Significant outcomes
When GWAS fails to be significant, limited sample sizes can still prove useful using a hypothesis-driven molecular pathway approach.
Already known genes involved in obesity may compose a molecular pathway at risk of excessive AIWG.
New genes (MRAP, SLC19A3, GPD1, NLK, ASH2L, NCOA6, PTPN1, LEPROT, GSTK1, ADIPOR1, MRAP2, RBBP5, WDR5, JAK2, CHD7, SDC3, SETDB1, PPARG and PCKS1) not previously associated with AIWG have been identified.
Limitations
With a limited sample size, some potential relevant genetic variations able to influence weight gain could have been undetected because of their limited impact on the phenotype investigated.
The CATIE study was not designed to test the hypothesis under analysis, that is, only limited information on weight gain pattern, height (hence no BMI information) and diet, etc. was available.
The definition of the molecular pathways is based on the current biological knowledge of gene function and our selection of target genes. Future studies may discover that genes included in the pathway backbone may not be functionally relevant after all, just as it cannot be excluded that potentially relevant genes may have been omitted in the selection process.
Introduction
According to WHO, 21 million people worldwide were affected by schizophrenia (SCZ) in 2016. It is estimated that the total amount of lost disability-adjusted life year caused by SCZ will be close to 17 million by 2020 (Murray & Lopez, Reference Murray and Lopez1996). SCZ affects both quality of life and lifespan (Solanki et al., Reference Solanki, Singh, Midha and Chugh2008). Drug-induced metabolic disorders likely contribute to the reduced lifespan (11–20 years less) and high incidence of cardiovascular disorders in SCZ (Correll et al., Reference Correll, Manu, Olshanskiy, Napolitano, Kane and Malhotra2009; Scigliano & Ronchetti, Reference Scigliano and Ronchetti2013; Kredentser et al., Reference Kredentser, Martens, Chochinov and Prior2014). The mainstay for the treatment of SCZ is second-generation antipsychotics (SGA). Weight gain, dyslipidaemia, type II diabetes and metabolic changes are common side effects of SGA. SGA may differ within their group in terms of metabolic liability; however, no matter the drug choice, it is an expected, problematic side effect (Patel et al., Reference Patel, Buckley, Woolson, Hamer, McEvoy, Perkins and Lieberman2009; Bak et al., Reference Bak, Fransen, Janssen, van Os and Drukker2014; Bressington et al., Reference Bressington, Mui, Tse, Gray, Cheung and Chien2016). Possible mechanisms that drive the antipsychotic-induced weight gain (AIWG) are many, involving interactions with serotonin, histamine, dopamine, adrenergic, cannabinoid and muscarinic receptors (Roerig et al., Reference Roerig, Steffen and Mitchell2011). Several studies support that the extend of AIWG may at least, in part, be genetically driven rendering certain individuals more susceptible than other (MacNeil & Müller, Reference MacNeil and Müller2016; Zhang et al., Reference Zhang, Lencz, Zhang, Nitta, Maayan, John, Robinson, Fleischhacker, Kahn, Ophoff, Kane, Malhotra and Correll2016). Nevertheless, more than every second patient will experience an excessive weight gain (>7%) when exposed to common treatments like olanzapine, quetiapine and risperidone (McEvoy et al., Reference McEvoy, Lieberman, Perkins, Hamer, Gu, Lazarus, Sweitzer, Olexy, Weiden and Strakowski2007). The biological and genetic mechanisms driving excessive AIWG are not fully understood, suggesting a phenotype of a more complex genetic architecture. If the genetic make-up for AIWG were identified, protocols could be implemented to reduce the risk, with beneficial outcome for patients and society. An amount of overlap between the genes related to AIWG and genes related to obesity in general can be observed, for example, SLC6A14, 5HTR2C, MC4R, etc. (Shams & Müller, Reference Shams and Müller2014; Miranda et al., Reference Miranda, Vetter, Genro, Campagnolo, Mattevi, Vitolo and Almeida2015; Lotta et al., Reference Lotta, Mokrosiński, Mendes de Oliveira, Li, Sharp, Luan, Brouwers, Ayinampudi, Bowker, Kerrison, Kaimakis, Hoult, Stewart, Wheeler, Day, Perry, Langenberg, Wareham and Farooqi2019). However, the overlap seems to go largely unnoticed as studies regarding AIWG are often contradicting at best (see Tables 1 and 2). The genetic make-up of common obesity includes alternations in genes related to energy metabolism, fatty tissue and hypothalamic function; such as: ADIPOQ, FTO, LEP, LEPR, INSIG2, MC4R, PCSK1 and PPARG (Walley et al., Reference Walley, Asher and Froguel2009; Choquet & Meyre, Reference Choquet and Meyre2011).
