The membrane-linked adaptor FRS2β fashions a cytokine-rich inflammatory microenvironment that promotes breast cancer carcinogenesis

To examine the role of FRS2β in vivo, we mutated Frs2β in mice by gene targeting (SI Appendix, Fig. S1 A and B). FRS2β is known to be expressed abundantly in brain tissues (16, 17). Immunoblotting of brain lysate confirmed that the mutant mice were deficient for FRS2β protein by using the anti-FRS2β antibody, which we previously evaluated the specificity for immunoblotting, immunofluorescence, and immunohistochemistry (20) (SI Appendix, Fig. S1C). The genotype frequencies of offspring born to the heterozygous breeding pairs were in agreement with Mendelian ratio (SI Appendix, Fig. S1D). The mutant mice grew normally and were fertile, with no gross abnormalities. Frs2β promoter activity was detected by β-galactosidase (LacZ) staining of the heterozygous for the Frs2β mutant allele (SI Appendix, Fig. S1A). In the adult brain, the staining was observed in neural cells in cerebral cortex and hippocampus and purkinje cells in cerebellum, for example (SI Appendix, Fig. S1 E and F). Similarly, immunohistochemistry with anti-FRS2β showed cytoplasmic staining in neural cells in hippocampus, as reported previously (20) (SI Appendix, Fig. S1G). In the mature female mammary tissues, the abundance of Frs2β transcript increased significantly during pregnancy and lactation and then decreased during the involution period after weaning (SI Appendix, Fig. S1H). LacZ-stained cells were observed in several alveolar cells in mammary tissues during lactation period (Fig. 1 A and B). Immunohistochemistry revealed that FRS2β was expressed in a small subset of luminal cells that are positive for cytokeratin 18, a luminal cell marker (21) (Fig. 1 C–E). However, FRS2β was not expressed in myoepithelial cells that are positive for cytokeratin 14, a myoepithelial cell marker (Fig. 1 C and E and SI Appendix, Fig. S1I). These data suggest that a subset of luminal cells in the mammary gland express FRS2β.

Whole-mount staining of the mammary gland revealed no gross structural abnormality in the mutant mice (SI Appendix, Fig. S2A). There were no significant differences in number and length of the mammary tree branches (SI Appendix, Fig. S2 B and C). Although we observed modest reduction of alveolar cells in Frs2β (^−/−) mammary tissues during lactation period (SI Appendix, Fig. S2 D and E), the production of milk of Frs2β (^−/−) mammary tissues should be sufficient for raising pups, since there were no differences in body weight of offspring born to either Frs2β (^+/+) or Frs2β (^−/−) mother mice (SI Appendix, Fig. S2F).

Although Frs2β (^−/−) mice have very minor phenotype in normal conditions, we hypothesized that FRS2β is involved in pathological conditions, such as tumorigenesis. To test this hypothesis, we used MMTV-ErbB2 (+) mammary tumor model mice. We found that the number of FRS2β–positive luminal cells in the mammary gland during lactation period was greater in MMTV-ErbB2 (+) mice than in MMTV-ErbB2 (−) mice (SI Appendix, Fig. S2 G and H). We crossed Frs2β mutant mice with the MMTV-ErbB2 (+) mice to generate MMTV-ErbB2 (+)/Frs2β (^+/+), MMTV-ErbB2 (+)/Frs2β (^±), and MMTV-ErbB2 (+)/Frs2β (^−/−) mice hereafter referred to as Frs2β (^+/+), Frs2β (^±), and Frsβ (^−/−), respectively. We detected palpable tumors in Frs2β (^+/+) mice at 23.4 ± 1.9 wk after birth, and 83% of mice had tumors (n = 8) after experiencing pregnancy and subsequent lactation after the age of 8 wk. On the other hand, tumors became palpable in virgin Frs2β (^+/+) mice at 32.6 ± 2.6 wk, and only 23.4% of mice had tumors (n = 8) (i.e., later and at a lower probability). Since pregnancy and subsequent lactation seem to be required for efficient tumorigenesis, we analyzed mice after pregnancy and lactation (Fig. 1F). We used MRI, which is sufficiently sensitive to detect very small cell masses, even those with 1-mm diameters (22). We began to observe small tumors (>5 mm in diameter) 5 to 8 wk after commencement of measurement (19 to 22-wk-old mice) (SI Appendix, Fig. S3A). Strikingly, the tumor growth rate was much lower in Frs2β (^−/−) mice than in Frs2β (^+/+) mice (Fig. 1 G and H), although tumor incidence was similar: 83.2% (n = 18) in Frs2β (^+/+) and 88.2% (n = 17) in Frs2β (^−/−). This observation indicates that FRS2β plays important roles in mammary tumorigenesis.

