Investigating metabolic rewiring in cancer can lead to the discovery of new treatment strategies for breast cancer subtypes that currently lack targeted therapies. In this study, we used MMTV-Myc–driven tumors to model breast cancer heterogeneity, investigating the metabolic differences between two histologic subtypes, the epithelial–mesenchymal transition (EMT) and the papillary subtypes. A combination of genomic and metabolomic techniques identified differences in nucleotide metabolism between EMT and papillary subtypes. EMT tumors preferentially used the nucleotide salvage pathway, whereas papillary tumors preferred de novo nucleotide biosynthesis. CRISPR/Cas9 gene editing and mass spectrometry–based methods revealed that targeting the preferred pathway in each subtype resulted in greater metabolic impact than targeting the nonpreferred pathway. Knocking out the preferred nucleotide pathway in each subtype has a deleterious effect on in vivo tumor growth, whereas knocking out the nonpreferred pathway has a lesser effect or may even result in increased tumor growth. Collectively, these data suggest that significant differences in metabolic pathway utilization distinguish EMT and papillary subtypes of breast cancer and identify said pathways as a means to enhance subtype-specific diagnoses and treatment strategies.

Significance:

These findings uncover differences in nucleotide salvage and de novo biosynthesis using a histologically heterogeneous breast cancer model, highlighting metabolic vulnerabilities in these pathways as promising targets for breast cancer subtypes.

Breast cancer remains the leading cause of cancer-related mortality among women worldwide despite recent trends in decreasing mortality in high income countries (1), which can be attributed to advances in early detection and treatment (2). Current treatment strategies for advanced breast cancer often include general chemotherapy and radiotherapy, with the use of targeted therapies, such as endocrine therapy, for specific breast cancer subtypes (3). These subtypes are often defined based on expression of specific receptors including the estrogen receptor (ER), progesterone receptor, and HER2, with an additional triple-negative breast cancer (TNBC) subtype characterized by the absence of these markers. Breast cancer subtypes can also be classified according to gene expression patterns (4, 5), which often overlap with definitions based on receptor status and other clinical findings (3, 5) and are further able to provide valuable prognostic information (6). However, targeted therapies are not available for all subtypes of breast cancer, and current rates of recurrence and development of resistance remain problematic (7, 8). It is becoming increasingly clear that breast cancer subtypes have differences in metabolism, and targeting these metabolic pathways could provide new targeted therapy options (9, 10).

Metabolic rewiring is a hallmark of cancer (11), and significant efforts have been made to identify metabolic vulnerabilities in cancer and leverage these findings to develop novel treatment strategies. Early work defining this concept was performed in the 1920s by Otto Warburg, who observed that tumor cells generally upregulate glycolysis even in aerobic conditions (12)—a phenomenon now known as the Warburg effect. One of the modern consequences of the Warburg effect is that targeting aerobic glycolysis, by pharmacologic inhibition of glycolytic enzymes and by limiting glucose availability through dietary restriction (13–15), is under investigation as a therapeutic strategy for many types of cancer. However, one of the challenges in using metabolic rewiring to treat cancer arises from the fact that cancer is a remarkably heterogeneous disease, and a few metabolic vulnerabilities are common to all cancers. This variability is clearly illustrated by breast cancer, which demonstrates heterogeneity on histologic, genetic, and metabolic levels (4, 9, 16, 17).

In addition to glycolysis, another metabolic pathway commonly targeted in cancer therapy is nucleotide biosynthesis. Nucleotides enable cellular proliferation by facilitating RNA and DNA production (18, 19), and are also required to balance basal rates of RNA turnover in all cells (20). Nucleotide biosynthesis occurs through two parallel metabolic pathways: (i) de novo nucleotide biosynthesis, which generates new nucleotides from precursors derived predominately from glucose and glutamine metabolism and is an energetically costly process, and (ii) nucleotide salvage, which allows free bases derived from catabolic processes to be recycled back into nucleotides and is significantly more energetically efficient (20).

In our current work, we investigate subtype-specific differences in nucleotide metabolism using two histologic mouse mammary tumor subtypes derived from the MMTV-Myc mouse tumor model: (i) MMTV-Myc epithelial–mesenchymal-transition (EMT) and (ii) MMTV-Myc papillary. This model system mimics the heterogeneity of human breast cancer (21), and subtypes of the MMTV-Myc model can be correlated with human cancer subtypes based on gene expression patterns: the EMT subtype strongly correlates with the claudin-low subtype, and the papillary subtype correlates more moderately with several human subtypes including basal and luminal breast cancer (22, 23). Because the claudin-low and basal subtypes both have poor prognosis (24, 25), we decided to focus on the corresponding MMTV-Myc EMT and papillary subtypes in this study.

We have previously used cell lines derived from this model system to identify metabolic differences between subtypes (26). Here, we build on this work by integrating genomic and metabolomic techniques to refine our understanding of the metabolic differences between the EMT and papillary subtypes. We find striking differences in nucleotide metabolism between the two subtypes: the EMT subtype prefers nucleotide salvage pathways, whereas the papillary subtype prefers de novo nucleotide biosynthesis. We further investigate the clinical significance of expressing genes related to de novo purine biosynthesis and salvage pathways, and evaluate the consequences of targeting these genes in each subtype using CRISPR/Cas9 gene editing techniques (27). We find that targeting the preferred metabolic pathway of each subtype generally caused the most substantial disruption on nucleotide metabolism and had subtype-specific effects on in vivo tumor growth. Notably, targeting the preferred pathway significantly reduced tumor growth, whereas targeting the nonpreferred pathway either had no effect on tumor growth or in some cases significantly increased tumor growth. These results highlight the metabolic heterogeneity of breast cancer subtypes and demonstrate the potential efficacy of tailoring therapies to inhibit subtype-specific metabolism.

Primary mouse tumors

All animal use was performed in accordance with institutional and federal guidelines. All studies and procedures involving animals were Institutional Animal Care and Use Committee approved. Primary MMTV-Myc EMT and MMTV-Myc papillary tumors were acquired as a gift from Dr. Eran Andrechek and have been previously described (28). Tumors were sectioned, formalin-fixed, and paraffin-embedded for histologic examination with hematoxylin and eosin staining. Wild-type EMT and papillary tumors were cryopreserved in a mixture of 90% FBS and 10% DMSO. Tumor-derived cell lines were established by mechanical dissociation of primary tumors using scissors, followed by culturing tumor pieces in cell culture media (29).

Metabolic profiling

Unlabeled, targeted metabolomics was performed as previously described (30). Briefly, cells were seeded in 6-well tissue culture plates at 50,000 cells/well and cultured for 48 hours. Cells were washed with saline (VWR, 16005–092), and metabolism was quenched by addition of cold methanol. Flash-frozen tumor tissue was pulverized using a liquid nitrogen–cooled mortar and pestle and cold methanol, and water was added to the tissue sample. The tissue samples were further processed using a Precellys Evolution homogenizer (Bertin Instruments) operating a single 10-second cycle at 10,000 rpm. Extracts were then transferred to 1.5 mL Eppendorf tubes, and cold chloroform was added to each tube and vortexed for 10 minutes at 4°C. The final metabolite extraction solvent ratios were methanol:water:chloroform (5:2:5). The polar phase was collected and dried under a stream of nitrogen gas. The dried metabolites were then resuspended in high performance liquid chromatography-grade water for analysis. LC-MS/MS analysis was performed with ion-pairing reverse-phase chromatography using an Ascentis Express column (C18, 5 cm × 2.1 mm, 2.7 μm, MilliporeSigma, 53822-U) and a Waters Xevo TQ-S triple quadrupole mass spectrometer. Mass spectra were acquired using negative mode electrospray ionization operating in multiple reaction monitoring mode. Peak processing was performed using MAVEN (31), and data for each sample were normalized to the mean signal intensity for all metabolites in the analysis. Metabolites were grouped by relationship to metabolic pathways. Heatmaps were generated using Cluster 3.0 (32) and exported using Java Treeview (33).

Gene expression analysis

Gene expression data for MMTV-Myc EMT and papillary data were downloaded from Gene Expression Omnibus (GEO) using accession number GSE15904. The following EMT CHP datasets were downloaded: GSM399180, GSM399202, GSM399204, GSM399217, GSM399226, GSM399235, GSM399238, GSM399252, and GSM399259. The following papillary CHP datasets were downloaded: GSM399183, GSM399184, GSM399196, GSM399197, GSM399200, GSM399216, GSM399222, GSM399234, GSM399241, and GSM399245. Gene set enrichment analysis (GSEA; ref. 34) was performed by converting gene expression data to the required file formats and using the GSEA software available to download from www.gsea-msigdb.org/gsea/index.jsp. Reactome (35) metabolism gene sets were identified as all participant and subparticipant gene sets under the Reactome Metabolism pathway (stable identifier R-HSA-1430728) and were downloaded from the MSigDB Canonical pathways collection (36). Differential gene expression was determined using Transcriptome Analysis Console (TAC) 4.0 software. Sample signals and statistical measurements were exported from TAC 4.0 software. Genes measured by multiple probes were individually numbered. Clustering was performed in Cluster 3.0 using log-transformed data, and genes were clustered using the uncentered correlation similarity metric and average linkage settings (32). Heatmaps were generated using Java Treeview (33).

