Adverse Drug Reactions of Olanzapine, Clozapine and Loxapine in Children and Youth: A Systematic Pharmacogenetic Review

Children and youth treated with antipsychotic drugs (APs) are particularly vulnerable to adverse drug reactions (ADRs) and prone to poor treatment response. In particular, interindividual variations in drug exposure can result from differential metabolism of APs by cytochromes, subject to genetic polymorphism. CYP1A2 is pivotal in the metabolism of the APs olanzapine, clozapine, and loxapine, whose safety profile warrants caution. We aimed to shed some light on the pharmacogenetic profiles possibly associated with these drugs’ ADRs and loss of efficacy in children and youth. We conducted a systematic review relying on four databases, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 recommendations and checklist, with a quality assessment. Our research yielded 32 publications. The most frequent ADRs were weight gain and metabolic syndrome (18; 56.3%), followed by lack of therapeutic effect (8; 25%) and neurological ADRs (7; 21.8%). The overall mean quality score was 11.3/24 (±2.7). In 11 studies (34.3%), genotyping focused on the study of cytochromes. Findings regarding possible associations were sometimes conflicting. Nonetheless, cases of major clinical improvement were fostered by genotyping. Yet, CYP1A2 remains poorly investigated. Further studies are required to improve the assessment of the risk–benefit balance of prescription for children and youth treated with olanzapine, clozapine, and/or loxapine.


Introduction
In child psychiatry, antipsychotic drugs (APs) are used to treat psychotic or mood disorders, as well as behavioral symptoms, despite limited evidence. Although APs are usually efficacious, the risk of adverse drug reactions (ADRs) associated with this class should be considered when initiating APs in this vulnerable population [1,2]. Treatment resistance is also a major concern [3]. Many intrinsic and extrinsic factors may influence the pharmacokinetics and pharmacodynamics of APs, such as sex, ancestry, puberty, dietary, and smoking habits [4][5][6][7], potentially leading to ADRs or lack of therapeutic effects. the administration of a selective enzyme substrate. These approaches brought us closer to personalized medicine, whereby the understanding of each patient's genetic profile may predict the occurrence of ADRs or lack of effect. This may be especially useful in specific populations [42], often excluded of clinical trials and of the classical field of evidence-based medicine.
Therefore, we aimed to review the pharmacogenetic variants underlying olanzapine, clozapine, and loxapine ADRs and/or efficacy in children and youth having undergone genotyping. Then, we assessed the most frequently investigated ADRs and genetic polymorphisms in this population. Finally, we assessed the specific effect of CYP1A2 variants in the occurrence of ADRs and/or lack of therapeutic effect.

Research
The PROSPERO International prospective register of systematic reviews was checked for similar systematic reviews. Due to our issue of concern never having been addressed, we have submitted the research protocol to the INPLASY International platform of registered systematic review and meta-analysis protocols (INPLASY202250025).
We have, therefore, conducted this systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 recommendations and checklist [43]. We further followed special methodological considerations regarding pediatric systematic reviews [44]. The following query was used: ((((adolescent* OR youth OR child* OR pedia* OR paedia*) AND (clozapine OR olanzapine OR loxapine) AND (pharmacogen* OR allele OR genotype* OR cytochrome* OR CYP1* OR CYP2* OR CYP3* OR CYP4*) AND (adverse drug reaction* OR adverse event* OR adverse reaction* OR side effect* OR secondary effect* OR after effect* OR tolerability OR safety)))). Two authors (D.M. and A.O.G.) separately conducted the research in PubMed, EMBASE, PsycINFO, and PsycArticles. Our query retrieved publications registered in the four selected databases up to 21 March 2022.
Relying on four electronic bibliographic databases, our extraction retrieved for each publication the source database, publication year, language, full list of authors' names, article title, DOI (Digital Object Information), journal title, abstract, and Medical Subject Headings (MeSH) terms associated. Two authors independently performed the preliminary two steps of proper article screening, with the results shown in the PRISMA flowchart ( Figure 1).
Before screening, duplicates were removed. First, the eligibility of the titles and abstracts of the articles identified by the initial query were checked. Next, full-text copies of the articles whose titles and abstracts met the inclusion criteria were retrieved. Then, to ensure compliance with the inclusion criteria, the yielded full-text articles were assessed for eligibility.
When the two reviewing authors could not obtain a consensus regarding an article, the disagreement was resolved through discussion. Lastly, data extraction was performed for all publications that met the inclusion criteria, including the study site(s), study type, characteristics of the subjects (age, sample size, sex distribution, ancestry, diagnosis), antipsychotic(s) of interest and its (their) dosing, other drugs administered, outcome(s) measured, gene variants assessed, their potential association(s) with the ADR(s), the pathophysiology involved, and the pharmacogenetic approach. For quality assessment needs, we also extracted data addressing the reasons for choosing the genes/SNPs to genotype (summaries of previous findings, reasons given for choosing the genes and SNPs genotyped, the adjustment methods for multiple testing, and the p-values provided for the associations), the sample size (details on calculation of sample size and on a priori power to detect effect sizes of varying degrees), the reliability of genotypes (description of the genotyping procedure, of the primers and of any quality control methods, previously reported genotype frequencies, blind of genotyping personnel to outcome status), missing genotype data (the extent and reasons for missing data, any checks for missingness at random performed, any imputation of missing genotype data, number of patients contributing to each analysis and consistence with sample size), population stratification (tests undertaken for cryptic population stratification and adjustment for in the analyses), Hardy-Weinberg Equilibrium testing (was it performed, and were deviating (or not) SNPs highlighted and excluded from further analysis where appropriate), and choice and definition of outcomes (clear definition of all outcomes investigated, justification, results shown).