Histological examination revealed that Frs2β (^+/+) tumors contained ample stroma, reminiscent of human breast cancer tissues (23) (Fig. 1 I, Upper, and SI Appendix, Fig. S3B). By contrast, very little stroma was observed in Frs2β (^−/−) tumors. High levels of smooth muscle actin (SMA)–positive CAFs were present in the stroma of Frs2β (^+/+) tumors but not in Frs2β (^−/−) tumors (Fig. 1 I, Lower and J). The tumor stroma is a major component of the tumor microenvironment. We next analyzed the amount of blood vessels, another important component of the tumor microenvironment, by measuring the amount of CD31-positive endothelial cells. We found that the amount of CD31-positive endothelial cells was significantly smaller in Frs2β (^−/−) tumors than in Frs2β (^+/+) tumors (SI Appendix, Fig. S3 C and D). These results indicate that FRS2β is required for formation of tumor stroma that includes blood vessels.

We next investigated if FRS2β plays roles in creating the mammary tissue microenvironment required for tumorigenesis prior to tumor formation. To test this hypothesis, we performed xenograft experiments in which Frs2β (^+/+) tumor cells were inoculated into 8-wk-old young virgin premalignant mammary tissues of Frs2β (^+/+) and Frs2β (^−/−) mice. We first removed the Frs2β (^+/+) tumors (∼1 cm in diameter) from MMTV-ErbB2 mice and cultured the tumor cells as spheres in serum-free suspension in sphere culture medium with a cytokine mixture, as CSCs are enriched under these conditions (24, 25) (Fig. 2A). To facilitate tumorigenesis, we used these tumor sphere cells for xenografts. After we made single-cell suspension from the tumor spheres, we inoculated them into 8-wk-old Frs2β (^+/+) and Frs2β (^−/−) virgin mammary tissues and monitored tumorigenesis (Fig. 2A). Strikingly, while large tumors formed rapidly in transplanted Frs2β (^+/+) mice, only very small cell masses (<3 mm in diameter) were observed in Frs2β (^−/−) mammary tissues (Fig. 2 B and C and SI Appendix, Fig. S4 A and B). Limiting dilution assay revealed that no tumor-initiating ability was observed in tumor cells inoculated in Frs2β (^−/−) mammary tissues (Fig. 2 D and E). These results indicate that tumor cells grew well in the Frs2β (^+/+) mammary tissue microenvironment but not in the Frs2β (^−/−) mammary tissue microenvironment. On the other hand, tumors did not form when Frs2β (^+/+) tumor cells were inoculated into the Frs2β (^+/+) male mammary fat pads (SI Appendix, Fig. S4C). We also inoculated 10,000 Frs2β (^+/+) tumor cells subcutaneously into the flanks of Frs2β (^+/+) female mice and found no palpable tumors after 35 d. From these findings, we confirmed that the Frs2β (^+/+) female mammary tissue microenvironment is essential for tumorigenesis. Therefore, premalignant mammary gland cells expressing FRS2β create a microenvironment that is essential for tumorigenesis.

