The Role of Genetic Variation of BMI, Body Composition, and Fat Distribution for Mental Traits and Disorders: A Look-Up and Mendelian Randomization Study

Anthropometric traits and mental disorders or traits are known to be associated clinically and to show genetic overlap. We aimed to identify genetic variants with relevance for mental disorders/traits and either (i) body mass index (or obesity), (ii) body composition, (and/or) (iii) body fat distribution. We performed a look-up analysis of 1,005 genome-wide significant SNPs for BMI, body composition, and body fat distribution in 15 mental disorders/traits. We identified 40 independent loci with one or more SNPs fulfilling our threshold significance criterion (P < 4.98 × 10–5) for the mental phenotypes. The majority of loci was associated with schizophrenia, educational attainment, and/or intelligence. Fewer associations were found for bipolar disorder, neuroticism, attention deficit/hyperactivity disorder, major depressive disorder, depressive symptoms, and well-being. Unique associations with measures of body fat distribution adjusted for BMI were identified at five loci only. To investigate the potential causality between body fat distribution and schizophrenia, we performed two-sample Mendelian randomization analyses. We found no causal effect of body fat distribution on schizophrenia and vice versa. In conclusion, we identified 40 loci which may contribute to genetic overlaps between mental disorders/traits and BMI and/or shape related phenotypes. The majority of loci identified for body composition overlapped with BMI loci, thus suggesting pleiotropic effects.

To address the association of obesity with mental disorders, both clinical and population-based studies have been conducted cross-sectionally and longitudinally (Richardson et al., 2003;Patalay and Hardman, 2019). Particularly well analyzed are associations between obesity and depression (Luppino et al., 2010;Berkowitz and Fabricatore, 2011;Mühlig et al., 2015;Milaneschi et al., 2019) and ADHD (Cortese et al., 2016;Nigg et al., 2016), respectively. For obesity and MDD, a bidirectional, partly sex dependent relationship applies, entailing that obesity can precede MDD and vice versa (Luppino et al., 2014;Mühlig et al., 2015;Coryell et al., 2016). An association with obesity was also reported for BD and schizophrenia (Gurpegui et al., 2012). Because of the adverse effect of weight gain inherent to several psychotropic drugs, especially atypical antipsychotics (Almandil et al., 2013), a reliable assessment of the respective associations can only be based on drug-naïve or previously untreated patients (Gurpegui et al., 2012). A meta-analysis for BMI and WHR in never or minimally treated patients with psychosis or schizophrenia found lower BMI and elevated WHR in patients compared to healthy controls. No differences in WC were observed (Shah et al., 2019).
Observed clinical relationships between anthropometric traits BMI, obesity, body composition, and fat distribution have partially been supported by the detection of significant genetic correlations; in addition, genetic correlations have gone beyond established clinical associations. Significant negative genetic correlations have been described for BMI (or obesity) with educational attainment, intelligence, and neuroticism as well as with schizophrenia and BD. Positive genetic correlations have been found between BMI (or obesity) and ADHD and depression (Bulik-Sullivan et al., 2015;Zheng, 2017;Hill et al., 2018;Richardson et al., 2019b) 1 : (a) The negative genetic correlations between body fat percentage (BF%) and educational attainment and schizophrenia should also be highlighted (Richardson et al., 2019b) At the functional level, several genes coding for proteins involved in synaptic function have been associated with substance use, obesity-related traits, responses to mental and physical stress, heart rate, and blood pressure measurements (Nikpay et al., 2012).
Thus, complex anthropometric and mental phenotypes may reveal extensive overlapping genetic factors. Research is clearly warranted to identify genetic variation involved both in the (1) regulation of body weight, body composition, and body fat distribution and (2) in mental disorders/traits. Particularly, body fat distribution is of great importance, because visceral fat (abdominal obesity) is the main factor underlying the clinical association between obesity and hypertension and type 2 diabetes mellitus (Snijder et al., 2006). Because stress is presumed to play an important role in the development of abdominal obesity (Björntorp, 2001;Geiker et al., 2018;van der Valk et al., 2018), as well as in the pathogenesis of mental disorders (Chrousos, 2009), a genetic link especially with body fat distribution may be particularly pronounced.
The purpose of the current study was to perform for the first time a systematic look-up analysis to assess the potential overlap of genome-wide significant loci for a range of anthropometric traits with mental disorders/traits. Methodologically similar to our previous study, which assessed the role of metabolite SNPs in mental disorders/traits (Hebebrand et al., 2018), we initially identified all genome-wide significant SNPs for the anthropometric traits. Look-ups were subsequently performed in GWAS data sets for 15 mental disorders/traits. For our initial look-up analysis we used data from the GWAS catalog in December 2017. We aimed for more detailed analyses in followup studies. Thus, to keep up with the rapid pace in the GWAS field, we have used the latest data set for each phenotype for our post hoc analyses [effect sizes and directions, Mendelian randomization; e.g., for BMI (Yengo et al., 2018), body fat distribution (Pulit et al., 2019) or schizophrenia (Pardinas et al., 2018), and MDD (Wray et al., 2018)].
