Background
Obesity is a growing health pandemic [
1] whose global prevalence has more than doubled in the last quarter-century [
2]. Excess body weight has been implicated in the etiology of a broad spectrum of human cancers [
3,
4], including breast cancer, and recent projections suggest that obesity (defined by the World Health Organization [
5] as a body mass index [BMI] ≥ 30.0 kg/m
2) may exceed tobacco as the largest modifiable risk factor for cancer incidence in the United States [
6]. Because breast cancer is already the most prevalent cancer in the United States and the leading cause of cancer mortality in women worldwide [
7,
8], the impact of increases in the prevalence of obesity on breast cancer incidence and mortality is a major concern.
Of the 8 million female cancer survivors alive today in the United States, nearly half are breast cancer survivors [
9]. Although long-term survival is generally favorable, up to 30% of breast cancer patients will eventually succumb to their disease, with the majority of deaths due to disease recurrence after a variable period of clinical remission following upfront multimodality therapy. Hence, most deaths resulting from breast cancer are due to an inability to effectively prevent tumor recurrence [
10]. Similarly to epidemiological data linking obesity with increased primary breast cancer risk, obesity is also associated with a greater risk of breast cancer recurrence [
11‐
16] (HRs ranging from 1.17 to 1.46 [
13‐
15]), as well as a 30–40% increased risk of breast cancer-related death [
11,
15,
17,
18]. Because the prevalence of obesity among U.S. women is > 40% [
19] and may be even higher among breast cancer survivors [
20], understanding the biological underpinnings of the association between obesity and recurrence risk is a critical unmet need.
The association between obesity and increased risk of breast cancer recurrence and death reflects a variety of biological and nonbiological factors, including delayed detection [
21], more advanced presentation at diagnosis [
11,
14,
22], and the use of suboptimal doses of chemotherapeutic agents relative to body size [
17,
23], as well as increased risks of second primary cancers and non-cancer-related causes of death [
24]. Notwithstanding these factors, the negative impact of obesity on recurrence and long-term survival persists independently of biases related to detection, primary disease burden, and treatment [
14,
18,
25], and it has been observed irrespective of menopausal status [
11,
18,
25‐
27] and across a spectrum of breast cancer subtypes [
17,
18,
25,
28].
Obesity is associated with dysregulation of multiple biologic pathways involving adipokines, insulin and insulin-like growth factor (IGF) signaling, endogenous sex hormone levels, and chronic inflammation [
29]. Despite epidemiologic evidence suggesting a role for each of these pathways in obesity-accelerated recurrence, mechanistic support for the hypothesized causal relationship between obesity and breast cancer has been derived almost exclusively from animal models for primary, rather than recurrent, tumorigenesis. In this regard, increasing evidence suggests that the pathways that drive breast cancer recurrence are likely to be distinct from those associated with primary tumorigenesis [
30‐
34]. Consequently, the mechanisms by which obesity affects breast cancer recurrence may differ from those involved in primary breast cancer development.
Rodent models of diet-induced obesity have been essential for demonstrating a causal relationship between obesity and accelerated primary tumorigenesis [
35‐
48]. However, because the vast majority of breast cancer deaths are due to recurrence, understanding the mechanistic basis for the association of obesity with tumor recurrence is essential. Unfortunately, progress toward this goal has been limited by inherent challenges in human trial design and the absence of an appropriate preclinical animal model.
To address this important gap, we have employed a previously validated transgenic mouse model for breast cancer recurrence to test the causal relationship between obesity and mammary tumor recurrence [
30‐
32,
49].
MMTV-rtTA;TetO-HER2/neu (
MTB/TAN) mice conditionally express the
HER2/neu oncogene and develop invasive mammary adenocarcinomas in a tissue-specific manner in response to chronic induction with doxycycline [
49,
50]. Following oncogene downregulation and pathway inhibition by doxycycline withdrawal, mammary tumors regress to a nonpalpable state in a manner analogous to the treatment of cancers with targeted therapies such as trastuzumab [
51]. However, a small population of residual tumor cells persist following tumor regression and reside in a dormant state [
30‐
32,
52]. Moreover, as occurs in patients with breast cancer, spontaneous local and distant recurrences arise from this reservoir of residual tumor cells following a variable period of latency [
30‐
32,
49,
52,
53].
