Presence of human breast cancer xenograft changes the diurnal profile of amino acids in mice

Human xenografts are extremely useful models to study the biology of human cancers and the effects of novel potential therapies. Deregulation of metabolism, including changes in amino acids (AAs), is a common characteristic of many human neoplasms. Plasma AAs undergo daily variations, driven by circadian endogenous and exogenous factors. We compared AAs concentration in triple negative breast cancer MDA-MB-231 cells and MCF10A non-tumorigenic immortalized breast epithelial cells. We also measured plasma AAs in mice bearing xenograft MDA-MB-231 and compared their levels with non-tumor-bearing control animals over 24 h. In vitro studies revealed that most of AAs were significantly different in MDA-MB-231 cells when compared with MCF10A. Plasma concentrations of 15 AAs were higher in cancer cells, two were lower and four were observed to shift across 24 h. In the in vivo setting, analysis showed that 12 out of 20 AAs varied significantly between tumor-bearing and non-tumor bearing mice. Noticeably, these metabolites peaked in the dark phase in non-tumor bearing mice, which corresponds to the active time of these animals. Conversely, in tumor-bearing mice, the peak time occurred during the light phase. In the early period of the light phase, these AAs were significantly higher in tumor-bearing animals, yet significantly lower in the middle of the light phase when compared with controls. This pilot study highlights the importance of well controlled experiments in studies involving plasma AAs in human breast cancer xenografts, in addition to emphasizing the need for more precise examination of exometabolomic changes using multiple time points.


Intracellular metabolites profiling varied in breast cancer cells. The metabolomic profiles of breast
cancer cells and normal cells were carried out over 24 h to identify the differences in concentration of intracellular metabolites. In vitro clustering analysis of the metabolites per each class was performed and revealed separation of intracellular metabolites between malignant and normal cells with no effect of time ( Figure S1). Amongst six classes of compounds analyzed using the AbsoluteIDQ p180 kit, we observed separated metabolite profiles with the AAs, lyso-and phosphatidylcholine, Biogenic amines and Sphingomyelins.
AAs are significantly altered in cancer cells across 24 h. 21 AAs were potentially detectable in both cell lines; with the exception of Citrulline (Cit) where the levels were not consistent and Ornithine (Orn) which was undetectable in normal breast cells (data not shown), most of the metabolites were stable and did not vary across 24 h. Whilst two AAs, Alanine (Ala) and Arginine (Arg), were reduced in MDA-MB-231 cells (p < 0.0001), the majority of the AAs including Asparagine (Asn), Glutamate (Glu), Glutamine (Gln), Glycine (Gly), Isoleucine (Ile), Leucine (Leu), Lysine (Lys), Methionine (Met), Phenylalanine (Phe), Serine (Ser), Threonine (Thr), Tryptophan (Trp), Tyrosine (Tyr) and Valine (Val) were found to be consistently increased (p < 0.0001) compared with MCF-10A cells (Fig. 1). At earlier time points, Aspartate (Asp) and Proline (Pro) levels were lower in cancer cells compared with higher levels in non-tumorigenic cells after 24 h (Fig. 1B). Histidine (His) was visibly higher in tumor cells at most time points albeit not significantly. MDA-MB-231 cells are known to have an aggressive phenotype with propensity to metastasis yet they do not have robust circadian rhythms 23 . We used the in vitro data set to analyze the AAs profile per time point. Table 1 shows the analysis of the AAs concentrations between experimental groups in each time point of cell growth (two-way ANOVA followed by Bonferroni's test). The intracellular AAs profile revealed that only cancer cells had significant alterations (p < 0.0001), but day and night time had no significant direct impact on AAs together with the association of length and growth of tumor. Plasma AAs are affected by time of day and presence of a tumor. Following in vivo experiments, plasma samples were taken since they are a reliable source of blood biomarkers. We observed no changes in the body weight of the mice over the 22 days of tumor-hosting ( Fig. 2A). The tumor volumes varied from 50 to 98 mm 3 and the tumor tissue did not show necrotic regions post excision (Fig. 2B).
