Runx1 and Runx2 act in concert to suppress Wnt/β-catenin-driven mammary tumourigenesis Authors: Alessandra I. Riggio, Kerri Sweeney, Robin Shaw, Amy Lawlor, Adiba Khan, Nicola Ferrari, Jayanthi Anand, Laura CA Galbraith, Kathryn Gilroy, Courtney Bull, Alex L. Young, Dimitris Athineos, Holly Hall, Farah Ghaffar, Mark Hughes, Claire A. Mitchell, Louise Mitchell, Colin Nixon, Peter D. Adams, Edward W. Roberts, Crispin J. Miller, Philip D. Dunne, Kirsteen J. Campbell, Ewan R. Cameron, Karen Blyth Published: 2026 ABSTRACT IntroductionThe genes encoding RUNX1 and its binding partner CBFβ are recurrently reported to be mutated in breast cancer, a major cause of mortality in women worldwide. However, the functional role for these proteins remains unproven.MethodsThe putative tumour suppressor role of Runx1 was investigated in genetic mouse models of breast cancer. Stem cell assays, immunohistochemistry and RNAseq analyses were applied to study biological and molecular mechanisms.ResultsRunx1 loss of function leads to accelerated disease onset and tumour development in breast cancer models. Combined deletion of Runx1 and Runx2 further resulted in mammary cells becoming exquisitely sensitive to WNT-driven transformation, with expedited emergence of multiple tumours. Runx1 ablation induces a stem cell-like phenotype in mammary epithelial cells, whilst transcriptomic analysis demonstrated activation of multiple oncogenic pathways, especially when Runx2 was co-deleted. Altered Runx expression in the mammary epithelium also drove alterations in the tumour immune microenvironment, with changes to neutrophil and macrophage populations.ConclusionsRunx1 restricts some forms of breast cancer and inhibits the full oncogenic potential of aberrant WNT signalling. Combined Runx1 and Runx2 loss dramatically accelerates disease progression suggesting that Runx2 can substitute for Runx1 in dampening the oncogenic effects of WNT signalling. Introduction Breast cancer is the commonest malignant disease affecting women and, despite significant improvement in survival rates, remains a major cause of mortality [1]. Rather than a single disease, breast cancer is a heterogeneous group of tumours defined by anatomical origin and classified by histopathological, clinical and molecular features [2]. Based on receptor status, breast tumours are categorised into three subtypes: hormone receptor positive (oestrogen receptor, ER; progesterone receptor, PR), human epidermal growth factor receptor type 2 (HER2) positive and triple negative (TN, ER-/PR-/HER2-) [3]. Molecular profiling has helped stratify these distinct diseases into separate intrinsic subtypes that largely, but not entirely, map to the well-established disease categories [4–6]. While clinically useful, it is oversimplistic to assume each of the histological subtypes is always defined by a common genetic landscape and indeed, molecular heterogenicity exists within well characterised types of breast cancer. The RUNX genes (RUNX1, -2 and -3) encode transcription factors that bind to the common co-factor CBFβ and modulate expression of many target genes, either directly through binding to the RUNX DNA consensus sequence or indirectly through protein-protein interactions with other partners [7]. The emergence of these genes as players in breast cancer is evident [8], yet their functionality in tumour development and progression is complex and contradictory [9, 10]. This is particularly true for RUNX1 where changes in gene expression have unpredictable outcomes on cell survival, differentiation and proliferation. Large-scale mutational and expression studies indicate that loss of RUNX1/CBFB is associated with a significant proportion of breast cancers. RUNX1 is a frequent mutational target, with genetic changes predicted to interfere with its DNA binding activity or binding to CBFβ, resulting in reduced function [11–13]. CBFB is also frequently mutated, supporting the notion that RUNX1 acts as a tumour suppressor in breast cancer [11, 12]. These putative loss-of-function mutations in RUNX1 and CBFB were overrepresented in luminal breast cancers, suggesting that the RUNX1/CBFβ complex may restrict some subtypes [7]. Mutational studies are supported by mRNA expression profiling and protein levels, where reduced RUNX1 is associated with breast cancer metastasis and high-grade tumours [14, 15]. In contrast to its tumour suppressor role, transcriptome analyses showed RUNX1 upregulation in some subtypes [16, 17] and enhanced protein expression is associated with poor outcome in TN tumours [18]. We, and others, have previously demonstrated that Runx2 (but not Runx3) is expressed in the mammary epithelium, although at lower levels than Runx1 [19, 20]. Importantly, loss-of-function mutations of CBFB would ablate both RUNX1 and RUNX2 activity. Although the role of RUNX2 in breast cancer has not been fully explored, it has been associated with pro-oncogenic properties [19] and induction of epithelial to mesenchymal transition (EMT)-like changes and bone metastasis [20–24]. There are, however, reports where RUNX2 overexpression was found to suppress colony formation of MCF7 cells [24]. We previously showed that RUNX2 was important for mammary reconstitution and stem cell function, whereby genetic deletion resulted in reduced potential to regenerate mammary glands and impaired capacity to form mammospheres in vitro [25]. Therefore, with an obvious yet unresolved role in breast cancer, investigation of Runx genes in physiologically relevant models is warranted. As genetically engineered mouse models (GEMMs) have proved useful in dissecting the interaction of oncogenic pathways [26], we sought to assess the putative tumour suppressor functions of Runx1, in vivo. Loss of Runx1 accelerated disease in specific oncogenic backgrounds, including a model with activated WNT signalling, an important alteration in breast cancer. Components of the WNT pathway, including β-catenin, are frequently overexpressed across breast cancer subtypes, associated with a stem-like phenotype and poorer outcomes [27–29]. We show functional overlap between Runx1 and Runx2 in restricting Wnt/β-catenin-driven tumorigenesis, with Runx-deficient β-catenin-expressing mammary cells being exquisitely prone to tumour formation. Runx loss increases stemness of mammary cells and unleashes the oncogenic potential of deregulated WNT/β-catenin signalling. Furthermore, loss of Runx in the epithelium alters the tumour microenvironment. Results Runx1 loss facilitates early tumour onset in mouse models of mammary cancer To investigate the possible tumour suppressor role of RUNX1 in breast cancer, we deleted Runx1 in preclinical mouse models. MMTV-PyMT is a well characterised breast cancer model [30], transcriptionally aligning with the luminal B subtype [31] and exhibiting early mammary hyperplasia progressing to highly aggressive tumours. To explore Runx1 loss on PyMT-driven tumourigenesis, we generated MMTV-PyMT;MMTV-Cre;Runx1fl/fl (hereafter PyMT + /R1KO) mice and MMTV-PyMT;MMTV-Cre;Runx1wt/wt (hereafter PyMT + /WT) controls. Given that PyMT-driven tumorigenesis and Cre-mediated recombination in this model could occur asynchronously and independently, a conditional tdRFP reporter [32] was introduced as a surrogate for MMTV-Cre mediated Runx1 allele excision (Supplementary Fig. 1A). Loss of Runx1 significantly (P < 0.001) accelerated formation of mammary tumours in PyMT + /R1KO mice (average time to tumour onset of 50 days, ± 10.9) compared to PyMT + /WT mice (average time to tumour onset of 61 days, ± 11.5), indicating that RUNX1 normally restricts the oncogenic pathways activated by PyMT (Fig. 1a). No significant difference in tumour volume between cohorts was observed at tumour onset (Supplementary Fig. 1A). To explore if Runx1 deletion was manifesting a selective advantage to the epithelial population primed for transformation, we assessed the proportion of tdRFP-positive cells in mouse mammary epithelial cells (MMECs) isolated from PyMT + /WT and PyMT + /R1KO mice at tumour notice, as well as from PyMT-negative counterparts at ~60 days. Our results showed a significantly (P < 0.01) increased percentage of tdRFP positivity in Runx1-deleted MMECs, regardless of the status of the PyMT oncogene (Fig. 1b).Fig. 1Loss of Runx1 facilitates early tumour onset in preclinical models of breast cancer.a Time to tumour onset of PyMT+ mice wild type (n = 44; WT) or deleted for Runx1 (n = 34; R1KO). Average time to tumour onset with SD shown: 61 days ( ± 11.5) for WT; 50 days ( ± 10.9) for R1KO mice. Each data point represents an individual mouse. NP: nulliparous. Statistical analysis performed with two-tailed unpaired t-test, ***P < 0.001. b Flow cytometric analysis of tdRFP expression in MMECs from PyMT + /WT (n = 6) and PyMT + /R1KO (n = 6) mice at tumour notice, and PyMT-negative controls (PyMT-/WT, n = 7; PyMT-/R1KO, n = 6) at ~60 days. Each data point represents percentage of tdRFP-positive MMECs, with smaller points representing individual glands (5-7 glands per mouse, see methods) and larger points representing the means from each mouse. Glands isolated from the same mouse are indicated by the same colour. Statistical analysis performed on the mean percentage of tdRFP-positive MMECs with ordinary one-way ANOVA with Sidak’s multiple comparisons test, **P < 0.01. c Time to tumour onset of BLG-Cre;Ctnnb1wt/lox(ex3) (B-cat + ) mice wild type (n = 12; WT) or deleted for Runx1 (n = 16; parous R1KO and n = 14; nulliparous R1KO). Average time to tumour onset with SD shown: 344 days ( ± 69.6) for parous WT; 250 days ( ± 59.3) for parous R1KO; 259 days ( ± 75.9) for nulliparous R1KO mice. Each data point represents an individual mouse. Parous (P) mice, filled circles; nulliparous (NP) mice, open circles. Statistical analysis performed with ordinary one-way ANOVA test with Dunnett’s multiple comparisons test, **P < 0.01. d Time to tumour onset in BLG-Cre;Trp53fl/fl (Trp53-/-) parous (P) and nulliparous (NP) mice wild type (n = 9 and n = 13, respectively; WT) and deleted for Runx1 (n = 8 and n = 15, respectively; R1KO). Each data point represents an individual mouse. Average time to tumour onset with SD shown. WT-P: 199 days ( ± 29.4); WT-NP: 195 days ( ± 26.7); R1KO-P: 195 days ( ± 15); R1KO-NP: 215 days ( ± 28.9). Statistical analysis performed with ordinary one-way ANOVA test with Sidak’s multiple comparisons test; ns, non-significant (P > 0.05). Loss of Runx1 facilitates early tumour onset in preclinical models of breast cancer. a Time to tumour onset of PyMT+ mice wild type (n = 44; WT) or deleted for Runx1 (n = 34; R1KO). Average time to tumour onset with SD shown: 61 days ( ± 11.5) for WT; 50 days ( ± 10.9) for R1KO mice. Each data point represents an individual mouse. NP: nulliparous. Statistical analysis performed with two-tailed unpaired t-test, ***P < 0.001. b Flow cytometric analysis of tdRFP expression in MMECs from PyMT + /WT (n = 6) and PyMT + /R1KO (n = 6) mice at tumour notice, and PyMT-negative controls (PyMT-/WT, n = 7; PyMT-/R1KO, n = 6) at ~60 days. Each data point represents percentage of tdRFP-positive MMECs, with smaller points representing individual glands (5-7 glands per mouse, see methods) and larger points representing the means from each mouse. Glands isolated from the same mouse are indicated by the same colour. Statistical analysis performed on the mean percentage of tdRFP-positive MMECs with ordinary one-way ANOVA with Sidak’s multiple comparisons test, **P < 0.01. c Time to tumour onset of BLG-Cre;Ctnnb1wt/lox(ex3) (B-cat + ) mice wild type (n = 12; WT) or deleted for Runx1 (n = 16; parous R1KO and n = 14; nulliparous R1KO). Average time to tumour onset with SD shown: 344 days ( ± 69.6) for parous WT; 250 days ( ± 59.3) for parous R1KO; 259 days ( ± 75.9) for nulliparous R1KO mice. Each data point represents an individual mouse. Parous (P) mice, filled circles; nulliparous (NP) mice, open circles. Statistical analysis performed with ordinary one-way ANOVA test with Dunnett’s multiple comparisons test, **P < 0.01. d Time to tumour onset in BLG-Cre;Trp53fl/fl (Trp53-/-) parous (P) and nulliparous (NP) mice wild type (n = 9 and n = 13, respectively; WT) and deleted for Runx1 (n = 8 and n = 15, respectively; R1KO). Each data point represents an individual mouse. Average time to tumour onset with SD shown. WT-P: 199 days ( ± 29.4); WT-NP: 195 days ( ± 26.7); R1KO-P: 195 days ( ± 15); R1KO-NP: 215 days ( ± 28.9). Statistical analysis performed with ordinary one-way ANOVA test with Sidak’s multiple comparisons test; ns, non-significant (P > 0.05). We next tested the effects of Runx1 deletion in a novel model of Wnt/β-catenin-driven breast cancer by crossing mice carrying a stabilised form of β-catenin (Ctnnb1wt/lox(ex3)) [33] onto the BLG-Cre line [34]. BLG-Cre;Ctnnb1wt/lox(ex3) mice (hereafter B-cat + ), either Runx1-proficient (hereafter B-cat + /WT) or Runx1-deficient (hereafter B-cat + /R1KO), were generated (Supplementary Fig. 1B). Loss of Runx1 resulted in a significant (P < 0.01) acceleration of Wnt/β-catenin-driven tumorigenesis, with B-cat + /R1KO mice developing tumours at an average onset of 250 days ( ± 59.3) as opposed to the average 344 days ( ± 69.6) of B-cat + /WT animals (Fig. 1c). Of note, Runx1 deletion removed the need for parity required to maximise BLG-Cre expression, with 16 out of 20 (80%) nulliparous B-cat + /R1KO animals aged to 325 days presenting with palpable tumours compared to only 1 out of 12 (8%) nulliparous B-cat + /WT mice aged to the same time. Furthermore, tumour onset in nulliparous B-cat + /R1KO animals occurred at the same latency as the corresponding parous cohort (259 days, ± 75.9; Fig. 1c). There was no difference in tumour volume between cohorts at tumour notice (Supplementary Fig. 1B). Thus, BLG-Cre-mediated loss of Runx1, in combination with constitutive activation of WNT/β-catenin, significantly accelerates mammary tumourigenesis. Interestingly, despite the tumour promoting effects of Runx1 deletion in both MMTV-PyMT and Wnt/β-catenin models, Runx1 loss did not accelerate tumour development in a BLG-Cre;Trp53fl/fl GEMM (Fig. 1d & Supplementary Fig. 1C), pointing towards possible redundancy in the tumour suppressive actions of Trp53 and Runx1. Genetic loss of Runx1 and Runx2 dramatically accelerates Wnt/β-catenin-driven mammary tumour development Given the clinical relevance of WNT/β-catenin in breast cancer [35, 36] and the co-occurrence of WNT/RUNX gene alterations [37], we further investigated the tumour suppressor phenotype shown by Runx1 loss in the Wnt/β-catenin-driven model. It was previously reported that RUNX1 has a role in the differentiation of mature ER+ luminal mammary epithelial cells [38], whilst our group showed that RUNX2 is associated with mammary stemness [25]. Given the auto and cross regulation between these proteins and that through a common binding site, they regulate the same targets, we sought to understand the effects of combined Runx1 and Runx2 loss, hypothesising that Runx2 might substitute for Runx1 loss in restricting the oncogenic potential of WNT/β-catenin. Accordingly, crossing B-cat+ mice with Runx1fl/fl;Runx2fl/fl animals (hereafter B-cat + /DKO) resulted in a dramatic acceleration of disease onset (P < 0.0001), with nulliparous B-cat + /DKO females developing lesions with a median tumour-free age of 56 days ( ± 14.9) compared to both parous B-cat + /R1KO (median tumour-free age of 266 days ± 59.3) and B-cat + /WT (median tumour-free age of 330 days ± 69.6) mice (Fig. 2a and Supplementary Fig. 1D). Indeed, tumour development was too rapid to permit even one round of pregnancy in B-cat + /DKO females. Loss of Runx2 alone (B-cat + /R2KO) did not affect the rate of tumour onset compared to B-cat + /WT mice (Fig. 2a), indicating that the extreme phenotype was only observed with loss of both genes.Fig. 2Combined loss of Runx1 and Runx2 dramatically accelerates Wnt/β-catenin-driven mammary tumour development.a Time to tumour onset in B-cat+ mice wild-type for Runx (n = 12; WT as in Fig. 1C), Runx1fl/fl (n = 16; R1KO as in Fig. 1c), Runx2fl/fl (n = 15; R2KO) and Runx1fl/fl;Runx2fl/fl (n = 32; DKO). Parous (P) mice, solid lines; nulliparous (NP) mice, dotted lines. Median survival of tumour-free mice with SD shown: WT-P mice, 330 days ( ± 69.6); R1KO-P mice, 266 days ( ± 59.3); R2KO-P mice, 367 days ( ± 65.4); DKO-NP mice, 56 days ( ± 14.9). Statistical analysis performed with log-rank (Mantel-Cox) test, ***P < 0.001; **** P < 0.0001. b Number of tumour-bearing mammary glands in B-cat+ cohorts from Fig. 2A harvested at clinical endpoint. Each data point represents an individual mouse. Parous (P) mice, filled circles; nulliparous (NP) mice, open circles. WT-P (n = 9); R1KO-P (n = 12); R2KO-P (n = 12); DKO-NP (n = 24). Statistical analysis performed with ordinary one-way ANOVA with Dunnett’s multiple comparisons test, *P < 0.05; ****P < 0.0001. c Wholemounts and H&E images of abdominal (#4) mammary glands from 6- and 9-weeks old B-cat+ mice from the WT, R1KO, R2KO and DKO cohorts. One representative image is shown per genotype (of n = 5–7 mice per genotype). Scale bars of wholemounts, 2000 µm; H&E images, 1000 µm. Arrow depicts lymph node. d ISH for Axin2 and IHC for KERATIN 5 (KRT5) and KERATIN 18 (KRT18) of 9-weeks mammary glands from B-cat+ mice of the WT, R1KO and DKO genotypes. One representative image is shown per genotype (of n = 5 mice per genotype). Scale bars, 100μm. Low power magnification of the same images is shown in Supplementary Fig. 3C. Bar graphs show quantification analysis of total positive stain area in tissues (left) and total positive stain area as a percentage of non-adipose tissue area (e.g. epithelial and stroma) (right). Each data point represents individual mice. Bars represent mean ± S.D. Statistical analysis performed with ordinary one-way ANOVA with Tukey’s multiple comparisons test, *P < 0.05, **P < 0.01. Combined loss of Runx1 and Runx2 dramatically accelerates Wnt/β-catenin-driven mammary tumour development. a Time to tumour onset in B-cat+ mice wild-type for Runx (n = 12; WT as in Fig. 1C), Runx1fl/fl (n = 16; R1KO as in Fig. 1c), Runx2fl/fl (n = 15; R2KO) and Runx1fl/fl;Runx2fl/fl (n = 32; DKO). Parous (P) mice, solid lines; nulliparous (NP) mice, dotted lines. Median survival of tumour-free mice with SD shown: WT-P mice, 330 days ( ± 69.6); R1KO-P mice, 266 days ( ± 59.3); R2KO-P mice, 367 days ( ± 65.4); DKO-NP mice, 56 days ( ± 14.9). Statistical analysis performed with log-rank (Mantel-Cox) test, ***P < 0.001; **** P < 0.0001. b Number of tumour-bearing mammary glands in B-cat+ cohorts from Fig. 2A harvested at clinical endpoint. Each