The clinical relevance of the
MTB/TAN genetically engineered mouse model is supported by several key findings. In particular, functional interrogation of this model has identified several pathways that contribute to tumor recurrence in mice, including NOTCH [
31], SPSB1 [
30], SNAIL [
54], CERK [
52], and PAR-4 [
32], each of which is strongly associated with risk of distant relapse in patients with breast cancer and in the direction predicted by studies in mice, as well as in a manner that is neither specific for local relapse nor restricted to a particular subtype of breast cancer. Furthermore, survival of minimal residual disease (MRD) in the mouse mammary gland following chemotherapy or targeted therapy parallels that of patients who receive neoadjuvant therapy but do not achieve pathological complete response. Indeed, in both mice and humans, survival of local residual tumor cells in the mammary gland following therapy is prognostic for relapse at distant sites [
55,
56]. Also of note, recurrent tumors arising in
MTB/TAN mice often lack human epidermal growth factor receptor 2 (HER2) overexpression, such that recurrence is driven by the activation of alternate growth and survival pathways [
30‐
32,
52‐
54,
57,
58]. This is paralleled by clinical observations that HER2+ primary breast cancers in patients frequently give rise to HER2− residual disease [
59‐
61] and HER2− recurrent tumors [
61]. This strongly suggests that residual tumor cells can survive and recur via HER2-independent pathways in both mice and humans. Finally, residual disease and recurrent tumors in
MTB/TAN mice often exhibit a triple-negative (estrogen receptor [ER]-negative, progesterone receptor [PR]-negative, HER2-negative) phenotype, which in patients is associated with an increased risk of recurrence [
62,
63]. In aggregate, these findings indicate that the biology of the
MTB/TAN model is neither specific for nor restricted to a particular subtype of human breast cancer, and is neither specific for nor restricted to local as opposed to distant sites of recurrence. This, in turn, supports the clinical relevance of this model and suggests that it is informative for the biology of residual tumor cells that survive selection pressures imposed by targeted therapy or the microenvironment at local or distant sites.
In the present study, MTB/TAN mice fed a high-fat diet developed phenotypic and physiologic features characteristic of human obesity, including excess body fat, hyperinsulinemia, impaired glucose tolerance, and dysregulated circulating adipokines. Consistent with this, obese mice experienced an accelerated rate, as well as an increased frequency, of mammary tumor recurrences compared with lean mice, and this was associated with the survival and persistence of an increased number of residual tumor cells in obese mice. These studies provide a preclinical model in which to study the association of obesity with breast cancer recurrence, demonstrate that this association is causal in nature, and suggest that it is associated with obesity-induced alterations in the survival and persistence of residual tumor cells.
Methods
Animals and orthotopic recurrence assays
All mouse experiments were performed in accordance with guidelines of the Institutional Animal Care and Use Committee (protocol number 803351) at the University of Pennsylvania (Philadelphia, PA, USA). All transgenic lines were created and maintained on an inbred
FVB/N background. Bitransgenic
MMTV-rtTA;TetO-HER2/neu (
MTB/TAN) mice were generated by cross-breeding, and tumors were generated by doxycycline induction of
HER2 transgene expression, as described previously [
31,
49]. Orthotopic recurrence assays in monotransgenic
TetO-HER2/neu (
TAN) mice were performed as described in
nu/nu mice and were carried out by injecting 1 × 10
6 MTB/TAN primary tumor cells into the number 4 mammary fat pad of recipient
TAN mice maintained on doxycycline. In contrast to the
MTB/TAN model,
TAN mice do not induce the
HER2 transgene or form tumors when exposed to doxycycline; however, the similar genetic background permits engraftment of doxycycline-sensitive
MTB/TAN tumor cells into the
TAN host.