The term "mental disorders" refers to disorders delineated in DSM-5 (American Psychiatric Association, 2013). Mental traits refer to quantitative personality/character, behavioral and socio-psychological traits. We use the overarching term mental phenotypes to refer to both mental disorders and mental traits. Apart from the genetic contribution, there are strong environmental, behavioral, and social components to some of these phenotypes (Meyer-Lindenberg and Tost, 2012).
To exemplarily investigate the potential causality of SNPs for body fat distribution in schizophrenia (and vice versa) in post hoc analyses, we performed a two-sample Mendelian randomization (MR) analysis with WHR adjBMI as exposure and the risk for schizophrenia as outcome. In the reverse MR we used the risk of schizophrenia (Richardson et al., 2019a) as exposure and WHR adjBMI as outcome.
The identified studies were independently reviewed by three authors (LL, TP, and JH) in order to exclude anthropometric traits that did not fit the purpose of the analysis. Single different ratings were resolved by consensus. This selection process resulted in the exclusion of some traits from further analyses with a focus on a disease (other than obesity), e.g., BMI in smokers, body mass in chronic obstructive pulmonary disease, BMI (change over time) in cancer, and osteoporosis.
To define independent loci we used a distance criterion of ±500 kb surrounding the middle SNP of a chromosomal locus (Locke et al., 2015). Consecutive numbers were assigned to each locus (Supplementary Table S4; for MDD loci see Supplementary Table S5). Furthermore, linkage disequilibrium (r 2 ) was checked within each locus and, in case of linkage equilibrium between two SNPs (D < 0.01, r 2 < 0.01) 5 , these were regarded as independent loci. Independent data sets for confirmatory look-up analyses were available for (a) educational attainment  [n = 111,114, downloaded on 4 December 2017 (Leslie et al., 2014), based on data from the CHARGE and COGENT consortia, and UK Biobank]. These study groups are independent of the 65 study groups included in the study of Okbay (2016b); and (b) neuroticism (Luciano et al., 2018; UK Biobank sample size; n = 329,821; 17 April 2018; Supplementary Table S3).
Genome-wide association studies for mental phenotypes and anthropometric traits have been conducted with study groups that mainly included people of European descent. However, single anthropometric studies were performed in Japanese (Okada et al., 2012;Akiyama et al., 2017), African (Monda et al., 2013;Graff et al., 2017), and East-Asian populations (Wen et al., 2014).
To provide the reader with a synopsis, Tables 1, 2 were derived from Supplementary Tables S4, S5, which illustrate the complete results of the look-ups. Tables 1, 2 depict one representative SNP for each study-wide significant locus, which in case of loci tagged by >1 SNP was identified by (1) if applicable, the highest number of mental phenotypes hits (P < 4.98 × 10 −5 ), (2) among these SNPs the one with the lowest P-value was chosen.
Based on supplementary files and summary statistics effect directions of the alleles for mental phenotypes and the anthropometric traits were described for each representative SNP (P < 4.98 × 10 −5 ) 6,7 . In addition, the effect directions of the alleles were included for each anthropometric trait, for which that particular locus showed at least one hit. For this purpose, summary statistics for HC, WC, WHR, HC adjBMI , WC adjBMI , WHR adjBMI (Shungin, 2015; European descent and sex-combined), BF% (Lu et al., 2016), and BMI (European descent; Locke et al., 2015;Yengo et al., 2018) were used.

Statistical Analyses
To control for the overall type I error rate in the look-up, Bonferroni correction was applied. We assumed a univariate multiple regression model for each mental phenotype and independence of SNPs. This entailed a Bonferroni correction for 1,005 tests (1,005 SNPs), resulting in a threshold P-value of 4.98 × 10 −5 . We did not prune SNPs in high LD to adhere to a conservative approach. For confirmatory analyses, the same P-value (4.98 × 10 −5 ) was considered as significant.