Selection Criteria
Data extraction relied on the following inclusion criteria:

1.
Studies including at least one child and/or adolescent and/or youth, therefore aged under 25, following the United Nations definition [45].

3.
Having experienced an adverse drug reaction/a lack of therapeutic effect linked to at least one of these treatments.

4.
Having undergone pharmacogenomic analysis/genotyping, the results of which are mentioned.

5.
Record issued from an English-language and peer-reviewed journal, for which fulltext was available We therefore excluded books (and chapters), commentaries, but also any published material that did not meet the original research criteria (e.g., systematic reviews, meta-analyses) [46]. However, considering the foreseeable paucity of evidence informing the review, we decided to include conference abstracts and editorial pieces [47].
To serve the same purpose, we have chosen to include studies including 'mixed' (both adult and pediatric) populations [44], with due regard to the age criterion: 'Studies including at least one child and/or adolescent, therefore aged under 25 .
Then, identical or overlapping patient cohorts were detected by the analysis of study site(s) and characteristics of the subjects, among others. The objectives and genetic variants investigated tended to differ across the reports, based on overlapping or identical cohorts, so we have chosen to include publications presenting redundant cohorts [39].
When the ancestry of patients (whose consideration is pivotal in genetics concerns) was not provided in a study, we hypothesized that it could be consistent with the study site, and reported it as such.
Studies were classified according to their methodology: case reports or case series, cohort studies [48], and case-control (or cross-sectional) studies [49]. We distinguished 'pediatric' studies, exclusively relying on pediatric samples, and 'mixed-population' studies, to present their respective characteristics (Tables 1 and 2) and quality assessments (Tables S1 and S2). Then, the whole studies were grouped according to the main classes of ADRs investigated (Tables 3-5).

Quality Assessment
The quality of the included pharmacogenetic studies was independently assessed by D.M. and A.O.G, relying on a tool adapted from Maruf et al. [13] and the checklist developed by Jorgensen and Williamson [50]. As stated above, we considered each article (irrespective of the potential redundancy of its (their) cohort(s)) for quality assessment. Indeed, methods may vary from an article to another, relying on identical or overlapping patient cohorts. Any case of discrepancy between their assessments was resolved through discussion.
The used tool addressed different issues of methodological quality: 1.
Choice and definition of outcomes (3 binary questions).
The purpose of open questions (sample size; study design) was to allow a quality visual check as a complement to the global score of each publication.
For each binary question, we answered: • 'Yes' if the study provided an adequate response. • 'No' if the response was not mentioned in the manuscript nor a method publication referenced by the authors. • 'N/A' (not applicable) if the response to the main (first) question of the issue of concern addressed is 'No'.
Consequently, each study received a quality score between 0 and 24, based on the summation of the 'Yes' answers. According to this approach, the higher the score, the higher the quality of a given study.

Study Selection
Selection and progressive elimination of the identified articles are summarized in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart provided in Figure 1. Our database query retrieved 406 records. Before screening, we removed 55 duplicates (see Methods). Then, 352 records were screened on the basis of their title and abstract. Among them, 72 publications were assessed for eligibility via the analysis of their full-text version. Finally, 32 records met the inclusion criteria of this systematic review.