For both
MTB/TAN and
TAN mice, primary tumors formed in the presence of HER2 signaling, after which oncogene downregulation induced by doxycycline withdrawal was used to simulate targeted therapy. All tumors regressed to a nonpalpable state, and mice were followed twice weekly until clinical recurrence or 250 days in the absence of recurrence [
31,
32]. Study diets included a high-fat diet (HFD; OpenSource D12492 [Research Diets Inc., New Brunswick, NJ, USA]: 60%, 20%, and 20% of calories from fat, carbohydrate, and protein, respectively) or a low-fat diet (LFD; OpenSource D12450B [Research Diets Inc.]: 10%, 70%, and 20% calories from fat, carbohydrate, and protein, respectively). The fat content of the HFD was derived from lard and thus was high in saturated fats. Mice were fed ad libitum.
Biometric parameters
The body weight of each mouse was measured twice weekly throughout the study. Body composition (total, fat and lean mass) was measured using the EchoMRI™ 3-in-1 Body Composition Analyzer (EchoMRI, Houston, TX, USA).
Glucose tolerance testing
Intraperitoneal glucose tolerance tests (IP-GTTs) were performed following complete primary tumor regression (i.e., 4 weeks following doxycycline de-induction) and at the time of clinical recurrence. A glucose bolus (10 μl/g body weight of 20% glucose solution) was administered following a 16-h overnight fast, and blood samples were measured by glucometer from the tail vein at 0, 30, 60, 90, and 120 minutes.
Quantification of serum biomarkers
Serum was collected by cardiac puncture 10 weeks following primary tumor regression, and selected biomarkers were evaluated by enzyme-linked immunosorbent assay (ELISA) according to manufacturers’ protocols. All samples were run in duplicate. The adipokine multiplex panel (insulin, leptin, resistin, tissue plasminogen activator 1) was assessed using the MADKMAG-71K assay kit (EMD Millipore, Billerica, MA, USA). Adiponectin levels were assessed using the EZMADP-60K assay kit (EMD Millipore). IGF-1 and insulin-like growth factor-binding proteins 1–7 (IGFBP1–7) were assessed using the 22-IG1MS-E01 assay kit (ALPCO, Salem, NH, USA) and the MIGFBPMAG-43K multiplex kit (EMD Millipore), respectively. Estradiol, testosterone, and sex hormone-binding globulin were measured using the ES180S-100 (Calbiotech, El Cajon, CA, USA), 55-TESMS-E01 (ALPCO), and CSB-E08233m (Biotrend, Destin, FL, USA) assays, respectively. Hepatocyte growth factor (HGF) was measured using the MHG00 assay kit (R&D Systems, Minneapolis, MN, USA). The inflammatory markers C-reactive protein (CRP), monocyte chemoattractant protein (MCP)-1, and corticosterone were assessed using the RH971CRP01MR (BioVendor, Brno, Czech Republic), MJE00 (R&D Systems), and 55-CORMS-E01 (ALPCO) assay kits, respectively. Multiplex ELISA plates were read on a Luminex 100 plate reader (Luminex Corp., Austin, TX, USA) and analyzed using Exponent software (Texture Technologies, Hamilton, MA, USA). Single ELISAs were read on the Molecular Devices M2 plate reader and analyzed using SoftMax Pro software (Molecular Devices, Sunnyvale, CA, USA).
Because of the low concentration of tumor necrosis factor (TNF)-α and interleukin (IL-6) in serum, these two cytokines were measured using single-molecule arrays (Simoa), an ultrasensitive ELISA technique (Quanterix, Lexington, MA, USA) [
64]. Serum was collected from mice with MRD at 4 weeks after de-induction, data were acquired using Simoa HD-1 Analyzer software version 1.5 (Quanterix) and analyzed using Excel (Microsoft, Redmond, WA, USA) and Prism 7 (GraphPad Software, La Jolla, CA, USA) software.
RNA isolation, qRT-PCR, and RNA sequencing
RNA was isolated from tissues using TRIzol reagent (Life Technologies/Thermo Fisher Scientific, Carlsbad, CA, USA) followed by the RNeasy Mini Kit (Qiagen, Valencia, CA, USA). Reverse transcription was performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems/Thermo Fisher Scientific, Foster City, CA, USA) according to the manufacturer’s instructions. qPCR was performed using the ViiA 7 Real-Time PCR System (Life Technologies) using 6-carboxyfluorescein (FAM)-labeled TaqMan probes for ErbB2 (Rn00566561_m1) and Tbp (Mm00446973_m1) (Applied Biosystems). Expression levels were normalized to Tbp.