Survival analysis

Survival curves were generated using Kaplan–Meier (KM) Plotter for Breast Cancer (37) using probe 209434_s_AT for PPAT and 213892_S_AT for APRT. For survival analysis using relapse-free survival (RFS), patients were separated by upper and lower terciles of expression using the trichotomization option to assess the effects of high and low gene expression without including patients with intermediate expression levels. For survival analysis using overall survival (OS), patients were dichotomized into upper and lower expression groups using median expression as a cutoff due to the smaller sample size of the OS datasets. Redundant samples were removed, and biased arrays were excluded as per the default quality control settings.

Cell lines and culture conditions

EMT and papillary tumor–derived cell lines were acquired as a gift from Dr. Eran Andrechek and have been previously described (28). Cell lines were cultured in DMEM (Corning 10–017-CM) with 25 mmol/L glucose without sodium pyruvate supplemented with 2 mmol/L glutamine (Corning, 25–005-CI), 10% heat-inactivated FBS (MilliporeSigma, 12306C), and 1% penicillin and streptomycin (Corning, 30–002-CI). Cells were maintained at 37°C with 5% CO2. Mycoplasma testing was performed using the Venor GeM Mycoplasma Detection Kit (MilliporeSigma, MP0025) and was last performed on August 11, 2020.

CRISPR/Cas9

Lentivirus-mediated CRISPR/Cas9 genome editing was used to achieve gene knockout (KO). Guide RNAs targeting APRT or PPAT were designed using the CRISPR-DO web application (38). Plasmids containing dual guide RNA, puromycin resistance, and Cas9 co-expression were acquired from VectorBuilder. Plasmids containing scramble guide RNA, puromycin resistance, and Cas9 coexpression were also acquired from VectorBuilder. APRT KO dual guide RNA sequences are guide (A) 5′-GTCGATCTTGCCGCTGTGCG-3′ and guide (B) 5′-GTGTGCTCATCCGGAAACAG-3′. PPAT KO dual guide RNA sequences are guide (A) 5′-CATACGAGGTACGCCACCAC-3′ and guide (B) 5′-TACGCGGTGCGAGATCCATA-3′ The nontargeting puromycin-resistant scramble guide RNA sequence is 5′-GCACTACCAGAGCTAACTCA-3′. Lentiviral envelope and packaging plasmids were acquired from addgene. The VSVG plasmid was a gift from Bob Weinberg (Addgene plasmid # 8454; http://n2t.net/addgene:8454; RRID:Addgene 8454). The psPAX2 plasmid was a gift from Didier Trono (Addgene plasmid # 12260; http://n2t.net/addgene:12260; RRID:Addgene 12260). To produce lentivirus, HEK293T cells seeded in 10-cm plates were transfected using lipofectamine 3000 (ThermoFisher Scientific, L3000015) with 10.0 μg lentivirus plasmids, 0.5 μg VSVG, and 5.0 μg psPAX2 plasmids. The following morning, fresh DMEM with 15% FBS and 1% P/S was added, and cells were grown for another 48 hours to generate virus. For transduction with lentivirus, the recipient EMT and papillary cells were seeded in 10-cm plates, and the supernatant of transfected HEK293T was collected and passed through 0.45 μm PVDF syringe filter. Note that 5 mL of the viral supernatant and 5 mL of fresh media were added to recipient cell plates with polybrene (Fisher Scientific, TR1003G) at a final concentration of 4 μg/mL. The cells were cultured for 24 hours followed by addition of fresh DMEM medium supplemented with 10% FBS and treatment for 10 days with 2 μg/mL puromycin for selection. After transduction, cell culture media were supplemented with 50 μmol/L nucleosides (adenosine, cytidine, guanosine, inosine, thymidine, and uridine) in DMSO across all conditions to provide extracellular nucleotides for cells with deficient de novo biosynthesis. The puromycin-selected cells were then resuspended to a concentration of 5 cells/mL and seeded 1 cell/well on 96-well plates. Surviving clones were expanded and analyzed for successful gene KO. Genomic DNA was extracted using DNeasy Blood and Tissue Kit (Qiagen) to check for successful gene editing. The following primer pairs were used for PCR expansion and sequencing (marked with *) of APRT guide A: 5′-GGGTCACTCTCCTGTCCTTG-3′ and 5′-AGGACAGAGCAGAGTTCGTC-3′*, APRT guide B: 5′-GAGCTGTTCAGAAGGCAGGT-3′*, and 5′-AGCGTTTCTGGGTGGTGTAA-3′, PPAT guide A: 5′-CTCAGGACGGTCAAGGCTAC-3′* and 5′-AAGATGCCTTTTGTCGGAGA-3′, and PPAT guide B: 5′-GCATACACCCCTCCTCAAGA-3′* and 5′-CATCAGAGACTGGCATAAGACG-3′. Tracking of Indels by Decomposition (TIDE) was used to evaluate successful gene editing (39).

Western blot analysis

Cell lysis and Western blot analysis were carried out according to standard protocols. The following dilutions of primary commercial antibodies were used as probes: 1:250 dilution of anti-APRT (Thermo Scientific, PA576741), 1:500 dilution of polyclonal anti-PPAT (Proteintech 15401–1-AP), 1:1,000 dilution of monoclonal anti-PPAT (Origene, TA504769), 1:1,000 dilution of anti-vinculin (Cell Signaling Technology, E1E9V). Anti-APRT and anti-vinculin antibodies were diluted in 5% BSA and incubated overnight at 4°C. Polyclonal anti-PPAT antibody 15401–1-AP was diluted in 5% milk and incubated for 60 minutes at room temperature, whereas monoclonal anti-PPAT antibody TA504769 was diluted in 5% milk and incubated overnight at 4°C per the manufacturer's recommendations. For secondary staining, horseradish peroxidase–linked anti-rabbit IgG (Cell Signaling Technology, 7074S) or anti-mouse IgG (7076S) was diluted in 5% nonfat milk at a dilution of 1:2,000 and incubated at room temperature for 1 hour. Blots were imaged by chemiluminescence after incubation with Clarity Western ECL substrate (Bio-Rad, 1705061) using a ChemiDoc Imaging system (Bio-Rad).

Isotope labeling studies

For isotope labeling experiments, DMEM without glucose or glutamine was prepared from powder (MilliporeSigma, D5030) and supplemented with 13C6-glucose (Cambridge Isotope Laboratories, CLM-1396) and unlabeled glutamine (MilliporeSigma, G8540). Labeled media were prepared with 10% dialyzed FBS (Sigma-Aldrich, F0392). Cells were then seeded and cultured as described above. Fresh cell culture media without nucleoside supplementation were added to cells for 1 hour prior to switching to isotope-containing media. Prior to metabolite extraction, media were switched to isotope-containing media, and samples were collected at T = 240 minutes. Metabolite extraction and analysis were performed as above. Labeling data were corrected for natural isotope abundance using IsoCor (40).

In vivo tumor studies

To generate tumors, monoclonal KO cell lines were injected in 50 μL of a 1:1 mixture of DMEM:Matrigel (Corning, 354262) at 500,000 cells/50 μL into the fourth mammary fat pad of syngeneic 6- to 8-week-old FVB mice. The resulting tumors grew to a size of 15 mm as measured by external calipers along the longest axis, at which time the tumors were harvested and fragmented into 3 mm pieces that were cryopreserved in a mixture of 90% FBS and 10% DMSO. Cryopreserved tumors were then thawed, washed in saline, and cut into 1–2 mm fragments for implantation into the fourth mammary fat pad of recipient mice. These reimplanted tumors were then measured by external caliper 3 times weekly starting at 7 days after implantation until the experimental endpoint at 24 days after implantation. Tumor size was calculated as cross-sectional area using measurements from the longest and shortest axes. Mice were monitored for humane endpoints throughout the experiment according to institutional guidelines. At 24 days, the tumors were collected, and a cross-section of each tumor was formalin fixed for histologic preparation.

