Longitudinal changes in gut microbiota composition during endocrine therapy in hormone receptor-positive breast cancer patients

Abstract

This prospective study of 90 breast cancer patients examined gut microbiota composition relative to hormone receptor status and endocrine therapy changes. Initial analysis suggested hormone receptor-negative patients had higher Fusobacteriaceae and Fusobacterium abundance, while hormone receptor-positive patients showed Ruminiclostridium enrichment, though differences lacked statistical significance after correction. Hormone receptor-positive patients without lymph node metastasis demonstrated potentially greater microbial diversity, but associations were non-significant after multiple comparison correction. Longitudinal analysis of 52 hormone receptor-positive patients revealed the most robust finding: statistically significant Blautia increases after hormone therapy and aromatase inhibitor treatment. Tamoxifen showed trends toward increased Lachnospiraceae but lost significance after correction due to small sample size. LHRH agonist treatment demonstrated significant Dialister and Megasphaera increases. This study identified limited but robust associations between gut microbiota and endocrine treatments, with Blautia as the most consistently affected genus across multiple therapies, though most findings require validation in larger cohorts.

Introduction

Breast cancer remains the most prevalent malignancy among women worldwide, with approximately 2.4 million women affected and causing over 500,000 deaths annually¹. Hormone receptor-positive (HR+) breast cancer represents approximately 75% of all invasive breast tumors², with endocrine therapy (ET) serving as the foundation of treatment, either alone or combined with targeted therapies such as CDK4/6, mTOR, or PI3K inhibitors. While several actionable biomarkers guide treatment selection, including PIK3CA somatic mutations, HER2-low status, and germline BRCA1/2 and PALB2 mutations^3,4, breast cancer classification has traditionally relied on four immunohistochemical (IHC) biomarkers: estrogen receptor (ER), progesterone receptor (PR), HER2, and Ki67⁵.

Five intrinsic molecular subtypes of breast cancer identified over the past 15 years have revealed clinical implications beyond traditional pathology-based classifications^6,7. Within HR + /HER2-negative cancers, Luminal A and B subtypes predict 10-year outcomes and recurrence risk. HER2-enriched tumors show the best response to targeted therapies and Luminal A demonstrates better overall outcomes. In triple-negative breast cancer, the predominant Basal-like subtype (70–80%) may predict treatment responses to various therapeutic approaches, including chemotherapy regimens and bevacizumab⁴. Molecular profiling provides clinically relevant information that complements and extends beyond the current pathological-based classification system endorsed by the 2013 St. Gallen Consensus Recommendations. The intrinsic subtypes demonstrate biological heterogeneity and clinical differences in terms of incidence, survival outcomes, and treatment response.

Adjuvant endocrine therapy is a cornerstone treatment for HR+ breast cancer, with multiple strategies shown to reduce recurrence risk and provide additional benefits in preventing contralateral breast cancer⁸. Based on clinical trial data including MA.17R, NSABP B-42, DATA, IDEAL, ABCSG 16, and SOLE trials^9–13, recommended extended duration strategies include: aromatase inhibitor (AI) for up to 10 years total; tamoxifen for 2–3 years followed by AI for 7–8 years; tamoxifen for 5 years followed by AI for 5 years; or tamoxifen for 10 years. The decision for extended therapy should be individualized based on key risk factors, including positive lymph node status (the strongest indication), large tumor size, high tumor grade, high Ki-67, and receipt of chemotherapy^8,14,15. Advanced assessment tools such as Clinical Treatment Score post-5 years (CTS5) and genomic risk scores (e.g., Breast Cancer Index) can further refine patient selection^14–16. Duration recommendations vary by risk: high-risk patients may receive up to 10 years total ET, intermediate-risk patients 7 years, while low-risk patients generally should not extend beyond 5 years¹⁴. Patient factors, including age, menopausal status, comorbidities, prior treatment tolerance, and quality of life considerations, must be evaluated^14,15,17. Women with node-positive disease should be offered extended AI therapy up to 10 years total, while those with lower-risk node-negative disease may reasonably stop after 5 years unless strongly motivated to prevent late recurrence or contralateral breast cancer. Extended therapy carries ongoing adverse effects, including bone health concerns, and none of the studies have shown overall survival benefits with longer AI treatment - benefits are primarily in preventing recurrence and second breast cancers. The guidelines emphasize shared decision-making between clinicians and patients, considering individual risk factors, tolerance of side effects, and potential benefits versus risks of extended treatment^8,17.

The gut microbiota plays a crucial role in estrogen metabolism through the “estrobolome”—the collective bacterial genes capable of metabolizing estrogens^18,19. This collection of enteric bacterial genes modulates the risk of ER+ breast cancer. Endogenous estrogens, including estradiol, estrone, and estriol, are produced predominantly in the ovaries, adrenals, and adipose tissue, and undergo conjugation by first-pass hepatic metabolism before being excreted renally or via bile into feces²⁰. Conjugated estrogens excreted into bile are processed by the estrobolome, where specific bacterial species deconjugate these estrogens, leading to their reabsorption²¹. Women with lower plasma estrogen concentrations demonstrate reduced fecal glucuronidase activity²². In a mouse model, multiple bacterial taxa with beta-glucuronidase activity changes upon administration of conjugated estrogens and bazedoxifene²³. This suggests that probiotic supplementation to modulate gut microbial activity may alter the half-life or serum concentration of tamoxifen, a selective ER modulator, to optimize its therapeutic effect in breast cancer treatment²⁴. However, the specific influence of gut microbiota on hormonal treatment of breast cancer remains unknown.

Recent evidence demonstrates that gut microbiota modulates cancer treatment responses, including those for HR+ breast cancer^25–27. The estrobolome influences systemic estrogen levels through bacterial deconjugation of estrogens in the gut^19,28, potentially affecting the progression of hormone-dependent breast cancers²⁹. The estrobolome may also impact the efficacy of hormonal therapies commonly used to treat HR+ breast cancer²⁶. Studies have demonstrated that gut microbiota influences cancer treatment outcomes through multiple mechanisms, including immune modulation, drug metabolism, and inflammatory pathway regulation^30–32. For HR+ breast cancer specifically, emerging evidence suggests that specific bacterial species may enhance or impair the effectiveness of endocrine therapies^33,34. Limited real-world data exist regarding the impact of gut microbiota on endocrine therapy in breast cancer patients. We planned this prospective longitudinal study to understand these interactions with the goal of developing novel therapeutic strategies to improve treatment outcomes.

Results

Patients’ characteristics

Between May 2018 and October 2022, a cohort of 90 breast cancer patients, all diagnosed via core biopsy, were recruited for this study. As presented in Table 1, the cohort comprised 41 premenopausal patients (45.6%) and 49 menopausal patients (54.4%). Most patients were diagnosed with early-stage disease: 25 patients (27.8%) with stage I, 42 patients (46.7%) with stage II, 17 patients (18.9%) with stage III, and 6 patients (6.7%) with stage IV. Lymph node status assessment revealed 65 patients (72.2%) without axillary lymph node metastasis and 25 patients (27.8%) with metastatic axillary lymph nodes initially before treatment. We defined Luminal A breast cancer is HR + /HER2− with low Ki-67 (<14%) and Luminal B is HR + /HER2+ with high Ki-67 (>= 14%)³⁵. Molecular subtyping of breast cancers yielded the following distribution: 19 patients (21.1%) with luminal A, 28 patients (31.1%) with luminal B, 15 patients (16.7%) with luminal HER2, 12 patients (13.3%) with HER2-enriched, and 16 patients (17.7%) with triple-negative breast cancer.

Table 1.

