Metabolomic studies of breast cancer in murine models: A review

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


Introduction
Breast cancer (BC) is the most frequently diagnosed cancer in women worldwide with an estimated incidence of ca. 24% in 2018 and the highest mortality related to cancer in women (ca. 15%) [1]. BC incidence increases with age and is mostly diagnosed in the 50-60 years-old group, although BC also affects younger age groups, usually in more aggressive forms [2]. In spite of the continuous methodological advances in BC diagnosis and therapy, research remains highly relevant regarding early identification of individuals at higher risk, monitoring of disease progression and early assessment of response to therapy. Cancer arises as a result of genetic and epigenetic alterations that promote changes in gene expression and protein function. These changes ultimately result in aberrant cellular metabolism which, in tandem with other cancer hallmarks, is involved in the carcinogenic process [3]. Deviant metabolism in cancer cells has given rise to important tools that are currently used in diagnostic such as positron emission tomography coupled with computed tomography (PET-CT) or magnetic resonance imaging (MRI) [4,5], both techniques enabling the detection of metabolic and molecular phenomena, which often occur before morphological alterations become detectable.
Metabolomics entails the comprehensive detection of low molecular weight (M w ) molecules (or metabolites) involved in intermediary metabolism and the changes in their levels upon perturbations such as disease, treatment, diet. In particular, disease triggers deviant metabolic behavior and metabolic profile changes detected in tissue, cells, or biofluids hence hold much promise in the search for new disease biomarkers [6,7]. Both untargeted and targeted metabolomic approaches employ high-throughput analytical techniques, namely nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS). NMR spectroscopy is a powerful tool capable of detecting a range of different types of metabolites simultaneously, providing valuable structural information with high reproducibility, albeit its J o u r n a l P r e -p r o o f 9 inter-laboratory comparison of metabolomics results, it is of paramount importance that standard operating procedures (SOPs) should be followed in every step of the workflow shown, particularly regarding sampling and storage protocols, sample handling prior and during analysis, and data mining procedures. Metabolomic studies of BC cell lines cultured in vitro will not be included here and the interested reader is directed to other texts for more detailed information on that subject [18,19]. Furthermore, both untargeted and targeted studies of metabolite sets (or signatures) will be discussed, leaving out reports of biochemical assessment of single compounds in BC-related biological samples.
Although the number of publications on metabolomics of BC mouse models is clearly lower compared to human cohorts (Figure 2), the interest in BC murine models has tended to increase in recent years, with MS metabolomics emerging as the more extensively used technique compared with NMR ( Figure 2a). Studies on murine models have mainly addressed tissue analysis (over 60% of all studies), whereas serum has been the most extensively studied biofluid (Figure 2b). In section 2 of this review, the main types of mouse models used in BC metabolomics research ( Figure   2c) will be briefly presented, with carcinogen-induced and spontaneous models comprising the majority of studies to date (Figure 2d). In subsequent sections, each model type will be discussed regarding the specific metabolomic strategies employed and main findings reported.
In general, subsections will be defined, either describing studies that evaluate tumor metabolic profiling in different contexts; or studies of response to therapeutic agents; or, when applicable, other types of studies. Within each subject, studies addressing local tumor metabolism will be firstly discussed, addressing ex vivo analysis of intact tissue or tissue extracts. This will be followed by presentation of biofluid studies, usually performed for systemic evaluation and non-invasive biomarker search.

