Obesity promotes the expansion of metastasis-initiating cells in breast cancer

C57BL/6 J, FVB/N, MMTV-PyMT (FVB/N) [16], and B6(Cg)-Rag2^tm1.1Cgn/J (Rag2−/−) [17] mice were housed in ventilated cages in the mouse husbandry of the University of Fribourg. For tumor cell grafting, cells were trypsinized, resuspended in complete medium and centrifuged at 1300 rpm. They were washed twice in PBS, counted and resuspended in 1:3 Matrigel:PBS for injection into the 4^th mammary fat pad. To mimic postmenopausal estrogen decrease, 5–7-week-old female mice were ovariectomized and 2 weeks later they were supplied with either a high-fat diet (HFD) or a normal (regular) diet (RD (60% and 10% fat content, respectively). Mice were treated with clodronate liposomes as previously descrived [18]. All the experiments were performed by trained researchers holding the necessary accreditations and in accordance to the Swiss Animal Welfare Regulations and approved by the Cantonal Veterinary Service of the Canton Fribourg (2015_07_FR).

The following antibodies and reagents were used: TER119, CD3 (17A2), CD4 (GK1.5), CD8a (53–6.7), CD19 (6D5), CD31 (MEC13.3), CD45 (30-F11), Ly6C (HK1.4), Ly6G (RB6-8C5), CD11b (M1/70) (Biolegend), CD31, PCNA (Santa Cruz Technologies), Cytokeratin 14 (Covance), CD11b, CD31, Ki67 (Abcam), α-SMA, β-Tubulin, β-Actin (Sigma), Vimentin (Lifespan Biosciences), N-Cadherin, E-cadherin, p21, p53 (Cell Signaling), hypoxia inducible factor 1 alpha (HIF1α) (Novus Biologicals) and PIMO (Hypoxiprobes).

EO771 [19] and Py230 [20] cell lines were obtained from the American Type Culture Collection (ATCC) and grown as recommended. Mouse tumor tissue was dissociated using a mixture of Liberase TH (Roche) and DNAse at 37 °C for 45 min. Cells were filtered, washed twice in 2 mM EDTA in PBS and twice in PBS and then seeded for culture.

For FACS analysis, tumor cells derived from tumor grafts (Py230 and EO771) or primary MMTV-PyMT tumors were obtained by disaggregating the tumors with Liberase TH (Roche) and DNAse at 37 °C for 45 min with agitation. Cells were then washed, filtered, stained with the appropriate antibodies for 30 min at 4 °C; 4',6-diamidino-2-phenylindole (DAPI) was used to stain and discard dead cells. Fluorescence was analyzed using a MACSQuant (Miltenyi) analyzer. FACS data were processed and analyzed using FlowJo.

Immunostaining was performed on 4-μm-thick paraffin sections. Antigen retrieval was induced by heating the samples to 95 °C for 30 min in citrate buffer, pH 6.0. After blocking, we incubated the sections with the indicated antibodies overnight at 4 °C, and then used the secondary fluorescently labeled antibodies Alexa Fluor 488, 567 and 647 (Molecular Probes, Invitrogen) or HRP-conjugated secondary antibodies (Dako). Fluorescent images were taken with a TCS-SP5 confocal microscope (Leica). Light images were taken with a widefield microscope (Leica).

Protein was extracted with complete radioimmunoprecipitation assay (RIPA) buffer, separated by electrophoresis, transferred to polyvinylidene fluoride (PVDF) membranes, blocked with 5% BSA and incubated overnight with primary antibodies. Immunoreactive bands were visualized using HRP-conjugated secondary antibodies (Cell Signaling).

RNA was prepared using the mini RNeasy kit (Qiagen). Complementary DNAs (cDNAs) were generated using oligo-T priming and the M-MLV transcriptase (H-) point mutant (Promega) and quantitative PCR (qPCR) was performed in a StepOnePlus thermocycler (Applied Biosystems) using the SYBR green PCR Master Mix (Kapa). A list of the primers used is shown in Additional file 1: TableS1.

