Introduction
Breast cancer is not a single disease with variable morphologic features, but rather a group of molecularly distinct neoplastic disorders [
1]. According to the gene expression-based intrinsic classification, breast carcinomas can be categorized into at least five subtypes: luminal A, luminal B, normal breast-like, human epidermal growth factor receptor 2 (HER2) enriched, and a basal-like subtype [
2,
3]. In addition to the distinctly different gene expression patterns, the subgroups also show significantly different clinical outcomes [
3], likely to be caused by alterations in specific cellular pathways. Moreover, tumors that appear to have similar diagnostic features, do not always respond to treatment in the same way. This can, among other factors, be caused by differences in mutational profile, signaling redundancy and the particular tumor microenvironment [
4].
Most of the existing
in vivo preclinical breast cancer models are established from a limited number of cell lines isolated from human tumors grown in cell culture before implantation into immunodeficient animals. These models do not reflect the breast cancer heterogeneity since they usually have a monomorphic, poorly differentiated histology and lack of tissue organization [
5]. A panel of patient-derived xenograft models has been established in which human breast tumor tissue has been engrafted directly into mice [
6‐
8]. Patient-derived xenograft models generally maintain key features of the original tumors, including histologic subtype, degree of differentiation, growth pattern, and gene expression profiles, even after several passages
in vivo [
5‐
10]. Furthermore, the drug response in these models shows a good correlation with the primary patient tumors [
5,
6,
11], and altogether the xenografts are representative model systems for studies of metabolic and genetic patterns in human breast cancer.
Abnormal choline metabolism is a well-known feature of breast cancer. An elevated total choline (tCho) signal can be observed using magnetic resonance spectroscopy (MRS) and is an
in vivo biomarker for malignant disease [
12]. In line with this, a reduction in tCho has been suggested as an
in vivo marker for response to treatment [
13]. High-resolution magic angle spinning (HR-MAS) MRS has proven to be a useful technique for assessment of choline (Cho) metabolism, as it allows detection of individual Cho metabolites in intact tissue specimens. High levels of Cho and phosphocholine (PCho), which are the main contributors to the tCho signal, have been demonstrated in cultured breast cancer cells [
14,
15], while high levels of glycerophosphocholine (GPC) have been detected in human breast cancer biopsies and xenografts [
16‐
18]. Cho metabolism has been shown to be altered following chemotherapy [
19,
20], and several enzymes involved in Cho metabolism have been identified as potential drug targets [
21,
22]. Despite the potential diagnostic value of Cho-containing compounds, the underlying mechanisms causing the alterations in Cho metabolism are not fully understood [
23]. Integration of metabolic abnormalities and altered gene expression profiles provides new insights into the underlying regulatory network. Elucidation of the biochemical mechanisms governing Cho metabolism may be useful in the development of prognostic and predictive tools in breast cancer management.
The purpose of this study was to map the metabolomic and transcriptomic characteristics of 34 patient-derived breast cancer xenografts, with a special focus on Cho metabolism. In order to evaluate the clinical relevance of the xenograft models for metabolism studies, human breast cancer biopsies from the corresponding molecular subtypes were analyzed using identical methods.
Discussion
In this study, metabolic and gene expression profiles of 34 patient-derived breast cancer xenograft models have been characterized and compared with patient breast cancer samples. The majority of the xenograft models was classified as basal-like and had a triple-negative receptor status. The gene expression profiling was consistent with IHC assessments previously reported for these tumors [
7,
24]. The luminal B/ER positive xenograft samples were characterized by a high PCho/GPC ratio. For the basal-like subgroup, a larger variation in the PCho/GPC ratio was found within the xenograft samples. The metabolic profiles of the xenografts corresponded well with the profiles obtained from human breast cancer tissue. The expression of genes associated with Cho metabolism was found to be different in luminal B and basal-like xenograft models, which also were in accordance with findings in the corresponding subgroups of human breast tumor tissue samples.
Eighteen of twenty four triple-negative samples were classified as basal-like cancers. These results corresponded well with findings from other studies, since approximately 90% of triple-negative breast carcinomas are classified as basal-like [
29,
37]. In addition, expression of estrogen receptors is a known feature of luminal A and B subtypes of breast cancer [
38], and all of the ER positive xenografts were found to be luminal A or B. Overall, the association between histopathological characteristics and intrinsic molecular subclassification was in accordance with previously published data [
39]. This confirms that the molecular subclassification of xenografts reflects the typical characteristics seen in human disease despite the presence of mouse stromal cells and thereby potentially different tumor/host interaction than in human tumors.
In concordance with the high PCho/GPC ratio in the luminal-like/ER positive xenografts, significantly higher PCho/GPC levels were found in the ER positive versus ER negative samples from breast cancer patients. These results are in agreement with findings in other studies of human breast cancer tissue and xenografts [
17,
40] but do not correspond to results from
in vitro studies. Studies of a panel of cultured cell lines have suggested that malignancy is associated with high PCho and low GPC levels [
14]. High GPC levels found
in vivo, both in xenograft tissue and clinical samples, suggests that this hypothesis has to be refined. Due to the discrepancy between
in vitro and
in vivo data, it is tempting to speculate that microenvironmental factors may play a role in the
in vivo regulation of Cho metabolism [
41,
42]. In addition, there is a possibility that high GPC concentrations could be linked to differences in driver mutations between ER positive and ER negative tumors. Luminal-like breast cancer is strongly associated with ER expression, and metabolism is likely tightly regulated by ER-mediated mechanisms. In basal-like breast cancer, the impact of ER (and HER2) –mediated signaling plays a smaller role. The activity in other signaling pathways, such as PI3K and MAPK, is, therefore, comparatively more important, resulting in a more heterogeneous metabolic profile.