Next, we examined the molecular mechanisms by which FRS2β expressed in luminal cells create a microenvironment that is essential for tumorigenesis. For this purpose, we cultured Frs2β (^+/+) and Frs2β (^−/−) premalignant mammary epithelial cells derived from 8- to 12-wk-old mice and compared their transcriptomic profiles using DNA microarrays. Gene set enrichment analysis revealed that gene sets related to NF-κB targets, stem cell function, and stroma were enriched in Frs2β (^+/+) cells relative to Frs2β (^−/−) cells (Fig. 2F). In contrast, the ERK pathway–related gene set was up-regulated in Frs2β (^−/−) cells relative to Frs2β (^+/+) cells, which was confirmed by increased expression levels of phosphorylated ERK1/2 in Frs2β (^−/−) cells by immunoblotting (SI Appendix, Fig. S4 D and E). This was expected because FRS2β inhibits ERK signaling (15).

Many genes encoding cytokines were up-regulated in Frs2β (^+/+) cells (SI Appendix, Table S1); among them, the top 10 genes at higher expression levels in Frs2β (^+/+) cells than in Frs2β (^−/−) cells were indicated (Fig. 2G). We then focused on Igf1 and Cxcl12, both of which were included in the gene sets related to NF-κB targets. Igf1 and Cxcl12 were also included in the gene sets related to stem cell function and stroma, respectively. qPCR confirmed that Igf1 and Cxcl12 transcripts were expressed at higher levels in Frs2β (^±) than in Frs2β (^−/−) premalignant mammary epithelial cells (Fig. 2H). Immunohistochemistry revealed that the protein levels of IGF1 and CXCL12 were higher in the cytoplasm of Frs2β (^+/+) premalignant luminal cells than in Frs2β (^−/−) cells (Fig. 2 I–L and SI Appendix, Fig. S5 A and B). Inflamed areas in which immune cells were accumulated were observed in the Frs2β (^+/+) premalignant mammary tissues but not in Frs2β (^−/−) premalignant mammary tissues (Fig. 2 M and N and SI Appendix, Fig. S5C). These results indicate that the premalignant mammary epithelial cells produce ample amounts of cytokines, including IGF1 and CXCL12, and that cytokine production was significantly reduced in FRS2β-deficient mammary epithelial cells.

To examine whether the up-regulation of these cytokines was due to overexpression of ErbB2, we compared the expression levels of IGF1 and CXCL12 in premalignant mammary tissues between MMTV-ErbB2(−)/Frs2β (^+/+) mice and MMTV-ErbB2(+)/Frs2β (^+/+) mice using immunohistochemistry and qPCR. We observed that the expression levels were up-regulated in MMTV-ErbB2(+)/Frs2β (^+/+) premalignant mammary tissues as compared to those of MMTV-ErbB2(−)/Frs2β (^+/+) mice (SI Appendix, Fig. S6 A–C). These findings indicate that overexpression of ErbB2 causes up-regulation of IGF1 and CXCL12 in the premalignant mammary tissues.

We next examined in which type of luminal cells FRS2β is expressed. We sorted the mammary epithelial cells derived from 10-wk-old virgin Frs2β (^+/+) or Frs2β (^−/−) mice by their expression of surface markers. It is known that the progenitor cells are enriched in the CD49f^low/CD24^high cell population (26) (SI Appendix, Fig. S4 F, Right, P3, and G, Left, P3) and that the luminal progenitor cells can be enriched by further fractionation with CD61 for the CD49f^low/CD24^high/CD61⁺ population (27) (SI Appendix, Fig. S4G, Middle, P5). After permealization of the CD49f^low/CD24^high/CD61⁺ cells, we next analyzed them by using anti–FRS2β antibody. We observed significant expression of FRS2β in 23.6% cells among the CD49f^low/CD24^+high/CD61⁺ luminal progenitor cell population (SI Appendix, Fig. S4G, Upper Right, P6). We confirmed that FRS2β was lost in CD49f^low/CD24^high/CD61⁺ luminal progenitor cell population derived from Frs2β (^−/−) mammary epithelial cells (SI Appendix, Fig. S4G, Lower Right, P6). These data suggest that a subset of luminal progenitor cells in the mammary gland express FRS2β.

