Induction of senescence during postpartum mammary gland involution supports tissue remodeling and promotes postpartum tumorigenesis

Abstract

Cellular senescence is an evolutionarily conserved stress response that contributes to tissue repair and tumor suppression, yet its accumulation is also linked to aging and disease. Whether physiological senescence can be exploited by oncogenic events to promote tumorigenesis is unknown. Postpartum mammary gland involution is a major adult tissue remodeling event, resembling wound healing, and is closely associated with postpartum breast cancer. Here, we show that during mammary gland involution in mice, a p16^Ink4a-dependent senescence response is induced in alveolar luminal cells. Eliminating senescent cells disrupts tissue remodeling and delays involution, demonstrating their physiological importance. However, in a postpartum breast cancer model where oncogenic activation coincides with involution, removing involution-associated senescent cells extended tumor latency. Mechanistically, senescent cells enhance tumor cell plasticity via the senescence-associated secretory phenotype, fostering metastasis. Our findings reveal that senescence, while required for postpartum tissue remodeling, can be hijacked to facilitate tumorigenesis, defining senescence as a unifying mechanism linking tissue repair to tumorigenesis.

Chiche, Djoual, Charifou and colleagues identify a dual role for cell senescence in postpartum tissue remodeling: senescent cells drive mammary gland involution, yet create a permissive niche for tumor initiation, revealing how a conserved repair program can be redirected toward cancer progression.

Main

Cellular senescence is a stress-induced cell state characterized by stable cell cycle arrest, resistance to apoptosis and a robust secretome, known as the senescence-associated secretory phenotype (SASP)^1,2. However, senescence is not a uniform identity as previously viewed; it is now recognized as highly heterogeneous phenotypes and pleiotropic functions that vary depending on cell type origin, inducers and tissue context³. The stable cell cycle arrest is a potent cell-intrinsic tumor suppression mechanism⁴, yet contributes to diminished regenerative capacity and tissue dysfunction during aging⁵. Extrinsically, senescent cells extensively modulate the microenvironment by secreting SASP factors and presenting cell-surface proteins^6,7. The functional duality positions senescence as a key player in numerous physiological and pathological processes^1,2, including embryonic development^8,9, tissue repair^10,11, fibrosis^12–14, cancer¹⁵ and aging¹⁶. Importantly, targeted elimination of senescent cells can delay the onset and improve many aging-associated pathologies^17–19. Despite extensive research into the pathological roles of senescence, its functional contributions to normal physiology remain poorly understood^20,21. Furthermore, it remains unclear whether physiological senescence can be hijacked by oncogenic processes to drive tumorigenesis.

The postpartum period involves extensive physiological changes with lasting implications for women’s health^22,23. Among these, the mammary gland (MG) undergoes dramatic remodeling during postpartum involution, a process that returns the gland to its pre-pregnant state after lactation (or directly after pregnancy if lactation does not occur)²⁴. During this process, all the newly formed lobuloalveolar structures generated during pregnancy are removed via apoptosis and lysosomal-mediated cell death^25–30. Concurrently, the gland undergoes extensive remodeling of the extracellular matrix (ECM) and blood vessels and adipocyte refilling, accompanied by a strong inflammatory response and immune infiltration^31–38, to restore mammary epithelium integrity. Of note, mammary involution has been proposed to share features with scarless wound healing, as the mammary epithelium after involution can respond properly to the subsequent pregnancy and lactation repetitively throughout the female’s reproductive lifespan, although the underlying mechanisms may differ.

While involution is critical for MG homeostasis and the successful establishment of future lactation³⁹, it also creates a transient pro-tumorigenic microenvironment⁴⁰ and is strongly implicated in postpartum breast cancer (PPBC), defined as breast cancer diagnosed within 5–10 years after childbirth^41–43. PPBC is associated with higher metastasis rates and poorer survival compared nulliparous women or those diagnosed during pregnancy, regardless of hormone receptor status^41–43. Notably, increased maternal age positively correlates with a higher incidence of PPBC⁴⁴. With the global trend of delayed childbirth, PPBC incidence and mortality rates are expected to rise markedly, emphasizing a critical unmet medical need. However, the mechanisms underlying the pro-tumorigenic features of involution remain not well understood.

Here, we provide evidence that certain alveolar luminal cells upregulate p16^INK4a to acquire senescence state during postpartum MG involution. Loss-of-function studies reveal that senescence is required for proper tissue remodeling during involution but, paradoxically, also promotes tumorigenesis and metastasis in a paracrine manner by enhancing tumor-initiating cell plasticity. Thus, our findings demonstrate that in a physiological context such as MG involution, senescence plays a dual role in both mediating tissue remodeling and facilitating tumorigenesis, suggesting that senescence might serve as a common mechanism linking wound healing to cancer.

Result

Cellular senescence is transiently induced at the onset of the tissue remodeling phase of MG involution

To investigate whether senescence occurs throughout MG postnatal development and the reproductive cycle, we performed senescence-associated β-galactosidase (SAβGal) staining on murine MGs collected at various developmental stages, including puberty (6 weeks), mature virgin (10 weeks and 12 months), pregnancy (pregnancy day 13.5 (p13.5) and pregnancy day 18.5 (p18.5)), lactation (lactation day 10 (Lac10)) and involution (Fig. 1a). Notably, robust SAβGal staining was observed in involuting MGs, while SAβGal⁺ cells were scarcely detectable at other stages (Fig. 1b–d). The involution process in rodents consists of two morphologically distinct phases²⁴. During the first 48 h, programmed cell death (PCD) occurs, while the alveolar lumen remains intact. This is followed by a phase of extensive tissue remodeling between 60 and 72 h, characterized by alveolar regression and adipocyte repopulation²⁴. To analyze the dynamics of SAβGal⁺ cells during involution, we initiated synchronized involution by removing pups after 10 d of lactation and collected MGs at different timepoints: involution day 1 (Inv1), Inv3, Inv4, Inv5, Inv7, Inv10, Inv14 and Inv28 (representing the resting MG) (Fig. 1a). Interestingly, SAβGal⁺ cells became apparent starting at Inv3, with staining intensity peaking at Inv7 and gradually diminishing thereafter (Fig. 1c(i–vii),e). No SAβGal⁺ cells were detected in the resting MG following the completion of involution (Inv28, Fig. 1c(vii)). Most SAβGal staining was spatially restricted to the alveolar region compared with the ductal epithelium.

Fig. 1. Transient induction of senescence during MG involution.

a, Experimental design highlighting the timepoints of MG collection at: puberty (6 weeks), adult virgin (10 weeks), pregnancy (p13.5, p18.5), lactation (10 d, Lac10/Inv0) and involution after 1, 3, 4, 5, 7, 10, 14 and 28 d/resting stage (Inv, Inv3, Inv4, Inv5, Inv7, Inv10, Inv14 and Inv28/RS) following the first pregnancy as well as after at least three rounds of reproductive cycles (MP-Inv5 and MP-RS), or in old virgin female mice (12 months). b, SAβgal staining in mammary glands during puberty and the virgin state, pregnancy and lactation. Puberty (n = 3), Y-virgin (n = 5), p13.5 (n = 3), p18.5 (n = 3), Lac10/Inv0 (n = 7). c, SAβgal staining in mammary glands during involution. Inv1 (n = 3), Inv2 (n = 2), Inv3 (n = 3), Inv5 (n = 5), Inv7 (n = 4), Inv10 (n = 2), Inv14 (n = 2), Inv28/1P-RS (n = 4). d, SAβgal staining during multiparous involution stages and one year old virgin state. MP-Inv5 (n = 3), MP-RS (n = 4), O-virgin (n = 5). Scale bar: 100 μm. e, Quantification of SAβgal⁺ cells related to total number of cells at different timepoints of MG stage: puberty (n = 3), Y-virgin (n = 5), p18.5 (n = 3), Lac10/Inv0 (n = 7), Inv1 (n = 3), Inv2 (n = 2), Inv3 (n = 3), Inv5 (n = 5), Inv7 (n = 4), Inv10 (n = 2), Inv14 (n = 2), Inv28/1P-RS (n = 4), MP-Inv5 (n = 3), MP-RS (n = 4), O-virgin (n = 5). Data are represented as mean ± s.d. Mann–Whitney test was performed. P = 0.5714. f, Principal component analysis (PCA) of transcriptomes from MGs collected at Inv0, Inv3 and Inv7 (n = 3 for each). Inv3 versus Inv0 P < 0.001; Inv7 versus Inv0 P < 0.001. g, GSEA plots for genes related to cellular senescence, using the SAUL_SEN_MAYO (Sen_Mayo) gene set, comparing Inv3 with Inv0 as well as Inv7 with Inv0. In total, 1,000 permutations using the ‘gene_set’ permutation type parameter were performed. Illustrations in a created with BioRender.com. NES, normalized enrichment score; NS, not significant.

Source data

To further validate our observations and characterize the senescence program associated with MG involution, we conducted bulk RNA sequencing (RNA-seq) on MGs from Inv0, Inv3 and Inv7. Principal component analysis revealed clear clustering of Inv0, Inv3 and Inv7 samples, indicating distinct biological differences across these stages (Fig. 1f). Gene set enrichment analysis (GSEA) demonstrated significant enrichment of senescence-associated genes⁴⁵ in both Inv3 and Inv7 compared with Inv0 (Fig. 1g). A heatmap depicting upregulated senescence-associated genes in Inv3 and Inv7 relative to Inv0 further highlighted this senescence signature (Extended Data Fig. 1a). Interestingly, we identified numerous genes crucial for the involution process, including ECM remodeling factors such as Mmp2, Mmp3, Mmp14 and Serpine1 (refs.^34,35); cytokines and chemokines⁴⁶ such as Cxcl12, Cxcl10, Ccl4, Cxcl16, Ccl3, Ccl2, Il18 and Ccl8; and growth factors^35,47, including Gdf15, Fgf2/7 and Igfbp4/5. Additionally, hallmark gene sets frequently associated with senescence, such as interferon responses⁴⁸, epithelial–mesenchymal transition⁴⁹, NF-κB signaling⁵⁰ and the complement system, were significantly enriched in Inv3 and Inv7 (Extended Data Fig. 1b). Collectively, these findings suggest that senescence is transiently induced during MG involution, coinciding with the onset of tissue remodeling.

Extended Data Fig. 1. Senescence-associated genes are upregulated during mammary gland involution.

