The American Psychiatric Association (APA) has updated its Privacy Policy and Terms of Use, including with new information specifically addressed to individuals in the European Economic Area. As described in the Privacy Policy and Terms of Use, this website utilizes cookies, including for the purpose of offering an optimal online experience and services tailored to your preferences.

Please read the entire Privacy Policy and Terms of Use. By closing this message, browsing this website, continuing the navigation, or otherwise continuing to use the APA's websites, you confirm that you understand and accept the terms of the Privacy Policy and Terms of Use, including the utilization of cookies.

×

Abstract

Objective:

Clozapine is the only effective medication for treatment-resistant schizophrenia, but its worldwide use is still limited because of its complex titration protocols. While the discovery of pharmacogenomic variants of clozapine metabolism may improve clinical management, no robust findings have yet been reported. This study is the first to adopt the framework of genome-wide association studies (GWASs) to discover genetic markers of clozapine plasma concentrations in a large sample of patients with treatment-resistant schizophrenia.

Methods:

The authors used mixed-model regression to combine data from multiple assays of clozapine metabolite plasma concentrations from a clozapine monitoring service and carried out a genome-wide analysis of clozapine, norclozapine, and their ratio on 10,353 assays from 2,989 individuals. These analyses were adjusted for demographic factors known to influence clozapine metabolism, although it was not possible to adjust for all potential mediators given the available data. GWAS results were used to pinpoint specific enzymes and metabolic pathways and compounds that might interact with clozapine pharmacokinetics.

Results:

The authors identified four distinct genome-wide significant loci that harbor common variants affecting the metabolism of clozapine or its metabolites. Detailed examination pointed to coding and regulatory variants at several CYP* and UGT* genes as well as corroborative evidence for interactions between the metabolism of clozapine, coffee, and tobacco. Individual effects of single single-nucleotide polymorphisms (SNPs) fine-mapped from these loci were large, such as the minor allele of rs2472297, which was associated with a reduction in clozapine concentrations roughly equivalent to a decrease of 50 mg/day in clozapine dosage. On their own, these single SNPs explained from 1.15% to 9.48% of the variance in the plasma concentration data.

Conclusions:

Common genetic variants with large effects on clozapine metabolism exist and can be found via genome-wide approaches. Their identification opens the way for clinical studies assessing the use of pharmacogenomics in the clinical management of patients with treatment-resistant schizophrenia.

Schizophrenia affects approximately 0.7% of the population (1), is characterized by disturbances in cognition, emotion, perception, and thought, and has a severe impact on quality and length of life (2). Around a third of patients experience treatment-resistant schizophrenia, a form of the disorder marked by severe functional impairment in which symptoms fail to respond adequately to at least two first-line antipsychotics (3). Clozapine is the most effective treatment (4) and the only medication approved for treatment-resistant schizophrenia. Despite extensive evidence supporting its effectiveness, clozapine remains underprescribed worldwide, including in countries with highly developed health services (5, 6). A key factor limiting clozapine’s use is its potential to induce severe adverse drug reactions, including agranulocytosis, which occurs in up to 1% of patients and necessitates regular hematological monitoring (7). Other adverse drug reactions, such as seizures, tachycardia, sedation, weight gain, and hypersalivation, have been associated with either clozapine dosage or plasma concentration (8).

Clinicians routinely use clozapine levels to assess adherence and guide dosage in the management of both therapeutic response and side effects (9). This is an important strategy, as adverse effects are the primary reason for clozapine discontinuation (10). However, there is high interindividual variability in plasma clozapine concentrations at given dosages (11), often as a result of the effects of concomitant medications (12), which complicates titration and presents challenges for any research that aims to assess the relationship between dosage, efficacy, and adverse drug reactions. Guidelines for therapeutic drug monitoring indicate that clozapine plasma concentrations in the range 0.35–0.60 mg/L are optimal for response (13), and concentrations higher than 0.60 mg/L have been linked to serious adverse drug reactions. Dose-response relationships between clozapine concentration and weight gain (14) or sedation (15) have been suggested, although this has not been seen for all adverse drug reactions (8). The accurate prediction of clozapine plasma levels therefore has important clinical implications. Sophisticated models incorporating lifestyle habits and metabolic indicators can explain up to 48% of the variance in clozapine levels in large patient samples (16, 17), but no individual factors other than age, smoking habits, and sex have been found to be of clinical value (11).

The use of genetic approaches to identify biomarkers that explain individual variability in drug metabolism and response is the basis of the field of pharmacogenomics. Although this discipline experienced strong growth in the mid-2000s, fostered by successes related to cardiology and oncology (18), translation of these results into clinical settings has proved challenging (19). This has been particularly true in psychiatry (20), and clozapine research in this area showcases some of the difficulties of carrying out robust pharmacogenomics studies. Clozapine is metabolized by the liver (21) with first-pass metabolism, driven primarily by the CYP1A2 enzyme, producing norclozapine (N-desmethylclozapine), a pharmacologically active compound that can reach up to 90% of the circulating concentration of clozapine (22, 23). Other metabolites have been identified, such as clozapine-N-oxide, formed by CYP3A4 (22), and N-glucuronides, which are secondary and tertiary metabolites produced by the UDP-glucuronosyltransferase (UGT) protein superfamily (24). These enzymes are all implicated in other drug metabolic pathways (25, 26) and have been the focus of the search for genetic variants associated with clozapine plasma concentrations. Previous studies have examined candidate polymorphisms in small (usually N<100) samples (27, 28). Promising results have been reported for variants of the ABCB1 drug transporter (28), but no finding has yet passed the threshold of genome-wide significance, which is now widely accepted as being required for robust association, even in candidate gene studies (29).

Here we report the first genome-wide association study (GWAS) of plasma concentrations of clozapine and its metabolites. By applying modern statistical modeling techniques, we exploited information from over 10,000 metabolite concentration assays taken from a sample of nearly 3,000 patients with treatment-resistant schizophrenia. We identified genome-wide significant polymorphisms that delineate clozapine’s metabolic pathways, and we discuss their relevance to the clinical management of treatment-resistant schizophrenia.

Methods

Sample

Data were acquired as part of the CLOZUK2 study (30) from individuals for whom clozapine was prescribed for treatment-resistant schizophrenia in the United Kingdom. Sample and data acquisition were arranged through collaboration with Leyden Delta (Nijmegen, the Netherlands), which monitors clozapine in the United Kingdom. The study was conducted in accordance with its U.K. National Health Service ethics permissions. The CLOZUK2 sample has been described fully elsewhere (30).