We then cultured premalignant mammary epithelial cells as mammospheres in which stem/progenitor cells are enriched (28). We found that the efficiency of mammosphere formation was greater in Frs2β (^+/+) premalignant mammary epithelial cells than in Frs2β (^−/−) cells in several passages (SI Appendix, Fig. S4 H and I). Our findings suggest that FRS2β contributes to growth ability of macroscopically normal mammary stem/progenitor cells.

Tumor sphere formation reflects the properties of CSCs, whose growth is dependent on cytokines in the culture (24). To determine whether IGF1 derived from premalignant mammary epithelial cells plays a role in tumor sphere formation, we cultured Frs2β (^+/+) tumor cells under serum-free suspension condition without the standard cytokine mixture in the presence or absence of premalignant Frs2β (^+/+) mammary epithelial cells (Fig. 3A). We observed tumor sphere formation by Frs2β (^+/+) tumor cells in the presence of Frs2β (^+/+) premalignant mammary epithelial cells but not in their absence (compare control versus not cocultured in Fig. 3 B and C). Treatment with an IGF1-neutralizing antibody greatly diminished tumor sphere formation by Frs2β (^+/+) tumor cells cocultured with Frs2β (^+/+) premalignant mammary cells (Fig. 3 B and C). These findings indicate that IGF1 derived from nearby Frs2β (^+/+) premalignant mammary epithelial cells plays an important role in tumor sphere formation. Thus, IGF1 derived from Frs2β (^+/+) premalignant mammary epithelial cells may support CSC growth.

To determine whether CXCL12 derived from the premalignant mammary epithelial cells plays roles in CAFs, we obtained Frs2β (^+/+) CAFs from the MMTV-ErbB2 mammary tumors. They had spindle morphology and were positively stained by anti-αSMA and anti-vimentin antibodies, which are typical features of CAFs (29) (SI Appendix, Fig. S7 A and B). We cocultured the Frs2β (^+/+) CAFs with Frs2β (^+/+) or Frs2β (^−/−) premalignant mammary epithelial cells (Fig. 3D). We observed significantly more migrated CAFs in cocultures with Frs2β (^+/+) premalignant mammary epithelial cells than in those with Frs2β (^−/−) cells (Fig. 3 E and F). CXCL12 binds to its receptors CXCR4 and CXCR7 (30). Although we did not observe significant effects on the migration of CAFs upon treatment with the CXCR4 inhibitor AMD3100 (100 ng/mL) or the CXCR7 inhibitor CCX771 (100 ng/mL) alone (SI Appendix, Fig. S7C), a combination of both inhibitors greatly diminished migration of CAFs in a CCX771 dose-dependent manner in the presence of the effective dose of AMD3100 (Fig. 3 G and H and SI Appendix, Fig. S7D). Furthermore, we observed increased migration of CAFs by addition of recombinant CXCL12 in the culture media in place of Frs2β (^+/+) premalignant mammary epithelial cells (SI Appendix, Fig. S7 E and F). These findings indicate that CXCL12 derived from nearby Frs2β (^+/+) premalignant mammary epithelial cells contributes to mobilization of CAFs. Taken together, the FRS2β–dependent increases in production of cytokines, including IGF1 and CXCL12, enable premalignant mammary cells to maintain CSC and mobilized CAF at the onset of tumorigenesis.