Table 1. Genes previously found to be associated with obesity
Table 2. Previous reports of SNPs and their association with AIWG
ADIPOQ, Adiponectin, C1Q And Collagen Domain Containing; FTO, FTO, Alpha-Ketoglutarate-Dependent Dioxygenase; LEP, Letpin; LEPR, Letpin receptor; INSIG2, Insulin-Induced Gene 2; MC4R, Melanocortin 4 Receptor; PCSK1, Proprotein Convertase Subtilisin/Kexin Type 1; PPARG, Peroxisome Proliferator-Activated Receptor Gamma.
In this study, we hypothesise that knowledge of obesity-prone genes could have value for the identification of AIWG-prone individuals. We test the hypothesis that the genetic variations associated with common obesity might pose an increased risk of excessive AIWG. We provide new information from a sample size too small for a genome-wide association study (GWAS) approach such as the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) trial.
Methods
The CATIE trial
The sample under analysis is the NIMH CATIE sample (NIMH contract NO1 MH90001). A total of 1460 schizophrenic patients were enrolled between January 2001 and December 2004. CATIE was a multi-phase randomised controlled trial of antipsychotic medications (olanzapine, perphenazine, quetiapine, risperidone or ziprasidone) involving individuals (Table 3) with SCZ followed for up to 18 months (Stroup et al., Reference Stroup, McEvoy, Swartz, Byerly, Glick, Canive, McGee, Simpson, Stevens and Lieberman2003; Lieberman et al., Reference Lieberman, Stroup, McEvoy, Swartz, Rosenheck, Perkins, Keefe, Davis, Davis, Lebowitz, Severe and Hsiao2005).
Table 3. Sample descriptions
(F = female; M = male; when the ± symbol is present, the mean and the standard deviation are reported on the left and on the right, respectively) – in order to conduct the covariate analysis, clinical variables found to be significantly associated with the outcome under analysis were used as covariates in the model – the statistics (T = student; R = correlation; F = ANOVA). Values are reported in bold when significantly associated with the outcome under analysis.
Fifty-one per cent (765 individuals, male = 556, mean age = 40.93 ± 11.03) of the CATIE participants provided a DNA sample and is the core data set of the present investigation. DNA samples were sent to the Rutgers University Cell and DNA Repository, where cell lines were established by Epstein–Barr virus (EBV) transformation. Sample DNA concentrations were quantified and normalised using PicoGreen dsDNA quantification kits (Molecular Probes, Eugene, OR, USA).
The analysis was conducted through the following steps.
GWAS
Outcome for the first-step analysis was maximum weight gain recorded during the CATIE. Clinical covariates were identified though a general linear model and were included in the genetic analysis as stratification factors when significantly associated with the outcome under analysis (Table 3). Quality checking was set as standard for this type of analysis (genotype call rate > 0.95; maf > 0.01; hwe < 0.0001), inflation factor was controlled by lambda values and imputation was run with the use of 1000 Genomes in a Plink environment. Pathway analysis was conducted at the SuperCluster PC at Aarhus University. The single variations associated with weight gain were identified (a nominal p threshold of 0.01 was chosen) together with the genes that harboured those variations. Genes were identified through the interrogation of the public available genetic data set within the specifications of R packages ReactomePA (Yu & He, Reference Yu and He2016), Bioconductor (Huber et al., Reference Huber, Carey, Gentleman, Anders, Carlson, Carvalho, Bravo, Davis, Gatto, Girke, Gottardo, Hahne, Hansen, Irizarry, Lawrence, Love, MacDonald, Obenchain, Oleś, Pagès, Reyes, Shannon, Smyth, Tenenbaum, Waldron and Morgan2015), biomaRt (Drost & Paszkowski, Reference Drost and Paszkowski2017) and GenABEL (Aulchenko et al., Reference Aulchenko, Ripke, Isaacs and van Duijn2007). Plink (Purcell et al., Reference Purcell, Neale, Todd-Brown, Thomas, Ferreira, Bender, Maller, Sklar, de Bakker, Daly and Sham2007) served for the genetic association test, default settings. For a full walk-through of the methods used, please see the Supplementary Material.
