefore warranted.

Conclusion
Morning chronotype protects against BC. Short sleep duration is associated with reduced risk of overall and luminal A BC, whereas long sleep increases the risk of triple-negative and luminal A BC. CXCL11 mediates part of the protective effect of short sleep on luminal A BC. Reverse MR further demonstrates that BC can reciprocally disrupt sleep. These findings clarify the sleep-inflammation-BC causal chain and provide an evidence base for preventive strategies centered on sleep management.

Statement of Ethics
Ethical approval and consent were not required as this study was based on publicly available data.

Conflict of Interest Statement
The authors have no conflicts of interest to declare.

Funding Sources
Funding sources of this study were Taizhou University School of Medicine Young Doctors Research and Teaching Pilot Fund Project, yxbz202405. School of medicine, Taizhou University, Taizhou, 318000, Zhejiang, China.

Author Contributions
Jiawei Zhou and Lijun Mao contributed to data analysis and drafting and revising the article, gave the final approval of the version to be published, and agreed to be accountable for all aspects of the work.