Local and systemic factors during lactation promote breast cancer
To study PABCs in vivo, we utilized a model we previously described [36] in which carcinoma cells are introduced directly into the fourth inguinal mammary glands of mice at the five major stages of postnatal development: post-pubertal (nulliparous), pregnancy, lactation, involution and upon full regression of the gland following involution (Figure 1A-C). To limit overlap of tumor growth across different developmental stages, we utilized a syngeneic model of triple negative basal like breast cancer that rapidly forms primary mammary carcinomas within three weeks (Figure 1C).
Consistent with earlier findings [16, 19], 4 T1-12B carcinoma cells formed larger tumors in involuting mammary glands relative to nulliparous glands (P = 0.03) (Figure 1D). However, tumors harvested from pregnant and lactating mice were on average 3.5-fold larger than their counterparts in nulliparous mice (201 mg versus 58 mg; P <0.01) and 2-fold larger than in involuting mice (201 mg versus 98 mg; P = 0.045) (Figure 1D, Additional file 2). These results suggest that factors present during pregnancy and lactation contribute more strongly to tumor growth than during involution. Due to the aggressive nature of the 4 T1-12B model, no obvious histopathological differences were observed between 4 T1-12B tumors that formed between the different stages of mammary gland development.
Circulating levels of estrogen and progesterone are at their highest during pregnancy, while the levels of prolactin, a potent mitogen for breast cancer cells, are at their peak during pregnancy and lactation [37, 38]. Since these hormones are powerful modulators of mammary epithelial cell proliferation, the accelerated breast tumor growth during pregnancy and lactation could be due to differences in systemic hormone levels. An alternative possibility is that local alterations in cellular composition and phenotypes within the mammary gland microenvironment could be sufficient to promote the observed increase in tumor growth during pregnancy and lactation. To evaluate the latter possibility, we co-mixed 4 T1-12B cells with the ASCs isolated from mice at each developmental stage and injected the co-mixture into nulliparous mice. Since nulliparous mice have comparable systemic hormone profiles, any differences in tumor growth would be attributable to the admixed stromal cells.
Although injection into mammary glands of pregnant, lactating and involuting mice accelerated tumor formation in the experiments above (Figure 1D), only ASCs derived from lactating mammary glands could accelerate 4 T1-12B mammary tumor growth in nulliparous mice (P <0.001, Figure 1E). To determine whether the difference in tumor growth rates might be associated with differential cell survival following implantation, injection sites of early 4 T1-12B tumors were evaluated and viable cells were observed within the mammary glands one week following injection, suggesting that the delay in early tumor growth was not due to immune clearance of cells or increased cell death. To test whether the tumor-enhancing effects of ASCs derived from lactating mammary glands extended to other breast epithelial models, we repeated the above experiment with COMMA-D cells, which are weakly to non-tumorigenic and when transformed also model triple negative basal like breast cancers [39]. While COMMA-D cells did not form large tumors when injected alone, they did grow slowly when co-mixed with ASCs from nulliparous, pregnant, involuting or regressed mammary glands. There was a trend towards faster growth in the tumors co-mixed with ASCs isolated from pregnant and involuting glands; however, this trend was not statistically significant. In contrast to ASCs from other stages of development, co-mixture with ASCs from lactating glands resulted in significant acceleration of tumor formation by COMMA-D cells (P = 0.03, Figure 1F). These findings indicate that while systemic factors likely play a key role in the development of tumors during pregnancy and involution, local alterations in either adipose stromal cell composition or function during lactation strongly contribute to breast cancer growth.
Characterization of the adipose stroma during mammary gland development
The proportion of the mammary gland occupied by stroma is known to vary significantly depending on the differentiation state of the adjacent epithelium [22]. We, therefore, undertook a detailed histological and cellular analysis of the adipose stroma across the different stages of post-natal development to characterize stromal adipocyte remodeling during different mammary epithelial differentiation stages. Quantification of adipocyte occupancy revealed that they occupy the largest proportion of the gland in nulliparous and regressed glands (97.2% and 96.1%, respectively), while adipocytes comprise the smallest proportion of the gland during lactation (11.4%, P <0.001, Figure 2A). Adipocytes present in nulliparous and regressed tissues were variably sized and were both multilocular and unilocular (Figure 2B). In contrast, adipocytes present in pregnant mice were uniquely unilocular and significantly larger in size, consistent with a lipid storage phenotype (P <0.001, Figure 2C). Notably, adipocytes present in lactating tissues were markedly distinguishable from adipocytes in any other developmental stage. Not only were lactation-associated adipocytes significantly smaller in size (P = 0.01) but, more importantly, they exhibited features consistent with lipolysis [40] including prominent membrane folding and vesicle formation (Figure 2B arrows).
