Neuroendocrine Neoplasms: Identification of Novel Metabolic Circuits of Potential Diagnostic Utility

Abstract

The incidence of neuroendocrine neoplasms (NEN) is increasing, but established biomarkers have poor diagnostic and prognostic accuracy. Here, we aim to define the systemic metabolic consequences of NEN and to establish the diagnostic utility of proton nuclear magnetic resonance spectroscopy (¹H-NMR) for NEN in a prospective cohort of patients through a single-centre, prospective controlled observational study. Urine samples of 34 treatment-naïve NEN patients (median age: 59.3 years, range: 36-85): 18 had pancreatic (Pan) NEN, of which seven were functioning; 16 had small bowel (SB) NEN; 20 age- and sex-matched healthy control individuals were analysed using a 600 MHz Bruker ¹H-NMR spectrometer. Orthogonal partial-least-squares-discriminant analysis models were able to discriminate both PanNEN and SBNEN patients from healthy control (Healthy vs. PanNEN: AUC = 0.90, Healthy vs. SBNEN: AUC = 0.90). Secondary metabolites of tryptophan, such as trigonelline and a niacin-related metabolite were also identified to be universally decreased in NEN patients, while upstream metabolites, such as kynurenine, were elevated in SBNEN. Hippurate, a gut-derived metabolite, was reduced in all patients, whereas other gut microbial co-metabolites, trimethylamine-N-oxide, 4-hydroxyphenylacetate and phenylacetylglutamine, were elevated in those with SBNEN. These findings suggest the existence of a new systems-based neuroendocrine circuit, regulated in part by cancer metabolism, neuroendocrine signalling molecules and gut microbial co-metabolism. Metabonomic profiling of NEN has diagnostic potential and could be used for discovering biomarkers for these tumours. These preliminary data require confirmation in a larger cohort.

Keywords: biomarkers; metabolic profiling; metabonomics; neuroendocrine neoplasms; neuroendocrine tumours; nuclear magnetic resonance; precision medicine.

Figure 1

Remark diagram of the cohort studied.* Exclusion criteria are: patients under the age of 18, pregnancy, patients undergoing neuroendocrine neoplasm (NEN)-specific systemic treatment, second malignancies, comorbidities requiring significant systemic treatment (e.g., immunosuppression), impaired renal function, poor compliance, missing consent for study participation. ^# 1 healthy control sample due to the corruption of the original data file; for NEN samples, 1 due to poor water suppression, 3 due to high intensity of an unknown drug signal, and 3 due to extreme misalignment of the spectra. PanNEN = pancreatic neuroendocrine neoplasm, SBNEN = small bowel neuroendocrine neoplasm, BMI = body mass index.

Figure 2

Ability of the OPLS-DA models to predict NEN status from healthy controls. (A) R²Y and Q²Y coefficients of the OPLS-DA models generated. (B) Receiver operating curves (ROC) and their associated area under the receiving operator curve (AUROC) values for Healthy vs. NEN (AUROC 0.94), (C) Healthy vs. PanNEN (AUROC 0.90) and (D) Healthy vs. SBNEN (AUROC 0.90). These demonstrate strong diagnostic utilities of the models for NEN and its subgroups.

Figure 3

OPLS-DA models comparing NEN and its subgroups to healthy control samples. Scores (A–C) and their respective loadings (D–F) plots of the model comparing healthy individuals versus all NEN patients (A,D), healthy individuals versus those with PanNEN (B,E) and healthy individuals versus those with SBNEN (C,F). The pseudospectra of the loadings plots are coloured according to correlation coefficient |r| for peak positions where positive false discovery rate (pFDR) < 0.1.

Figure 4

Boxplot of metabolites with significant differences between NEN and NEN subgroups compared to healthy control: (A) Hippurate, (B) Trigonelline, (C) metabolite related to niacin, (D) S-methyl-L-cysteine sulfoxide, SMCSO, related metabolite, (E) 2-Hydroxybutyrate, (F) Trimethylamine N-oxide, TMAO, (G) 4-Hydroxyphenylacetate, (H) Phenylacetylglutamine, PAG, (I) Kynurenine, (J) 5-Hydroxyindoleacetic acid, 5HIAA. Numbers in the title correspond to the chemical shift in ppm of the metabolite signal used for integration and the letters describe the signal multiplicity: singlet, s, doublet, d, doublet, dd, triplet, t The symbols * and # indicate the levels of significance compared to healthy control based on Wilcoxon’s rank sum test (*) and OPLS-DA model correlation (#), respectively: * and # for pFDR < 0.05; ** and ## for pFDR < 0.01; *** for pFDR < 0.001.

Figure 5

Scores plot of cross-validated principal component analysis model of spectra from PanNEN patients with and without any metastases. (A) Component 1 vs. 2; (B) Component 1 vs. 3.