Main murine models used in BC research: characteristics and complementarity
Murine models used in BC research may be divided into two main types: immunodeficient (xenotransplant or xenograft) models that entail cells or tissue of human origin transplanted into immunocompromised mice; and immunocompetent models that can be subdivided into spontaneous or carcinogen-induced tumors (Figure 2c). Spontaneous tumors may arise naturally or as a result of specific genetic modifications introduced in genetically engineered models (GEM). Models based on either spontaneous or carcinogeninduced tumors are also referred to as syngeneic (or allografts) meaning that the original tumors are transplanted into mice with the same genetic background.
Xenograft models enable the study of human cancer cells implanted into immunodeficient mice to prevent host immune reaction and consequent rejection.
J o u r n a l P r e -p r o o f BC cell lines previously established in vitro [22]. Consequently, CDX models usually present faster tumor manifestation, compared with PDX models, and the large range of available cell lines that represent different molecular subtypes of BC enables CDX models to be successfully used in different contexts of preclinical BC research [22]. However, translation of CDX models into the clinic is limited, since these models usually originate relatively homogeneous tumors, thus not recapitulating the typical heterogeneity of human tumors.
Furthermore, since the cell lines used have previously adapted to in vitro growth, they may not faithfully mimic human cancer cell biology. PDX models involve transplantation of either human cancer cells or tumor fragments. This is a challenging task that involves obtaining viable cells with tumor-initiating capacity from fresh or cryopreserved human tumors, and their successful engraftment and growth may typically require many months. An advantage of PDX models is that they maintain many of the relevant original characteristics of the primary tumor, such as growth kinetics, histological features, cancer cell heterogeneity, invasiveness, metastatic capacity and response to therapy. Therefore, this type of model is particularly suited for studies in search of biomarkers of drug response/resistance, as well as of anticancer potential of new therapeutic drugs [22]. Although PDXs allow for personalized treatments to be tested, the use of these models is still somewhat limited due to the typical long time required for the tumor to become established in the host [23].
Immunocompetent BC models faithfully reflect tumor-stroma interactions and contemplate immunosurveillance and immunoediting mechanisms that regulate tumor progression. These characteristics make these models particularly useful for immunotherapy studies. A disadvantage of immunocompetent models is that they do not involve human J o u r n a l P r e -p r o o f 13 oncogenes or reduced expression/loss of function of tumor suppressor genes [16,17]. The more conventional GEMs are oncogene-driven transgenic mice, characterized by the same genotype in all cells of the organism, which preferentially reflects familial/hereditary BC. To circumvent this and more closely mimic the heterogeneity of the human disease, GEM models are often generated using tissue-specific promoters, involving the selective activation of oncogenes and deletion of tumor suppressors [22]. This approach allows for oncogenes such as Neu/ErbB2, cyclinD1, Ras, Myc and Want1 to be activated only in the mammary gland tissue [20]. The most frequently used promoters to develop these conditional GEM models comprise the mouse mammary tumor virus-long terminal repeat (MMTV-LTR), the C3 promoter and the whey acidic protein (WAP) promoter, used to drive expression of oncogenes only in mammary epithelial cells. Overall, GEM models may be clinically relevant since they develop tumors spontaneously and localized in the correct site, thus enabling the study of initiation and progression, as well as of therapeutic response. However, GEM models are expensive to setup, originate low numbers of tumors, with long latency, limited heterogeneity and which often do not metastasize [25]. Carcinogen-induced models are developed through the inoculation of the mice with compounds that induce DNA damage, so that the carcinogenic process is initiated by mutations in the DNA. The resulting multiple primary cancers may be used to evaluate molecular mechanisms involved in tumor initiation, promotion and progression [26]. A wide array of chemical carcinogens have been employed for induction of histologically distinct murine mammary tumors, namely 7,12dimethylbenz[a]anthracene (DMBA) [27], N-methyl-N-nitrosourea (MNU) [28], medroxyprogesterone acetate (MPA) [29] and 17-β-estradiol [30]. DMBA is a polycyclic aromatic hydrocarbon with the ability to develop well-differentiated mammary adenocarcinomas in mice and rats, these tumors being hormone responsive and morphologically similar to human breast carcinomas [27]. MNU acts as a methylating agent and is widely used to induce both estrogen-dependent and estrogen-independent carcinomas, in both mice and rats [28]. In BALB/c mice, MPA induces metastatic ductal mammary carcinomas that retain estrogen, progesterone and prolactin receptors and are hormone responsive [29]. Furthermore, estrogens are known to induce mammary tumors in rats [30], J o u r n a l P r e -p r o o f 14 The recognized importance of choline metabolism in BC [32], and the possibility of measuring choline compounds in vivo through low resolution magnetic resonance spectroscopy (MRS), has led to the comparison of in vivo measurements with ex vivo analysis of tumor tissue, aiming to characterize the profile of choline compounds in more detail.
Specifically, in vivo localized 31 P-MRS was used to quantify phosphocholine (PC) and glycerophoshopcholine (GPC) in MCF-7 and MDA-MB-231 CDX models developed in SCID mice [33] and these results were compared with ex vivo high resolution 1 H-and 31 P-NMR analysis of the corresponding tumor extracts ( Table 1). The data indicated that mammary tumors were characterized by increased choline uptake, increased choline kinase activity, and increased phosholipase-mediated turnover of phosphatidylcholine (PtdCho).
These metabolites contributed to the increase in PC observed in vivo. Another study [34] was based on previous observations that the glycerophosphodiester phosphodiesterase domain containing 5 (GDPD5) enzyme seems to be overexpressed in highly malignant estrogen receptor-negative cell lines and that PC/GPC ratio tends to be increased in metastatic cell lines. The study compared profile changes upon GDPD5 silencing in MDA-MB-231 cell line using in vivo 31 P-MRS and ex vivo 31 P-NMR of tumor tissue and cell extracts ( Table 1). The results showed that GDPD5-silencing alters the profile of choline compounds by increasing the levels of GPC and phosphoethanolamine (PE) and reducing PC/GPC ratio, thus approaching a possible non-malignant metabolic phenotype of breast epithelial cells.