Genotyping and Imputation

Genotyping of the CLOZUK2 sample was performed by deCODE Genetics (Reykjavik, Iceland), using an Illumina HumanOmniExpress-12 array. Quality control and analyses were performed using PLINK, version 1.9 (31), unless otherwise specified. Quality control followed standard GWAS protocols (32), including the removal of samples and markers with >2% missingness or homozygosity (F>0.2). After quality control, 7,287 individuals with data from 698,442 markers remained in the data set.

The Haplotype Reference Consortium panel, accessed through the Michigan Imputation Server (33, 34), was used for genotype imputation. Because using this service to impute X chromosome data was not possible at the time of this study, genotype data from the X chromosome were imputed locally using the Cardiff University RAVEN cluster (35). For this, the SHAPEIT/IMPUTE2 algorithms (36) and a combination of the 1000 Genomes phase 3 (1KGPp3) and UK10K reference panels (37) were used. These two approaches have been shown to perform similarly when imputing GWAS variants (33), which traditionally have minor allele frequencies (MAFs) >1%. After imputation, 20 million single-nucleotide polymorphisms (SNPs) with imputation quality scores (MaCH r2 or INFO) >0.8 remained in the data set.

Selection of Individuals for Analysis

In order to select a sample of individuals with homogeneous genetic ancestry, we selected a custom panel of ancestry informative markers, as described previously (38). Briefly, using principal components and a classification algorithm based on linear discriminant analysis, we identified 5,900 CLOZUK2 individuals with >90% probability of European ancestry. Individuals who did not fulfill this criterion were excluded from all further analyses, given the small number of non-European individuals with plasma concentration data.

From the imputed genotypes, we retained SNPs with MAFs ≥1% and Hardy-Weinberg equilibrium p values ≤10−6, leaving 7.5 million SNPs for analysis.

Clozapine and Norclozapine Levels in Blood

Clozapine and norclozapine plasma concentration assays were conducted at Magna Laboratories (Ross-on-Wye, U.K.) and were determined by liquid chromatography mass spectrometry using standard procedures for clozapine assays, as summarized in the Supplementary Methods section of the online supplement.

Curation of Plasma Concentration Data

The clozapine and norclozapine plasma concentrations of the CLOZUK2 sample formed a data set of assays taken at 15,504 time points from 3,986 unique individuals. Data available for each time point included age of patient, daily clozapine dose, date and time of last dose, date and time the blood was sampled, and measured clozapine and norclozapine levels. Clozapine or norclozapine concentrations <0.05 mg/L (corresponding to the minimum detection thresholds of the high-performance liquid chromatography instrument, and indicating nonadherence) were removed. We also excluded outliers outside the 99th percentile of plasma concentrations. We noted that removing data from a broader range of the extremes of the plasma concentration distributions did not meaningfully alter our results. Finally, we removed assays for which the blood was sampled <6 hours or >24 hours since the last clozapine dose taken, because a 6- to 24-hour postdose measurement interval is recommended for clozapine monitoring to ensure adequate drug absorption (11). After this process, 10,353 assays from 2,989 individuals (range, 1–42 per individual) remained (see Figure S1 in the online supplement). Overall, 2,022 individuals (67.8% of the total) had assay data from more than one time point. We did not undertake analyses specifically on steady-state levels (39), given that we could not strictly determine this in the available data, but we noted that when we restricted our analyses to patients for whom more than one clozapine measurement was available (and hence who were more likely to have taken the medication for a longer period), the results were unchanged; indeed, the data from the majority of our patients with more than one assay measurement spanned a much longer period (median time, 361 days).

Generation of Plasma Concentration Phenotypes

We examined three primary metabolic outcome variables: the plasma concentrations of clozapine and norclozapine and the ratio of clozapine to norclozapine (the “metabolic ratio”) (16). To make maximum use of the data available at multiple time points, we employed a regression modeling framework to combine data from multiple assays into a single phenotype per individual, as detailed in the Supplementary Methods section of the online supplement. Briefly, we identified the best-fitting distribution for each metabolic outcome variable and used this to specify a random-effects model controlling for known predictors of clozapine metabolism (clozapine dose, time between dose and assay, and age at assay). From this model, random-effects coefficients were extracted for each individual; this corresponds to the variation in plasma concentrations for that individual (clozapine, norclozapine, and their ratio), independent of the effects of the known predictors (dose, time since dose, and age); this coefficient was then used as our primary outcome phenotype for our GWAS.

GWAS of Plasma Concentrations

The approach described above that we used to derive our primary outcome allowed us to use standard methods to perform a GWAS of clozapine concentration, norclozapine concentration, and their metabolic ratio. We undertook the GWAS on the CLOZUK2 imputed data, using the “leave-one-chromosome-out” linear mixed model (LMM-LOCO) implemented in GCTA (Genome-wide Complex Trait Analysis), version 1.26 (40). This analysis requires genotype relatedness matrices to control for family and population structure, which we calculated from nonimputed genotypes to avoid introducing biases due to imputation accuracy. Sex was used as a covariate for this analysis. (Summary statistics from these GWASs are available for download at http://walters.psycm.cf.ac.uk/.)

Because we did not have information on other factors known to influence drug plasma concentrations, such as weight or cigarette smoking habits, we performed secondary sensitivity analyses controlling for proxy measures based on polygenic risk scores (PRSs) for those traits (see the Supplementary Methods section of the online supplement). From each LMM-LOCO analysis, we identified approximately independent index SNPs (r2=0.1) using the PLINK linkage disequilibrium (LD) clumping procedure (p<10−4 and distance <3000 kb).

Identification of Putatively Causal SNPs and Genes

For each genome-wide significant locus (p<5×10−8), FINEMAP, version 1.1 (41), was used to pinpoint putatively causal SNPs. These were defined as individual SNPs with a posterior probability (PPFINEMAP) higher than 95%. In the absence of such SNPs, a list of credible SNPs was compiled, which included those with a cumulative PPFINEMAP of 95%. Sets of credible SNPs were generated using a setting of one expected causal SNP per locus (k=1), and annotated to function using the Ensembl Variant Effect Predictor tool (42). To attempt a more accurate identification of putatively causal genes in these loci, we also analyzed gene expression from multiple tissues, including liver, using data from GTEx, version 7 (43) (see the Supplementary Methods section of the online supplement). Information about hepatic promoters, enhancers, and topologically associated domains was retrieved from Schmitt et al. (44) and added to the SNP-based annotations.