General Characteristics
The most represented study type was cohort studies (20 reports; 62.5%). Sample sizes ranged from single cases (case reports) to 1445 patients (case-control study). Among articles for which the ancestry was provided, 90.9% involved Caucasian/European/White populations. It was not reported in 10 records (31.3%). Diagnosis of the included patients was provided in 32 records (96.9%), mainly represented by psychotic disorders (29 reports; 93.5%). In 11 studies (34.3%), genetic assessment relied on studying cytochromes. Olanzapine was the most commonly used AP (24 reports; 75.0%). The most frequent ADR was weight gain and metabolic syndrome (MetS), investigated in more than half of the studies (18 reports; 56.3%). Lack of therapeutic effect accounted for 8 reports (25.0%) and neurological ADRs for 7 reports (21.8%). Comparing study sites and characteristics of the populations, we noticed several overlaps between the included articles. Indeed, Nussbaum et al. in both studies ( [51,52]), as well as Le Hellard et al. [53] and Jassim et al. [54] relied on identical cohorts, respectively. To a lesser extent, Le Hellard et al. included the Theisen et al. [55] cohort; the Gagliano et al. [56] cohort overlapped with the Tiwari et al. [57] cohort; and the Quteineh et al. [58] and Saigi et al. [59] cohorts were both overlapping the Choong et al. [60] cohort.
The mean quality assessment score (see Methods) of the 32 included studies was 11.3/24 (±2.7). The scores ranged from 6 (a case series) to 18 (a cohort study). In all studies, a literature review was undertaken, whose findings were summarized, as well as the reasons for choosing the genes and SNPs genotyped. The method of adjustment for multiple testing was described in 13 records (40.6%). Precise p-values were provided for all associations in 25 records (78.1%). Regarding sample size, details on its calculation were given in one (3.1%) study (a cohort study). Details were given regarding the a priori power to detect effect sizes of varying degrees in 5 publications (15.6%). Almost all records described the genotyping procedure (31; 96.9%). Primers and quality control methods were described in 8 (25.0%) and 6 (18.8%) studies, respectively. Previously reported genotype frequencies were quoted in 9 publications (28.1%). Genotyping personnel was blinded to outcome status in one study (a cohort study) (3.1%). The extent of missing data was summarized in 9 studies (28.1%), among which 6 gave the reasons for missing data (66.7%). No study reported checks for missingness at random, nor imputed missing genotype data. All studies quoted the number of patients contributing to each analysis (32; 100%), which agreed to samples sizes in 24 studies (75.0%). No study presented tests for cryptic population stratification. Hardy-Weinberg Equilibrium (HWE) was tested in 18 reports (56.3%). Among them, the presence (or the absence) of deviating SNPs was highlighted and excluded from further analysis in 17 studies (94.4%). Finally, all studies provided definitions, justifications for their choices, and results for all outcomes investigated (32; 100%).