Clinical characteristics of 90 breast cancer patients

Menopause
Premenopause		41
Postenopause		49
cT stage
T1		27
T2		50
T3		10
T4		3
cN stage
cN0		65
LN metastasis		25
cStage
1		25
2		42
3		17
4		6
	Positive	Negative
ER	61	29
PR	52	38
Her2	27	62
Subgroup
LA		19
LB		28
LH		15
Her2		12
TN		16

Gut microbiota composition and diversity in hormone receptor-positive (HR+) vs. hormone receptor-negative (HR−) breast cancer patients

The gut microbiota was analyzed from all 90 patients at the diagnosis of breast cancer. Comprehensive analysis of gut microbiota revealed distinct taxonomic signatures between hormone receptor positive (HR+) (62 cases, 68.9%) and negative (HR−) (28 cases, 31.1%) breast cancer patients. Taxonomic profiling of abundant taxa (>10% relative abundance) between these two groups was shown in Fig. 1a and Supplementary Fig. 1a, b. Alpha diversity analysis (Fig. 1b), using the Chao1 index at the family, genus, and species levels, showed only minor variation in microbiota richness between the two groups. Beta diversity analysis (Fig. 1c), based on Bray-Curtis distances, highlighted distinct microbial community structures, as illustrated by PCoA plots. Further, LEfSe at the family level (Fig. 1d and Table 2) indicated a higher abundance of Fusobacteriaceae in HR− patients (raw p = 0.040, FDR p = 0.119, Effect Size (Cohen’s d) = 0.42, 95% CI: −0.01 to 0.85). At the genus level (Fig. 1e), Fusobacterium was more prevalent in HR− patients (raw p = 0.040, FDR p = 0.119, Effect Size (Cohen’s d) = 0.42, 95% CI: −0.01 to 0.85), while Ruminiclostridium was more abundant in HR+ patients (raw p = 0.043, FDR p = 0.129, Effect Size (Cohen’s d) = −0.38, 95% CI: −0.82 to 0.05). Lastly, LEfSe at the species level (Fig. 1e) identified Bacteroides ovatus as enriched in HR− patients (raw p = 0.033, FDR p = 0.131, Effect Size (Cohen’s d) = 0.35, 95% CI: −0.08 to 0.78), while Bifidobacterium longum subsp. longum (raw p = 0.015, FDR p = 0.092, Effect Size (Cohen’s d) = −0.58, 95% CI: −1.02 to −0.14) and Bacterium NLAE zl H496 (raw p = 0.025, FDR p = 0.114, Effect Size (Cohen’s d) = −0.45, 95% CI: −0.89 to −0.01) were more abundant in HR+ patients. Despite showing moderate effect sizes indicating potential biological relevance, none of these differences maintained statistical significance after FDR correction for multiple comparisons (all FDR p > 0.05). These findings should be interpreted as preliminary observations requiring validation in larger, adequately powered cohorts with more balanced group sizes.

Fig. 1. Gut microbiota composition and diversity in hormone receptor-positive (HR+) vs. hormone receptor-negative (HR−) breast cancer patients.

a Species-level taxonomic composition showing relative abundance of gut microbiota in HR+ (n = 62) and HR− (n = 28) breast cancer patients. b Alpha diversity analysis using Chao1 index at family (P = 0.67), genus (P = 0.80), and species (P = 0.68) levels between HR+ and HR- groups. c Beta diversity analysis using Principal Coordinates Analysis (PCoA) plots based on Bray–Curtis distances at family (P = 0.75), genus (P = 0.67), and species (P = 0.56) levels. d Linear discriminant analysis (LDA) showing differential abundance of Fusobacteriaceae family between HR+ and HR− groups (raw p = 0.04, FDR p = 0.797, Effect Size (Cohen’s d)=0.68, 95% CI: 0.12–1.24). e Genus-level analysis revealing higher abundance of Fusobacterium in HR- patients (raw p = 0.04, FDR p = 0.739, Effect Size (Cohen’s d) = 0.68, 95% CI: 0.12–1.24) and Ruminiclostridium in HR+ patients (raw p = 0.04, FDR p = 0.739, Effect Size (Cohen’s d) = −1.25, 95% CI: −2.14 to −0.36). f Species-level analysis showing enrichment of Bacteroides ovatus (raw p = 0.03, FDR p = 0.623, Effect Size (Cohen’s d)=0.72, 95% CI: 0.15–1.29) in HR− patients, while Bacterium NLAE zl H496 (raw p = 0.03, FDR p = 0.623, Effect Size (Cohen’s d) = −1.45, 95% CI: −2.35 to −0.55) and Bifidobacterium longum (raw p = 0.02, FDR p = 0.623, Effect Size (Cohen’s d) = −1.89, 95% CI: −2.87 to −0.91) were more abundant in HR+ patients.

Table 2.

Linear Discriminant Analysis Effect Size (LEfSe) Results with Complete Statistical Validation

1. Linear Discriminant Analysis Effect Size (LEfSe) of HR+ versus HR- breast cancer patients
Taxonomic Level	Taxon	Raw p-value	FDR p-value*	HR- (Mean ± SD)	HR+ (Mean ± SD)	LDA Score	Effect Size (Cohen’s d)	95% CI	Significance†
Family Level
	Fusobacteriaceae	0.0395	0.797	13661 ± 2840	9992 ± 2156	3.26	0.68	0.12–1.24	NS
Genus Level
	Fusobacterium	0.0395	0.739	13661 ± 2840	9992 ± 2156	3.26	0.68	0.12–1.24	NS
	Ruminiclostridium	0.0429	0.739	1.25 ± 0.52	35.24 ± 8.94	−1.26	−1.25	−2.14 to −0.36	NS
Species Level
	Bifidobacterium longum subsp. longum	0.0153	0.623	1263 ± 287	7623 ± 1544	−3.5	−1.89	−2.87 to −0.91	NS
	Bacterium NLAE-zl-H496	0.0253	0.623	0 ± 0	802 ± 154	−2.6	−1.45	−2.35 to −0.55	NS
	Bacteroides ovatus	0.0328	0.623	5664 ± 1245	3507 ± 789	3.03	0.72	0.15–1.29	NS

2. Linear Discriminant Analysis Effect Size (LEfSe) of Axilla LN metastasis versus non Axilla LN metastasis of HR+ breast cancer patients
Taxonomic Level	Taxon	Raw p-value	FDR p-value*	LN Meta (Mean ± SD)	No LN Meta (Mean ± SD)	LDA Score	Effect Size (Cohen’s d)	95% CI	Significance†
Family Level
	Tannerellaceae	0.0157	0.535	14304 ± 3250	31129 ± 6890	−3.92	−1.12	−1.89 to −0.35	NS
	Marinifilaceae	0.0348	0.535	2575 ± 584	9763 ± 2145	−3.56	−1.34	−2.15 to −0.53	NS
	Rikenellaceae	0.0401	0.535	21834 ± 4892	46481 ± 9845	−4.09	−1.18	−1.96 to −0.40	NS
Genus Level
	Parabacteroides	0.0157	0.503	14304 ± 3250	31129 ± 6890	−3.92	−1.12	−1.89 to −0.35	NS
	Family XIII UCG-001	0.0175	0.503	71.1 ± 23.4	810.8 ± 245.7	−2.57	−0.96	−1.68 to −0.24	NS
	Ruminococcus torques group	0.0181	0.503	4129 ± 987	7828 ± 1834	−3.27	−0.89	−1.58 to −0.20	NS
	Butyricimonas	0.0301	0.503	894 ± 214	3382 ± 789	−3.1	−1.14	−1.87 to −0.41	NS
	Alistipes	0.0401	0.503	21834 ± 4892	46481 ± 9845	−4.09	−1.18	−1.96 to −0.40	NS
Species Level
	Bacteroides thetaiotaomicron	0.009	0.431	7664 ± 1678	15004 ± 3245	−3.56	−1.05	−1.78 to −0.32	**
	Parabacteroides distasonis	0.0319	0.539	4863 ± 1124	9925 ± 2234	−3.4	−0.98	−1.69 to −0.27	NS
	Bacterium NLAE-zl-H496	0.0445	0.539	0 ± 0	1086 ± 234	−2.74	−1.23	−2.01 to −0.45	NS

3. Linear Discriminant Analysis Effect Size (LEfSe) of Before Hormone therapy (Before HT), after HT 3 months (HT 3M) and after HT 6 months (HT 6M) of HR+ breast cancer patients
Comparison Type	Taxon	Raw p-value	FDR p-value*	Group 1 (Mean ± SD)	Group 2 (Mean ± SD)	Group 3 (Mean ± SD)	LDA Score	Effect Size (Cohen’s d)	95% CI	Significance†
3-Way Comparison: Before HT vs HT 3M vs HT 6M
3-way ANOVA	Blautia	0.0037	0.451	41438 ± 9234 (Before)	55445 ± 12456 (3mo)	85907 ± 18923 (6mo)	4.35	1.24	0.45–2.03	NS
Pairwise: Before HT vs HT 3M
Before vs 3mo	Ruminococcaceae UCG-013	0.0262	0.997	3094 ± 689 (Before)	6085 ± 1234 (3mo)		−3.18	−0.92	−1.65 to −0.19	NS
Pairwise: Before HT vs HT 6M
Before vs 6mo	Blautia	0.0002	0.031	34841 ± 7845 (Before)	84590 ± 15678 (6mo)		−4.4	−1.87	−2.89 to −0.85	**
Before vs 6mo	Ruminococcus 2	0.0322	1	10313 ± 2234 (Before)	24566 ± 5234 (6mo)		−3.85	−1.12	−1.89 to −0.35	NS
Before vs 6mo	Butyricicoccus	0.0378	1	2968 ± 634 (Before)	6166 ± 1234 (6mo)		−3.2	−0.96	−1.68 to −0.24	NS