Estimating the Effect of Individual SNPs on Plasma Concentration

The modeling we undertook to derive our primary GWAS phenotype produces genetic effect sizes in LMM-LOCO that are related to the residuals used as our primary outcome rather than to the raw assay data, which would be easily interpretable. To explicitly estimate genetic effects on the scale of our clozapine metabolite plasma concentrations, we extracted the minor allele counts (“allelic dosages”) for genome-wide significant SNPs of each GWAS for each individual. Mixed regression models were fitted for each outcome variable, including as covariates clozapine dose, time since dose, age, sex, allelic dosage, and the first 20 principal components derived from the genotype data using PC-AiR (45). A random-effect covariate was used to capture individual-level variance in this model. The effect of allelic dosage on plasma concentration was then estimated within the same mixed linear regression framework used to generate the GWAS phenotypes, which produces a more meaningful result from a clinical point of view. Model-fitting statistics (e.g., variance explained by fixed and random effects, proportion of variance explained by single SNPs) were determined as described in the Supplementary Methods section of the online supplement.

Locating Shared Associations With Other Metabolic Traits

We used gwas-pw, version 0.21 (46), to identify genetic markers associated with clozapine and norclozapine concentrations that were also associated with the concentrations of other metabolites in the KORA/TWINSUK study (47) in a genome-wide context. We focused on the 285 metabolites and xenobiotics confidently identified in that study, disregarding unknown compounds and metabolite ratios. Using the summary statistics from a GWAS of each metabolite, we generated co-localization posterior probabilities (PPCOLOC) with our clozapine and norclozapine summary statistics. Probabilities were calculated for all individual SNPs with complete data, to identify shared effects inside and outside genome-wide significant loci, thus allowing us to pinpoint more robust shared effects than would emerge from examining a limited number of loci.

Analysis of Human Metabolic Pathways

To study the genetic component of clozapine metabolism in the context of the human metabolic network, we retrieved the most recent metabolome reconstruction, RECON 2.2, capturing 5,324 metabolites and 1,675 genes (48). We grouped genes into subsystems (e.g., “extracellular transport,” “steroid metabolism”), resulting in 79 gene sets. One additional set was created from 203 genes analyzed in a recent drug metabolism study (49), representing known pharmacokinetic-relevant enzymes and receptors. Gene set enrichment analysis was performed with MAGMA, version 1.06 (50), using the “multi” method to calculate gene-wide p values from GWAS summary statistics. Within each analysis, gene set p values were corrected using the family-wise error rate with 100,000 permutations.

Results

Genome-Wide Significant SNPs Associated With Clozapine Plasma Concentrations

The GWAS of clozapine levels identified a single genome-wide significant association at rs2472297, an intergenic variant between CYP1A1 and CYP1A2 (Figure 1A; Table 1; see also Table S1 and Figure S2 in the online supplement). Analysis of GTEx hepatocyte expression data did not relate this signal to any particular gene (see the Supplementary Methods section of the online supplement), although rs2472297 has previously been associated with CYP1A2 activity on the basis of its effect on caffeine metabolite concentrations (51). In the mixed-model analysis, the minor allele of this variant was shown to be associated with reduced clozapine plasma concentrations, with a proportion of variance explained (PVE) of 1.47% (Table 2, Figure 2; see also the Supplementary Methods section). Model-fitting statistics for the complete groups of fixed and random effects are listed in Table S2 in the online supplement.

FIGURE 1.

FIGURE 1. Manhattan plots of the genome-wide association studies of clozapine and norclozapine levels and their metabolic ratio

TABLE 1. Association statistics of the index SNPs for each phenotype and LD-independent locus and for fine-mapped missense variants in high LD with each index SNPa

Phenotype and LocusSNPAlleleAnnotationGWAS p
Clozapine
 chr15:74817689–75404506rs2472297TIntergenic4.35×10–10
Norclozapine
 chr4:69542100–70312793rs11725502TIntergenic5.47×10–15
 chr4:69542100–70312793rs61750900TMissense8.91×10–15
 chr2:234611523–234676118rs2011425GMissense8.37×10–9
Metabolic ratio
 chr4:69542100–70387482rs10023464TIntergenic8.72×10–66
 chr4:69542100–70387482rs61750900TMissense1.69×10–64
 chr10:96098093–96974830rs12767583TIntronic4.64×10–14
 chr10:96098093–96974830rs1126545TMissense1.02×10–13

aGWAS=genome-wide association study; LD=linkage disequilibrium; SNP=single-nucleotide polymorphism.

TABLE 1. Association statistics of the index SNPs for each phenotype and LD-independent locus and for fine-mapped missense variants in high LD with each index SNPa

Enlarge table

TABLE 2. Effect sizes of genetic, demographic, and clinical covariates as estimated with linear mixed regression modelinga

Phenotype and CovariateBetaSEp
Clozapine
 rs2472297 (T)–0.0890.0132.40×10–11
 Clozapine daily dose (mg)0.0024.03×10–5<1×10–300
 Time since last dose (hours)–0.0090.0028.11×10–6
 Patient age (years)0.0040.0040.384
 Patient sex (male)–0.1470.0191.31×10–14
Norclozapine
 rs61750900 (T)–0.1490.0183.17×10–17
 rs2011425 (G)–0.1120.0193.34×10–9
 Clozapine daily dose (mg)0.0023.67×10–5<1×10–300
 Time since last dose (hours)6.81×10–40.0020.701
 Patient age (years)–0.0030.0040.444
 Patient sex (male)–0.1200.0173.61×10–12
Metabolic ratio
 rs61750900 (T)0.2120.0125.01×10–70
 rs1126545 (T)0.0780.0105.96×10–14
 Clozapine daily dose (mg)–1.49×10–42.40×10–55.22×10–10
 Time since last dose (hours)–0.0140.0011.03×10–33
 Patient age (years)0.0070.0030.006
 Patient sex (male)–0.0160.0120.164

aAll models also included age squared and 20 genotype principal components as fixed effects (omitted).

TABLE 2. Effect sizes of genetic, demographic, and clinical covariates as estimated with linear mixed regression modelinga

Enlarge table
FIGURE 2.

FIGURE 2. Effect of the rs2472297 genotype on clozapine plasma levels, at different daily clozapine dosesa

a For this analysis, only the last time point of each individual in the CLOZUK2 sample was used. For each interval of daily clozapine dose, average plasma concentrations and standard deviations are shown. Values inside the central point represent the number of individuals within each genotype/interval category.

