DISCUSSION

This study analysed the metabolome of 43 healthy volunteers in order to identify novel endogenous biomarkers associated with CYP2D6 activity, using non-targeted metabolomics assays applied to urine and plasma samples. Drug-induced CYP2D6 inhibition and the search for CYP2D6 polymorphisms were taken into account for the identification of endogenous metabolites. Five endogenous metabolites could potentially act as probe for testing CYP2D6 activity as shown in Table 2 . All of them were significantly reduced after seven days of paroxetine intake compared to baseline. These results were reinforced and validated thanks to the significant reduction observed for the five candidates in the PM group compared to the EM-UM group. One of them had a similar mass-to-charge ratio to the urinary biomarker M1 (m/z 444.3102) reported earlier (Tay-Sontheimer et al., 2014). In their study, M1 was below detection limit in PM paediatric subjects and its levels were reduced by drug-induced inhibition of CYP2D6 in adults. However, only urine samples were tested and we hypothesise that their methodology might not have been sensitive enough to detect the four other compounds highlighted in our study since they have lower signals than m/z444.3108. Until now, they did not characterize the chemical structure ofm/z 444.3108. The formal identification of metabolites is one of the main challenges of untargeted metabolomics (Alonso et al., 2015; Ivanisevic and Thomas, 2018). Multiple databases have been developed and are regularly updated to help scientists with this process. But a large proportion of the compounds, which have been revealed by untargeted metabolomic profiling, remains unidentified (Blaženović et al., 2018). For structure elucidation, fragmentation mass spectra is required to improve confidence in metabolite identification (e.g. matching of an experimental MS/MS spectrum with a reference fragmentation spectrum) (Alonso et al., 2015). Analysis of isotope patterns is another mean for the determination of the most likely elemental composition of metabolites. SIRIUS 4 is a software that combines both isotope pattern and MS/MS data through fragmentation tree (Dührkop et al., 2019). It was successfully used within the framework of this project, allowing the elucidation of molecular formulas. The physicochemical properties of the five hits show that they are most likely lipids, one of them being a lipid glucuronide. More precisely, since they all contain a nitrogen element, they could potentially belong to fatty amides class. Interestingly, anandamide, a fatty acid amide that belongs to the class of endocannabinoids, is known to be metabolized by the CYP2D6 enzyme into 20-hydroxyeicosatetranoic acid ethanolamide as well as 5,6-, 8,9-, 11,12- and 14,15- epoxyeicosatrienoic acid ethanolamides as demonstrated through in vitro experiments using recombinant CYP2D6 (Farrell and Merkler, 2008; Snider et al., 2008). Furthermore, it is generally known that a majority of CYP2D6 substrates are lipophilic bases with a protonable nitrogen atom (Ingelman-Sundberg, 2005). In this work, metabolomics databases (i.e. METLIN and LIPID MAPS) revealed the following matches: N-linoleoyl dopamine for m/z 416.3159 and 17-phenyl trinor PGF2α diethyl amide for m/z 444.3108, two fatty acid amides compounds. However, MS/MS fragmentation patterns were not concordant. More in-depth analyses are required to achieve a detailed structure elucidation since the monoisotopic masses and MS/MS fragmentation patterns of the five hits appear to be unknown from the metabolomics databases used. Complete structure elucidation will require complementary analytical methods such as nuclear magnetic resonance (NMR) after extraction and purification of samples through preparative HPLC (Dias et al., 2016). Conveniently, all biomarker candidates are present in the urine. Large volumes of urine are relatively easy to obtain, which makes it an optimal starting material for purification and concentration for NMR structure identification (Whiley et al., 2019).
The targeted assay in PRM mode confirmed the results obtained from the untargeted LC-MS based metabolomics approach. It showed a significant decrease in the intensity of the features intensity after paroxetine intake compared to baseline. It is worth noting that signals from UM volunteers are often increased rather than decreased during the inhibition phase compared to baseline especially for m/z220.1543, 416.3159, 432.3108 and 444.3108 as seen in Figure 4a and 5a . In addition of being an inhibitor of CYP2D6, paroxetine is also a substrate for this enzyme. As demonstrated in numerous studies, UM subjects have lower concentrations of paroxetine than EM patients (Hicks et al., 2015). CYP2D6 inhibition may therefore be less strong in such participants and may account for inconsistent results. However, the UMRDEM/DOR does not display such pattern. It is then likely that the identification of the structure of endogenous substrates would allow refining these results by measuring the ratios of endogenous substrate/metabolite.
Even when using a more sensitive MS-based PRM experiment, we were still unable to detect signals in PM subjects for the features m/z416.3159 in plasma, 432.3108 in urine and 444.3108 in both. The same observation was reported for m/z 444.3108 (Tay-Sontheimer et al., 2014). Therefore, if certain metabolites are absent among this population group, it may explain differences in behaviour (e.g. impulsivity, anxiety) and disease susceptibility observed between PM and other individuals (Peñas-LLedó and LLerena, 2014). Additionally, log(area/creatinine) in urine and log(area) in plasma of the 5 endogenous compounds were both significantly correlated with CYP2D6 AS and log(UMRDEM/DOR). This confirms the potential ability of these compounds to accurately predict CYP2D6 activity. The best correlation with CYP2D6 AS was observed for m/z 597.3382 (rs = 0.710) in plasma (Figure 5d ). It is worth noting that the Spearman’s rank correlation coefficients measured in this study for all five hits were always lower than that observed for UMRDEM/DOR (rs = 0.791) (Figure 1 ). As previously explained, more significant correlations are expected once metabolites and substrates are fully identified, since measurement of the metabolic ratio between a specific substrate and a metabolite could be achieved similarly to UMRDEM/DOR. Additionally, the use of a normalized ratio (substrate/metabolite), could correct for potentially highly variable metabolite concentrations due to the influence of different factors of variability, such as circadian rhythm, diet, physiological or pathophysiological conditions. As an example, urinary excretion of 6β-hydroxycortisol, a CYP3A metabolite, varies considerably throughout the day, showing a strong diurnal rhythm. However, when normalized to cortisol, the variation is no longer significant within the day (Lee, 1995). Similarly, plasma 4β-hydroxycholesterol, another CYP3A metabolite, is ideally normalized to cholesterol in order to provide more robust data (Mao et al., 2017; Aklillu et al., 2020).
Once the couple metabolite/substrate is fully identified and quantifiable, specificity for CYP2D6 enzyme should be assessed using, for example, recombinant enzyme assays in order to exclude the influence of other metabolic pathways. In this study, no correlation with CYP2C9, CYP2C19 and CYP3A genotypes were observed with the identified biomarkers. In addition, no impact of oral contraceptives (moderate CYP1A2 and CYP3A inhibitor) (Samer et al., 2013) was observed on mean relative intensities of the five endogenous CYP2D6 metabolites (data not shown). But these results provide only some preliminary insights into the CYP2D6 specificity and further in vitro studies should be performed to fully validate specificity of the metabolic pathway.
In conclusion, non-targeted metabolomics enabled the identification of five potential endogenous CYP2D6 metabolites presumably related structurally to the class of fatty amides, including a glucuronide compound. Each of these candidate biomarkers could map the functionality of this enzyme. Further studies will focus on complete structure elucidation using complementary analytical method such as NMR. Once identified and validated, noninvasive prediction of CYP2D6 activity based on these candidates could greatly improve current phenotyping strategies by being readily available at any time and completely bypassing the need of administering exogenous components and thus the risk of adverse events.