Pediatric Studies
Cohort studies accounted for 41.6% of pediatric studies (n = 5), followed by case reports and case series (4 studies; 33.3%). Sample sizes ranged from single cases (2 case reports) to 279 patients (a cohort study). The population was aged 3 to 20 years old. Ancestry was not reported in most publications (7 studies; 58.3%). All studies in which ethnicity was reported included Caucasian/European/White populations and African/Black populations (5; 100%). Patients' diagnosis was mentioned in 11 studies (91.6%); psychotic disorders in 8 of them (72.7%) and mood disorders in 5 of them (45.5%). Cytochromes were genotyped in a great majority of reports (9; 75.0%). Olanzapine was mentioned in nearly all the publications (11; 91.6%). Among the studied ADRs, 5 studies were related to inadequate efficacy (41.7%), 4 (33.3%) to weight gain or MetS, and 3 (25.0%) to neurological symptoms. Detailed characteristics of the included pediatric studies are provided in Table 1.
For pediatric studies, the average quality assessment score was 9.1/24 (±1.7), ranging from 6 (a case series) to 13 (a cohort study). The adjustment for multiple testing was described in one-fourth of the studies (3; 25.0%), and precise p-values were provided for all associations in one-half of the studies (6; 50.0%). No pediatric study provided details on the calculation of the sample size nor on the a priori power to detect effect sizes of varying degrees. The genotyping procedure was described in nearly all the publications (11; 92.0%). However, no study described the primers nor the quality control methods used. Previously reported genotype frequencies were quoted in 4 studies (33.3%). No study reported blinding of the genotyping personnel to outcome status. One study (1; 8.3%) summarized the extent of missing data (a cohort study), but justifications were not provided. The number of patients contributing to analyses agreed to the sample size in 10 studies (83.3%). HWE was tested in one study (a cohort study), where the absence of deviation was highlighted (1; 8.3%). The comprehensive quality assessment for pediatric studies is displayed in Table S1.
For mixed population studies, the mean quality assessment score was 12.6/24 (± 2.4), lying between 8 (a case-control study) and 18 (a cohort study). The method used to adjust for multiple testing was described in one-half of the studies (10; 50.0%). Precise p-values were provided for all associations in almost all studies (19; 95.0%).The calculation of sample size was detailed in one study (1; 5.0%) and the a priori power to detect effect sizes of varying degrees was detailed in 5 studies (5; 20.0%). All studies described the genotyping procedure (20; 100%). Primers were described in 8 studies (40.0%), and quality control methods in 6 studies (30.0%). Previously reported genotype frequencies were quoted in one-fourth of the studies (5; 25.0%). Genotyping personnel was blinded to outcome status in one study (a cohort study) (5.0%). The extent of missing data was summarized in 8 reports (40.0%), among which 6 justified it (75.0%). The number of patients contributing to the analyses agreed to sample size in 14 studies (70.0%). HWE was tested in 17 reports (85.0%), among which almost all (16; 94.1%) underlined the presence (or absence) of deviating SNPs and excluded them from further analysis when appropriate. The comprehensive quality assessment for mixed population studies is displayed in Table S2.     Table 3. Among studies specifically assessing antipsychotic-induced weight gain (AIWG), 2 were pediatric studies (14.3%) and 12 were mixed-population studies (85.7%). Both pediatric and mixed studies accounted for half (2; 50.0%) of the reports addressing MetS.
In 2014, Nussbaum et al. [51] found that CYP2D6 wt/*4 (intermediate metabolizer-IM) children had a significant increase in weight gain when compared to the patients without *4 allele, after six months of administration of atypical APs (p < 0.001). Likewise, Thümmler et al. [3] reported the case of a CYP2D6 *4/*41 (poor metabolizer-PM) 14-year-old female who showed weight gain and binge-eating behaviors when treated with clozapine and loxapine. According to the findings of Menus et al. [61], a moderate/high risk of obesity in patients treated with clozapine was significantly more frequent in low CYP3A4 expressers (13.6% of CYP3A4 low expressers, 1.5% of CYP3A4 normal/high expressers, OR = 13.5 (95% CI 1.2-147.9), n = 87, p = 0.045). However, there was no association between CYP1A2 or CYP3A4 expression and blood glucose or lipid levels (p > 0.1). By contrast, in low CYP3A4 expressers, a significant correlation was found between the clozapine serum concentration and blood glucose level (r = 0.52, n = 20, p = 0.02).