Response to therapy
A large number of metabolomic studies in CDX models have addressed the effects of different therapies, mostly by analyzing tumor metabolic response to drugs, either by ex vivo direct tissue analysis (using HRMAS NMR) or by analysis of tissue extracts (using both NMR and MS-based methods). Initial low resolution in vivo 31 P-MRS studies of tamoxifen effects in MCF-7 xenografts revealed increases in the ratio of nucleotide triphosphates to inorganic phosphate, NTP/Pi [35]. Subsequent studies included high resolution ex vivo studies of therapy effects through the direct analysis of mammary tumors by 1 H HRMAS NMR [36,37] ( Table 1). Ex vivo 1 H HRMAS NMR provides remarkably enhanced spectral resolution that enables the identification of a high number of low M w metabolites (giving rise to narrow J o u r n a l P r e -p r o o f samples have higher levels of GPC, PCho and Choline. Adapted with permission from reference [65]. A study of a CDX model of mouse mammary epithelial cell lines transformed with activated Neu/ErbB2 oncogene (NMuMG-NT2197) reported the analysis of tissue extracts by GC-MS metabolomics [38], in tandem with other biochemical measurements. The study addressed the role of the ErbB family receptor tyrosine-kinase inhibitors, combined with guanides, on the metabolism of MNuMG-NT2197 tumors grown in Nude mice. The ErbB1/ErbB2 inhibitor lapatinib and the anti-diabetic drug phenformin were tested alone and in combination and results revealed that, when administrated alone, each drug induced opposite effects on tricarboxylic acid (TCA) cycle intermediates (Table 1). On the other hand, lapatinib/phenformin combination seemed to have a synergetic effect on metabolism and, overall, the results led the authors to propose that cancer cells seem to display significant metabolic plasticity, in particular related to aspartate, asparagine, serine synthesis and glutamine metabolism, which may limit the efficacy of this type of therapy combinations [38].
Regarding biofluid metabolomics, a serum and urine MS-based metabolomics study [39] of MCF-7 CDXs developed in female BALB/c Nude mice showed that volatile oil extracted from the Saussurea lappa Decne plant, and some of its components (Table 1)   J o u r n a l P r e -p r o o f