Mendelian Randomization
We performed a Mendelian randomization study to exemplarily investigate the causality between body fat distribution and schizophrenia. Body fat distribution data were obtained from the latest GWAS analysis (Pulit et al., 2019;694,649 individuals), which explained about 3.9% of the variance in WHR adjBMI . As 6 https://www.snpedia.com/index.php/SNPedia 7 https://www.ncbi.nlm.nih.gov/snp 8 https://portals.broadinstitute.org/collaboration/giant/index.php/GIANT_ consortium_data_files 9 ftp://ftp.ebi.ac.uk/pub/databases/gwas/summary_statistics/LuY_26833246_ GCST003435 data source for the risk for schizophrenia the GWAS analysis; with 40,675 cases and 64,643 controls (145 independent loci) was used (Pardinas et al., 2018). SNP-based heritability was estimated at 24% (Pardinas et al., 2018;Pettersson et al., 2019). We conducted a two-sample MR analysis with WHR as risk factor (exposure) and schizophrenia as outcome variable and the reverse analysis. In order to conduct a MR study, three core assumptions must be fulfilled (Haycock et al., 2016;König and Greco, 2018). Therefore, we conducted an F-test to test the weakness of the instrument (Del Greco et al., 2015;. We examined horizontal pleiotropy by estimating the intercept of Egger's regression . To test heterogeneity of the instrument variable, we used Cochran's Q-statistic. We carried out MR using two statistical methods to estimate overall ratio effects: the inverse-variance weighted method (IVW) and the MR-Egger method. IVW assumes that the all ratio estimates provide independent evidence on the causal effect and there is no pleiotropic effect. So IVW assumes that all genetic variants are valid instrumental variables. There is no intercept term in the regression model . In MR-Egger the intercept term is estimated as part of the analyses. The intercept term can be interpreted as the average pleiotropic effect of a genetic variant included in the analyses. A non-zero intercept from MR-Egger shows that there is directional pleiotropy, or that instrumental variable assumptions are violated, or both . Consideration of MR-Egger estimates and intercept terms, and comparison of IVW and MR-Egger estimates is helpful to interpret results of a MR analysis .
We performed a power analysis to estimate whether our analysis, given sample size, proportion of cases in the study and the proportion of variance explained, is sufficient to detect a true casual effect (Brion et al., 2013). In order to investigate the relationship between study accuracy and effect size, we created a funnel plot (Haycock et al., 2016;Katikireddi et al., 2018). To examine whether an individual data point (SNP) has a large influence on the regression coefficients, we calculated the IVW regression by leaving each genetic variant out in turn . To visualize the results, forest and scatter plots were used. We also performed MR with Robust Adjusted Profile Score (RAPS), which estimator is more robust to pleiotropy (Zhao et al., 2019). All tests were performed using the software "R, " version 3.5.2 and R-packages for performing 2-sample MR 10 and RAPS. 11,12
The seven of the 40 loci that were not identified via BMI SNPs pertained to different measures of body fat distribution adjusted for BMI (WC adjBMI , HC adjBMI , and/or WHR adjBMI ; Table 2). All but one locus were associated with schizophrenia.
We analyzed effect directions of all alleles of significant SNPs (P < 4.98 × 10 −5 ). For the SNPs in Table 1, we looked up the effect directions for all phenotypes in respective locus (BMI, body composition, body fat distribution, and mental phenotypes) in the respective summary statistics (Locke et al., 2015;Shungin, 2015;Lu et al., 2016;Yengo et al., 2018). As expected, the effect directions were always identical for BMI, obesity, non-adjusted HC and WC, respectively. Table 3 provides an overview of the effect directions of all detected significant (P < 4.98 × 10 −5 ) SNP associations separated according to associations with BMI (irrespective of additional associations to other traits; for loci see Table 1) or with unique associations pertaining to BMI adjusted measures (for loci see Table 2). Opposite effect directions of alleles associated with a higher BMI were observed twice as often for educational attainment, intelligence, and schizophrenia than concordant directions.

Confirmatory Analyses
We performed confirmatory look-up analyses for educational attainment and neuroticism in the most recent and independent GWAS datasets for SNPs that fulfilled our P-value threshold of P < 4.98 × 10 −5 . No independent data sets were available for other mental phenotypes (Supplementary Table S3). Twenty two of the 38 SNPs for educational attainment revealed P-values below P < 4.98 × 10 −5 (10 of 15 loci). For neuroticism, five of seven SNPs (three of four independent loci) were confirmed in an independent dataset (Supplementary Table S6).

