Search for Early Pancreatic Cancer Blood Biomarkers in Five European Prospective Population Biobanks Using Metabolomics

Abstract

Most patients with pancreatic cancer present with advanced disease and die within the first year after diagnosis. Predictive biomarkers that signal the presence of pancreatic cancer in an early stage are desperately needed. We aimed to identify new and validate previously found plasma metabolomic biomarkers associated with early stages of pancreatic cancer. Prediagnostic blood samples from individuals who were to receive a diagnosis of pancreatic cancer between 1 month and 17 years after sampling (N = 356) and age- and sex-matched controls (N = 887) were collected from five large population cohorts (HUNT2, HUNT3, FINRISK, Estonian Biobank, Rotterdam Study). We applied proton nuclear magnetic resonance-based metabolomics on the Nightingale platform. Logistic regression identified two interesting hits: glutamine (P = 0.011) and histidine (P = 0.012), with Westfall-Young family-wise error rate adjusted P values of 0.43 for both. Stratification in quintiles showed a 1.5-fold elevated risk for the lowest 20% of glutamine and a 2.2-fold increased risk for the lowest 20% of histidine. Stratification by time to diagnosis suggested glutamine to be involved in an earlier process (2 to 5 years before diagnosis), and histidine in a process closer to the actual onset (<2 years). Our data did not support the branched-chain amino acids identified earlier in several US cohorts as potential biomarkers for pancreatic cancer. Thus, although we identified glutamine and histidine as potential biomarkers of biological interest, our results imply that a study at this scale does not yield metabolomic biomarkers with sufficient predictive value to be clinically useful per se as prognostic biomarkers.

Conflict of interest statement

Restrictions apply to the availability of data generated or analyzed during this study to preserve patient confidentiality or because they were used under license. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.

Copyright © 2019 Endocrine Society.

Figures

Figure 1.

Schematic overview of the sample set used for data analysis and the different data analysis approaches performed in the current study. aAny individual containing missing values in metabolomics measurements or phenotypical information were assumed to be missing at random and were removed from the data set. bAny individual containing missing values in phenotypical information were removed from the data set. PC, pancreatic cancer.

Figure 2.

Concentrations (logarithmic scale) of (A–D) glutamine and (E–H) histidine in the blood circulation in controls and cases, that is, those individuals who developed pancreatic cancer within a time window after blood sampling. (A and E) Distribution of the concentrations of controls (light blue) and cases (dark blue) in the different cohorts analyzed (EGCUT, FR, HUNT2, HUNT3, RS). (B and F) Distribution of concentrations in nondiabetics (light blue) and individuals diagnosed with T2DM (dark blue). (C and G) Distribution of concentrations in controls and cases sampled within 2 y before diagnosis, between 2 and 5 y before diagnosis, and >5 y before diagnosis. (D and H) Distribution of concentrations in nonfasting individuals (light blue), individuals who had a meal between 4 and 8 h before blood draw (dark blue), and fasting individuals (green, last meal was >8 h before blood draw). Box plots reflect the distribution of the concentrations in individual samples, including the middle quartiles (25th to 75th percentile of the data points are in the boxes); the horizontal band; the median value; the lower whiskers representing the data points up to 1.5 × the interquartile range (IQR) below the 25th percentile; the upper whiskers representing the data points up to 1.5 × IQR above the 75th percentile; the data points outside these ranges plotted as individual data points.

Figure 3.

Forest plots from random effects meta-analysis across different cohorts for (A) glutamine and (B) histidine. The meta-analysis was performed on the β coefficients and SD from the logistic regressions run for each cohort separately. In the logistic regression, pancreatic cancer status was modeled as a function of log-transformed and standardized metabolite concentration, sex, age, BMI, smoking status, T2DM, and fasting status. Shown are the estimated effect size, the SE on this estimate, the estimated OR and the CI on this ratio, the weight of the individual cohort on the calculation of the final estimate, the heterogeneity measure (modeling differences between cohorts), and the unadjusted and Bonferroni–Holm-corrected P values for the respective metabolites.

Figure 4.

Receiver operator curves for classification of pancreatic cancer cases (sampled up to 5 y before diagnosis) and controls for (A) training set (70% of all individuals) and (B) performance testing set (30% of all individuals unseen during the variable selection). In red, the null model is shown in which only the clinical covariates (sex, age, BMI, smoking status, T2DM, and fasting status) were included in the regression. In blue, the alternative model is shown where the metabolites selected by the LASSO regression were included in addition to the clinical covariates. The AUCs are indicated, as well as the specificity (1 − false-positive rate) at 70% sensitivity.