Evaluation of the concentrations of all plasma metabolites in the eight time points revealed that AAs were the only metabolite class significantly altered between control and tumour bearing mice. (Fig. 1) Analysis of variance showed that in vivo, plasma AAs levels in mice were significantly affected not only by tumor presence (p < 0.0015) and diurnal variation (p < 0.0001), but also by the time/tumor interaction (p < 0.0001).
In contrast to that observed early in the day, the middle of the light phase showed significantly lower plasma AAs levels in tumor-bearing mice compared with control mice. These include Ala (p < 0.01); Arg (p < 0.01); Asn (p < 0.01); Cit (p < 0.01); Lys (p < 0.001); Met (p < 0.05); Orn (p < 0.05); Phe (p < 0.05); Pro (p < 0.001); Ser (p < 0.001); Val (p < 0.05). In early dark phase, we observed a number of AAs slightly decreased in tumor-bearing mice, whereas healthy mice showed significant increase of Asn (p < 0.001); Cit (p < 0.05); Ile (p < 0.001); Leu (p < 0.01); Orn (p < 0.05); Phe (p < 0.05); Pro (p < 0.01); Ser (p < 0.01); Val (p < 0.01). At ZT15, the plasma levels of AA in both groups were similar with no statistical significance (Fig. 2C). The overall observed C max of Arg, Asp, Lys, Met, and Orn was significantly higher in plasma of tumor-bearing mice compared with control animals. The C max occurred at different times in both tumor bearing mice and non-tumor bearing mice and the C min occurred at the same time.
Although clustering analyses clearly separated the profile of AAs between tumor-bearing and non-tumor bearing mice across 24 h, a noticeable reduction in plasma AAs levels by tumor-bearing animals occurred during advanced light and dark phases (Fig. 3). During the day, we identified four main clusters of AAs varying among the groups (Gln, Ala-Lys, Gly, and other AAs), whereas in the dark phase the main distinct clusters were Gln, Ala, Lys-Gly-Leu-Val, and other AAs (Fig. 3 A). Whilst most AAs were significantly increased in tumor bearing animals during the initial light phase, a significant reduction of AAs occurred during dark phases. Examination of AAs concentrations throughout the day and night, showed a different metabolite profile pattern in the presence of tumor. A significant reduction of plasma AAs was observed in the dark phase, for the following AAs (Arg, Asp, Lys, Gly, Ser, and Tyr) (Fig. 3B).
Enrichment analysis for altered metabolic pathways. The enrichment analysis was based on AAs profile depicting at least four time points with significant differences between the in vitro experimental groups and the shifting plasma profiles in the early light phase. Thus, the enrichment analysis was performed with Arg, Asn, Cit, Ile, Leu, Lys, Orn, Phe, Pro, and Ser, and displayed 11 human metabolic pathways in which those AAs are significantly involved. The highest significant involvement and most enriched pathways Arg, Pro and Asp www.nature.com/scientificreports/  www.nature.com/scientificreports/ processes including AAs amino acids catabolism, cell growth, proliferation, migration, invasion, metastasis and synthesis of biomolecules (Fig. 4).