Molecular pathway analysis
As a second step, genes previously related to obesity were subjected to an enrichment analysis. Genes classically associated with obesity (metabolism, fatty tissue function and hypothalamic regulation and the hypothalamic-pituitary-adrenal (HPA) axis) were chosen to create a backbone of the molecular pathway (Table 1). The genes shown in Table 1 were used as input for Cytoskape, GeneMANIA and further enriched the original pathway. As a result, a complete molecular pathway was identified by Cytoskape and then tested for enrichment in the data set.
SNPs relating to the Cytoskape-generated pathway
The enrichment analysis was conducted using the R software suite (R Foundation for Statistical Computing, 2013), through the packages Bioconductor (Huber et al., Reference Huber, Carey, Gentleman, Anders, Carlson, Carvalho, Bravo, Davis, Gatto, Girke, Gottardo, Hahne, Hansen, Irizarry, Lawrence, Love, MacDonald, Obenchain, Oleś, Pagès, Reyes, Shannon, Smyth, Tenenbaum, Waldron and Morgan2015) and ReactomePA (Yu & He, Reference Yu and He2016). The analysis of clinical covariates was conducted prior to the genetic tests and, when found significantly associated with the phenotype (age, years of treatment, years at the moment of presentation) under analysis was included as covariates for genetic tests.
Permutation test to confirm the validity of our findings
As a third step of the analysis, a permutation test was conducted by selecting 105 molecular pathways randomly identified in the CATIE genetic database and testing their enrichment against the index pathway (p = 0.05) generated by Cytoskape. A permutation test was deemed mandatory in order to abate the risk of false-positive findings. The selected molecular pathways had the same length and the same number of SNPs as the index pathway in order to limit the possibility of bias selection.
Result
Mean weight gain was 6.48 ± 7.50 kg throughout the study, independently from the drug treatment (olanzapine, perphenazine, quetiapine, risperidone or ziprasidone) delivered to single individuals. Significant covariates were age, years of treatment and years at the moment of presentation which were all associated with increased weight gain (see Table 3).
GWAS analysis
Initially, a set of 495.172 SNPs were available from the CATIE study, after pruning and quality assessment 170.841 SNPs were imputed, resulting in a total of 4.268.977 SNPs. As a result, from the first step of the analysis, none of the variations under analysis reached a genome-wide significant level when tested for weight gain throughout the trial. The Manhattan plot illustrates that no SNP was of significance; however, a trend of association was observed for both rs822391 (ADPIOQ) and rs2071045 (LEP). A lambda value of 1.002 allowed for the ruling out of major stratification factors. From the Q–Q plot, a good correlation between expected values and observed values indicates a high-quality data, see Fig. 1 for the result of the GWAS and Fig. 2 for the Q–Q plot.
Fig. 1. GWAS results of Manhattan plot. It is illustrated that no single SNP was of significance; however, a trend of association was observed for both rs822391 (ADPIOQ) and rs2071045 (LEP).
Fig. 2. Q–Q plot analysis. The R qqman package (Turner, 2017) was instrumental for creating the picture. Estimated λ for the p distribution was 1002. This was calculated using the GenABEL package, ‘estlambda’ function (Aulchenko et al., Reference Aulchenko, Ripke, Isaacs and van Duijn2007).
Hypothesis-driven enrichment analysis
ADIPOQ, LEP, LEPR, PPARG, FTO, MC4R, PCKS1 and INSIG2 are considered classical genes associated with obesity (Hinney & Hebebrand, Reference Hinney and Hebebrand2008; Enns et al., Reference Enns, Taylor and Zahradka2011; Sarzynski et al., Reference Sarzynski, Jacobson, Rankinen, Carlsson, Sjöström, Bouchard and Carlsson2011; Kasim et al., Reference Kasim, Huri, Vethakkan, Ibrahim and Abdullah2016). Based on the 8 backbone genes, a molecular pathway was created with 20 new genes giving a total of 28 genes (Fig. 3 and Table 4): MRAP, SLC19A3, GPD1, NLK, POMC, ASH2L, NCOA6, PTPN1, LEPROT, GSTK1, ADIPOR1, MRAP2, RBBP5, WDR5, AGRP, SOCS3, JAK2, CHD7, SDC3 and SETDB1. A total of 2067 SNPs were harboured by genes belonging to the pathway under analysis. The prevalence of variations significantly associated with the outcome under analysis (significance level set at 0.01, not GWAS significance level) was higher than expected by chance (n = 44, expected = 21 (= 0.01 × 2067)).