Histologic analyses

All histologic preparation and IHC staining were performed by the Investigative HistoPathology Laboratory at Michigan State University. Ki67 staining was measured using multiple images taken from distinct, non-necrotic regions of each tumor and evaluated as follows. For each tumor, at least four color images from distinct regions were acquired using an Olympus BX41 microscope operated at 10x magnification and saved as TIFF image files. Image processing was performed in ImageJ 1.52p (Fiji distribution). The color images were first deconvoluted into H (hematoxylin) and DAB (diaminobenzidine) color channels using Color Deconvolution (“H DAB” deconvolution matrix). Deconvoluted H and DAB images were saved as new TIFF images. For each image, smoothing was applied 5 times, then Auto Local Threshold was performed using Bernsen's algorithm (window size 15, contrast threshold 15) to detect stained nuclei. Stained nuclei were counted using Analyze Particles (minimum size 150, minimum circularity 0.3). The above steps were looped over all images. To check that threshold parameters were appropriate, several output images were manually inspected to confirm that visually identifiable nuclei were properly counted. The percent Ki67+ nuclei was calculated as the ratio of DAB-stained nuclei counts (representing proliferating cells) to H-stained nuclei counts (representing all cells) for each image, and averaged across all images for each experimental group. Terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) assays were evaluated using a single image of the full tumor cross-section to determine the proportion of necrotic area to non-necrotic area of each tumor. Images were acquired using a Leica M165FC stereo microscope operated at 1x magnification and saved as TIFF image files. TUNEL assay images were also processed using ImageJ. Images were duplicated, and color thresholding was used to select either the TUNEL+ area (image 1) or the entire tumor area (image 2). The percent TUNEL+ area was calculated as the ratio of image 1 area to image 2 area for each tumor and averaged across all tumors within each experimental group.

Statistical analyses

Statistical analyses were performed using unpaired Student t test except where otherwise noted. P values were adjusted in R using the p.adjust() function to account for multiple hypothesis testing using the Benjamini–Hochberg procedure (metabolites) or Hommel procedure (tumor measurements). All error bars presented are SD. All figures except survival curves and heatmaps were generated using GraphPad Prism.

Metabolite pool sizes and gene expression patterns of MMTV-Myc mouse mammary tumors implicate differences in nucleotide metabolic pathway activity between subtypes

To identify differences in metabolic pathway activities between EMT and papillary mouse mammary tumor subtypes, we integrated a metabolomics analysis with publicly available gene expression data (28). Metabolites were extracted from flash-frozen tumor sections of known histologic subtype (Fig. 1A) and quantitated using LC-MS/MS. We found metabolites involved in the pentose phosphate pathway (PPP) and metabolites related to nucleotide metabolism to be significantly different between EMT and papillary tumors (Fig. 1B; Supplementary Table S1). Notably, PPP intermediates including gluconolactone, ribose 5-phosphate, ribulose 5-phosphate, and sedoheptulose phosphate are uniformly elevated in the EMT subtype compared with papillary. The PPP serves several important functions including: (i) production of ribose 5-phosphate, which can be used for nucleotide biosynthesis or converted to glycolytic intermediates; (ii) production of reducing equivalents in the form of NADPH; and (iii) generation of erythrose 4-phosphate, which can also be converted to glycolytic intermediates (41). Several metabolites related to nucleotide metabolism are also different between EMT and papillary tumors (Fig. 1B; Supplementary Table S1). For example, EMT tumors have higher levels of inosine monophosphate (IMP), adenine, and inosine compared with papillary tumors. Adenine and inosine are both intermediates in breakdown and salvage pathways of nucleotide metabolism, and IMP is an intermediate for purine biosynthesis. To investigate how these metabolite levels reflect differences in gene expression, we downloaded gene expression data for the EMT and papillary tumors from GEO (42) and applied GSEA (34) using metabolism-related gene sets from the Reactome database (35). This analysis revealed that genes involved in the PPP (Fig. 1C) and nucleobase biosynthesis (Fig. 1C) are both significantly enriched for lower expression in the EMT subtype compared with the papillary subtype. Therefore, the higher levels of PPP metabolites and IMP observed in EMT tumors (Fig. 1B) likely reflect accumulation due to decreased flux through the PPP and nucleobase biosynthesis pathways. Together, these data agree with our previous in vitro findings, where we observed lower nucleotide biosynthesis in EMT cells compared with papillary cells (26). When considered with our previous results, these findings further demonstrate that EMT and papillary tumors exhibit significant differences in nucleotide metabolism in vivo.

Figure 1.

Metabolic profiles and gene expression patterns indicate differences in nucleotide metabolism between MMTV-Myc EMT and papillary tumor subtypes. A, Representative histology images of the EMT and papillary tumor subtypes. B, Heatmap indicating relative metabolite differences between EMT and papillary tumors. Yellow and blue boxes indicate increased or decreased metabolite levels relative to the average of the papillary subtype, respectively. Metabolites with statistically significant differences (P < 0.05) are bolded and marked with asterisks (*). Statistical comparisons are listed in Supplementary Table S1. C, GSEA for PPP genes is significantly enriched (P = 0.014; FDR q = 0.16) for low expression in EMT tumors vs. papillary tumors. D, GSEA for genes involved in nucleobase biosynthesis are significantly enriched (P = 0.039; FDR q = 0.18) for low expression in EMT tumors vs. papillary tumors.

Figure 1.

Metabolic profiles and gene expression patterns indicate differences in nucleotide metabolism between MMTV-Myc EMT and papillary tumor subtypes. A, Representative histology images of the EMT and papillary tumor subtypes. B, Heatmap indicating relative metabolite differences between EMT and papillary tumors. Yellow and blue boxes indicate increased or decreased metabolite levels relative to the average of the papillary subtype, respectively. Metabolites with statistically significant differences (P < 0.05) are bolded and marked with asterisks (*). Statistical comparisons are listed in Supplementary Table S1. C, GSEA for PPP genes is significantly enriched (P = 0.014; FDR q = 0.16) for low expression in EMT tumors vs. papillary tumors. D, GSEA for genes involved in nucleobase biosynthesis are significantly enriched (P = 0.039; FDR q = 0.18) for low expression in EMT tumors vs. papillary tumors.

Close modal

Expression of nucleotide salvage genes is increased in the EMT subtype

To further characterize gene expression differences in nucleotide metabolism between the EMT and papillary tumors, we used the TAC software. We filtered the gene list to include nucleotide metabolism and the PPP genes as denoted within the Reactome database. Based on our GSEA results (Fig. 1C and D), we expected genes involved in de novo nucleotide biosynthesis and PPP to have higher expression in the papillary subtype. Indeed, many genes in the PPP including PGLS, PGD, RPE, RPIA, TKT, and TALDO1 have significantly higher expression in the papillary subtype (Supplementary Fig. S1). For the EMT subtype, we find that the gene with highest relative expression is UPP1, with 18-fold higher expression in the EMT subtype versus the papillary subtype (Supplementary Table S2). UPP1 encodes uridine phosphorylase 1, an enzyme involved in pyrimidine salvage. Together with the observation of higher nucleotide salvage pathway intermediates adenine and inosine in EMT (Fig. 1B), this suggests the EMT subtype has higher activity of the nucleotide salvage pathway. The cytoskeletal protein vimentin, a mesenchymal marker with increased expression in the EMT subtype (Supplementary Fig. S2A), has been shown to associate with UPP1 and form enzymatically functional complexes (43). In addition, papillary tumors have significantly higher MYC expression than EMT tumors (Supplementary Fig. S2B). MYC regulates expression of numerous genes in nucleotide metabolism, including de novo biosynthesis genes (20). Therefore, we decided to focus our analysis on genes involved in de novo nucleotide biosynthesis and nucleotide salvage pathways. Hierarchical clustering revealed two major groupings of genes as illustrated by the dendrogram in Fig. 2A. The first, smaller group included many nucleotide salvage genes with significantly higher expression in EMT tumors and the second, larger group predominately contained de novo biosynthesis genes with significantly lower expression in EMT compared with papillary tumors. Therefore, EMT tumors show a relative preference for nucleotide salvage, whereas papillary tumors prefer de novo biosynthesis. We also considered GSEA results for the nucleotide salvage pathway, however, this gene set as a whole was not significantly enriched for the EMT subtype despite several genes within this pathway having significantly higher expression in the EMT subtype (Supplementary Fig. S3; Supplementary Table S2). A simplified pathway overview summarizing our results highlights the potential metabolic preferences for nucleotide biosynthesis that are specific to each subtype, with the papillary subtype preferring de novo biosynthesis and the EMT subtype preferring nucleotide salvage (Fig. 2B).

Figure 2.

Expression of nucleotide salvage genes is higher in the EMT subtype, and expression of de novo biosynthesis genes is higher in the papillary subtype. A, Heatmap depicting expression of genes related to nucleotide metabolism. Genes are sorted by hierarchical clustering and color-coded by relationship to nucleotide metabolism pathways. Row labels are the sample IDs from the original dataset. Data are normalized to the average expression of the papillary subtype for each gene. Genes with statistically significant differences (FDR P < 0.05) are marked with asterisks (*). Statistical comparisons are listed in Supplementary Table S2. B, Summary of nucleotide biosynthesis pathway. Metabolic intermediates and genes are marked according to subtype-specific relationships.

Figure 2.

Expression of nucleotide salvage genes is higher in the EMT subtype, and expression of de novo biosynthesis genes is higher in the papillary subtype. A, Heatmap depicting expression of genes related to nucleotide metabolism. Genes are sorted by hierarchical clustering and color-coded by relationship to nucleotide metabolism pathways. Row labels are the sample IDs from the original dataset. Data are normalized to the average expression of the papillary subtype for each gene. Genes with statistically significant differences (FDR P < 0.05) are marked with asterisks (*). Statistical comparisons are listed in Supplementary Table S2. B, Summary of nucleotide biosynthesis pathway. Metabolic intermediates and genes are marked according to subtype-specific relationships.

