2.1 Source of biospecimens

As described in detail previously,¹⁰ the mouse tissues were provided by the Mouse Model Repository for Cancer Biomarker Discovery led by Christopher J. Kemp, PhD, at the Fred Hutchinson Cancer Research Center, sponsored by National Cancer Institute Mouse Proteomics Initiative (http://proteomics.cancer.gov/programs/completed/mouseinitiative10). Tissues were collected and cryopreserved from mice of Pten-KO genotype and their wild type (WT) litter-/cage-mates in 129 background from April 2005 to April 2007.

The frozen tissue specimens were shipped to Texas Tech University Health Sciences Center School of Pharmacy, Amarillo, on dry ice (received on August 2, 2011) and stored at −80°C until analyses. The litter-/cage-mate paired tissues allowed comparison of the genotypic differences with strict control of housing and environmental factors, breeding stock, and age variations among the Pten-KO and WT mice. For the Pten-KO and matched prostate specimens obtained from mice aged 20–22 weeks, we divided the frozen tissues into smaller portions and used the leftover potions from the mRNA microarray omics for the metabolomics. The extraction protocols for each omic were unique to its platform and had to be performed separately. As described in our previous paper,¹⁰ the proteomics analyses were carried out on a different batch of WT and KO matched pairs at the younger age of 12–15 weeks. Due to the frozen nature of tissues obtained from the National Cancer Institute (NCI) repository, histology was not available. However, the pathology was assumed to involve invasive adenocarcinoma based on published work of the originator laboratory of Hong Wu, PhD,⁸ with significant infiltration of myeloid-derived immune cells and inflammatory cells.⁹

By the same procurement procedure as the Pten-KO pairs, three pairs of frozen TRAMP tumor and WT prostate in litter/cage-mates of 28, 29, and 31 weeks of age were also obtained. The matched pairs of mice were sacrificed on respective dates when a palpable pelvic mass was detected in the transgenic sibling. Similarly, lack of histology information, the TRAMP tumors were presumed poorly differentiated neuroendocrine carcinomas (NECa) based on literature from groups of authoritative experimental/veterinary pathologists ¹¹ and our own.¹² Due to the frozen nature of tissues as described above, we were not able to obtain specific lobes for the WT prostate of either model. We presume the prostate tissues sampled in our metabolomics work were mixture of different lobes.

2.2 Tissue sample extract preparation

The frozen tissue samples were homogenized in 30 volumes (v/w) of precooled (−80°C) 80% methanol in water and stored at −80°C for 16 h. The homogenates were centrifuged at 14,000 g for 10 min in a refrigerated centrifuge at 4°C. The supernatants were saved at −80°C and analyzed by the following protocol.

2.3 Metabolomics

The protocol published by Yuan et al.¹³ was adapted to profile aqueous endogenous metabolites from the tissue homogenate supernatants prepared in 80% methanol/water. The platform used hydrophilic interaction liquid chromatography with positive/negative ion switching to monitor 289 Q1/Q3 transitions from a single LC-MS/MS acquisition with a 5 ms dwell time and a 1.55 s cycle time. A UHPLC system (Nexera, Shimadzu) was coupled to a hybrid triple quadruple linear ion trap mass spectrometer (QTRAP 5500, AB Sciex). Chromatographic separation was performed on an XBridge BEH Amide Column (hydrophilic interaction liquid chromatography, 100 × 4.6 × 3.5 µm, Waters). The column oven temperature was kept at 40°C.

To ensure sensitive detection within the linear dynamic range, each extract was analyzed by multiple UHPLC-MS/MS runs with the injection volume of 10 and 1 µl. The abundance of each endogenous metabolite varied over a wide range, some of them were not able to be detected when 1 µl extract was injected; whereas the signal of some metabolites was saturated when 10 µl extract was injected. For 1 µl volume injections, all were done in duplicate. For 10 µl volume injections, most was done in duplicates, some in triplicates. The identification of metabolites was performed as per the Yuan report.¹³

2.4 Data curation and statistical metrics

The average/mean of the replicate measurements (peak area) was calculated and used for computation of the paired ratio (indicative of relative abundance) of each LC-MS/MS verified metabolite between Pten-KO or TRAMP prostate tumor and respective WT litter-/cage-mate. One-tailed, paired t test was administered on the peak areas in matched pairs (null hypothesis difference of tumor and WT pairs = 0) or on log-transformed peak area (null hypothesis ratio = 1 or logTumor-logWT = 0). Given the small number of pairs and inherent large variance among individual mice for metabolite analyses, p ≤ 0.1 or one-tail log paired p ≤ 0.1was used as statistical threshold for differential abundance for metabolites between tumor and WT litter-/cage-mate samples and for inclusion in enrichment analyses. For exception in the Pten-KO model, nine metabolites of clinical relevance and with the ratio changes for all three pairs concordant in direction were included for analyses.

2.5 Bioinformatic and metabolic enrichment analyses

We performed the statistical analysis and the enrichment analysis utilizing MetaboAnalyst 4.0 (link: www.metaboanalyst.ca/).¹⁴ The partial least-squares discriminant analysis (PLS-DA) model was used to highlight differences between Pten-KO or TRAMP tumors versus respective WT mouse prostate. For the creation of the heat maps, the selected metabolites were also hierarchically clustered by distance measure using Pearson's correlation and clustering algorithm using average linkage (clustering uses the centroids of the observations). The values were scaled metabolite-wise in columns for heatmap visualization. For the metabolic enrichment analyses, location-based metabolite sets and pathway-associated metabolite sets in MetaboAnalyst's enrichment analysis (MSEA) module were selected as a metabolite library for MSEA. The metabolic pathway map of the selected differential metabolites and significantly changed metabolic pathways was analyzed and visualized by iPath 3.0 software (https://pathways.embl.de/ipath3.cgi).

3.1 Aqueous metabolome of Pten-KO prostate

Using the UHPLC-MS/MS platform, we identified 177 endogenous metabolites in the mouse prostate aqueous extracts (Table S1, Raw data). With the exception of a detectable increase of cholesteryl sulfate, it was apparent for a lack of detection of hydrophobic lipids, fatty acids, cholesterol, and cholesteryl fatty acid esters. Table 1 summarized 44 selected metabolites (20 greater and 24 less in Pten-KO prostate than in WT prostate) that met criteria of statistical threshold or concordance of change directions described in Section 2, 2 (more details in Table S1, select metabolites). A supervised PLS-DA of the variance of these select metabolites between the two genotypes showed an obvious separation of WT prostate tissue (green crosses) and Pten-KO mutant prostate (red triangles) (Figure 1A). The heatmap of the top 25 metabolites (14 more abundant, 11 less abundant than WT) (Figure 1B) showed distinctive clustering of the metabolites between the genotypes.

Some metabolites were remarkably increased in the Pten-KO prostate tumors (Table 1). They included many ribosyl- and deoxyribosyl nucleotides (ADP, ATP, IDP, dAMP, dCMP, dGDP, dGTP), cytosine and nucleotide derivatives (AICA-ribonucleotide, NADP+, UDP-d-glucuronate), uric acid (a purine metabolite), glutathione (GSH) and its oxidized disulfide (GSSG) and key cysteine precursor cystathionine. Beside the increased abundance of cholesteryl sulfate, Pten-KO prostate contained a much-elevated level of ornithine, a known precursor to polyamines and a product of arginase to catalyze the release of urea from arginine (urea cycle).

On the other hand, metabolites with diminished abundance in the Pten-KO prostate (Table 1) included amino acid derivatives (N-acetyl-l-alanine, N-acetyl-glutamate, acetyl-lysine, S-ribosyl-l-homocysteine), purine base or nucleoside (adenosine, guanine), glycolysis intermediates, phosphorylcholine intermediates as well as other arginine and glycine metabolites such as creatine, creatinine, and sarcosine.

3.2 Bioinformatic analyses of Pten-KO prostate metabolome data set and integration with key proteomic and transcriptomic changes

We used MESA to identify possible pathways and biologically meaningful patterns enriched in the metabolomic data (Tables 2 and S2). The functional enrichment analysis yielded nine significant metabolic pathways (two upregulated and seven downregulated pathways) with p < 0.05 (Table 2). Purine metabolism and glutathione metabolism were the two upregulated pathways. The seven downregulated pathways were glycerophospholipid metabolism; arginine and proline metabolism; glycine, serine, and threonine metabolism; alanine, aspartate, and glutamate metabolism; glycerolipid metabolism, glycolysis, or gluconeogenesis; and pyruvate metabolism. Location-based metabolite sets MESA predicted that these metabolic pathways were mainly associated with mitochondria, lysosome, and prostate (Figure 1C). Not surprisingly, mitochondria and, to a less extent, lysosomes are the major subcellular organelles for providing building blocks as well as regulators of metabolic flux and redox activity in the cell.¹⁵

Based on mammalian biochemical reactions, Figure 2 mapped the majority of detected metabolite changes (bold for increased in Pten-KO; shaded white for decreased in Pten-KO) within the interconnected metabolic pathways, shunts, and cycles, focusing on glucose fates, GSH/GSSG, and their constituent amino acids glutamate, cysteine and glycine, and arginine/ornithine/urea cycle. The Pten-KO prostate shunted glucose away from glycolysis, as indicated by decreased dihydroxy-acetone-phosphate and glycerate and decreased glycerol/choline pool (decreased glycerol-3-phosphate, glycerophosphocholine, and phosphorylcholine), and from the flux through the citrate cycle, as indicated by diminished supply of acetyl-CoA to the mitochondria (decreased dephopsho-CoA, and decreased acetylation of glutamate and select amino acids and glucosamine) and a modest decrease of oxaloacetate. Instead, glucose was channeled toward pentose phosphate cycle (increased gluconate) for NADPH production and for the pentose moieties essential for anabolic synthesis or salvage of purine and pyrimidine nucleotides for nucleic acids crucial cell proliferation; and to the glucuronidation pathway for Phase II drug metabolism (increased UDP-glucuronate). The Pten-KO prostate increased the overall synthesis of GSH and redox fluxes of GSH/GSSG pair. Increased cystathionine would fuel a steady supply of cysteine for GSH synthesis. Another key amino acid glycine for GSH synthesis was likely made more available by diversion away from its S-adenosyl-methionine-driven methylation to generate sarcosine and the production of creatine via arginine:glycine aminotransferase (AGAT). For GSH/GSSG redox and GSH driven functions, we have reported increased protein abundance of glutathione disulfide reductase (GSR), glutathione S-transferases omega-1 (GSTO1) and A4 (GSTA4) and mRNA increases of Gsto1, glutathione peroxidase 2 gene (Gpx2), and hexose-6-phosphate dehydrogenase (H6pd) in the Pten-KO prostate.¹⁰

In addition to diversion from creatine/creatinine production, arginine was likely shunted away from nitric oxide synthesis (T cell maturation) to allow arginase to produce the polyamine precursor ornithine (part of the urea cycle). The mRNA level for Arg1 (arginase 1) gene was much increased in our macromolecular omics analyses of Pten-KO prostate,¹⁰ consistent with and supportive of infiltration of myeloid-derived (immune) suppressive cells (MDSC) and inflammatory cells.⁹ Unfortunately, our metabolomics platform failed to detect any polyamines or their further metabolites (no detection in Table S1).

3.3 Comparison with reported metabolomics datasets, and human relevance

As a measure of cross validation, we compared our data set with the metabolomics analysis reported by Zabala-Letona et al. of Pten-KO mouse prostate.¹⁶ They reported that the most prominently increased metabolites were higher-order polyamines, such as N-acetyl-spermidine and N-acetyl-spermine, which were not detected by our platform (Table S1). Four metabolites (IDP, ADP, uric acid, and cholesteryl sulfate) matched our panel of 20 increased metabolites (Table 1). In contrast to our finding of elevated abundance of NADP+ and glutathione (GSH), they reported decreased levels of both metabolites in the Pten-KO prostate tumors. Among the 24 decreased metabolites, they reported five that matched the same change trajectory, that is, N-acetylglucosamine, creatinine, phosphorylcholine, adenosine, and sn-glycerol-3-phosphoate (Table 1). However, they reported increased levels of glycerate, hydroxyphenylpyruvate, and N-acetyl-l-alanine in the Pten-KO prostate. These discrepancies could be caused by the difference in the mass spectrometer type (QTRAP 5500 mass spectrometer (ours) vs. Agilent 6520 QToF mass spectrometer¹⁶), as well as the prostate tissue sample acquisition and extract preparation methods.

Regarding reported metabolite changes in human PCa specimens, NADP+, ornithine, glutathione, adenosine, and N-acetyl-glutamate were found to change in the same directions as in our Pten-KO mouse data set (Table 1). On the other hand, uric acid, GSSG, N-acetylglucosamine, glycerate, and sn-glycerol-3-phosphate did not change in the same patterns between human PCa and our profiled mouse Pten-KO prostate (Table 1). In particular, both the human PCa datasets from Jung et al.¹⁷ and Ren et al.¹⁸ could not detect ATP and most of other high-energy phosphorylated nucleotides (e.g., dGTP, dGDP, IDP, ADP, etc.). Previously, many studies failed to show the consistent extraction of high-energy phosphorylated compounds (e.g., ATP and NADPH) known to be susceptible to degradation.¹⁹ A mixture of acetonitrile:methanol:water (40:40:20) with 0.1 M formic acid and subsequent neutralization with ammonium bicarbonate was considered as an effective solvent system for both quenching and extraction.¹⁹ The solvent systems used for their metabolite extractions (Ren paper: formic acid, ammonium bicarbonate, and methyl tert-butyl ether¹⁸; Jung paper: water/ethanol/dichloromethane/acetonitrile¹⁷) could be suboptimal for high-energy compounds.

3.4 Aqueous metabolome of TRAMP NECa and comparison with Pten-KO prostate

In the TRAMP NECa, our aqueous metabolomics platform detected increased levels of cystathionine, GSH and GSSG, and purine and pyrimidine nucleotides of both ribose and deoxyribose, yet without a significant change of ornithine and even a decrease of cholesteryl sulfate over their paired counterpart WT prostate (Tables 1 and S3). We detected decreased levels in the TRAMP NECa of adenosine, creatine and creatinine, N-acetyl-glucosamine, glycerate, phosphorylcholine (PC), and glycerophosphocholine (GPC), as in the Pten-KO prostate. These concordant changes in both the Pten-KO and the TRAMP mouse prostate cancer models suggested that these metabolites could play common roles in prostate carcinogenesis, or the changes could be an indication of the disruption of normal prostate metabolic functions. An earlier study using (1)H-NMR showed that in TRAMP mice of 28 weeks of age, reduced tumor levels of citrate (49%), choline (33%), PC (57%), GPC (66%), and glycerophosphoinositol (61%) were observed relative to age-matched WT prostate (p < 0.05).²⁰

However, in contrast to the Pten-KO prostate, prominent increases of many amino acids and their metabolites were detected in the TRAMP NECa (Table S3). Indeed, pathway enrichment analyses (Table S4) showed that five of six significantly (p < 0.05) upregulated metabolic pathways were related to the metabolisms of methionine; glycine and serine; homocysteine degradation; aspartate; and glutamate. The other was purine metabolism.

Figure 3A showed a clear separation of the malignant TRAMP NECa from the WT prostate specimens by supervised PLS-DA of the variance of the differential metabolites between TRAMP (red triangles) and paired WT mouse prostate tissue (green crosses). The heatmap of the top 25 differential metabolites (18 more abundant, seven less abundant than WT) (Figure 3B) showed distinctive clustering of the metabolites between NECa and the normal prostate. Location-based metabolite sets MESA predicted that the altered TRAMP metabolic pathways were mainly associated with peroxisome and lysosome subcellular organelles and the implicated organs were related to NE signaling and innervations (nerve cells; neuron; brain; prostate, pancreas, heart, and muscles) (Figure 3C).

Figure 4 mapped the majority of detected metabolite changes (bold for increased in TRAMP NECa; shaded white for decreased in TRAMP NECa) within the interconnected metabolic pathways, shunts, and cycles. TRAMP NECa appeared to shunt glucose away from glycolysis and citrate cycle, even more so than the Pten-KO prostate. Instead, glucose was channeled toward the glycosylation (increased UDP-glucose) and glucuronidation (increased UDP-glucuronate) pathways. As in the Pten-KO prostate, the TRAMP NECa increased GSH and GSSG, in association with increased cystathionine to increase cysteine, and shunting glycine away from degradation (i.e., decreased creatine and creatinine). Several distinctions from Pten-KO prostate were noteworthy. Foremost, the elevated pools of many amino acids would undoubtedly provide ample substrates for protein synthesis. Besides supporting protein synthesis, the extraordinarily elevated aspartate (Asp) and asparagine (Asn) and elevated carbamoyl-Asp might fuel the de novo syntheses of uracil and subsequent conversion to uridine, cytidine, and the further downstream ribosyl and deoxyribosyl pyrimidine nucleotides for RNA and DNA syntheses. In addition, the increased S-adenosyl-methionine (as well as its de-methylated product S-adenosyl-homocysteine) was interpreted as providing abundant methyl donor for the methylation epigenetic regulations of DNA and other macromolecules. An earlier study had shown DNA hypermethylation of poorly differentiated TRAMP tumors.²¹ Furthermore, in the TRAMP NECa, elevated levels of UDP-ethanolamine and UDP-choline plus depleted levels of ethanolamine and choline and choline phospho intermediary metabolites might be consistent with potential de novo synthesis of phosphatidyl lipids required for cell membranes in the malignant cells. Moreover, TRAMP NECa decreased arginine degradation (urea level) and did not increase ornithine production, nor uric acid but increased a purine degradation product allantoate which fell further downstream of uric acid.

4.1 Limitations

The major caveat for the employed metabolome platform was a bias for hydrophilic metabolites, excluding almost all lipids, fatty acids, and sterols and their hydrophobic metabolites. Another weakness was the lack of anatomical (different prostate lobes) and cell-type distribution information of the metabolites due to the homogenization of each tissue sample. Future efforts should include lipodomics for the hydrophobic metabolites, as well as imaging MS approaches that provide metabolite mapping resolution to cellular or subcellular organelle level on tissue microscopic slice.