The GWAS of norclozapine levels identified two genome-wide significant loci (Figure 1B; Table 1; see also Table S1 and Figure S3 in the online supplement). The first was indexed by rs72846859, an intergenic variant upstream of UGT2B10. FINEMAP revealed a complex association signal in this region, with 171 credible SNPs (see Table S3 in the online supplement), including a missense variant (Asp/Tyr) in UGT2B10, rs61750900. LD between the index and missense variants was high (r2=0.964), and given its higher prior probability of causality (see the Supplementary Methods section), we incorporated rs61750900 into the mixed regression model of norclozapine plasma levels. For the second genome-wide significant locus, the index SNP was rs2011425, a missense variant (Leu/Val) in UGT1A4, which also obtained the highest FINEMAP probability of 47 credible SNPs (see Table S3 in the online supplement). The minor alleles of both SNPs were associated with lower norclozapine plasma levels, with a PVE of 2.32% for rs61750900 and 1.15% for rs2011425 (Table 2).

The GWAS of the clozapine-to-norclozapine metabolic ratio identified three independent genome-wide significant associations at two distinct loci (Figure 1C; Table 1; see also Table S1 and Figure S4 in the online supplement). Two of these LD-independent SNPs (rs10023464, rs7668556) tagged a locus on chromosome 4 that includes seven genes of the UGT2 family. The remaining SNP, rs12767583, is an intronic variant in CYP2C19. FINEMAP showed that both loci harbor complex association signals, returning 65 and 102 credible SNPs, respectively (see Table S4 in the online supplement). At each locus, the set of credible SNPs included a missense variant in high LD (r2>0.9) with the top FINEMAP SNP, one of which (rs61750900) was also genome-wide significant in the norclozapine analysis. Both missense variants, rs61750900 (PVE, 9.48%) and rs1126545 (CYP2C18 Thr/Met; PVE, 1.85%), were incorporated into a log-normal model, where their minor allele dosage was shown to increase the clozapine-to-norclozapine ratio (Table 2).

Secondary GWAS analyses controlling for smoking and body mass index PRSs produced results very similar to those reported here, with no gain or loss of genome-wide significant signals (see the Supplementary Methods section of the online supplement). Also, confirming observations from a previous study conducted using multiple assays (52), we found that our approach of using mixed-model residuals as a GWAS phenotype produced results similar to those obtained with the use of summary statistics (averages or maximum values), but with tighter standard errors, resulting in improved significance for individual loci (see Figure S6 in the online supplement).

Co-Localization Analysis of Metabolite Levels

We employed a co-localization procedure to test whether SNPs implicated in clozapine levels might also affect the plasma concentrations of other compounds, as this can provide insight into the causal mechanisms behind these signals and reveal metabolic convergences and potential clinically important interactions. Analysis of the clozapine level GWAS showed that the association at the CYP1A2 locus, indexed by rs2472297, was also observed (PPCOLOC > 94%) in GWASs of five xenobiotic metabolites: caffeine, theophylline, 7-methylxanthine, paraxanthine, and Leu-Pro cyclopeptide. All of these are putative biomarkers of coffee consumption (53): the first four are implicated in caffeine metabolism, and Leu-Pro cyclopeptide is a component of roasted coffee. On the basis of these results, we carried out a polygenic score analysis to obtain a surrogate metric of daily coffee intake (see the Supplementary Methods section of the online supplement), which we found to be significantly associated with all of our phenotypes (see Table S5 in the online supplement).

In the analysis of the norclozapine level GWAS, the xenobiotic metabolite pelargonate co-localized at the main UGT2B10 locus (PPCOLOC=96.97%), and caffeine and theophylline co-localized at the CYP1A1/CYP1A2 locus (PPCOLOC>98%), which for norclozapine is indexed by rs2472297 (p=3.52×10−5). Although pelargonate is a component of some commercial coffee varieties, it is not part of the caffeine metabolic pathway, but of the wider system of fatty acid metabolism, as is UGT2B10 (54). Interestingly, pelargonate shares structural similarities with the antiepileptic valproate, and coadministration of valproate with clozapine has recently been shown to reduce norclozapine plasma levels (55).

Genome-Wide Enrichment of Metabolic Pathways

After correction for multiple testing, five gene sets were significant in our analysis of RECON biochemical pathways (Table 3; see also Table S6 in the online supplement). Vitamin A (retinol) metabolism was the top enriched pathway for both clozapine and norclozapine, and linoleate metabolism was the second norclozapine pathway and the top pathway for the clozapine/norclozapine ratio. In clozapine and norclozapine, we also observed significant family-wise error rate–corrected enrichment for the set of 203 drug-metabolizing enzymes; repeating the enrichment analysis using this gene set as a covariate removed all other gene set signals from the clozapine GWAS, while vitamin A and linoleate (a fatty acid) remained significant in the gene set analyses of norclozapine and the metabolic ratio (see Table S7 in the online supplement).

TABLE 3. Gene sets surviving family-wise error rate correction from the MAGMA gene set analysis of the RECON metabolic pathwaysa

Phenotype and Gene SetGenes (N)βSEp (MAGMA)p (Corrected)
Clozapine
 Retinol metabolism330.6350.182.07×10–40.016
Norclozapine
 Retinol metabolism331.080.04483.95×10–103.12×10–8
 Linoleate metabolism161.50.04332.30×10–76.00×10–5
 Arachidonate metabolism220.5520.01871.17×10–40.009
 Steroid metabolism410.5280.02441.53×10–40.012
Metabolic ratio
 Linoleate metabolism161.540.04455.95×10–71.30×10–4
 Chondroitin degradation100.7530.01722.07×10–40.039

aThe family-wise error rate correction threshold for p was 0.05.

TABLE 3. Gene sets surviving family-wise error rate correction from the MAGMA gene set analysis of the RECON metabolic pathwaysa

Enlarge table

Discussion

This is, to our knowledge, the first GWAS of clozapine metabolite plasma concentrations, which we carried out in 2,989 European individuals, the majority of whom had been assayed at multiple time points. Using statistical modeling to take advantage of all the available data, and a linear mixed-model GWAS approach, we provide the first robust evidence that alleles of specific CYP and UGT genes contribute to clozapine pharmacokinetics. This represents an advance from previous inconclusive studies (28), mostly based on candidate marker surveys, and clarifies the relevance of common genetic variation in the proteins implicated in the clozapine metabolic route, which has been a matter of extensive debate. More specifically, our results support the hypothesis that the genetic architecture of clozapine metabolism may be driven by a few variants of large effect (see Figure S7 in the online supplement), in line with other well-studied metabolic traits (56).