Fig. 3. The molecular pathway under analysis: A graphic representation by clusters of molecular pathways associated with weight gain. The picture provides a representation of the molecular pathways that are in strict functional association with those found to be associated with weight gain during antipsychotic treatment. Dark circles are the initial eight genes, where the grey circles are suggested genes by Cytoscape and their molecular pathways.
Table 4. Genes under analysis in the molecular pathway
Table generated using R to access gene_ensembl.
Table 5 reports a selection of SNPs associated with the hypothesis-driven genes found in the CATIE sample associated with the weight gain. These SNPs can be prioritised in further research for their functional role and their association with the risk of gaining weight when patients are treated with antipsychotics.
Table 5. Single-nucleotide polymorphisms of the chosen genes, with the strongest statistical association with weight gain in the CATIE sample
ECM, extracellular matrix; SNP, single-nucleotide polymorphism.
Variations with a significant impact on expression adipose tissue and with a strong statistically significant association with the weight gain in CATIE trials. EQTLs data were retrieved from https://www.gtexportal.org/.
Confirmation/permutation analysis
As a result, from the third step of analysis, the higher prevalence of SNPs associated with excessive AIWG retrieved from the second step of analysis resisted the permutation test (a permutated p-value = 0.05 for enrichment was retrieved from the pathway under analysis).
Discussion
With a molecular pathway approach, we were able to show a significant enrichment in variations harboured by a selected group of genes with relevance to common obesity. The results strengthen the hypothesis that genetic pre-disposition for obesity may increase the risk of excessive AIWG.
From the molecular pathway generated, the genes ADIPOQ, INSIG2, LEP, LEPR, MC4R, FTO, AGRP, POMC and SOCS3 (Piao et al., Reference Piao, Park, Li, Shin, Shin, Kong, Shrestha, Tran, Hur, Kim and Park2014) all have previously been associated with AIWG. PCSK1 and PPARG showed to be significantly enriched in variations in the current molecular pathway and although associated with weight gain and obesity, this is to our knowledge the first time PCSK1 and PPARG are associated with AIWG. PCSK1 has a crucial importance on the appetite and metabolism regulation, as the protein (PC1/3) coded by PCSK1 converts the proteins of proinsulin and proglucagon to their active biological forms (Rouillé et al., Reference Rouillé, Kantengwa, Irminger and Halban1997). Genetic variations in PCSK1 have been associated, although heavily debated, with sensitivity to obesity and in particular childhood obesity, severe obesity and race (Choquet & Meyre, Reference Choquet and Meyre2011; Choquet et al., Reference Choquet, Kasberger, Hamidovic and Jorgenson2013; Kulanuwat et al., Reference Kulanuwat, Phonrat, Tungtrongchitr, Limwongse, Chongviriyaphan, Tungtrongchitr and Santiprabhob2014; Dušátková et al., Reference Dušátková, Zamrazilová, Aldhoon Hainerová, Atkinson, Sedláčková, Lee, Včelák, Bendlová, Kunešová and Hainer2015; Kulanuwat et al., Reference Kulanuwat, Santiprabhob, Phonrat, Limwongse, Tungtrongchitr, Chongviriyaphan and Tungtrongchitr2015; Nordang et al., Reference Nordang, Busk, Tveten, Hanevik, Fell, Hjelmesæth, Holla and Hertel2017). In relation to AIWG, the presence of a catecholamine like dopamine or antipsychotics like haloperidol has shown to influence the levels of PC1/3 and could point towards a possible connection with AIWG not yet discovered (Day et al., Reference Day, Schafer, Watson, Chrétien and Seidah1992; Oyarce et al., Reference Oyarce, Hand, Mains and Eipper1996; Helwig et al., Reference Helwig, Vivoli, Fricker and Lindberg2011). PPARG encodes a member of the peroxisome proliferator-activated receptor subfamily of nuclear receptors. PPARG plays a key