METHODS

Study design and population

This study protocol (NCT04188028) was approved by the Geneva Research Ethics Committee and the Swiss Agency for Therapeutic Products (Swissmedic). All participants provided written informed consent before inclusion. Protocol conception and trial conduct were performed in accordance with the Declaration of Helsinki ethical principles and the Good Clinical Practice guidelines of the International Congress of Harmonization.
Inclusion criteria were the following: age between 18-65 years, body mass index between 18-27 kg m-2, CYP2D6 genotype activity score (AS) = 0 or ≥ 1 (Gaedigk et al., 2008), reliable contraception during the whole study, including a barrier method.
Exclusion criteria included: pregnancy/breastfeeding, any pathology, drug or food affecting CYP activity, tobacco consumption (≥ 10 cigarettes/day), alcohol intake 2 days prior to session 1 and during paroxetine intake, hepatic impairment, medical history of chronic alcoholism or abuse of psychoactive drugs, regular use of psychotropic substances, drug sensitivity, psychiatric disorders, and Beck Score ≥ 10 (question related to suicide >0).
Forty-three healthy volunteers participated in this study, which was conducted in two sessions. Each session included the oral administration of 5 mg dextromethorphan (DEM) (BEXIN syrup, Spirig Healthcare, Egerkingen, Switzerland) to participants after an overnight fast and urine collection for 4 hours following the administration of DEM for CYP2D6 phenotyping. For metabolomic analyses, prior to DEM administration, urine samples were also collected over a full 24-hour period and venous blood samples were collected in tubes containing EDTA (BD Vacutainer, Plymouth, UK) immediately before DEM ingestion. Breakfast was served 1 hour after DEM intake. At session 2, the study course was similar but participants were asked to take 20 mg (10 mg for poor metabolizer (PM) subjects) of paroxetine (PAROXETIN‐MEPHA, Basel, Switzerland), a time-dependant inhibitor, every morning for one week (7 doses in total) with the breakfast (Storelli et al., 2019). Participants were specifically asked about the time at which paroxetine tablets were taken and were asked to bring back empty blister packs to verify compliance. For women participating in the study, a pregnancy test was performed at inclusion and at each session prior to any medication administration. Plasma was obtained through centrifugation at 2,750g for 10 min. All blood and urine samples were stored at −80°C until analysis.

Quantification of dextromethorphan and dextrorphan

Subsequent to chemical hydrolysis and liquid-liquid extraction (Daali et al., 2008), DEM and dextrorphan (DOR) were quantified in urine by liquid chromatography-tandem mass spectrometry (Sciex, Darmstadt, Germany). CYP2D6 phenotype was determined based on the urinary metabolic ratio dextromethorphan to dextrorphan (UMRDEM/DOR) as follows: PM phenotype (UMRDEM/DOR ≥ 0.3), intermediate metabolizer (IM) phenotype (UMRDEM/DOR between 0.03-0.3), extensive metabolizer (EM) phenotype (UMRDEM/DOR between 0.003-0.3) and ultrarapid metabolizer (UM) phenotype (UMRDEM/DOR < 0.003) (Gaedigk et al., 2008).

CYP450 genotyping

Genomic DNA was extracted from whole blood (200 µl) using the QIAamp DNA Blood Mini Kit (Qiagen, Hombrechtikon, Switzerland). Fourteen CYP2D6 allelic variants were screened using the TaqMan® OpenArray® PGx Panel (Thermo Fisher Scientific, Waltham, USA) performed on the QuantStudio 12K Flex real‐time PCR system in compliance with the manufacturer’s instructions. The following mutations were considered: 2850C>T, 4180G>C, 2549delA (*3), 100C>T (*4, *10), 1846G>A (*4A). 1707delT (*6), 2935A>C (*7). 1758G>T (*8), 2613_2615delAGA(*9), 124G>A (*12), 1758G>A (*14). 1023C>T (*17), 3183G>A (*29), 2988G>A (*41). Regarding CYP2C9 the following mutations were measured: 430C>T, 3608C>T (*2), 1075A>C, 42614A>C (*3). CYP2C19*2 (681G>A, 19154G>A), CYP2C19*3 (636G>A, 17948G>A) and CYP2C19*17 (806C>T) were also determined, as well as CYP3A4*22 (15389C>T) and CYP3A5*3 (6986A>G).
CYP2D6 Taqman® Copy Number Assay (assay ID: Hs00010001_cn targeting exon 9, Applied Biosystems, Foster City, USA) was performed on a 7900HT Fast Real-Time PCR System (Applied Biosystems, Thermo Fisher Scientific, CA, USA) instrument for the detection of gene deletion (*5 allele) and duplication.
CYP2C9, CYP2C19, CYP2D6 and CYP3A activity scores were assigned using previously developed scoring system (Gaedigk et al., 2008; Elens et al., 2011; Karnes et al., 2020; Lima et al., 2020). Values of 0, 0.5, 1 and 1.5,were assigned to the non-functional, reduced function, fully functional and gain-of-function alleles, respectively (Tay-Sontheimer et al., 2014). In the case of CYP2D6, the values for alleles with two or more gene copies are multiplied by the number of gene (Gaedigk et al., 2018). Summing the values of the two alleles gives the AS of a genotype (Gaedigk et al., 2018).