The CYP1A1/CYP1A2 SNP rs2472297 associated with clozapine plasma concentrations lies in an intergenic region rich in binding sites for the aryl hydrocarbon receptor (AHR) protein, sites that are collectively known as “xenobiotic response elements” (57). AHR binding is known to induce the expression of CYP enzymes in hepatocytes in response to the detection of many compounds, and thus variation in the regulatory function of AHR provides a strong candidate mechanism underpinning this association. While a causal association cannot be made solely on these grounds, disruption of normal AHR binding has also been suggested as explaining the association between variants at this locus and caffeine plasma levels, which may also influence coffee consumption (51). Previous studies of clozapine levels and candidate polymorphisms at this locus have focused on common alleles within CYP1A2 (27, 28), none of which has been shown to influence its expression (58). We also note that we find no support for other candidate genes from the literature, including the ABCB1 variant rs1045642 (p=0.84), which was previously reported as associated with clozapine plasma concentrations in smaller (N<100) samples (28). Both of these examples demonstrate the limitations of candidate SNP approaches that have been common in psychiatric pharmacogenomics to date and support genome-wide analysis to capture both coding and noncoding functional elements.

The results of our regression modeling show that the genetic modifiers of clozapine levels are comparable in impact to other known clinical and demographic variables, with their effect sizes being of the same magnitude as sex (Table 1, Table 2). An example to illustrate these effects and place them in clinical context is the observation that carrying one minor allele of rs2472297 at CYP1A1/CYP1A2 is associated with a reduction in clozapine plasma concentrations roughly equivalent to a decrease in clozapine by 50 mg/day, and homozygosity for the minor allele is equivalent to a reduction by 100 mg/day (Figure 2). Similar effects were found for the missense SNPs associated with norclozapine levels (see Figure S8 in the online supplement). The impacts on clozapine metabolite concentrations captured by these SNPs warrant their further study within the context of personalized drug therapy, given their potential clinically significant impact on dosing.

In following up the results of the GWAS, we sought genomic regions associated with clozapine metabolism that have also been identified as influencing metabolism of other compounds. We identified a strong relationship between the genetics of clozapine and caffeine metabolism, a finding with potential clinical relevance. A link between clozapine and caffeine metabolism was first proposed on the basis of findings indicating that the results of caffeine clearance tests, used as an index of CYP1A2 activity, correlate with clozapine clearance (59). While there have not been large-scale studies in clinical settings, the available evidence suggests that caffeine interacts competitively with clozapine, causing heavy coffee drinkers to have higher baseline clozapine plasma levels (60). Among factors that may have obscured this finding in previous research are the proven correlation between smoking and coffee consumption (61) and the observation that even decaffeinated coffee may lower the activity of some hepatic enzymes (62). In this regard, our analysis of metabolic-genetic association data showed commonalities, which hint at potential interactions, with several compounds related to coffee and caffeine. Remarkably, loci outside of the widely studied CYP1A2 region seem to jointly affect both coffee consumption habits and plasma concentration of clozapine metabolites, as we have shown using a polygenic score approach (see the Supplementary Methods section of the online supplement). However, given that we did not have access to coffee or caffeine consumption data, we could not assess the degree to which caffeine may mediate or moderate the genetic associations with clozapine metabolite levels. Nonetheless, our results add to the existing evidence of the potential clinical importance of the interaction between the metabolic pathways for clozapine and caffeine.

Our data can also be interpreted in the light of the proposed mechanistic link between smoking tobacco and clozapine metabolism, which is thought to result from induction by tobacco of CYP1A2 activity, which in turn increases the first-pass metabolism of clozapine (63). Current guidelines state that patients on clozapine need to be more closely monitored if they stop smoking, as their plasma levels can suddenly rise as the CYP1A2 induction fades. This effect, also seen with other medications, has been attributed to the effect of polycyclic hydrocarbons present in tobacco smoke, rather than a direct action of nicotine (64). Thus, nonsmoke alternatives to tobacco, such as nicotine patches and e-cigarettes, are generally considered not likely to interact with clozapine treatment. However, we have shown that genetic variants in UGT enzymes, which are responsible for nicotine glucuronidation, also have a role in clozapine metabolism. Specifically, we have highlighted a missense polymorphism in UGT2B10, previously shown to result in impaired enzymatic function (65), as a credible causal variant for influencing norclozapine plasma levels. This enzyme has also been shown to be a substrate of several antipsychotic drugs with structural properties similar to clozapine’s (66). Given that nicotine is a specific high-affinity inhibitor of UGT2B10, our results support the possibility of nicotine-clozapine interactions in the glucuronidation excretion pathway (26, 66), which should be investigated in more detail.

One of the limitations of this study is that our regression models do not explain as much variance in plasma concentrations as previous studies (16, 17). However, we note that these studies included the clozapine/norclozapine metabolic ratio as a covariate of clozapine plasma concentrations. Given our data, and considering all fixed-effect covariates (see Table S2 in the online supplement), this addition would have increased the variance explained by our mixed model from 19.28% to 32.34%, but at the cost of adding collinearity and hindering its interpretability. In any case, these models likely represent a lower bound of variance explained for clozapine plasma concentrations, given that we lacked individual measures of some known predictors of clozapine metabolism, including smoking and weight. We attempted to address this limitation in the discovery GWAS by using a novel application of PRSs as genetically informative proxies of these measures (see the Supplementary Methods section of the online supplement). While this did not affect the results, we acknowledge that these are just markers of the exposures and do not capture their full effects. A further limitation is the lack of detailed individual-level data on concomitant medications that could interact with clozapine. Coprescription of such medications (e.g., carbamazepine and fluvoxamine) has been shown to be rare given their potential for clinically important interactions (67), and hence it does not seem feasible that such coprescription could be an important source of bias in our findings. Furthermore, given that the absence of detailed individual-level exposure data is known to obscure the detection of genetic influences in metabolic enzymes (68), our finding of detectable GWAS signals is reassuring. A final limitation is the potential that those who have had their clozapine levels taken may be an unrepresentative sample of all those taking clozapine, which would constitute a form of selection bias. In examining this issue, we did not detect differences between those with or without clozapine therapeutic drug monitoring assays in the distribution of age, sex, and several PRSs (schizophrenia, IQ, body mass index, smoking). Nonetheless, we cannot rule out other selection effects, and thus our findings should be interpreted as relevant to the population in which clozapine levels are monitored.