Mendelian Randomization
We performed a MR analysis to investigate the causal effect of body fat distribution, measured as WHR adjBMI , on risk of schizophrenia (for data on the 328 genome wide significant SNPs associated with WHR adjBMI , see Supplementary Table S9). With F = 85.907 the F-statistic indicates strong instrumental variables. After removing palindromic and ambiguous SNPs (n = 9), we included 319 SNPs in the analysis. The overall estimates, calculated by IVW or MR Egger, did not reveal associations between WHR adjBMI and risk of schizophrenia (IVW, b = 0.0062, P = 0.9107; Supplementary Table S10 and Supplementary Figures S4-S6), although in the single SNP analyses, 47 SNPs showed a significant result (Supplementary Table S11). Sensitivity analyses using the leave-one-out approach confirmed the lack of associations (Supplementary Figure S7). There was no evidence for pleiotropy (MR-Egger intercept = −0.0003194; P = 0.8991). The heterogeneity test was significant [Q(df = 317) = 860.0176, P = 1.12 × 10 −51 ]. Power analysis revealed that our MR analyses had an 80% power to detect an odds ratio (OR) of 1.09 and 100% power to detect an OR of 1.28 for schizophrenia (Supplementary Figure S8). The more robust method RAPS also showed no causal effect of body fat distribution on schizophrenia [RAPS main function: beta(hat) = 0.0059, SE = 0.03291, P = 0.858; RAPS overdispersed function: beta(hat) = −0.0744, SE = 0.1115, P = 0.504].
To test reverse causality, we performed an MR analysis where we considered the risk of schizophrenia as a risk factor (for information on the association of 142 SNPs, see Supplementary Table S12) and WHR adjBMI as an outcome. After removing palindromic and ambiguous SNPs (n = 4), we included 138 SNPs in the analysis. F-test revealed that the instruments were strong (F = 233.896). The heterogeneity test was significant [Q(df = 136) = 771.8842, P = 1.60 × 10 −89 ], as was the test for directional horizontal pleiotropy (MR-Egger intercept = 0.00539, P = 0.0077).
The potential relevance of the identified loci seems to be substantiated by the fact that (a) 15 SNPs loci (seven for schizophrenia, six for educational attainment, and two for intelligence) at 12 loci met the genome-wide significance threshold and (b) 11 of 18 loci were confirmed for intelligence and neuroticism in independent data sets. Nevertheless, future GWAS and meta-analyses will need to show that our significant hits with P-values between P < 4.98 × 10 −5 and 5 × 10 −8 indeed represent true findings for the respective mental phenotypes. Thirty-three loci associated with mental phenotypes were identified via BMI/obesity SNPs (Table 1); for 16 of these loci, the same or additional SNPs identified via other anthropometric traits were co-localized. Thus, as expected based on high correlations between the anthropometric traits (Schneider, 2010) the SNPs for BMI unadjusted traits of body composition and body fat distribution (WC, HC, WHR, BF%, and BFM) associated with mental phenotypes were always co-associated with the loci identified via BMI SNPs. Only seven loci were unique for BMI adjusted measures of body fat distribution ( Table 2). Because of the co-occurrence of nine loci identified both via BMI and measures of body fat distribution adjusted for BMI, we looked up the lead SNPs of these seven loci ( Table 2) in the most recent GWAS meta-analyses for BMI (Yengo et al., 2018; Supplementary Table S7); interestingly, all but one locus (locus no. 38: rs1727294) revealed a P-value <0.05, one of which was even genome-wide significant (locus no. 39: rs12317176). Clearly, the overall frequent co-identification of loci for BMI and BMI adjusted measures of body fat distribution indicates an overlap between BMI and BMI adjusted body fat distribution loci with respect to the respective mental phenotypes.
Recently, Kaur et al. (2018) conducted look-up analyses in several somatic and mental disorders using genome wide significant WHR adjBMI SNPs as measurement for body fat distribution/abdominal obesity. In accordance with our results, they found that two WHR adjBMI variants in 3p21.1 (see locus 5 in Table 1) are associated with schizophrenia (PBMI1 and ITIH1) and BD (ITIH1) (Kaur et al., 2018); our results indicate that BMI, HC adjBMI , and WC adjBMI , too, are associated with these mental disorders. Additionally, they reported that LOC105375015 (rs2596500) was associated with WHR adjBMI and schizophrenia, a finding we could not confirm as our analyses did not include proxy SNPs.
There are numerous indications that the clinical association between abdominal obesity and cardiometabolic risk is largely independent of BMI (Després, 2006;van Vliet-Ostaptchouk et al., 2013). For the latest genetic study on body fat distribution (WHR adjBMI ; Pulit et al., 2019), the potential for collider bias resulting from conditioning WHR on BMI was analyzed. They found that most WHR adjBMI SNPs (311 of 346) are associated with stronger standard deviation effect sizes for WHR than for BMI. Each additional (weighted) WHR adjBMIraising allele was associated with a higher WHR and with a lower BMI (Pulit et al., 2019). We looked up the 346 SNPs associated with WHR adjBMI (Pulit et al., 2019) in the recent GWAS for BMI (Yengo et al., 2018): 263 SNPs overlapped between the two data sets (Yengo et al., 2018;Pulit et al., 2019), 20 of which were genome-wide significant for both WHR adjBMI and BMI. In light of the high number of SNPs identified in our study that have an effect on both BMI and WHR adjBMI and the less than 10% overlap (20/263) of BMI and WHR adjBMI (Yengo et al., 2018;Pulit et al., 2019). SNPs we postulate that loci having independent effects on both anthropometric traits have a higher probability of playing a role in mental phenotypes.