Data were analyzed using GraphPad Prism 6. Means were compared using the unpaired Student t test. Samples were analyzed using Mann-Whitney’s non-parametric test if the data were not normally distributed (with normality assessed using the D’Agostino-Pearson omnibus normality test). When comparing more than two variables, we performed one-way or two-way analysis of variance (ANOVA). To isolate differences between groups in ANOVA, we performed Fisher’s least significant difference (LSD) test. We tested correlation using Pearson’s correlation coefficient or Spearman’s nonparametric correlation analysis depending on the data distribution. The p values are indicated for each experiment. Error bars in the figures indicate standard deviation unless stated otherwise in the figure legends. Significant differences between experimental groups are indicated with asterisks as follows: *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001.

In order to recapitulate postmenopausal obesity and assess how it affects breast cancer progression, we first generated an experimental model following the strategy schematically depicted in Additional file 2: Figure S1A. Ovariectomizing C57BL/6 J mice and feeding them with a high-fat diet (HFD, 60% fat content) significantly increased weight gain compared to non-ovariectomized HFD-fed mice and ovariectomized or non-ovariectomized mice fed with a regular diet (RD) (Additional file 2: Figure S1B). Between 20 and 25 weeks of age, the difference between the mean of the final weight in both groups was 39.7% (Additional file 2: Figure S1C). In addition, obese mice developed the common systemic conditions frequently observed in the HFD mouse model, such as hyperinsulinemia (data not shown) [21]. Obesity is mainly associated with estrogen receptor alpha-positive (ERα⁺) breast tumors [22]. To mimic human disease, we next performed syngeneic transplants in the mammary fat pad of C57BL/6 J mice with two different murine breast cancer cell lines that are hormone-sensitive in vivo, EO771 and Py230 [23, 24], and studied primary tumor growth and progression. As shown in Fig. 1a and b, E0771 and Py230 tumors in the HFD group grew significantly bigger. As with humans, in rodents the susceptibility to gain weight in response to obesogenic diets differs substantially between individuals [25, 26]. This variability is reflected within our experimental groups, since neither the RD nor the HFD body weights follow a normal distribution but are negatively and positively skewed, respectively (p < 0.0068; n = 29 and p < 0.007; n = 35, Additional file 2: Figure S1D and E). Nevertheless, our analyses revealed that body weight moderately correlated with tumor mass (Fig. 1c), which is again in agreement with observations in humans [27]. Interestingly, metastasis was also significantly increased in obese mice (Fig. 1d, e), even when there was no significant correlation between the size of the tumor and the number of metastatic foci in our control groups (r = 0.29, p = 0,22). Comparing same-sized tumors rendered similar results (Additional file 2: Figure S1F and G). To understand whether this increase in metastasis was due to tumor or host derived factors, we injected Celltracker-labeled Py230 cells into the tail vein of lean and obese mice and studied lung colonization using FACS after 2 h, as the time point for initial tumor cell trapping/seeding, and after 48 h, when most cells have extravasated. Our results show that there are no major differences in initial seeding and extravasation in lean compared to obese mice (Fig. 1f). Furthermore, we did not observe significant differences in the number of metastatic colonies formed upon tail vein injection, though there was a slight, non-significant trend toward more metastasis formation in obese mice (Fig. 1g, h). Taken together, these results demonstrated that obesity in ovariectomized mice promotes the formation of larger tumors and increased lung metastasis formation in the two models tested.

Recent clinical data suggest that a diet rich in unsaturated fatty acids correlates with breast cancer risk independently of body mass index (BMI) [28], particularly in postmenopausal women [29]. However, it is still not clear whether the diet itself contributes to a poor prognosis in those patients with breast cancer or whether obesity is required. We then aimed at assessing whether the effects observed on tumor growth and metastasis in our model were due to obesity or to the diet. It is well-known that alternatively activated macrophages (M2) protect against obesity and insulin resistance [30]. We therefore reasoned that using M1/Th1 and M2/Th2-biased mouse strains [31] would allow us to discriminate the relevance of the diet versus obesity in our setting. Hence, we switched to the FVB/N mouse strain, an archetypical M2/Th2-biased mouse strain, in which we could use the PyMT tumor model for consistency purposes. We ovariectomized female mice, fed them with either a RD or HFD and performed syngeneic transplants with MMTV-PyMT tumor-derived cells.