In summary, our analysis has allowed us to dissect the clozapine metabolic pathway using genetic and pharmacokinetic data. We have also demonstrated commonalities with the metabolism of other biological compounds, in particular nicotine and caffeine, that highlight relevant facets of metabolism and indicate potential interactions of clinical importance. Furthermore, our findings indicate avenues for next-stage clinical studies to determine the utility of pharmacogenomic testing as a complement to clozapine monitoring procedures, with the potential to have an impact on clinical care through improved titration and dosing and minimizing of adverse drug reactions.

MRC Centre for Neuropsychiatric Genetics and Genomics, Division of Psychological Medicine and Clinical Neurosciences, School of Medicine, Cardiff University, Cardiff, Wales (Pardiñas, Nalmpanti, Pocklington, Legge, Medway, Zammit, Owen, O’Donovan, Walters); Magna Laboratories, Ross-on-Wye, U.K. (King); Leyden Delta, Nijmegen, the Netherlands (Jansen, Helthuis); the Centre for Academic Mental Health, Population Health Sciences, Bristol Medical School, University of Bristol, Bristol, U.K. (Zammit); the National Institute for Health Research, Biomedical Research Centre, University of Bristol, Bristol, U.K. (Zammit); and the Department of Psychosis Studies, Institute of Psychiatry, Psychology, and Neuroscience, King’s College London (MacCabe).
Send correspondence to Dr. Walters ().

Drs. Helthuis and Jansen are full-time employees of Leyden Delta. Dr. King is a full-time employee of Magna Laboratories. Drs. Pocklington, Owen, O’Donovan, and Walters are supported by a collaborative research grant from Takeda (Takeda played no part in the conception, design, implementation, or interpretation of this study, which was completed prior to the funding award). The other authors report no financial relationships with commercial interests.

Supported by grants from the Medical Research Council (MRC) to Cardiff University (MR/L010305/1; G0800509; MR/L011794/1 and MC_PC_17212). The project has also received funding from the European Union’s Seventh Framework Programme (279227). Dr. MacCabe is partly funded by the National Institute for Health Research, Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London.

The authors acknowledge Leyden Delta and Magna Laboratories for supporting the CLOZUK2 sample collection, anonymization, and data preparation (Andy Walker and Anouschka Colson); deCODE genetics (Hreinn Stefansson) for genotyping of the CLOZUK2 sample; the Cardiff University MRC Centre Core Team (Lucinda Hopkins, Lesley Bates, and Catherine Bresner) for laboratory sample management; and the Cardiff University MRC Centre HPC team (Mark Einon) and Cardiff University Advanced Research Computing division (Wayne Lawrence) for support with the use and setup of computational infrastructures.

References

1 McGrath J , Saha S , Chant D , et al. : Schizophrenia: a concise overview of incidence, prevalence, and mortality. Epidemiol Rev 2008; 30: 67– 76 Crossref, MedlineGoogle Scholar

2 Owen MJ , Sawa A , Mortensen PB : Schizophrenia. Lancet 2016; 388: 86– 97 Crossref, MedlineGoogle Scholar

3 Howes OD , McCutcheon R , Agid O , et al. : Treatment-resistant schizophrenia: Treatment Response and Resistance in Psychosis (TRRIP) working group consensus guidelines on diagnosis and terminology. Am J Psychiatry 2017; 174: 216– 229 LinkGoogle Scholar

4 Essali A , Al-Haj Haasan N , Li C , et al. : Clozapine versus typical neuroleptic medication for schizophrenia. Cochrane Database Syst Rev 2009; 1: CD000059 MedlineGoogle Scholar

5 Kelly DL , Wehring HJ , Vyas G : Current status of clozapine in the United States. Shanghai Jingshen Yixue 2012; 24: 110– 113 Google Scholar

6 Latimer E , Wynant W , Clark R , et al. : Underprescribing of clozapine and unexplained variation in use across hospitals and regions in the Canadian province of Québec. Clin Schizophr Relat Psychoses 2013; 7: 33– 41 Crossref, MedlineGoogle Scholar

7 Atkin K , Kendall F , Gould D , et al. : Neutropenia and agranulocytosis in patients receiving clozapine in the UK and Ireland. Br J Psychiatry 1996; 169: 483– 488 Crossref, MedlineGoogle Scholar

8 Nair B , MacCabe JH : Making clozapine safer: current perspectives on improving its tolerability. Future Neurol 2014; 9: 313– 322 CrossrefGoogle Scholar

9 Best-Shaw L , Gudbrandsen M , Nagar J , et al. : Psychiatrists’ perspectives on antipsychotic dose and the role of plasma concentration therapeutic drug monitoring. Ther Drug Monit 2014; 36: 486– 493 Crossref, MedlineGoogle Scholar

10 Legge SE , Hamshere M , Hayes RD , et al. : Reasons for discontinuing clozapine: a cohort study of patients commencing treatment. Schizophr Res 2016; 174: 113– 119 Crossref, MedlineGoogle Scholar

11 Flanagan RJ : A practical approach to clozapine therapeutic drug monitoring. CMHP Bulletin 2010; 2: 4– 5 Google Scholar

12 Spina E , Hiemke C , de Leon J : Assessing drug-drug interactions through therapeutic drug monitoring when administering oral second-generation antipsychotics. Expert Opin Drug Metab Toxicol 2016; 12: 407– 422 Crossref, MedlineGoogle Scholar

13 Hiemke C , Bergemann N , Clement HW , et al. : Consensus Guidelines for Therapeutic Drug Monitoring in Neuropsychopharmacology: Update 2017. Pharmacopsychiatry 2018; 51: 9– 62 Crossref, MedlineGoogle Scholar

14 Simon V , van Winkel R , De Hert M : Are weight gain and metabolic side effects of atypical antipsychotics dose dependent? A literature review. J Clin Psychiatry 2009; 70: 1041– 1050 Crossref, MedlineGoogle Scholar

15 Ramos Perdigués S , Sauras Quecuti R , Mané A , et al. : An observational study of clozapine induced sedation and its pharmacological management. Eur Neuropsychopharmacol 2016; 26: 156– 161 Crossref, MedlineGoogle Scholar