The results from the gene expression profiles indicated that the genes involved in Cho metabolism were differentially expressed in luminal B compared to basal-like xenograft samples. The luminal B xenograft samples were found to have a higher expression of PLCD4, GDPD3 and GPD1L, while the basal-like samples were characterized with higher expression of PLCG2, PNPLA3 and PLCE1. In addition, the concentrations of Cho, PCho and GPC were correlated with the expression of different genes in different breast cancer subgroups. This suggests that luminal B and basal-like breast cancer may have different mechanisms regulating Cho metabolism. The concordance between the gene expression profiles from the xenograft models and breast cancer tissue samples confirmed the assumption that these xenografts are representative models of human breast cancer.
Although tCho is proposed as an
in vivo biomarker in breast cancer, the regulation of Cho metabolism is not fully understood. The current consensus is that the transport of Cho into cancer cells is increased compared to normal cells.
In vitro, increased expression of various Cho transporter proteins has been demonstrated, and
in vivo PET studies using [
11C]-choline or [
18 F]-fluorocholine have indicated increased uptake of Cho both in preclinical models and clinical studies [
43‐
46]. Furthermore, it has been shown that
CHKA is upregulated in several cancers, and that
CHKA expression correlates with PCho concentration
in vitro [
45,
47]. In our study, the positive correlation between
CHKA expression and PCho concentration in xenograft tumors was confirmed. The regulation of GPC is poorly understood [
48], which is a challenge as this metabolite contributes significantly to the tCho signal measured by
in vivo MRS. Several studies have demonstrated that GPC may be a potential biomarker for response to treatment [
21,
49‐
52], and it is, therefore, necessary to elucidate the mechanisms responsible for regulating GPC concentration
in vivo. We found a positive correlation between
CHKA and GPC concentration, which is not surprising as PCho and GPC concentrations are positively correlated. This suggests that malignant transformation and upregulation of
CHKA leads to a general increase in PtdCho turnover, which is reflected by a high concentration of both precursor (PCho) and degradation products (GPC) of this cell membrane component. A positive correlation between
GDPD5 and PCho concentration was also found which is in accordance with previous studies suggesting that
GDPD5 may be upregulated in ER negative breast cancer [
53]. As GDPD5 has been suggested to catalyze GPC degradation, the positive correlation between
GDPD5 and GPC concentration was not anticipated. These results suggest that
GDPD5 may be a general marker for abnormal Cho metabolism, but not necessarily regulating GPC concentration. This interpretation is further supported by the positive correlation between expression of
CHKA and
GDPD5.
Various phospholipase enzymes are involved in degradation of PtdCho to GPC, PCho, and Cho but the roles of the various isoforms are still not fully elucidated. Several phospholipases are upregulated in cancer compared to normal tissue [
54‐
57], but their impact on the concentration of Cho-containing metabolites is poorly understood [
58]. The results of this study indicate that expression of phospholipases varies significantly between the xenograft models within the different breast cancer subgroups. However, the large number of phospholipase isoforms and their complex biology makes it difficult to interpret the significance of these differences. Numerous studies have demonstrated complex and often reciprocal interactions between oncogenic signaling pathways and enzymes involved in Cho metabolism [
23]. Several enzymes involved in Cho metabolism, including CHK, PLC, PLD, and PLA2, have been shown to be affected by RAS-mediated signaling [
59,
60]. MYC and HIF1 have also been shown to be involved in the regulation of CHKA [
61]. When we observe differences in the Cho metabolic and gene expression profiles between cancer subtypes, it may be caused by specific oncogenic signaling pathways that are more frequently upregulated in some subtypes.
More than 10 years after the first report on molecular fingerprints in breast cancer [
2], there are still active discussions on the optimal strategy for subtyping breast cancer. Several research groups advocate an integrated approach, where data from several –omics platforms are combined for identification of clinically relevant subtypes [
62,
63]. Since MR metabolomics reflect tumor microenvironment to a larger degree than other –omics techniques, it can contribute to improved understanding of the underlying biology in different breast cancer subtypes. Including metabolic profiles in the criteria for novel breast cancer subtypes may, therefore, bring us closer to personalized breast cancer treatment.
In this study, gene expression profiling and metabolomic analysis of 34 patient-derived xenograft models demonstrated significant difference between luminal B and basal-like breast cancer. Similar patterns both in metabolic profiles and expression of Cho genes were found in the xenograft models and human breast cancer with corresponding molecular subtype. This panel of patient-derived xenograft models, therefore, represents a unique and valuable tool for studies of molecular properties associated with sensitivity or resistance to chemotherapy or targeted anticancer drugs. It also allows further studies of the unique biology of the different subtypes of breast cancer, which may be important for future clinical applications based on molecular fingerprints.
Competing interests
The authors declared they have no competing interests.
Authors’ contributions
MTG was involved in the study design, acquired and analyzed the HR-MAS MRS data and drafted the manuscript. NS performed the gene expression analysis, interpreted the microarray data and was involved in drafting of the manuscript. SAM was involved in the study design and in drafting of the manuscript. EAR performed statistical analyses and interpreted the data. EB performed gene expression analysis and interpreted microarray data. AK coordinated the xenograft samples. BS was involved in the HR-MAS MRS protocol and supervised the analyses. TFB was involved in the study design and in statistical analyses. ALBD contributed with expertise in molecular biology techniques. GMM and OE were involved in the coordination of xenografts. TS was involved the study design and in the interpretation of the microarray data. EM was responsible for the establishment of the xenograft models and delivery of samples. ISG designed and coordinated the study. All authors critically revised, read and approved the final manuscript.