Close modal

Expressions of key de novo and salvage genes are correlated with worse patient outcomes

Our results thus far illustrate the possibility for distinct histologic subtypes to utilize different pathways to meet the same metabolic demand for nucleotides. We next sought to determine the potential clinical relevance associated with expression of these genes in human breast cancer. We focused on genes phosphoribosyl pyrophosphate amidotransferase (PPAT) and adenine phosphoribosyltransferase (APRT) because they encode the rate-limiting step for de novo purine biosynthesis and salvage of the purine base adenine, respectively, and our findings show that PPAT expression is significantly higher in the papillary subtype and APRT expression is significantly higher in the EMT subtype (Fig. 2B; Supplementary Table S2). In addition, APRT protein levels are higher in EMT tumors; we were unable to obtain reliable results for PPAT protein levels due to the lack of suitable antibodies (Supplementary Fig. S4). We used KM plotter, which generates KM curves using patient data mined from GEO datasets (37), to generate survival curves with patients stratified by relative gene expression of PPAT and APRT. We find that in general, patients with high expression of both PPAT and APRT have worse RFS than patients with low expression of these genes (Fig. 3A). This trend is also observed when patients are further divided according to intrinsic subtype. High expression of both PPAT and APRT is similarly significant for patients with luminal A (Fig. 3B) and luminal B (Fig. 3C) breast cancer. However, for patients with HER2+ (Fig. 3D) and basal (Fig. 3E) breast cancer, high expression of PPAT is no longer associated with decreased RFS, whereas high expression of APRT remains significant. These trends are also generally maintained when OS of patients is considered (Supplementary Fig. S5A–S5E), with the exceptions being that high PPAT expression and high APRT expression are no longer associated with worse OS in luminal B and HER2+ patients, respectively. These results highlight the potential importance of nucleotide metabolism in breast cancer and further suggest that de novo purine biosynthesis may be most important for luminal breast cancer, whereas salvage may be relevant for all breast cancer subtypes.

Figure 3.

Expressions of de novo nucleotide biosynthesis gene PPAT and nucleotide salvage gene APRT are strongly associated with RFS across breast cancer subtypes. Kaplan–Meier survival curves for all patients with breast cancer (A) and specific breast cancer subtypes (B–E). Statistically significant relationships (P value < 0.05) are bolded and marked with asterisks (*).

Figure 3.

Expressions of de novo nucleotide biosynthesis gene PPAT and nucleotide salvage gene APRT are strongly associated with RFS across breast cancer subtypes. Kaplan–Meier survival curves for all patients with breast cancer (A) and specific breast cancer subtypes (B–E). Statistically significant relationships (P value < 0.05) are bolded and marked with asterisks (*).

Close modal

Knocking out de novo and salvage genes disrupts cell metabolism in a subtype-specific manner

To further investigate the importance of nucleotide biosynthesis genes PPAT and APRT in our model, we targeted each gene using CRISPR/Cas9 gene editing (27) in EMT and papillary tumor-derived cell lines. We concurrently generated puromycin-resistant control cell lines for each subtype with a nontargeting scramble guide RNA. Clonal lines for each subtype, KO, and puromycin-resistant scramble control (PSC) were isolated by serial dilution, and successful gene editing was confirmed by TIDE analysis (Supplementary Fig. S6; ref. 39). Western blots were also performed to determine successful KO by protein expression (Supplementary Fig. S7A). Although the APRT antibody worked well, the PPAT bands were inconclusive, with multiple faint bands near the predicted molecular weight of PPAT. We therefore also performed isotope labeling studies to functionally assess how 13C-glucose is incorporated into purine biosynthesis in these cell lines. We reasoned that the M-5 isotopolog of ATP represents production from either pathway, because the M-5 isotopolog of ATP is predominately derived from a fully labeled PRPP molecule with a fully unlabeled adenine nucleobase, and both de novo and salvage pathways utilize PRPP as a substrate. In contrast, the M1–4 and M6–10 isotopologs require labeling of the adenine base, which is only attained through de novo biosynthesis. Hence, we distinguish ATP isotopologs as unlabeled (M-0), ATP that may be derived from either de novo or salvage pathways (M-5), and ATP that could only be derived from de novo biosynthesis (M1–4 and M6–10). Using this approach, we find that, compared with controls, knocking out the salvage gene APRT resulted in increased labeling of isotopologs of ATP that can only be derived from de novo biosynthesis in both subtypes (Fig. 4; Supplementary Table S3), which is expected because a larger proportion of ATP is now derived from de novo biosynthesis instead of salvage. In addition, targeting the de novo biosynthesis gene PPAT caused significantly decreased labeling of ATP isotopologs derived from de novo biosynthesis (Fig. 4; Supplementary Table S3). However, a small amount of labeling into de novo biosynthesis-specific ATP isotopologs is observed in the PPAT KO cell lines for each subtype, implying that a small degree of de novo purine biosynthesis may still occur in these lines. Nevertheless, these results provide strong evidence that the metabolic activity of nucleotide salvage and de novo biosynthesis have been significantly decreased in the APRT and PPAT KO cell lines, respectively.

Figure 4.

13C-Isotope incorporation from glucose into ATP biosynthesis is altered after targeting de novo and salvage genes. Percent 13C-isotope labeling in ATP is shown for PSC cells, de novo nucleotide biosynthesis gene PPAT knockout cells (PPAT KO), and nucleotide salvage gene APRT knockout cells (APRT KO). White boxes represent the unlabeled (M-0 isotopolog) proportion of ATP. Light gray boxes represent the M-5 isotopolog, which can be derived from either de novo or salvage pathways. Dark gray boxes represent the sum of all other isotopologs of ATP (M1–4 and M6–10), which are derived from de novo ATP biosynthesis. Data are displayed as mean ± SD, N = 3. *, P < 0.05. Statistical comparisons are listed in Supplementary Table S3.

Figure 4.

13C-Isotope incorporation from glucose into ATP biosynthesis is altered after targeting de novo and salvage genes. Percent 13C-isotope labeling in ATP is shown for PSC cells, de novo nucleotide biosynthesis gene PPAT knockout cells (PPAT KO), and nucleotide salvage gene APRT knockout cells (APRT KO). White boxes represent the unlabeled (M-0 isotopolog) proportion of ATP. Light gray boxes represent the M-5 isotopolog, which can be derived from either de novo or salvage pathways. Dark gray boxes represent the sum of all other isotopologs of ATP (M1–4 and M6–10), which are derived from de novo ATP biosynthesis. Data are displayed as mean ± SD, N = 3. *, P < 0.05. Statistical comparisons are listed in Supplementary Table S3.

Close modal

We also analyzed the abundance of a wide range of metabolites in CRISPR-edited cell lines relative to the control line of each respective subtype using the targeted LC-MS/MS method described above. This comparison makes it possible to determine whether metabolic alterations occur in EMT or papillary cells when the preferred or nonpreferred pathway for nucleotide metabolism of each subtype is targeted. These changes within each subtype can then be compared with the metabolic alterations in the other subtype to determine whether the changes are subtype-specific based on metabolic pathway preference. We found significant differences between cell lines across several metabolic pathways (Supplementary Fig. S8; Supplementary Table S4). As expected, the most consistently altered metabolites include PPP-related metabolites (Fig. 5A), nucleoside triphosphates (NTP; Fig. 5B), and deoxynucleoside triphosphates (dNTP; Fig. 5C). Notably, the relative abundance of these metabolites is generally most different when the preferred metabolic pathway for each subtype has been targeted. For example, ATP (Fig. 5B) and dATP (Fig. 5C) levels are significantly decreased in the papillary PPAT KO cell line compared with the papillary control and APRT KO lines, whereas the EMT PPAT KO is similar to EMT control for these metabolites. In addition, the papillary PPAT KO cells have lower levels of most PPP intermediates, but significantly higher levels of phosphoribosyl pyrophosphate (PRPP; Fig. 5A), which is used by both de novo biosynthesis and salvage pathways and is produced from the PPP intermediate ribose-5-phosphate. Most NTPs and dNTPs are also decreased in the papillary PPAT KO line compared with the control or APRT KO line (Fig. 5B and C). The exception to this is the pyrimidine uridine triphosphate (UTP), which is significantly increased in the papillary PPAT KO compared with control and APRT KO lines (Fig. 5B). These metabolic effects may be explained by principles of metabolic regulation. PRPP elevations generally feed forward to stimulate synthesis of both purines and pyrimidines. Because de novo purine biosynthesis is blocked, pyrimidine biosynthesis is instead increased until UTP levels rise high enough to feedback inhibit pyrimidine biosynthesis. Purine nucleotides, such as AMP, also feedback inhibit PRPS (20). Because the papillary PPAT KO cells have lower levels of AMP (Supplementary Table S4), this feedback inhibition is removed, enabling increased flux through the PPP pathway to generate PRPP with subsequent depletion of PPP intermediates.