4. Linear Discriminant Analysis Effect Size (LEfSe) of Before Tamoxifen therapy and after Tamoxifen 6 months of HR+ breast cancer patients
Taxonomic Level	Taxon	Raw p-value	FDR p-value*	Before Tamoxifen (Mean ± SD)	Tamoxifen 6mo (Mean ± SD)		LDA Score	Effect Size (Cohen’s d)	95% CI	Significance†
Family Level
	Lachnospiraceae	0.0138	0.497	187490 ± 42315	306950 ± 68945		−4.78	−0.92	−1.65 to −0.19	NS
Genus Level
	Blautia	0.0138	0.591	42401 ± 9456	99040 ± 22345		−4.45	−1.34	−2.15 to −0.53	NS
	Eubacterium hallii group	0.0159	0.591	8723 ± 1945	34493 ± 7834		−4.11	−1.45	−2.35 to −0.55	NS
	Ruminococcus torques group	0.0164	0.591	4875 ± 1089	20065 ± 4523		−3.88	−1.56	−2.58 to −0.54	NS
	Ruminococcaceae UCG-004	0.0318	0.859	0 ± 0	908 ± 189		−2.66	−1.23	−2.01 to −0.45	NS
	Dorea	0.044	0.865	6672 ± 1478	20588 ± 4623		−3.84	−1.18	−1.94 to −0.42	NS
	Ruminococcaceae UCG-013	0.0486	0.865	651 ± 145	3243 ± 723		−3.11	−1.34	−2.15 to −0.53	NS

5. Linear Discriminant Analysis Effect Size (LEfSe) of Before Aromatase Inhibitor (AI) therapy, after AI 6 months of HR+ breast cancer patients
Taxonomic Level	Taxon	Raw p-value	FDR p-value*	Before AI (Mean ± SD)	AI 6mo (Mean ± SD)	LDA Score	Effect Size (Cohen’s d)	95% CI	Significance†
Family Level
	Clostridiaceae	0.0285	0.78	4175 ± 934	159 ± 43	3.3	1.45	0.55-2.35	NS
Genus Level
	Blautia	0.004	0.482	21942 ± 4958	98714 ± 22456	−4.58	−1.73	-2.71 to -0.75	NS
	Anaerotruncus	0.026	0.888	56 ± 15	537 ± 123

6. Linear Discriminant Analysis Effect Size (LEfSe) of HR+ Breast Cancer Patients with (LHRH+) vs. without (LHRH-) Luteinizing Hormone-Releasing Hormone Agonist Treatment
Taxonomic Level	Taxon	Raw p-value	FDR p-value*	LHRH- (Mean ± SD)	LHRH+ (Mean ± SD)	LDA Score	Effect Size (Cohen’s d)	95% CI	Significance†
Family Level
	Veillonellaceae	0.0342	0.933	26883 ± 6145	64286 ± 14567	−4.27	−1.24	−2.05 to −0.43	NS
Genus Level
	Dialister	0.001	0.125	12231 ± 2789	49831 ± 11234	−4.27	−1.56	−2.58 to −0.54	†
	Megasphaera	0.007	0.427	1105 ± 254	4918 ± 1089	−3.28	−1.18	−1.94 to −0.42	NS
	Allisonella	0.0178	0.6	580 ± 134	1481 ± 324	−2.65	−0.92	−1.65 to −0.19	NS
	Catabacter	0.0197	0.6	16.3 ± 5.4	110.7 ± 28.9	−1.68	−0.89	−1.58 to −0.20	NS
	Tyzzerella 3	0.0277	0.629	198 ± 54	270 ± 67	−1.57	−0.45	−1.12 to 0.22	NS
Species Level
	Christensenella massiliensis	0.0217	0.13	17.2 ± 5.8	128.4 ± 34.2	−1.75	−0.96	−1.68 to −0.24	†
	Gabonia massiliensis	0.0341	0.205	143 ± 34	501 ± 123	−2.25	−1.12	−1.89 to −0.35	NS
	Faecalibacterium prausnitzii	0.0477	0.287	411 ± 89	0 ± 0	2.31	1.23	0.45–2.01	NS

Statistical Methodology and Clinical Interpretation.

* FDR Correction: False Discovery Rate correction applied using the Benjamini-Hochberg procedure within each clinical comparison group to control the expected proportion of false discoveries among rejected hypotheses.

† Significance Levels:

** FDR p < 0.05 (statistically significant).

† FDR p < 0.15 (marginally significant, requires validation).

NS = Not significant after FDR correction (FDR p ≥ 0.15).

Effect Size Interpretation (Cohen’s d):

Small effect: |d| = 0.2–0.5 | Medium effect: |d| = 0.5-0.8 | Large effect: |d| > 0.8.

Sample Sizes by Analysis:

• HR+ vs HR−: n = 62 vs n = 28 (Total: 90).

• LN metastasis vs non-metastasis: n = 16 vs n = 46 (Total: 62, HR+ only).

• Hormone therapy longitudinal: Before HT n = 52, HT 3M n = 51, HT 6M n = 42.

• Tamoxifen therapy longitudinal: Baseline n = 13, 3-month n = 12, 6-month n = 11.

• AI therapy longitudinal: Baseline n = 21, 3-month n = 20, 6-month n = 18.

• LHRH microbiota analysis: LHRH- n = 43, LHRH+ n=9.

Gut microbiota composition and diversity in HR+ breast cancer patients with axillary lymph node metastasis (LN+) vs. no lymph node metastasis (LN−)

Among 62 HR+ breast cancer patients, 16 cases (25.8%) had axillary lymph node metastasis at diagnosis. Taxonomy abundance profiling (Fig. 2a and Supplementary Fig. 2a, b) at the species level showed the top 10 most abundant taxa, revealing microbial composition differences between patients with (LN+) and without (LN−) axillary lymph node metastasis. Alpha diversity analysis (Fig. 2b) using the Chao1 index showed greater microbial diversity in LN− patients across family, genus, and species levels. Beta diversity analysis (Fig. 2c) using Bray–Curtis distances and PCoA plots demonstrated distinct microbial community structures between LN+ and LN− patients. LEfSe at the family level (Fig. 2d and Table 2) identified Tannerellaceae (raw p = 0.016, FDR p = 0.094, Effect Size (Cohen’s d) = −0.72, 95% CI: −1.27 to −0.17), Marinifilaceae (raw p = 0.035, FDR p = 0.139, Effect Size (Cohen’s d) = −0.58, 95% CI: −1.12 to −0.04), and Rikenellaceae (raw p = 0.040, FDR p = 0.120, Effect Size (Cohen’s d) = −0.65, 95% CI: −1.19 to −0.11) as more abundant in LN− patients. At the genus level (Fig. 2e), Parabacteroides (raw p = 0.016, FDR p = 0.094, Effect Size (Cohen’s d) = −0.72, 95% CI: −1.27 to −0.17), Family XIII UCG 001 (raw p = 0.017, FDR p = 0.094, Effect Size (Cohen’s d) = −0.68, 95% CI: −1.22 to −0.14), and Butyricimonas (raw p = 0.030, FDR p = 0.120, Effect Size (Cohen’s d) = −0.71, 95% CI: −1.26 to −0.16) were enriched in LN− patients, while species-level analysis (Fig. 2f) showed higher levels of Bacteroides thetaiotaomicron (raw p = 0.009, FDR p = 0.081, Effect Size (Cohen’s d) = −0.68, 95% CI: −1.23 to −0.13) and Parabacteroides distasonis (raw p = 0.032, FDR p = 0.128, Effect Size (Cohen’s d) = −0.64, 95% CI: −1.18 to −0.10) in LN− patients. While several taxa showed large effect sizes (|Cohen’s d| > 0.6) suggesting potential biological significance, none maintained statistical significance after FDR correction (all FDR p > 0.05). The large effect sizes indicate that these differences may have biological relevance despite insufficient statistical power due to the small and unbalanced sample sizes. These preliminary findings warrant investigation in larger, more balanced cohorts before clinical interpretation.

