Increased BMI and late-life mobility dysfunction; overlap of genetic effects in brain regions

How obesity earlier in life impacts upon mobility dysfunctions in late life is not well understood. Pernicious effects of excess weight on the musculoskeletal system and mobility dysfunctions are well-recognized. However, increasingly more data support the link of obesity to overall motor defects that are regulated in the brain. To assess the causal relationship between body mass index (BMI) at midlife and performance of the Timed Up-and-Go test (TUG) in late life among a population-based longitudinal cohort of Chinese adults living in Singapore. We evaluated genetic predispositions for BMI in 8342 participants who were followed up from measurement of BMI at average 53 years, to TUG test (as a functional mobility measure) 20 years later. A robust 75.83% of genetically determined BMI effects on late-life TUG scores were mediated through midlife BMI (Pindirect-effect = 9.24 × 10−21). Utilizing Mendelian randomization, we demonstrated a causal effect between BMI and functional mobility in late life (βIVW = 0.180, PIVW = 0.001). Secondary gene enrichment evaluations highlighted down-regulation of genes at BMI risk loci that were correlated with poorer functional mobility in the substantia nigra and amygdala regions as compared to all other tissues. These genes also exhibit differential expression patterns during human brain development. We report a causal effect of obesity on mobility dysfunction. Our findings highlight potential neuronal dysfunctions in regulating predispositions on the causal pathway from obesity to mobility dysfunction.


INTRODUCTION
Global prevalence of obesity [body mass index (BMI) of ≥30 kg/m 2 in the general population and ≥27.5 kg/m 2 in Asians] has tripled over the past four decades, imposing enormous burden on public health [1]. Obesity is associated with a range of chronic diseases [2]. Mechanical issues resulting from increased weight also impacts an individual's quality of life [3]. Mobility functions are critical to many activities that are necessary for independent living and higher BMI is contemporaneously associated with poorer functional mobility [4,5].
How obesity results in mobility dysfunction has not been precisely elucidated. Increased inflammation and insulin resistance in individuals with obesity may be associated with poorer physical function and muscle strength [6,7]. Also, persistent excess weight loaded onto the musculoskeletal system may lead to biomechanical impairments [8,9]. Nevertheless, there is limited knowledge linking altered biomechanics due to obesity and musculoskeletal injury. Additionally, whether such biomechanical changes are a cause or a result of related conditions, such as osteoarthritis, are unclear [8,9].
Besides the musculoskeletal burdens due to increased weight, recent data suggest on a relationship between obesity and inferior motor coordination [10,11]. Importantly, defects in motor coordination have been reported from very early developmental stages in life, potentially before the onset of obesity-related biomechanical dysfunctions. As the control of motor and movement functions are regulated in the brain, it would be important to consider the role of obesity in the central nervous system (CNS) control of mobility functions.
Intriguingly, much of the genetic predisposition to obesity is also believed to regulate gene expression patterns in brain regions, including those that are key in controlling movement functions [12][13][14]. These findings suggest that mobility dysfunctions among individuals with obesity may not simply be a result of excess weight that impacts on the musculoskeletal system but there may be an underlying overlap between neuronal circuits that govern both obesity predispositions and motor dysfunctions.
We employed a Mendelian randomization (MR) framework to evaluate for a causal relationship between BMI and performance of the Timed Up-and-Go test (TUG) in late life among a longitudinal cohort of Chinese adults. TUG is a reliable tool to evaluate mobility in older adults [15] and performance on the TUG is affected by agility, static/dynamic balance, and executive function [16,17]. Subsequently, we evaluated gene sets, regional genes at BMI risk loci that were correlated with poorer mobility function, in established brain-related databases to understand gene expression enrichments and dysfunctional pathways that may correlate with both increased obesity and subsequent mobility dysfunctions.