For schizophrenia, we identified the strongest overlap with body fat distribution traits adjusted for BMI in our look-up analysis ( Table 2). Most clinical studies (Sugawara et al., 2012;Konarzewska et al., 2014;Strassnig et al., 2017), but not all (Kim et al., 2017) have found a higher BF%, BFM, visceral adipose tissue mass in normal weight patients with schizophrenia as compared to healthy controls. These clinical observational studies either considered antipsychotic medication as a covariable (Strassnig et al., 2017), discuss it as a justification for observed differences (Kim et al., 2017), or mention the inclusion of patients on antipsychotics as a limitation (Sugawara et al., 2012). Besides general limitations of observational studies such as residual confounding, directions of potential causal links between body fat distribution and schizophrenia remain unclear. MR assumes that genetic variants are not confounded by behavioral, socioeconomic, and physiological factors and genetic variants will not be influenced by the onset of disease (Davey Smith and Ebrahim, 2005). Therefore, MR can be regarded as a randomized controlled trial provided by nature. For this reason, we applied MR analysis to the most recent data to test whether there is an overall causal association between BMI adjusted body fat distribution and schizophrenia.
The MR analysis did not reveal evidence for an overarching effect of adjusted body fat distribution on the risk for this mental disorder. The power analysis revealed that our MR had a 100% power to detect an OR of 1.28 for schizophrenia. This OR is similar as those observed for single SNPs calculated in GWAS: e.g., the T-allele of rs3130820 (the SNP with the highest effect in GWAS by Pardinas et al., 2018) increases the risk of getting a diagnosis of schizophrenia by 1.281 times (SE = 0.018, P = 2.12 × 10 −44 ); the OR of the T-allele at rs12416331 is 1.157 (SE = 0.017, P = 7.09 × 10 −18 ). Thus, the power of our MR was sufficient to detect a clinically relevant risk for schizophrenia. In the reverse analyses about the causal effect of schizophrenia on body fat distribution, the non-zero intercept from MR-Egger indicates that there is the directional pleiotropy or the assumptions for instrumental variable are violated, or both. The Egger method showed a significant effect, but the IVW method showed none. The RAPS method, which is more robust to pleiotropy, showed no effect. Our MR analyses suggest that there is no causal effect of schizophrenia on body fat distribution.
Our study now shows that specific genetic effects at individual loci may well play a role in associations between both BMI adjusted body fat distribution and schizophrenia.
In the following, we provide the reader with an insight into potential functional implications of five of the 40 loci which we deem of particular interest 13,14,15,16,17,18: (1) Locus 1p31.1 (rs3101336, rs7550173, NEGR1; locus 1 in Table 1) was significantly associated with both MDD and depressive symptoms, educational attainment, and intelligence. NEGR1 (neuronal growth regulator 1 gene), a well-known obesity-associated gene (Willer, 2009), is associated with white matter integrity (Dennis et al., 2014), and shows high expression in several brain tissues (highest median expression in frontal cortex, Supplementary Figure S1). NEGR1 is a member of the immunoglobulin domain-containing glycosylphosphatidylinositol (GPI)anchored neural cell adhesion molecules (IgLON; Karis et al., 2018) and involved in neural cell communication and synapse formation (Joo et al., 2019). Recently, NEGR1 was also found to play a role in intracellular cholesterol trafficking (Joo et al., 2019). Overall, NEGR1 seems to play a pleiotropic role due to its involvement in synaptogenesis, neuron maturation, and neurite outgrowth, which all represent essential mechanisms for sustaining brain function and development (Pischedda and Piccoli, 2015). Significantly, elevated levels of NEGR1 transcripts were found in the dorsolateral prefrontal cortex of patients with schizophrenia compared to healthy controls (Karis et al., 2018).