Few studies investigated the potential link between lipid homeostasis and polymorphisms of genes involved in energy. Indeed, Le Hellard et al. [53] found a strong association (p = 0.0003-0.00007) between three genetic polymorphisms localized within or near the INSIG2 gene (rs17587100, rs10490624, and rs17047764) and AIWG in patients treated with clozapine. Choong et al. [60] found that carriers of the CRTC1 (rs3746266) G allele had a lower BMI than noncarriers (AA genotype) (p = 0.001, p = 0.05, and p = 0.0003, respectively, in the three samples). When excluding patients taking other weight gain-inducing drugs, G allele carriers (n = 98) had a 1.81 kg/m 2 lower BMI than noncarriers (n = 226; p < 0.0001). This association was more marked in women aged under 45 years, with a 3.87 kg/m 2 lower BMI in G allele carriers (n = 25) compared with noncarriers (n = 48; p < 0.0001). In patients treated with clozapine, Jassim et al. [54] found a marked association between AIWG and 6 genetic polymorphisms in ADIPOQ, among which only 2 showed both allelic and genotypic association. Body Mass Index (BMI) changes were, to a lesser extent, associated with one marker in PRKAA1 (rs10074991), by an allelic (p = 0.011) and genotypic (p = 0.004) association, as well as three markers in PRKAA2 (rs4912411, p = 0.044; rs7519509, p = 0.043; rs10489617, p = 0.036). In PRKAG2, one marker (rs17714947, p = 0.020) displayed allelic association with AIWG, while another marker (rs7800069, p = 0.0008) showed genotypic association. By contrast, Gagliano et al. [56] analyzed 16 tag SNPs across the PRKAR2B gene in a sample of patients treated with clozapine or olanzapine. Patients displaying the minor allele of the polymorphism PRKAR2B (rs9656135) had a mean weight increase of 4.1%, whereas patients without this allele had an increase of 3.4%, but this association did not remain significant after correcting for multiple testing. Quteineh et al. [58] found that only male carriers of the HSD11β1 (rs846906) T allele had significantly higher waist circumference and triglycerides (TG), and lower high-density lipoprotein cholesterol (HDL) (p corrected = 0.028). This allele was also associated with a higher risk of antipsychoticinduced MetS at 3 months of follow-up (OR = 3.31 (95% CI 1.53-7.17), p corrected = 0.014). When studying patients treated with APs, the impact of 52 SNPs previously associated with BMI changes, Saigi et al. [59] found that CADM2 (rs13078807) showed a nominal association with BMI over time (p = 0.01), with a 1.04 increase in BMI per additional risk allele after 12 months of treatment. The genetic polymorphisms HSD11β1 (rs3753519) (p = 0.00001) and CRTC2 (rs8450) (p = 0.04) were also associated with a risk of an increase in BMI.
Regarding genotyping of 5-HT2C (serotonin) receptor, Theisen et al. [55] found no association between the 5-HT2C receptor (rs3813929)-759C allele and weight gain after 12 weeks of clozapine treatment in 97 patients with schizophrenia. Notwithstanding, among patients treated with olanzapine and genotyped for 5-HT2C receptor (rs518147), Godlewska et al. [62] found that significantly less patients with -697C (3/51, p ≤ 0.0006) and no patient with -759T (0/28, p ≤ 0.002) alleles experienced a BMI increase ≥10%. In an analysis of body weight change after 4 months of clozapine treatment, Hong et al. [63] showed no relationship with the histamine receptor H1 genotype (rs2067467). The analysis of DRD2 -141C (rs1799732) by Lencz et al. [64] in patients treated with APs showed that deletion carriers gained significantly more weight over time (time-by-genotype interaction, p = 0.024). Tiwari et al. [57] showed a nominal association of the CNR1 (rs806378) polymorphism with weight gain in patients treated with clozapine or olanzapine. T allele (minor allele) carriers gained more weight (5.96%) than the CC carriers (2.76%, p ≤ 0.008), which can be translated into approximately 2.2 kg more weight gain in patients carrying the T allele (CC vs. CT + TT, 2.21 ± 4.51 vs. 4.33 ± 3.89 kg; p ≤ 0.022). When searching for an association of COMT Val158Met (rs4680) variants with MetS, Cote et al. [65] found that atypical AP-treated children with the Met allele had higher systolic (p = 0.014) and diastolic (p = 0.034) blood pressure, and higher fasting glucose concentrations (p = 0.030) compared with children with the Val/Val genotype.
In atypical AP-treated children, Devlin et al. [66] found an association between the MTHFR (rs1801133) 677T allele with MetS (p ≤ 0.05) (OR 5.75 [95% CI 1.18-28.12]). Dong et al. [67] found that the A2BP1 (rs1478697) polymorphism was significantly associated with AIWG caused by olanzapine (p = 0.0012, Bonferroni corrected p = 0.0048). This association was replicated in another sample, including 208 first-episode and drugnaïve patients presenting with schizophrenia after a 4-week treatment with olanzapine (p = 0.0092, Bonferroni corrected p = 0.0368). Pouget et al. [68] found no association between TSPO (rs739092, rs5759197, rs138911, rs113515, rs6971, rs6973, rs80411 and rs138926) polymorphisms and weight change.    In low CYP3A4 expressers, a significant correlation was found between clozapine serum concentration (or daily dose) and blood glucose level The relative activity of CYP1A2 and CYP3A4 is assumed to determine which enzyme has a greater role in clozapine metabolism. 5-HT2C antagonism has been reported to be a mechanism underlying atypical AIWG + norclozapine has a greater antagonist effect on 5-HT2C receptors than the parent compound = positive correlation between BMI and norclozapine/clozapine ratios.