A. Heatmap of RNA-seq transcriptome analysis for 38 selected senescence-associated genes (customized gene set) that are significantly enriched in both Inv3 and Inv7 comparing to Inv0 (based on the GSEA analysis), n = 3 for each group. B. Bi-directional bar chart of HALLMARK gene sets, comparing Inv3 to Inv0, and Inv7 to Inv0. Absence of bar indicates absence of significant difference between experimental conditions.

Source data

Involution-associated senescence does not accumulate with age

Senescent cells accumulate during aging in various tissues¹. A recent study has reported that aged mammary stem/progenitor cells acquire a gene signature associated with cellular senescence⁵¹. We questioned whether senescent cells might accumulate in the MG upon multiple pregnancies (multiparity). To address this question, we compared SAβGal⁺ cells across several groups, including involuting MGs following the first pregnancy (1P-Inv5, ~3 months), resting MGs after one round of involution (1P-RS, ~4 months), involuting MGs from multiparous mice (MP-Inv5, ~11 months, corresponding to approximately 35 years in humans), resting MGs post-multiparity (MP-RS, ~12 months, corresponding to approximately 40 years in humans) and glands from nulliparous mice (O-virgin, ~12 months) (Fig. 1c–e). We did not observe an increase in SAβGal⁺ cells in involuting multiparous MGs (MP-Inv5) compared with first pregnancy MGs (IP-Inv5) (Fig. 1f). Furthermore, we detected negligible SAβGal⁺ cells in MGs from 1-year-old female mice, regardless of parity status. These findings suggest that the senescence program associated with involution does not accumulate or amplify with multiple pregnancies during reproductive aging.

Alveolar luminal cells are the major senescent cell type

To identify which cell types undergo senescence during MG involution, we employed C₁₂FDG, a fluorogenic substrate for β-galactosidase⁵², alongside cell-specific markers. Stromal cells were identified as CD31⁻, CD45⁻, CD24⁻, α6⁻; basal cells as CD31⁻, CD45⁻, CD24^low, α6^high; and luminal cells as CD31⁻, CD45⁻, CD24^high, α6^low. Using flow cytometry, we found that approximately 6% of stromal cells, 15% of basal cells and 52% of luminal cells were C₁₂FDG⁺ at Inv5 compared with Inv0 (Extended Data Fig. 2b). Although our antibody panel could not distinguish alveolar epithelial cells from ductal epithelial cells, histological staining indicated that most SAβGal⁺ cells are alveolar epithelial cells. Next, we conducted co-staining of SAβGal with the basal cell marker Keratin 5 (KRT5) and the luminal cell marker Keratin 8 (KRT8) at Inv3. Consistently, we found that the majority of SAβGal⁺ cells were Krt8⁺ alveolar luminal cells (Fig. 2a), which are negative for both the apoptotic marker cleaved caspase-3 (CC3) and the proliferation marker Ki67 (Extended Data Fig. 2c,d). Together, our findings indicate that alveolar luminal cells are the primary senescent cell type during MG involution.

Extended Data Fig. 2. p16 is upregulated in involuted mammary gland.

A. Experimental design of C₁₂FDG staining in dissociated involuting mammary glands at Inv0 and Inv5. B. Representative FACS analysis of C₁₂FDG staining in stromal, basal and luminal cells, from mammary gland of mice at Inv0 (red) or Inv5 (blue) (n = 3). C-D. Representative pictures of co-staining SAβGal with CC3 or Ki67 on WT mammary glands from Inv3 females (n ≥ 2). Scale bar: 100 μm. E-F. Heatmaps derived from RNA-seq analysis showing the expression profiles of 4 selected cell-cycle regulator genes (E) and 19 selected TP53 target genes (F). All displayed genes are significantly differentially expressed in Inv3 and Inv7 relative to Inv0, as determined by differential expression analysis; n = 3 for each group. G. Representative pictures of co-staining of p16^INK−GFP with Krt5 on INKBRITE involuting mammary gland from Inv0 (n = 4) and Inv7 (n = 3). Scale bar: 50 mm. Quantification represents p16^{INKBRITE−GFP+} cells. H. Experimental design of in vivo experiment. p16Cre^ERT2; R26^mT/mG mice were treated intraperitoneally with tamoxifen (TMX) at 2 mg.kg-1.day-1 at Inv2.5 and Inv3.5 to label p16+ cells. Mammary glands were collected at Inv5. Representative pictures of p16^mGFP with gH₂AX staining on mammary glands of p16Cre^ERT2; R26^mT/mG mice at Inv5 (n = 3). Illustrations in a created with BioRender.com.

Source data

Fig. 2. Alveolar luminal cells are the major senescent cell type.

a, Representative pictures of co-staining SAβGal with Krt5 or Krt8 markers on wild-type MGs from Inv3 female mice (n = 2). Scale bar: 100 μm. b, qPCR analysis of Cdkn2a (p16Ink4a), Cdkn2a (p19ARF) and Cdkn1a genes from MGs collected from Inv0 to Inv28 (n = 4 for each timepoint). Data are represented as mean ± s.d relative to Inv0. One-way ANOVAs were performed. **P = 0.0021, *P = 0.015, ****P < 0.0001. c, Representative pictures of co-staining of p16^INK−GFP with KRT5 and γH₂AX on INKBRITE involuting MG from Inv0 (n = 4) and Inv3 (n = 4). Scale bar: 100 μm. d, Quantification of p16^INK−GFP+ cells (green), γH₂AX⁺ cells (red) and double-positive p16^INK−GFP+; γH₂AX⁺ cells (yellow) on INKBRITE involuting MG from Inv0 (n = 5), Inv3 (n = 4) and Inv7 (n = 3). Data are represented as mean ± s.d. e, Experimental design of ex vivo culture of organoids. f, SAβGal staining of involuting organoids from involuting (BOM, d18) and lactation control (LM, d18) (n = 3). Scale bar: 100 μm. g, qPCR analysis of Cdkn2a (p16Ink4a) and Cdkn1a of organoids from involution-like and lactation-like processes. n = 3 biological replicates. Data are represented as mean ± s.d. Two-way ANOVAs were performed. *P = 0.0134, **P = 0.0039. ANOVA, analysis of variance; d, day.

Source data

p16^Ink4a upregulation, but not p21^CIP1, is associated with mammary involution-induced senescence

Interestingly, our bulk RNA-seq analysis revealed upregulation of Cdkn2a, Rb1, Trp53 and various Trp53 target genes⁵³, but not Cdkn1a, at Inv3 and Inv7 compared with Inv0 (Extended Data Fig. 2e,f). To validate these findings, we performed quantitative PCR (qPCR) on RNA extracted from whole MGs at different involution timepoints. Results showed significant upregulation of both p16^INK4a and p19^Arf at Inv7, while p21 was transiently upregulated at Inv1 and returned to the basal levels by Inv3 (Fig. 2b). We further investigated p16^INK4a expression using a sensitive reporter mouse model, INKBRITE, where the Cdkn2a (p16Ink4a) promoter drives expression of multiple copies of H2B-GFP⁵⁴. At Inv0, few GFP⁺ cells were observed within alveolar structures marked by Krt5 or in the stromal compartment (Fig. 2c, left panel, and Extended Data Fig. 2g). Importantly, at Inv7, a significantly higher proportion of GFP⁺ cells was detected, predominantly within the alveolar structures facing the lumen (Fig. 2d and Extended Data Fig. 2g, lower panel). Next, we examined whether involution-induced senescence was triggered by DNA damage. While we observed γH2AX⁺ cells at Inv3, predominantly in alveolar luminal cells, GFP⁺ and γH₂AX⁺ cells were mutually exclusive (Fig. 2c). Furthermore, we generated a p16Cre^ERT2 (ref. ⁵⁵); R26^mTomato/mGFP lineage-tracing model that enables specific labeling of p16^high cells with mGFP upon tamoxifen administration. Immunofluorescence staining revealed that most p16^mGFP+ cells lacked nuclear γH₂AX foci, and conversely, the majority of γH₂AX⁺ cells were not p16^mGFP+ (Extended Data Fig. 2h). Quantifications in both mouse models confirmed the low frequency of double-positive cells, indicating that senescence during early involution occurs largely independently of canonical DNA damage responses (Fig. 2d and Extended Data Fig. 2h). Together, these findings suggest that involution-induced senescence is linked to p16^Ink4a upregulation, independent of p21, and occurs most likely independently of genotoxic stress.

We recently established a mammary organoid system that models mammary postnatal developmental stages including pregnancy, lactation and involution ex vivo⁵⁶. To test whether senescence is also induced during the involution-like process in this organoid model, we performed whole-mount SAβGal staining on organoids cultured in either lactation-like medium (LM, Lac: containing prolactin) or involution-like medium (basal organoid medium (BOM), Inv: without hormone) (Fig. 2e,f). Histological analysis revealed that involution organoids were notably smaller than lactation organoids, with significantly more SAβGal⁺ cells in the involution organoids (Fig. 2f). Consistently, qPCR analysis confirmed significant upregulation of Cdkn2a (p16Ink4a) in involution-like organoids, with no significant change in Cdkn1a expression (Fig. 2g). Thus, the organoid model supports the in vivo finding that p16^Ink4a, but not Cdkn1a, is associated with senescence during involution, and the organoid system can serve as a simplified ex vivo model to study involution-associated senescence.

Systemic lactogenic hormones prevent senescence induction during involution

Next, we investigated upstream signals regulating senescence induction during involution. Organoid experiments indicated that prolactin withdrawal triggers senescence (Fig. 2e–g). To assess whether systemic lactogenic signals are important for senescence induction in vivo, we sealed a set of teats of lactating mice to initiate local involution while maintaining systemic lactogenic hormone levels (Fig. 3a and Extended Data Fig. 3a,b). It is known that involution-associated PCD is triggered locally²⁶, and consistent with previous findings, we observed distinct morphological differences between involuting and sealed MGs at the onset of irreversible involution (day 3, Fig. 3b and Extended Data Fig. 3c,d). The sealed MGs displayed intact alveolar structures without tissue remodeling, unlike the involuting MGs. Importantly, no SAβGal⁺ cells or Cdkn2a (p16Ink4a) upregulation were detected in sealed MGs compared with involuting MGs at Inv3 (Fig. 3b and Extended Data Fig. 3b) or at Inv5 (Extended Data Fig. 3d). These results suggest that systemic lactogenic hormone withdrawal upon weaning is required for senescence induction during involution.