(2) CYP17A1 (cytochrome P450 family 17 subfamily a member 1 gene) in 10q24.33 (locus 19 in Table 1) is one of the well-known genes associated with adult hypertension (Levy et al., 2009). We found an association with BMI, WC, HC, WC adjBMI , HC adjBMI , and both schizophrenia and MDD. CYP17A1 has previously been found to be associated with schizophrenia (Ripke, 2014); our detected association is novel for MDD. In the recently published multivariate study on the wellbeing spectrum, CYP17A1 was associated with depressive symptoms and neuroticism (Baselmans et al., 2019). A potentially overarching functional implication of this locus in both somatic and mental disorders is supported (Van Woudenberg et al., 2015) by the fact that in adolescent males but not females, a CYP17A1 SNP was associated with enhanced blood pressure reactivity to mental stress (Van Woudenberg et al., 2015). CYP17A1 is predominantly expressed in adrenal glands (Supplementary Figure S2) and is a bifunctional key regulatory enzyme in the steroidogenic pathway with 17α-hydroxylase and 17,20-lyase activity (Yoshimoto and Auchus, 2015). CYP17A1 is essential for the formation of all endogenous androgens (Peng et al., 2019) and many glucocorticoids (Kmetova Sivonova et al., 2017). Key substrates are pregnenolone and progesterone, which are transformed to the androgen precursor's dehydroepiandrosterone and androstenedione (Kim et al., 2016). Several findings indicate a sexrelated susceptibility for schizophrenia with males at higher risk, e.g., earlier age of onset compared to a lower susceptibility for women due to a potentially neuroprotective role of estrogens (Seeman, 1996) via effects on dopamine (DA) neurotransmission (Di Paolo, 1994;Sanchez et al., 2010;Frau et al., 2014;Godar and Bortolato, 2014). It is hypothesized that androgenic metabolites of testosterone, especially dihydrotestosterone (DHT) and androstenedione, may facilitate the development of schizophrenia-related symptoms. Quantitative real-time PCR in post-mortem brain slices of MDD patients showed a significant decrease in mRNA levels of CYP17A1 (Qi et al., 2018) and genetic variants of the rs743572 polymorphism of CYP17A1 were correlated with the intensity of depressive symptoms in post-menopausal women (Loja-Chango et al., 2016), altogether suggesting a role of CYP17A1 in mood disorders. (3) At the locus 16p.11.2 (see locus 26 in Table 1) higher BMI, BF%, WC, HC, HC adjBMI , and higher risk for obesity were negatively associated with intelligence and educational attainment in line with previous evidence (Jacquemont et al., 2011). Genetic variations [SNPs, copy number variants (CNVs), and larger deletions/insertions] at 16p.11.2 have been shown to be associated with developmental delay, ASD, ADHD, schizophrenia, BD, epilepsy, intracranial volume, brain development, congenital anomalies, altered satiety response, energy imbalance, underweight, and morbid obesity (McCarthy et al., 2009;Bochukova et al., 2010;Fernandez et al., 2010;Walters et al., 2010;Jacquemont et al., 2011;Volckmar et al., 2012;Zufferey et al., 2012;Egger et al., 2014;Qureshi et al., 2014;Volckmar et al., 2015;Maillard et al., 2016;Giuranna et al., 2018;Martin-Brevet et al., 2018;Sonderby et al., 2018;Niarchou et al., 2019). The consequences of 16p11.2 CNV and larger deletions and duplications on health are broad (Crawford et al., 2019;Niarchou et al., 2019). Because the CNVs are rare (Stefansson et al., 2014), large studies and complex analytical methods are needed to detect them in GWAS (Nothen et al., 2019). The SNPs we found in 16p11.2 are assigned to genes IL27, NUPR1, ATXN2L, SH2B1, TUFM, SBK1, SULT1A2, ATP2A1, APOB48R, AC138894.2, ATP2A1-AS1, and RABEP2 (see Supplementary Table S6); however, a total of 41 genes may be affected by the CNV in this region (Stefansson et al., 2014). The protein encoded by IL27 (Interleukin 27) is one of the subunits of a heterodimeric cytokine complex and is in many aspects closely tied to the cytokines IL-12 and IL-23. IL-27 may play a role in the rapid initiation of a response to an inflammatory challenge (Pflanz et al., 2002). Neuro-inflammatory processes as a result of stress-induced immune system activation (Halaris, 2019) are discussed as potential mechanisms in the pathogenesis of mental disorders, e.g., depression (Berk et al., 2013). This hypothesis was recently supported by systems genomics studies (Sharma, 2016) and clinical intervention studies with COX-2 inhibitors and aspirin (Muller, 2019). A negative phenotypical correlation of the level of systemic inflammation biomarkers with general intelligence in children also has been observed . Inflammation and metabolic risk factors might increase the risk of psychosis in some people (Perry et al., 2019), inflammatory mechanisms may also play a role in the pathophysiology of stress-related and anxiety disorders (Lasselin et al., 2016). (4) In 3p12.1 we found an association between BMI (obesity) and educational attainment (rs2325036, CADM2; see locus 6 in Table 1). Cell adhesion molecule 2 (CADM2) belongs to the immunoglobulin superfamily and interacts with other cytoskeletal proteins (Biederer, 2006). It is predominantly expressed in the central and peripheral nervous systems (Pellissier et al., 2007;Morris et al., 2019; Supplementary Figure S3) and regulates, e.g., synapse formation (Rathjen et al., 2017). Obesity risk variants (Yan et al., 2018;Morris