role in the regulation of lipid and glucose metabolism (Franks et al., Reference Franks, Jablonski, Delahanty, Hanson, Kahn, Altshuler, Knowler and Florez2007). It is assumed that PPARG expresses its role in regulation of weight through a modulation of genes associated with body weight homeostasis and insulin signalling (Lyche et al., Reference Lyche, Nourizadeh-Lillabadi, Karlsson, Stavik, Berg, Skåre, Alestrøm and Ropstad2011). PPARG is a main regulator for the development of adipose cells as well as a factor in the pathology of numerous diseases including obesity, atherosclerosis and cancer (Tontonoz & Spiegelman, Reference Tontonoz and Spiegelman2008; Hetherington & Cecil, Reference Hetherington and Cecil2010). Frequently used medicines that target the receptor cover areas of diabetes mellitus, atherosclerosis and lately as an anticancer drug (Dang et al., Reference Dang, Jiang, Gong and Guo2018). Although variations in PPARG are well documented for influencing weight gain, its contribution to AIWG seems so far to be less obvious (Brandl et al., Reference Brandl, Tiwari, Zai, Chowdhury, Lieberman, Meltzer, Kennedy and Müller2014). This could be due to the complexity of gene–gene interactions. For instance, through inhibition of PPARG, a reduction of LEP expression takes place, when the patient is administered the antidiabetic thiazolidinediones (Catalano et al., Reference Catalano, Mauro, Bonofiglio, Pellegrino, Qi, Rizza, Vizza, Bossi and Andò2011). Alternatively, activation of PPARG and the membrane receptor GRP120 ameliorates the adipose inflammation and insulin resistance caused by local hypoxia in adipocytes (Hasan et al., Reference Hasan, Ohmori, Konishi, Igarashi, Hashimoto, Kamitori, Yamaguchi, Tsukamoto, Uyama, Ishihara, Noma, Tokuda and Kohno2015). Hence, in the case of PPARG, obesity and AIWG are not a simple activation/inhibition model.
Interestingly, of the 20 new genes identified, 3 (SOCS3, AGRP and POMC) had previously been associated with AIWG. Suppressor of cytokine signalling 3 (SOCS3) is a negative regulator of leptin signalling (Piao et al., Reference Piao, Park, Li, Shin, Shin, Kong, Shrestha, Tran, Hur, Kim and Park2014). The protein from agouti-related neuropeptide (AGRP) antagonises melanocortin-4 and melanocortin-4 receptors, thereby regulating the hypothalamic feeding behaviour, previously associated with obesity and obesity susceptibility (Krashes et al., Reference Krashes, Koda, Ye, Rogan, Adams, Cusher, Maratos-Flier, Roth and Lowell2011). POMC encodes a preproprotein, and deficiencies in this gene are associated with obesity and adrenal insufficiency. A complex protein depending on its cleavage site influences energy metabolism, pigmentation and inflammation. Deficiencies in this gene among others are associated with obesity and adrenal insufficiency. Interestingly, up/down regulations of POMC and AGRP seem to depend on acute/chronic administration and drug choice (Fernø et al., Reference Fernø, Varela, Skrede, Vázquez, Nogueiras, Diéguez, Vidal-Puig, Steen and López2011; Ehrlich et al., Reference Ehrlich, Leopold, Merle, Theophil, Haag, Lautenschlager and Schaefer2012; Weston-Green et al., Reference Weston-Green, Huang and Deng2012; Kursungoz et al., Reference Kursungoz, Ak and Yanik2015; Lian et al., Reference Lian, De Santis, He and Deng2015; Rojczyk et al., Reference Rojczyk, Pałasz and Wiaderkiewicz2015), and the reason behind this observation could be due to different experimental set-ups or perhaps that antipsychotics need to be administered at threshold levels to influence the anorexigenic POMC, while AGRP may be affected earlier or more potent by other factors, for example, leptin or reduced response from MC4R. No clear SNPs significance to AIWG has been observed (Chowdhury et al., Reference Chowdhury, Souza, Tiwari, Brandl, Sicard, Meltzer, Lieberman, Kennedy and Müller2014). So far, this leaves an unclear picture of function and relevance of POMC and AGRP in regard to AIWG.