Neurological Symptoms: Movement Abnormalities and Seizures
Our query retrieved two studies investigating seizures (28.6%) and five studies addressing movement abnormalities (71.4%), as shown in Table 4. One pediatric and one mixed population study assessed antipsychotic-induced seizures (50%). In addition, two pediatric (40%) and thee mixed studies (60%) investigated movement abnormalities.
Baumann et al. [69] reported an epileptiform seizure, which occurred in a 16-yearold female treated with sertraline and olanzapine. She was found to be CYP3A5 *3/*3 (though, with a preserved CYP3A activity), CYP2B6 *6/*6, and CYP2D6 *4/*4 (PM). Indeed, the resulting high sertraline plasma levels added to the olanzapine treatment could have contributed to the onset of the seizure. Prows et al. [70] found that patients' combined phenotype (generated via CYP2C19 and CYP2D6 phenotypes) was associated with the number of ADRs (p = 0.03). Combined PMs treated with psychotropics had the highest number of ADRs (among which EPS was classified as a severe ADR), and combined ultrarapid metabolizers (UMs) had the lowest number of ADRs. By contrast, Thümmler et al. [3] reported the case of a CYP2D6 (>2N) UM 16-year-old male that presented EPS when treated by olanzapine and clozapine. Their case series also mentioned the case of a 14-year-old female, CYP2D6 *4/*41 (PM), who presented numerous ADRs, including EPS, akathisia, and dystonia, when treated with clozapine and loxapine. In patients treated with psychotropic drugs, Vandel et al. [71] observed a higher percentage of carriers of a genotype with CYP2D6 unfunctional alleles in the group of patients suffering from extrapyramidal ADRs than in the symptom-free patient group (p < 0.00001).
Beyond cytochromes, Kohlrausch et al. (2008) [72] found that, in patients treated with clozapine, carriers of the T825 allele of the GNB3 (rs5443) polymorphism had a higher risk to present a convulsion episode (p = 0.007). Ivashchenko et al. [73] observed that patients with HTR2A (rs6313) C allele (42.1 vs. 0%, p = 0.003), but also patients with DRD2 (rs1800497) T allele, more often complained of tremor (50 vs. 21.6%, p = 0.039). However, these associations could not be confirmed because of coincidence with higher dosing of antipsychotics. In patients treated with APs, Nicotera et al. [74] found that the COMT Val158Met (rs4680) G/A (Val/Met) genotype was almost exclusively represented in patients presenting with persistent dystonia.