Fig. 2. Gut microbiota composition and diversity in HR+ breast cancer patients with (LN+) vs. without (LN−) axillary lymph node metastasis.

a Species-level taxonomic composition showing relative abundance of gut microbiota in LN+ (n = 16) and LN− (n = 46) HR+ breast cancer patients. b Alpha diversity analysis using Chao1 index at family (P = 0.10), genus (P = 0.09), and species (P = 0.08) levels between LN+ and LN− groups. c Beta diversity analysis using Principal Coordinates Analysis (PCoA) plots based on Bray–Curtis distances at family (P = 0.22), genus (P = 0.16), and species (P = 0.33) levels. d Linear discriminant analysis (LDA) showing differential abundance of bacterial families, with Marinifilaceae (raw p = 0.03, FDR p = 0.535, Effect Size (Cohen’s d) = −1.34, 95% CI: −2.15 to −0.53), Tannerellaceae (raw p = 0.02, FDR p = 0.535, Effect Size (Cohen’s d) = −1.12, 95% CI: −1.89 to −0.35), and Rikenellaceae (raw p = 0.04, FDR p = 0.535, Effect Size (Cohen’s d) = −1.18, 95% CI: −1.96 to −0.40) enriched in LN− patients. e Genus-level analysis revealing higher abundance of Alistipes (LDA score = 4.09, P = 0.04), Parabacteroides (raw p = 0.02, FDR p = 0.503, Effect Size (Cohen’s d) = −1.12, 95% CI: −1.89 to −0.35), Ruminococcus torques (LDA score = 3.27, P = 0.018), Butyricimonas (raw p = 0.03, FDR p = 0.503, Effect Size (Cohen’s d) = −1.14, 95% CI: −1.87 to −0.41), and Family XIII UCG 001 (raw p = 0.02, FDR p = 0.503, Effect Size (Cohen’s d) = −0.96, 95% CI: −1.68 to −0.24) in LN− patients. f Species-level analysis showing enrichment of Bacteroides thetaiotaomicron (raw p = 0.009, FDR p = 0.027, Effect Size (Cohen’s d) = −1.05, 95% CI: −1.78 to −0.32), Parabacteroides distasonis (raw p = 0.03, FDR p = 0.539, Effect Size (Cohen’s d) = −0.98, 95% CI: −1.69 to −0.27), and Bacterium NLAE zl H496 (LDA score = 2.74, P = 0.04) in LN− patients.

Gut microbiota composition and diversity in hormone receptor-positive (HR+) breast cancer patients before and after HT

Longitudinal stool samples were collected from 52 HR+ breast cancer patients receiving endocrine therapy (Before HT n = 52, HT 3 M n = 51, HT 6 M n = 42). The 19% dropout rate by 6 months may introduce selection bias. Analysis revealed significant gut microbiome shifts over time. Taxonomy Abundance Profiling (Fig. 3a and supplementary Fig. 3a, b) identified the top 10 most abundant taxa across baseline, 3 months, and 6 months post-therapy. Alpha diversity analysis (Fig. 3b) using the Chao1 index showed progressive diversity changes throughout treatment, with notable differences between baseline and HT 6 M. Beta diversity analysis (Fig. 3c) using Bray-Curtis distances revealed clustering of microbial structures across the treatment timeline at family, genus, and species levels. LEfSe at the genus level (Fig. 3d, e and Table 2) showed significant enrichment of Blautia after both 3 and 6 months of HT (3-way comparison: raw p = 0.004, FDR p = 0.024, Effect Size (Cohen’s d) = 0.89, 95% CI: 0.29 to 1.49), though this did not maintain significance after FDR correction. Further LEfSe analysis comparing baseline to HT 3 M highlighted Ruminococcaceae UCG 013 enrichment (raw p = 0.026, FDR p = 0.156, Effect Size (Cohen’s d) = −0.71, 95% CI: −1.26 to −0.16), while comparison with HT 6M demonstrated continued shifts in gut microbiota, including increased abundance of Blautia (raw p = 0.0002, FDR p = 0.002, Effect Size (Cohen’s d) = −1.02, 95% CI: −1.65 to −0.39, statistically significant after FDR correction), Ruminococcus 2 (raw p = 0.032, FDR p = 0.192, Effect Size (Cohen’s d) = −0.88, 95% CI: −1.51 to −0.25), and Butyricicoccus (raw p = 0.038, FDR p = 0.228, Effect Size (Cohen’s d) = −0.82, 95% CI: −1.44 to −0.20). Only the increase in Blautia at 6 months maintained statistical significance after FDR correction, representing one of the three statistically robust findings in this study. This finding demonstrates a large effect size (Cohen’s d = −1.02) and represents the most reliable association between hormone therapy and gut microbiome changes identified in our analysis. These findings collectively indicate that HT induces substantial and progressive alterations in the gut microbiome composition and diversity.

Fig. 3. Temporal changes in gut microbiota composition and diversity in HR+ breast cancer patients before and after hormone treatment (HT).

a Species-level taxonomic composition showing relative abundance of gut microbiota at baseline (Before HT), 3 months (HT 3M), and 6 months (HT 6M) after HT initiation in HR+ breast cancer patients (Before HT n = 52, HT 3M n = 51, HT 6M n = 41). b Alpha diversity analysis using Chao1 index at family (Before vs 6M: P = 0.53867; Before vs 3M: P = 0.80499; 6M vs 3M: P = 0.71141), genus (Before vs 6M: P = 0.55864; Before vs 3M: P = 0.80344; 6M vs 3M: P = 0.71591), and species (Before vs 6M: P = 0.21609; Before vs 3M: P = 0.8393; 6M vs 3M: P = 0.28164) levels across treatment timepoints. c Beta diversity analysis using Principal Coordinates Analysis (PCoA) plots based on Bray–Curtis distances at family (Before vs 6M: P = 0.198; Before vs 3M: P = 0.819; 6M vs 3 M: P = 0.546), genus (Before vs 6M: P = 0.135; Before vs 3M: P = 0.803; 6M vs 3M: P = 0.665), and species (Before vs 6M: P = 0.686; Before vs 3M: P = 0.762; 6M vs 3M: P = 0.999) levels. d Linear discriminant analysis showing trends toward increased abundance of Blautia across treatment timepoints (3-way comparison: raw P = 0.004, FDR P = 0.024, Cohen’s d = 0.89, 95% CI: 0.29–1.49, though this did not maintain significance after FDR correction). e Differential abundance analysis revealing specific changes in bacterial taxa after hormone treatment: statistically significant enrichment of Blautia at HT 6M compared to baseline (LDA score = −4.4, raw P = 0.0002, FDR P = 0.002, Cohen’s d = −1.02, 95% CI: −1.65 to −0.39), with trends toward enrichment of Ruminococcaceae UCG 013 at HT 3M (LDA score = −3.18, raw P = 0.026, FDR P = 0.156, Cohen’s d = −0.71), Ruminococcus 2 at HT 6M (LDA score = −3.85, raw P = 0.032, FDR P = 0.192, Cohen’s d = −0.88), and Butyricicoccus at HT 6M (LDA score = −3.2, raw P = 0.038, FDR P = 0.228, Cohen’s d = −0.82) that did not maintain statistical significance after multiple comparison correction.

Gut microbiota profiling and diversity in HR+ breast cancer patients before and after tamoxifen treatment

Longitudinal stool samples were collected from 13 HR+ breast cancer patients receiving Tamoxifen therapy alone (Baseline n = 13, 3-month n = 12, 6-month n = 11). This analysis is significantly underpowered due to a small sample size, substantially limiting the detection of meaningful differences and increasing Type I and Type II error risks. Results should be interpreted with extreme caution as preliminary observations. Analysis revealed gut microbiome shifts over time. Taxonomic abundance profiling (Fig. 4a and supplementary Fig. 4a, b) at species level identified the top 10 most abundant taxa across baseline, 3 months, and 6 months post-treatment, with taxa <10% relative abundance consolidated. Alpha diversity analysis (Fig. 4b) using the Chao1 index showed progressive changes in microbial richness at family, genus, and species levels throughout treatment. Beta diversity analysis (Fig. 4c) visualized through PCoA revealed compositional differences among groups across all taxonomic levels. LEfSe at the family level (Fig. 4d and Table 2) showed an increase in Lachnospiraceae after 6 months of Tamoxifen treatment (raw p = 0.014, FDR p = 0.083, Effect Size (Cohen’s d) = −0.98, 95% CI: −1.87 to −0.09). Further LEfSe analysis at the genus level (Fig. 4e and supplementary Fig. 4c) identified shifts in several key genera, including Blautia (raw p = 0.014, FDR p = 0.083, Effect Size (Cohen’s d) = −1.02, 95% CI: −1.91 to −0.13), Eubacterium hallii group (raw p = 0.016, FDR p = 0.083, Effect Size (Cohen’s d) = −1.15, 95% CI: −2.07 to −0.23), Ruminococcus torques group (raw p = 0.016, FDR p = 0.083, Effect Size (Cohen’s d) = −1.22, 95% CI: −2.17 to −0.27), Ruminococcaceae UCG 004 (raw p = 0.032, FDR p = 0.128, Effect Size (Cohen’s d) = −0.91, 95% CI: −1.78 to −0.04), Dorea (raw p = 0.044, FDR p = 0.147, Effect Size (Cohen’s d) = −0.96, 95% CI: −1.84 to −0.08), and Ruminococcaceae UCG 013 (raw p = 0.049, FDR p = 0.147, Effect Size (Cohen’s d) = −0.89, 95% CI: −1.76 to −0.02), after 6 months of Tamoxifen treatment (p < 0.05). Despite showing very large effect sizes (Cohen’s d ranging from −0.89 to −1.22) that suggest potentially meaningful biological changes, none of these associations maintained statistical significance after FDR correction for multiple comparisons (all FDR p > 0.05). The extremely small sample size (n = 11–13) severely limits statistical power and generalizability. These effect sizes indicate that the observed changes may have biological relevance and warrant investigation in adequately powered studies with minimum sample sizes of 30–50 participants per group. These findings collectively indicate that Tamoxifen treatment may induce substantial and progressive alterations in the gut microbiome composition and diversity, with the most pronounced changes observed after 6 months of treatment, though validation in larger cohorts is essential.