Our results show that FVB/N mice did not gain weight after 12 weeks on the HFD regimen (Fig. 2a). Contrary to Py230-injected C57BL/6 mice, in which RD and HFD tumor growth rates diverge very early (Fig. 2b), we found that FVB/N mice, tumors did not differ in growth kinetics between the RD and HFD groups (Fig. 2c). Recruitment and activation of resting macrophages into proinflammatory ones in the adipose tissue requires previous infiltration by CD8⁺ effector T cells [8]. Therefore, we argued that the absence of lymphocytes in an M1/Th1 biased strain should be sufficient to prevent obesity and rescue the obesity-mediated effects on tumor growth depicted in Fig. 1. Indeed, our results demonstrate that C57BL/6 J Rag2^−/− mice, which lack T and B cells but not macrophages, did not become obese after 12 weeks of HFD (Fig. 2d). Consistent with the lack of T cells, overall tumor growth was faster in C57BL/6 Rag2^−/− mice than in FVB/N mice. However, Py230 tumors did not progress faster in C57BL/6 Rag2^−/− mice fed with HFD compared to RD controls (Fig. 2e). Moreover, in contrast to HFD-fed wild-type C57BL/6 J controls, the peritumoral adipose tissue of HFD-fed FVB/N mice had fewer crown-like structures - histologic arrangements composed of macrophages and dead or dying adipocytes that define white adipose tissue inflammation (Fig. 2f) [32]. Likewise, the peritumoral adipose tissue of obese wild-type C57BL/6 J mice had higher expression of monocyte chemoattractants such as Ccl2 (Fig. 2g), which is in agreement with human data [5]. Overall, these results indicate that in our experimental model, obesity promotes primary tumor growth and metastasis formation, while HFD in the absence of obesity is not sufficient to do this.

We then aimed at investigating possible reasons for faster primary tumor progression in obese mice. Not surprisingly, we observed an increase in the fraction of proliferating cancer cells in early-stage tumors in obese mice (Fig. 3a). To study the association between obesity and faster tumor progression, we next analyzed tumor angiogenesis. A number of reports show that in obesity, angiogenesis cannot cope with adipose tissue growth [33‐37]. We hypothesized that this might be mirrored in tumors, since the mammary gland is mainly composed of adipose tissue and tumors are surrounded by and in close contact with adipose tissue. In agreement with this, we found fewer vessels and lower fractions of CD31⁺ cells in tumors in obese mice (Fig. 3b, c and Additional file 2: Figure S2A). To understand the impact of decreased angiogenesis on oxygen levels in tumors in HFD-fed mice, we injected mice with pimonidazole and found higher hypoxic regions in tumors in obese mice (Fig. 3d). Furthermore, hypoxia in tumors from obese mice led to the accumulation of HIF1α (Additional file 3: Figure S2B), which consequently activated the transcription of specific hypoxia-target genes (Fig. 3e). Interestingly, HIF1α is known to be highly activated in triple-negative breast cancer (TNBC) [38, 39], a subset of aggressive breast cancers most of which are high-grade and present a high risk of metastasis and recurrence [40]. Indeed, histological analyses revealed that in obese mice the tumor mass was less differentiated, more often lacking glandular structures and possessing bigger nuclei (Additional file 3: Figure S2C). In addition, Py230 tumors in obese mice showed a consistent reduction in ERα, human epidermal growth factor receptor 2 (HER2), GATA3 and cytokeratin 18 and a gain in vimentin and c-Myc expression (Fig. 3f and Additional file 3: Figure S2D), suggestive of differentiation into more aggressive TNBC tumors. Overall, our results indicate that obesity causes reduced angiogenesis and triggers hypoxia in primary tumors, which promotes tumor progression.