Discussion
Our study shows that mice bearing human breast cancer MDA-MB-231 xenografts exhibit broad differences in the rhythm of plasma AAs compared with the non-tumor bearing mice. The C max of most AAs is observed in the light phase before mid-day in tumor bearing mice in contrast to that occurring in the middle of the dark phase in control animals. There is no real difference in C max yet the C min is decreased in tumor bearing mice where it occurs at the same time as the control mice. In this study, we emphasized that measurement of plasma AAs levels at a single time could lead to opposite results depending on time-sampling. We reported a clear metabolite separation between malignant and normal conditions in vitro, across all time points evaluated. There were no rhythmic variations in metabolites between normal and cancer cells over the four time points across 24 h. Meanwhile, in vivo, metabolites were clustered according to time and at the opposite corresponding time points. In our xenograft model, we did not observe any change in the body masses of mice hosting the tumor. In these animals, specific rhythmic AAs variation was clearly observed between the groups and during the analysis with eight time points across 24 h (Fig. 2). In tumor-bearing mice, we observed lower levels of selected AAs in late dark phase which tended to increase until the early light phase; in contrast to healthy mice where AAs levels were higher in late dark phase and decreased plasma levels in early light phase. Tumor-bearing mice showed peak levels of more than half of the AAs with a significant shifting profile of those selected AAs in early light phase (ZT0 and ZT3). However, in late light phase (ZT6), all plasma AAs levels of tumor-bearing mice, excluding aspartate, were decreased compared with healthy mice. At ZT9, the plasma levels of selected AAs decreased and showed no significant difference between the groups. In early dark phase (ZT12), www.nature.com/scientificreports/ tumor-bearing mice had significantly lower plasma levels of selected AAs compared with control mice. Interestingly, at ZT15, all the AAs showed similar concentrations with no significance in both experimental groups. None the less, the other metabolite classes analyzed showed two or one time point(s) with significant differences between the groups and yet did not show the similar profile alteration between the groups, observed with the AAs. In the current study, food was provided ad libitum and the likelihood of animals feeding at different times is unlikely. Should this be the reason for our observed differences? If various feeding times had affected the metabolic profiles, one would also expect to see parallel differences on other classes of metabolites (e.g., carnitines) which was not observed suggesting that variation in food intake is not responsible for the changes observed.
The reason for the current observation is not clear at all. Different clocks are present in different tissues and it has recently been shown that metabolism in tumor cells is driven primarily by the local tissue metabolism 25 . Of note, the gene expression rhythm profile in the liver of mice bearing the 4T1 breast cancer xenograft was modified and a subset of clock genes showed a secondary peak when compared with control animals 26 . In this study, there was no evidence that tissue or circulating amino acids were altered. However, in another study with genetically engineered mice, the development of lung adenocarcinoma driven by conditional activation of the KrasG12D mutation and loss of the p53 tumor suppressor led to altered circadian patterns of transcription of a subset of genes in the liver 27 . This phenomenon was observed without alterations to the core circadian clock machinery. These transcriptional changes resulted in altered patterns of liver metabolism across the circadian cycle, including insulin signaling, glucose production, and lipid metabolism. None of these studies evaluated the circulating metabolites. It is therefore possible that the presence of a breast cancer MDA-MB-231 cells-originated tumor affects liver metabolism resulting in altered plasma AAs; this needs further validation 26 .
Changes in AAs metabolism and concentration in cancer patients have been consistently investigated 20,28 . A study with plasma AAs reported decreased levels in pancreatic and breast cancers and these levels were correlated with an enhanced consumption of AAs by the tumor 20 . Furthermore, several AAs have been reported with lower levels in patients with breast, gastric, and thyroid cancer 29,30 . Conversely, plasma AAs levels have been shown to be increased in a number of studies considering the early stages of cancer. In addition, plasma AAs levels have been shown to be tumor-dependent and to correlate with specific breast cancer subtypes 21,31,32 . Previous investigation focusing on breast cancer using a metabolomics approach showed that AAs metabolism represent some of the changes in metabolic activity of several pathways associated with breast cancer 33 . It also should be noted that cellular uptake of AAs correlates with cancer cell proliferation as their exogenous supply may critically affect the proliferative activity, survival or tumorigenic potential of cancer cells 34 .
Indeed, solid tumors provide a nutrient-poor environment where cells often outgrow their small molecules from the host blood supply. It is possible that cancer metabolism affects the circulating AAs levels and its availability may be clock-controlled [17][18][19] . The enrichment analysis shows major significant association with three metabolic pathways such as Arginine and Proline, Glycine and Serine metabolism, and Urea Cycle, all involved with molecular biosynthesis and degradation. In addition to the Warburg effect, metabolism of AAs, i.e. glutamine, serine, aspartate, and proline, has been shown to contribute to tumor metabolic reprogramming [35][36][37] . Altered levels of Arg plus Orn and Cit may have affected the urea cycle in our animal model. This pathway has a plethora of important self-maintained enzymatic steps such as argininosuccinate synthetase 1 (ASS1) and ornithine transcarbamylase (OTC). These enzymes have been reported in humans and mice, displaying clockdependent activity in a rhythmic manner, which contributes to the temporal plasma variation and biosynthesis of these AAs 17 . Circadian rhythm dependency has also been documented regarding homocysteine levels as being an enzyme component cyclically coordinated by the clock genes 37 .