Study population
The Singapore Chinese Health Study (SCHS) is a prospective, populationbased cohort designed to evaluate determinants of chronic diseases in Chinese adults living in Singapore [18]. In brief, 63,257 participants aged 45-74 (mean 53) years old were recruited from 1993 to 1998. Study participants were either of the two major dialect groups of Chinese in Singapore (Hokkien or Cantonese) and resided in government-built housing estates. Among consenting survivors, follow-up interviews were conducted every 5-6 years.
Participants were re-contacted during the first follow-up interviews (1999)(2000)(2001)(2002)(2003)(2004), and approximately half consented to donate blood or buccal samples. The third follow-up interviews at late life were conducted from 2014 to 2017 and involved in-person interviews along with anthropometric measurements and physical tests. In total, 17107 surviving participants aged 61-96 years participated in the third follow-up. All studies were approved by the Institutional Review Board at National University of Singapore, and written informed consent was obtained from all participants.

BMI and other covariates at midlife
During recruitment interviews, participants self-reported their height and weight that was used to compute BMI (kg/m 2 ) at midlife. During baseline interviews, information on sociodemographic characteristics were obtained through a structured questionnaire. Characteristics recorded included level of education (no formal education, primary, secondary and above), history of smoking (never, former, current smoker), frequency of alcohol consumption (never or monthly, weekly, daily), amount of physical activity (0 h, 1-3 h, 4 or more hours per week), and history of physiciandiagnosed comorbidities: heart attack, stroke, diabetes, and cancer.

Mobility function at late life
During the third follow-up (2014-2017), we used the TUG test to assess mobility function. From an initial sitting position on a chair, participants were instructed to stand, then walk at a usual pace towards a marker on the floor 3 m away, then subsequently turn around and walk back to the chair, and finally to sit back down. A stopwatch was used to record the time the participants took to complete this test to the nearest second. Participants wore their usual footwear and could use any walking aids that they normally required. Two trials were performed and the faster time was used.

Genetically determined midlife BMI
Description for the genome-wide association study genotyping, quality control (QC) procedures and imputation have been published previously [19,20]. In brief, 27,308 samples were genotyped using Illumina Global Screening Array and 2136 samples were genotyped using Illumina HumanOmniZhonghua Bead Chip [21]. Additional autosomal SNPs were imputed with IMPUTEv2 using the cosmopolitan 1000 Genomes reference panels (Phase 3). Samples with call-rate <95.0% (N = 212), extremes in heterozygosity (N = 278), one sample from each cryptic related pair (N = 2995) and outliers from reported ethnicity detected through principal component analysis (N = 356) were excluded. In total, 25603 participants had valid GWAS genotyping data. Among them, 8342 completed the third follow-up interview and TUG measurement and were included in the current study.
SNPs to generate the weighted genetic risk score (wGRS) for BMI were selected from a previous MR study [22]. Among the 565 reported BMI SNPs, 559 reached genome-wide level of significance (P < 5 × 10 −8 ). The final wGRS consisted of 362 SNPs that were polymorphic and common [minor allele frequency > 0.01] in our East-Asian population and passed QC in SCHS (Supplementary Table 1). To compute the wGRS, each SNP was coded based on the number of BMI increasing risk alleles and weighted by the individual effect estimate from the original GWAS studies [12,22,23].

Statistical analyses
Normally distributed variables were presented as mean ± standard deviation (SD), while categorical variables were presented as proportions of study population. Midlife BMI and TUG at late life were presented as median [Interquartile range (IQR)], and these were normalized using rankbased inverse normal transformation. Multivariable linear regression was used to evaluate associations between wGRS, midlife BMI, and TUG at late life. We used the "sem" command in Stata to construct structural equation models for mediation analyses. Post-estimation command "estat teffects" provided the decomposition of total, direct, and indirect effects [24]. In models with wGRS as the exposure of interest, we adjusted for age, sex, and the first three principal components. In models where midlife BMI was the exposure or mediator, we additionally adjusted for time from BMI measurement to TUG, dialect group, levels of education, smoking status, alcohol consumption and physical activity, and history of heart attack, stroke, diabetes, or cancer. Analyses were conducted using Stata/SE14.2. Two-sided P < 0.05 was considered statistically significant.
We included MR-Egger [28] and simple and weighted median [29] methods as sensitivity analysis. MR-Egger intercept tests were performed to detect horizontal pleiotropy [28] and an intercept of zero (P intercept > 0.05) was considered as absence of pleiotropic bias. Heterogeneity within MR-Egger was assessed by Rucker's Q [25]. Weak instrument bias was assessed by calculating the variation between individual genetic variant estimates for BMI (I 2 GX ) [30].

