Animal models
The MAS98.06 and MAS98.12 patient-derived breast cancer xenograft models were established at the Institute of Cancer Research, Oslo University Hospital, as previously described [18]. The models have previously been classified as luminal B and basal-like molecular subtypes, respectively [19]. MAS98.12 has a triple-negative phenotype, whereas MAS98.06 is estrogen- and progesterone-receptor positive, and strongly dependent on estradiol supplement for tumor growth [19, 20]. Tumor tissue was bilaterally and orthotopically transplanted into 5- to 6-week-old (18–20 g) female Hsd:Athymic Nude-Foxn1nu mice. The animals were kept under pathogen-free conditions at a temperature between 19 and 22 °C, humidity between 50 and 60%, 20 air changes/h, and a 12-h light/dark cycle. The animals were fed RM1 diet (Scanbur BK, Karlslunde, Denmark) and distilled tap water ad libitum. The drinking water was supplemented with 17-β-estradiol at a concentration of 4 μg/ml in order to replicate the conditions described in [18]. An overview of the experimental design is shown in Additional file 1.
All procedures and experiments involving animals were approved by the Norwegian Animal Research Authority (FOTS ID: 7713 and 9126) and carried out according to the European Convention for the Protection of Vertebrates used for Scientific Purposes.
Tumor growth inhibition
Following bilateral transplantation, mice carrying MAS98.06 and MAS98.12 tumors (experiment 1) were kept until tumor volume reached approximately 60 mm3 (60.3 mm3 ± 32.4 mm3). Mice from both models were randomized to receive either CB-839 (200 mg/kg) or drug-free vehicle two times per day for up to 28 days (n = 5 per group). CB-839 (Calithera Biosciences, CA, USA) was dissolved in 10% cyclodextrin/saline solution and administered orally per gavage. Tumor length (l) and width (w) were measured using a digital caliper, and tumor volumes were calculated using the formula \( V=\frac{1}{6}\left(\uppi \times l\times {w}^2\right) \). The mice were sacrificed at day 14 (MAS98.12) and day 28 (MAS98.06) when untreated tumors approached the upper volume limit. Untreated tumors were collected for histopathological examination and HR MAS NMR (natural abundance) analysis.
Gene expression analysis
Gene expression analysis was performed (experiment 2) using previously generated and published data [21]. Briefly, RNA was isolated from tumor tissue from six animals from each xenograft models and hybridized to 4 × 44 k Agilent Whole Human Genome Oligo Microarrays according to the manufacturer’s protocol. The microarray data was normalized and analyzed using R(v 2.9.0) and the LIMMA Bioconductor package [22], normalized and log2 transformed. The microarray data is accessible through GEO Series accession number GSE37543. A total of 60 genes were selected for analysis, based on the KEGG maps central carbon metabolism in cancer (map 05230), arginine and proline biosynthesis (M00330), and glutathione biosynthesis (M00118) [23], which outline potential metabolic fates of glutamine in human cells.
Normalized and log2-transformed data was imported into Qlucore Omics Explorer 3.3 (Qlucore AB, Lund, Sweden) for statistical analysis. Testing for differential expression of genes between the xenograft models was performed using t tests with Empirical Bayesian correction of the test statistics [22]. To account for multiple testing, an adjusted q value of 0.05 (using Benjamini & Hochberg’s false discovery rate) was defined as the threshold for statistical significance [24].
The heatmap was generated in R (v 3.3.2) using RStudio (v 1.1.447). Hierarchical clustering was performed using the in-house made R-package Clustermap [25]. In brief, median-centered and log2-transformed RPPA data were clustered using Euclidean distance and complete linkage. For heatmap visualization of the data, values are normalized to the range [− 1, 1] by application of a nonlinear sigmoid transformation f(x) = tanh(x). This limits the visual dominance of outlier values while maintaining the order of the values, since f is strictly increasing.