Other studies
In relation to studies on CDX models other than for tumor characterization or response to therapy, a serum 1 H-NMR metabolomics study aimed to characterize the effects of hypoxia in MDA-MB-231 cells and in the plasma obtained from NIH-III Nude mice injected with the same cells to promote metastatic spread [41]. Although comparison of cell culture and plasma results is not straightforward, the authors noted some similarities e.g. regarding relative amino acid levels (Table 1), thus suggested to potentially represent general markers of hypoxia. In addition, a recent MS-based study, also using MDA-MB-231 CDXs, investigated the effects of the contaminant bisphenol F (BPF, a usual alternative to bisphenol A in many industrial applications) on tissues where BC metastases commonly arise, namely kidney and liver [42].
Results showed that exposure to BPF induced alteration of the levels of glutathione (reduced form, GSH), amino acids and lipid levels, possibly suggesting reprogrammed GSH biosynthesis and glycolysis alterations, as well as reprogramming of lipid metabolism. The authors claimed that metabolomics may aid in the understanding of BFP toxicity mechanisms, here particularly considered concomitantly with the presence of BC metastatic behavior.

Metabolomics of patient derived xenograft models (PDX)
To the best of our knowledge, metabolomics of PDX models has been carried out only through the direct analysis of breast tumors using 1 H HRMAS NMR (Table 2) (and not through biofluid analysis), either for comparison between the metabolic profiles of basal-like and luminal tumors or to investigate tumor response to different therapies.

Characterization of tumor-related metabolic profiles
One of the initial studies in this context aimed to compare the abnormal choline metabolism typical of BC tumors [32], between basal-like and luminal PDXs developed from the MAS98-12 and MAS98.06 primary BCs, respectively [43].  (Table 2).

Response to therapy
The main reports on PDX models in the context of therapy investigation have, to the best of our knowledge, arisen from a single team of researchers (Table 2) seemed to be specific of BC subtype (Table 2).
Another study, investigated the effects of inhibitors of different components of the phosphatidylinositol 3-kinase (PI13K) pathway, which is often found activated in BC cells. J o u r n a l P r e -p r o o f 22 In this context, the effects of two of these inhibitors, MK-2206 and BEZ235, on the metabolome of basal-like tumors, compared to luminal tumors, was investigated by 1 H HRMAS NMR [47]. Both drugs induced significant growth inhibition in basal-like tumors, but not in luminal ones, consistently with a strongly deviant metabolic behavior noted for the former, particularly in terms of lactate, GPC and PCho levels ( Table 2). The authors suggested that this set of compounds may potentially be used as a biomarker for early therapy monitoring. A more recent report [48] dealt with the idea of using inhibitors of the PI3K/AKT/mTOR pathway, which is typically altered in triple negative breast cancer (TNBC). In particular, the metabolic effect of mTOR inhibitor everolimus was assessed in a large cohort of TNBC PDXs (n=103) [48]. Results revealed that, after treatment, changes in specific metabolite levels seemed indicative of reduced glycolytic lactate production and glutaminolysis ( Table 2 and Figure 4), consistently with inhibition of the PI3K/AKT/mTOR pathway. Interestingly, the metabolic metabolic variations observed (specified in Table 2) seemed to be more significant in treatment responders, compared to non-responders, thus opening enticing possibilities for individualized early monitoring of treatment response.   [51].
In addition to the above, most metabolomic studies of spontaneous animal models with naturally occurring tumors have, to our knowledge, addressed biofluids, with two reports on urine and six on serum/plasma, two of the latter dedicated to therapy response (Table 3) to be discussed in the following section. Urine has been studied by NMR before and after development of spontaneous mammary tumors in SHN mice [52] and results unveiled that the  (CNS) burdens. In relation to biomarker specificity, these results seemed to suggest the interesting possibility of urine metabolic profile being used to specifically differentiate brain metastasis, from other types. The sera of mice in similar models (animals injected with 4T1 cells with different metastatic potential) was also investigated through NMR metabolomics to attempt the identification of biomarkers of malignancy associated to spontaneous metastasis [54,55]. The subpopulation described by the more metastatic phenotype (sLe xnegative, compared to sLe xpositive) was characterized by higher levels of lactate and tCho in serum [54]. A second report added that BC with different metastatic behavior could be successfully  Some indication of deviant lipid metabolism during BC metastasizing to the lung was also observed through FTIR analysis of the plasma of BALB/c mice injected with 4T1 cells [57].
Spectral changes became more clearly detected with the aid of PCA and second derivative records of FTIR spectra and showed that, at the early metastatic stages, changes occur in PLs unsaturation levels and secondary protein structure, with a suggestion of enhanced synthesis of sugar-containing biomolecules and amyloid proteins. Subsequent later stages seemed to involve clear changes in most spectral regions, namely affecting content and structure of lipids and carbohydrates, while suggesting a decrease in total protein content (Table 3).