Fig. 3. Impaired senescence delays MG involution.

a, Experimental design of involuting (blue) or teat sealed (red) MGs, while other MGs are maintained in lactation. MGs were collected 3 d after sealing and compared with MGs from mice undergoing normal involution. b, Representative pictures of whole-mount carmine and SAβGal staining from involuting or sealed MGs at day 3, on cryosections, counterstained with Fast Red. Scale bar: 100 μm. Automatic quantifications of SAβGal staining at day 3, from involuting (black, n = 2) or sealed (red, n = 3) MGs. Data are represented as mean ± s.d. c, Experimental design of experiment with BlgCre; p16^f/f mice. d, Representative pictures of SAβGal staining (left panel) from p16^f/f or Blg^Tg/+; p16^f/f MGs at Inv5, on cryosections, counterstained with Fast Red. Scale bar: 100 μm. Automatic quantification of SAβGal staining (right panel), p16^f/f (n = 7); Blg^Tg/+; p16^f/f (n = 9). Data are represented as mean ± s.d. Unpaired t-test and Mann–Whitney test were performed. *P = 0.0229. e, Representative pictures of α-SMA staining (left panel) from p16^f/f or Blg^Tg/+; p16^f/f MGs at Inv5. Scale bar: 100 μm. Quantification of remaining epithelium (right panel), p16^f/f (n = 9); Blg^Tg/+; p16^f/f (n = 12). Data are represented as mean ± s.d. Mann–Whitney test was performed. *P = 0.0184. f, Representative pictures of perilipin staining from p16^f/f or Blg^Tg/+; p16^f/f MGs at Inv5. Scale bar: 100 μm. g–i, Quantification of adipose tissue proportion (g), number of adipocytes per unit area (h) and average of adipocyte area per unit area (i), n = 3 mice per group. Data are represented as mean ± s.d. Unpaired t-test was performed. *P = 0.0204 (g); **P = 0.0078 (h); **P = 0.0066 (i). j, Distribution of adipocyte size from p16^f/f or Blg^Tg/+; p16^f/f MGs at Inv5, n = 3 mice per group. Data are represented as mean ± s.d. Illustrations in a created with BioRender.com.

Source data

Extended Data Fig. 3. Systemic lactogenic hormones prevent senescence induction during involution.

A. Proper sealing controlled by typical morphology during harvesting of mammary glands from lactating & sealed (left) or involuting (right) mice. B. qPCR analysis of p16, WAP, and CSN2 genes from mammary glands collected at day 3, from involuting (black, n = 6), sealed (red, n = 3) or lactating (blue, n = 3) tissues. Data are represented as mean ± standard deviation. Unpaired t-test was performed. p-values *<0.05. C-D. Whole mount mammary glands after carmine (left panel) and SAβGal staining on cryosections (right panel), from sealed or involuting females at day 2 (C) and day 5 (D). Scale bar: 400 μm or 100 μm. E. Representative pictures of SAβGal staining from p16^f/f or Blg^Tg/+; p16^f/f mammary glands at Inv3 and Inv4, on cryosections, counterstained with FastRed. Scale bar: 100 μm. F. Automatic quantification of SAβgal+ cells in p16^f/f (n = 5) and Blg^Tg/+; p16^f/f (n = 4). Data are represented as mean ± standard deviation. Unpaired t-test was performed.

Source data

p16^Ink4a-dependent senescence induction ensures involution dynamic

To determine the functional relevance of senescence in mammary involution, we used both genetic and pharmaceutical approaches to perturb the senescence program. First, we generated BlgCre^Tg/+; p16^f/f mice, an alveolar luminal cell-specific Cdkn2a (p16Ink4a) knockout model, to assess p16’s role in senescence during involution (Fig. 3c). The BlgCre transgene drives Cre recombinase expression under the β-lactoglobulin (Blg) promoter, which is active in luminal alveolar cells of the MG during lactation⁵⁷. In BlgCre^Tg/+; p16^f/f mice, Cre expression leads to the specific deletion of the p16^Ink4a locus in luminal alveolar cells, whereas p16^f/f littermates serve as a negative control with intact Cdkn2a (p16Ink4a) expression. In BlgCre^Tg/+; p16^f/f mice, we observed a delayed onset of tissue remodeling compared with p16^f/f controls (Extended Data Fig. 3e). Importantly, BlgCre^Tg/+; p16^f/f MGs exhibited significantly fewer SAβGal⁺ cells at Inv5 compared with littermate controls (Fig. 3d), although a minor, nonsignificant reduction in SAβGal⁺ cells was also observed at Inv4 (Extended Data Fig. 4e,f). Hence, these findings demonstrate that alveolar luminal cell senescence during involution is p16^Ink4a-dependent.

Extended Data Fig. 4. Delayed involution is resolved at later timepoints.

A-B. Whole mount mammary glands after carmine (left panel); SAβGal on cryosections (middle panel) and α-SMA staining (right panel) from p16^f/f or Blg^Tg/+; p16^f/f mammary glands at Inv10 or Inv28. Scale bar: 400 μm or 100 μm. C. Automatic quantification of SAβGal+ cells at Inv10 (p16^f/f (n = 4); Blg^Tg/+; p16^f/f n = 3)) and Inv28 (n = 4 per group). Data are represented as mean ± standard deviation. Unpaired t-tests were performed. D. Quantification of α-SMA for remaining epithelium in percentage per picture at Inv10 (p16^f/f (n = 4); Blg^Tg/+; p16^f/f n = 3)) and Inv28 (n = 4 per group). Data are represented as mean ± standard deviation. Unpaired t-tests were performed.

Source data

Next, we evaluated the functional consequence of impaired senescence induction during MG involution, using alveolar size and adipocyte area as metrics to evaluate the involution kinetic. Interestingly, at Inv5, BlgCre^Tg/+; p16^f/f MGs retained more epithelial structures and exhibited a reduced adipocyte area compared with controls (Fig. 3e–g). Additionally, BlgCre^Tg/+; p16^f/f MGs contained much more, but markedly smaller, adipocytes than controls (Fig. 3h–j), suggesting a link between reduced senescence and impaired adipocyte re-differentiation and hypotrophy.

To determine whether reduced senescence induction affects the progression or completion of involution, we extended our analyses to later timepoints (Inv10 and Inv28) in BlgCre^Tg/+; p16^f/f mice (Extended Data Fig. 4). We assessed MG morphology by whole-mount carmine staining, performed histological analyses and quantified SAβGal⁺ cells. These complementary analyses revealed no significant differences between control and BlgCre^Tg/+; p16^f/f glands in tissue remodeling, suggesting that the absence of p16^INK4a-associated senescence does not impair the resolution of involution. Collectively, these findings demonstrate that p16-dependent senescence is crucial for the onset of tissue modeling during MG involution; however, completion of the process may rely either on residual senescent cells or on compensatory mechanisms such as apoptosis⁸, underscoring the robustness of the involution program.

ABT-263 treatment delays MG involution

Next, we used ABT-263^18,58, a BH3 mimetic targeting anti-apoptotic Bcl-2 family proteins, to deplete senescent cells during involution. Mice were treated with ABT-263 daily for 3 d before, and MGs were collected at Inv3, Inv4 and Inv5 (Fig. 4a). Notably, ABT-263 treatment led to a significant increase in apoptotic (CC3⁺) cells at Inv5 (Fig. 4b) and a decrease in SAβGal⁺ cells, suggesting apoptosis was induced in senescent cells (Fig. 4b,c and Extended Data Fig. 5b). Notably, ABT-263 did not affect STAT3 activation, a key mediator of PCD during mammary involution⁵⁹ (Extended Data Fig. 5a). Consistent with the BlgCre^Tg/+; p16^f/f mice, ABT-263-treated MGs exhibited delayed alveolar regression and reduced adipocyte refilling at Inv4 and Inv5 compared with controls (Fig. 4d–g). However, unlike the BlgCre^Tg/+; p16^f/f model, we observed no change in adipocyte numbers, except at Inv3 (Extended Data Fig. 5c), indicating differences between genetic senescence prevention of senescence induction and pharmacological clearance of senescent cells. These findings indicate that ABT-263 effectively removes senescent cells and delays the involution process. Taken together, both genetic and pharmaceutical approaches demonstrated that senescence induction is important for proper MG involution.

Fig. 4. ABT-263 treatment delays MG involution.

a, Experimental design of in vivo treatment with ABT-263 at 50 mg kg⁻¹ d⁻¹ or vehicle. Mice were force-fed every day for 3 d before MG collection at Inv3, Inv4 or Inv5. b,c, Representative pictures of MG sections from control or ABT-263-treated mice at Inv5, respectively stained for CC3 (b) and SAβGal on cryosection (c). Scale bar: 100 μm. Quantifications of CC3 (b) and SAβGal staining (c). CC3 staining: control n = 4; ABT-263 n = 6. SAβGal staining: control n = 10; ABT-263 n = 8. Data are represented as mean ± s.d. Unpaired t-tests were performed to analyze the effect of ABT-263 treatment, respectively, on the number of CC3⁺ or SAβGal⁺ cells. **P = 0.005 (b), **P = 0.0015 (c). d,f, Representative pictures of MG sections from control or ABT-263-treated mice at Inv3 to Inv5, respectively stained for α-SMA (d), or perilipin (f). Scale bars: 100 μm. e,g, Quantifications for remaining epithelium (e) and proportion of adipose tissue (g) in percentage. α-SMA staining: Inv3/Inv4 n = 5 for each group; Inv5 control n = 4, Inv5 ABT-263 n = 6. Perilipin staining: Inv3 n = 5 for each group; Inv4/Inv5 control n = 4; Inv4/Inv5 ABT-263 n = 5. Data are represented as mean ± s.d. Two-way ANOVAs were performed. *P = 0.0117, **P = 0.0047 (e), *P = 0.0186 (g). Illustrations in a created with BioRender.com.

Source data

Extended Data Fig. 5. ABT-263 treatment delays mammary gland involution.

A. Representative pictures of mammary gland sections from control or ABT-263 treated mice at Inv3, stained for p-STAT3. Scale bar: 100 μm. Quantification of p-STAT3 staining, n = 5 mice per group. Data are represented as mean ± standard deviation. Unpaired t-test was performed. B. Representative pictures of mammary gland sections from control or ABT-263 treated mice at Inv3 and Inv4, stained for SAβGal on cryosection. Scale bar: 100 μm. Automatic quantification, Inv3 Control n = 4; Inv3 ABT-263 n = 5; Inv4 n = 5 mice per group. Data are represented as mean ± standard deviation. Unpaired t-tests were performed to analyze the effect of ABT-263 treatment on the number of SAβGal⁺ cells. p-values *<0.05. C. Quantification of the number of adipocytes per picture, Inv3 n = 5 mice per group; Inv4/Inv5 Control n = 4, ABT-263 n = 5. Data are represented as mean ± standard deviation. Unpaired t-tests were performed. p-values *<0.05.

