Differential effects on neurodevelopment of FTO variants in obesity and bipolar disorder suggested by in silico prediction of functional impact: An analysis in Mexican population

Abstract Introduction Several studies indicate that polygenic obesity is linked to fat‐mass and obesity‐associated (FTO) genetic variants. Nevertheless, the link between variants in FTO and mental disorders has been barely explored. The present work aims to determine whether FTO genetic variants are associated with bipolar disorder and obesity, and to perform an in silico prediction of variant‐dependent functional impact on the developing brain transcriptome. Methods Four hundred and forty‐six Mexican mestizos were included in a genetic association analysis. SNP‐sequence kernel association test and linear mixed models were implemented for genetic association assessment. For functional impact prediction, we analyzed the mapping of regulatory elements, the modification of binding sites of transcription factors and the expression of transcription factors in the brain developing transcriptome, searching on different databases. Results In the set‐based analysis, we found different associated regions to BD (bipolar disorder) and obesity. The promoter flanking region of FTO intron 1 was associated with differential effects on BMI, while intron 2 of RPGRIP1L and FTO upstream regions were associated with BD. The prediction analysis showed that FTO BD‐associated variants disturb binding sites of SP1 and SP2; obesity‐associated variants, on the other hand, disturb binding sites of FOXP1, which are transcription factors highly expressed during prenatal development stages of the brain. Conclusion Our results suggest a possible effect of FTO variants on neurodevelopment in obesity and bipolar disorder, which gives new insights into the molecular mechanism underlying this association.


| INTRODUC TI ON
One of the strongest associations to obesity was established in 2007, when in a genome-wide association study (GWAS) for type 2 diabetes, common variants of the FTO gene showed an association with this trait, but subsequent analyses revealed that this association was modulated by body mass index (BMI) (Frayling et al., 2007;Scuteri et al., 2007). The relation between FTO common variants and obesity has been replicated in diverse populations worldwide, including Mexican population (Andreasen et al., 2008;Dina et al., 2007;Do et al., 2008;Grant et al., 2008;Saldana-Alvarez et al., 2016;Villalobos-Comparan et al., 2008).
Obesity-related illnesses, such as cardiovascular diseases (CVD), have been established as leading causes of death in individuals with bipolar disorder (BD) (Goldstein et al., 2011;Osby, Brandt, Correia, Ekbom, & Sparen, 2001;Weiner, Warren, & Fiedorowicz, 2011). It is suggested that some drugs used for treating individuals with BD cause weight gain, which has been settled as one possible explanation of the frequent comorbidity between CVD risk, obesity, and BD (Keck et al., 2007;Shah, Shen, & El-Mallakh, 2006). A recent study in Mexican individuals diagnosed with BD suggests that common FTO variants could modulate BMI in these patients and there could be a possible interaction between mood-stabilizers (Diaz-Anzaldua et al., 2015). Nevertheless, a high prevalence of overweight has been reported in some treatment-naïve patients, which suggests that there are other factors involved in this relationship (Maina, Salvi, Vitalucci, D'Ambrosio, & Bogetto, 2008). In these sense, some studies have reported that BD patients have a diminished ability to modulate self-mood by over-eating (Elmslie, Mann, Silverstone, Williams, & Romans, 2001;Fleet-Michaliszyn et al., 2008;Wildes, Marcus, & Fagiolini, 2006). Supporting the evidence that a FTO genetic variation could have an effect on BMI modulation by mood-state regulation in patients with BD, an association with FTO variants has also been found in patients with major depressive disorder and eating-disorders, (Castellini et al., 2017;Hung et al., 2014;Muller et al., 2012). Therefore, we decided to evaluate the association of FTO polymorphisms with bipolar disorder and obesity using aggregate variation; additionally, we also performed an in silico prediction analysis of associated variants to search for their possible functional impact on developing brain transcriptome.

| Study population
We included a total of 446 Mexican mestizo individuals. The sample was divided into two groups based on their psychiatric diagnosis.

| Ethical statement
In an interview, the participants were informed about the aims of the study and signed a written-informed consent. The protocol was ap-
All missing genotypes were imputed, spanning a genomic region on chromosome 16 (including the last introns of RPGRIP1L and the first introns of FTO) that started on 53700000 bp and ended on 53900000 bp from the p telomere (based on the hg19 coordinates) using Beagle software and the 1000 Genomes Phase 3 as reference panel (Auton et al., 2015;Browning & Browning, 2016). In order to establish the phasing of the input data, Beagle performed a phasing of the introduced unphased genotypes prior to the imputation process; it used the phase reference panel just as previously described by Browning & Browning (2007).
For the statistical analysis, we only considered variants with a MAF > 1% and a p value > 0.0001 for a χ2 test for Hardy-Weinberger equilibrium. After imputation, we obtained a total of 1,422 variants; nevertheless, due to the previously established criteria, only 340 variants remained and were included in the following analysis.