data point represents an individual mouse. Parous (P) mice, filled circles; nulliparous (NP) mice, open circles. WT-P (n = 9); R1KO-P (n = 12); R2KO-P (n = 12); DKO-NP (n = 24). Statistical analysis performed with ordinary one-way ANOVA with Dunnett’s multiple comparisons test, *P < 0.05; ****P < 0.0001. c Wholemounts and H&E images of abdominal (#4) mammary glands from 6- and 9-weeks old B-cat+ mice from the WT, R1KO, R2KO and DKO cohorts. One representative image is shown per genotype (of n = 5–7 mice per genotype). Scale bars of wholemounts, 2000 µm; H&E images, 1000 µm. Arrow depicts lymph node. d ISH for Axin2 and IHC for KERATIN 5 (KRT5) and KERATIN 18 (KRT18) of 9-weeks mammary glands from B-cat+ mice of the WT, R1KO and DKO genotypes. One representative image is shown per genotype (of n = 5 mice per genotype). Scale bars, 100μm. Low power magnification of the same images is shown in Supplementary Fig. 3C. Bar graphs show quantification analysis of total positive stain area in tissues (left) and total positive stain area as a percentage of non-adipose tissue area (e.g. epithelial and stroma) (right). Each data point represents individual mice. Bars represent mean ± S.D. Statistical analysis performed with ordinary one-way ANOVA with Tukey’s multiple comparisons test, *P < 0.05, **P < 0.01. Examination of B-cat + /DKO mice showed a very distinct presentation, with multiple, sometimes coalescing tumours, as opposed to B-cat + /WT, B-cat + /R1KO and B-cat + /R2KO animals presenting with a smaller number of larger lesions (Supplementary Fig. 2A). Accordingly, a significant increase in the number of tumour bearing glands was observed in B-cat + /DKO (n = 9/10 on average; P < 0.0001) compared to B-cat + /WT (n = 2/10 on average) mice (Fig. 2b). A modest, but significant, difference was also observed between B-cat + /R1KO (n = 4/10 on average; P < 0.05) and B-cat + /WT controls (Fig. 2b). This pathological profile, combined with faster onset of disease, indicates that the threshold for Wnt/β-catenin-driven mammary tumour development is significantly lower with simultaneous loss of Runx1 and Runx2 activity. To track early neoplastic changes, we examined wholemounts and H&E images in virgin females. Activation of Wnt/β-catenin signalling resulted in mild hyperplasia in control B-cat + /WT mice, as well as in B-cat + /R1KO and B-cat + /R2KO cohorts (Fig. 2c). However, B-cat + /DKO mice exhibited dramatic changes, with hyperplastic outgrowths and disrupted architecture (Fig. 2c). As this phenotype was not observed in the absence of activated β-catenin (Supplementary Fig. 2B), loss of RUNX function alone does not overtly interfere with mammary gland development. Rather, RUNX proteins act in concert to suppress Wnt/β-catenin-driven expansion of mammary epithelial cells. We investigated if levels of activated β-catenin were altered in Runx-deleted (B-cat + /R1KO and B-cat + /DKO) glands using the WNT-target gene Axin2 as a readout, since nuclear β-catenin was difficult to quantify (see Supplementary Fig. 3A-B). B-cat + /DKO mammary glands showed increased numbers of Axin2 positive cells (Fig. 2d, Supplementary Fig. 3), however no significant difference was observed between genotypes when normalised to levels of hyperplasia (Fig. 2d). Mammary glands of all genotypes had positive areas for luminal (KRT18) and basal (KRT5) markers (Fig. 2d, Supplementary Figs. 3C, 4A), and while we cannot confirm the cell-of-origin in the hyperplastic lesions, it was notable that it was KRT5-positive cells in which nuclear β-catenin and Axin2-positive cells dominated (Fig. 2d, Supplementary Figs. 3C, 4A). Therefore, we can speculate that the presence of oncogene expression in these cells implicates them in the early expansion of hyperplastic lesions. Loss of RUNX expands a population of mammary epithelial cells with stem-like properties in the presence of activated WNT/β-catenin signalling Given the early changes described above and the increased tumour-bearing glands at endpoint in B-cat + /DKO animals (Fig. 2b), we speculated that combined loss of Runx1 and Runx2 could alter stem/progenitor population(s) to favour Wnt/β-catenin-driven preneoplastic expansion and outgrowth of mammary tumours. Thus, we isolated MMECs from 7 to 9 weeks old B-cat+ mice and performed colony formation cell assay. Stingl et al. [39] demonstrated that fractions enriched for cells capable of mammary gland regeneration in vivo (putative stem cells referred to as mammary repopulating units, MRUs) can be distinguished from the more numerous cells capable of colony formation in vitro (putative progenitor cells referred to as mammary colony forming units, Ma-CFU) by the type of colonies formed. While Ma-CFU give rise to spherical acinar structures, MRUs generate solid, irregular-shaped colonies in Matrigel culture. Interestingly, B-cat + /DKO and B-cat + /R1KO cells showed significantly reduced (P < 0.0001) acinar-like colonies in favour of a significant increase (P < 0.0001) in solid-like colonies compared to B-cat + /WT MMECs (Fig. 3a). These results demonstrate that loss of RUNX function expands a pool of cells enriched for stem-like properties. Interestingly, the effects of Runx1 deletion were as potent as combined Runx1 and Runx2 loss, suggesting the dominance of RUNX1 in curtailing stem/progenitor populations when cultured ex vivo.Fig. 3Loss of RUNX function alters mammary stem/progenitor populations.a Graph (top) and representative pictures (bottom) of colony-forming cell assay showing relative proportions of acinar- and solid-like colonies from MMECs isolated from 7–9 weeks old B-cat+ cohorts; Runx1 wild type (n = 5; WT); R1KO (n = 4); DKO (n = 3). Mean with standard error of N ≥ 3 independent experiments, each with 12 technical replicates per condition. One representative image shown per cohort (n ≥ 3 biological replicates per cohort). Acinar and solid-like colonies depicted by white and black arrows, respectively. Statistical analysis performed with two-way ANOVA with Dunnett’s multiple comparisons test, ****P < 0.0001. Scale bar, 500 µm. b Analysis of mammary RNA-seq signatures derived from gene set enrichment analysis (GSEA) of FACS-sorted tdRFP-positive MMECs from 9-weeks old B-cat+ females comparing: Runx1 knockout (B-cat + /R1KO; n = 5) versus wildtype (B-cat + /WT; n = 4), Runx1/Runx2 knockout (B-cat + /DKO; n = 5) versus B-cat + /WT, and B-cat + /DKO versus B-cat + /R1KO. Analysed gene sets include LIM Mammary Stem Cell Up (https://www.gsea/msigdb.org/gsea-msigdb/mouse/geneset/LIM_MAMMARY_STEM_CELL_UP.html), LIM Mammary Mature Luminal Up (https://www.gsea-msigdb.org/gsea/msigdb/mouse/geneset/LIM_MAMMARY_STEM_CELL_UP.html), and LIM Mammary Luminal Progenitor Up (https://www.gsea-msigdb.org/gsea/msigdb/mouse/geneset/LIM_MAMMARY_STEM_CELL_UP.html). Each bar represents normalised enrichment score (NES) for a specific gene set under each comparison, with higher NES values indicating stronger enrichment of the gene set in the corresponding condition. c Normalised gene expression levels of selected genes in FACS-sorted tdRFP-positive MMECs as described for Fig. 3b. Individual data points represent expression levels per sample, with central line indicating the median. Gene expression normalised for sequencing depth, with significant differences between conditions assessed with Wald test and shown as p-values adjusted for multiple testing with Benjamini–Hochberg method. Both procedures performed with DESeq2; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Loss of RUNX function alters mammary stem/progenitor populations. a Graph (top) and representative pictures (bottom) of colony-forming cell assay showing relative proportions of acinar- and solid-like colonies from MMECs isolated from 7–9 weeks old B-cat+ cohorts; Runx1 wild type (n = 5; WT); R1KO (n = 4); DKO (n = 3). Mean with standard error of N ≥ 3 independent experiments, each with 12 technical replicates per condition. One representative image shown per cohort (n ≥ 3 biological replicates per cohort). Acinar and solid-like colonies depicted by white and black arrows, respectively. Statistical analysis performed with two-way ANOVA with Dunnett’s multiple comparisons test, ****P < 0.0001. Scale bar, 500 µm. b Analysis of mammary RNA-seq signatures derived from gene set enrichment analysis (GSEA) of FACS-sorted tdRFP-positive MMECs from 9-weeks old B-cat+ females comparing: Runx1 knockout (B-cat + /R1KO; n = 5) versus wildtype (B-cat + /WT; n = 4), Runx1/Runx2 knockout (B-cat + /DKO; n = 5) versus B-cat + /WT, and B-cat + /DKO versus B-cat + /R1KO. Analysed gene sets include LIM Mammary Stem Cell Up (https://www.gsea/msigdb.org/gsea-msigdb/mouse/geneset/LIM_MAMMARY_STEM_CELL_UP.html), LIM Mammary Mature Luminal Up (https://www.gsea-msigdb.org/gsea/msigdb/mouse/geneset/LIM_MAMMARY_STEM_CELL_UP.html), and LIM Mammary Luminal Progenitor Up (https://www.gsea-msigdb.org/gsea/msigdb/mouse/geneset/LIM_MAMMARY_STEM_CELL_UP.html). Each bar represents normalised enrichment score (NES) for a specific gene set under each comparison, with higher NES values indicating stronger enrichment of the gene set in the corresponding condition. c Normalised gene expression levels of selected genes in FACS-sorted tdRFP-positive MMECs as described for Fig. 3b. Individual data points represent expression levels per sample, with