Source data

Senescent cells contribute to macrophage recruitment during involution

To gain insights on the role of senescence in involution, we further analyzed our bulk RNA-seq datasets. GSEA revealed enriched biological processes related to immune cell chemotaxis and migration (Fig. 5a,b). We took advantage of the involuting organoid system to determine the expression of relevant genes (Fig. 5c). The expression of Csn2, important for milk production, was significantly induced by prolactin and reduced after prolactin withdrawal⁵⁶ (Extended Data Fig. 6a). Consistently, Cdkn2a (p16Ink4a) expression was significantly upregulated post prolactin withdrawal (Extended Data Fig. 6b). Interestingly, we found that chemokines critical for monocyte/macrophage recruitment and ECM modifiers important for tissue remodeling were upregulated in involuting organoids compared with lactational controls (Fig. 5d and Extended Data Fig. 6b), suggesting that senescent epithelial cells are at least one of the sources producing these factors.

Fig. 5. Involution-induced senescence is required for macrophage recruitment.

a, GSEA (bar chart) comparing Inv3 with Inv0; top 20 significantly enriched gene sets from the Gene Ontology – Biological Process (GO-BP) database. b, GSEA plots for genes related to monocyte/macrophage chemoattraction, using the GOBP_MONOCYTE_CHEMOTAXIS (upper panel), Q < 0.001, and GOBP_MACROPHAGE_MIGRATION (lower panel), Q = 0.008, gene sets, comparing Inv3 with Inv0. In total, 1,000 permutations using the ‘gene_set’ permutation type parameter were performed. c, Design of the organoid culture experiment. d0 to d6: FGF₂ treatment to induce growth of organoids. d6 to d10: addition of LM to mimic lactation. d10 to d18: BOM to mimic involution. d, qPCR analysis of Ccl2, IL6, Csf1, Cxcl1 and Cxcl5 genes from mammary organoids collected at d14 and d18 relative to d10. n = 4 for each timepoint. Data are represented as mean ± s.d. e, Representative pictures of MG sections from p16^f/f or Blg^Tg/+; p16^f/f mice at Inv5, stained for F4/80, and its quantification. n = 4 mice per group. Data are represented as mean ± s.d. Unpaired t-test was performed. **P = 0.0038. Scale bar: 50 μm. f, Design of in vitro Transwell experiment. To study BMDM attraction, we collected CM from: growing organoids treated with FGF₂ (growing); lactogenic organoids treated with prolactin (lactation-like) and involuted organoids cultured in BOM (involution-like). Pictures show representative images of DAPI staining of attracted BMDMs. g, Quantifications of the numbers of nuclei (n = 3 independent experiments). Data are represented as mean ± s.d. One-way ANOVAs were performed to analyze the effects of growing, lactation-like and involution-like media on the attraction of BMDM cells. *P = 0.0363, NS P = 0.0664. FDR, false discovery rate.

Source data

Extended Data Fig. 6. Involution-induced senescence is required for macrophage recruitment.

A. qPCR analysis of Csn2 gene from mammary organoids collected at d6, d10, d14 and d18. n = 4 for each timepoint. Data are represented as mean ± standard deviation. Two-way ANOVAs were performed. p-values < 0.05 (d6 vs d10, d10 vs. d14 and d10 vs. d18). B. qPCR analysis of Serpine1, Plaur, Spp1, Ccn1, Ccn5, Ptgs2, Mmp3, Mmp14, Timp1 and p16 genes from mammary organoids collected at d14 and d18. n = 4 for each timepoint. Data are represented as mean ± standard deviation relative to d10. Two-way ANOVAs were performed. p-values < 0.05 (d10 vs. d14 and d10 vs. d18). C. Representative pictures of mammary gland sections from control or ABT-263 treated mice at Inv3 and Inv5, stained for F4/80. Scale bar: 50 μm. D. Quantification of F4/80+ cells in Control (n = 4) or ABT-263 (n = 5) mice at Inv3 and Inv5 (n = 4 per group). Data are represented as mean ± standard deviation. Mann-Whitney test was performed. p-values *<0.05. E. Representative flow cytometry staining plots of CD45 and CD64 expression on mammary gland cells isolated from control or ABT-263 treated involuted Inv3 and Inv5 females. F. Proportion of CD45⁺ and CD64⁺ cells in mammary gland isolated from control or ABT-263 treated involuted Inv3 (n = 3 per group) and Inv5 (n = 5 per group) females. Data are represented as mean ± standard deviation. Unpaired t-test was performed. p-values *<0.05.

Source data

Macrophages are known to infiltrate involuting MGs at the onset of the tissue remodeling phase, a process essential for adipocyte repopulation and involution progression^60,61. Considering the intimate relationship between macrophages and senescent cells⁶², we hypothesized that senescent epithelial cells recruit macrophages during involution. Supporting this, we found significantly fewer F4/80⁺ macrophages in BlgCre^Tg/+; p16^f/f involuting MGs compared with control littermates at Inv5 (Fig. 5e). Consistently, ABT-263-treated involuting MGs showed fewer CD45⁺ immune cells and macrophages (CD64⁺ or F4/80⁺) compared with controls at Inv3 and Inv5 (Extended Data Fig. 5c–f).

To functionally assess whether epithelial cells undergoing involution can actively recruit immune cells, we performed a Transwell migration assay using bone-marrow-derived macrophages (BMDMs). Conditioned medium (CM) was collected from organoids cultured under three different conditions mimicking key MG states: FGF₂-treated (growing), lactation-like and involution-like. BMDMs seeded in the upper chamber showed significantly increased migration toward CM from involution-like organoids, compared with the other two conditions. These results suggest that epithelial cells in an involution-like state secrete soluble factors that promote macrophage recruitment, supporting the hypothesis that senescence-associated signals contribute to shaping the local immune microenvironment during post-lactational remodeling (Fig. 5f,g). Collectively, these data suggest that senescent epithelial cells play a key role in recruiting macrophages during involution to ensure a proper involution dynamic.

Involution-associated senescence promotes postpartum tumorigenesis

Next, we set to explore the potential implication of involution-associated senescence in breast cancer. Women diagnosed with breast cancer within 5–10 years postpartum exhibit a worse prognosis compared with age-matched nulliparous women or those diagnosed during pregnancy, independent of hormone receptor status^40,41. It has been proposed that the involuting MG microenvironment promotes breast cancer metastasis^44,63,64. Given senescent cells can extensively rewire the microenvironment to influence cancer progression in a context-dependent manner^7,15,65,66, we investigated whether senescence associated with MG involution contributes to a pro-tumorigenic microenvironment.

To test this hypothesis, we utilized the MMTV-neu mouse model (Fig. 6a). Consistent with existing literature⁶⁴, we observed that a single round of pregnancy and subsequent involution significantly reduced tumor latency (Fig. 6b). Surprisingly, repeated rounds of pregnancy mitigated the impact of parity on tumorigenesis (Fig. 6b). Previous studies suggest that concurrent pregnancy delays the involution process⁶⁷. To examine whether concurrent pregnancy affects involution-associated senescence, we mated MMTV-neu mice and dispatched them into two groups randomly. We removed the male mice from the control group after confirming pregnancy (control group), while we left the male mice with the other group of pregnant female mice (concurrent pregnancy group).

Fig. 6. Involution-induced senescence promotes postpartum tumorigenesis.

a, Experimental design of in vivo experiment with MMTV-neu mice. Mice were mated at the age of 10–12 weeks and allowed to go to lactation and natural involution, 1 or 3–4 times (called 1P or 3–4P for the number of pregnancies). Some of the littermates were maintained virgin to compare the tumor free latency. b, Tumor free latency curve comparing virgin versus 1 pregnancy (1P) versus 3–4 pregnancies (3–4P) from the first date of delivery used as starting point for the virgin littermate. n = 20 for virgin; n = 9 for 1P; n = 6 for 3–4P. Curve comparison Gehan–Breslow–Wilcoxon test was performed. ****P < 0.0001 (virgin versus 1P), ***P = 0.0002 (3–4P versus 1P). c, Representative pictures of MG sections from control or concurrent pregnancy mice at Inv3 and Inv7, stained for SAβGal on cryosection. Scale bar: 100 μm. d, Automatic quantification of SAβGal staining, Inv3 n = 2 mice per group; Inv7 n = 3 mice per group. Data are represented as mean ± s.d. Unpaired t-test was performed. **P = 0.0017. e, Experimental design of in vivo experiment with MMTV-neu mice treated with ABT-263 at 50 mg kg⁻¹ d⁻¹ or vehicle. Mice were force-fed every day for 2 weeks following synchronized involution and tumor development was monitored every week. All mice were harvested when the first of them reached the humane endpoint (9 weeks after synchronized involution). f, Tumor free latency curve comparing 1 pregnancy + vehicle (1P + Veh) versus 1 pregnancy + ABT-263 (1P + ABT-263) from the date of delivery used as starting point. n = 4 for 1P + Veh; n = 4 for 1P + ABT-263. Curve comparison (Gehan–Breslow–Wilcoxon test) was performed. *P =0.0293. g, Experimental design of in vivo experiment with MMTV-neu mice treated with ABT-263 at 50 mg kg⁻¹ d⁻¹ or vehicle. Mice were force-fed every day for 2 weeks following synchronized involution and tumor development was monitored every week. h, Tumor free latency curve comparing 1P + Veh versus 1P + ABT-263 from the date of delivery used as starting point. n = 13 for 1P + Veh; n = 10 for 1P + ABT-263. Curve comparison (Gehan–Breslow–Wilcoxon test) was performed. *P = 0.0118. Illustrations in a, e and g created with BioRender.com. Veh, vehicle.

Source data

Upon delivery, all female mice from both groups established lactation successfully, and we confirmed that female mice kept with the male mice (concurrent pregnancy) conceived again during lactation. We then removed pups from both groups to initiate synchronized involution and collected MGs at Inv3 and Inv7 (Extended Data Fig. 7a). Histological analysis and STAT3 activation confirmed a delayed involution process in the concurrently pregnant female mice⁶⁷ (Extended Data Fig. 7b,c). Notably, we observed a significant reduction in SAβGal⁺ cells in the involuting MGs of concurrently pregnant female mice (Fig. 6c,d). Our animal records further indicated that the multiparous MMTV-neu mice remained pregnant during involution. Thus, we hypothesized that bypassing involution-associated senescence might underlie the diminished impact of parity on tumorigenesis.

Extended Data Fig. 7. Concurrent pregnancy delays involution-induced senescence.