Lack of Therapeutic Effect
Among studies addressing lack of therapeutic effect (Table 5), pediatric and mixed studies each accounted for a half (4; 50%).

Others
Studies investigating other ADRs were represented by a majority of pediatric studies (5; 62.8%), the remaining 3 (37.5%) relying on mixed-population samples.
Vandel et al. [71] showed a higher percentage of genotypes, including at least one allele characterized by an extensive enzyme metabolic capacity for CYP2D6 in the symptom-free group (86%) in comparison with 45.4% in the group suffering from EPS. The genotypes deprived from extensive functional alleles were more frequent (54.4%) in the group of patients suffering from EPS than in the other group (14%).
Butwicka et al. [76] reported the case of a 16-year-old male who experienced a neuroleptic malignant syndrome while being treated by olanzapine. This patient displayed a CYP2D6 *4/*4 (PM) genotype, leading to a decreased CYP2D6 activity. Nussbaum et al. [51] found that patients showing a CYP2D6 wt/*4 genotype presented a higher BMI than patients showing a wt/wt genotype. A difference across these groups was also noted for insulin values. Nussbaum et al. [52] further noted that the PANSS score in the CYP2D6 wt/*4 group was higher than in the wt/wt group. Indeed, the first patients would have exhibited no adequate drug response.
As stated above, Thümmler et al. [3] described five young patients with pharmacoresistant mental health disease who displayed CYP2D6 abnormalities: three patients were >2N UM and two patients were PM with *4/*41 and *3/*4 polymorphisms. Major psychotropic ADRs were found in four patients (EPS, akathisia, dystonia, binge eating and weight gain, hepatic cytolysis, galactorrhea, and constipation inter alia).
Grădinaru et al. [77] found that, in CYP2D6 poor and intermediate metabolizers, the use of atypical APs led to a significant increase in prolactin levels from baseline to 18 months. In IMs, the mean level of prolactin was higher than in EMs at each time point except baseline. After 6 months of AP treatment, IMs displayed a significant increase in prolactin level, over EMs.
Ivashschenko et al. [73] noted an increased dream activity in CYP2D6 IMs compared to NMs (54 vs. 22%; p = 0.043). CYP2D6 was not significantly associated with a change in the mean score of the PANSS between 1 and 14 days of treatment.
In the case series of Berel et al. [11], the second patient presented a CYP2D6 IM phenotype and a CYP3A5 *1/*1 polymorphism, and these profiles could have contributed to previous high aripiprazole and low haloperidol plasma levels.
In Ivashschenko et al.'s study [73], CYP3A5*3 polymorphism was not significantly associated with changes in the mean score of the PANSS between 1 and 14 days of treatment.
In Prows et al.'s study [70], while a significant association between combined phenotype (CYP2D6 and CYP2C19) and BIS was found, no relationship was detected between CYP2C19-predicted metabolizing phenotype and BIS (p = 0.57). Nonetheless, a relationship between CYP2C19-predicted metabolizing phenotype and the number of ADRs was observed (p = 0.01). CYP2C19-predicted metabolizing phenotype has also been linked to the type of ADRs (severe vs. mild vs. none, p = 0.04).
In the study of Berel et al. [11], the third patient was found to display a CYP2C9*1/*3 heterozygous genotype. Leading to a CYP2C9 IM phenotype, it could partly explain the low clozapine plasma levels.
Berel et al. [11] reported in their case series two 11-year-old patients with low clozapine plasma levels, which were found to be CYP1A2 UM (CYP1A2*1F/*1F and CYP1A2*1/*1F, respectively). Therefore, this issue has been corrected by the adjunction of fluvoxamine, a potent CYP1A2 inhibitor. Menus et al. [61] demonstrated a contribution of CYP1A2 to norclozapine production (0.86 ± 0.55 vs. 1.17 ± 0.70, p = 0.0007). Yet, no association was found between CYP1A2 expression and blood glucose, TG, or cholesterol (total, HDL, and LDL) levels in patients (p > 0.1). Similarly, CYP1A2 expression has not been linked with obesity (p > 0.1). None of the ADRs reported by patients was influenced by their CYP1A2 expression (p > 0.1).
In the case report of Baumann et al. [69], CYP2B6 *6/*6 homozygosity added to a PM CYP2D6 phenotype and to an olanzapine co-prescription, may have favored the occurrence of the epileptiform seizure.