Untargeted metabolomics analysis by LC-HRMS

300 µL of methanol/ethanol (50:50) containing hydrocodone-D6 and phenobarbital-D5 100 ng/mL (internal standards for positive and negative modes, respectively) were added to 100 µL of urine or plasma for protein precipitation. Samples were centrifuged for 20 min at 16,000g. The supernatant was then evaporated under a stream of nitrogen, reconstituted in 100 µL of 10% methanol and 5 μL were injected into the LC-MS system.
Non-targeted metabolomics analyses were carried out using a LC system Ulimate 3000 coupled to a Q Exactive Plus system (Thermo Scientific Fisher, Bremen, Germany).(Forchelet et al., 2018; Kowalczuk et al., 2018) Separation was performed with a Kinetex® C18 column (50 × 2.1 mm, 2.6 µm) from Phenomenex (Brechbühler, Switzerland) with mobile phases consisting of water (A) and methanol (B) both containing 0.1% formic acid. The flow rate was fixed at 0.3 mL/min over 13 minutes. Gradient program was set as follows: 2% B (0-0.3 minutes), 2-98% B (0.3-6 minutes), 98-100% B (6-9 minutes), 100-2% B (9-9.1 minutes), and 2% B (9.1-13 minutes). Quality controls (i.e. pooled aliquots of all clinical study samples) were included in the analytical sequence at regular intervals. Data was acquired in a full scan mode in both positive and negative polarities. The parameters were set as follows: the capillary voltage at 3.2 kV and 2.5 kV in positive and negative mode, respectively, sheath and auxiliary gas flow rate at 40 and 10 respectively, capillary temperature at 320 °C and S-lens RF level at 50.

Untargeted metabolomics Data and Statistical Analysis

The raw UPLC-HRMS files were converted to .mzXML format using MSConvert (ProteoWizard 3.0, http://proteowizard.sourceforge.net/) and pre-processed using the XCMS Online platform for features detection, chromatogram alignment, isotope annotation and data visualization (https://xcmsonline.scripps.edu).
All data transformation and statistical analyses were performed using MetaboAnalyst (https://www.metaboanalyst.ca/). Data were sum-normalized, Pareto-scaled and log-transformed. Subsequently, features were filtered, and only those with a CV less than 20% in the QC samples were selected. Isotopes were filtered out and finally (Kim et al., 2018), ions of zero intensity in >20% of all participants in both sessions were excluded.
Zero values were replaced by the half of the minimum value found for the corresponding hit (Xia and Wishart, 2011). Principal Component Analysis (PCA) was performed using QC samples to assess performance and stability of the system. Data Volcano plots were generated in order to filter metabolites that displayed both significant fold changes (≥ 1.5 or ≤ 0.67) and statistical significance (FDR adjusted P -value < 0.05) between the control and the inhibition session in non-PM subjects (n = 37). The significant features obtained were then filtered out according to genotype: fold changes of relative intensity in the CYP2D6 EM-UM group (n = 37) compared to the PM group (n = 6) ≤ 0.67 or ≥ 1.50 (P -value < 0.05). The data and statistical analysis comply with recommendations of the British Journal of Pharmacology on experimental design and analysis (Curtis et al., 2018).