A. Experimental design of in vivo experiment with MMTV-neu mice. Mice were mated at the age of 10-12 weeks and allowed to go through lactation. Synchronized involution was induced at day 10 of lactation by removing the pups. One group was kept with males to enable a concurrent pregnancy before beginning of involution while the second group remain separated from males. Mammary glands were collected at Inv3 and Inv7. B. Representative pictures of mammary gland sections from control or concurrent pregnancy mice at Inv3 and Inv7, stained for hematoxylin and eosin. Scale bar: 100 μm. C. Representative pictures of mammary gland sections from control or concurrent pregnancy mice at Inv3 and Inv7, stained for p-STAT3. Scale bar: 100 μm. D. Representative pictures of SAβGal co-stained with Krt5 (left panel) or Krt8 (right panel) on control or ABT-263 treated mice involuted at Inv5. Scale bar: 50 μm. E. Representative pictures of mammary gland sections from control or ABT-263 treated involuted MMTV-neu females at Inv5, stained for SAβGal on cryosections. Scale bar: 100 μm (left panel). Automatic quantification of SAβGal staining at Inv5 (n = 4 females per group). Data are represented as mean ± standard deviation (right panel). F. Whole mount mammary glands after carmine staining from virgin (14weeks) or involuting females at day 5 (16 weeks). Scale bar: 500 μm. G. Representative pictures of mammary gland tumor from MMTV-neu mice treated or not with ABT-263, 9 weeks after synchronized involution. Illustrations in a created with BioRender.com.

Source data

To test this hypothesis, we treated MMTV-neu mice with ABT-263 for 2 weeks at the beginning of involution (Fig. 6e,g). First, to determine the cellular identity of senescent cells in the MMTV-neu mouse model, we performed immunohistochemistry (IHC) for basal (K5) and luminal (K8) keratins on MGs at Inv5 from both control and ABT-263-treated mice. As observed in wild-type mice, SAβGal⁺ cells predominantly localized to the luminal (K8-positive) epithelial compartment, and ABT-263 treatment markedly reduced the frequency of SAβGal⁺ cells (Extended Data Fig. 7d,e). Importantly, ABT-263 administration significantly reduced the incidence of palpable tumors 9 weeks after synchronized involution and delayed tumor onset in parous mice compared with controls (Fig. 6e–h and Extended Data Fig. 7g), supporting a role of involution-associated senescence in promoting mammary tumorigenesis. Finally, to exclude the possibility that ABT-263 eliminates pre-existing tumor cells, we examined the morphology of MGs from 14–16-week-old nulliparous and involuting MMTV-neu female mice and confirmed the absence of detectable tumors at the time of treatment (Extended Data Fig. 7f).

Involution-associated senescence promotes tumor plasticity via SASP

Next, we sought to elucidate the mechanisms by which involution-induced senescence promotes tumorigenesis. Previously, we showed that senescence promotes cellular plasticity in the context of cellular reprogramming^68,69, and the activation of Rank signaling induces senescence to promote mammary tumorigenesis⁷⁰. Therefore, we investigated whether senescence promotes tumor plasticity by establishing a CM-based tumor organoid system. Specifically, we generated involuting organoids from both control and BlgCre^Tg/+; p16^f/f mice, collecting CM every 2 d for a total of three times after day 14 (day 14, day 16 and day 18) to culture 4T1 tumor organoids derived from virgin Balb/c hosts (Fig. 7a). Notably, tumor organoids cultured with CM from BlgCre^Tg/+; p16^f/f mice exhibited significantly fewer invasive events than those cultured with control CM (Fig. 7b,c). Similar results were obtained using the MMTV-PyMT mammary tumor model (Extended Data Fig. 8a,b), indicating that involution-associated senescence promotes tumor invasion in a paracrine manner.

Fig. 7. Involution-induced senescence promotes tumor invasion.

a, Experimental design of in vitro experiment. After collection, CM from p16^f/f or Blg^Tg; p16^f/f was used to cultivate organoids isolated from 4T1 tumor (transplantation in virgin Balb/c female mice). b, Representative brightfield pictures of 4T1 tumor organoids growing from d0 to d4. c, Quantification of invasion events per organoid at d4, n = 20 tumor organoids per condition. Mann–Whitney test was performed. **P = 0.0031. d, Experimental design of in vivo experiment for 4T1 tumor cell transplantation in Balb/c mice treated with ABT-263 at 50 mg kg⁻¹ d⁻¹ or vehicle. Mice were injected when virgin or at day 2 of involution and force-fed for 10 d. e, Representative pictures of lung sections from control or ABT-263-treated mice 4 weeks after transplantation, stained for hematoxylin and eosin. Scale bar: 1 mm. n = 6 for Virgin + Veh; n = 6 for Virgin + ABT-263; n = 5 for Inv2 + Veh; n = 6 for Inv2 + ABT-263. f, Diagram of lung metastasis frequency in control or ABT-263-treated mice 4 weeks after transplantation. n = 6 for Virgin + Veh; n = 6 for Virgin + ABT-263; n = 5 for Inv2 + Veh; n = 6 for Inv2 + ABT-263. g, Design of metastatic colonies experiment. Lung, liver and brain were collected 4 weeks after 4T1 transplantation. Organs were dissociated to obtain single-cell suspensions and seeded in selection medium containing 6-TG. Representative pictures of metastatic colonies obtained from a lung dissociation culture from control or ABT-263-treated mice, 4 weeks after transplantation, stained with Giemsa. Quantification of the numbers of colonies from the lungs of control mice or mice treated with ABT-263. n = 6 for Inv2 + Veh; n = 6 for Inv2 + ABT-263. Data are represented as mean ± s.d. Paired t-tests were performed. *P = 0.0491. Scale bar: 0.5 cm.

Source data

Extended Data Fig. 8. Involution-induced senescence promotes tumor invasion.

A. Representative brightfield pictures of PyMT tumor organoids growing from d0 to d2. B. Quantification of invasion event per organoid at d2, n = 20 tumor organoids per condition. Unpaired t-tests were performed. p-values ***<0.001. C-D. Representative pictures of metastatic colonies obtained from liver or brain dissociation culture from control or ABT-263 treated mice 4 weeks after transplantation, stained with Giemsa (left panels). Quantification of the number of colonies from control or mice treated with ABT-263 (right panels). Unpaired t-tests were performed. p-values *<0.05. Scale bar: 0.5 cm.

Source data

Involution-associated senescence promotes mammary tumor metastasis in the lung

Finally, to investigate whether involution-associated senescence could promote the tumor metastatic potential in vivo, we adapted a PPBC mouse model focused on the tumor-promoting effects of involution⁶⁴. We injected 4T1 cells into both virgin and involuting (Inv2) Balb/c mice, treating half of the groups with ABT-263 as described above (Fig. 7d). Importantly, we found that ABT-263 treatment significantly suppressed lung macroscopic metastasis in hematoxylin and eosin staining (Fig. 7e,f). Notably, no difference was observed in the virgin groups, indicating that ABT-263 specifically targets senescent cells in the involuting microenvironment without affecting tumor cells directly (Fig. 7f).

To further validate our initial observations, we performed an independent experiment using a clonogenic assay⁷¹ for 4T1 cells, which provides a sensitive measure of metastatic burden (Fig. 7d). Mice were orthotopically injected with 4T1 cells and treated with or without ABT-263 during involution. After 4 weeks, lungs, liver and brain were collected, dissociated and plated in the presence of 6-thioguanine (6-TG), to which 4T1 cells are resistant. This selection-based approach enabled specific detection and quantification of clonogenic metastatic 4T1 cells in distant organs. Strikingly, ABT-263-treated mice displayed a marked reduction in metastatic colonies across multiple organs compared with untreated controls (Fig. 7g and Extended Data Fig. 8c,d). These findings confirm and extend our histological analysis, providing strong evidence that senolytic treatment during involution effectively limits metastatic dissemination. Taken together, these data support the hypothesis that involution-associated physiological senescence enhances the plasticity of tumor-initiating cells to facilitate tumor invasion and metastasis.

Collectively, our study uncovers a pivotal dual role for senescence during postpartum MG involution, demonstrating its importance in both physiological tissue remodeling and tumorigenesis.

Discussion

Here, we report that senescence is transiently induced during postpartum MG involution, a physiological process critical for restoring MG homeostasis after pregnancy. We identify that alveolar luminal cells are the primary cell type undergoing p16-dependent senescence, which is likely driven by systemic hormone withdrawal. Loss-of-function experiments demonstrate that senescence facilitates tissue remodeling processes to ensure proper involution. Unexpectedly, we also find that involution-associated senescence promotes PPBC progression by enhancing tumor-initiating cell plasticity through paracrine signaling. These findings reveal a previously unappreciated dual role for senescence in tissue repair and tumorigenesis, demonstrating how physiological senescence can be exploited by oncogenic events. This highlights the potential for targeting involution-associated senescence as a new therapeutic strategy to mitigate PPBC progression.

Involution-associated senescence shares notable parallels with senescence in decidual stromal cells of the cycling human endometrium⁷², which is required for successful implantation⁷³. Both processes are triggered by fluctuations in female sex hormones, mediated by Trp53 and Cdkn2a upregulation, and play important roles in scarless tissue remodeling. However, their hormonal regulation differs. Brighton et al. showed that progesterone induces differentiation and senescence in human primary endometrial stromal cells during decidualization⁷², while in MG, prolactin withdrawal is required for senescence induction (Figs. 2f,g and 3b). This context-dependent role of sex hormones underscores their importance as potential triggers of senescence. Interestingly, senescence induction in alveolar luminal cells appears to be DNA damage-independent (Fig. 2c,d), consistent with developmental senescence¹⁶. Future work will be needed to delineate the molecular pathways that trigger mammary alveolar luminal cell senescence.

A recent study reported that aged MG stem cells exhibited a senescence-like gene signature⁵¹. Curiously, our study showed that involution-associated senescent cells do not accumulate with age, which suggests that involution-associated senescence is a highly controlled and programmed process, distinct from the stochastic nature of aging-associated senescence. In the future, it could be interesting to know whether senescent cells accumulate in the uterus during the premenopausal period.

Our RNA-seq analysis revealed that involution-associated senescence is linked to upregulation of Cdkn2a and Trp53. Notably, Trp53^−/− and Cdkn2a (p19Arf)^−/− mice exhibit delayed mammary involution, highlighting the importance of these pathways in this process⁷⁴. Importantly, acute deletion of p16 in alveolar luminal cells partially prevented senescence induction and impaired the involution process (Fig. 3). These results suggest that potential compensatory mechanisms may obscure phenotypes in germ-line knockout models. It remains plausible that p19^Arf or other members of the INK4 family also play critical roles in this process, warranting further investigation into their contributions.