central line indicating the median. Gene expression normalised for sequencing depth, with significant differences between conditions assessed with Wald test and shown as p-values adjusted for multiple testing with Benjamini–Hochberg method. Both procedures performed with DESeq2; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. To help identify MMECs with simultaneous Wnt/β-catenin activation and Runx loss, B-cat+ mice (i.e. WT, R1KO and DKO) were crossed with the tdRFP line. Consistent with the preneoplastic changes, a trend towards more tdRFP-positive cells was observed in B-cat + /R1KO and B-cat + /DKO cohorts compared to B-cat + /WT, albeit the difference did not reach significance (Supplementary Fig. 4B). RNA-seq analysis was then performed on flow-sorted tdRFP-positive MMECs lacking Runx1, or both Runx1 and Runx2. Interestingly, an enrichment of gene sets aligning with mammary stem cells and negatively associated with mature luminal and progenitor populations, was observed in MMECs isolated from B-cat + /R1KO mice compared to B-cat + /WT animals (Fig. 3b). The same expression profile showed a stronger enrichment in MMECs from B-cat + /DKO mice (Fig. 3b), indicating that combined loss of Runx1 and Runx2 further shifts MMECs to a stem-like state, in vivo. Accordingly, expression levels of luminal markers (i.e. Krt8 and Krt18) were reduced and basal markers (i.e. Krt5 and Krt14) increased in B-cat + /DKO cells compared to their WT counterpart (Fig. 3c), in line with a shift to more solid-like colonies and the gene expression signature of MRUs [39]. In addition, genes important in cancer stem cells, EMT and aggressiveness, such as Aldh1a, Lgr4, Lgr6 and Snai2 [40–43], were mildly and markedly increased in B-cat + /R1KO and B-cat + /DKO cells, respectively, compared to B-cat + /WT MMECs (Fig. 3c). Altogether, these data indicate a shift towards an earlier lineage stage when Wnt/β-catenin activation occurs with simultaneous Runx loss. Runx gene loss increases mammosphere formation in murine epithelial cells To study RUNX function in a more tractable system, we used CRISPR-Cas9 to target Runx1 and Runx2 separately and in combination, in the non-tumorigenic, mammary HC11 cell line [44] (Supplementary Figs. 5A & 5C). There was no obvious change in phenotype or growth characteristics when cells with Runx1 and/or Runx2 loss were grown in 2D conditions (Supplementary Fig. 5D), nor when Runx1 was overexpressed (Supplementary Figs. 5B & 5D). However, Runx1 loss significantly increased mammosphere formation when HC11 cells were grown in suspension (Fig. 4a), whilst the reverse was observed upon Runx1 overexpression (Fig. 4b). By contrast, loss of Runx2 reduced mammosphere number (Fig. 4c), in line with previous results showing Runx2 driving a stem cell phenotype [25]. Combined loss of Runx1 and Runx2 increased mammosphere formation similar to that with Runx1 loss (Fig. 4c), emphasising that RUNX1 is the predominant factor moving cells out of the stem cell compartment, akin to that observed in ex vivo cultures (Fig. 3a). Whilst the ultimate test for stemness is the ability to form ductal outgrowths in cleared fat pads, we did not observe in vivo reconstitution with HC11 cells (empty vector or R1KO), possibly due to the fact these cells are already committed to luminal differentiation [44]. Given the exquisite relationship between Runx and Wnt/β-catenin signalling revealed above, recombinant WNT3A ligand was added to mammosphere cultures. While WNT3A increased mammosphere formation in HC11 cells wild-type for Runx, as expected, it further potentiated stemness of R1KO and DKO clones (Fig. 4d). Altogether, these results suggest complementary effects of Wnt/β-catenin signalling and loss of Runx1 (and Runx2) on mammosphere formation.Fig. 4Runx gene loss increases mammosphere formation in HC11 cells.a Scatter dot-plot of primary (left) and secondary (right) mammospheres of HC11 cells with empty vector (Empty) or independent CRISPR/Cas9-deleted Runx1 clones (R1KO_a, R1KO_b, R1KO_c). Results show normalised mammosphere counts per well relative to the average mammosphere count in control (Empty) group. Smaller points represent technical replicates, larger points the mean of each experimental repeat, with coloured points used to differentiate experimental replicates (also for b–d below). For primary mammospheres, there were 12–48 technical replicates per condition per experimental replicate (n = 8). Error bars indicate normalised mean of all technical replicates ± SD (Empty, 100 ± 11.5; R1KO_a, 158.1 ± 17.9; R1KO_b, 165.4 ± 20.7; R1KO_c, 166.3 ± 20.0). For secondary mammospheres, there were 12-24 technical replicates per condition for each experimental repeat (n = 4). Error bars indicate normalised mean of all technical replicates ± SD (Empty, 100.1 ± 9.6; R1KO_a, 179.1 ± 28.9; R1KO_b, 175.6 ± 31.1; R1KO_c, 174.5 ± 34.1). Statistical analysis performed with ordinary one-way ANOVA with Dunnett’s multiple comparisons test, ****P < 0.0001. b Scatter dot-plot of primary mammospheres of HC11 cells overexpressing Runx1 (Runx1P1) or empty vector control (Empty). Results show mammosphere counts relative to the average mammosphere count in control group (Empty). Error bars indicate normalised mean of all technical replicates ± SD (Empty, 100 ± 18.6; Runx1P1, 73.5 ± 14.3) and data are representative of n = 4 experimental repeats. For each experiment, there were 8–12 technical replicates per condition. Statistical analysis performed with two-tailed unpaired t-test, ****P < 0.0001. c Scatter dot-plot of primary mammospheres of HC11 cells first deleted for Runx2 and then transduced with lentiCRISPRv2-Neo empty vector (Empty) or containing a Runx1-targeting gRNA plasmid to generate an independent Runx1-deleted (R1KO_d), two Runx2-deleted (R2KO_a, R2KO_b) and two Runx1/Runx2-deleted (DKO_a, DKO_b) clones. Results show normalised mammosphere counts per well relative to the average mammosphere count in control group (empty). Error bars are normalised mean of all technical replicates with SD (Empty, 100 ± 7.4; R1KO_d, 137.1 ± 9.1; R2KO_a, 80.1 ± 7.6; R2KO_b, 78.6 ± 9.6; DKO_a, 120.6 ± 7.9; DKO_b, 116.4 ± 7.7) and data are representative of N = 3 experimental repeats. For each experiment, there were 12 technical replicates per condition. Statistical analysis performed with ordinary one-way ANOVA with Dunnett’s multiple comparisons test, ****P < 0.0001. d Scatter dot-plot of primary mammospheres of HC11 cells as for c in the presence or absence of recombinant WNT3A ligand. Error bars show mean of all technical replicates ± SD for vehicle treated (Empty, 120.5 ± 6.2; R1KO_D, 145.4 ± 7.0; DKO_a, 136.8 ± 7.7; DKO_b, 137.7 ± 7.9) and WNT3A treated (Empty, 152.5 ± 8.1; R1KO_D, 191.3 ± 9.0; DKO_a, 193.4 ± 10.2; DKO_b, 193.1 ± 9.5) clones, of N = 2 experimental repeats. For each experiment, 12 technical replicates were tested per condition. Statistical analysis performed with ordinary one-way ANOVA with Sidak’s multiple comparisons test, ****P < 0.0001. Runx gene loss increases mammosphere formation in HC11 cells. a Scatter dot-plot of primary (left) and secondary (right) mammospheres of HC11 cells with empty vector (Empty) or independent CRISPR/Cas9-deleted Runx1 clones (R1KO_a, R1KO_b, R1KO_c). Results show normalised mammosphere counts per well relative to the average mammosphere count in control (Empty) group. Smaller points represent technical replicates, larger points the mean of each experimental repeat, with coloured points used to differentiate experimental replicates (also for b–d below). For primary mammospheres, there were 12–48 technical replicates per condition per experimental replicate (n = 8). Error bars indicate normalised mean of all technical replicates ± SD (Empty, 100 ± 11.5; R1KO_a, 158.1 ± 17.9; R1KO_b, 165.4 ± 20.7; R1KO_c, 166.3 ± 20.0). For secondary mammospheres, there were 12-24 technical replicates per condition for each experimental repeat (n = 4). Error bars indicate normalised mean of all technical replicates ± SD (Empty, 100.1 ± 9.6; R1KO_a, 179.1 ± 28.9; R1KO_b, 175.6 ± 31.1; R1KO_c, 174.5 ± 34.1). Statistical analysis performed with ordinary one-way ANOVA with Dunnett’s multiple comparisons test, ****P < 0.0001. b Scatter dot-plot of primary mammospheres of HC11 cells overexpressing Runx1 (Runx1P1) or empty vector control (Empty). Results show mammosphere counts relative to the average mammosphere count in control group (Empty). Error bars indicate normalised mean of all technical replicates ± SD (Empty, 100 ± 18.6; Runx1P1, 73.5 ± 14.3) and data are representative of n = 4 experimental repeats. For each experiment, there were 8–12 technical replicates per condition. Statistical analysis performed with two-tailed unpaired t-test, ****P < 0.0001. c Scatter dot-plot of primary mammospheres of HC11 cells first deleted for Runx2 and then transduced with lentiCRISPRv2-Neo empty vector (Empty) or containing a Runx1-targeting gRNA plasmid to generate an independent Runx1-deleted (R1KO_d), two Runx2-deleted (R2KO_a, R2KO_b) and two Runx1/Runx2-deleted (DKO_a, DKO_b) clones. Results show normalised mammosphere counts per well relative to the average mammosphere count in control group (empty). Error bars are normalised mean of all technical replicates with SD (Empty, 100 ± 7.4; R1KO_d, 137.1 ± 9.1; R2KO_a, 80.1 ± 7.6; R2KO_b, 