CYP2D6 biomarkers identification

Metabolites molecular formulas were investigated further using MS fragmentation and isotope pattern analysis with SIRIUS 4.0.1 (Dührkop et al., 2019). The glucuronide metabolites were enzymatically deconjugated prior to MS fragmentation using the β-Glucuronidase /Arylsulfatase mixture from Roche Diagnostics (Mannheim, Germany) in sodium acetate buffer (1 M, pH 5.0) at 37°C (Schmidt et al., 2013). Main metabolomics databases: LIPID MAPS® (https://www.lipidmaps.org/), METLIN (https://metlin.scripps.edu/) and HMDB (http://www.hmdb.ca/) were then used to assist in the identification of molecular structures of significant features on the basis of the available experimental data (i.e. exact molecular weights, molecular formulas, fragmentation patterns).

Production Reaction Monitoring analysis

To improve sensitivity, validate and refine results, a semi-quantitative method using HRMS-based PRM was developed. The chromatographic separation was performed using a LC system Vanquish coupled to a Q Exactive Focus system (Thermo Scientific, Bremen, Germany). The preparation of urine and plasma samples as well as the chromatographic and mass spectrometry conditions were identical to those of the metabolomics analyses, except that the extracts were concentrated twice (reconstitution in 50 µL of 10% methanol). Hydrocodone-d6 at 15 ng/mL was used as internal standard. The resolution was set at 17′500 for the fragmentation experiments with an AGC target of 5e4 and a maximum IT of 100 ms. The NCE values for each compound were set individually and urinary creatinine concentration was used for data normalization.

Production Reaction Monitoring Data and Statistical analysis

Comparisons of two dependent and independent groups were performed using paired and unpaired t test (two-tailed), respectively. Measures of associations were established using Spearman’s rank correlation. The statistical analyses were performed using GraphPad Prism 8.0.1 software (San Diego, USA). A P value below 0.05 was considered statistically significant. The data and statistical analysis comply with recommendations of the British Journal of Pharmacology on experimental design and analysis (Curtis et al., 2018).

Nomenclature of Targets and Ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Harding et al., 2018), and are permanently archived in the Concise Guide to PHARMACOLOGY 2019/20 (Alexander et al., 2019).

RESULTS

Healthy subjects

A total of 43 healthy subjects were enrolled (Table 1 ). Forty-two completed the study while one among the PM subjects only attended the control session. The mean age was 24 (range 19-29) with a slightly higher proportion of females (55.8 %, n = 24). Among women participants, 45.8% used oral contraceptive pill (n = 11).

CYP2D6 genotype and phenotype

Based on genotyping analyses, 6 participants were classified as PM subjects (genetic‐predicted activity score (gAS) = 0), 33 as EM subjects (1 ≤ gAS ≤ 2) and 4 as UM subjects (gAS > 2) (Gaedigk et al., 2008; Crews et al., 2014). UMRDEM/DOR were measured to establish CYP2D6 phenotype. As illustrated in Figure 1a , mean CYP2D6 activity was significantly reduced after paroxetine intake compared to control session (P < 0.0001), demonstrating an effective inhibition of CYP2D6 activity by paroxetine. In addition, a significant Spearman’s rank correlation coefficient ofr s = 0.791 (P < 0.0001) was found between the logarithm of UMRDEM/DOR and AS groups (Figure 1b ).

Untargeted metabolomics analysis

Using untargeted metabolomics assays, the goal of this project was to identify biomarkers of CYP2D6 in urine and plasma reflecting the activity of the enzyme. Figure 2 shows the flowchart of the data analysis, from data extraction to statistical analysis. After the filtering steps, 8926 and 5997 ions in plasma and urine, respectively, were processed for statistical analysis, including sum-normalisation, log-transformation and Pareto scaling. PCA scores plot revealed a tight clustering of QC samples (Supplementary Figure S1 ) indicating high experimental quality for both urine and plasma samples.
As seen in Table 2 , five endogenous metabolites were significantly decreased in urine and/or plasma during the CYP2D6 inhibition phase relative to baseline with the following m/z in positive mode: 220.1545, 416.3159, 432.3108, 444.3108 and 597.3382 (595.3236 in negative mode). The largest reduction was observed form/z 597.3382 (0.13-fold). In parallel, their intensity was significantly lower in PM subjects than in EM-UM volunteers (fold changes ≤ 0.67). These observations strongly imply that these five features are metabolites produced via the CYP2D6 enzyme.