For RNA sequencing (RNA-Seq), samples were processed using the High-Throughput Sequencing Core at the Children’s Hospital of Philadelphia in collaboration with Beijing Genomics Institute. RNA-Seq was performed on the HiSeq 4000 platform (Illumina, San Diego, CA, USA) with 30 million 100-bp paired-end reads per sample. The quality of raw data was evaluated by FastQC [
65] and reads from samples passing quality control parameters were aligned to the mm10 mouse reference genome using the STAR aligner [
66]. Aligned reads were counted at the gene level using featureCounts [
67]. Normalization of read counts and differential expression analysis between sample groups were performed using DESeq2 [
68].
A mouse obesity signature was generated by comparing RNA-Seq data from parametrial fat tissues of HFD-Obese and LFD mice (n = 6/arm). The signature was defined as the 679 genes differentially expressed between the two arms using DESeq2 at a fold change cutoff > 1.5 and a false discovery rate cutoff < 0.1. Signature scores were defined as weighted mean expression of the 679 signature genes in a BMI-stratified human dataset (Gene Expression Omnibus [GEO] accession number GSE27949), where the weights were 1 for genes higher in expression in HFD-Obese mice and − 1 for genes higher in expression in LFD mice. Expression levels in GEO accession number GSE27949 were first log2-transformed and then standardized by gene to a mean of 0 and an SD of 1 before signature score calculation. Mapping between mouse and human genes was done using HomoloGene build 68.
Droplet digital PCR
DNA was extracted from the fourth mammary gland of killed TAN mice following primary tumor regression. Total DNA was purified according to the manufacturer’s instructions (DNeasy Blood & Tissue Kit, catalogue number 69506; Qiagen). Briefly, tissue specimens were homogenized in 180 μl of Buffer ATL (Qiagen) and 20 μl of proteinase K and incubated at 56 °C until completely lysed. Subsequently, 200 μl of Buffer AL and 200 μl of ethanol (100%) were added to the tissue lysate, and the mixture was centrifuged through DNeasy Mini spin columns (6000 × g for 1 minute). The flow-through was discarded, and centrifugation was repeated twice using 500 μl of Buffer AW1 (Qiagen) followed by 500 μl of Buffer AW2 (Qiagen). The DNeasy Mini spin column was moved to a clean 2-ml microcentrifuge tube, and 100 μl of Buffer AE (Qiagen) was added directly onto the DNeasy membrane and centrifuged (6000 × g for 1 minute) for elution.
Droplet digital PCR (ddPCR) was performed using the RainDrop Plus™ System (RainDance Technologies, Billerica, MA, USA). The reaction was carried out in a final volume of 30 μl, which included 15 μl of TaqMan™ Genotyping Master Mix (catalogue number 4371355; Applied Biosystems), 1.5 μl of rtTA primers/probe mix 20× (TaqMan probe labeled with FAM; Applied Biosystems), 1.2 μl of droplet stabilizer 20× (catalogue number 30-07026; RainDance Technologies), and 10 μl of sample nucleic acid solution (sample prediluted to 100 ng/μl) and H2O to reach the final volume. Reaction mixtures were placed into the sample wells of the RainDrop Source chip (RainDance Technologies), and droplets were formed in the RainDrop Source droplet generator (RainDance Technologies). After processing, the droplets were transferred to a thermal cycler for PCR amplification with the following thermal profile: hold at 95 °C for 10 minutes, 45 cycles of 95 °C for 15 seconds and 60 °C for 1 minute (ramp 0.5 °C/second), 1 cycle at 98 °C for 10 minutes, and hold indefinitely at 12 °C. After amplification, the Sense chip was loaded on the RainDrop Sense droplet reader (RainDance Technologies) to read droplets in all sample wells. RDA II software (RainDance Technologies) was used to provide absolute quantification of target DNA.