Next, we examined the molecular mechanisms that induce expression of IGF1 and CXCL2 in premalignant mammary tissues. Since Igf1 and Cxcl12 were among list of the identified NF-κB target gene (Fig. 2F), we investigated if the activation of NF-κB is involved in the production of these cytokines. To this end, we cultured Frs-2β (^+/+) premalignant mammary epithelial cells and treated them with DHMEQ, a specific inhibitor of NF-κB (Fig. 4A). Treatment with DHMEQ inhibited the expression of Igf1, Cxcl12, and IκBα, a well-established NF-κB–inducible gene, in a dose-dependent manner (Fig. 4B). We further confirmed these results by using two unrelated NF-κB inhibitors, TPCA-1 and SC514. Treatment with TPCA-1 or SC514 inhibited the expression of Igf1, Cxcl12, and IκBα in a dose-dependent manner, similar to treatment with DHMEQ (SI Appendix, Fig. S8). We next searched potential NF-κB binding sites in the 5′ regulatory regions of Igf1 and Cxcl12 genes by using ChIP-Atlas (http://chip-atlas.org) and found them in each region. Chromatin immunoprecipitation assay showed that RelA, an NF-κB subunit, was strongly recruited to the 5′ regulatory regions of Igf1 and Cxcl12 genes in Frs-2β (^+/+) premalignant mammary epithelial cells compared to those of Frs-2β (^−/−) (Fig. 4C). This suggests that NF-κB activation plays important roles in the expression of IGF1 and CXCL12 in premalignant mammary epithelial cells.

We then examined activation of NF-κB in premalignant mammary tissues in vivo. Immunoblotting of lysates from premalignant mammary tissues revealed that higher amounts of the NFκB components RelA and p50 in the nucleus, a higher level of phosphorylated IKK-β, and a lower level of IκBα in Frs2β (^+/+) tissues relative to Frs-2β (^−/−) tissues (Fig. 4 D and E). Moreover, immunohistochemistry revealed that RelA was localized to the nucleus in a much greater proportion of Frs2β (^+/+) than Frs2β (^−/−) premalignant luminal cells (Fig. 4 G, Left and Middle and H; compare values indicated by left bar versus those of middle bar). Treatment with DHMEQ in vivo dramatically decreased the number of Frs2β (^+/+) premalignant luminal cells harboring RelA in the nucleus (Fig. 4 F, G, Right, and H; compare Frs2β (^+/+) and DHMEQ − versus Frs2β (^+/+) and DHMEQ +) and inhibited expression of Igf1 and Cxcl12 transcripts in premalignant mammary tissues (Fig. 4I). These results suggest that NF-κB activation in luminal cells plays important roles in the expression of IGF1 and CXCL12 in premalignant mammary epithelial cells in vivo. It thus appears that FRS-2β triggers activation of NF-κB in the premalignant luminal cells, thereby inducing the production of cytokines including IGF1 and CXCL12 that in turn activate NF-κB in an autocrine and/or paracrine manner to spread the effects of NF-κB activation to the surrounding mammary epithelial cells.

To determine whether IGF1 and CXCL12 expressed in premalignant mammary epithelial cells contribute to tumorigenesis, we treated Frs2β (^+/+) xenograft mice with the IGF1-neutralizing antibody or a combination of CXCR4 and CXCR7 inhibitors, which block both the receptors for CXCL12 (hereinafter called “CXCL12 inhibitor”). Treatment with either the IGF1-neutralizing antibody or the CXC12 inhibitor significantly suppressed tumorigenesis and combined treatment with the IGF1-neutralizing antibody, and the CXC12 inhibitor yielded the strongest suppression of tumor volume and weight (Fig. 4 I–L). We used effective doses of the inhibitors that were well tolerated without toxic effects, as previously reported (31, 32). These results indicate that the production of IGF1 and CXCL12 in premalignant mammary tissues creates a microenvironment that is necessary for tumorigenesis.