Response to therapy
Regarding studies of spontaneous models with naturally-occurring tumors to investigate response to therapy, a recent NMR study of mammary tumor extracts obtained from female BALB/c mice implanted with the 4T1 cell line [58] was carried out in order to    A second report on serum NMR [60] followed up on the idea that drug encapsulation may induce more efficient drug delivery and described an NMR analysis of serum and multiple organ extracts showing that delivery of combined Dox and paclitaxel (Tax) encapsulated in methoxy poly(ethylene glycol)-poly(lactide-co-glycolide) (mPEG-PLGA-DT) nanoparticles performed significantly more efficiently, compared to administration of drug free forms [60].

NATURALLY-OCCURRING TUMORS Tissues and extracts
Besides characterizing the effect of untreated mammary tumors, compared to healthy controls, in terms of enzymatic expression levels and some metabolites (Table 3) relatable to different organs, the authors noted that mPEG-PLGA-DT treatment stabilizes fumarate levels (suggested to relate to reduced Dox cardiotoxicity) and activated NADPH synthesis through the PPP (possibly to tackle oxidative stress induced by exposure to nanoparticles).
Combination of Dox and Tax in their free forms or mPEG-PLGA-DT encapsulated showed similar tumor inhibition efficacies, however, the encapsulated form performed better in counteracting tumor-induced metabolic alterations. Furthermore, mPEG-PLGA-DT was also related to improved drug efficacy, meliorated pro-inflammatory effect and reduced cardiotoxicity, thus making it an interesting option to conventional Dox/Tax administration.

Genetically engineered models (GEM)
Notably, metabolomics studies using GEM murine models of BC have been carried out in the last four years, to our knowledge, and have only addressed local tumor metabolism (through the study of tissue extracts, with the exception of one record of HRMAS direct tissue characterization), that is, no reports of biofluids metabolomics in this context have been found.