Statistical analyses
Unpaired Student’s t tests were used to analyze normally distributed data. Mann-Whitney U tests were used when data were not normally distributed. Two-way analysis of variance was used to compare paired data. Log-rank tests were used to analyze survival curves. P values < 0.05 were considered statistically significant. HRs with 95% CIs were calculated for all survival curves.
Discussion
Developing more effective strategies to improve long-term survival for patients with breast cancer would be enabled by a deeper understanding of the biologic pathways governing tumor dormancy and cancer recurrence. In this regard, mechanistic studies aimed at gaining an understanding of the biological basis for the positive association between obesity and recurrence risk have been hampered by a lack of animal models. In this paper, we describe a mouse model for studying the impact of diet-induced obesity on breast cancer recurrence that recapitulates key features of the human disease. First, mice fed a HFD exhibit physiological changes analogous to those observed in humans, including increased body weight and fat mass, hyperinsulinemia, fasting hyperglycemia and impaired glucose tolerance consistent with insulin resistance, dysregulated circulating adipokine levels, and characteristic gene expression changes in adipose tissue. Second, paralleling epidemiological observations in breast cancer survivors, obese mice exhibit an increased incidence and an accelerated rate of mammary tumor recurrence. To our knowledge, this is the first experimental model for studying the impact of obesity on tumor recurrence for any cancer type. Third, our finding that diet-induced obesity accelerates breast cancer recurrence in both intact and orthotopic mouse models suggests that at least part of the effects of obesity on recurrence risk are non-cancer cell autonomous and are thus likely due to obesity-induced changes in the tumor cell microenvironment. Last, our observation that obesity results in an increased number of residual tumor cells that survive and persist following tumor regression suggests a novel hypothesis to explain the biological mechanisms by which obesity might result in accelerated tumor recurrence.
Whether the epidemiological association between obesity and decreased RFS observed in humans reflects a causal relationship, or is instead a consequence of the many nonbiological factors that can affect diagnosis and treatment in obese patients, has remained open to debate. Our observation that obese mice fed a HFD exhibit an increased rate and incidence of spontaneous tumor recurrence compared with LFD or HFD-Lean mice provides direct evidence that the relationship between obesity and risk of tumor recurrence is causal in nature. Further, our observations that obese mice experience accelerated breast cancer recurrence compared with genetically identical nonobese mice exposed to the same HFD (i.e., HFD-Lean), and that LFD and HFD-Lean mice exhibit similar rates of recurrence, strongly suggest that the increased rates and incidence of recurrence observed in HFD-Obese mice are due to obesity per se rather than to the HFD to which both HFD-Obese and HFD-Lean mice were exposed. Accordingly, this model should permit the identification of mechanisms and biomarkers for the impact of obesity on recurrence risk that are separable from those of dietary fat.
The presence of residual disease in the breast or local lymph nodes in patients with breast cancer following completion of neoadjuvant chemotherapy is highly predictive of disease relapse [
55,
56]. In this regard, a striking finding of our experiments is the observation that obese mice harbor an increased number of residual tumor cells following tumor regression, compared with HFD-Lean or LFD controls, and that these increases parallel the changes in recurrence risk observed in HFD-Obese, HFD-Lean, and LFD cohorts. These data suggest that obesity may alter the dynamics of residual tumor cells by increasing the number of cells that survive and persist following treatment of the primary tumor (i.e., HER2 inhibition), such that the pool of residual tumor cells “at risk” for reentry into the cell cycle is increased and, by extension, the risk of recurrence is increased. Consistent with this hypothesis, we have previously demonstrated that genetic or pharmacological interventions that reduce the number of residual tumor cells that persist following tumor regression result in a reduced rate and incidence of tumor recurrence [
30‐
32]. These findings provide a potential biological mechanism by which obesity may increase recurrence risk.
Ostensibly, obesity could impact recurrence risk by increasing the number of residual tumor cells that survive therapy, by enriching residual tumor cell populations for phenotypically aggressive cells that are more likely to recur, or both. Although it remains to be determined whether the increase in residual tumor cell abundance that we observed fully accounts for the obesity-induced increase in recurrence risk, or whether this increased pool of residual tumor cells is also enriched for phenotypically aggressive cells, the model we describe should enable these possibilities to be distinguished.