Given the essential involvement of inflammation in obesity [41], we next aimed at understanding how decreased angiogenesis and hypoxia modulate the immune compartment in the primary tumor. FACS analyses revealed that tumors from HFD mice contained 23% less CD11b⁺F4/80⁺ macrophages (Additional file 4: Figure S3A), which are mostly M1 macrophages in the C57BL/6 model (Additional file 4: Figure S3B). In contrast, the population of CD11b⁺F4/80⁻ cells showed a 31% increase in tumors from HFD mice. This population consists of CD11b⁺Ly6C^medGr1⁺ neutrophils and CD11b⁺Ly6C^high monocytes (Fig. 4a). We confirmed these results by performing western blot analyses and found increased CD11b protein in tumor tissue lysates from HFD mice, compared to tumors from RD mice (Additional file 4: Figure S3C). Of note, this increase was not seen in tumors grown in C57BL/6 Rag2−/− or FVB/N mice fed with HFD diet (Additional file 4: Figure S3D and E), which underscores again the immunological differences between these strains. We then reasoned that if fast-growing tumors in HFD-fed mice contain fewer M1 macrophages and more tumor associated neutrophils (TANs) compared to tumors growing in RD-fed mice, macrophage may be protective against tumor growth. To test this hypothesis, we treated mice with clodronate liposomes to deplete macrophages. Indeed, clodronate liposome treatment boosted primary tumor growth in HFD-fed mice (Fig. 4b) and did not reduce metastasis (Additional file 4: Figure S3F). These results suggest that in our model, macrophages do not contribute to promote tumor progression and metastatic spread, regardless of their essential involvement in obesity.

In order to evaluate the importance of the effects of the microenvironment on the tumor cells, we performed cross-transplantation of tumor cells from HFD-fed and RD-fed into RD-fed and HFD-fed mice, respectively. Interestingly, we observed that tumor cells derived from obese mice grew faster in lean recipient mice compared to cells derived from lean mice (Fig. 4c). As expected, grafting into obese mice further boosted growth of both transplanted cell populations. These results uncoupled immediate microenvironmental effects from tumor cell effects and indicate that the obese tumor microenvironment exerts contextual and sustained effects on tumor cells.

Neutrophils are known to migrate to ischemic tissues and to contribute to the epithelial-to-mesenchymal transition (EMT) [42]. EMT is a process involved in invasion and metastasis and produces cancer stem cells (CSC) [43], a subpopulation of cells that we and others have previously shown to lead metastatic colonization [15]. Indeed, tumors from HFD mice consistently lost E-cadherin and had an increase in N-cadherin and vimentin, three hallmarks of EMT (Fig. 4d and Additional file 5: Figure S4A). This effect was not observed in FVB/N tumors (Additional file 5: Figure S4B). In agreement with TANs being associated with EMT, we identified strong correlation between the expression of CD11b and N-cadherin and anti-correlation with E-cadherin in primary tumors (Additional file 5: Figure S4D).

All these assays were performed with equal-sized tumors to avoid potential confounding effects due to more rapid tumor growth in HFD-fed mice (Additional file 5: Figure S4C). EMT is a distinctive feature of claudin-low tumors, a particular subset of TNBC that is enriched in CSC-related genes [44]. Since aggressive subtypes of breast cancer such as TNBC and basal-like tumors are associated with mutations in p53 [23, 45‐47], we next stained for p53, a surrogate marker for its mutational status. Our results indicate that tumors in obese mice have a higher number of p53-positive cells (Additional file 5: Figure S4E). As a consequence, they also show significantly lower levels of p21(WAF1/CIP1), an important target of p53 responsible for cell cycle arrest (Additional file 5: Figure S4A).

Claudin-low tumors are also characterized by a loss of cell-cell junction proteins. Therefore, we next performed qPCR analyses in tumors from RD and HFD mice using a number of genes from the cell-cell junction organization gene set M820 from the MSigDB database [48], as previously described [23]. The results confirmed that the obese microenvironment triggers a process that leads to the rapid expansion of claudin-low tumors (Additional file 5: Figure S4F).

Finally, to test whether the effects of obesity on the primary tumor are essential for the late steps of cancer metastasis, we digested tumors from RD and HFD mice and injected 5 × 10⁵ tumor cells via the tail vein into tumor-free, RD mice. Our results demonstrate that tumor cells derived from obese mice metastasize more to the lungs compared to cells derived from lean mice (Fig. 4e), i.e. tumors from obese mice contain more CSC with metastasis-initiating capacity. Our data provide direct evidence that the primary tumor microenvironment of obese mice generates more tumor cells with lung metastasis-initiating capacity.