To determine whether the metabolic characteristics of MAS98.06 and MAS98.12 xenografts are representative of the luminal B and basal-like subtypes of breast cancer, respectively, we accessed a previously published gene expression data set that in total includes 19 basal-like and 7 luminal B PDX models [19]. Gene expression of SLC1A5, GLS1, GLUL, and GLUD1 was accessed and are displayed as waterfall plots in Additional file 2. The microarray data is available at the Gene Expression Omnibus (GEO) with accession number GSE44666.
Immunohistochemistry
IHC staining
Seven proteins (ALDH18A1, GLS1, GLUD1, GS, Myc, PYCR1, SLC1A5) were selected for protein expression analysis based on prior knowledge on their relevance in glutaminolysis. Firstly, glutamine transporters ensure uptake of glutamine into the cells, of which neutral amino acid transporter B(0) (coded by the gene SLC1A5 (Solute Carrier Family 1 Member 5), hereby abbreviated as SLC1A5) has received high attention since it has been shown that increased expression correlates with poor patient prognosis in many cancer types [9, 26,27,28]. Glutamine synthetase (GS) is the enzyme that catalyzes the conversion from glutamate to glutamine. Absence of GS is correlated with high GLS1 activity and can be associated with glutamine addiction in invasive and aggressive breast cancer phenotypes [29, 30]. Some cancer cells also rely on glutamate dehydrogenase 1 (GLUD1)-mediated Glu deamination to fuel the TCA cycle [31]. Glutamine can be converted to proline via glutamate, and the enzymes pyrroline-5-carboxylate reductase 1 (PYCR1) and aldehyde dehydrogenase 18 family member A1 (ALDH18A1) are found to be key enzymes in the conversion [6, 32, 33]. The proto-oncogene Myc is shown to be a key regulator of glutaminolysis affecting glutamine uptake, GLS1 activity, and proline metabolism [4, 26, 34,35,36].
Tumors from untreated mice (experiment 1) were fixed in 10% neutral-buffered formalin and embedded in paraffin. Information about primary antibodies are presented in a table in Additional file 3. GLUD1, SLC1A5, ALDH18A1, and PYCR1 staining were performed at the Cellular & Molecular Imaging Core Facility (CMIC), NTNU, Norway. GLS1, GS, and cMYC staining were performed at Covance Laboratories Inc., (Greenfield, USA). The following protocols were applied: Tumor specimens were cut into 4 μm sections which were dried, deparaffinized, and rehydrated. Heat-induced antigen retrieval was then performed for 10 (Covance, Antigen Retrieval Solution, Leica, AR9661) or 20 (CMIC, Target Retrieval Solution: Dako, low pH 6, K8005) minutes. Endogenous peroxidase activity was quenched with peroxidase block (H2O2). Sections were incubated with primary antibodies (Additional file 3) for 15 (Covance) or 40 (CMIC) mins in room temperature. Immunohistochemical reactions were visualized as specified by the vendor using either (Covance) Dako Rabbit Envision+HRP with DAB+ for use with rabbit primary antibodies (K4011, Dako) or (CMIC) Bond PDAB Reagent Kit (Leica DS9800). Sections were counterstained with hematoxylin.
IHC evaluation
Immunohistochemical markers were evaluated by a semiquantitative approach used to assign the histo-score (H-score) for each tumor. The H-score is given as the sum of the percentage of staining multiplied by an ordinal value corresponding to the intensity level (0 = negative, 1 = weak, 2 = moderate, 3 = strong): [1 × (% cells 1+) + 2 × (% cells 2+) + 3 × (% cells 3+)]. The final score, ranging from 0 to 300, gives more relative weight to higher intensity labeling in a given tumor sample [24, 37, 38].
An experienced pathologist from Covance Inc. scored slides labeled with antibodies against GLS1, GS, and cMYC. Two researchers (MTG and SAM) scored each slide from GLUD1, SLC1A5, ALDH18A1, and PYCR1 independently in a blinded manner. Final H-score is mean H-score ± SD from the two researchers. Some MAS98.06 tumors had central necrosis, and areas with necrosis were excluded from the analysis.