| Genetic association tests
For the genetic association of genetic variants with BD and BMI phenotype, we performed two different approaches: (a) a set-based association test, with variants grouped by the mapping of regulatory and coding sequences to the gene; and (b) a single-variant analysis, implementing a linear mixed model. For the set-based association test, the SNP sequence kernel association test (SKAT) was implemented in R (RRID:SCR_001905); also, linear mixed models were implemented on the genome-wide complex trait analysis (GCTA) software to analyze single-variants (Yang, Lee, Goddard, & Visscher, 2011;Wu et al., 2011). The BD phenotype was coded as a categorical variable, BMI was considered a continuous variable expressed in kg/m 2. , obesity criteria followed previous reports (Hubry & Hu, 2015). A SKAT and mixed model p < 0.05 was considered statistically significant after adjustment for age, gender and genetic relationship matrix (GRM), as well as after FDR correction for multiple testing (i.e. throughout the text p-values are reported after FDR multiple testing correction). The GRM was calculated based on the algorithm implemented using the efficient mixed-model association expedited software (EMMAX) and all imputed variants (Zhou & Stephens, 2012).
Imputed variants that showed a positive association were validated through Sanger sequencing in 20% of the samples.

| Prediction of functional effects
The functional impact of variants was predicted using Variant Effect  TA B L E 1 Descriptive characteristics of the population levels at differential development stages of the human brain (from prenatal stages to adulthood) measured in different cerebral areas.
Using BrainSpan we explored transcription factors that showed sites disrupted by associated variants which are highly expressed in the prenatal brain and scarcely expressed in postnatal brain.

| RE SULTS
On average, individuals of the BD group were younger than those of the BDC group, but this difference did not reach statistical significance. The overview descriptive characteristics of the sample are shown in Table 1. Regarding the mean BMI, there were no significant differences between the two groups; however, after BMI stratification, the BD group showed a higher prevalence of overweight than the BDC; while the BDC group showed a higher frequency of obesity (chi-square test, p < 0.05).

| Differential association of variants in FTO between obesity and bipolar disorder
Set-based association analysis with BD as the phenotype trait revealed that two regions reached a statistically significant association, intron 2 of RPGRIP1L and the FTO upstream region. A summary of the set-based association analysis is reported in Table 2. When we considered BMI as a continuous variable for the whole sample (BD and BDC groups), variants in intron 1 promoter flanking of FTO reached statistical significance (p < 0.05) and after diagnosis stratification this association remained only in the BD group (p < 0.05), even when the BDC group had a higher rate of obesity.
The analysis of single-variants revealed that seven variants reached a statistically significant association with BD (p < 0.05); these BD-associated variants were mapped to intron 2 of RPGRIP1L, FTO upstream region and intron 1 promoter flanking of FTO. When we considered obesity (BMI > 30 kg/m 2 ) as phenotype, 14 variants in the promoter flanking region immersed in FTO intron 1 reached a statistically significant association (p < 0.05). Rs7205859, in the promoter flanking region immersed in intron 1 of FTO, was the only variant associated to BD and obesity (p < 0.05).

| Prediction of functional impact ("in silico") of variants in the developing brain
In order to correlate associated variants with expression-regulatory elements that could alter brain development, we performed a query of the associated variants to different databases. After mapping the associated variants with expression-regulatory elements in DNA, we found that BD-associated variants are in high linkage disequilibrium with an enhancer located in intron 1 of RPGRIP1L; while obesity-associated variants are in high linkage disequilibrium with an enhancer located in intron 1 of FTO, suggesting a possible differential role of FTO variants in BD and obesity. The inclusion of genomic information at different levels is summarized in Figure 1. We also mapped variants that could modulate binding sites of transcription factors with a query to SNP2TFBS database. BD-associated vari-