In the molecular pathway generated, FTO is the sole gene, not directly connected to the other genes. The reason for this could either be that FTOs full function yet to be understood or that the mechanisms behind its role in obesity are more elusive than that of the other genes in the molecular pathway. FTO encodes the FTO alpha-ketoglutarate-dependent dioxygenase exact, and the physiological function of this gene is not completely known. Evidence links FTO with BMI, obesity and a possible participation in central energy homeostasis, as an upregulation of 41% of FTO in the hypothalamus is observed in food-deprived rats (Fredriksson et al., Reference Fredriksson, Hägglund, Olszewski, Stephansson, Jacobsson, Olszewska, Levine, Lindblom and Schiöth2008; Zhao et al., Reference Zhao, Yang, Sun, Zhao and Yang2014). It is, however, known that an increased function of FTO is associated with obesity in both humans and animals, whereas a reduced function results in growth retardation (Yeo, Reference Yeo2014; Zhao et al., Reference Zhao, Yang, Sun, Zhao and Yang2014). The FTO genotype has a major effect on body weight in chronically treated patients with SCZ, something not observed in first-episode patients (Reynolds et al., Reference Reynolds, Yevtushenko, Gordon, Arranz, San and Cooper2013). One could speculate that the inconsistent findings reflect a biphasic correlation with other factors throughout the life of patients, for example, between alleles and BMI (Hardy et al., Reference Hardy, Wills, Wong, Elks, Wareham, Loos, Kuh and Ong2010).
The hypothesis-driven pathway was emulated from current knowledge in the field. The genes selected for the backbone of the molecular pathway are in concordance with previous articles of similar type (Enns et al., Reference Enns, Taylor and Zahradka2011; Sarzynski et al., Reference Sarzynski, Jacobson, Rankinen, Carlsson, Sjöström, Bouchard and Carlsson2011; Kang et al., Reference Kang, Lee, Han, Soh and Hong2014). Genes were chosen that are related to energy metabolism, feeding behaviour, adipose tissue and leptin response, but excluding genes with inconclusive reports and genes belonging to vast molecular pathways, for example, tropic factors SH2B1 (Choquet & Meyre, Reference Choquet and Meyre2011). While our molecular pathway gives statistical sound evidence, the embracement of an increasing number of genes and their variations will at the same time escalate the complexity in how the variations relate to functional changes. This makes it possibly less useful in a clinical setting, at least at the current time.
Simultaneously, a hypothesis-driven molecular pathway approach is of course subjected to selection bias as a major limitation. The bias will have a high impact on the results and a molecular pathway analysis is by itself a less precise tool than, for example, GWAS. With this in mind, and based on the assumption of a polygenetic risk of drug-induced weight gain and the limited power of the sample, as expected, no single SNP was significant when using a GWAS approach. A trend of association was observed for both rs2071045 (LEP, p = 0.091) and rs822391 (ADPIOQ, p = 0.078) (Fig. 1). These findings are of particular relevance, as LEP was previously found to be associated with weight gain in animal and human investigations (Erez et al., Reference Erez, Tirosh, Rudich, Meiner, Schwarzfuchs, Sharon, Shpitzen, Blüher, Stumvoll, Thiery, Fiedler, Friedlander, Leiterstdorf and Shai2011); however, more studies lean towards the idea that the contribution from polymorphisms in LEP may be of minor relevance for AIWG (Creta et al., Reference Creta, Fabbri and Serretti2015; Klemettilä et al., Reference Klemettilä, Kampman, Seppälä, Viikki, Hämäläinen, Moilanen, Mononen, Lehtimäki and Leinonen2015; Vasudev et al., Reference Vasudev, Choi, Norman, Kim and Schwarz2017).