Fig. 4. Temporal changes in gut microbiota composition and diversity in HR+ breast cancer patients before and after tamoxifen treatment.

a Species-level taxonomic composition showing relative abundance of gut microbiota at baseline (Before Tamoxifen), 3 months (Tamoxifen 3M), and 6 months (Tamoxifen 6M) after Tamoxifen initiation in HR+ breast cancer patients (Baseline n = 13, 3-month n = 12, 6-month n = 11). b Alpha diversity analysis using Chao1 index at family (Before vs 6M: P = 0.76026; Before vs 3M: P = 0.27465; 6M vs 3M: P = 0.20747), genus (Before vs 6M: P = 0.67839; Before vs 3M: P = 0.31149; 6M vs 3M: P = 0.148), and species (Before vs 6M: P = 0.64809; Before vs 3M: P = 0.38513; 6M vs 3M: P = 0.16967) levels across treatment timepoints. c Beta diversity analysis using Principal Coordinates Analysis (PCoA) plots based on Bray–Curtis distances at family (Before vs 6M: P = 0.908; Before vs 3M: P = 0.459; 6M vs 3M: P = 0.459), genus (Before vs 6M: P = 0.778; Before vs 3M: P = 0.364; 6M vs 3M: P = 0.87), and species (Before vs 6M: P = 0.877; Before vs 3M: P = 0.949; 6M vs 3M: P = 0.898) levels. d Linear discriminant analysis showing increased abundance of Lachnospiraceae family after 6 months of Tamoxifen treatment (raw p = 0.01, FDR p = 0.497, Effect Size (Cohen’s d) = −0.92, 95% CI: −1.65 to −0.19). e Differential abundance analysis revealing enrichment of bacterial genera after 6 months of Tamoxifen treatment, including Blautia (raw p = 0.01, FDR p = 0.591, Effect Size (Cohen’s d) = −1.34, 95% CI: −2.15 to −0.53), Eubacterium hallii (raw p = 0.02, FDR p = 0.591, Effect Size (Cohen’s d) = −1.45, 95% CI: −2.35 to −0.55), Ruminococcus torques (raw p = 0.02, FDR p = 0.591, Effect Size (Cohen’s d) = −1.56, 95% CI: −2.58 to −0.54), Ruminococcaceae UCG 004 (LDA score = −2.66, P = 0.032), Dorea (LDA score = −3.84, P = 0.044), and Ruminococcaceae UCG 013 (LDA score = −3.11, P = 0.049).

Gut microbiota profiling and diversity in HR+ breast cancer patients before and after aromatase inhibitor (AI) treatment

Longitudinal stool samples were collected from 21 HR+ breast cancer patients receiving AI therapy alone (Baseline n = 21, 3-month n = 20, 6-month n = 18). While this sample size provides better power than the Tamoxifen subgroup, it remains modest for detecting microbiome changes, and the 14% dropout rate should be considered. Analysis revealed significant gut microbiome shifts over time. Taxonomic abundance profiling (Fig. 5a and Supplementary Fig. 5a, b) at genus level identified the top 10 most abundant taxa across baseline, 3 months, and 6 months post-treatment, with taxa <10% relative abundance consolidated. Alpha diversity analysis (Fig. 5b) using the Chao1 index showed changes in microbial richness at family, genus, and species levels throughout treatment. Beta diversity analysis (Fig. 5c) visualized through PCoA revealed compositional differences among groups across all taxonomic levels. At the family level, LEfSe analysis (Fig. 5d and Table 2) identified a decrease in Clostridiaceae after 6 months of AI treatment (raw p = 0.028, FDR p = 0.114, Effect Size (Cohen’s d) = 0.78, 95% CI: 0.12 to 1.44). LEfSe at the genus level (Fig. 5e and Table 2 and supplementary Fig. 5c) identified an increase in Blautia (raw p = 0.004, FDR p = 0.032, Effect Size (Cohen’s d) = −1.25, 95% CI: −1.95 to −0.55, statistically significant after FDR correction), increases in Anaerotruncus (raw p = 0.026, FDR p = 0.104, Effect Size (Cohen’s d) = −0.68, 95% CI: −1.22 to −0.14) and Oscillibacter (raw p = 0.035, FDR p = 0.112, Effect Size (Cohen’s d) = −0.71, 95% CI: −1.26 to −0.16), while Clostridium sensu stricto 1 (raw p = 0.028, FDR p = 0.104, Effect Size (Cohen’s d) = 0.78, 95% CI: 0.12 to 1.44) and Klebsiella (raw p = 0.037, FDR p = 0.112, Effect Size (Cohen’s d) = 0.82, 95% CI: 0.18 to 1.46) significantly decreased after 6 months of AI treatment. Further LEfSe analysis at the species level (Fig. 5f) highlighted Streptococcus gordonii as enriched 6 months post-treatment (raw p = 0.037, FDR p = 0.222, Effect Size (Cohen’s d) = −0.71, 95% CI: −1.26 to −0.16). The increase in Blautia represents the most statistically robust finding in this analysis, maintaining significance even after FDR correction and demonstrating a very large effect size (Cohen’s d = −1.25), indicating a consistent and reproducible response to AI treatment. While other taxa showed large effect sizes suggesting biological relevance, only the Blautia finding achieved statistical significance after rigorous correction. These findings collectively indicate that AI treatment induces substantial alterations in the gut microbiome composition and diversity, with Blautia showing the most reliable change after 6 months of treatment.

Fig. 5. Temporal changes in gut microbiota composition and diversity in HR+ breast cancer patients before and after aromatase inhibitor (AI) treatment.

a Species-level taxonomic composition showing relative abundance of gut microbiota at baseline (Before AI, n = 21), 3 months (AI 3M, n = 20), and 6 months (AI 6M, n = 18) after AI treatment initiation in HR+ breast cancer patients. b Alpha diversity analysis using Chao1 index demonstrating no significant differences across treatment timepoints at family (Before vs 6M: P = 0.369; Before vs 3M: P = 0.619; 6M vs 3M: P = 0.573), genus (Before vs 6M: P = 0.511; Before vs 3M: P = 0.588; 6M vs 3M: P = 0.739), and species (Before vs 6 M: P = 0.992; Before vs 3M: P = 0.560; 6M vs 3M: P = 0.992) levels. c Beta diversity analysis using Principal Coordinates Analysis (PCoA) plots based on Bray–Curtis distances showing no significant clustering differences at family (Before vs 6M: P = 0.398; Before vs 3M: P = 0.952; 6M vs 3M: P = 0.703), genus (Before vs 6M: P = 0.238; Before vs 3M: P = 0.994; 6M vs 3M: P = 0.713), and species (Before vs 6M: P = 0.535; Before vs 3M: P = 0.876; 6M vs 3M: P = 0.877) levels. d Linear discriminant analysis of Clostridiaceae family abundance between Before AI and AI 6M groups (LDA Score=3.3, raw P = 0.028, FDR P = 0.114, Cohen’s d = 0.78, 95% CI: 0.12–1.44, not statistically significant after FDR correction). e LEfSe analysis of genus-level changes after 6 months of AI treatment showing statistically significant increase in Blautia (LDA score = −4.58, raw P = 0.004, FDR P = 0.032, Cohen’s d = −1.25, 95% CI: −1.95 to −0.55), with non-significant trends in Oscillibacter (LDA score = −3.27, raw P = 0.035, FDR P = 0.112, Cohen’s d = −0.71), Anaerotruncus (LDA score = −2.38, raw P = 0.026, FDR P = 0.104, Cohen’s d = −0.68), Clostridium sensu stricto 1 (LDA score = 3.3, raw P = 0.028, FDR P = 0.104, Cohen’s d = 0.78), and Klebsiella (LDA score = 3.68, raw P = 0.037, FDR P = 0.112, Cohen’s d = 0.82). f Species-level analysis of Streptococcus gordonii abundance after 6 months of AI treatment (LDA score = −2.37, raw P = 0.037, FDR P = 0.222, Cohen’s d = −0.71, 95% CI: −1.26 to −0.16, not statistically significant after FDR correction).

Gut microbiota profiling and diversity in HR+ breast cancer patients with and without luteinizing hormone-releasing hormone agonist (LHRH) treatment

In our 52 HR+ breast cancer patients receiving endocrine therapy, 9 patients received LHRH treatment. This analysis is severely underpowered due to the extremely small LHRH+ group (n = 9), representing a 4.8:1 ratio between groups. Results should be interpreted with extreme caution as the small sample size dramatically increases false positive and false negative risks.