To assess the functional relevance of senescence during MG involution, we employed both a genetic approach using BlgCre; p16^f/f mice and a pharmacological strategy with ABT-263. Both models effectively reduced senescent cell burden and delayed involution, but with notable differences. In BlgCre^Tg/+; p16^f/f mice, effects became evident at Inv5, whereas ABT-263 reduced senescence markers by Inv3, although visible changes in alveolar regression and adipocyte repopulation appeared only at Inv4. These differences likely reflect the distinct modes of action: the genetic model selectively prevents p16^Ink4a-dependent senescence induction in alveolar luminal cells, while ABT-263 broadly eliminates senescent cells via apoptosis, regardless of the molecular mediator. Interestingly, a significantly increased adipocyte number was observed only in BlgCre^Tg/+; p16^f/f mice, suggesting a specific role for p16-positive senescent cells in regulating adipogenesis.

Although the absence of a reliable murine p16Ink4a antibody limited direct assessment after ABT-263 treatment (Supplementary Note), reduced SAβGal staining supports effective senescent cell clearance. However, we recognize several limitations of this pharmacological approach. First, we cannot fully exclude potential pleiotropic effects of ABT-263 or its impact on nonsenescent cell types, which may contribute to differences between models, particularly regarding adipocyte remodeling. Second, in our study, ABT-263 was administered once daily for 3 d, beginning at involution day 0.5, 1.5 or 2.5, with samples collected at days 3, 4 and 5, respectively. Across all regimens, senescence load was consistently reduced. Since senescence is induced around day 3, these findings cannot exclude the possibility that ABT-263 also interferes with senescence induction in addition to clearance, and the clearance kinetics may further depend on drug availability during this window. Finally, as a systemic drug, ABT-263 may exert off-target effects in other organs that indirectly influence MG remodeling. Together, these considerations highlight distinctions between pharmacological and genetic models of senescent cell depletion, and the importance of complementary strategies to unravel the diverse functions of senescent cells in a physiological context. Future studies are needed to refine the temporal and cell type specificity of senolytic treatments.

Developmental and programmed senescence, such as observed during embryonic development^8,9,75, the female reproductive cycle^72,73 and postpartum involution, including MG and uterus⁷⁶, occur within immune-privileged microenvironments. However, it remains unclear whether senescence actively establishes this immunosuppressive state. Our GSEA revealed enrichment of pathways regulating immune cell chemotaxis, leading us to propose that senescent epithelial cells recruit macrophages during the irreversible phase of involution. Further characterization of the immune landscape will provide critical insights into the specific role of senescence in establishing immune suppression during postpartum tissue remodeling. Additionally, while both natural killer cells⁷¹ and macrophages⁷⁶ are known to eliminate senescent cells in physiological contexts, it remains unknown which immune cells are required for regulating the fate of involution-associated senescence. Understanding these interactions will be pivotal in unraveling the immune-modulatory roles of senescence in postpartum tissue remodeling, which will have direct implications in tumorigenesis.

Finally, we demonstrated that the physiological senescence program can be exploited by oncogenic processes. In this study, we specifically focused on paracrine senescence in promoting tumor-initiating cell plasticity. Given the extensive body of research highlighting the distinctive features of involution^41–43, it would be important to investigate whether and how senescence contributes to the establishment of a pro-tumorigenic microenvironment. Recent studies highlight shared immunosuppressive mechanisms between pregnancy and oncogenesis⁷⁷, and senescent cancer-associated fibroblasts have been shown to mediate immune suppression to facilitate cancer progression^65,78. These findings suggest potential parallels between the roles of senescence in physiological tissue remodeling and tumorigenesis, warranting further investigation.

Methods

Mice and experimental procedures

Experiments involving animals were approved in accordance with French legislation in compliance with European Communities Council Directives (A 75-15-01-3) and the regulations of Institut Pasteur Animal Care Committees (Comité d’Éthique en Expérimentation Animale (CETEA)). The study was performed by certified individuals and carried out in accordance with the principles of the Basel Declaration.

Animals were housed in the Institut Pasteur animal facilities accredited by the French Ministry of Agriculture for performing experiments on live rodents. Work on animals was performed in compliance with French and European regulations on care and protection of laboratory animals (EC Directive 2010/63, French Law 2013-118, 6 February 2013). All experiments were approved by the Ethics Committee no. 89 (CETEA Institut Pasteur). Only female mice were used in this study.

For experiments involving tumor-bearing mice, animals were euthanized upon reaching humane endpoints in accordance with applicable regulations, with tumors not exceeding 1,500 mm³ in volume.

Wild-type inbred strain C57BL/6J mice were bred at the Monod Animal Facility of the Institut Pasteur. INKBRITE mice⁵⁴ were kindly provided by Dr Tien Peng (University of California, San Francisco) and bred in a C57BL/6 genetic background. BlgCre mice⁷⁹ were obtained from The Jackson Laboratory (strain no. 012620) and bred in a mixed genetic background, C57BL/6; 129. p16-floxed mice⁸⁰ were obtained from the MMRRC (strain no. 043540-UNC) and bred in a mixed genetic background, C57BL/6; 129. BlgCre mice were crossed with Cdkn2a (p16Ink4a)-floxed mice to delete p16^INK4a specifically in Blg-expressing cells. These mice are named BlgCre^+/Tg; p16^f/f in this manuscript. Cdkn2a (p16Ink4a)Cre^ERT2 mice⁵⁵ were kindly provided by Dr Makoto Nakanishi (University of Tokyo, Japan) and bred in a C57BL/6 genetic background. Rosa26^mTomato/mGFP mice⁸¹ were obtained from The Jackson Laboratory (strain no. 007576) and bred in a mixed genetic background, C57BL/6; 129. Cdkn2a (p16Ink4a)Cre^ERT2 mice were crossed with Rosa26^mTomato/mGFP mice to mark and trace p16⁺ cells. These mice are named p16Cre^ERT2; R26^mT/mG in this paper. MMTV-neu mice⁸² were obtained from The Jackson Laboratory (strain no. 005038) and bred in an FVB/N genetic background. For the 4T1 tumor cell transplantation, Balb/c mice were obtained from Charles River Laboratories. For part of the tumor organoid invasion experiments, MMTV-PyMT mice⁸³ were obtained from The Jackson Laboratory (strain no. 002374) and bred in an FVB/N genetic background. All animals were bred from the Central Animal Facility of the Institut Pasteur under standard conditions at the C2RA, Institut Pasteur; specific values for temperature, humidity and light cycle were those routinely maintained by the facility.

Kinetics

Tissues from pubertal mice were collected at 6 weeks old; tissues from virgin mice were collected at 8–10 weeks old (Y-virgin) or 12–14 months old (O-virgin); tissues from pregnant mice were collected at p13.5 and p18.5; tissues from multiparous mice were collected at the resting stage from female mice coming from breeding cages, after 3–4 rounds of pregnancy. For the involution experiments, adult virgin female mice were mated between 8 and 10 weeks old. To homogenize lactation stimuli, litters were evenly distributed to six pups per female mouse after birth and fed by the mother for 10 d. Then, pups were physically separated to allow a synchronized involution of MGs. Only female mice establishing proper lactation were analyzed after synchronized involution. The day of removal of the progeny was considered as the beginning of involution, also called Lac10 or Inv0. The involution process was left to progress from Inv0 to Inv28 or resting stage.

ABT-263 treatment

ABT-263 (Tebu cat. no. T2101)¹⁸ was initially resuspended in DMSO at 100 mg ml⁻¹, aliquoted and stored at −20 °C for <1 month. ABT-263 was then freshly diluted by adding forced feeding mixture (60% Phosal; 30% PEG400; 10% EtOH) of vehicle (same mix with DMSO only) upon use. Mice were force-fed by oral gavage daily with ABT-263 at 50 mg kg⁻¹ for 1–3 d before MG collection or for 10–14 d in tumor development experiments.

Sealing

Teat sealing was performed as previously described²⁶. Briefly, after delivery, female mice were allowed to lactate for 10 d. Hair remover was applied to clean the nipple area of MGs before sealing. After manually restraining the animal, a 2–3-μl drop of veterinary tissue adhesive (3M Vetbond cat. no. 1469C) was applied on the nipple and left to dry before release of the mouse in the cage with the litter. The manipulation was repeated 3 h later to ensure a full sealing of the nipple. Maintenance of lactation on counterpart open teats was checked every day. MGs were collected 2 or 3 d after teat closure.

Concurrent pregnancy

To allow a concurrent pregnancy, female mice were kept with male mice after delivery. Only female mice that were pregnant again the first week of lactation were considered for concurrent pregnancy experiments.

Whole-mount analyses, histology and immunolabeling

For morphological analyses, mammary fat pads were spread on glass slides and fixed overnight in a solution of methacarn (60% methanol; 30% chloroform; 10% acetic acid). MGs were washed twice in EtOH 70% and once in H₂O and incubated overnight in Carmine Alum stain (Stem Cell Technologies cat. no. 07070) at room temperature. Then, MGs were washed in H₂O followed by EtOH 70% and processed for embedment in paraffin. During the second clearing with xylenes, pictures of whole-mount stained MGs were taken using a binocular magnifier.

For histological analyses, MGs, tumors or organs were fixed overnight in formalin (or methacarn as previously described) and embedded in paraffin. Then, 5-μm-thick sections were cut and de-waxed for hematoxylin/eosin coloration or immunolabeling (IHC or immunofluorescence). When specified, an antigen retrieval was performed using a retriever (Electron Microscopy Sciences cat. no. 2100). For immunochemistry, an inhibition of endogenous peroxidase was performed using H₂O₂ 1% during 10 min before a 1-h blocking step. The co-staining with SAβgal (IHC) was performed using SAβgal staining of MG samples. For immunofluorescence analyses on cryopreserved samples, slides of 10-μm-thick OCT cryopreserved MGs were removed from −80 °C storage and dried at room temperature for 30 min. Samples were fixed in a fixation solution of PFA 4% (Electron Microscopy Sciences cat. no. 15714) in PBS at room temperature for 15 min before the immunofluorescence (see below).

The primary antibodies used are listed in Supplementary Table 1.