De novo Pro, Ser and Gly synthesis has been shown to be critical in several types of cancer 37,38 . Pro catabolism is reported in in vivo metastasis formation through high expression of proline dehydrogenase (PRODH) compared to primary breast cancers of patients and mice. In fact, Pro can be interconvertible with Glu, in which glutamic-c-semialdehyde (GSA) is utilized as an intermediate, which is derived from Glu or Pro and converted into Orn, serving as a precursor for Arg synthesis in the Urea cycle 39 . Meanwhile, Ser metabolism may be an important factor for breast cancer as Ser biosynthetic pathways were upregulated in high metastatic breast cancer 40 . Glycine metabolism is also involved in cancer cell proliferation being able to promote tumorigenesis and malignancy where it provides carbon and nitrogen for energy 41 . Research into plasma AAs levels started decades ago with Bennegård et al. (1984) which examined 18 cancer patients with more than 7% weight loss and found a significant decrease in Pro and Ser levels 42 . Watanabe et al. (1984) have also investigated PFAA concentrations in 14 cirrhotic patients with hepatocellular carcinoma (HCC) and found significant decreases in the levels of Ser 43 . A well described phenomenon that affects plasma amino acid in cancer is cachexia. Plasma levels of AAs have been shown to increase or decrease in various studies undergoing significant muscle loss 44 . These differences could potentially be due to sampling time collection but also to the fact that it is difficult to trace the source of effects in plasma (e.g., the tumor or a peripheral effect such as cachexia or liver metabolism alteration). In our in vivo study, the tumors were relatively small and no body weight loss was observed suggesting no associated cachexia.
In our study, there was no sign of rhythmic intracellular concentrations in MD-MBA-231 and non-tumorigenic mammary cells within 24 h of standard cell culture. Conversely, the plasma AAs in in vivo experiments showed an evident and significant cyclical behaviour during diurnal variation in mice. Thus, AAs were observed being increased or decreased depending on daytime. The lack of rhythmic variation in our in vitro study is likely due to the cells having their own biological clocks or the tissue culture conditions (different from in vivo xenografts), which are peripheral and non-synchronized within any central circadian oscillator. The mice might have their circulating metabolites under their own controlling central oscillator. Once in the xenograft model, the cells could have resynchronized their clock according to the new host's central oscillator 45 . We suggest that depending on the time we may verify an increased or a decreased level which may be a general tumor association with the diurnal effect. www.nature.com/scientificreports/ There are no currently reported data supporting a circadian control of Pro and Asp metabolism and the enriched pathways. Most of the pathways such as Met, Tyr, Phe and Biotin metabolism, Val, Leu, Ile and Lys degradation, carnitines synthesis, ammonia recycling are subjected to metabolic reprogramming to sustain cell growth, proliferation, and oncogenes expression in many cancer types; these are essential to accumulate building blocks for the construction of new cellular components [46][47][48][49] .