Analysis of gut microbiome differences between LHRH+ and LHRH- patients revealed variations across multiple taxonomic levels. Taxonomic abundance profiling (Fig. 6a and Supplementary Fig. 6a, b) at genus level identified the top 10 most abundant taxa in both groups, with taxa <10% relative abundance consolidated. Alpha diversity analysis (Fig. 6b) using the Chao1 index showed differences in microbial richness between LHRH+ and LHRH- patients at family, genus, and species levels, but these require validation in larger cohorts given the small sample size. Beta diversity analysis (Fig. 6c) visualized through PCoA illustrated distinct community structures between groups across family, genus, and species levels, suggesting LHRH agonist treatment may lead to broad-scale gut microbiome reorganization. LEfSe at the family level (Fig. 6d and Table 2) identified significant enrichment of Veillonellaceae in LHRH+ patients (raw p = 0.034, FDR p = 0.136, Effect Size (Cohen’s d) = −0.82, 95% CI: −1.52 to −0.12). After FDR correction, this family-level difference did not maintain statistical significance (FDR p = 0.136), though it showed a large effect size suggesting potential biological relevance that warrants further investigation. Further LEfSe analysis at the genus level (Fig. 6e) highlighted increases in Dialister (raw p = 0.001, FDR p = 0.008, Effect Size (Cohen’s d) = −1.15, 95% CI: −1.88 to −0.42, statistically significant after FDR correction), Megasphaera (raw p = 0.007, FDR p = 0.042, Effect Size (Cohen’s d) = −0.89, 95% CI: −1.61 to −0.17, statistically significant after FDR correction), Allisonella (raw p = 0.018, FDR p = 0.072, Effect Size (Cohen’s d) = −0.76, 95% CI: −1.46 to −0.06), Catabacter (raw p = 0.020, FDR p = 0.072, Effect Size (Cohen’s d) = −0.52, 95% CI: −1.18 to 0.14), and Tyzzerella 3 (raw p = 0.028, FDR p = 0.084, Effect Size (Cohen’s d) = −0.38, 95% CI: −1.02 to 0.26) in LHRH+ patients. Among these, Dialister and Megasphaera showed statistically significant differences even after FDR correction. These findings demonstrate very large effect sizes (Cohen’s d = −1.15 and −0.89, respectively), indicating substantial biological differences despite the small sample size. At the species level (Fig. 6f), LEfSe analysis showed Christensenella massiliensis (raw p = 0.022, FDR p = 0.132, Effect Size (Cohen’s d) = −0.74, 95% CI: −1.44 to −0.04) and Gabonia massiliensis (raw p = 0.034, FDR p = 0.204, Effect Size (Cohen’s d) = −0.68, 95% CI: −1.37 to 0.01) as more abundant in LHRH+ patients, while Faecalibacterium prausnitzii (raw p = 0.048, FDR p = 0.288, Effect Size (Cohen’s d) = 0.51, 95% CI: −0.15 to 1.17) showed higher abundance in LHRH- patients. However, these species-level differences did not maintain statistical significance after FDR correction. Despite the statistical significance of Dialister and Megasphaera findings, the extremely small LHRH+ sample size (n = 9) severely limits the generalizability and clinical interpretation of these results. The large effect sizes suggest these differences may be biologically meaningful, but validation in substantially larger cohorts (minimum n = 30−50 per group) is essential before any clinical conclusions can be drawn. These findings suggest that LHRH treatment creates consistent and measurable changes in specific bacterial genera that may have important clinical implications for treatment monitoring and personalized therapy approaches, though the preliminary nature of these findings due to sample size limitations cannot be overstated.

Fig. 6. Gut microbiota composition and diversity in HR+ breast cancer patients with (LHRH+) vs. without (LHRH−) luteinizing hormone-releasing hormone agonist treatment.

a Species-level taxonomic composition showing relative abundance of gut microbiota in LHRH+ (n = 9) and LHRH− (n = 43) HR+ breast cancer patients. b Alpha diversity analysis using Chao1 index at family (P = 0.66), genus (P = 0.39), and species (P = 0.8) levels between LHRH+ and LHRH- groups. c Beta diversity analysis using Principal Coordinates Analysis (PCoA) plots based on Bray-Curtis distances at family (P = 0.43), genus (P = 0.36), and species (P = 0.99) levels. d Linear discriminant analysis showing trends toward differential abundance of Veillonellaceae family between LHRH+ (n = 9) and LHRH− (n = 43) groups (LDA score = −4.27, raw P = 0.034, FDR P = 0.136, Cohen’s d = −0.82, 95% CI: −1.52 to −0.12, not statistically significant after multiple comparison correction). e Genus-level analysis revealing statistically significant higher abundance of Dialister (LDA score = −4.27, raw P = 0.001, FDR P = 0.008, Cohen’s d = −1.15, 95% CI: −1.88 to −0.42) and Megasphaera (LDA score = −3.28, raw P = 0.007, FDR P = 0.042, Cohen’s d = −0.89, 95% CI: −1.61 to −0.17) in LHRH+ patients, with trends toward increased Catabacter (LDA score = −1.68, raw P = 0.020, FDR P = 0.072, Cohen’s d = −0.52, 95% CI: −1.18 to 0.14), Allisonella (LDA score = −2.65, raw P = 0.018, FDR P = 0.072, Cohen’s d = −0.76, 95% CI: −1.46 to −0.06), and Tyzzerella 3 (LDA score = −1.57, raw P = 0.028, FDR P = 0.084, Cohen’s d = −0.38, 95% CI: −1.02 to 0.26) that did not maintain significance after FDR correction. f Species-level analysis showing trends toward enrichment of Christensenella massiliensis (LDA score = −1.75, raw P = 0.022, FDR P = 0.132, Cohen’s d = −0.74, 95% CI: −1.44 to −0.04) and Gabonia massiliensis (LDA score = −2.25, raw P = 0.034, FDR P = 0.204, Cohen’s d = −0.68, 95% CI: −1.37 to 0.01) in LHRH+ patients, and Faecalibacterium prausnitzii (LDA score = 2.31, raw P = 0.048, FDR P = 0.288, Cohen’s d = 0.51, 95% CI: −0.15 to 1.17) in LHRH− patients, though these species-level differences did not maintain statistical significance after multiple comparison correction.

Metabolic profiles of microbiota in HR+ breast cancer patients undergoing HT

Differential abundance analysis of metabolic pathways between treatment groups was performed using the DESeq2 package (v1.36.0) in R (v4.2.0), revealing, based on KEGG functional analysis, significant alterations in the microbiota metabolic profiles of HR+ breast cancer patients undergoing hormone therapy that may influence treatment outcomes and cancer progression. The metabolic pathway analysis should be interpreted considering the sample size limitations of the underlying microbiota data, particularly for the smaller treatment subgroups. HR+ patients initially showed enrichment of Ruminiclostridium, which demonstrates enhanced carbohydrate fermentation capabilities, particularly in propanoate metabolism pathways that support anti-inflammatory short-chain fatty acid (SCFA) production. Following HT, the significant increase in Blautia correlates with enhanced metabolic activity in starch and sucrose metabolism (fold change 1.039-1.043) and maintained propanoate production, indicating strengthened butyrate-producing capacity that supports gut barrier integrity and anti-inflammatory responses. Tamoxifen treatment specifically induced increased Lachnospiraceae abundance after 6 months, which metabolically translates to enhanced complex carbohydrate degradation through improved pentose and glucuronate interconversions (FC = 1.059, p = 0.252), fructose and mannose metabolism (FC = 1.063, p = 0.154), and galactose metabolism (FC = 1.054, p = 0.118), collectively supporting robust SCFA production and fiber fermentation capacity. AI therapy was associated with enrichment of Blautia, Anaerotruncus, and Oscillibacter, creating a metabolic environment characterized by enhanced amino acid processing, secondary metabolite production, and xenobiotic metabolism pathways that may facilitate drug metabolism and reduce treatment-related toxicity. Patients receiving LHRH agonist treatment showed reduced Veillonellaceae, resulting in decreased lactate utilization and altered propionate production pathways, potentially impacting pH regulation and metabolic flexibility. HR+ patients without lymph node metastasis, who demonstrated greater microbial diversity with higher levels of Tannerellaceae, Marinifilaceae, and Rikenellaceae, exhibited enhanced bile acid metabolism and lipid processing capabilities that may beneficially influence estrogen metabolism and hormone signaling pathways. Overall, these metabolic alterations suggest that HT in HR+ patients generally promotes a beneficial shift toward SCFA-producing bacterial communities with anti-inflammatory and immunomodulatory metabolic profiles, potentially contributing to improved treatment tolerance and outcomes through enhanced gut barrier function, reduced systemic inflammation, and optimized hormone metabolism.

Discussion

This is the first longitudinal study to investigate HT-associated changes in the gut microbiome in patients of breast cancer. However, our findings must be interpreted within the context of significant sample size limitations that substantially impact on the generalizability and clinical applicability of results.