SAβgal staining

For SAβgal staining on whole-mounts, MGs or organoids were incubated in a fixation solution of PFA 2% (Electron Microscopy Sciences cat. no. 15714) and glutaraldehyde 0.2% (Sigma cat. no. 49629) in PBS at room temperature during, respectively, 30 min or 15 min, on a swing. After brief washes in PBS, MGs were incubated overnight (only 6 h for the organoids) at 30 °C in the SAβgal solution, followed by washes in PBS, and post-fixed during 8 h in 10% formalin on a swing. Then, MGs were briefly washed in PBS, dehydrated and embedded in paraffin. Organoids were embedded in 3% low-gelling-temperature agarose. After solidification, samples were dehydrated before paraffin embedding. Tissue sections (5 μm thick) were then counterstained with Nuclear Fast Red (Vector cat. no. H-3403) for 2–5 min, dehydrated and mounted with Eukitt (Sigma cat. no. 49629) or used for IHC co-staining.

For SAβgal staining on cryosections, sections of 10 μm (cryopreserved MGs) in OCT (Leica cat. no. 14020108926) were removed from −80 °C storage and dried at room temperature for 30 min. Samples were fixed in a fixation solution of PFA 2% (Electron Microscopy Sciences cat. no. 15714) and glutaraldehyde 0.2% (Sigma cat. no. 49629) in PBS at room temperature during 15 min. After three PBS washes of 5 min, MGs were incubated for 6 h at 30 °C in an SAβgal solution. After incubation, samples were washed in PBS, post-fixed in PFA 4% (Electron Microscopy Sciences cat. no. 15714) for 15 min and washed again in PBS. Tissue sections were then counterstained with Nuclear Fast Red (Vector cat. no. H-3403) for 2–5 min, and dehydrated by one 5-min bath of EtOH 95% and two 5-min EtOH 100% baths before being mounted with Eukitt (Sigma cat. no. 49629).

SAβgal solution contains: 40 mM citrate buffer pH 6.0 (VWR cat. no. 28027; Sigma cat. no. C7129); 5 mM K₃Fe(CN)₆ (Sigma cat. no. P8131); 5 mM K₄Fe(CN)₆ (Sigma cat. no. P9387); 2 mM MgCl₂ (Sigma cat. no. M8266); 150 mM NaCl (Sigma cat. no. 31434); 0.1% NP40 (Sigma cat. no. I8896); 0.5 mg ml⁻¹ X-gal (Roche cat. no. 10703729001); and ultrapure Milli-Q water.

Isolation of primary mammary epithelial organoids

Primary mammary organoids were prepared from a pool of thoracic and inguinal MGs (without visible lymph node) isolated from at least three 8–12-week-old female mice as previously described^56,84.

Briefly, the mice were euthanized by cervical dislocation or CO₂ exposition. The MGs were removed and homogenously minced into approximately 1-mm³ pieces before being transferred to a 50-ml tube of digestion solution containing 2 mg ml⁻¹ collagenase A (Roche cat. no. 11088793001), 2 mg ml⁻¹ trypsin (Dutcher Dominique cat. no. P10-022100), 5% fetal bovine serum (FBS) (Gibco cat. no. 10270-106), 5 μg ml⁻¹ insulin (Sigma cat. no. I6634-100MG), 50 μg ml⁻¹ gentamicin (Sigma cat. no. G1397) and 1 × GlutaMAX (Gibco cat. no. 35050-038) in DMEM/F12 (Gibco cat. no. 21331-020), and incubated for 30–40 min at 37 °C with shaking at 100 rpm.

The resulting pellets of digested samples were centrifuged and successively treated with solutions of 0.5 mg ml⁻¹ dispase II (Roche cat. no. 13 75 2000) and 100 μg ml⁻¹ DNase I (Sigma cat. no. D4527-40KU) for 3–5 min at room temperature and exposed to five rounds of differential centrifugation at 450g for 10 s, which resulted in separation of epithelial (organoid) and stromal fractions. The organoids were resuspended in BOM containing 1 × GlutaMAX (Gibco cat. no. 35050-038), 1 × Insulin-Transferrin-Selenium (Gibco cat. no. 41400-045) and 1 × penicillin/streptomycin (Gibco cat. no. 15140-122) in DMEM/F12 (Gibco cat. no. 21331-020) and kept on ice for quantification on a slide.

Three-dimensional culture of mammary organoids

Isolation and culture

Freshly isolated primary mammary organoids were resuspended in an appropriate volume of growth factor-reduced Matrigel (BD Biosciences cat. no. 354230) and plated. To prepare the cell culture plate, we used a small volume of Matrigel (10–20 μl per well) to cover the central part of the bottom of wells of a 24-well cell culture plate. We then incubated the plate at 37 °C for 15 min. Then, we seeded, respectively, 200 or 400–500 organoids per drop for histology and gene expression analysis (50 μl of Matrigel per drop). We incubated the plate for 30–60 min in an incubator at 37 °C, 5% CO₂. Finally, we added pre-warmed BOM to each well to cover the Matrigel drop. For organoid growing, we added 2.5 nM FGF₂ (Gibco cat. no. PMG0034) in BOM medium for 6 d. We changed medium every 3 d. To induce lactogenic differentiation, we added 1 μg ml⁻¹ prolactin (Sigma cat. no. SRP4688) + 1 μg ml⁻¹ hydrocortisone (Sigma cat. no. H6909) in BOM medium during 4–14 d. We changed LM every 2 d. To induce an involution-like process, we cultivated organoids in only BOM and changed the medium every 2 d.

CM collection

For tumor invasion experiments, CM, containing soluble proteins and exosomes/extracellular vesicles, was collected from organoid culture at day 16 and day 18 of the involution-like process from BlgCre^+/+; p16^f/f (control) and BlgCre^Tg/+; p16^f/f organoids. After 5 min of centrifugation at 400g to acquire the supernatant, aliquoted CM was stored at −20 °C.

For the macrophage migration assay (see below), wild-type mammary organoids were isolated and cultivated as described previously in FGF₂ (growing), prolactin (lactation-like) or BOM (involution-like). At 24 h before the beginning of the assay, all media were replaced by BOM only and CM from those three organoid stages was collected, centrifuged at 400g and used freshly or stored at −20 °C.

Macrophage migration assay

BMDM culture

Total bone marrow was obtained by flushing the femur and tibia from wild-type C57BL/6J mice with DMEM + GlutaMAX (Gibco cat. no. 11594446). Cells were cultured in DMEM + GlutaMAX containing 20% heat-inactivated FBS (Gibco cat. no. 10270-106), 30% of CM derived from the L929 cell line (European Collection of Authenticated Cell Cultures; enriched in CSF-1) and 1 × penicillin/streptomycin (Gibco cat. no. 15140-122) for 6–7 d.

Transwell assay

Macrophages from BMDM were collected in DMEM/F12 (Gibco cat. no. 21331-020) and plated on Transwells (50,000 cells per Transwell (Corning cat. no. 3422)). The bottom parts of Transwells were filled with CM from organoids cultured with FGF₂ (Growing), prolactin (lactation-like) or BOM (involution-like). After 8 h, Transwell membranes were fixed with 4% PFA for 15 min, and permeabilized for 10 min in Triton 0.5%. The upper surfaces of the Transwells were scraped to remove the macrophages that did not migrate, and the bottom parts were stained with DAPI (1:1,000, Sigma-Aldrich cat. no. D9542) for 10 min. Then, Transwell membranes were cut and mounted in Immu-mount (Fisher Scientific cat. no. 10662815) between a slide and a coverslip, and images were acquired using an Olympus IX83 microscope and quantified using ImageJ software.

Isolation and culture of mammary tumor organoids

Isolation

Primary mammary tumor organoids were prepared from a tumor, originated from transplanted 4T1 cells in a nulliparous Balb/c mouse or from MMTV-PyMT mice, as previously described⁸⁵.

Briefly, the mice were euthanized by cervical dislocation or CO₂ exposition. The mammary tumor was removed and homogenously minced into approximately 1-mm³ pieces before being transferred to a 50-ml tube of digestion solution containing 2 mg ml⁻¹ collagenase A (Roche cat. no. 11088793001), 2 mg ml⁻¹ trypsin (Dutcher Dominique cat. no. P10-022100), 5% FBS (Gibco cat. no. 10270-106), 5 μg ml⁻¹ insulin (Sigma cat. no. I6634-100MG), 50 μg ml⁻¹ gentamicin (Sigma cat. no. G1397) and 1 × GlutaMAX (Gibco cat. no. 35050-038) in DMEM/F12 (Gibco cat. no. 21331-020), and incubated for 1 h at 37 °C with shaking at 180 rpm.

The resulting pellets of digested samples were centrifuged and treated with a solution of 100 μg ml⁻¹ DNase I (Sigma cat. no. D4527-40KU) for 3–5 min at room temperature and exposed to four rounds of differential centrifugation at 400g for 3 s, which resulted in separation of epithelial (organoid) and stromal fractions. The organoids were resuspended in BOM and counted.

Culture

Freshly isolated mammary tumor organoids were resuspended in an appropriate volume of 1:1 growth factor-reduced Matrigel (BD Biosciences cat. no. 354230) and neutralized Collagen Type I (Corning cat. no. 354236) and plated. To prepare the cell culture plate, a small volume of Matrigel (10–20 μl per well) was used to cover the central part of the bottom of wells of a 24-well cell culture plate. The plate was incubated at 37 °C for 15 min. Then, 100 organoids per drop were used for time-lapse imaging (50 μl of the mix of Matrigel/Collagen I per drop). The plate was incubated for 30–60 min in an incubator at 37 °C, 5% CO₂ to allow polymerization of the Matrigel/Collagen I drops. Finally, pre-warmed CM, collected from primary organoid cultures at day 16 of the involution-like process from BlgCre^+/+; Cdkn2a (p16Ink4a)^f/f (control) and BlgCre^Tg/+; Cdkn2a (p16Ink4a)^f/f organoids, complemented with 2.5 nM FGF₂ (Gibco cat. no. PM60034), was added in each well to cover the Matrigel/Collagen I drop for 4 d.