We next examined how FRS2β induces activation of NF-κB. Immunofluorescence staining for FRS2β showed a punctate cytoplasmic distribution in higher-magnification (Fig. 1D) reminiscent of its localization on intracellular vesicles, as previously reported (20). Similarly, it was reported that ErbB2 is localized not only on the plasma membrane but also on intracellular vesicles (33). To examine whether FRS2β is localized on the early endosomes, we stained cultured mammary epithelial cells derived from MMTV-ErbB2 (+)/Frs2β (^+/+) mice or MMTV-ErbB2 (−)/Frs2β (^+/+) wild-type mice with antibodies against FRS2β and an early endosomal marker EEA1. These antibodies were validated in this study (Fig. 1D and SI Appendix, Figs. S1 C, G, I and S2G) and in previous studies (34). We examined colocalization of FRS2β and EEA1 in FRS2β–positive cells. Higher levels of colocalization of FRS2β and EEA1 were observed in cells derived from MMTV-ErbB2 (+)/Frs2β (^+/+) than MMTV-ErbB2 (−)/Frs2β (^+/+) wild-type mice (Fig. 5A), indicating that FRS2β is localized on the early endosomes more efficiently in ErbB2-overexpressing cells. These results suggest that ErbB2 transported from the plasma membrane recruits FRS2β on the early endosomes since FRS2β binds to ErbB2 (15). This finding led us to hypothesize that ErbB2–FRS2β triggers activation of NF-κB on the early endosomes.

We focused on NEMO, a scaffold subunit of the IKK complex (3). To demonstrate the specificity of the anti-NEMO antibodies, we compared the immunofluorescent staining of cells following RNA interference knockdown of Nemo. Immunofluorescence staining with anti-NEMO antibodies showed a punctate cytoplasmic distribution, which was abolished following Nemo knockdown (SI Appendix, Fig. S9 A and B). We thus confirmed the specificity of the anti-NEMO antibody. We next stained cells derived from MMTV-ErbB2 (+)/Frs2β (^+/+) mice with antibodies against NEMO and EEA1. Notably, higher levels of colocalization of NEMO and EEA1 were observed in FRS2β–positive cells than in FRS2β-negative cells (Fig. 5B). This suggests that NEMO is localized on early endosomes more efficiently in FRS2β-expressing cells. These cells were stained with antibodies against TAK1, an upstream activator of the IKK complex, or IKK-β and EEA1. Higher levels of colocalization of TAK1 or IKK-β and EEA1 were observed in FRS2β-positive cells than in FRS2β-negative cells (SI Appendix, Fig. S9 C and D). This suggests that TAK1 and IKK-β are also more efficiently localized on early endosomes in FRS2β-expressing cells. We further stained cells derived from MMTV-ErbB2 (+)/Frs2β (^+/+) or MMTV-ErbB2 (−)/Frs2β (^+/+) wild-type mice and examined the colocalization of NEMO and EEA1 in FRS2β–positive cells. We found a greater degree of colocalization between NEMO and EEA1 in ErbB2-overexpressing cells (Fig. 5C). Finally, elevated colocalization of FRS2β and NEMO on early endosomes was observed in cells derived from MMTV-ErbB2 (+)/Frs2β (^+/+) mice than MMTV-ErbB2 (−)/Frsβ (^+/+) wild-type mice (Fig. 5D). Collectively, it is likely that FRS2β facilitates recruitment of TAK1 and IKK complex that includes NEMO and IKK-β to the early endosomes, resulting in the activation of the IKK complex (Fig. 5E and SI Appendix, Fig. S9E).