Characterization of tumor-related metabolic profiles
Many biochemical studies have been carried out using transgenic mice in order to target particular aspects of the carcinogenesis or metastatic processes, while encompassing the important role of microenvironment on the spontaneously formed tumors. One report considered MMTV-NEU-NT transgenic mice, known to produce mammary tumors by overexpressing the activated form of the Neu/HER2 oncogene (the mouse homologous to the ERBB2 gene found amplified in many human BC), to study the effect of -hydroxybutyrate MS metabolomics was used to search for oncogene-specific metabolic signatures, compared to healthy control mice. Some metabolite changes seemed to be general to tumors, compared to mammary tissue, (increased levels of some TCA intermediates, lactate, glucose-6phosphate, cholesterol and cytidine, however specific changes were also noted for the cohorts with different initiating oncogenes ( Table 3). The authors singled out the C3-Tag cohort as producing a more robust specific metabolic tumor signature, which the authors state may hold prognostic value, if translated to human cohorts.
Regarding research on BC metastatic behavior, NMR metabolomics of lung tissue extracts was used to gauge lung metastatic progression in BC transgenic MMTV-PyMT mice [51] (Table 3), unveiling that, at earlier stages (micrometastasis), a possible metabolic marker comprised significant increases in lactate, alanine, glutamate and creatine. At later stages of the metastatic process, increased levels of m-inositol and the GPC/PC ratio, together with decreased valine levels, seem to become important. Thus, the authors advanced the possibility of following the extension of lung metastases through measurement of deviant behavior in glycolysis and creatine metabolism, as well as amino acid and PLs metabolism. Furthermore, transgenic mice have been used to investigate the role of sialic acid in the metabolism associated with metastatic BC [63]. Mice implanted with cells lacking a sialic acid-encoding gene, and observed to be less metastatic, were compared with mice with upregulated sialic acid metabolism, using MS-based metabolomics. The metabolic changes observed in tumor extracts for the two main groups (more metastatic and less metastatic) ( Table 3) indicated that the absence of the targeted gene does inhibit sialic acid activation, which seems to result in a lesser metastatic behavior. The tailoring of sialic acid pathways is proposed as a possible way to reduce the extent of metastatic BC.
J o u r n a l P r e -p r o o f polyamines and nucleotides production, occurring in tandem with enhanced levels of TCA and urea cycle intermediates. Notably, lipid metabolism was also observed to change from high triglycerides content in mammary gland to high PL levels in the tumors, which is consistent with the important contribution of surrounding adipose tissue in the mammary gland, compared to tumor samples.

Response to therapy
To our knowledge, the first metabolomics report on GEMs of BC aimed to investigate the potential of NMR metabolomics of intact tissue (using HRMAS NMR) to find biomarkers of resistance to docetaxel treatment [65] (Figure 3 J o u r n a l P r e -p r o o f

Metabolomics of carcinogen-induced models
Most of the metabolomic studies of immunocompetent carcinogen-induced BC models have, to our knowledge, addressed exposure to the environmental pollutants 7,12-

dimethylbenz[a]anthracene (DMBA) and N-methyl-N-nitrosourea (MNU), along with fewer
reports on other carcinogens, namely 3-methylcholantrene and 17β-estradiol (Table 4). In these studies, initial characterization of tumor tissue or extracts has, in recent years, been complemented by plasma/serum metabolomics studies. Also, most studies on these models have addressed the metabolic response to different therapeutics, along with the study of particular diets or simply the search for tumor metabolic biomarkers, compared to controls; in this section, all studies will be discussed for the corresponding common carcinogenic compound.

3-Methylcholantrene-induced
The first studies on carcinogen-induced models were, to our knowledge, carried out on 3-methylcholantrene-induced models, using 31 P-NMR to characterize tumor extracts, in tandem with in vivo 31 P-MRS [67][68][69], during time-course treatment with the alkylating drug CPA [67] and the anti-metabolite 5-flurouracil (5-FU) [68] (Table 4). In the former case, CPA was chosen because the drug is believed to have a similar mechanism of action as radiation and the authors aimed to pinpoint the metabolic changes expected with radiation therapy. In vivo 31 P MRS revealed that the ratio of a composite NMR peak partially arising from PE over PC, represented by (PME´)/PC (where PME´ corresponds to a down-field spectral component arising from phosphomonoesters), was noted to increase with treatment time. Ex vivo 31 P-NMR showed that such an increase was due to PC decrease, in tandem with an increase in an overlapped resonance probably arising from nucleotide monophosphates. CPA treatment also induced increased levels of PDEs, GPC and glycerophosphoethanolamine (GPE), compared to untreated tumors. These changes in PME resonances were expected with basis on previous radiation studies. A similar strategy applied by the same authors established that 5-FU induces an increase in PE, rather than a PC decrease, as measured ex vivo. Increased levels of PDEs, GPC and GPE were also noted as a result of treatment [68]. More recently, the same murine model was used to test different doses of NAD synthesis inhibitor FK866, which is known to initiate apoptosis without damaging DNA, thus, potentially working as an anticancer drug without inducing tumor resistance [69]. In tandem with in vivo NAD decreased levels and increased PME signals measured by 1 H-decoupled 31 P-MRS, 31  Still regarding biofluids metabolomics in the context of DMBA-induced mammary tumors, a plasma metabolomics (MS-based) study [75] investigated the claim that the ubiquitous amino acid derivative taurine may have anti-cancer activity in BC. Results showed that taurine administration significantly reduced mammary tumor rate and metabolomic results unveiled a set of 23 circulatory metabolites with changed levels (Table 4), which suggested that anti-cancer taurine activity may induce deviant behaviors in energy, amino acid and nucleic acid metabolisms.