Figure 5.

Metabolite levels are most affected by targeting the preferred nucleotide biosynthetic pathway for each subtype. The abundance of metabolites related to the PPP (A) and nucleotides (B–C) is most altered within each subtype when nucleotide salvage gene APRT is knocked out in EMT (left half of each graph) and when de novo biosynthesis gene PPAT is knocked out in papillary (right half of each graph). Data are displayed relative to the PSC for each subtype and represent mean ± SD, N = 3. *, P < 0.05. Statistical comparisons are listed in Supplementary Table S4.

Figure 5.

Metabolite levels are most affected by targeting the preferred nucleotide biosynthetic pathway for each subtype. The abundance of metabolites related to the PPP (A) and nucleotides (B–C) is most altered within each subtype when nucleotide salvage gene APRT is knocked out in EMT (left half of each graph) and when de novo biosynthesis gene PPAT is knocked out in papillary (right half of each graph). Data are displayed relative to the PSC for each subtype and represent mean ± SD, N = 3. *, P < 0.05. Statistical comparisons are listed in Supplementary Table S4.

Close modal

Decreased PPP intermediates, increased PRPP, and alterations in NTP and dNTP levels are not observed in the EMT PPAT KO cells, likely due to the metabolic preference of the EMT subtype to salvage nucleotides. Indeed, targeting the salvage pathway caused significant metabolic alterations in the EMT APRT KO cells compared with the control: higher levels of most nucleotides (Fig. 5B and C) suggest that EMT APRT KO cells are forced to switch to de novo biosynthesis when their preferred means of obtaining nucleotides via salvage is inhibited. In the papillary APRT KO line, nucleotide levels are not significantly changed from control levels (Fig. 5B and C). Taken together, the above results indicate that the greatest impact on nucleotide metabolism is achieved when the preferred nucleotide biosynthesis pathway of each subtype is inhibited, whereas inhibiting the nonpreferred pathway has minimal effects.

Targeting nucleotide de novo biosynthesis and salvage genes affects tumor growth in a subtype-specific manner

To determine the in vivo effects of targeting the preferred nucleotide biosynthesis pathway for each subtype, we monitored tumor growth of KO and control cell lines injected in mice. Control or KO cells were first injected into the mammary fat pad of syngeneic mice to generate tumors, then the resulting tumors were resected, and fragments of these tumors were orthotopically implanted into new cohorts of mice to monitor tumor growth over time. This was performed because implantation of tumor fragments, rather than direct injection of tumor cells, resulted in less variability in the lag time of tumor growth. As expected, the EMT tumors grew slowest when the preferred nucleotide salvage pathway gene APRT was targeted: EMT APRT KO tumors were significantly smaller (762.8 ± 108.4 mm2, N = 5) at 24 days after implantation as compared with the PPAT KO tumors (982.7 ± 116.1 mm2, N = 5). The EMT PPAT KO tumors also grew slower than the PSC tumors (1344.6 ± 141.7 mm2, N = 6), which were the largest at 24 days after implantation (Fig. 6A; Supplementary Table S5). Consistent with the reliance of the papillary subtype on de novo nucleotide biosynthesis, targeting PPAT prevented papillary cells from growing tumors in vivo (Fig. 6B; Supplementary Table S5). Surprisingly, targeting the nonpreferred nucleotide salvage gene APRT caused papillary tumors to grow larger (1161.8 ± 155.8 mm2, N = 5) than the PSC tumors (514.0 ± 114.0 mm2, N = 5) at 24 days after implantation. Taken together, these results indicate that de novo nucleotide biosynthesis is a critical metabolic pathway for papillary tumors, and further demonstrate targeting a nonpreferred metabolic pathway could have the unintended side effect of increasing tumor growth.

Figure 6.

Tumor growth for each subtype is decreased after knocking out the preferred nucleotide metabolism pathway. In vivo growth curves for EMT (A) and papillary (B) tumors. Slowest growth for EMT is observed when the preferred nucleotide salvage pathway gene APRT is targeted (APRT KO). The EMT PPAT KO tumors also grew slower than the PSC tumors. Consistent with the reliance of the papillary subtype on de novo nucleotide biosynthesis, targeting PPAT prevented papillary cells from growing tumors in vivo (PPAT KO). Surprisingly, targeting the nonpreferred nucleotide salvage gene APRT caused papillary tumors to grow larger than the PSC tumors. Data are displayed as mean ± SD. *, P < 0.05. Statistical comparisons are listed in Supplementary Table S5.

Figure 6.

Tumor growth for each subtype is decreased after knocking out the preferred nucleotide metabolism pathway. In vivo growth curves for EMT (A) and papillary (B) tumors. Slowest growth for EMT is observed when the preferred nucleotide salvage pathway gene APRT is targeted (APRT KO). The EMT PPAT KO tumors also grew slower than the PSC tumors. Consistent with the reliance of the papillary subtype on de novo nucleotide biosynthesis, targeting PPAT prevented papillary cells from growing tumors in vivo (PPAT KO). Surprisingly, targeting the nonpreferred nucleotide salvage gene APRT caused papillary tumors to grow larger than the PSC tumors. Data are displayed as mean ± SD. *, P < 0.05. Statistical comparisons are listed in Supplementary Table S5.

Close modal

To determine whether differences in tumor sizes are attributable to changes in proliferation or cell death, IHC analysis was performed. Ki67 staining and TUNEL assays were performed to measure proliferation and necrosis within the tumors, respectively. As expected, Ki67 staining is directly proportional to tumor growth in each subtype. The EMT PSC tumors have significantly more Ki67+ nuclei than both KOs, and the EMT APRT KO tumors have significantly fewer compared with PPAT KOs (Fig. 7A; Supplementary Table S6). This indicates that the APRT KO EMT tumors grow slower due to decreased proliferation. In papillary tumors, the APRT KOs have significantly more Ki67+ nuclei than the PSC tumors (Fig. 7B; Supplementary Table S6), showing these tumors grow more quickly due to increased proliferation. TUNEL assays show that in the EMT subtype, both KOs were significantly more necrotic than the control tumors (Fig. 7C; Supplementary Table S7). In the papillary subtype, no difference in staining was observed between control and APRT KO tumors (Fig. 7D; Supplementary Table S7), indicating that the observed differences in tumor growth are not due to differences in tumor necrosis.

Figure 7.

IHC analysis reveals decreased proliferation in slower growing tumors. A and B, IHC analysis for Ki67 staining in EMT (A) and papillary (B) tumors. EMT PSC tumors have significantly more Ki67+ nuclei than both KOs, and the EMT APRT KO tumors have significantly fewer compared with PPAT KOs, indicating the APRT KO EMT tumors grow slower due to decreased proliferation. In papillary tumors, the APRT KOs have significantly more Ki67+ nuclei than the PSC tumors, showing these tumors grow more quickly due to increased proliferation. C and D, TUNEL assays for EMT (C) and papillary (D) tumors show that in the EMT subtype, both KOs were significantly more necrotic than the PSC tumors. In the papillary subtype, no difference in staining was observed between control and APRT KO tumors, indicating that the observed differences in tumor growth are not due to differences in tumor necrosis. Data are displayed as mean ± SD. *, P < 0.05. Statistical comparisons are listed in Supplementary Tables S6 and S7.

Figure 7.

IHC analysis reveals decreased proliferation in slower growing tumors. A and B, IHC analysis for Ki67 staining in EMT (A) and papillary (B) tumors. EMT PSC tumors have significantly more Ki67+ nuclei than both KOs, and the EMT APRT KO tumors have significantly fewer compared with PPAT KOs, indicating the APRT KO EMT tumors grow slower due to decreased proliferation. In papillary tumors, the APRT KOs have significantly more Ki67+ nuclei than the PSC tumors, showing these tumors grow more quickly due to increased proliferation. C and D, TUNEL assays for EMT (C) and papillary (D) tumors show that in the EMT subtype, both KOs were significantly more necrotic than the PSC tumors. In the papillary subtype, no difference in staining was observed between control and APRT KO tumors, indicating that the observed differences in tumor growth are not due to differences in tumor necrosis. Data are displayed as mean ± SD. *, P < 0.05. Statistical comparisons are listed in Supplementary Tables S6 and S7.