78.6 ± 9.6; DKO_a, 120.6 ± 7.9; DKO_b, 116.4 ± 7.7) and data are representative of N = 3 experimental repeats. For each experiment, there were 12 technical replicates per condition. Statistical analysis performed with ordinary one-way ANOVA with Dunnett’s multiple comparisons test, ****P < 0.0001. d Scatter dot-plot of primary mammospheres of HC11 cells as for c in the presence or absence of recombinant WNT3A ligand. Error bars show mean of all technical replicates ± SD for vehicle treated (Empty, 120.5 ± 6.2; R1KO_D, 145.4 ± 7.0; DKO_a, 136.8 ± 7.7; DKO_b, 137.7 ± 7.9) and WNT3A treated (Empty, 152.5 ± 8.1; R1KO_D, 191.3 ± 9.0; DKO_a, 193.4 ± 10.2; DKO_b, 193.1 ± 9.5) clones, of N = 2 experimental repeats. For each experiment, 12 technical replicates were tested per condition. Statistical analysis performed with ordinary one-way ANOVA with Sidak’s multiple comparisons test, ****P < 0.0001. Loss of Runx1 and Runx2 results in the activation of oncogenic pathways To understand the pathways driving accelerated Wnt/β-catenin-driven oncogenic development in the absence of both Runx genes, we explored RNA-seq data of the tdRFP-positive sorted MMECs. In line with the milder phenotype and modest reduction in tumour development observed between B-cat + /WT and B-cat + /R1KO groups, only 88 differentially expressed genes reached significance upon Runx1 loss (Supplementary Table 1). By contrast, 1312 genes were significantly different between B-cat + /WT and B-cat + /DKO cohorts (Supplementary Table 2), demonstrating a greatly altered gene expression in the absence of both Runx genes (Fig. 5a). Of note, 75/88 (85%) significantly altered genes within the B-cat + /R1KO set were shared with the B-cat + /DKO cohort (Fig. 5a). Genes altered only in the Runx1-deleted condition tended to be developmental regulators (e.g. Meis2, Zfp949, Nrp1, Capn12, Rdh10) and glycoproteins (e.g. Stc1, Abdh12b). When we interrogated gene set enrichment analysis and the Hallmark data sets using a significance cut-off of P < 0.01 and a false discovery rate of P < 0.05, 41 Hallmark gene sets were found to be enriched in the B-cat + /DKO dataset compared to B-cat + /WT, whilst only 28 were significantly enriched in the B-cat + /R1KO group relative to B-cat + /WT. Notably, the majority of these gene sets (24/28) were also shared with the B-cat + /DKO cohort (Fig. 5a).Fig. 5Combined loss of Runx1 and Runx2 results in the activation of key oncogenic pathways.RNA-seq analysis of MMECs isolated from 9-weeks old B-cat + /WT (n = 4), B-cat + /R1KO (n = 5) and B-cat + /DKO (n = 5) females. Each sample comprised of pooled tdRFP-positive MMECs isolated from multiple mice of the same genotype (see “Methods”). a Venn diagrams of the number of significantly altered differentially expressed genes (DEGs) (left) and enriched Hallmark Pathways identified by GESA (right) comparing B-cat + /R1KO against B-cat + /WT, and B-cat + /DKO against B-cat + /WT. The differential expression of a gene was considered significant if it had an absolute fold change >1.5 and adjusted P < 0.05, calculated using Wald test with DESeq2 package. Pathways’ significance determined based on a false discovery rate (FDR) q-value threshold of 0.05. Circles represent sets of DEGs/pathways identified in B-cat + /R1KO and B-cat + /DKO; intersecting region shows DEGs/pathways common to both conditions; non-overlapping areas represent DEGs/pathways unique to each condition. b Lollipop plot displaying normalised enrichment score (NES) from GSEA of the Hallmark Gene Sets comparing B-cat + /R1KO and B-cat + /DKO to B-cat + /WT. Each lollipop represents a gene set, with NES on the y-axis and the gene sets on the x-axis. Only gene sets that were significant (i.e. FDR q-value of 0.05 or less) in at least one of the conditions were used, with the position of the lollipop indicating the NES, and the size of the bubble corresponding to the −log10 of the FDR-adjusted q-value. Gene ranking for the analysis was based on the signal-to-noise ratio. These plots highlight the most significantly enriched gene sets in each condition, with larger bubbles indicating stronger statistical significance. Combined loss of Runx1 and Runx2 results in the activation of key oncogenic pathways. RNA-seq analysis of MMECs isolated from 9-weeks old B-cat + /WT (n = 4), B-cat + /R1KO (n = 5) and B-cat + /DKO (n = 5) females. Each sample comprised of pooled tdRFP-positive MMECs isolated from multiple mice of the same genotype (see “Methods”). a Venn diagrams of the number of significantly altered differentially expressed genes (DEGs) (left) and enriched Hallmark Pathways identified by GESA (right) comparing B-cat + /R1KO against B-cat + /WT, and B-cat + /DKO against B-cat + /WT. The differential expression of a gene was considered significant if it had an absolute fold change >1.5 and adjusted P < 0.05, calculated using Wald test with DESeq2 package. Pathways’ significance determined based on a false discovery rate (FDR) q-value threshold of 0.05. Circles represent sets of DEGs/pathways identified in B-cat + /R1KO and B-cat + /DKO; intersecting region shows DEGs/pathways common to both conditions; non-overlapping areas represent DEGs/pathways unique to each condition. b Lollipop plot displaying normalised enrichment score (NES) from GSEA of the Hallmark Gene Sets comparing B-cat + /R1KO and B-cat + /DKO to B-cat + /WT. Each lollipop represents a gene set, with NES on the y-axis and the gene sets on the x-axis. Only gene sets that were significant (i.e. FDR q-value of 0.05 or less) in at least one of the conditions were used, with the position of the lollipop indicating the NES, and the size of the bubble corresponding to the −log10 of the FDR-adjusted q-value. Gene ranking for the analysis was based on the signal-to-noise ratio. These plots highlight the most significantly enriched gene sets in each condition, with larger bubbles indicating stronger statistical significance. Given that combined loss of Runx1 and Runx2 resulted in a dramatically accelerated oncogenic phenotype, we were interested in those gene sets that distinguished the B-cat + /DKO group from the B-cat + /R1KO one. These fell into two groups. The first refers to gene sets that showed an enriched score when the B-cat + /DKO dataset was compared against the B-cat + /R1KO one (Fig. 5b), including MYC, mTORC1, TGFβ, EMT, IL6/JAK/STAT3, IL2/STAT5 and WNT/β-catenin signalling (Fig. 5b), possibly contributing to the more aggressive phenotype. The second group includes gene sets negatively enriched in B-cat + /R1KO MMECs compared to the B-cat + /WT ones, yet positively enriched in B-cat + /DKO cells. Changes in these pathways (e.g. TNFA, hypoxia, PI3K/AKT/MTOR, ROS and apoptosis) might be key to explaining the dramatic differences observed upon simultaneous Runx1 and Runx2 loss (Fig. 5b). Deletion of Runx in mammary epithelial cells alters the immune microenvironment To investigate the wider effect of Runx deletion on the tumour microenvironment, RNA-seq was also carried out on whole mammary tissue from 9-week-old mice. PCA showed that samples within each genotype grouped together (Fig. 6a). In keeping with the B-cat + /R2KO cohort having a comparable tumour phenotype to the B-cat + /WT cohort (Fig. 2a-b), these two groups were transcriptionally aligned, while the B-cat + /DKO one was the most divergent (Fig. 6a). Of all genotypes, B-cat + /R1KO samples were slightly more dispersed (Fig. 6a), possibly reflecting higher heterogeneity with a greater requirement for additional transforming events in Runx1-deleted cells. Surprisingly, MetaCore pathway analysis revealed that 7/10 of the most significantly enriched pathways in the B-cat + /DKO cohort related to immune pathways (Fig. 6b, lighter bars). These data suggested to us that loss of Runx1 and Runx2 in Wnt/β-catenin positive MMECs might alter the immune landscape of surrounding tissues. Deconvolution analysis using two different approaches was applied. First, the Mouse Microenvironment Cell Population counter (mMCP-counter) [45] revealed 16 major immune and stromal cell types using gene expression profiles that identify and quantify different cellular populations. B-cat + /DKO glands showed a markedly increased presence of immune cells within the mammary gland comprised of macrophages, granulocytes and T cells, including CD8 T cells (Fig. 6C). We then followed up with CIBERSORTx [46] and referenced our RNA-seq data against that of Bach et al. [47] who used single cell RNA-seq to fractionate different epithelial, stromal and immune cell types within the murine gland. Three different CD8 + T cell (CD8-1, CD8-2 and CD8-3) and two different CD4 + T cell (CD4-1 and CD4-2) subtypes were identified based on gene expression profiles [47]. We noted clear alterations of these subsets in B-cat + /DKO glands, with a decrease in CD8-1, CD8-2, CD4-2 and CTL populations and an increase in CD8-3 and T-reg cells (Fig. 6d). Our data also demonstrated that B-cat + /DKO glands are enriched for a macrophage subtype termed Mo3, previously identified as tissue resident macrophages and associated with mammary ducts [48]. Guided by the altered populations in CIBERSORTx, 9-weeks mammary glands were stained for immune populations, where it was found that neutrophils (S100A8 + ) and macrophages (F4/80 + ) were indeed significantly increased in the