et al., 2019) were associated with increased CADM1 and CADM2 expression in the hypothalamus with a complex interplay of CADM1 and CADM2 (Rathjen et al., 2017). Reduced expression promoted a negative energy balance and weight loss in mice (Rathjen et al., 2017). The expression of both CADM1 and CADM2 in several neuronal circuits suggests an intertwining of body weight maintenance with motor control, but also behavioral aspects, cognition, and memory (Rathjen et al., 2017). In addition to associations with obesity risk genes (SNP rs13078807 located in the second intron of CADM2) novel associations of CADM2 genetic variants with systolic blood pressure (SBP), CRP levels , as well as with diverse psychological traits (neuroticism, mood instability, and risk-taking) have been detected. Interestingly, other loci identified by SNPs for BMI and/or body fat distribution in the current study are involved in the immune system, too (Supplementary  (Hinney et al., 2013). Evidence from animal models links the MC4R to eating behavior in stress-induced anxiety-like or depression-like behavior, anhedonia, and corticosterone secretion (Lim et al., 2012;Liu et al., 2013a;Sabban et al., 2015). MC4R appears to interact with the SAPAP3 gene regulating compulsive behavior and body weight (Xu et al., 2013). A higher rate of anxiety and compulsiveness in patients with AN might be mediated by the MC4R (Xu et al., 2013). The MC4R (rs17782313-C) has also been associated with elevated scores for emotional eating and craving for food (Yilmaz et al., 2015). Gain of function variants in MC4R, which have a weight lowering effect (Wang et al., 2010), have been associated with binge eating disorder (BED) in overweight subjects. Apart from polymorphisms exerting weak effects on BMI 166 MC4R mutations mostly entailing a loss in function have been identified up to date mainly in obese individuals (Hinney et al., 2013). Apart from obesity, mutation carriers showed increased lean mass, increased linear growth, hyperphagia, hyperinsulinemia, and ADHD (Loos et al., 2008;Albayrak et al., 2011;Hinney et al., 2013).

LIMITATIONS
Limitations of our study include: (i) we conducted a lookup analysis using summary statistics, so that we could not analyze covariates. (ii) The number of hits was higher for mental phenotypes with higher heritability, higher number of genome-wide significant loci, and/or higher numbers of subjects in the respective GWAS. Therefore, we suspect a type 2 error for phenotypes with a currently low number of identified loci. (iii) We had performed a Bonferroni correction for the 1,005 SNPs. As a drawback true associations may have been missed.
On the other hand, we did not correct for the number of analyzed traits. (iv) We were unable to systematically consider the ancestry/ethnicity of the study groups comprising the diverse GWAS. However, as most data were derived from individuals of European descent this should represent a minor issue only. For those loci, which were detected in non-European study groups, we checked P-values and effect directions in recent GWAS conducted in individuals of European descent. (v) The progress in GWAS research leads to the release of many new analyses every year. For most phenotypes the selected GWAS data sets are the most recent ones; it was, however, not possible to include the most recent analyses, e.g., for WHR adjBMI (Pulit et al., 2019), BMI (Yengo et al., 2018), and schizophrenia (Pardinas et al., 2018) in our look-up analyses. We were, however, able to use these most recent studies for MR-analysis and additional look-ups. (vi) Sex-and age-stratified analyses were not possible as the respective data sets were not available for all traits despite evidence for differences in genetic regulation for BMI and body fat distribution depending on age and sex (van Vliet-Ostaptchouk et al., 2013;Winkler et al., 2015;Pulit et al., 2019;Rask-Andersen et al., 2019). For body shape (WHR adjBMI ), age effects were not found (Winkler et al., 2015). (viii) It needs to be noted that across the literature for body fat distribution the term "adjusted for BMI" is not defined uniformly. It can either mean that the measurement (WC, HC, and WHR) is directly adjusted for BMI or that the estimation of the SNP effects in a GWAS was adjusted for BMI. For future GWAS a clear definition of the term "adjusted for BMI" is needed. (viii) We defined loci mainly via a central SNP and a surrounding region (±500 kb). However, some loci might actually represent more than one independent locus. We attempted to detect independent hits by analysis of linkage disequilibrium; nevertheless, independent SNPs within a given locus may have escaped detection. (ix) A clinically relevant aspect of the relationship between BMI and mental disorders is the susceptibility to weight gain as a side effect of treatment with especially atypical antipsychotic drugs. Because the GWAS data for the anthropometric traits are mostly based on "healthy" individuals, large scale confounding via use of psychotropic drugs a priori appears unlikely. (x) Study samples for some of the examined phenotypes are not completely independent. Sample overlap can potentially affect the results of cross-trait analysis. Since our analyses were conducted using summary statistics it is not possible to determine an exact number of overlapping participants between the underlying GWAS used in our analyses.
