Metabonomics in NETs · Evolving landscape of neuroendocrine tumor disease: Predicting tumor behaviour using metabolic profiling

Evolving landscape of neuroendocrine tumor disease: Predicting tumor behaviour using metabolic profiling

Neuroendocrine tumors (NET) comprise a heterogeneous group of neuroendocrine neoplasms(NEN) arising from different cells distributed throughout the human body which share a common neuroendocrine phenotype. The majority of such tumors originate from either the pancreas or the small bowel. They represent an important clinical issue for two reasons: firstly, 40-95% are metastatic at diagnosis, and secondly, evidence-based best practice strategies are scarce. A critical unmet need in the management of patients with NET is the lack of a sensitive and specific set of biomarkers which would offer a base for accurate diagnosis, allow patient-specific prognostication, determine response to treatment, and assess tumor dynamics in the advanced stages of disease. Systems medicine approaches integrate high throughput “-omics” technologies into diagnostic platforms for use at the point of care. Metabonomics describes the response of complex metabolic systems to perturbations through time, typically measured in blood, urine or tissue with analytical platforms such as nuclear magnetic resonance (NMR) spectroscopy or mass spectrometry (MS). It is possible to map these responses by applying a suite of supervised and unsupervised statistical tools to complex data sets in both targeted and untargeted manners. This approach therefore provides data on a series of interacting metabolic networks operating within multiple body compartments, facilitating analysis of the continuum of metabolic processes contributing to the overall metabo-type. Metabonomic approaches have now been extensively used in various clinical conditions including cancer, where this technology was shown to outperform standard tumor markers for hepatocellular carcinoma and also discriminate between early- and late-stage colorectal cancers. With our project we attempted to apply a metabolic phenotyping approach to the analysis of NET to determinate its clinical utility as a novel biomarker. The work was performed at Imperial College London which provides an optimal scientific environment for this type of research. As a European Neuroendocrine Tumor Society (ENETS) Centre of Excellence for Neuroendocrine Tumors, more than 100 new NET patients are referred to the Imperial clinical team annually. The MRC–NIHR Phenome Centre at Imperial is run by a world-leading metabolic phenotyping team and is a national resource for the biomedical research community to undertake metabolic phenotyping. The Imperial Tissue Bank ensures state-of-the art storage of biological samples and their clinical information. In our pilot study, by using urine samples from 28 treatment-naïve patients with different types of NET and 17 healthy individuals serving as a control group, we were able to demonstrate that metabonomic analysis has the potential to not only distinguish between those healthy and diseased, but also between different groups of tumors (training set). Urine samples were analysed using NMR and MS. Partial least squares-discriminant analysis (PLS-DA) score plots of 1D 1HNMR urinary spectra demonstrated strong clustering of samples according to the presence or absence of NET (Figure 1 – see attachment and Scientific publications*). Receiver operating characteristics (ROC) curve analysis demonstrated reasonable sensitivity and good specificity of the model. An orthogonal projection to latent structure-discriminant analysis (OPLS-DA) pseudo-loadings plot was used to visualize the metabolic variations between the classes (Figure 2 – see attachment and Scientific publications*). By using OPLS-DA we were able to distinctly separate the pancreatic NET and the small bowel NET classes. The 7-fold cross validation model had a diagnostic sensitivity of 71.4%, a specificity of 80%, and an area under the curve (AUC) of 0.85. In our validation study we developed a robust database for the collection of clinical patient data and registered the sample collection as a sub-collection of the Imperial College Tissue Bank. Patient details and sample details were entered onto the database of the Tissue Bank. Sample pipeline construction and definition of the optimum sampling conditions were performed. We collected biological samples (blood, urine, and tissue where applicable) of 54 new NET patients and 21 age- and sex-matched healthy controls. The urinary metabotypes of patients and healthy controls were analysed and compared between different subtypes of NET (pancreas vs. small bowel) and different stages of the disease. The results validated the results achieved in the training set. A urinary metabolite panel was designed. Metabolites whose 1 H resonances appeared in those regions of NMR spectra containing the most significant differences between samples generated from patients and controls and from different sub- groups of patients were identified using the proprietary B-Bioref-Code (BrukerBioSpin) metabolite database as well as the in-house built databank. These metabolites will be utilised to develop a robust and reproducible metabonomic signature with the potential to serve as a novel biomarker for NET. We hypothesise that this new tumor marker will diagnostically outperform chromogranin A, which is currently used a standard blood-based tumor marker for NET. Having discovered such novel metabolic biomarkers, we will then translate these into clinical tools for use in stratified medicine. Baseline characterisation and follow-up of NET patient cohorts with metabolic phenotyping will be used to improve patient stratification; e.g. in the selection for specific treatments, early identification of poor responders to treatment or those with poor prognosis, who could then be offered alternative or intensified treatment options. The Imperial group has shown that monitoring complex patient journeys in a hospital environment by “patient journey metabolic phenotyping” is already a realistic prospect for improving patient care. Multivariate longitudinal modelling of relationships between phenotype variations at different stages for multiple patients from the same disease/hospital journey enables linkage of subject heterogeneity prior to treatment to post-interventional therapeutic outcome. This allows derivation of probabilistic models for patient stratification, optimisation of the choice and form of intervention for future patients, and identifies new prognostic biomarkers and drug targets. Metabolic profiles can be modelled in relation to routinely collected clinical data to provide a reference framework for novel diagnostics and targeting treatment modalities. Cross-integration of genetic and metabolic phenotype data of an individual patient offers the genuine possibility of future low-cost stratified medicine and personalised health care. Contact details Professor Andrea Frilling Head ENETS Centre of Excellence for NET Department of Surgery and Cancer Division of Surgery Imperial College London Hammersmith Campus Du Cane Road London W12 0HS e-mail: a.frilling@imperial.ac.uk www.imperial.ac.uk Professor Jeremy K. Nicholson Head Department of Surgery and Cancer Head MRC-NIHR Phenome Centre Department of Surgery and Cancer Imperial College London South Kensington Campus Exhibition Road London SW7 2AZ e-mail: j.nicholson@imperial.ac.uk www.imperial.ac.uk

Data: CORDIS, © European Union