16 Rostami-Hodjegan A , Amin AM , Spencer EP , et al. : Influence of dose, cigarette smoking, age, sex, and metabolic activity on plasma clozapine concentrations: a predictive model and nomograms to aid clozapine dose adjustment and to assess compliance in individual patients. J Clin Psychopharmacol 2004; 24: 70– 78 Crossref, MedlineGoogle Scholar

17 Couchman L , Bowskill SVJ , Handley S , et al. : Plasma clozapine and norclozapine in relation to prescribed dose and other factors in patients aged <18 years: data from a therapeutic drug monitoring service, 1994–2010. Early Interv Psychiatry 2013; 7: 122– 130 Crossref, MedlineGoogle Scholar

18 Russell Teagarden J , Stanek EJ : On pharmacogenomics in pharmacy benefit management. Pharmacotherapy 2012; 32: 103– 111 Crossref, MedlineGoogle Scholar

19 Lauschke VM , Milani L , Ingelman-Sundberg M : Pharmacogenomic biomarkers for improved drug therapy-recent progress and future developments. AAPS J 2017; 20: 4 Crossref, MedlineGoogle Scholar

20 de Leon J : Have we successfully implemented CYP2D6 genotyping in psychiatry? Expert Opin Drug Metab Toxicol 2017; 13: 1201– 1203 Crossref, MedlineGoogle Scholar

21 Jann MW , Grimsley SR , Gray EC , et al. : Pharmacokinetics and pharmacodynamics of clozapine. Clin Pharmacokinet 1993; 24: 161– 176 Crossref, MedlineGoogle Scholar

22 Eiermann B , Engel G , Johansson I , et al. : The involvement of CYP1A2 and CYP3A4 in the metabolism of clozapine. Br J Clin Pharmacol 1997; 44: 439– 446 Crossref, MedlineGoogle Scholar

23 Lameh J , Burstein ES , Taylor E , et al. : Pharmacology of N-desmethylclozapine. Pharmacol Ther 2007; 115: 223– 231 Crossref, MedlineGoogle Scholar

24 Breyer-Pfaff U , Wachsmuth H : Tertiary N-glucuronides of clozapine and its metabolite desmethylclozapine in patient urine. Drug Metab Dispos 2001; 29: 1343– 1348 MedlineGoogle Scholar

25 Zanger UM , Schwab M : Cytochrome P450 enzymes in drug metabolism: regulation of gene expression, enzyme activities, and impact of genetic variation. Pharmacol Ther 2013; 138: 103– 141 Crossref, MedlineGoogle Scholar

26 Rowland A , Miners JO , Mackenzie PI : The UDP-glucuronosyltransferases: their role in drug metabolism and detoxification. Int J Biochem Cell Biol 2013; 45: 1121– 1132 Crossref, MedlineGoogle Scholar

27 Bersani FS , Capra E , Minichino A , et al. : Factors affecting interindividual differences in clozapine response: a review and case report. Hum Psychopharmacol 2011; 26: 177– 187 MedlineGoogle Scholar

28 Krivoy A , Gaughran F , Weizman A , et al. : Gene polymorphisms potentially related to the pharmacokinetics of clozapine: a systematic review. Int Clin Psychopharmacol 2016; 31: 179– 184 Crossref, MedlineGoogle Scholar

29 Rietveld CA , Conley D , Eriksson N , et al. : Replicability and robustness of genome-wide-association studies for behavioral traits. Psychol Sci 2014; 25: 1975– 1986 Crossref, MedlineGoogle Scholar

30 Pardiñas AF , Holmans P , Pocklington AJ , et al. : Common schizophrenia alleles are enriched in mutation-intolerant genes and in regions under strong background selection. Nat Genet 2018; 50: 381– 389 Crossref, MedlineGoogle Scholar

31 Chang CC , Chow CC , Tellier LC , et al. : Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience 2015; 4: 7 Crossref, MedlineGoogle Scholar

32 Anderson CA , Pettersson FH , Clarke GM , et al. : Data quality control in genetic case-control association studies. Nat Protoc 2010; 5: 1564– 1573 Crossref, MedlineGoogle Scholar

33 McCarthy S , Das S , Kretzschmar W , et al. : A reference panel of 64,976 haplotypes for genotype imputation. Nat Genet 2016; 48: 1279– 1283 Crossref, MedlineGoogle Scholar

34 Das S , Forer L , Schönherr S , et al. : Next-generation genotype imputation service and methods. Nat Genet 2016; 48: 1284– 1287 Crossref, MedlineGoogle Scholar

35 Advanced Research Computing @ Cardiff (ARCCA). Raven Supercomputing Cluster. https://www.cardiff.ac.uk/research-equipment/facilities/view/arcca-raven-supercomputing-cluster Google Scholar

36 van Leeuwen EM , Kanterakis A , Deelen P , et al. : Population-specific genotype imputations using minimac or IMPUTE2. Nat Protoc 2015; 10: 1285– 1296 Crossref, MedlineGoogle Scholar

37 Huang J , Howie B , McCarthy S , et al. : Improved imputation of low-frequency and rare variants using the UK10K haplotype reference panel. Nat Commun 2015; 6: 8111 Crossref, MedlineGoogle Scholar

38 Legge SE , Pardiñas AF , Helthuis M , et al. : A genome-wide association study in individuals of African ancestry reveals the importance of the Duffy-null genotype in the assessment of clozapine-related neutropenia. Mol Psychiatry 2019; 24: 328– 337 Crossref, MedlineGoogle Scholar

39 Grundmann M , Kacirova I , Urinovska R : Therapeutic drug monitoring of atypical antipsychotic drugs. Acta Pharm 2014; 64: 387– 401 Crossref, MedlineGoogle Scholar

40 Yang J , Lee SH , Goddard ME , et al. : GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet 2011; 88: 76– 82 Crossref, MedlineGoogle Scholar

41 Benner C , Spencer CCA , Havulinna AS , et al. : FINEMAP: efficient variable selection using summary data from genome-wide association studies. Bioinformatics 2016; 32: 1493– 1501 Crossref, MedlineGoogle Scholar

42 McLaren W , Pritchard B , Rios D , et al. : Deriving the consequences of genomic variants with the Ensembl API and SNP effect predictor. Bioinformatics 2010; 26: 2069– 2070 Crossref, MedlineGoogle Scholar