The rs822391 SNP in ADPIOQ (adiponectin) has been related to body size in pre-menopausal women (Slattery et al., Reference Slattery, Lundgreen, Hines, Wolff, Torres-Mejia, Baumgartner and John2015). Other ADPIOQ SNPs studies have showed an association with AIWG, but nothing of significance (Jassim et al., Reference Jassim, Fernø, Theisen, Haberhausen, Christoforou, Håvik, Gebhardt, Remschmidt, Mehler-Wex, Hebebrand, Lehellard and Steen2011; Wu et al., Reference Wu, Zhao, Shao, Ou and Chang2011; Brandl et al., Reference Brandl, Tiwari, Zai, Chowdhury, Lieberman, Meltzer, Kennedy and Müller2014). However, ADIPOQ may be affected directly by drugs, and independent reports show a negative correlation between weight gain and serum adiponectin (ADPIOQ) when treating with neuroleptica in vivo (Cuerda et al., Reference Cuerda, Merchan-Naranjo, Velasco, Gutierrez, Leiva, de Castro, Parellada, Giráldez, Bretón, Camblor, García-Peris, Dulín, Sanz, Desco and Arango2011; Soeiro-de-Souza et al., Reference Soeiro-de-Souza, Gold, Brunoni, de Sousa, Zanetti, Carvalho, Gattaz, Machado-Vieira and Teixeira2014). Similar was shown in vitro, when stimulating cultured adipose cells with imipramine or lithium at therapeutic levels (Löffler et al., Reference Löffler, Landgraf, Körner, Kratzsch, Kirkby and Himmerich2016). To summarise, the data for ADIPOQ and its alleles are conflicting at best as well as when serum levels of ADPIOQ are monitored directly. Possible explanation for the discrepancies could be lack of multiple testing in previous studies, different ethnicities between studies and the heterogeneities of samples (Jassim et al., Reference Jassim, Fernø, Theisen, Haberhausen, Christoforou, Håvik, Gebhardt, Remschmidt, Mehler-Wex, Hebebrand, Lehellard and Steen2011; Wu et al., Reference Wu, Zhao, Shao, Ou and Chang2011; Li et al., Reference Li, Xu, Tian, Wang, Jiang, Zhang, Wang, Yang, Gao, Song, He, Zhang, Li and Li2017).
Lastly, some of the variations found to be significantly associated with the phenotype under analysis were also reported to have a functional role in the expression of the genes they are harboured by, in the adipose tissue. This finding may be interpreted as a consistent and independent report of the relevance of a list of SNPs in driving the AIWG. The results are reported in Table 5.
Conclusion
The results from the current contribution confirm that variations in the genes related to obesity and the molecular pathway composed from them increase the risk of AIWG. Despite decades of research in genetics, we are currently not able to identify patients at risk of excessive AIWG, supporting the notion of a complex polygenetic nature of this phenomenon. Even in larger studies, the panoramic view of GWAS has so far provided limited evidence for easily identifiable phenotypes. As to why this is, one could speculate that the AIWG phenotype is seen in individuals where the total genetic make-up, from combined variations in metabolically important genes, exceeds a ‘threshold’. In which case, it would be impossible to locate a SNP ‘culprit’. This is similar to what is generally known of pharmacological effect/side effect, that is, 20–40% of all side effects are caused by gene variations identifiable in just one individual (Ingelman‐Sundberg, Reference Ingelman‐Sundberg2001). Therefore, as shown in the current work, a molecular pathway approach may be a more proficient tool for describing complex polygenetic phenotypes even with limited sample sizes. Hence, downstream effects when introducing medicine to genetic alterations become increasingly complex.
Supplementary material
For supplementary material for this article, please visit https://doi.org/10.1017/neu.2019.41
Acknowledgements
The authors thank the researchers involved in the initial CATIE clinical trial (NIMH NO1 MH90001), led by Jeffrey A. Lieberman, M.D., T. Scott Stroup, M.D., M.P.H. and Joseph P. McEvoy, M.D. The CATIE trial was funded by a grant from the National Institute of Mental Health (N01 MH900001) along with MH074027 (PI PF Sullivan). Genotyping was funded by Eli Lilly and Company. They also thank the CATIE participants and their families. The authors thank the genome unit at the Aarhus University for sharing the computational power necessary for the calculations in the present contribution (http://genome.au.dk/).
Authors contributions
HTC contributed to the writing of the introduction, to the discussion and interpretation of the final result and to the research of bibliography. BK contributed to the writing of the introduction and to the discussion. AD contributed to the research of the bibliography, to the statistical analysis of data and with AL and to the general supervision of the paper.
Financial support
The present contribution was made possible by the Psychiatric Research Unit West, Herning, Denmark.
Conflicts of interest
There are no conflicts of interest.