Primary mammary epithelial cell preparation and flow cytometry analysis

Mammary epithelial cell isolation

To isolate mammary epithelial cells, mammary fat pads from Lac10 (or Inv0) and Inv5 female mice were collected and mechanically dissociated with scissors and scalpels before an enzymatic digestion for 1.5 h at 37 °C, with shaking at 140 rpm, in a solution containing 3 mg ml⁻¹ Collagenase A (Roche cat. no. 11088793001) and 100 U ml⁻¹ hyaluronidase (Sigma cat. no. H3884) in CO₂-independent medium (Gibco cat. no. 18045-054), complemented with 5% FBS (Gibco cat. no. 10270-106), 1 × GlutaMAX (Gibco cat. no. 35050-038) and 1 × penicillin/streptomycin (Gibco cat. no. 15140-122). Digested tissues were centrifuged for 5 min at 400g and the supernatant was discarded. After a wash with CO₂-independent medium and a centrifugation, each pellet was resuspended with 2 ml of pre-warmed PBS + 0.25% trypsin (Dutscher cat. no. P10-022100) and 0.1% EDTA Versen (Biochrom cat. no. L2113) for 1 min at room temperature. After a second wash in CO₂-independent medium + 5% FBS and a centrifugation, each pellet was resuspended in 2 ml of CO₂-independent medium + 5% FBS containing 5 mg ml⁻¹ of dispase II (Roche cat. no. 13752000) and 100 μg ml⁻¹ DNase I (Sigma cat. no. D4527-40KU) and incubated for 5 min at 37 °C. After a third wash in CO₂-independent medium + 5% FBS and a centrifugation, pellets were resuspended in 2 ml of cold ammonium chloride solution (Stem Cell Technologies cat. no. 07800) and immediately centrifuged. After a resuspension in CO₂-independent medium, the cell solution was filtered through a nylon mesh cell strainer with 40-μm pores (Corning cat. no. 352340) before immunolabeling and cell sorting.

Immunolabeling

Freshly dissociated mammary cells were incubated at 4 °C for 30 min with antibodies. The following conjugated antibodies were used: anti-CD45-APC 1:100 (BioLegend cat. no. 103112, clone 30-F11); anti-CD31-APC 1:100 (BioLegend cat. no. 102510, clone MEC13.3); anti-CD24-BV421 1:50 (BioLegend cat. no. 101826, clone M1/69); anti-α6-PE.Cy7 1:50 (BioLegend cat. no. 313622, clone GoH3); anti-CD64-Alexa Fluor 647 1:100 (BD Pharmingen cat. no. 558539, clone X54-5/7.1).

After incubation, cells were washed twice with CO₂-independent medium. Labeled cells were sorted using a MoFlo Astrios EQ (Beckman Coulter) and data were analyzed with FlowJo software.

C₁₂FDG staining

Before immunolabeling, C₁₂FDG staining was performed on an isolated mammary cell suspension following the C₁₂FDG protocol previously described to stain senescent cells⁸⁶. Following supplier guidelines, Bafilomycin A1 (Tebu cat. no. 21910-2060) was resuspended in DMSO to a concentration of 0.1 mM and stored at −20 °C and C₁₂FDG (Fisher cat. no. 11590276) was resuspended in DMSO at a concentration of 20 mM and stored at −20 °C. After mammary cell dissociation, pellets of isolated cells were resuspended in 1 ml of CO₂-independent medium (Gibco cat. no. 18045-054). To neutralize the acidic pH of lysosomes, cells were treated with 100 nM Bafilomycin A1 for 1 h at 37 °C, 5% CO₂. Upon use, C₁₂FDG 20 mM stock was first diluted with DMSO to a temporary stock of 2 mM in fresh culture media and added to the cell solution at a concentration of 33 µM. Cells were incubated for 2 h at 37 °C, 5% CO₂ (Eppendorf tubes were manually inverted every 30 min to homogenize the solution). After incubation, cells were washed twice with CO₂-independent medium and centrifuged for 5 min at 400g. Resuspended pellets were then stained with antibodies compatible with the green FITC-emitted signal (wavelength 488 nm) from C₁₂FDG.

4T1 transplantation experiment

For the tumor transplantation experiment, 10,000 4T1 cells were injected (50 μl) subcutaneously into the mammary fat pad of virgin or Inv2 Balb/c female mice. 4T1 cells were resuspended in PBS. Mice were monitored twice a week and harvested 4 weeks after transplantation.

Quantification of distant-site metastasis

Taking advantage of the intrinsic 6-TG resistance of 4T1 tumor cells, we assessed metastatic burden in various organs (lung, liver, brain) by explanting the tissues, dissociating the cells, plating them in medium containing 6-TG and subsequently quantifying the number of clonogenic 6-TG-resistant tumor cells according to the protocol in ref. ⁷¹. Briefly, 4 weeks after 4T1 transplantation, mice were euthanized by cervical dislocation or CO₂ exposition. The different organs (lung, liver and brain) were collected, washed in HBSS (Gibco cat. no. 14025-092) and homogenously minced into approximately 1-mm³ pieces before being transferred to a 50-ml tube of digestion solution containing: (1) 2 mg ml⁻¹ Collagenase I (Sigma cat. no. SCR103), 2 mg ml⁻¹ hyaluronidase (Sigma cat. no. H3884) and 25 μg ml⁻¹ DNase I (Sigma cat. no. D4527-40KU) in HBSS (Gibco cat. no. 14025-092), and incubated for 45 min to 1 h at 37 °C with shaking at 140 rpm, for liver dissociation; or (2) 2 mg ml⁻¹ Collagenase A (Roche cat. no. 11088793001), 1 mg ml⁻¹ elastase (Worthington cat. no. LS002292) and 25 μg ml⁻¹ DNase I (Sigma cat. no. D4527-40KU) in HBSS (Gibco cat. no. 14025-092), and incubated for 1.5 h to 2 h at 37 °C with shaking at 140 rpm, for lung and brain dissociation.

Digested tissues were centrifuged for 5 min at 500g and the supernatant was discarded. Pellets from liver were resuspended in 3 ml of cold ammonium chloride solution (Stem Cell Technologies cat. no. 07800) and immediately centrifuged. After three washes with HBSS medium (Gibco cat. no. 14025-092) and centrifugation, each pellet was resuspended with complete IMDM + GlutaMAX medium (Gibco cat. no. 31980-030) supplemented with 10% FBS (Gibco cat. no. 10270-106), 1 × penicillin/streptomycin (Gibco cat. no. 15140-122) and 60 μM 6-TG. Cell solutions were filtered through a nylon mesh cell strainer with 100-μm pores (Corning cat. no. 352360) and seeded in p100 tissue culture dishes (3–5 plates per organ) for 2 weeks.

After 2 weeks, clonogenic metastatic colonies were fixed by adding methanol for 5 min, washed with water and stained with 10% diluted Giemsa stain (Sigma cat. no. 32884) for 20 min. After washes with distilled water, plates were left to dry before scanning and quantification using Ilastik and Fiji softwares.

RNA isolation and real-time qPCR

Total RNA from mammary tissue or organoid samples was extracted, respectively, with Trizol Reagent (Invitrogen cat. no. 15596026) or RNeasy Micro Kit (Qiagen), following manufacturer instructions. Reverse transcription was performed using a High-capacity cDNA reverse transcription kit (Thermo Fisher Scientific). Quantitative real-time PCR was performed using 5 ng of complementary DNA and 5 pmol each of the forward and reverse gene-specific primers in LightCycler SYBR Green I Master mix (Roche) on a LightCycler 480 II (Roche). All reactions were performed at least in duplicate and in a total of at least three independent assays. Relative gene expression was calculated using the ΔΔC_t method and the values were normalized to the housekeeping gene Gapdh. All primers were ordered from Sigma-Aldrich and were either previously described in publications or newly designed. Their specificity and efficiency were verified before use. The primer sequences (5′–3′) used are listed in Supplementary Table 2.

RNA-seq data analysis

Quality control and preprocessing

Raw sequencing reads were subjected to quality control using FastQC (v.0.11.9). Low-quality bases and adapter sequences were trimmed using TrimGalore (v.0.6.7) with the following parameter: --stringency 4. Reads passing the quality filters were retained for further analysis.

Quantification of gene expression

Gene expression levels were quantified using RSEM (v.1.3.3) against the Mus musculus reference genome, GRCm39.

Differential gene expression analysis

Differential expression analysis was performed using edgeR v.3.28.1⁸⁷ in R v.3.5 (https://www.r-project.org/). Genes with an adjusted P value (false discovery rate using Benjamini–Hochberg method) < 0.05 and a |log₂(fold change) | > 1 were considered significantly differentially expressed.

Functional enrichment analysis

To identify enriched biological pathways, normalized counts were subjected to pathway enrichment analysis via GSEA using the official GSEA software v.4.3.3 from the Broad Institute^21,88. In total, 1,000 permutations using the ‘gene_set’ permutation type parameter were performed for the HALLMARK⁸⁹; WikiPathways⁹⁰; REACTOME⁹¹; Gene Ontology – Biological Process, Cellular Component and Molecular Function^92,93; and SAUL_SEN_MAYO⁴⁵ gene sets. All gene sets were retrieved from the mouse Molecular Signatures Database⁹⁴. Gene sets with an adjusted P value (false discovery rate) < 0.05 and a |normalized enrichment score | > 1 were considered significantly different.

Visualization and data interpretation

Enrichment plots were generated by the GSEA software^88,95 using the parameters described above. Principal component analysis plots for the first two principal components and bar plots for significant HALLMARK pathways were generated using ggplot2 v.2.1.0 (https://ggplot2.tidyverse.org/) in R v.3.5 (https://www.r-project.org/). The heatmaps were manually redrawn in Excel (https://www.microsoft.com/en-gb/microsoft-365/excel/?legRedir=true) to visualize the core enriched genes, using relative expression values for better clarity and presentation.

Statistics and reproducibility

Statistical analysis was performed using the GraphPad Prism software, and statistical tests used are specified in the figure legends. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. The number of independent biological replicates is indicated as n. Data distribution was formally tested, and appropriate statistical analysis was performed.

No statistical methods were used to pre-determine sample sizes; no data were excluded from the analyses; no method of randomization was used to assign animals to experimental groups; the investigators were not blinded to allocation during experiments and outcome assessment. Data collection and analysis were not performed blind to the conditions of the experiments.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

A.C., E.C. and L.D. performed most of the experimental work and contributed to most of the experimental design, data analysis and discussions. Shuoyang Wang, L.T. and M.S. made critical experimental contributions. Shaoxiang Wangand J.C. performed the bioinformatic analysis. H.L. supervised the study, designed the experiments and interpreted the data together with A.C., and wrote the manuscript. All authors discussed the results and commented on the manuscript.

Peer review

Peer review information

Nature Aging thanks Victoria Findlay and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

The bulk RNA-seq data have been deposited in the Gene Expression Omnibus (GEO; accession number GSE285005). Raw data underlying all figures are provided as source data files. All other data supporting the findings of this study are included within the article and its Supplementary Information, and any additional clarifications are available from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s43587-025-01058-y.

Supplementary information

The online version contains supplementary material available at 10.1038/s43587-025-01058-y.