Tumors also can affect circadian rhythms even in tumor-free organs. For example, mice bearing triple negative breast cancer can impact the hepatic circadian gene expression causing alterations in several genes including the core clock genes Rev-Erba (Nr1d1), PER2, RORγ, and CLOCK 26 . Dysregulation of these physiological oscillations can further result in oxidative stress, polyploidy, and inflammation. The presence of a non-metastatic melanoma significantly impaired the biological clock of tumor-adjacent skin and affected the oscillatory expression of genes involved in light-and thermo-reception, proliferation, melanogenesis, and DNA repair. The expression of tumor molecular clock was significantly reduced compared to healthy skin but still displayed an oscillatory profile (attenuated PER1 and BMAL1 oscillations) 50,51 . Although not affecting the core clock machinery, Masri et al. also found that lung cancer induced significant shifts in liver circadian rhythms, both at the level of transcripts and metabolites including AAs in tumor-bearing mice. Yet, nutrient addiction of a tumor subtype can differ. For instance, when glutamine levels are limiting, cell proliferation can be alternatively driven by aspartate or asparagine 52 . This establishes an interesting paradox whereby limiting the availably of a specific nutrient may not inhibit cancer cell proliferation, as tumors can switch their fuel preference. It should be emphasized that plenty of these nutrients are rhythmically offered over the day, thus it may be possible that fuel utilization of tumors may differ based on rhythmic availability of nutrients [53][54][55] .
The recent crosstalk between the gut microbiota and cancer emerged to directly and/or indirectly correlate with tumor development, treatment, and prognosis [56][57][58] . Disruptions to gut microbiota balance or "dysbiosis" is implicated in a growing list of cancers and biological processes including host cell proliferation and apoptosis, immune system function, chronic inflammation, oncogenic signaling, hormonal, and detoxification pathways 57,58 . Most of the studies report the main effects of gut microbiota on cancer development correlating with the inflammation caused by bacterial infectious agents 59,60 ; the indirect effects are often linked to genomic instability through direct genotoxic effects on DNA and by modulating epigenetic mechanisms 60 .
The development of breast cancer has been associated with intestinal microbiota dysbiosis since certain gut bacteria alter the production of the beneficial anticancer metabolites and disrupt estrogen metabolism 56 . There are plenty estrogen-dependent and non-estrogen-dependent functions of the gastrointestinal microbiome involved in the production of bioactive metabolites 56 . Also, regulation of free AAs across the course of its digestion and absorption is influenced by the resident gut microbiota and could reflect on the distribution in the gastrointestinal tract in mice 61 . Moreover, the gut microbiota performs a key function in producing AAs, and this includes de novo biosynthesis 61 . For instance, some AAs are essential for carbon skeletons such Leu, Trp, and His which cannot be synthetized by the cell and are required not only from the diet but also from the intestinal microbiota 62 . However, we reported, through our in vivo model, only alterations in the Leu levels. The role of AAs is also reported regarding the sulphate reduction that might result in genomic DNA damage. These AAs from the dietary protein contribute to H 2 S production by sulfate-reducing bacteria (SRB) and has been shown to have genotoxic effects in vitro 60 . Likewise, biogenic amines such as cadaverine were observed having an impact on the Trace amino acids receptors TAARs in mice with breast cancer 59 ; in our study, we observed no alterations on the biogenic amines profile. Further investigation on how small molecules can affect cancer progression, particularly AAs, might be helpful in defining the functional relationship between microbiota and host. It is extremely necessary to classify AAs profile derived from tumor or the host physiology through metabolomics analysis based on liquid biopsies to increasingly correlate with their diurnal variation.
Caveats and limitations of the current study rely on the sections of experiments which were performed: i) We observed a not class-attributable variation between cancer cells (in vitro analysis) which were more spread compared to normal breast cells. The impairment regarding the replicability of the AAs profiling in cancer cells may be due to its normal variation which were also described by other studies. Our PLSDA models presented the lack of more accurate parameters (Supplementary figure 1); however, some metabolomics studies have reported a similar profile pattern in terms of more spread separation regarding cancer conditions. Past studies have already shown not only similar spread metabolite profile in cancer cells, but also metabolite variations observed in urine samples of cancer patients compared with healthy groups 63-65 . ii) The in vivo study limitation includes the use of a single animal model of tumors and is needed to confirm these findings in different human tumor xenograft models. Furthermore, although our data shows that AAs profile is not significantly altered by feed, the lack of a deeper investigation of the food intake habits may hinder the interpretation of AAs profile.