We next examined Frs2β (^+/+) and Frs2β (^−/−) mammary tumors during malignant progression (Fig. 1 G–I). Immunohistochemistry revealed that a few FRS2β-positive tumor cells were observed in Frs2β (^+/+) tumor tissues, while no FRS2β–positive cells were observed in Frs2β (^−/−) tumor tissues as expected (Fig. 6A). Since CSCs are thought to originate from luminal progenitor cells (24, 35), it is reasonable to assume that FRS2β is expressed in a part of CSCs and tumor cells. Expression levels of Igf1 and Cxcl12 transcripts were higher in Frs2β (^±) tumors than in Frs2β (^−/−) tumors (Fig. 6B). Immunohistochemistry confirmed that expression levels of IGF1 and CXCL12 were higher in Frs2β (^+/+) tumors than in Frs2β (^−/−) tumors (Fig. 6 C and D). The efficiency of tumor sphere formation was greater in Frs2β (^+/+) tumor cells than in Frs2β (^−/−) tumor cells (SI Appendix, Fig. S10 A and B). Frs2β (^+/+) tumor sphere formation was diminished by treatment with inhibitors for IGF1 receptor, a receptor for IGF1, linsitinib, or AEW541 (SI Appendix, Fig. S10 C and D). These findings suggest that IGF1 derived from tumors cells support CSC growth. We further examined whether IGF1 and CXCL12 contributed to CSC growth using knockdown of Igf1 and Cxcl12 by siRNAs in tumor cells. We found that the efficiency of tumor sphere formation was significantly diminished in cells after the knockdown (SI Appendix, Fig. S10 E–H). These results suggest that both IGF1 and CXCL12 in tumor cells contribute to the growth of CSCs and tumorigenesis. Collectively, FRS-2β expressed not only in premalignant epithelial cells but also in tumor cells may contribute to IGF1 and CXCL12 production, leading to CSC growth and establishment of the stroma-rich tumor microenvironment during malignant progression.

Finally, we examined the expression of FRS2β in human breast cancer tissues in tissue array by immunohistochemistry. There was a positive correlation between expression levels of FRS2β and the amount of stroma (Fig. 6 E and F). Furthermore, the analysis of published gene expression profiles (36) revealed that patients with higher expression levels of FRS2β in breast cancer tissues had poorer prognosis (Fig. 6G). Since the dataset included information about immunohistochemical staining for estrogen and progesterone receptors in cancer tissues, we stratified the patients with or without these receptors in their cancer tissues. We found that higher-expression levels of FRS2β had a poorer prognosis among patients with estrogen or progesterone-positive breast cancer (SI Appendix, Fig. S11 A–D).

We analyzed expression levels of FRS2β in different subtypes of breast cancer tissues using a METABLIC dataset consisting of 2,000 primary breast cancers (37). Expression levels of FRS2β were significantly lower in the HER2-positive subtype than in other subtypes (SI Appendix, Fig. S11E). To further stratify the HER2 subtype, we used another method, expression-based classification (37). We found that expression levels of FRS2β were significantly lower in HER2-positive and hormone receptor–negative subtype than in other subtypes (SI Appendix, Fig. S11F). These findings raise the hypothesis that a lower amount of FRS2β might be sufficient for HER2-positive breast cancer tissues and that hormone receptor signaling might contribute to an increase in the expression levels of FRS2β.

All images and data shown are representative of several experiments performed on different days. Each experiment was repeated independently with similar results. Statistics were computed using GraphPad Prism (version 8). For normally distributed datasets, significance was calculated by unpaired, two-tailed Student’s t test. Data are presented as means ± SEM or means ± SD. For nonparametric datasets, Mann–Whitney U tests were performed. Data are presented as means ± SEM or means ± SD. The tumor growth datasets were analyzed using the Bonferroni-corrected two-way ANOVA to compute statistical significance. Data are presented as means ± SEM. Tumor-initiating cell frequency was analyzed by Extreme Limiting Dilution Analysis software (bioinf.wehi.edu.au/software/elda/). The quantification of immunoblotting results was performed using ImageJ software. Kaplan–Meier survival curves were analyzed by log-rank test. Quantification of staining results of tissue array was performed using WinROOF software. Pearson’s and Spearman’s correlation (two sided) were performed to determine correlation between expression levels of FRS-2β in tumor cells and amount of stroma in tumor tissues.

Mice were handled according to the guidelines of the Institute of Medical Science at The University of Tokyo and Kanazawa University. The experiments were approved by the ethics committees for Animal Research at the Institute of Medical Science, The University of Tokyo, and Kanazawa University.