N-Methyl-N-nitrosourea (MNU)-induced
In the context of the role of nutrition on cancer development, the effect of different types of edible beans on mammary tissue and BC tumor lipid metabolism was investigated in a MNU-induced murine model [76] using MS-based metabolomics of tissue extracts (Table   4). Both mammary and tumor tissues were found to vary in the metabolic profile of  (Table 4). In addition, the authors proposed a relationship between increased 5-methylcytidine levels and decreased N-acetylcitrulline (urea cycle intermediate) levels with the presence of mammary cancers [77]. NMR metabolomics of rat sera considered the specific roles of cyclooxygenase-2 (COX-2) and 5-lipoxygenase (5-LOX) inhibitors (zaltoprofen and zileuton, respectively), either in monotherapy or in combination (Table 4). Inhibitor combination was found to have beneficial effects in cell proliferation (decreased), histopathological architecture, apoptosis and oxidative stress control, as well as restoring the levels of several metabolites to levels typical of controls [78].

17β-Estradiol-induced
To our knowledge, only one report of metabolomics has recently addressed estrogen (17β-estradiol) -induced mammary tumors [30]. This study addressed serum PL profiles of tumor-bearing rats, compared to healthy controls, having mainly found that higher levels of lysophosphatidylcholines (LPCs) as well as PtdCho seemed to characterize tumors. The authors emphasized the importance of PLs as biomarkers in BC tumors and advanced LPCs as being particularly associated to estrogen-induced tumorigenesis.  [30] J o u r n a l P r e -p r o o f

Conclusion
In recent years, metabolomics has been increasingly applied to murine models of BC in later years, using high throughput analytical techniques either based on Nuclear Magnetic Resonance (NMR) or Mass Spectromety (MS). Both tissue and biofluids have been analyzed, although in some types of models, reports focused primarily on the former. The aims of tissue studies mostly seek mechanistic knowledge of carcinogenesis, metastasis development and response/resistance to therapies. On the other hand, biofluid metabolomics usually has aimed to find non-invasive biomarkers for early BC detection and prognosis. We propose that, in future developments, increased consideration of both tissue and biofluids together may be carried out (rather than just one of these biological matrices), in order to obtain concomitant biochemical knowledge on local and systemic metabolisms associated to the disease. In addition, the clear complementarity of NMR-or MS-based methods justifies their increased tandem use to characterize polar and apolar metabolomes together (rather than only one of these alone, in each study), particularly in tumor tissues, so that increased metabolic information becomes accessible. Finally, a typical stumbling block in metabolomics applies also in the context of this paper, and regards the need to gear future research towards the unambiguous interpretation of putative biochemical hypotheses advanced to tentatively explain changes in metabolite levels given by metabolomic results. This may be achieved by in tandem biochemical/biological measurements (e.g. enzymatic activities or protein expression levels) of possible protein players and/or isotopic tracing studies, which have the ability of highlighting particular deviant pathways as a means to explain complex metabolome responses. Unambiguous understanding of the origin of potential metabolite biomarkers is certain to aid in the translation of metabolomic findings to the clinic and is, thus, a necessary follow-up of most metabolomic studies of BC in murine models.      Reprinted with permission from reference [51]. .Reprinted with permission from reference [56].