Discussion
Our review aimed to assess whether pharmacogenetic mechanisms underly the occurrence of olanzapine, clozapine, and loxapine ADRs in children and youth. Several included publications investigated the genes involved in neurotransmission (COMT [65,74,80], serotonin receptors/transporters [55,62,73], dopamine receptors [64,73]), and in energy and lipid homeostasis (AMP-K related genes [54,56], HSD11β1 [58,59]), mostly regarding weight gain (or MetS). However, findings regarding possible associations were sometimes conflicting. While COMT Val158Met (rs4680) genetic polymorphism may have influenced epigenetic regulation and, therefore, decreased activity of COMT, contributing to a deleterious effect in adults [81], Cote et al. [65] found no significant association in children. Whereas Theisen et al. [55] retrieved no association between the 5-HT2C receptor gene (rs3813929) polymorphism and clozapine-induced weight gain, Godlewska et al. [62] found a protective effect of -759T and -697C alleles. In antipsychotic-naive patients, Houston et al. [82] did not find similar associations. However, highlighting the possible association of DRD2 polymorphisms with increased weight gain, their findings supported Lencz et al.'s [64] conclusions. Otherwise, while our query yielded one study addressing the role of HLA gene variations in DILI (Ocete-Hita et al.) [79], we did not retrieve similar approaches regarding clozapine-induced neutropenia and agranulocytosis that formerly have been investigated [83].
Cytochromes genotyping (and phenotyping) was the preferred approach when investigating ADRs, especially in pediatric studies. Studies relying on large sample size underlined increased weight gain [51], prolactin levels [77], risk of EPS [71], and impaired treatment response [52] in patients deprived from at least one functional allele for CYP2D6, resulting in increased drug exposure. While the findings regarding movement abnormalities and lack of therapeutic effect concur with existing evidence [84,85], AIWG [86] and hyperprolactinemia [87] were not consistently linked with CYP2D6 impairments. However, olanzapine is mostly metabolized by CYP1A2 (and to a lesser extent by CYP2D6 and CYP3A4) [88,89], clozapine is mainly metabolized by CYP3A4 and CYP1A2 (with CYP2D6 playing a minor role) [16,90], and loxapine is primarily metabolized by CYP1A2 (then by CYP3A4 and CYP2D6) [19]. Despite the fact that Menus et al. [61] found no association between CYP1A2 expression and any ADR, some variants have been formerly linked to tardive dyskinesia [91,92] and to an increased risk of insulin and lipid elevation [93].
Indeed, some of these discrepancies may originate from several limitations of the evidence included in our review. First, we chose to focus on studies involving children and youth, often characterized by smaller samples and thus lack of power to show an existing difference, and lower-evidence study designs (case reports/series). Several large cohorts were (at least partially) overlapping, therefore lowering the total size of the investigated population. Second, we aimed to assess the pharmacogenetic causes of ADRs related to olanzapine, clozapine, and loxapine, whereas several of our largest sample size studies investigated atypical APs indiscriminately. Furthermore, Thümmler et al. [3] only reported a case of patients treated with loxapine, which may be due to French-specific prescription behaviors [23,24]. Third, apart from metabolic changes, ADRs were subject to heterogeneous outcome measurements (EPS, clinical improvement), which may have prevented us from direct comparisons between different studies. Fourth, most studies lacked consideration for potential interacting factors with AP-induced side effects, such as co-treatments, inflammation, weight change, dietary habits, smoking, and/or consumption of caffeine. These factors may be prevailing, especially in transitioning-age youths, and are important to consider. Fifth, our quality assessment of the studies (see Methods), relying on a tool adapted from the checklist by Jorgensen and Williamson [50], yielded an average score of 11.3/24. Overall, some issues of concern were the lack of information upon quality control methods, handling of missing data, and population stratification. In studies including children and youth only, lack of adjustments for multiple testing and of HWE testing were frequent additional flaws, therefore lowering the mean quality score of these studies (9.1/24). Furthermore, the quality assessment tool we relied on may be used as a checklist for further pharmacogenetic studies, to improve the comprehensiveness of the presented results.
In fact, in addition to proper pediatric studies, and considering the foreseeable scarce body of evidence among this population, we accepted to include studies involving at least one youth patient (see Methods) [44]. Thus, while broadening the study population, it may have lowered the impact of the children's metabolic characteristics. As stated above, the features of the included studies did not permit a strict comparison, preventing any meta-analysis. Nevertheless, our grouping strategy, relying on the main ADR classes (see Methods), enabled qualitative assessments. As a flaw inherent to systematic reviews, reporting bias limits the interpretation of our findings, even if several studies showed negative results. Furthermore, as the overall quality of evidence could not be estimated with reference methods such as GRADE [94], the methodological quality of our included pharmacogenetic studies was assessed via a tool adapted from the checklist of Jorgensen and Williamson [50] (see Methods). Then, a quality assessment was conducted among pediatric and mixed-population studies, allowing us to detect the main issues of concern in each study category. For each database query, the two screening steps and the quality scoring were subject to a dual assessment (D.M. and A.O.G.), which may have limited sources of bias.
While findings in children and youth pharmacogenetics are conflicting regarding olanzapine, clozapine, and loxapine, the benefits of genotyping in clinical use may be limited by lack of sufficient evidence, the barriers to routine use, and overall impact [95]. However, the dose-effect relationship is significantly influenced by cytochromes, holding sway over exposure to the medication [96]. Yet, in comparison with CYP2D6, CYP1A2 remains less investigated, while olanzapine and clozapine's ADRs are serious. Furthermore, cases of major clinical improvement were fostered by CYP1A2 genotyping [11], although its benefit is not collective yet. The use of advanced technologies, such as WGS, might provide an interesting complement, broadening the research spectrum in psychiatric disorders [40,41]. From this perspective, further studies addressing the cytochromes' and other genes' (involved in energy homeostasis, metabolism, neurotransmission inter alia) impact should consider potential polypharmacy and intercurrent modifications in the metabolism of children and youth. Further studies may provide insights into possible cross-talks between the pathways associated with ADRs and GABA-A signaling, identifying new drug targets and therefore paving the way for the development of new antipsychotic drugs with variable receptor affinities. These drugs could constitute alternatives to thienobenzodiazepines, dibenzodiazepines, and dibenzoxazepines, and improve the acceptability of treatments. Phenotypical variations due to ancestry and/or infrequent cytochrome variants should also be taken into account by studying larger pediatric samples that originate from different countries. Determined by genetics, but influenced by the environment, CYP1A2 and its interactions should be further investigated, to improve assessment of the risk-benefit balance in children and youth treated with olanzapine, clozapine, and loxapine.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/ph15060749/s1, Table S1 Quality assessment of included pediatric studies; Table S2 Quality assessment of included mixed population studies.