To examine further the mammary stroma, ASCs were isolated from mammary fat pads of several mice at different developmental stages. Although ASCs consist of heterogeneous populations of cell types including adipocyte stem cells, preadipocytes, mature adipocytes and other stromal cells, we found that the ASCs from all developmental stages isolated from three different mice all expressed similar levels of mature adipocyte markers including FABP4 and HSL (Figure 2D,E) indicating that the representation of mature adipocytes within the cultures was similar. However, when examined for their ability to accumulate and store lipid, Oil Red O staining revealed that ASCs from lactating glands failed to accumulate lipid while ASCs from nulliparous, pregnant, involuting and regressed glands all induced significant lipid accumulation under identical conditions (Figure 2F,G).
Since the lactating-derived ASCs failed to accumulate lipid, we reasoned this might be due to (1) the ability to retain lipid (lipid metabolism) differed between ASCs isolated from different developmental stages, (2) the induction of adipogenic differentiation differed between ASCs isolated from different developmental stages, or (3) there were significant differences in the representation of cell types within ASCs isolated from different developmental stages that could affect lipid accumulation. To evaluate these possibilities, we assayed ASCs from mammary glands of nulliparous, pregnant, lactating and regressed mice for lipolyis, adipogenesis and cellular composition.
Lipid metabolism is maintained by the highly regulated balance of lipogenesis and lipolysis [41]. During lipid catabolism, lipids are broken down into glycerol and free fatty acids, a process that can be measured biochemically. We quantified the levels of glycerol released from ASCs under differentiation conditions and compared them to the expression level of lipoprotein lipase, an enzyme required for lipogenesis [42]. ASCs isolated from lactating mammary glands exhibited a significantly higher ratio of lipolysis to lipogenesis compared to nulliparous- and pregnant-derived cells (P = 0.01, Figure 2H, Additional file 3).
Adipocyte differentiation involves a stepwise program triggered by the suppression of Klf2 expression with the concomitant induction of PPARγ and FABP4 [43]. We observed that ASCs isolated from all developmental stages, including lactation, were able to activate PPARγ and FABP4 expression and repress Klf2 expression in response to differentiation conditions (Figure 2I). However, compared to ASCs from other stages, the magnitude of PPARγ and FABP4 induction in ASCs from lactating and involuting mammary glands was reduced, suggestive of an attenuated, although not blocked, differentiation potential of adipocytes.
We examined ASC cultures for the presence of other contaminating mesenchymal cell types that could affect adipogenesis including preadipocytes (Pref1), endothelial cells (CD31), angioblasts (CD34), mesenchymal stem cells (Sca-1 and CD29) and macrophages (F4/80). However, no significant differences were observed between ASCs derived from lactating mammary glands compared to any of the other developmental stages that could account for their failure to undergo adipogenesis or lipid accumulation (Additional file 4). Furthermore, gene-expression profiling of ASCs from nulliparous, involuting and lactating mammary glands from multiple mice indicated that very few genes were differentially expressed (Additional files 5, 6 and 7). GSEA further revealed that ASCs from lactating mammary glands were highly enriched in genes associated with adipocytes including linoleic acid metabolism (P <0.0001), phospholipase A2 activity (an enzyme that releases fatty acids from glycerol, P <0.0001) and lipid homeostasis (Figure 2J, P = 0.012). Together, these results suggest that ASCs from lactating mammary glands (hereafter termed lactation-derived adipose cells (ASC-Ls)) are enriched in adipocytes that are lipolytic, unable to accumulate lipid and are unable to fully execute the adipogenic differentiation program.
Tumor-promoting functions of ASC-Ls
To determine the functional properties of ASC-Ls their gene expression signature was queried for known biological process by GSEA. ASC-Ls were enriched in genes involved in extracellular matrix production, immune/inflammatory response and cytokine-cytokine receptor interactions compared to nulliparous-derived adipose cells (ASC-Ns) or involuting derived adipose (ASC-Is) cells (Figure 3A, 3B; Additional files 8, 9, 10 and 11). In addition, compared to ASC-Ns or regressed-derived adipose cells (ASC-Rs), ASC-Ls were highly enriched in the gene set termed ‘tumor angiogenesis’ (P <0.0001) and, to a lesser degree, ‘angiogenesis’ (Figure 3A, 3B; Additional files 8, 9 and 10). This finding prompted us to assess whether the tumor promoting activity of ASC-Ls might be through angiogenesis.