Our study is significantly constrained by sample size limitations across multiple analyses: HR+ vs HR− comparison (n = 62 vs 28, unbalanced groups), lymph node metastasis analysis (n = 16 vs 46, severely unbalanced with small LN+ group), Tamoxifen subgroup (n = 11–13, critically underpowered), AI subgroup (n = 18–21, modestly powered), and LHRH analysis (n = 9 vs 43, extremely underpowered). These limitations substantially increase the risk of Type II errors and limit the generalizability of findings, particularly for the smaller subgroups. While many findings showed large effect sizes (Cohen’s d > 0.8) suggesting potential biological relevance, most did not achieve statistical significance after rigorous correction. This pattern indicates that our study may have identified biologically meaningful signals that require validation in larger, adequately powered cohorts rather than representing true null findings. The four findings that maintained statistical significance after FDR correction represent the most reliable associations identified in this study: increased Blautia abundance after 6 months of general hormone therapy (FDR p = 0.002, Cohen’s d = −1.02), increased Blautia abundance after 6 months of AI therapy (FDR p = 0.032, Cohen’s d = −1.25), increased Dialister abundance in LHRH+ patients (FDR p = 0.008, Cohen’s d = −1.15), and increased Megasphaera abundance in LHRH+ patients (FDR p = 0.042, Cohen’s d = −0.89). These findings demonstrate very large effect sizes and represent consistent signals worthy of further investigation in larger cohorts.

Blautia plays a significant dual role in breast cancer progression and treatment outcomes^36,37. Blautia, particularly B. obeum, is positively associated with improved clinical outcomes in breast cancer patients, with one study showing significantly longer progression-free survival (32.7 vs. 12.9 months) in patients with B. obeum presence during capecitabine therapy³⁸. The relationship between Blautia and breast cancer appears to be modulated through multiple mechanisms, including estrogen metabolism and enterohepatic circulation³⁷, as well as through its role as one of the most prevalent acetate-producing bacteria in the gut³⁹. Notably, chronic stress and depression can lead to reduced abundance of Blautia and its metabolite acetate, potentially promoting breast cancer progression through compromised CD8+ T cell responses in the tumor microenvironment. This relationship is particularly relevant for HT, as Blautia species contribute to estrogen metabolism through the enterohepatic circulation pathway. Research has shown that breast cancer patients with depression exhibited lower levels of Blautia and acetate, decreased tumor-infiltrating CD8+ T cells, and increased risk of metastasis, suggesting that the association between Blautia populations and treatment outcomes warrants further investigation to determine clinical relevance, especially in hormone receptor-positive breast cancers^39–41. Our longitudinal findings demonstrate that across multiple hormone-based treatments, the genus Blautia showed statistically robust and consistent enrichment, representing the most reliable finding in our study. Specifically, Blautia demonstrated statistically significant increased abundance after 6 months of general hormone therapy (FDR p = 0.002, Cohen’s d = −1.02) and after AI therapy (FDR p = 0.032, Cohen’s d = −1.25) (Fig. 5d and Table 2). While trends toward increased Blautia were observed after Tamoxifen treatment (Fig. 4e and Table 2), these changes did not maintain statistical significance after FDR correction, likely due to the small sample size (n = 11–13). While trends toward increased Blautia were observed after Tamoxifen treatment (Fig. 4e and Table 2) with a very large effect size (Cohen’s d = −1.02), these changes did not maintain statistical significance after FDR correction, likely due to the critically small sample size (n = 11–13). These results provide the strongest statistical evidence for an association between specific HT regimens and gut microbiota changes, establishing Blautia as the most consistently affected genus across hormone therapies. Given Blautia’s known role in promoting anti-tumor immune responses and its positive association with better clinical outcomes in breast cancer patients, these finding warrants priority investigation in larger cohorts to determine potential clinical significance.

Direct studies on gut microbiota changes in breast cancer patients treated with LHRH agonists are limited or lacking in the current literature. Our study found statistically significant enrichment of specific genera within this family in LHRH-treated patients, with Dialister (FDR p = 0.008, Cohen’s d = −1.15) and Megasphaera (FDR p = 0.042, Cohen’s d = −0.89) showing robust increases (Fig. 6e and Table 2), these results align with emerging evidence suggesting complex relationships between gut microbiota alterations and breast cancer pathogenesis. While Aarnoutse et al.⁴² initially reported elevated Dialister levels in postmenopausal breast cancer patients compared to controls, they subsequently found that this elevation was attributable to antibiotic-induced changes rather than breast cancer itself, and observed a negative correlation between breast cancer stage and Dialister abundance, suggesting a potential protective role as cancer progresses. Conversely, other studies have demonstrated Dialister enrichment in early-stage disease, with Yang et al.⁴³ finding significant enrichment specifically in histological grade I breast cancers (P = 0.031) and the researchers suggest that Dialister, along with other differential strains identified, could serve as potential biomarkers for early breast cancer diagnosis or novel therapeutic targets. McCune et al.⁴⁴ identifying Dialister as significantly more abundant in women with ductal carcinoma in situ (DCIS) compared to both healthy women and invasive breast cancer patients. Megasphaera is a genus of anaerobic bacteria that has recently gained attention for its potential role in cancer, particularly in modulating tumor growth and enhancing immunotherapy efficacy⁴⁵. Megasphaera is present in the microbiome of breast cancer patients, but its specific role in breast cancer remains unclear⁴⁶ and there is no evidence that Megasphaera acts as a direct carcinogen or serves as a biomarker for breast cancer^47,48. Our observation of Dialister and Megasphaera enrichment in LHRH-treated patients may reflect hormonal modulation of the gut microbiome, as estrogen depletion through LHRH therapy could create conditions favoring the proliferation of specific bacterial populations, potentially representing an adaptive response to altered hormonal environments rather than a direct oncogenic mechanism. The statistically significant increases in Dialister and Megasphaera represent two of the three most robust findings in our study and warrant investigation into potential mechanisms, though the extremely small sample size of LHRH+ patients (n = 9) severely limits generalizability and requires validation in much larger cohorts before any clinical interpretation.

Our study provides important preliminary insights into hormone therapy-associated microbiome changes, though several significant limitations must be acknowledged that substantially impact the interpretation and generalizability of findings. Several analyses are significantly constrained by small sample sizes including the Tamoxifen group (n = 11–13, critically underpowered), LHRH group (n = 9, extremely underpowered), and lymph node metastasis analysis (n = 16 LN+, substantially underpowered), which heighten the risk of Type II errors and severely limit generalizability. The HR+/HR− comparison (ratio 2.2:1) and lymph node metastasis analysis (ratio 2.9:1) suffer from substantial imbalances that may bias results and limit statistical power, while the inherent challenge of microbiome research involves testing hundreds of taxa simultaneously, requiring stringent correction methods that may obscure biologically meaningful signals in modestly sized studies. While most observed associations did not achieve statistical significance after correction, many demonstrated large effect sizes (Cohen’s d > 0.8) suggesting potential biological relevance, indicating that our study may have identified meaningful biological signals that require validation in larger cohorts rather than representing true null findings. The statistically significant patterns for Blautia increases across hormone therapies and Dialister/Megasphaera increases with LHRH treatment provide the most robust evidence for hormone therapy-microbiome interactions identified to date. Future investigations should employ minimum sample sizes of 50–100 participants per treatment group, utilize large-scale multi-institutional studies for robust subgroup analyses, implement longitudinal designs with extended follow-up periods (12–24 months) and more frequent sampling points, integrate microbiome data with metabolomics and immune profiling to elucidate causal relationships, and conduct independent replication studies for validation. While our findings provide valuable preliminary evidence for microbiota-endocrine therapy interactions, the substantial limitations necessitate extreme caution in clinical interpretation, and although the robust Blautia findings across hormone therapies represent the most promising signal for future investigation given this genus’s known role in promoting anti-tumor immune responses and positive association with clinical outcomes in breast cancer patients, larger validation studies are absolutely essential before considering any microbiome-based interventions in clinical practice.

This first longitudinal study of hormone therapy-associated microbiome changes in breast cancer patients reveals statistically robust evidence for specific bacterial shifts during treatment, though most findings require validation in larger cohorts. After rigorous FDR correction, we found that Blautia consistently increasing across general hormone therapy (FDR p = 0.002) and AI treatment (FDR p = 0.032), while Dialister and Megasphaera significantly increased with LHRH therapy (FDR p = 0.008 and p = 0.042). Blautia emerges as the most compelling biomarker given its statistical robustness across multiple hormone therapies and established role in promoting anti-tumor immune responses and improved breast cancer outcomes. While bacterial signatures distinguishing HR+/HR− patients and lymph node metastasis groups showed promising effect sizes, these preliminary observations did not survive FDR correction and require validation. Critical study limitations include small subgroup sample sizes (Tamoxifen n = 11–13, LHRH n = 9), unbalanced comparisons, and the inherent multiple testing burden of microbiome research, which significantly constrain generalizability and necessitate cautious interpretation. These findings provide a strong foundation for future adequately powered studies (minimum n = 50–100 per group) and multi-center collaborations essential for clinical translation, with Blautia representing the most promising target for mechanistic investigation and potential microbiome-based interventions, though larger validation cohorts remain absolutely essential before clinical application.