Close modal

To validate the monoclonal tumor growth findings, tumor growth of additional, distinct clonal lines for each subtype was measured (Supplementary Figs. S9 and S10; Supplementary Table S5). Four additional papillary PPAT KO clones were tested with ATP labeling comparable with Fig. 4, as well as one additional EMT PPAT KO clone (Supplementary Fig. S11; Supplementary Table S8). Two additional confirmed APRT KO clones for each subtype were also injected into mice for in vivo testing (Supplementary Fig. S7B). To validate the controls, the tumor growth of an additional PSC clone and wild-type tumors of each subtype were measured. The clonal PSC tumors for the EMT subtype grew similarly and are larger than the EMT wild-type tumors at 24 days after implantation, indicating the clonal selection process may select for more aggressive clones of this subtype. In addition, the two EMT PPAT KOs grew comparably and were also similar in size to the wild-type EMT tumors. However, the additional two EMT APRT KO clonal cell lines failed to generate tumors (Supplementary Fig. S9A; Supplementary Table S5). For the papillary subtype, one APRT KO clone again grew more quickly than the control tumors, whereas the remaining clone grew similarly to the PSC and wild-type papillary tumors (Supplementary Fig. S9B; Supplementary Table S5). Consistent with the results shown in Fig. 6, the four additional papillary PPAT KO clonal cell lines also failed to generate tumors. Taken together, our findings demonstrate the importance of targeting subtype-specific metabolic vulnerabilities to effectively control tumor growth. In addition, inhibiting a nonpreferred metabolic pathway not only fails to reduce tumor growth, but can have the detrimental effect of increasing tumor growth.

In this study, we used a combination of genomic and metabolomic techniques to identify subtype-specific metabolic preferences in nucleotide metabolism in the EMT and papillary tumor subtypes derived from the MMTV-Myc mouse model. We discovered that the EMT subtype prefers nucleotide salvage, whereas the papillary subtype relies on de novo nucleotide biosynthesis. We also investigated patient outcomes and identified that high expression of the nucleotide salvage gene APRT is correlated with worse RFS across breast cancer subtypes, whereas high expression of the de novo biosynthesis gene PPAT is associated with worse outcomes in patients with luminal breast cancer. We further characterized the metabolic effects of targeting both the preferred and nonpreferred pathways in the EMT and papillary subtypes and demonstrate the effect of knocking out these pathways on the in vivo tumor growth of each subtype.

Our results demonstrate that targeting the preferred metabolic pathway for nucleotide biosynthesis reduces tumor growth in both EMT and papillary tumors. An alternate interpretation of our results is that dependence on different nucleotide pathways could be related to the basal growth rate of the tumors. However, our findings do not support this possibility. The EMT subtype grows faster and prefers nucleotide salvage; however, causing the papillary subtype to rely on nucleotide salvage by targeting de novo biosynthesis (PPAT) does not cause it to grow faster. Similarly, targeting salvage does not universally cause tumors to grow slower: in fact, targeting salvage (APRT) caused 2 out of 3 independent papillary tumors to grow faster (Supplementary Fig. S9B).

Sustained proliferation is a hallmark of cancer (11), and to achieve this, cancer cells have a high requirement for nucleotide biosynthesis. Indeed, targeting nucleotide metabolism has long been used as a staple of cancer therapy with early examples including the folate analog methotrexate (MTX) and the pyrimidine analog 5-fluorouracil (5FU). MTX inhibits dihydrofolate reductase and blocks one-carbon metabolism that is essential for several de novo biosynthetic reactions (44), and 5FU inhibits thymidylate synthase, which catalyzes the de novo production of thymidine monophosphate (45). Other compounds targeting de novo nucleotide biosynthesis including 6-mercaptopurine (46), leflunomide (47), and brequinar (48) are also currently approved or under investigation as cancer therapeutics. Notably for our model, an active metabolite of 6-mercaptopurine inhibits PPAT (46) and could prove effective at inhibiting growth of the papillary subtype. However, 6-mercaptopurine and several other de novo nucleotide metabolism–targeting compounds including 5FU and gemcitabine are activated by the nucleotide salvage pathway, and downregulation of this pathway could provide a potential resistance mechanism to these compounds (49, 50).

In our model, the papillary subtype has increased MYC signaling compared with the EMT subtype (28), and its metabolic preference for de novo nucleotide biosynthesis highlights the role of MYC as a master regulator of nucleotide biosynthesis (51, 52). MYC amplification is a common feature of many human cancers (53) and occurs in 15.7% of breast cancers (54). In TNBC specifically, it has been shown that chemotherapy with doxorubicin adaptively upregulates de novo pyrimidine biosynthesis and cotreatment of TNBC xenografts with doxorubicin, and the de novo pyrimidine biosynthesis inhibitor leflunomide is more effective at treating TNBC tumors than doxorubicin alone (55). Upregulated de novo purine biosynthesis, directed by MYC signaling, has also been implicated as a key metabolic pathway in glioblastoma, and targeting de novo purine biosynthesis genes improved survival and reduced tumor burden in an in vivo model of glioblastoma (56). These studies and ours strongly suggest that further development of compounds targeting de novo nucleotide biosynthesis will be useful to treat many types of cancer.

Our results show that targeting nucleotide salvage also attenuates tumor growth in cancers that prefer this pathway, such as the EMT subtype MMTV-Myc tumors (Fig. 6A). The EMT subtype has previously been correlated with the claudin-low subtype of human breast cancer based on gene expression patterns (22, 23), and additional studies should be performed to determine whether nucleotide salvage is also a metabolic vulnerability in claudin-low breast cancer. EMT has also been associated with metastasis (57), however, we did not observe a metastatic phenotype in this study. The MMTV-Myc model is not highly metastatic, as lung metastasis only occurs in 25% of MMTV-Myc tumors. Although the EMT subtype does exhibit gene expression patterns associated with increased metastatic potential, the metastatic lesions do not display an EMT histology (28). Therefore, our model is not well suited to study the potential effects of targeting nucleotide metabolism on metastasis. In addition, manipulation of the preferred metabolic pathways did not cause interconversion of histologic subtypes.

Nucleotides and related metabolites are abundant in the extracellular space and serve important biological functions: purines play a significant role as signaling molecules (58), and pyrimidine release by tumor-associated macrophages has been shown to mediate gemcitabine resistance in animal models of pancreatic cancer (59). Therefore, the uptake and utilization of these metabolites should be further investigated as therapeutic targets. Unfortunately, there are currently very few available drugs that target nucleotide salvage. Two indirect examples of salvage inhibitors are dilazep and dipyridamole. These compounds act through inhibition of equilibrative nucleoside transporters (ENT) and function as vasodilators, prevent platelet aggregation, and are currently approved to treat cardiovascular disease (60). ENT inhibition indirectly blocks nucleotide salvage pathways by preventing uptake of nucleosides and nucleobases. ENTs also mediate the uptake of nucleoside analogs like gemcitabine (61), which means ENT inhibition as a means to block nucleotide salvage would not be compatible for combination therapy with these drugs. Our findings support the development of therapeutic compounds to specifically target nucleotide salvage pathways. This could prove particularly beneficial for patients diagnosed with claudin-low breast cancer, which carries a poor prognosis and does not have targeted therapies (24).

One limitation of our study is that we are only able to assess the metabolic effects of targeting these pathways in vitro. This is important to consider, as metabolic features identified in vitro do not always translate in vivo (62). However, it is impossible to assess metabolic effects of targeting PPAT in the papillary subtype tumors in vivo, as these lines do not generate tumors, and we are therefore limited to considering these metabolic effects in vitro. This shortcoming does not detract from our ultimate finding that tumor growth of each subtype is most decreased by targeting their preferred pathway for nucleotide metabolism in vivo, consistent with our previous work detailing 13C-glucose and 13C-glutamine labeling patterns in purine and pyrimidine nucleotides for EMT and papillary subtypes (26). Our results further reveal the concerning possibility that targeting a nonpreferred pathway can cause an increase in tumor growth. Specifically, when APRT was targeted in the papillary subtype, two of three clones grew tumors surprisingly fast, whereas the remaining clone grew comparably with control tumors (Fig. 6; Supplementary Figs. S9 and S10). In our current study, we used tumors and cell lines from histologically pure samples, however, this is not always the case in spontaneous tumors. Specifically regarding the MMTV-Myc mouse model, spontaneous tumors develop with a wide variety of histologies, including mixed tumors composed of multiple subtypes in one region (28). If we consider a possible mixed tumor that is predominately EMT with a minor papillary component, our results indicate that treating it by inhibiting nucleotide salvage alone would likely be ineffective for the papillary component and could even have the unintended side effect of increasing the growth of the papillary portion of the tumor. One implication of this finding is that, for a mixed tumor exhibiting both EMT and papillary histologies, it may be safer to target de novo biosynthesis rather than the salvage pathway, because although EMT subtype cells prefer salvage, blocking de novo biosynthesis still has a small inhibitory effect on tumor growth (Fig. 6A); on the other hand, blocking the salvage pathway in papillary subtype cells can have the opposite and undesirable effect of increasing tumor growth (Fig. 6B).

In human breast cancer, intratumor heterogeneity can manifest in many ways, including on morphologic and genomic levels (63). The importance of this heterogeneity is particularly notable when considering biomarker expression; for example, current recommendations report a positive finding if at least 1% of tumors cells are positive for the ER (64). Because the degree of ER positivity is also directly correlated with patient outcomes following antiendocrine treatment (65), it is clear that the intratumor heterogeneity of this biomarker has important clinical implications. Based on our present findings, metabolic vulnerabilities can be used to design new treatments for breast cancer subtypes. However, the possibility of inadvertently stimulating tumor growth by improperly targeting metabolism should also be considered further, especially in recognition of the significant heterogeneity of breast cancer. Further work should be directed at determining whether subtypes of human breast cancer, which are known to exhibit different metabolic features (9), have differences in metabolic vulnerabilities, and whether targeting nonpreferred pathways is detrimental.