In summary, our report highlights the importance of the frequency of sampling and the potential effect of evaluating single or few time-points data which may not be appropriate in monitoring accurately tumor-related changes of circulating metabolites through daily rhythm. Our findings reveal an important variation of amino acids levels in MD-MB-231 breast cancer xenografts over the 24 h in contrast to a very low variation observed in isolated breast cancer cells. Therefore, additional experiments are needed to substantiate this effect in additional human tumor xenograft models of various genetic backgrounds. Although obvious advantages using human xenografts exist, the use of allograft syngeneic mouse models with reduced genetic and metabolic differences could generate valuable information to properly decipher the biological rhythms in cancer investigation from preclinical to clinical trials. . Animals were kept in pathogen-free conditions, at 21 to 25 °C and exposed to a normal diurnal variation under 12 h of light and 12 h of dark with food and water available ad libitum. MDA-MB-231 human tumor breast cells were grown, harvested and re-suspended with serum free media at a concentration of 6 × 10 7 cells per mL. Mice were subcutaneously inoculated with 100 µL of cell suspension in the fat pad, which requires more time for consistent growth compared with hind flank [22][23][24][25] , in addition to accounting for reduced tumor volumes. The animals were randomly divided into two groups: non-tumor bearing (n = 5) and tumor bearing (n = 5), which were partitioned into 8 time points within 24 h. The control group received 100µL of vehicle solution to subject the animals to the same stress handling. Mice were periodically weighed and monitored during the experimental period and showed no variation within the groups. Tumor volumes were measured by sample, at each time point by digital caliper and calculated based on the measurements by the formula: tumor volume = 0.5 (length × width 2 ).
Plasma sampling for targeted metabolomic analysis. Mice were anesthetized with ketamine-xylazine combination and euthanized by open cardiac puncture at eight time points (Zeitgeber time-ZT) every three hours over a 24 h period: 06:00 am (ZT0), 09:00 am (ZT3), 12:00 pm (ZT6), 03:00 pm (ZT9), 06:00 pm (ZT12), 09:00 pm (ZT15); 12:00 am (ZT18) and 03:00 am (ZT21). During the dark time points, the sampling was performed under systematic and standardized steps. We avoided artificial light inside and near the vivarium, by using red dim light to illuminate the room whilst blood sampling was carried out on the bench. The blood was collected using a heparinized syringe and the plasma was subsequently extracted by centrifugation at ~ 5,000 RCF for 2 min at 4 °C. The supernatant was recovered and stored at-80 °C until further analysis. Mice were not fasted, not only because of the number of animals but also due to the number of time points. The sampling frequency of most chronobiology studies is typically 4-6 h and we achieved more time-points and collection over a longer duration. Therefore, it would be rather difficult to coordinate the sampling with the fasting conditions as well.
Targeted metabolomic analysis. Metabolites were measured using the AbsoluteIDQ p180 targeted metabolomics kit (Biocrates Life Sciences AG, Innsbruck, Austria) which covers 6 classes of metabolites including amino acids, biogenic amines, acylcarnitines, phosphatidylcholines, lysophosphatidylcholines, and sphingomyelins, on a Waters Xevo TQ-S mass spectrometer coupled to an Acquity H-Class LC system (Waters Corporation, Milford, MA, USA). Data analysis. Data were processed with MassLynx V4.1 and validated by MetIDQ software (Biocrates Life Sciences AG, Innsbruck, Austria). The first analysis were performed through Principal Component Analysis (PCA) followed by Partial Last Squares Discriminant Analysis (PLS-DA) by Metabonalyst 5.0. In sequence, heatmaps analysis per each class of metabolite were also performed through Metabonalyst 5.0 (Supplementary figures). The p values (< 0.05) used were generated from the rhythmic analysis for every single AAs between the groups in each time points for both assays (in vitro and in vivo); two-way analysis of variance adjusted by Bonferroni test was performed using GraphPad Prism. For the enrichment pathway evaluation we still used Metaboanalyst 5.0 database. The heatmaps used for clustering analyses (Fig. 3) were performed using the web tool Morpheus (https:// softw are. broad insti tute. org/ morph eus). www.nature.com/scientificreports/