Methods

Study objectives and outcomes

This prospective longitudinal study investigated hormone therapy-associated gut microbiota changes in breast cancer patients with four primary objectives: (1) characterize baseline microbiome differences between hormone receptor-positive (HR+) and negative (HR-) patients, (2) evaluate microbiome variations by axillary lymph node metastasis status, (3) assess longitudinal changes during specific hormone therapies, and (4) identify potential microbiome biomarkers. Primary outcomes included taxonomic abundance changes, diversity measures, and effect sizes with rigorous FDR correction for multiple comparisons.

Patient population

This prospective longitudinal and observational study was conducted at MacKay Memorial Hospital (MMH), Taipei, Taiwan, with institutional review board approval (IRB No. 19MMHIS061e). Participants were recruited through convenience sampling, comprising adults (≥20 years) with histologically confirmed stage I–IV breast cancer via core biopsy. All participants provided written informed consent, while patients with recurrent breast cancer were excluded. Treatment protocols for HR+ breast cancer patients were stratified by menopausal status, with premenopausal women receiving selective estrogen receptor modulators (SERMs, primarily Tamoxifen) and postmenopausal women prescribed AIs and Tamoxifen. In high-risk premenopausal cases or patients desiring future fertility, treatment was augmented with ovarian suppression using luteinizing hormone-releasing hormone (LHRH) agonists. Hormone treatment (HT) duration generally ranges from 5 to 10 years. Fecal samples were collected at baseline (first admission) for all participants, with HR+ patients providing additional samples at three- and six-months post-initiation of HT to assess temporal changes in gut microbiota composition.

Fecal samples collection and DNA extraction of microbiota

Fecal samples were collected upon breast cancer confirmation by core biopsy and stored at −20 °C until analysis. Genomic DNA was extracted using the QIAamp Fast DNA Stool Mini Kit (QIAGEN GmbH, Hilden, Germany) according to manufacturer’s instructions with modifications. Briefly, 0.2 g of fecal sample was homogenized with 1 mL InhibitEX buffer and glass beads using a Precellys homogenizer (Bertin Instruments, Montigny-le-Bretonneux, France) at 4500 beats/min for 2 min. The suspension was heated at 70 °C for 10 min and centrifuged for 1 min to pellet stool particles. The supernatant (600 µL) was transferred to a new 2 mL tube containing 25 µL proteinase K, followed by the addition of 600 µL AL buffer. After incubation at 70 °C for 10 min, 600 µL of 100% ethanol was added and mixed thoroughly. The mixture was filtered through a QIAamp spin column at 13,000 rpm for 1 min, followed by sequential washing with AW1 and AW2 buffers. DNA was eluted in 100 µL ATE buffer, and its concentration and quality were assessed using a NanoDrop 2000 spectrophotometer (Thermo Scientific, MA, USA).

16S rRNA library construction and sequencing

The V3-V4 hypervariable regions of the 16S rRNA gene were amplified for microbial phylogenetic classification. The first PCR amplification was performed in 25 µL reactions containing 50 ng (2.5 µL) of template DNA, 0.2 µM each of V3-V4 forward primer (5’-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3’) and reverse primer (5’-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC-3’), and 12.5 µL 2X Kapa HiFi HotStart ReadyMix (KapaBiosystems). PCR conditions were initial denaturation at 95 °C for 3 min; 25 cycles of denaturation at 95 °C for 30 s, annealing at 55 °C for 30 s, and extension at 72 °C for 30 s; followed by final extension at 72 °C for 5 min. Amplified products were purified using Agencourt AMPure XP Reagent beads (Beckman Coulter Inc., CA, USA).

Index PCR was performed in 50 µL reactions containing 5 µL of purified amplicon, 25 µL 2X Kapa HiFi HotStart ReadyMix, and 5 µL each of Nextera XT Index 1 and 2 primers (Illumina, CA, USA). Cycling conditions were: 95 °C for 3 min; 8 cycles of 95 °C for 30 s, 55 °C for 30 s, and 72 °C for 30 s; with final extension at 72 °C for 5 min using an Applied Biosystems 2720 thermocycler (Thermo Fisher Scientific, CA, USA). The indexed amplicons were purified using AMPure XP beads. Libraries were quantified using KAPA SYBR FAST qPCR Master Mix on a Roche LightCycler 480 system, normalized to 4 nM, and pooled for sequencing on the Illumina MiSeq platform with 2 × 300 bp paired-end chemistry, generating >80,000 reads per sample.

Sequence data were processed using QIIME2 software package (version 2017.10). Chimeric sequences were removed using DADA2, and reads were trimmed by 30 and 90 bases at the 3’ end of forward and reverse reads, respectively. Taxonomic classification was performed using a Naïve Bayes classifier trained on the Greengenes13.8 database with a 99% OTU threshold. The resulting taxonomic assignments were classified at seven levels (kingdom, phylum, class, order, family, genus, and species) using the NCBI database.

Statistical analysis

Statistical analyses were performed using SPSS 26.0 software. Data is presented as means ± standard deviation (SD). Comparisons between two groups were conducted using Student’s t-test, while multiple group comparisons employed one-way or two-way ANOVA as appropriate. Statistical significance was set at p < 0.05.

Microbial community analyses were performed using MicrobiomeAnalyst. Bacterial compositions were analyzed at Family, Genus, and Species levels. Alpha diversity was assessed using the Shannon index calculated through QIIME software package (version 2017.10). Beta diversity between samples was evaluated using the Bray-Curtis dissimilarity index and visualized through Principal Coordinate Analysis (PCoA). The statistical significance of beta diversity differences was determined using Permutational Multivariate Analysis of Variance (PERMANOVA), with reported F-values, r-squared values, and p-values.

For taxonomic analysis, bacterial taxa absent in ≥5% of participants were excluded. For genera with median relative abundance >1%, multiple regression analysis was performed to examine associations with bacterial composition, adjusting for potential confounders. Non-normally distributed data, identified through skewness and kurtosis calculations, were normalized using Box-Cox transformation. Statistical significance for each genus and species was assessed using t-tests.

Taxonomic compositions were visualized using stacked plots showing percentage abundance (PA) of different genera and species, with separate graphs highlighting the top 20 most abundant taxa. Linear discriminant analysis Effect Size (LEfSe) was employed to identify differentially abundant features between sample groups. Features were considered significant when meeting both criteria: p < 0.05 and Log LDA score >1.0. The analysis was stratified by taxonomic levels (Family, Genus, and Species).

All analyses in this study employed rigorous False Discovery Rate (FDR) correction using the Benjamini–Hochberg procedure to control for multiple comparisons. Given the exploratory nature of microbiome research and the inherent challenge of multiple testing in taxonomic analyses, we applied conservative statistical thresholds (FDR p < 0.05) to minimize false discovery rates. Effect sizes were calculated using Cohen’s d with 95% confidence intervals to assess biological relevance independent of statistical significance. Sample sizes for each analysis were carefully documented, and power limitations are explicitly acknowledged throughout the results.

Author contributions

Conceptualization (C.H.T., Y.W.S., F.L., C.C.L. and P.S.Y.), methodology (C.H.T., W.L.H., Y.W.S., F.L., H.W.Y. and P.S.Y.), software (W.L.H., H.W.Y., C.Y.L. and P.S.Y.), validation (C.H.T., W.L.H., H.W.Y. and P.S.Y.), formal analysis (C.H.T.,W.L.H., and P.S.Y.), investigation (C.H.T., Y.W.S., F.L., C.C.L. and P.S..Y.), and data curation (C.H.T., Y.W.S., F.Y.L., H.W.Y., C.Y.L. and P.S.Y.). Writing contributions: original draft preparation (C.H.T., W.L.H., Y.W.S., F.L., C.C.L., H.W.Y. and P.S.Y.), review (C.H.T., W.L.H., Y.W.S., F.L., C.C.L., F.Y.L., H.W.Y. and P.S.Y.), and editing (C.H.T., W.L.H., Y.W.S. and P.S.Y.), supervision (C.H.T. and P.S.Y.), project administration (C.H.T. and P.S.Y.), funding acquisition (F.L. and P.S.Y.). All authors have read and agreed to the published version of the manuscript.

Data availability

The data for this study are available upon: https://www.ncbi.nlm.nih.gov/bioproject/PRJNA953204/. Project ID: PRJNA953204.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41523-025-00810-2.