| D ISCUSS I ON
FTO is one of the genes known as a post-GWAS gene, being mainly associated with obesity. Obesity is a complex process in which different systems and organs participate, including the brain. In recent studies, it has been hypothesized that genetic variants affecting brain development could have a high impact on obesity, establishing a possible link between molecular mechanisms modulating obesity and brain disorders (Hara et al., 2001;McCammon, Blaker-Lee, Chen, & Sive, 2017;Yeo & Heisler, 2012). In this study, we analyzed a genetic association of FTO variants with BD and obesity using a variant aggregation strategy, and we also performed an in silico prediction analysis of the possible functional impact of these variants in the developing brain.
There is a known hypothesis that explains the high obesity rates in patients with BD, which relates to the use of drugs related to weight gain in treatment schemes of these patients (Keck et al., 2007;Shah et al., 2006). Nevertheless, another hypothesis supported by the fact that naïve treatment patients have higher rates of obesity than the general population, established that the link between BD and obesity could also be explained by a disrupted brain biological process in individuals with BD (McElroy & Keck, 2012).
In this line of evidence, we found that FTO variants associated with BMI only remained in BD patients, even when the BDC group had F I G U R E 1 Summary of the in silico predicted functional analysis reduced longitudinal functioning in the medial prefrontal cortex and this reduction correlated with higher rates of impulsivity, behavioral disinhibition, and risky decision-making, which are common symptoms in BD (Chuang et al., 2015). These FTO-associated changes in brain-behavior relationship lead us to hypothesize that the link between BD symptomatology and eating behavior (particularly the way in which eating is controlled) could be a process modulated by FTO genetic variants. However, we did not search for eating-related endophenotypes that could lead us to more precise FTO-associated behavioral changes in BD patients.

Mixed Linear Model Association Results
The molecular mechanism underling the association of FTO variants with obesity remains largely unrevealed; some in vivo and in vitro studies applying different molecular strategies have returned inconclusive results. The FTO protein is an Alk-B like DNA/ RNA demethylase, FTO knock-out mice survive until adulthood stages and exhibit no differential weight changes when compared to wild-type mice, suggesting that alterations in protein structure do not indicate a risk of obesity (Han et al., 2010;Ronkainen et al., 2015;Stratigopoulos et al., 2008). One of the most accepted hypotheses to explain the molecular mechanism takes into consideration that the expression regulation of FTO and other nearby genes (RPGRIP1L and IRX3) is a process controlled by an obesity-associated region in intron 1 of FTO, which we also replicated in this study (Ronkainen et al., 2015;Smemo et al., 2014;Stratigopoulos et al., 2014). Nevertheless, divergent results in animal models have been reported. Some results indicate a possible dysregulation of IRX3 whereas others point to RPGRIP1L. The lack of consistency among reports may be due to the use of dissimilar stages of tissue development when analyzing gene expression. The brain is a complex organ which has differential spatial-temporal gene expression patterns depending on the developmental stage, this differential pattern could be a key regulatory factor during the wiring development process of this organ (Chedotal & Richards, 2010;Chou et al., 2016;French, Tan, & Pavlidis, 2011). In the present analysis, we found a different variant association pattern when we considered BD or obesity as a phenotype trait, which suggests differential modulating effects of these variants in these diseases. In an attempt to yield new insights of a possible mechanism underlying the association of FTOobesity risk variants with BD, we performed an in silico prediction algorithm consulting different databases. The in silico prediction analysis showed that FTO-risk variants could have a direct effect on modulating binding sites of transcription factors such as SP1, SP2 and FOXP1 during brain development, which suggests a possible effect of these transcription factors on a FTO-risk variant dependent on brain development mechanism. Supporting these data, in animal models it has been described that alterations in expression levels of these transcription factors (SP1, SP2 and FOXP1) have a direct effect reducing the morphogenesis of the brain (Li et al., 2015;Valin, Cook, Ross, Saklad, & Gill, 2009 Variant previously associated to obesity (Hassanein et al., 2010;Li et al., 2013).
TA B L E 3 Report of variants that disrupted a transcription factor binding site (Precious et al., 2016). A possible link between striatum and obesity has been found in a neuroimaging study using positron emission tomography, and reported that the striatum was the most altered in diet-induced obesity, which reinforces a possible link between the brain and obesity (Michaelides et al., 2018). Notwithstanding, more studies should be performed to achieve a conclusive explanation of the relationship between these transcription factors and FTO-risk variants during the dynamics of brain development.
One of the limitations of the present work is that patients analyzed were not drug-naïve which could be a confounding factor in this population; nevertheless, this work reports an association of FTO variants with bipolar disorder. To the best of our knowledge, this is the first study reporting differential FTO-risk for obesity and bipolar disorder. Also, we performed a prediction analysis of the regulatory function of these FTO-risk variants, to gain new insight into the molecular functioning of this region in the developing brain. However, it is necessary to perform more functional in vivo and in vitro analyses to identify the mechanism underling these associations.

| CON CLUS ION
The present work found an association of FTO variants with bipolar disorder. Also, we performed a prediction analysis of the regulatory function of these FTO-risk variants, to give new insights into the molecular functioning of this region in the developing brain.

ACK N OWLED G M ENTS
We thank all the participants of this study.

CO N FLI C T O F I NTE R E S T S
The authors declared no conflict of interest.