Studies in genetically obese and diet-induced obese rodent breast cancer models have demonstrated that obesity leads to accelerated primary tumorigenesis in models of ER+ and triple-negative (ER−PR−HER2−) breast cancer [
36‐
38,
41,
42,
45,
47]. In contrast, HER2-overexpressing tumor models have been used only infrequently and, in agreement with human epidemiological observations, have not demonstrated an increased risk of primary tumors in the setting of obesity [
35]. To our knowledge, however, rodent studies have not previously been performed to determine the impact of obesity on breast cancer recurrence.
Although obesity-induced alterations in several molecular pathways and circulating biomarkers have been hypothesized to underlie the epidemiological association between obesity and breast cancer recurrence risk, none have been proven at a functional level. In this regard, the serum biomarker findings we report provide several avenues for further experimentation. For example, obesity-induced hyperleptinemia and increases in insulin/IGF-1 levels, as observed in the present study and in other models, can lead to activation of phosphoinositide 3-kinase and mammalian target of rapamycin (mTOR) signaling and promote primary tumor growth [
92]. Conversely, pharmacological blockade of the mTOR pathway has been reported to reverse these obesity-induced effects in a model of Wnt1-induced primary mammary tumorigenesis [
37]. By extension, these observations suggest a potential link between obesity-induced increases in mTOR pathway activation and recurrence risk that could provide a rationale for the therapeutic use of mTOR inhibitors to prevent recurrence.
In contrast to hyperleptinemia and increased insulin/IGF-1 signaling, circulating levels of estrogen and testosterone in this mouse model did not differ significantly between obese and lean mice. Increased levels of endogenous estrogens due to aromatization of androgens in adipose tissue in obese postmenopausal women have previously been hypothesized to promote breast cancer recurrence [
93,
94]. If true, obesity-induced effects on breast cancer recurrence would be predicted to be limited to ER+ breast cancer. In contrast, however, one of the largest meta-analyses of 43 primarily observational cohort studies revealed poorer breast cancer-specific survival in obese women even after adjusting for ER, PR, or HER2 overexpression [
18]. In addition, it is worth noting that although the association of obesity with an increased risk of ER+ breast cancer could reflect dysregulation of the endogenous ER pathway, it could instead represent a manifestation of luminal breast cancers, which in humans are typically ER+. Insofar as the
MTB/TAN mouse model is a model for luminal breast cancer, it is interesting to consider the possibility that the mechanisms by which obesity impacts risk of recurrence reflect the impact of obesity on the biology of luminal breast cancers.
With respect to HER2-overexpressing tumors, researchers in a phase III clinical trial (N9831) that incorporated trastuzumab adjuvant therapy for HER2+ breast cancer observed poorer breast cancer-specific survival in the setting of obesity [
20]. Similar findings were observed for both ER+HER2+ tumors in a meta-analysis of eight neoadjuvant chemotherapy trials [
95] and ER−HER2+ tumors in a pre-trastuzumab era study [
96]. In aggregate, these findings indicate that obesity is associated with an increased risk of recurrence for HER2+ breast cancers in patients, a conclusion that is consistent with our observations in mice and that further supports the utility of this model. Additional studies are required to evaluate the applicability of these findings for specific subtypes of human breast cancer, including those that are ER+, HER2−, or triple-negative.
As noted above, rodent models to date have been used to explore the effects of obesity on breast cancer exclusively with respect to primary tumorigenesis. Accordingly, the mouse model we describe provides several advantages over previously used models. Beyond its focus on spontaneous recurrent mammary tumorigenesis, which accounts for the lion’s share of breast cancer mortality in patients, benefits of the current model include the use of immunocompetent mice, use of diet-induced rather than genetic forms of obesity, and inclusion of diet responders (i.e., HFD-Obese) and nonresponders (i.e., HFD-Lean), thereby permitting separation of effects of obesity from those of a HFD per se. Together, these characteristics provide a new tool for exploring those aspects of breast cancer progression that are responsible for the vast majority of deaths resulting from this disease.