Primary mouse heart endothelial cells (MHECs) were treated with conditioned medium (CM) harvested from ASC-Ns, ASC-Rs or ASC-Ls and examined for proliferation, wound healing and tube formation. Increased proliferation of MHECs was observed following stimulation with CM from ASC-Ls (P <0.001) compared to cells treated with CM from ASC-Rs (Figure 3C). MHEC tube formation was also significantly augmented in the presence of CM from ASC-Ls compared to CM from ASC-Ns or ASC-Rs (P <0.001, Figure 3D, Additional file 12A). Furthermore, wound healing assays revealed that MHECs migrated significantly faster when treated with CM from ASC-Ls when compared to CM from ASC-Ns or ASC-Rs (P = 0.01, Figure 3E).
To determine if the increased angiogenic and inflammatory activity of ASC-Ls was contributing to their ability to accelerate tumor formation, 4 T1-12B cells were co-mixed with ASC-Rs or ASC-Ls and injected into nulliparous mice. Macrophage recruitment and tumor angiogenesis was assessed by immunofluorescence (IF). No significant difference was found in the number of F4/80 positive macrophages recruited to tumors derived from 4 T1-12B cells alone, 4 T1-12B cells co-mixed with ASC-Ls or 4 T1-12B cells co-mixed with ASC-Rs (Additional file 12B). However, there was a 10-fold increase in the number of CD31 positive endothelial cells present within tumors derived from 4 T1-12B cells co-mixed with ASC-Ls (P <0.001, Figure 3F) compared to tumors from 4 T1-12B cells alone or from 4 T1-12B cells co-mixed with ASC-Rs.
Given these findings, the expression of secreted pro-angiogenic growth factors implicated in angiogenesis (vascular endothelial growth factor-A, -C, -D (VEGF-A, VEGF-C, VEGF-D), matrix metalloproteinase-9 (MMP-9), IL-6, placental-like growth factor (PlGF) and platelet-derived growth factor (PDGF-C)) was examined in ASC-Ls (Figure 3G and Additional file 12C). Only the levels of VEGF-D and IL-6 were significantly elevated in ASC-Ls compared to ASC-Ns or ASC-Rs. Consistent with the function of these factors, tumors derived from 4 T1-12B cells co-mixed with ASC-Ls showed an increase in Lyve-1 staining (a marker of lymphatic endothelial cells) compared to tumors derived from 4 T1-12B cells co-mixed with ASC-Rs (Figure 3H). Collectively, these results indicate that ASC-Ls accelerate tumor growth through their increased angiogenesis.
Phenotypic modulation of ASC-Ls is regulated by crabp1
To determine how ASC-Ls are regulated, we conducted analyses of the ranked genes that were differentially expressed between ASC-Ls, ASC-Ns and ASC-Rs. In doing so, we found that crabp1 was one of the most significantly differentially expressed genes in ASC-Ls compared to ASC-Ns or ASC-Rs (Figure 4A). Crabp1 has been well studied for its role as a binding protein for retinoic acid [44, 45], but its role in regulating lipid metabolism or adipocyte differentiation is less well understood [46].
In agreement with the microarray analysis, we confirmed that crabp1 mRNA and protein levels were highly expressed in ASC-Ls (Figure 4B,C). Crabp1 mRNA was overexpressed approximately five-fold in ASC-Ls (L) compared to any other developmental stage (N, P, R: P = 0.003; I: P = 0.04). Likewise, crabp1 protein was most abundantly expressed in ASC-Ls. Crabp1 expression was also examined within mammary tissues from different developmental stages by IF (Figure 4D). Abundant levels of crabp1 protein expression were found localized within adipocytes in mammary glands during lactation (arrows) but not in adipocytes from mammary glands of nulliparous or regressed mice.
To determine whether crabp1 expression is directly regulating the adipogenic phenotype of ASC-Ls, lentiviral-mediated short hairpin-inhibition was used to target crabp1 expression. Crabp1 gene expression was reduced by 86% when compared to cells treated with scrambled shRNAs (P = 0.01; Figure 4E). Crabp1 inhibition led to a dramatic increase in lipid accumulation under differentiation conditions (Figure 4F,G). Furthermore, inhibition of crabp1 expression resulted in a marked decrease in the levels of glycerol release relative to lipoprotein lipase expression from cultures of adipocytes upon differentiation compared to control ASC-Ls (Figure 4H, P = 0.03).
To determine whether phenotypic modulation though crabp1 was responsible for the tumor promoting ability of ASC-Ls, 4 T1-12B tumor cells were co-mixed with ASC-Ls in which crabp1 was inhibited and injected into nulliparous mice. Tumor cells co-mixed with shCrabp1-ASC-Ls grew significantly slower (P = 0.02, Figure 4I) and were significantly smaller (P <0.001, Figure 4J) than tumors derived from 4 T1-12B cells co-mixed with either control ASC-Ls or shScram-ASC-Ls. In addition a three-fold decrease in VEGF-D expression was observed upon inhibition of crabp1 expression when compared to shScram-ASC-Ls cells (P = 0.01, Figure 4K). Endothelial tube formation was also inhibited in endothelial cells treated with media isolated from shCrabp1-ASC-L cells (Figure 4L). Taken together, these findings indicate that the tumor promoting phenotype of ASC-Ls, which can be modulated by crabp1, is important for angiogenesis.