B-cat + /DKO cohort compared to B-cat + /WT, even when normalised to the increased level of hyperplastic lesions (Fig. 6e). No significant difference in CD4 or Foxp3 levels were observed. Altogether, these results indicate that loss of Runx in Wnt/β-catenin-driven MMECs results in both tumour-intrinsic cell changes, as well as in significant perturbations to non-tumour cells capable of altering the local environment.Fig. 6Deletion of Runx genes in mammary epithelial cells alters the immune microenvironment.a PCA plot of the top 500 most variable genes showing RNA-seq expression profiles for whole mouse mammary glands harvested from 9-weeks old B-cat+ mice. Each point represents an individual mouse: Runx wildtype (B-cat + /WT; black; n = 5), B-cat + /R1KO (blue; n = 5), B-cat + /R2KO (green; n = 5), and B-cat + /DKO (red; n = 4). PC1 and PC2, explaining 55% and 16% of the variance, are on the x- and y-axes, respectively. b Top enriched pathways from MetaCore’s Canonical Pathway Maps representing signalling and metabolic pathways are shown (for samples as in a). The -log10 of the FDR-corrected p-value indicates the probability of the observed intersection between the DEGs and the pathway gene set occurring by chance, as determined by a hypergeometric test. A higher -log10 p-value suggests a stronger association between the DEGs and the pathway “DKO vs WT” refers to DEGs in the Runx1/Runx2 double knockout compared to wildtype. c Heatmap displaying z-scored mMCP scores for immune and stromal cell populations in Runx wildtype (B-cat + /WT; n = 5) versus Runx1/Runx2 double knockout (B-cat + /DKO; n = 4) 9-weeks old mammary glands (as in A). Each column represents an individual mouse. d Heatmap of CIBERSORTx immune cell fractions in Runx wildtype (B-cat + /WT; n = 5) versus Runx1/Runx2 double knockout (B-cat + /DKO; n = 4) 9-weeks old mammary glands (as in a and c). CIBERSORTx immune cell fractions calculated using a signature matrix created from single cell sequencing of BLG-cre;Brca1f/f;Trp53+/- murine mammary glands from Bach et al. [47]. Z-scores are representative of the gene expression relative to the mean expression, indicated by a diverging scale running from blue to red. Each column represents an individual mouse. e IHC for S100A8, F4/80, CD4 and ISH for Foxp3 of 9-weeks mammary glands from B-cat + /WT and B-cat + /DKO mice (top). One representative image is shown per genotype (of n = 5 mice per genotype). Scale bars, 100μm. Bar graphs (bottom) show quantification analysis of average number of positive cells for S100A8, CD4, Foxp3 or positive stain area for F4/80 (left) and average positive stain proportional to classified tissue area (e.g. epithelial and stromal cells) (right). Each data point represents individual mice. Bars represent mean ± S.D. Statistical analysis performed with unpaired t test, *P < 0.05, **P < 0.01. Deletion of Runx genes in mammary epithelial cells alters the immune microenvironment. a PCA plot of the top 500 most variable genes showing RNA-seq expression profiles for whole mouse mammary glands harvested from 9-weeks old B-cat+ mice. Each point represents an individual mouse: Runx wildtype (B-cat + /WT; black; n = 5), B-cat + /R1KO (blue; n = 5), B-cat + /R2KO (green; n = 5), and B-cat + /DKO (red; n = 4). PC1 and PC2, explaining 55% and 16% of the variance, are on the x- and y-axes, respectively. b Top enriched pathways from MetaCore’s Canonical Pathway Maps representing signalling and metabolic pathways are shown (for samples as in a). The -log10 of the FDR-corrected p-value indicates the probability of the observed intersection between the DEGs and the pathway gene set occurring by chance, as determined by a hypergeometric test. A higher -log10 p-value suggests a stronger association between the DEGs and the pathway “DKO vs WT” refers to DEGs in the Runx1/Runx2 double knockout compared to wildtype. c Heatmap displaying z-scored mMCP scores for immune and stromal cell populations in Runx wildtype (B-cat + /WT; n = 5) versus Runx1/Runx2 double knockout (B-cat + /DKO; n = 4) 9-weeks old mammary glands (as in A). Each column represents an individual mouse. d Heatmap of CIBERSORTx immune cell fractions in Runx wildtype (B-cat + /WT; n = 5) versus Runx1/Runx2 double knockout (B-cat + /DKO; n = 4) 9-weeks old mammary glands (as in a and c). CIBERSORTx immune cell fractions calculated using a signature matrix created from single cell sequencing of BLG-cre;Brca1f/f;Trp53+/- murine mammary glands from Bach et al. [47]. Z-scores are representative of the gene expression relative to the mean expression, indicated by a diverging scale running from blue to red. Each column represents an individual mouse. e IHC for S100A8, F4/80, CD4 and ISH for Foxp3 of 9-weeks mammary glands from B-cat + /WT and B-cat + /DKO mice (top). One representative image is shown per genotype (of n = 5 mice per genotype). Scale bars, 100μm. Bar graphs (bottom) show quantification analysis of average number of positive cells for S100A8, CD4, Foxp3 or positive stain area for F4/80 (left) and average positive stain proportional to classified tissue area (e.g. epithelial and stromal cells) (right). Each data point represents individual mice. Bars represent mean ± S.D. Statistical analysis performed with unpaired t test, *P < 0.05, **P < 0.01. Discussion RUNX1 and CBFB have been reported to be genetically altered in breast cancer, with many of these changes believed to be loss-of-function mutations [11, 12, 38]. In addition, loss of RUNX1 function is associated with disease progression [14, 15, 49]. Despite this, it is unclear how the RUNX1/CBFβ complex protects against tumour development. In this study, we demonstrate that Runx1 loss accelerates mammary tumourigenesis in GEMMs driven by the PyMT oncogene or by sustained Wnt/β-catenin signalling, providing definitive in vivo proof that Runx1 fulfils a tumour suppressor role in breast cancer, at least in the context of these oncogenic pathways. However, this does not appear to be a universal phenomenon as no tumour acceleration was observed when Runx1 was deficient in mammary tumours arising from Trp53 loss. This was surprising given that Runx1 loss in MMECs had previously been linked with enrichment of genes associated with p53 activation [38]. The observation that arrested development of luminal epithelial cells lacking Runx1 could be rescued in the absence of p53, resulting in hyperproliferation of ER+ cells, led the authors to speculate that these events may pre-dispose to breast cancer [38, 50]. However, our results suggest that the tumour suppressive functions of Runx1 and Trp53 might overlap, consistent with TP53 and RUNX1 mutations only partially co-occurring in METABRIC or TCGA breast cancer datasets. Similarly, others reported that CBFB and TP53 mutations were also mutually exclusive in these datasets [51]. Heightened WNT signalling is a common finding in many breast tumours [52], with aberrant β-catenin activation reported to be preferentially associated with poor clinical outcome in patients with invasive breast cancer, regardless of molecular subtypes [28]. Moreover, deregulated WNT signalling is linked with aggressive tumour features, including metastasis and the maintenance of cancer stem cells [53]. In the presence of activated β-catenin, Runx1 deficiency promotes tumourigenesis and enhances a stem-like phenotype in both MMECs in vivo, as well as HC11 cells in vitro. These results agree with others who have shown that Runx1 drives the differentiation of stem/progenitor mammary epithelial cells [38, 54] and predicted that RUNX1 loss would result in the expansion of such cells [54]. Although loss of Runx2 did not change the rate of tumour formation, accelerated tumour onset was observed when β-catenin expressing MMECs were deficient for both Runx1 and Runx2. These results indicate that, in the absence of Runx1, Runx2 may curtail tumour development. Although different family members have unique roles during lineage programming and cell fate decisions, this largely arises from their distinct spatial and temporal expression patterns. That the Runx genes have significant functional redundancy is not surprising, given they share a common binding site and regulate a common gene set, as established in a number of different systems [55–57]. Our results show that loss of Runx2 can promote Wnt/β-catenin-driven mammary tumourigenesis, but only in combination with Runx1 loss. It is possible that, when both genes are functional, Runx2 may act to repress the effects of Runx1 to keep cells in a stem-like state. However, the enhanced stem cell signature observed here when both genes are absent suggests a cryptic role for Runx2 in negatively regulating the stem cell phenotype, which is only revealed in the absence of Runx1. Together, these suggest that both Runx1 and Runx2 act in concert to restrain the oncogenic potential and stem-like properties of aberrant Wnt/β-catenin. RNA-seq analysis revealed that upon β-catenin activation, many more genes are differentially expressed and many more oncogenic pathways activated when both Runx genes are lost. Particularly intriguing are those gene sets negatively enriched in cells that have lost Runx1 compared to activated β-catenin alone, but that are positively enriched when both Runx genes are lost (e.g. TNFA and hypoxia). Thus, the rapid