In conclusion, our findings demonstrate that distinct histologic subtypes of breast cancer exhibit different metabolic vulnerabilities in terms of their preferred nucleotide biosynthesis pathways, and that inhibiting the preferred pathway greatly affects metabolism as well as in vivo tumor growth. Crucially, we also show that targeting the nonpreferred pathway is not only less effective in controlling tumor growth but may have the opposite effect of increasing tumor growth. Our results underscore a critical need to elucidate the distinct metabolic preferences of different breast cancer subtypes in order to design effective targeted therapies for each subtype.

No disclosures were reported.

M.P. Ogrodzinski: Data curation, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. S.T. Teoh: Formal analysis, validation, investigation, visualization, methodology, writing-review and editing. S.Y. Lunt: Conceptualization, resources, supervision, funding acquisition, visualization, methodology, project administration, writing-review and editing.

The authors thank Deanna Broadwater, Elliot Ensink, and Hyllana Medeiros for helpful discussions and critical reading of this article. The authors thank Eran Andrechek for providing primary MMTV-Myc EMT and MMTV-Myc papillary tumors. The authors also thank the MSU Mass Spectrometry and Metabolomics Core and the MSU Investigative HistoPathology Laboratory.

This work was supported by the Office of the Assistant Secretary of Defense for Health Affairs, through the Breast Cancer Research Program, under Award No. W81XWH-15-1-0453 to S.Y. Lunt. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the Department of Defense. This work was also supported by the Spectrum Health MD/PhD Fellowship and the Aitch Foundation Graduate Fellowship to M.P. Ogrodzinski.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Torre
LA
,
Siegel
RL
,
Ward
EM
,
Jemal
A
. 
Global cancer incidence and mortality rates and trends—an update
.
Cancer Epidemiol Biomarkers Prev
2016
;
25
:
16
27
.
2.
Althuis
MD
,
Dozier
JM
,
Anderson
WF
,
Devesa
SS
,
Brinton
LA
. 
Global trends in breast cancer incidence and mortality 1973–1997
.
Int J Epidemiol
2005
;
34
:
405
12
.
3.
Senkus
E
,
Kyriakides
S
,
Ohno
S
,
Penault-Llorca
F
,
Poortmans
P
,
Rutgers
E
, et al
Primary breast cancer: ESMO clinical practice guidelines for diagnosis, treatment and follow-up
.
Ann Oncol
2015
;
26
:
v8
v30
.
4.
Perou
CM
,
Sorlie
T
,
Eisen
MB
,
van de Rijn
M
,
Jeffrey
SS
,
Rees
CA
, et al
Molecular portraits of human breast tumours
.
Nature
2000
;
406
:
747
52
.
5.
Bastien
RR
,
Rodríguez-Lescure
Á
,
Ebbert
MT
,
Prat
A
,
Munárriz
B
,
Rowe
L
, et al
PAM50 breast cancer subtyping by RT-qPCR and concordance with standard clinical molecular markers
.
BMC Med Genet
2012
;
5
:
44
.
6.
Coates
AS
,
Winer
EP
,
Goldhirsch
A
,
Gelber
RD
,
Gnant
M
,
Piccart-Gebhart
M
, et al
Tailoring therapies—improving the management of early breast cancer: St Gallen international expert consensus on the primary therapy of early breast cancer 2015
.
Ann Oncol
2015
;
26
:
1533
46
.
7.
Early Breast Cancer Trialists' Collaborative Group
. 
Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials
.
Lancet North Am Ed
2005
;
365
:
1687
717
.
8.
Gonzalez-Angulo
AM
,
Morales-Vasquez
F
,
Hortobagyi
GN
. 
Overview of resistance to systemic therapy in patients with breast cancer
. In:
Breast Cancer Chemosensitivity
.
Yu
D
,
Hung
M-C
, editors.
New York
:
Springer
; 
2007
. p.
1
22
.
9.
Ogrodzinski
MP
,
Bernard
JJ
,
Lunt
SY
. 
Deciphering metabolic rewiring in breast cancer subtypes
.
Transl Res
2017
;
189
:
105
22
.
10.
Luengo
A
,
Gui
DY
,
Vander Heiden
MG
. 
Targeting metabolism for cancer therapy
.
Cell Chem Biol
2017
;
24
:
1161
80
.
11.
Hanahan
D
,
Weinberg
RA
. 
Hallmarks of cancer: the next generation
.
Cell
2011
;
144
:
646
74
.
12.
Warburg
O
. 
The metabolism of carcinoma cells
.
J Cancer Res
1925
;
9
:
148
63
.
13.
Freedland
SJ
,
Mavropoulos
J
,
Wang
A
,
Darshan
M
,
Demark‐Wahnefried
W
,
Aronson
WJ
, et al
Carbohydrate restriction, prostate cancer growth, and the insulin‐like growth factor axis
.
Prostate
2008
;
68
:
11
9
.
14.
Poff
A
,
Koutnik
AP
,
Egan
KM
,
Sahebjam
S
,
D'Agostino
D
,
Kumar
NB
.
Targeting the Warburg effect for cancer treatment: ketogenic diets for management of glioma
.
Semin Cancer Biol
2017
;
56
:
135
48
.
15.
Pelicano
H
,
Martin
D
,
Xu
R
,
Huang
P
. 
Glycolysis inhibition for anticancer treatment
.
Oncogene
2006
;
25
:
4633
46
.
16.
Sørlie
T
,
Perou
CM
,
Tibshirani
R
,
Aas
T
,
Geisler
S
,
Johnsen
H
, et al
Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications
.
Proc Natl Acad Sci U S A
2001
;
98
:
10869
74
.
17.
Ellis
I
,
Galea
M
,
Broughton
N
,
Locker
A
,
Blamey
R
,
Elston
C
. 
Pathological prognostic factors in breast cancer. II. Histological type. Relationship with survival in a large study with long‐term follow‐up
.
Histopathology
1992
;
20
:
479
89
.
18.
Kondo
M
,
Yamaoka
T
,
Honda
S
,
Miwa
Y
,
Katashima
R
,
Moritani
M
, et al
The rate of cell growth is regulated by purine biosynthesis via ATP production and G1 to S phase transition
.
J Biochem
2000
;
128
:
57
64
.
19.
Sigoillot
FD
,
Berkowski
JA
,
Sigoillot
SM
,
Kotsis
DH
,
Guy
HI
. 
Cell cycle-dependent regulation of pyrimidine biosynthesis
.
J Biol Chem
2003
;
278
:
3403
9
.
20.
Lane
AN
,
Fan
TW-M
. 
Regulation of mammalian nucleotide metabolism and biosynthesis
.
Nucleic Acids Res
2015
;
43
:
2466
85
.
21.
Stewart
TA
,
Pattengale
PK
,
Leder
P
. 
Spontaneous mammary adenocarcinomas in transgenic mice that carry and express MTV/myc fusion genes
.
Cell
1984
;
38
:
627
37
.
22.
Hollern
DP
,
Andrechek
ER
. 
A genomic analysis of mouse models of breast cancer reveals molecular features of mouse models and relationships to human breast cancer
.
Breast Cancer Res
2014
;
16
:
R59
.
23.
Hollern
DP
,
Swiatnicki
MR
,
Andrechek
ER
. 
Histological subtypes of mouse mammary tumors reveal conserved relationships to human cancers
.
PLos Genet
2018
;
14
:
e1007135
.
24.
Prat
A
,
Parker
JS
,
Karginova
O
,
Fan
C
,
Livasy
C
,
Herschkowitz
JI
, et al
Phenotypic and molecular characterization of the claudin-low intrinsic subtype of breast cancer
.
Breast Cancer Res
2010
;
12
:
R68
.
25.
Rakha
EA
,
Ellis
IO
. 
Triple-negative/basal-like breast cancer
.
Pathology
2009
;
41
:
40
7
.
26.
Ogrodzinski
MP
,
Teoh
ST
,
Lunt
SY
. 
Metabolomic profiling of mouse mammary tumor-derived cell lines reveals targeted therapy options for cancer subtypes
.
Cell Oncol
2020
;
1
11
.
27.
Cong
L
,
Ran
FA
,
Cox
D
,
Lin
S
,
Barretto
R
,
Habib
N
, et al
Multiplex genome engineering using CRISPR/Cas systems
.
Science
2013
;
339
:
819
23
.
28.
Andrechek
ER
,
Cardiff
RD
,
Chang
JT
,
Gatza
ML
,
Acharya
CR
,
Potti
A
, et al
Genetic heterogeneity of Myc-induced mammary tumors reflecting diverse phenotypes including metastatic potential
.
Proc Natl Acad Sci U S A
2009
;
106
:
16387
92
.
29.
Mitra
A
,
Mishra
L
,
Li
S
. 
Technologies for deriving primary tumor cells for use in personalized cancer therapy
.
Trends Biotechnol
2013
;
31
:
347
54
.
30.
Ogrodzinski
MP
,
Teoh
ST
,
Yu
L
,
Broadwater
D
,
Ensink
E
,
Lunt
SY
. 
Measuring the nutrient metabolism of adherent cells in culture
. In:
Methods in Molecular Biology
.
Fendt
SM
,
Lunt
S
, editors.
New York
:
Humana Press
; 
2019
. p.
37
52
.
31.
Melamud
E
,
Vastag
L
,
Rabinowitz
JD
. 
Metabolomic analysis and visualization engine for LC−MS data
.
Anal Chem
2010
;
82
:
9818
26
.
32.
De Hoon
MJ
,
Imoto
S
,
Nolan
J
,
Miyano
S
. 
Open source clustering software
.
Bioinformatics
2004
;
20
:
1453
4
.
33.
Saldanha
AJ
. 
Java Treeview—extensible visualization of microarray data
.
Bioinformatics
2004
;
20
:
3246
8
.
34.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
35.
Fabregat
A
,
Sidiropoulos
K
,
Garapati
P
,
Gillespie
M
,
Hausmann
K
,
Haw
R
, et al
The Reactome pathway knowledgebase
.
Nucleic Acids Res
2016
;
44
:
D481
D487
.
36.
Liberzon
A
,
Birger
C
,
Thorvaldsdóttir
H
,
Ghandi
M
,
Mesirov
JP
,
Tamayo
P
. 
The molecular signatures database hallmark gene set collection
.
Cell Syst
2015
;
1
:
417
25
.
37.
Györffy
B
,
Lanczky
A
,
Eklund
AC
,
Denkert
C
,
Budczies
J
,
Li
Q
, et al
An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients
.
Breast Cancer Res Treat
2010
;
123
:
725
31
.
38.
Ma
J
,
Köster
J
,
Qin
Q
,
Hu
S
,
Li
W
,
Chen
C
, et al
CRISPR-DO for genome-wide CRISPR design and optimization
.
Bioinformatics
2016
;
32
:
3336
8
.
39.
Brinkman
EK
,
Chen
T
,
Amendola
M
,
van Steensel
B
. 
Easy quantitative assessment of genome editing by sequence trace decomposition
.
Nucleic Acids Res
2014
;
42
:
e168
.
40.
Millard
P
,
Letisse
F
,
Sokol
S
,
Portais
JC
. 
IsoCor: correcting MS data in isotope labeling experiments
.
Bioinformatics
2012
;
28
:
1294
6
.
41.
Stincone
A
,
Prigione
A
,
Cramer
T
,
Wamelink
MM
,
Campbell
K
,
Cheung
E
, et al
The return of metabolism: biochemistry and physiology of the pentose phosphate pathway
.
Biol Rev
2015
;
90
:
927
63
.
42.
Barrett
T
,
Wilhite
SE
,
Ledoux
P
,
Evangelista
C
,
Kim
IF
,
Tomashevsky
M
, et al
NCBI GEO: archive for functional genomics data sets—update
.
Nucleic Acids Res
2012
;
41
:
D991
5
.
43.
Russell
RL
,
Cao
D
,
Zhang
D
,
Handschumacher
RE
,
Pizzorno
G
. 
Uridine phosphorylase association with vimentin. Intracellular distribution and localization
.
J Biol Chem
2001
;
276
:
13302
7
.
44.
Walling
J
. 
From methotrexate to pemetrexed and beyond. A review of the pharmacodynamic and clinical properties of antifolates
.
Invest New Drugs
2006
;
24
:
37
77
.
45.
Longley
DB
,
Harkin
DP
,
Johnston
PG
. 
5-Fluorouracil: mechanisms of action and clinical strategies
.
Nat Rev Cancer
2003
;
3
:
330
8
.
46.
Karran
P
,
Attard
N
. 
Thiopurines in current medical practice: molecular mechanisms and contributions to therapy-related cancer
.
Nat Rev Cancer
2008
;
8
:
24
36
.
47.
Fox
RI
,
Herrmann
ML
,
Frangou
CG
,
Wahl
GM
,
Morris
RE
,
Strand
V
, et al
Mechanism of action for leflunomide in rheumatoid arthritis
.
Clin Immunol
1999
;
93
:
198
208
.
48.
Xu
X
,
Williams
JW
,
Shen
J
,
Gong
H
,
Yin
D-P
,
Blinder
L
, et al
In vitro and in vivo mechanisms of action of the antiproliferative and immunosuppressive agent, brequinar sodium
.
J Immunol
1998
;
160
:
846
53
.
49.
Zhang
N
,
Yin
Y
,
Xu
S-J
,
Chen
W-S
. 
5-Fluorouracil: mechanisms of resistance and reversal strategies
.
Molecules
2008
;
13
:
1551
69
.
50.
Klanova
M
,
Lorkova
L
,
Vit
O
,
Maswabi
B
,
Molinsky
J
,
Pospisilova
J
, et al
Downregulation of deoxycytidine kinase in cytarabine-resistant mantle cell lymphoma cells confers cross-resistance to nucleoside analogs gemcitabine, fludarabine and cladribine, but not to other classes of anti-lymphoma agents
.
Mol Cancer
2014
;
13
:
159
.
51.
Liu
Y-C
,
Li
F
,
Handler
J
,
Huang
CRL
,
Xiang
Y
,
Neretti
N
, et al
Global regulation of nucleotide biosynthetic genes by c-Myc
.
PLoS One
2008
;
3
:
e2722
.
52.
Stine
ZE
,
Walton
ZE
,
Altman
BJ
,
Hsieh
AL
,
Dang
CV
. 
MYC, metabolism, and cancer
.
Cancer Discov
2015
;
5
:
1024
39
.
53.
Nesbit
CE
,
Tersak
JM
,
Prochownik
EV
. 
MYC oncogenes and human neoplastic disease
.
Oncogene
1999
;
18
:
3004
16
.
54.
Deming
SL
,
Nass
SJ
,
Dickson
RB
,
Trock
BJ
. 
C-myc amplification in breast cancer: a meta-analysis of its occurrence and prognostic relevance
.
Br J Cancer
2000
;
83
:
1688
95
.
55.
Brown
KK
,
Spinelli
JB
,
Asara
JM
,
Toker
A
. 
Adaptive reprogramming of de novo pyrimidine synthesis is a metabolic vulnerability in triple-negative breast cancer
.
Cancer Discov
2017
;
7
:
391
.
56.
Wang
X
,
Yang
K
,
Xie
Q
,
Wu
Q
,
Mack
SC
,
Shi
Y
, et al
Purine synthesis promotes maintenance of brain tumor initiating cells in glioma
.
Nat Neurosci
2017
;
20
:
661
.
57.
Pastushenko
I
,
Blanpain
C
. 
EMT transition states during tumor progression and metastasis
.
Trends Cell Biol
2019
;
29
:
212
26
.
58.
Di Virgilio
F
,
Adinolfi
E
. 
Extracellular purines, purinergic receptors and tumor growth
.
Oncogene
2017
;
36
:
293
303
.
59.
Halbrook
CJ
,
Pontious
C
,
Kovalenko
I
,
Lapienyte
L
,
Dreyer
S
,
Lee
H-J
, et al
Macrophage-released pyrimidines inhibit gemcitabine therapy in pancreatic cancer
.
Cell Metab
2019
;
29
:
1390
9
.
60.
Noji
T
,
Karasawa
A
,
Kusaka
H
. 
Adenosine uptake inhibitors
.
Eur J Pharmacol
2004
;
495
:
1
16
.
61.
Mackey
JR
,
Mani
RS
,
Selner
M
,
Mowles
D
,
Young
JD
,
Belt
JA
, et al
Functional nucleoside transporters are required for gemcitabine influx and manifestation of toxicity in cancer cell lines
.
Cancer Res
1998
;
58
:
4349
57
.
62.
Davidson
SM
,
Papagiannakopoulos
T
,
Olenchock
BA
,
Heyman
JE
,
Keibler
MA
,
Luengo
A
, et al
Environment impacts the metabolic dependencies of ras-driven non-small cell lung cancer
.
Cell Metab
2016
;
23
:
517
28
.
63.
Turashvili
G
,
Brogi
E
. 
Tumor heterogeneity in breast cancer
.
Front Med
2017
;
4
:
227
.
64.
Hammond
ME
,
Hayes
DF
,
Dowsett
M
,
Allred
DC
,
Hagerty
KL
,
Badve
S
, et al
American Society of Clinical Oncology/College of American Pathologists guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer
.
J Clin Oncol
2010
;
28
:
2784
95
.
65.
Morgan
DA
,
Refalo
NA
,
Cheung
KL
. 
Strength of ER-positivity in relation to survival in ER-positive breast cancer treated by adjuvant tamoxifen as sole systemic therapy
.
The Breast
2011
;
20
:
215
9
.