Gene enrichment
BMI SNPs were stratified into 2 groups for tissue enrichment analyses. Group 1 included SNPs that were associated with increased BMI and slower TUG times (N = 190) and group 2 included SNPs (N = 153) that were associated with increased BMI and faster TUG times (Supplementary Table 1). Regional genes at these loci (within 50 kb by default) [22] were mapped using the SNP2GENE function in FUMA (v1.3.7) [31]. Subsequently, the two gene lists were employed for the GENE2FUNC in FUMA to evaluate tissue enrichment, which is tested using pre-defined differentially expressed genes (DEG) across 54 tissue types from GTEx (v8) [32]. Tissue DEGs were calculated by performing two-sided t-test for any one tissue against all others. DEGs were defined as genes with P adj < 0.05 after Bonferroni correction and absolute log fold change >0.58. Up-and down-regulated DEGs were determined by evaluating the sign of t-statistics. Significant DEGs from the study were extracted for further evaluations, as above, with gene expression data from 11 general developmental stages of brain samples using Brainspan [33]. Hypergeometric tests were performed to evaluate overrepresentation of gene sets in GO biological process. Overrepresented biological processes with a P adj < 0.05 were considered significant.  (Fig. 1a, Supplementary Table 2). Midlife BMI was positively associated with TUG at late life in a dose-dependent manner (P trend = 1.99 × 10 −58 , Fig. 1b, Supplementary Table 3); when compared to counterparts in the normal weight group, participants who were underweight had faster TUG at late life (β = −0.081, P = 0.042) while participants with obesity had significantly slower TUG (β = 0.538, P = 4.65 × 10 −54 ).

Demographic characteristics of prospective SCHS participants
The BMI wGRS was positively associated with TUG time at late life in a stepwise manner (P trend = 3.38 × 10 −4 ); when compared to counterparts with the lowest wGRS quartile, participants in the third and highest quartiles of wGRS had significantly slower TUG times at late life (β = 0.063, P = 0.018 for Q3; β = 0.084, P = 0.002 for Q4) (Fig. 1c, Supplementary Table 4).
Increased BMI was causally associated with slower TUG 75.83% of wGRS's total effect on TUG was mediated through midlife BMI (β indirect-effect = 0.002, P indirect-effect = 9.24 × 10 −21 ) (Supplementary Table 5, Fig. 2). In contrast, the direct effect of wGRS on TUG was not statistically significant (β direct-effect = 0.001, P direct-effect = 0.489), indicating that the observed effects of the BMI wGRS on TUG through horizontal pleiotropy paths other than midlife BMI, were insignificant.
MR was performed to interrogate the casual relationship between BMI and mobility dysfunction at late life. 19 SNPs were detected as outliers in Radial MR (P Q < 0.05) and excluded (Supplementary Table 6, Supplementary Fig. 1). The causal estimate obtained from the IVW method indicated that each 1-unit increase in normalized midlife BMI resulted in a slower normalized TUG at late life (β IVW = 0.180, P IVW = 0.001, Table 2).
Absence of uncorrelated horizontal pleiotropy was indicated by an insignificant MR-Egger's intercept (P intercept = 0.448, Table 2). The MR-Egger method obtained a marginally significant causal effect, with a larger effect size compared to IVW (β egger = 0.280, P egger = 0.050, Table 2). Rucker's Q (P Het = 0.999) was insignificant and a high I 2 GX value of 95.60% was observed. The median-based methods returned similar results to the IVW, although the P value for the simple median method was just above the significance threshold (P simple-median = 0.051, P weighted-median = 0.038, Table 2).

Enrichment of BMI-related genes in brain regions and development stages
BMI increasing alleles for 190 SNPs were associated with slower TUG times and were selected as group 1. The remaining 153 SNPs were associated with faster TUG times and were included in group 2 (Supplementary Table 1). The number of regional genes mapped to groups 1 and 2 genetic loci were 1188 and 998, respectively. BMI-related genes from group 1 were observed to be significantly down-regulated in lymphocytes (P adj = 0.003), substantia nigra (P adj = 0.022) and amygdala (P adj = 0.024) brain regions [32] (Fig. 3a, