Metabolic Changes in Flatfish Hepatic Tumours Revealed by NMR-Based Metabolomics and Metabolic Correlation Networks Andrew D. Southam,† John M. Easton,‡ Grant D. Stentiford,§ Christian Ludwig,| Theodoros N. Arvanitis,‡ and Mark R. Viant*,† School of Biosciences, University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom, School of Engineering, University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom, Centre for Environment, Fisheries and Aquaculture Science (Cefas), Weymouth Laboratory, Weymouth, DT4 8UB, United Kingdom, and The Henry Wellcome Building for Biomolecular NMR Spectroscopy, CRUK Institute for Cancer Studies, University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom Received May 13, 2008
Histopathologically well-characterized fish liver was analyzed by 800 MHz 1H NMR metabolomics to identify metabolic changes between healthy and tumor tissue. Data were analyzed by multivariate statistics and metabolic correlation networks, and results revealed elevated anaerobic metabolism and reduced choline metabolism in tumor tissue. Significant negative correlations were observed between alanine-acetate (p ) 3.0 × 10-5) and between proline-acetate (p ) 0.003) in tumors only, suggesting alanine and proline are utilized as alternative energy sources in flatfish liver tumors. Keywords: Environment • dab flatfish • hepatocellular adenoma • cancer • fingerprinting • profiling • metabolic networks • FDR • false discovery rate • succinate
Introduction Flatfish are an ideal species to use for environmental monitoring as they live in close proximity with the ocean floor where toxicants and carcinogens can accumulate in the sediment. Recent environmental monitoring studies of flatfish using liver histopathology to identify neoplastic lesions have revealed that liver tumor prevalence can exceed 20% at some offshore sites,1,2 while prevalence in estuarine regions can be significantly higher.3,4 Little research has, however, been directed toward the molecular characterization of these tumors to identify biomarkers and gain insight into the pathways altered by the disease. Emerging ‘omics’-based technologies that attempt to describe changes in gene, protein or metabolite patterns can provide considerable mechanistic insight.5 Here we have applied NMR-based metabolomics to flatfish livers collected as part of the UK Clean Seas Environmental Monitoring Programme (CSEMP) in an attempt to identify biomarkers and, more crucially, to use the observed metabolic changes to reveal the pathways that are altered. Metabolomics has previously been used successfully to identify metabolic changes that occur in mammalian tumors,6,7 and it represents a powerful technique to characterize biochemical pathways altered in tumor tissue.8 However, very little information is currently available regarding tumors in fish; thus, the metabolic changes commonly seen in mammalian tumors * To whom correspondence should be addressed. Mark R. Viant, Phone: +44-(0)121-414-2219. Fax: +44-(0)121-414-5925. E-mail:
[email protected]. † School of Biosciences, University of Birmingham. ‡ School of Engineering, University of Birmingham. § Weymouth Laboratory. | The Henry Wellcome Building for Biomolecular NMR Spectroscopy, CRUK Institute for Cancer Studies, University of Birmingham. 10.1021/pr800353t CCC: $40.75
2008 American Chemical Society
are now introduced as these may help to rationalize the changes observed in flatfish liver tumors. A frequent metabolic occurrence in several types of tumor, including liver, is increased anaerobic respiration and flux through glycolysis, known as the Warburg effect.9 It initially arises due to the tumor tissue growing faster than the blood vessels, leading to poor perfusion and a hypoxic environment. However, as tumor tissue develops, an increase in anaerobic respiration occurs even in the presence of oxygen, which reflects the cellular demand for intermediate products of glycolysis that can be used for anabolic synthesis of nucleic acids and phospholipids.7,10 This change in respiration state is thought to be initiated by the transcription factor, hypoxia-inducible factor-1 (HIF-1), which often shows increased activity in tumor cells and is responsible for expression of many glycolytic enzymes and enzymes involved in cellular growth and angiogenesis.11 Also, metabolites involved in choline phospholipid metabolism tend to be elevated in tumor tissue reflecting the increased demand for cellular phospholipids for membrane and organelle synthesis, and previous studies have shown that characteristic changes in levels of choline containing metabolites can be indicative of the tumor developmental stage and reflect how aggressively the tumor is growing.12,13 The metabolic changes previously observed in choline phospholipid metabolism and energy respiration in the tumor metabolome highlight the very different metabolic demands of the tumor cell compared to the quiescent cell. Metabolic processes involved in anabolic growth in the tumor cell appear to be of high priority, even if this compromises the efficiency of energy production, and such anabolic processes will ultimately determine the speed at which cells can grow and proliferate. Journal of Proteome Research 2008, 7, 5277–5285 5277 Published on Web 11/11/2008
research articles
Southam et al. a
Table 1. Location of Capture, Sex and Associated Lesions of the Flatfish Used in This Study fish ID
sampling site
sex
tumor type
associated lesions
1 2 3 4 5 6 7 8 9 10
Liverpool Bay, Irish Sea Liverpool Bay, Irish Sea Liverpool Bay, Irish Sea Liverpool Bay, Irish Sea Burbo Bright, Irish sea Inner Cardigan Bay, Irish Sea Burbo Bright, Irish Sea St. Bee’s, Irish Sea Inner Cardigan Bay, Irish Sea North Dogger Bank, North Sea
F M M M F F F M F F
HCA HCA HCA HCA HCA HCA HCA HCA HCA HCA
bFCA, MMA MMA MMA, FIB MMA bFCA, MMA, NEC eFCA, MMA, INF, NEC bFCA, MMA, INF, NEC INF, FIB, APOP, MMA -
a Key: HCA, hepatocellular adenoma; bFCA, basophilic focus of cellular alteration; eFCA, eosinophilic focus of cellular alteration; MMA, melanomacrophage aggregates; INF, inflammatory lesion; NEC, coagulative necrosis; APOP, apoptosis; FIB, hepatocellular fibrillar inclusions.
Here, the characterization of significant changes within the tumor metabolome, including identification of altered metabolic pathways, was addressed using two complementary approaches. First, NMR-based metabolic fingerprinting was used as an exploratory approach to identify specific phenotypes for the healthy and diseased livers.14-16 No prior knowledge of the metabolic content of the spectra is required; instead, metabolites are only identified, quantified and evaluated statistically once they have been shown to contribute toward discrimination of phenotypes. In addition, we utilized a metabolic profiling approach17-19 involving quantification of all identifiable metabolites within the NMR spectra. This second approach enabled correlation network analysis,20,21 which builds metabolic networks based upon the correlations between all pairs of metabolites within a single phenotype. The results are visualized as a network where metabolites are represented as nodes and connections between nodes indicate a significant correlation between two metabolites. Networks can be generated for each phenotype (e.g., healthy and tumor) and differences between these can indicate that correlations are lost or gained between phenotypes. Loss (or gain) can suggest that an element of metabolic regulation has been lost (or gained) between the two metabolites in question. Specifically, we have applied high-resolution 800 MHz 1H NMR-based metabolomics to histopathologically well-characterized livers dissected from a wild-caught species of marine flatfish, the dab (Limanda limanda), utilized as a disease sentinel in European marine monitoring programs. The aims of the study were to identify metabolic biomarkers that indicate tumor presence and to use metabolic networks to characterize disruptions to the metabolome caused by the tumor phenotype. Both healthy and tumor tissue were sampled from each fish liver, such that direct comparison of these ‘paired’ samples would reveal specific changes associated with the tumor phenotype and not be confounded by interindividual variation between environmentally sampled animals. By integrating the results from the more traditional multivariate statistics with the metabolic correlation networks, we aimed to generate rich biochemical insight into changes in the liver tumor metabolome of flatfish.
Materials and Methods Collection of Fish Liver. Several hundred dab flatfish were captured at United Kingdom Clean Seas Environmental Monitoring Program (CSEMP) sites during June and July of 2002 using 30-min tows of a standard Granton trawl by the Centre for Environment, Fisheries and Aquaculture Science (Cefas, 5278
Journal of Proteome Research • Vol. 7, No. 12, 2008
Weymouth, U.K.), and of these fish, 10 exhibited hepatocellular adenoma and were used in this study (Table 1). Upon landing, dab flatfish were immediately removed from the catch and placed into flow-through tanks containing aerated seawater. Following external disease assessment, each fish was sacrificed and the liver visually assessed for the presence of nodules. When present, nodules and a piece of adjacent healthy liver tissue were resected according to the methods of Feist et al.2 Sections of the nodules and healthy liver tissue samples were frozen immediately for metabolomic analysis (and remained frozen at -80 °C until analysis), while the remainder of these samples were fixed for histological analysis to confirm the presence or absence of lesions. Histopathology. Fixed samples were processed to wax in a vacuum infiltration processor and tissue sections were cut at 3-5 µm on a rotary microtome and then mounted onto glass slides before staining with hematoxylin and eosin. Stained sections were analyzed by light microscopy (Eclipse E800, Nikon, U.K.) and diagnosis of liver tumor type followed established guidelines for flatfish liver.2 Ten samples of tumor and corresponding nontumor were selected, and all histological data of interest was noted (Table 1). All lesions were classified as hepatocellular adenoma due to their defined nodular form, thickened trabecular structure, relative lack of melanomacrophage aggregates, dilated blood vessels and relative absence of atypical nuclear and cellular profiles. Preparation of Liver Extracts and Spiked Samples. Each liver was homogenized in ice-cold methanol (8 µL/mg wet tissue mass, all solvents HPLC grade, Fisher Scientific) and water (2.5 µL/mg) for 1 min using a Potter homogenizer (Fisher Scientific). Next, chloroform (8 µL/mg) and water (4 µL/mg) were added, the sample was homogenized for a further 1 min, and the biphasic mixture was centrifuged (1500g) for 10 min. The upper (polar) and lower (nonpolar) phases were isolated and frozen at -80 °C. Prior to NMR analysis, polar extracts were dried using a centrifugal concentrator (Thermo Savant, Holbrook, NY), resuspended in 650 µL of 0.1 M, pH 7.4, D2O phosphate buffer containing 0.5 mM trimethylsilyl-propionate (TMSP), and then centrifuged (5000g). A total of 600 µL of the supernatant was transferred into a 5 mm diameter NMR tube. The nonpolar extracts were not used in this study. To confirm the identities of those metabolites that lacked characteristic spin-spin coupling patterns (postinitial analysis), metabolite standards were individually spiked into 550 µL aliquots of a pooled polar liver extract. NMR Spectroscopy. Samples were analyzed by one-dimensional 1H NMR spectroscopy using an INOVA 800 MHz
Metabolic Changes in Flatfish Hepatic Tumours spectrometer (Varian) and room temperature probe. Sample temperature was 22 °C and presaturation was utilized to suppress the residual water. Specifically, one-dimensional spectra were obtained using a 60° pulse, 8.5 kHz spectral width and a 3-s relaxation delay, with 320 transients collected into 32 768 data points. All data sets were zero filled to 65 536 data points and exponential line-broadenings of 0.5 Hz were applied before Fourier transformation. Spectra were calibrated (TMSP peak at 0.0 ppm), then manually phased and baseline corrected. The samples spiked with metabolite standards were analyzed using the same spectrometer and pulse sequence, except it was equipped with a cryogenic probe and therefore the number of transients was reduced to 80. All other settings were identical. Multivariate Analysis. Spectra were read into ProMetab,22 custom written software in Matlab (version 7, The MathWorks, Natick, MA), and were truncated to a 0.2-10.0 ppm range. The residual water peak (4.70-5.23 ppm) was removed, and then spectra were normalized to a total spectral area of unity and a generalized log transformation (with transformation parameter λ ) 1 × 10-7) was applied.23,24 The full resolution spectral data (29 731 data points per spectrum after processing) were meancentered and subject to principal components analysis (PCA) and partial least-squares discriminant analyses (PLS-DA) using the PLS_Toolbox in Matlab (Eigenvector Research). Binning was not required as the NMR peaks did not shift across multiple spectra due to, for example, variable pH. Upon the basis of the multivariate analyses, peaks with large loadings were subsequently identified and quantified. This was conducted on the non-log transformed spectra using NMR Suite (Chenomx, professional version 4.6).19 Metabolic Correlation Network Analysis. With the use of the original spectral data, the NMR Suite software was used to identify and quantify as many metabolites as possible within the spectra. The quantified metabolic data for each spectrum was normalized using the same scaling parameter as for the multivariate analysis section (above) and then read into R (R Foundation for Statistical Computing, version 2.4.1). Next, Pearson correlation coefficients were calculated for all permutations of pairs of metabolites. The 10 spectra corresponding to each of the healthy and tumor phenotypes were analyzed separately. For each phenotype, the resulting 595 pairwise correlations between the 35 identified metabolites were then ranked according to their r-values and a False Discovery Rate (FDR)