43 Carithers LJ , Ardlie K , Barcus M , et al. : A novel approach to high-quality postmortem tissue procurement: the GTEx project. Biopreserv Biobank 2015; 13: 311– 319 Crossref, MedlineGoogle Scholar

44 Schmitt AD , Hu M , Jung I , et al. : A compendium of chromatin contact maps reveals spatially active regions in the human genome. Cell Reports 2016; 17: 2042– 2059 Crossref, MedlineGoogle Scholar

45 Conomos MP , Miller MB , Thornton TA : Robust inference of population structure for ancestry prediction and correction of stratification in the presence of relatedness. Genet Epidemiol 2015; 39: 276– 293 Crossref, MedlineGoogle Scholar

46 Pickrell JK , Berisa T , Liu JZ , et al. : Detection and interpretation of shared genetic influences on 42 human traits. Nat Genet 2016; 48: 709– 717 Crossref, MedlineGoogle Scholar

47 Shin S-Y , Fauman EB , Petersen A-K , et al. : An atlas of genetic influences on human blood metabolites. Nat Genet 2014; 46: 543– 550 Crossref, MedlineGoogle Scholar

48 Swainston N , Smallbone K , Hefzi H , et al. : Recon 2.2: from reconstruction to model of human metabolism. Metabolomics 2016; 12: 109 Crossref, MedlineGoogle Scholar

49 Schärfe CPI , Tremmel R , Schwab M , et al. : Genetic variation in human drug-related genes. Genome Med 2017; 9: 117 Crossref, MedlineGoogle Scholar

50 de Leeuw CA , Mooij JM , Heskes T , et al. : MAGMA: generalized gene-set analysis of GWAS data. PLOS Comput Biol 2015; 11: e1004219 Crossref, MedlineGoogle Scholar

51 Cornelis MC , Kacprowski T , Menni C , et al. : Genome-wide association study of caffeine metabolites provides new insights to caffeine metabolism and dietary caffeine-consumption behavior. Hum Mol Genet 2016; 25: 5472– 5482 MedlineGoogle Scholar

52 Hoffmann TJ , Theusch E , Haldar T , et al. : A large electronic-health-record-based genome-wide study of serum lipids. Nat Genet 2018; 50: 401– 413 Crossref, MedlineGoogle Scholar

53 Guertin KA , Loftfield E , Boca SM , et al. : Serum biomarkers of habitual coffee consumption may provide insight into the mechanism underlying the association between coffee consumption and colorectal cancer. Am J Clin Nutr 2015; 101: 1000– 1011 Crossref, MedlineGoogle Scholar

54 Turgeon D , Chouinard S , Bélanger P , et al. : Glucuronidation of arachidonic and linoleic acid metabolites by human UDP-glucuronosyltransferases. J Lipid Res 2003; 44: 1182– 1191 Crossref, MedlineGoogle Scholar

55 Hommers L , Scharl M , Hefner G , et al. : Comedication of valproic acid is associated with increased metabolism of clozapine. J Clin Psychopharmacol 2018; 38: 188– 192 Crossref, MedlineGoogle Scholar

56 Timpson NJ , Greenwood CMT , Soranzo N , et al. : Genetic architecture: the shape of the genetic contribution to human traits and disease. Nat Rev Genet 2018; 19: 110– 124 Crossref, MedlineGoogle Scholar

57 Corchero J , Pimprale S , Kimura S , et al. : Organization of the CYP1A cluster on human chromosome 15: implications for gene regulation. Pharmacogenetics 2001; 11: 1– 6 Crossref, MedlineGoogle Scholar

58 Ingelman-Sundberg M , Sim SC , Gomez A , et al. : Influence of cytochrome P450 polymorphisms on drug therapies: pharmacogenetic, pharmacoepigenetic, and clinical aspects. Pharmacol Ther 2007; 116: 496– 526 Crossref, MedlineGoogle Scholar

59 Bertilsson L , Carrillo JA , Dahl ML , et al. : Clozapine disposition covaries with CYP1A2 activity determined by a caffeine test. Br J Clin Pharmacol 1994; 38: 471– 473 Crossref, MedlineGoogle Scholar

60 Raaska K , Raitasuo V , Laitila J , et al. : Effect of caffeine-containing versus decaffeinated coffee on serum clozapine concentrations in hospitalised patients. Basic Clin Pharmacol Toxicol 2004; 94: 13– 18 Crossref, MedlineGoogle Scholar

61 Treur JL , Taylor AE , Ware JJ , et al. : Associations between smoking and caffeine consumption in two European cohorts. Addiction 2016; 111: 1059– 1068 Crossref, MedlineGoogle Scholar

62 Xiao Q , Sinha R , Graubard BI , et al. : Inverse associations of total and decaffeinated coffee with liver enzyme levels in National Health and Nutrition Examination Survey, 1999–2010. Hepatology 2014; 60: 2091– 2098 Crossref, MedlineGoogle Scholar

63 Lowe EJ , Ackman ML : Impact of tobacco smoking cessation on stable clozapine or olanzapine treatment. Ann Pharmacother 2010; 44: 727– 732 Crossref, MedlineGoogle Scholar

64 Smith TL , Mican LM : What to do when your patient who takes clozapine enters a smoke-free facility. Curr Psychiatr 2014; 13: 55 Google Scholar

65 Chen G , Blevins-Primeau AS , Dellinger RW , et al. : Glucuronidation of nicotine and cotinine by UGT2B10: loss of function by the UGT2B10 codon 67 (Asp>Tyr) polymorphism. Cancer Res 2007; 67: 9024– 9029 Crossref, MedlineGoogle Scholar

66 Pattanawongsa A , Nair PC , Rowland A , et al. : Human UDP-glucuronosyltransferase (UGT) 2B10: validation of cotinine as a selective probe substrate, inhibition by UGT enzyme-selective inhibitors and antidepressant and antipsychotic drugs, and structural determinants of enzyme inhibition. Drug Metab Dispos 2016; 44: 378– 388 Crossref, MedlineGoogle Scholar

67 Heald A , Livingston M , Yung A , et al. : Prescribing in schizophrenia and psychosis: Increasing polypharmacy over time. Hum Psychopharmacol 2017; 32: e2579 CrossrefGoogle Scholar

68 Matthaei J , Tzvetkov MV , Strube J , et al. : Heritability of caffeine metabolism: environmental effects masking genetic effects on CYP1A2 activity but not on NAT2. Clin Pharmacol Ther 2016; 100: 606– 616 Crossref, MedlineGoogle Scholar