Structure identification

All MS/MS fragmentation patterns are shown in Figure 3 . Due to the presence of a glucuronide moiety, the feature m/z 597.3382 was also enzymatically hydrolysed prior to MS/MS fragmentation, resulting in m/z 421.3061 (-176.0321 Da).
MS/MS fragmentation of m/z 220.1543 was difficult to obtain because several compounds with close mass (i.e. m/z 220.0966, 220.1329) co-elute. At 10 eV it is however possible to observe four losses of water (-18.0109 Da): 202.1436, 184.1331, 166.1226 and 148.1120.
The compounds m/z 416.3159, 432.3108 and 444.3108 have identical fragmentation patterns, with a major fragment at 98.0967. The skeletal structure appears therefore similar for these three compounds, which differ by one or two atoms: carbon loss between 444.3108 and 432.3108 (-12.0000 Da), oxygen loss between 432.3108 and 416.3159 (-15.9949 Da), and carbon-oxygen loss between 444.3108 and 416.3159 (-27.9949 Da). Regarding the feature m/z 444.3108, we observed the same MS/MS fingerprint described by Tay-Sontheimer et al. (fragment ions atm/z: 98.0967, 370.2733, 206.1900, 56.0501, 55.0549, 150.1275 and 81.0703) (Tay-Sontheimer et al., 2014).
As described above, one of the major fragments of m/z 597.3382 is 421.3061, which corresponds to neutral loss of a glucuronide moiety. Chromatograms before and after hydrolysis shown in Supplementary Figure S2 confirm the presence of a glucuronide since the conjugated peak (m/z 597.3382) decreases and the deconjugated peak (m/z 421.3061) increases after hydrolysis.
Molecular formula for all compounds was obtained through fragmentation and isotope pattern analysis using SIRIUS 4.0.1. Results are described in Table 3. Fragmentation trees are presented inSupplementary Figure S3-S8 . Interestingly, all the features contain one or two nitrogen atoms. Using exact mass, mass spectral databases were employed for potential features annotation and identification as shown in Table 3 . Results were, however, inconclusive in HMDB (no results for the query masses). In LIPID MAPS, only m/z 416.3159 showed a potential match: N-linoleyl dopamine. Nonetheless, this lipid normally has a characteristic and predominant fragment at 137, which was not observed, making this finding unlikely (Thomas et al., 2009). In METLIN, the feature m/z 444.3108 matches a prostaglandin derivative: 17-phenyl trinor PGF2α diethyl amide. However, MS/MS fragmentation patterns are not concordant.
Based on these findings and the physicochemical properties of these compounds, in particular their high retention times and molecular weight, it is likely that they belong to lipid species, including a lipid glucuronide.

Relative quantification using a PRM method

MS parameters were first optimized to enhance data quality using parallel reaction monitoring (PRM) mode. In order to obtain the optimal collision energy (CE) for each metabolite, the NCE were considered individually using a step of 10. The results are presented inTable 4 . All analytes were detectable in the urine samples of EM and UM subjects, with the exception of PM volunteers regardingm/z 432.3108 and 444.3108. Metabolites corresponding tom/z 220.1543 and 432.3108 were not measurable in any of the plasma samples. Also, metabolites with m/z 416.3159 and 444.3108 were not detectable in plasma samples of PM participants.
The results confirmed the significant down-regulation observed after paroxetine intake compared to the control session for all five hits in both urine and plasma samples (Figure 4a and 5a ). Likewise, the compounds were not detectable or significantly down-regulated in PM subjects compared to EM-UM participants(Figure 4b and 5b ). Going even further, significant correlations were observed between log(area) and log(UMRDEM/DOR) (Figure 4c and 5c ). Significant correlations between log(area) and CYP2D6 activity score were also described, with UM subjects (gAS > 2) having the highest mean intensities (Figure 4d and 5d ). AllP were < 0.05.
Mean relative intensity of m/z 220.1543, 416.3159 and 597.3382 was unchanged in PM subjects after paroxetine intake compared to baseline, indicating that changes are solely due to CYP2D6 inhibition (Figure 6) . Indeed, PM individuals do not express the enzyme CYP2D6. Therefore, a down-regulation in PM subjects would indicate the presence of false positives rather than changes due to CYP2D6 inhibition.
No significant correlation was observed with any CYP450 activity score other than CYP2D6 (Supplementary Table S1).