Glutamine metabolism
Another group of mice (experiment 3) carrying MAS98.06 (n = 12, 30.0 ± 10.3 mm2) or MAS98.12 (n = 11, 41.4 ± 10.6 mm2) were randomly distributed to receive CB-839 or drug-free vehicle for 2 days as described above. Three hours after the final treatment, animals received an intravenous infusion of 13C-enriched glutamine (99% enrichment, Cambridge Isotope Laboratories) while under isoflurane anesthesia, as described in [39]. The mice received 1.2 mg [5-13C] glutamine/g body weight, dissolved in sterile PBS. A 3-min bolus of 0.3 mg/g body weight was followed by continuous infusion of 0.005 mg/g body weight/min for 180 min, in a total infusion volume of 0.8 to 1.1 ml (depending on body weight). The mice were sacrificed, and tumor tissue samples were collected, snap frozen, and stored in liquid nitrogen until NMR analysis.
NMR experiments
NMR spectroscopy of tumor tissue samples
Tumor samples from both 13C glutamine-labeled (n = 5 to six per group, experiment 3) and unlabelled (N.A.: natural abundance, n = 3 per group, experiment 1) tumors (39.9 ± 1.1 mg) were cut to fit into a 50-μl zirconium HR MAS rotor (4-mm diameter). Lock reference containing D2O with formate (25 mM) was added to the rotor (~ 16 μl). The HR MAS MR spectra were recorded using a Bruker Advance DRX600 spectrometer (14.1 T) (Bruker Biospin GmbH, Germany) containing a 1H/13C MAS probe. Samples were spun at 5 kHz at magic angle, and the temperature was kept at 4 °C during the whole experiment. NMR spectra were acquired using the following NMR sequences and acquisition parameters: One-dimensional 1H NOESY pulse sequence with water presaturation (Bruker; noesygppr1d). Acquisition time was 2.7 s, repetition time 6.7 s, sweep width was 30 ppm, and 128 scans were acquired. The 13C MR spectra were acquired using a single pulse experiment, with 1H decoupling applied during recycle delay and acquisition (Bruker; zgpg30). The flip angle was 30°, acquisition time 0.9 s, repetition time 1.9 s, sweep width 250 ppm, and 16 k scans were obtained. Total acquisition time per sample was 9 h.
Analysis of NMR spectra
NMR spectra were Fourier transformed after application of line broadening (1H: 0.3 Hz, 13C: 1 Hz for tumors), and the chemical shift scale was calibrated to a reference peak 1H: alanine at 1.48 ppm, 13C: [5-13C] glutamine ([5-13C] Gln) at 180.4 ppm.
For quantification of tumor metabolites from 1H NMR spectra, ERETIC2 (Bruker), which is based on PULCON (PULse length-based CONcentration determination), was applied [40]. Each spectrum was scaled to sample mass, and a total of four metabolites (alanine, glutamate, glutamine, and lactate) were quantified (relative) using Chenomx software (version 8.1, Alberta, Canada). Tumor 13C NMR spectra were analyzed using MATLAB R2017a (The Mathworks, Inc., USA). Baseline correction was applied using an asymmetric least squares algorithm [41], and the NMR spectra were scaled to NMR sample mass. For determination of 13C-enriched peaks in tumors, 20 relevant peaks from the 13C NMR spectra were selected and integrated, and peaks that were significantly higher (Student’s t test, p < 0.05) in [5-13C] labeled compared to natural abundance tumors were considered 13C enriched. The amount of eight 13C-labeled metabolites in the tumors were calculated by subtracting natural abundance spectra from 13C-enriched spectra. The results are shown in Additional file 4 (panel c).
The different experimental groups (13C-enriched controls versus natural abundance controls from both models, 13C-enriched CB-839-treated versus 13C-enriched controls from both models, and 13C-enriched MAS98.06 controls versus 13C-enriched MAS98.12 controls) were compared statistically using Student’s t test. P values less than 0.05 were considered statistically significant.