tumour onset of multiple clones might be driven both by the expansion of a stem-like population of mammary cells and activation of different oncogenic pathways that are only unleashed in the absence of RUNX function. An additional finding was the marked enrichment of immune pathways observed in B-cat + /DKO glands before the appearance of gross lesions. Intriguingly, T-regs were one of the cell types predicted to be enriched in glands with simultaneous epithelial loss of Runx1 and Runx2, suggesting the establishment of an immune suppressive environment, albeit there was only a trend to increased Foxp3+ immunostaining in B-cat + /DKO samples. Whilst the mMCP analysis revealed a general T cells increase, refined CIBERSORTx deconvolution indicated that certain T cells subtypes were reduced, whilst others expanded, possibly reflecting that, overall, there was no gross increase in CD4+ staining in B-cat + /DKO samples. Future work will understand whether changes in T cell subtypes restrict or promote tumorigenesis. It is possible that the increase in Mo3 cells displayed by B-cat + /DKO mice may be secondary to ductal proliferation, as others have shown that hormone induced branching can drive expansion of this population [48]. Intriguingly, ductal macrophages have been associated with an immunosuppressive function within the gland, characterized by an expression profile aligning with tumour associated macrophages [48, 58]. Immunohistochemistry (IHC) of B-cat + /DKO glands confirmed a significant increase in neutrophil and macrophage populations. Thus, the altered immune populations might have the potential to influence the cellular milieu that nurtures emerging B-cat + /DKO tumours, a concept worth further investigation considering that WNT-driven tumours are characteristically renowned to be immune-excluded [59, 60]. In summary, loss of RUNX function within Wnt/β-catenin-driven mammary epithelial cells alters the threshold for oncogenic transformation and changes the immune response towards this expanding population of abnormal cells. How these different effects interact and contribute to rapid tumour development will be the focus of future studies. Materials and methods Animal procedures Mouse experiments were performed with ethical approval from the University of Glasgow Animal Welfare and Ethical Review Board under the Animal (Scientific Procedures) Act project licence authority (708645 and PP6345023 to K Blyth). Mice were housed in a pathogen-free facility in individual ventilated cages on a 12 h light/dark cycle with continual access to food, water and environmental enrichment. Genotyping was carried out by Transnetyx (Cordova, US). Mouse euthanasia was humanely performed. Transgenic lines and genetic models of mammary cancer Mouse Mammary Tumour Virus (MMTV)-Cre [61] and MMTV-Polyomavirus middle T antigen (PyMT) [30] lines were provided by WJ Muller. Tandem dimer red fluorescent (Gt(ROSA)26Sortm1Hjf) knock-in Cre-reporter line (hereafter tdRFP) was obtained from the European Mouse Mutant Archive [32]. Beta-lactoglobulin (BLG)-Cre [34]; Ctnnb1wt/lox(ex3) [33]; Trp53fl/fl [62]; Runx1fl/fl [63] and Runx2fl/fl [25] alleles as previously published. MMTV-PyMT mice were crossed with MMTV-Cre;Runx1fl/fl to generate PyMT + /R1KO and PyMT + /WT lines, backcrossed 10 generations to FVB/N [Charles River, UK]. BLG-Cre;Ctnnb1wt/lox(ex3) mice, bred with Runx1fl/fl and Runx2fl/fl lines, were intercrossed to establish cohorts maintained on an outbred background (C57BL/6 J; FVB/NCrl) as confirmed by Transnetyx MiniMUGA [64]. BLG-Cre;Trp53fl/fl mice, bred with Runx1fl/fl and Runx1wt/wt colonies, were backcrossed 10 generations to FVB/N [Charles River, UK]. To maximize BLG-Cre expression [34], females were mated at 11–13-weeks for two rounds of pregnancy, except for three BLG-Cre;Ctnnb1wt/lox(ex3);Runx2fl/fl (B-cat + /R2KO) females mated at 7–9-weeks. Tumour onset, tumour volume and clinical endpoint analyses Mice were palpated for tumour formation twice weekly. Tumour onset was defined as lesions ≥2 mm and ≤7 mm. Mice with tumours >7 mm at time of notice were excluded from analysis. Tumours were measured with callipers and volume (mm3) calculated as follows: Volume = (width x width x length)/2, with investigator blinded to genotype. Mice were euthanised when tumours reached endpoint (largest diameter at 15 mm or ulcerated) and number of tumour-bearing glands (of 10 in total) counted. Estimation of sample size based on previous experience. Numbers for tumour onset included cohorts used in different analyses (survival, RNAseq, IHC). Randomisation methods were not included as experimental groups were determined by genotype. Wholemounts and histological analyses Virgin females were sacrificed at 6- and 9-weeks. For wholemount staining, right-side abdominal (#4 R) mammary glands were excised, spread on glass slides, air-dried and placed overnight in Carnoy’s solution (6-parts EtOH/3-parts CHCl3/1-part glacial acetic acid). Slides were washed in EtOH/15 min (70%, 50%, 25%), rinsed in dH20/10 min and stained overnight in Carmine Alum. Glands were washed in EtOH/15 min (70%, 95%, 100%), cleared in Xylene, mounted with Pertex Mounting Media and coverslipped. Wholemounts were photographed with a Zeiss stereomicroscope. For H&E staining, left abdominal glands (#4 L) were formalin-fixed, paraffin-embedded (FFPE), sectioned at 4 µm and mounted on glass. Immunohistochemistry (IHC), in-situ hybridisation (ISH) and co-Immunofluorescence (co-IF) were carried out on 4 µm FFPE sections oven heated at 60 ⁰C/2 h. Antibodies for β-catenin [610154, BD Biosciences, 1:250], CD4 [14-9766-82, eBioscience, 1:500], F4/80 [ab6640, Abcam, 1:200], KERATIN 5 [905501, Biolegend, 1:1500], KERATIN 18 [ab181597, Abcam, 1:1000] and S100A8 [MAB3059, R&D systems, 1:2000] were used on a Leica Bond Rx autostainer. All FFPE sections underwent on-board dewaxing and epitope retrieval. Ready-to-use secondary antibodies were applied for 30 min as follows: rat ImmPRESS secondary solution [MP-7404, Vector Labs] for CD4, F4/80 and S100A8; mouse Envision [K4001, Agilent] for B-Catenin and rabbit Envision [K4003, Agilent] for KERATIN 5 and KERATIN 18. Sections were visualised with DAB and counterstained with haematoxylin. ISH detection for Axin2 (400338) mRNA was performed using RNAScope 2.5 LSx (Brown) detection kit [322700; Bio-Techne]. Foxp3 [12653, Cell Signaling; 1:200] underwent heat-induced epitope retrieval [K8005, Agilent], stained with secondary (rabbit envision) and visualised with DAB counterstained with hematoxylin. For KERATIN 5/KERATIN 18 co-IF, KERATIN 5 was applied (1:750/1 h), rinsed, then incubated with goat anti-rabbit AF555 [A21428, Thermo Fisher Scientific; 30 min]. KERATIN 18 (1:1000/1 h) was applied, rinsed, then incubated with goat anti-rabbit [AF647 Thermo Fisher Scientific; 30 min]. For β-Catenin/KERATIN 5 co-IF, β-Catenin was applied (1:100/1 h), rinsed, then incubated with goat anti-mouse AF555 [A21422, Thermo Fisher Scientific; 30 min]. KERATIN 5 (1:750/1 h) was applied, rinsed, then incubated with goat anti-rabbit AF647 [A21244, Thermo Fisher Scientific; 30 min]. Counterstaining done with DAPI. Confocal images were collected on a Nikon A1R point-scanning confocal microscope, built on an inverted Ti Eclipse stand. Images were acquired using a 20x/0.75 Plan-Apo objective lens with a confocal pinhole diameter of 27 µm. Multi-channel images were captured sequentially: DAPI (nuclear marker) using 404 nm excitation and 425–475 nm emission bandwidth, Alexa Fluor 555 using 561 nm excitation and 570–620 nm emission and Alexa Fluor 647 using 636 nm excitation and 663–738 nm emission. Images were collected with a 1.73x zoom with an image size of 1024 × 1024 pixels, yielding a pixel size of 0.36 × 0.36 um, and a 1.0 us pixel dwell time. Images were acquired using the software NIS-Elements AR 4.5 (Nikon). Image quantification HALO (Indica Labs) software was used for image quantification with investigator blinded to genotype. Positive stain area of Axin2, F4/80, CD4, KERATIN 5 and KERATIN 18 stains were classified using the Random Forest machine learning algorithm. S100A8 and Foxp3 positive stains were quantified with the Indica Labs Cytonuclear v2.0.9 algorithm. Quantification of Axin2, KERATIN 5 and KERATIN 18 was done on whole slide images (WSI). For F4/80, CD4, S100A8 and Foxp3 stains, representative hyperplastic regions of interest (area = 581369.68μm2, n = 4–9 per sample) from WSI were selected. Average positive stain and average positive stain relative to classified tissue (Average % Pos/Total Tissue) were calculated as readouts. Mouse mammary epithelial cell (MMEC) extraction Mammary glands (with removed lymph nodes) were dissected, placed in serum-free working solution—1X F-12 Nutrient Mixture (Ham), 1X Penicillin-Streptomycin-Glutamine (PSG)—chopped [Mcllwain, UK] and tissue paste incubated (1.5 h/37 °C) with serum-free digestion solution—1X Ham, 1X PSG, 10% Collagenase (300U/ml)/Hyaluronidase (100U/ml)—in a shaking incubator (100 rpm). Organoids were collected through centrifugation (350 g/5 min), while red blood cells lysed in 0.8% NH4Cl (RT/5 min) and removed by aspiration. To remove stromal cells, pellet was re-suspended in 10% Foetal Bovine Serum (FBS)-supplemented working solution, plated and incubated (37 °C/1 h). Floating epithelial organoids were isolated and centrifuged (350 g/5 min). Pellet was washed with 1X Phosphate-Buffered Saline (PBS), re-suspended in 0.5–1 ml dissociation solution—1X PBS, 0.25% Trypsin, 1 mM EGTA, 0.1 mg/ml DNase I recombinant, RNase-free 10,000U—and incubated (37 °C/10 min) to obtain a single cell suspension. MMECs purified by filtering (40 µm) and centrifugation (350 g/5 min). Flow cytometric analysis and fluorescence-activated cell sorting (FACS) For flow cytometry analysis of tdRFP expression, MMECs were isolated from PyMT + /WT and PyMT + /R1KO mice at tumour notice (average of 61 and 50 days, respectively). An equivalent time-point ( ~ 60 days) was chosen for PyMT-negative cohorts. One abdominal gland (#4 R) was retained for histology. Remaining glands were analysed as individual samples (except for thoracic glands combined due to their overlap, i.e. #2/3 R and #2/3 L). 