To assess whether the overlap can be assumed, we looked at whether GWAS studies included in our analysis used data from the same study groups. The GWAS study on body fat distribution (Pulit et al., 2019), which we used for our MR analysis included a total of 29 study groups of the PGC and of UK Biobank. Four cohorts (B58C-WTCCC, TWINSUK, EGCUT, and KORA) are also mentioned as sources of controls for GWAS for schizophrenia (Pardinas et al., 2018), which we used as outcome phenotype. For cases with schizophrenia CLOZUK1, CLOZUK2, and CardiffCOGS cohorts and the PGC cohorts were used. These study groups were apparently not used for the GWAS on body fat distribution (Pulit et al., 2019).
Two-sample MR assumes independent samples. Violation of this assumption will result in increased Type 1 error and biased effect estimates . For a continuous outcome, bias due to sample overlap is a linear function of the proportion of overlap between the samples with IVW method. Bias due to sample overlap may be more serious for the MR-Egger method. In a simulation analyses for binary outcome with sample overlap in the controls only -as it is the case in our study -there was no detectable bias in the IV estimates even with extremely weak instruments, nor was there any inflation of type 1 error rates . Accordingly, in our MR study, the sample overlap between controls of schizophrenia GWAS and study participants for GWAS on risk variable (i.e., WHR adjusted for BMI) do not lead to a bias for estimated causality in two sample MRs with binary outcome .
Sample overlap can also influence the results of the look-up analysis. As an example we compared studies that were included in the GWAS on BMI (158 studies in total; Locke et al., 2015) and studies that were included in the GWAS on educational attainment (63 studies in total; Okbay, 2016b). This comparison revealed that nineteen studies were the same.
We also checked the overlapping of the study groups included in the GWAS on BMI (GIANT; Locke et al., 2015) and in the GWAS on schizophrenia (PGC; Ripke, 2014). It is possible that a certain number of controls for the GWAS on schizophrenia were the same individuals included in GWAS on BMI (overlapping cohorts were EGCUT, MIGen, WTCCC, KORA, HRN, MGS, and deCODE).
The sample overlap between the GWAS on psychiatric disorders due to comorbidity should also be considered. This topic had previously been investigated using data from PGC (LeBlanc et al., 2018). They found that the overlap was less than 2% for eight psychiatric disorders. Data sets in the study were the same as in our study for four disorders (ADHD, AN, MDD, and schizophrenia). Nevertheless, we can probably assume that the sample overlap is of a similar magnitude for studies included in our analyses.

CONCLUSION
We identified several loci that were associated with BMI, body composition and/or body fat distribution, and mental disorders/traits. The SNPs for body composition and body fat distribution (WC, HC, WHR, BF%, and BFM) associated with mental disorders/traits were all simultaneously associated with BMI. Hence, an isolated effect of body fat distribution on mental phenotypes implied by clinical and epidemiological studies (Luppino et al., 2010;Berkowitz and Fabricatore, 2011;Gurpegui et al., 2012;Luppino et al., 2014;Mühlig et al., 2015;Cortese et al., 2016;Coryell et al., 2016;Nigg et al., 2016), was not substantiated. In the MR analysis, we did not detect a general causal effect of body fat distribution on schizophrenia. Our study has confirmed some known cross-phenotype associations (Brainstorm et al., 2018) but has also given some new clues for genetic loci involved in both anthropometric traits and mental phenotypes.

DATA AVAILABILITY STATEMENT
The datasets for this article are publicly available (see section "Materials and Methods").

AUTHOR CONTRIBUTIONS
All authors contributed to the writing of the manuscript and approved of the final version. TP planned and supervised the project, performed the analyses, discussed the results and wrote the manuscript; LN performed the analyses and drafted the first version of the manuscript; B-HL advised and reviewed the MR analysis; AH interpreted the results and was part of the writing team; JA discussed the results and was part of the writing team; LT contributed to analyses; RN performed the MR analyses; CA provided critical input on early versions of the manuscript; LL discussed the results and commented on the manuscript; JH planned and supervised the project, discussed the results, and wrote/finalized the manuscript.