Breast carcinoma-associated fibroblasts phenocopy ASC-Ls
In examining the functional properties of ASC-Ls, GSEA revealed that ASC-Ls were unexpectedly enriched in genes expressed in CAFs (Figure 5A, P = 0.03). Additionally, an enrichment of genes associated with tumorigenesis (P = 0.01) and breast cancer progression (P = 0.04) was also found in ASC-Ls (Figure 5A and Additional files 13, 14 and 15). These findings suggest that CAFs may share similarities with ASC-Ls and likewise be regulated by the same mechanisms that regulate ASC-Ls. Indeed, consistent with their known angiogenic functions [47], CAF-CM significantly stimulated both the proliferation (P < 0.001, Figure 5G) and migration (P = 0.002, Figure 5H) of HUVECs compared to control cells isolated from disease-free reduction mammoplasty tissues.
When examined for features of ASC-Ls including elevated crabp1expression, expression of adipocyte markers, features of lipolysis and a failure to undergo adipogenic differentiation, CAFs exhibited remarkable similarities to ASC-Ls. Compared to the normal adipose stroma of disease-free human breast tissue (Figure 5B left), the adipocytes adjacent to human xenografted tumors (Figure 5B middle) or adjacent to breast carcinomas in patient tumors (Figure 5B right) frequently exhibited features of lipolytic adipocytes including membrane blebbing and decreased adipocyte diameter. Isolated CAFs from breast tumor specimens (n = 6) were examined for the expression of adipocyte markers and lipid accumulation and compared to adipose cells isolated in the identical manner from reduction mammoplasty adipose (RMA, n = 8). Notably, CAFs expressed leptin and FABP4, two markers of mature adipocytes (Figure 5C, D), yet failed to accumulate lipid and undergo adipogenesis compared to RMAs (P = 0.03, Figure 5E).
Immunofluoresence of orthotopic xenograft breast tumors revealed that crabp1 expression was absent in tumor cells (green) but highly expressed within the desmoplastic fibroblastic tumor-associated stroma (red) in all samples tested (Figure 5F, Additional file 16). Crabp1 was also expressed within the tumor-associated stroma of patient-derived primary human breast cancer tissues (Figure 5F, Additional files 16, 17 and 18).
Phenotypic modulation of CAFs is regulated by crabp1
Since these above findings indicate that CAFs exhibit many of the molecular and functional features of ASC-Ls, we wanted to determine whether the expression of crabp1 could also regulate the tumor promoting phenotype of CAFs. To this end, we generated CAF-like (CAF-L) cells as previously described [26], since we were unable to genetically manipulate primary CAFs due to their limited lifespan in culture. Consistent with prior findings [26], CAF-L cells exhibited many of the features of primary CAFs including the ability to accelerate tumor growth (P = 0.04, Figure 6A).
Inhibition of human crabp1 in CAF-L cells using lentiviral-mediated short hairpin led to a reduction of crabp1 gene expression by 96% compared to cells infected with control shRNA (shScr) (Figure 6B). Notably, inhibition of crabp1 in CAF-L cells led to increased intracellular lipid accumulation (P = 0.03, Figure 6C). To determine if phenotypic modulation of fibroblasts was restricted to CAF-L cells, we also inhibited crabp1 expression in NIH3T3 fibroblasts, which express abundant levels of crabp1 (Additional file 19). Similar to the findings in CAF-L cells, knockdown of crabp1 in NIH3T3 cells led to a dramatic induction of adipogenesis and accumulation of lipid under differentiation conditions (P = 0.03, Figure 6D).
To determine whether the phenotypic switch of CAFs affects their ability to promote tumor growth, the Basal B SUM159 breast cancer cell line was co-mixed with either shScr-CAF-L or shCrabp1-CAF-L cells and injected into NOD-SCID mice. Indeed, phenotypic modulation of CAF-L cells abolished their tumor promoting activity (Figure 6E, P = 0.01). This failure to accelerate SUM159 xenografts was attendant with a significant reduction in the density of the tumor-associated vasculature (Figure 6F) and resulted in extensive and significant regions of centralized tumor necrosis, which was essentially absent in tumors from SUM159 cells co-mixed with shScr-CAF-L cells (Figure 6G, P = 0.03). Since tumor necrosis is a well-established feature of limited angiogenesis, these findings, combined with those above, support the notion that crabp1 expression affects the tumor promoting activities of CAFs by modulating, in part, tumor angiogenesis.