7 samples were analysed from PyMT+ cohorts (#1 R, #1 L, #2/3 R, #2/3 L, #4 L, #5 R and #5 L), 5 samples from PyMT-negative mice due to scarcity of MMECs in #1 and #5 glands, which were combined (i.e. #1RL, #2/3 R, #2/3 L, #4 L, #5RL). For flow cytometry and FACS of tdRFP-positive cells in B-cat+ cohorts, ten glands were pooled from each 8–9-weeks mouse and analysed together. MMECs were re-suspended in FACS buffer—1X PBS, 2 mM EDTA pH 8 and 1% FBS—blocked with Fc receptor antibody [Biolegend, 101319, 1:100/15 min], labelled with antibody cocktail/30 min—BV421 anti-mouse CD31 [BD Pharmingen, 551262, 1:200], BV421 anti-mouse CD45 [BD Pharmingen, 559864, 1:200], FITC anti-mouse CD24 [BD Pharmingen, 553262, 1:100]—and stained with DAPI [Life Technologies, D1306, 1:1000/5 min]. For each experiment, tdRFP-negative MMECs were used to set gates. Flow cytometry analysis done on an Attune Nxt Flow Cytometer [Invitrogen], with FACS using a BD FACSAria™ III Cell Sorter. Data was analysed with FlowJo (version 10.2). Colony forming cell assay MMECs from 7 to 9-weeks B-cat+ mice were plated onto 24-well plates (5 × 103 viable cells/well), embedded in 25 µl of Matrigel and cultured with 500 µl of medium—1X EpiCult-B Basal Medium Mouse, 10% EpiCult-B Proliferation Supplements, 5% FBS, 1% PSG, 4 µg/ml Heparin, 10 ng/ml mouse epidermal growth factor (mEGF), 10 ng/ml mouse fibroblast growth factor-2 (mFGF-2). MMECs were extracted from all ten glands per mouse and fed with 500 µl of fresh medium at day 3. Representative pictures were taken at day 7–9 and number of acinar- and solid-like colonies counted using an Olympus CKK41 microscope (4x lens). HC11 cell assays HC11 and HEK293T cells [ATCC, USA] were maintained in RPMI 1640 ( + 10% FBS, 1% PSG, 5 µg/ml Insulin, 10 ng/ml mEGF) or DMEM ( + 10% FBS, 1% PSG) media, respectively. HC11 cells authenticated by STR profiling in May 2025. For 2D growth assays, trypsinised cells were plated (5 × 103 viable cells/ml) and supplemented with fresh medium every 2–3 days. For each condition, four 24-wells were counted every 24 h for 2–4 experimental repeats. For mammosphere assays, 5 × 103 viable HC11 cells/ml in mammosphere medium—RPMI 1640, 1% PSG, 20 ng/ml mEGF, mFGF-2, 4 µg/ml heparin, 1X B27 Supplement—were cultured in 24-well Ultra-Low Attachment plates at 5% CO2/37 °C. Cells were supplemented with mEGF and mFGF-2 (20 ng/ml) every 3 days. Recombinant mouse WNT3A protein [100 ng/ml; Abcam, UK] was added to mammosphere medium and supplemented every 3 days. Number of mammospheres per well was counted at day 7 using an Olympus CKX41 Inverted Phase Contrast Microscope (10x lens). Runx1 overexpression & CRISPR/CAS9 gene targeting For Runx1 overexpression, HEK293T cells were transfected with 10 µg of pBABE-Puro (Empty or mRunx1P1), 7.5 µg of psPAX2 [Addgene, USA] and 4 µg of pVSV-G [Addgene] plasmids using calcium phosphate with medium replacement 24 h post-transfection. Virus-containing media, filtered through 0.45 µM, were used to transduce HC11 cells with 10 µg/ml polybrene. Cells were selected with 1.5 µg/ml Puromycin. For CRISPR/CAS9-mediated targeting, guide RNAs (gRNAs) targeting Runx1 and Runx2 were designed using the Zhang Lab tool [MIT, Boston, MA]. pX-459-Puro (Empty or Runx1-targeting gRNAs) constructs were used to transfect HC11 cells using Lipofectamine™ 2000 Transfection Reagent. Three different Runx1-targeting gRNAs were tested (mhRunx1, mRunx1_A, mRunx1_B). Three independent Runx1-deleted (R1KO_a, R1KO_b, R1KO_c) HC11 clones were generated using the mRunx1_B gRNA sequence (5’-CGGTGCGCACTAGCTCGCCA). For Runx1/Runx2 deletion, HC11 cells were transfected with pX-459-Puro (Empty or Runx2-targeting gRNA, i.e. mhRunx2: 5’-GAGCGACGTGAGCCCGGTGG) constructs, then transduced with lentiCRISPRv2-Neo (Empty or Runx1-targeting gRNA, i.e. mRunx1_B) vectors as above. Upon antibiotic selection (800 µg/ml G418 Sulphate/Geneticin) and single-cell cloning, an independent Runx1-deleted (R1KO_d), two independent Runx2-deleted (R2KO_a, R2KO_b) and two independent Runx1/Runx2-deleted (DKO_a, DKO_b) HC11 clones were generated. Immunoblotting and antibodies HC11 cells were lysed in Pierce RIPA Buffer. Protein extracts, resolved on 4-12% NuPAGE Novex Bis-Tris 1.0 mm Mini Protein Gels [Thermo Fisher Scientific, USA], were transferred to Hybond-ECL nitrocellulose membranes [Sigma-Aldrich, USA]. The following antibodies were used: RUNX1 [#8529, D4A6, Cell Signaling, USA, 1:1000]; RUNX2 [#8486, D1H7, Cell Signaling, 1:1000]; GAPDH [#3683, 14C10, Cell Signaling, 1:1000] and horseradish peroxidase-conjugated anti-rabbit secondary antibody [#7074, Cell Signaling, 1:10 000]. Densitometry analysis performed using Fiji ImageJ software (version 1.53). RNA sequencing RNA extraction was performed using the RNAqueous™-Micro Total RNA Isolation Kit [Invitrogen] and residual genomic DNA removed using the DNA-free™ system. tdRFP-positive sorted cells were pooled from 3-6 individual mice and combined into a single 20 µl RNA sample. NanoDrop 2000 Spectrophotometer was used to quantify and assess purity of extracted RNA. RNA quality was assessed on an Agilent 2200 TapeStation System using RNA ScreenTape Analysis. Samples with RNA integrity number (RIN) values ≥ 7 were used wherever possible (18/19 samples in Fig. 6) and RIN values ≥ 6 (8/14 samples) for tdRFP + FACS cells (Figs. 3 and 5). RNA-seq libraries were prepared using SMARTer Stranded Total RNA-Seq Kit v2—Pico Input Mammalian [Takara Bio, USA]. Library quality was assessed using Agilent 2200 TapeStation System with a High Sensitivity D1000 ScreenTape Analysis and quantity measured with a Qubit Fluorometer. Libraries were sequenced by W. Clarke at CRUK Scotland Institute Molecular Technologies Facility via the Illumina NextSeq 500 System with paired-end (2 × 36) sequencing using the MiniSeq High Output Reagent Kit (75-cycles). Illumina sequence data were demultiplexed using bcl2fastq (version 2.20). Quality checks on raw RNA-Seq data files were done using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and FastQ Screen [65]. RNA-Seq paired-end reads were aligned to the GRCm38.93 version of the mouse genome and annotation [66] using HiSat2 [67] and sorted using Samtools [68]. Aligned genes were identified using HTSeq [69]. Expression levels were determined and statistically analysed using the R environment (https://www.r-project.org/) and utilising packages from the Bioconductor data analysis suite [70]. Differential gene expression was analysed based on the negative binomial distribution using DESeq2 [71] and adaptive shrinkage using Apeglm [72]. Identification of enriched biological functions was achieved using g:Profiler [73], GESA from the Broad Institute [74] and MetaCore from Clarivate Analytics (https://portal.genego.com/). Computational analysis was documented at each stage using MultiQC [75], Jupyter Notebooks (https://eprints.soton.ac.uk/403913/) and R Notebooks (https://posit.co/). CIBERSORTx analysis Signature matrix was generated from an annotated single-cell RNA-seq dataset of murine glands [47] to provide a reference to impute cell fractions from gene expression profiles. Signature matrix and normalised bulk RNA-seq datasets were run through CIBERSORTx to deconvolute the bulk RNA-seq datasets (https://cibersortx.stanford.edu). CIBERSORTx job was run under Impute Cell Fractions settings, relative run mode, >100 permutations for significance analysis, and with quantile normalisation disabled. CIBERSORTx results were exported as CSV files and imported into R Studio (version 4.3.2) for visualization. mMCP counter analysis mMCPCounter (mMCPcounter.estimate) is a function within the mMCPcounter (version 1.1.0) R package in R Studio [45]. mMCPcounter.estimate was used to generate scores for 16 cell populations. Heatmaps of cell population Z-scores were plotted using the ComplexHeatmap (version 2.14.0) R package. To compare statistics of the boxplots for the different cell populations, the stats (version 4.2.2) R package was used to perform a t test between each of the conditions. Statistical analysis Statistical analyses performed using GraphPad Prism (version 10). Specific statistical tests used are indicated in figure legends. All error bars represent mean ± SD, unless otherwise indicated. Supplementary information Supplementary Figures Supplementary Tables legend Supplementary Table 1 Supplementary Table 2 Supplementary Figures Supplementary Tables legend Supplementary Table 1 Supplementary Table 2