Matching Drug Metabolites from Non-Targeted Metabolomics to Self-Reported Medication in the Qatar Biobank Study

Abstract

Modern metabolomics platforms are able to identify many drug-related metabolites in blood samples. Applied to population-based biobank studies, the detection of drug metabolites can then be used as a proxy for medication use or serve as a validation tool for questionnaire-based health assessments. However, it is not clear how well detection of drug metabolites in blood samples matches information on self-reported medication provided by study participants. Here, we curate free-text responses to a drug-usage questionnaire from 6000 participants of the Qatar Biobank (QBB) using standardized WHO Anatomical Therapeutic Chemical (ATC) Classification System codes and compare the occurrence of these ATC terms to the detection of drug-related metabolites in matching blood plasma samples from 2807 QBB participants for which we collected non-targeted metabolomics data. We found that the detection of 22 drug-related metabolites significantly associated with the self-reported use of the corresponding medication. Good agreement of self-reported medication with non-targeted metabolomics was observed, with self-reported drugs and their metabolites being detected in a same blood sample in 79.4% of the cases. On the other hand, only 29.5% of detected drug metabolites matched to self-reported medication. Possible explanations for differences include under-reporting of over-the-counter medications from the study participants, such as paracetamol, misannotation of low abundance metabolites, such as metformin, and inability of the current methods to detect them. Taken together, our study provides a broad real-world view of what to expect from large non-targeted metabolomics measurements in population-based biobank studies and indicates areas where further improvements can be made.

Ratios of Acetaminophen Metabolites Identify New Loci of Pharmacogenetic Relevance in a Genome-Wide Association Study

“…For this study we used all samples of our previous study on medication use in QBB [14] for which at least five of the nine acetaminophen metabolites were reported (N = 520, Table 1). The participants in this group were aged between 18 and 76 years with an average age of 40.1 years, and 308 samples were from female study participants (59.2%).…”

“…A total of 3000 EDTA blood plasma samples were analyzed on the non-targeted metabolomics platform of Metabolon Inc. (Morrisville, NC, USA) at the Anti-Doping Laboratory-Qatar (ADLQ) as previously described [11,14]. Briefly, samples were analyzed on an LC-MS system, consisting of a Waters ACQUITY ultra-performance liquid chromatography (UPLC) unit and a Thermo Scientific Q-Exactive high resolution/accurate mass spectrometer, interfaced with a heated electrospray ionization (HESI-II) source and Orbitrap mass analyzer operated at 35,000 mass resolution.…”

“…The provided dataset comprised a total of 1159 biochemicals, 937 compounds of known identity (named biochemicals) and 222 unidentified compounds (unknowns). Full details of the QBB metabolomics data are provided in [14]. In this study we only use data from 520 QBB study participants for whom at least five of the nine acetaminophen metabolites were detected (see Table 4 in [14]).…”

“…A second challenge resides in the relative quantification of metabolites with high missingness: due to the large sample numbers required for GWAS, metabolomics measurements are required to be run in batches over a significant period of time, which can lead to differences between batches of several orders of magnitude, especially when changes in liquid chromatography columns are required or multiple instruments are used. Non-targeted approaches have several potential methods to estimate appropriate correction factors [11,12], which appears to work reasonably for metabolites with a low degree of missingness [13] but does not perform well when only a few data points per batch are available for normalization [14].…”

“…We previously used ratios between metabolite pairs in GWAS [8,10,15] and found that this approach could strengthen association signals significantly, partly because they implicitly correct for technical and biological variation, and partly because ratios approximate reaction rates between substrate and product pairs and therefore represent close proxies for enzyme activity, which is assumed to be affected by the genetic variants [16]. We also recently analyzed drug metabolites from non-targeted metabolomics and their overlap with self-reported medication in the Qatar Biobank (QBB) study [14], where we identified nine distinct acetaminophen metabolites that were detected together in over 520 study participants (Table 1).…”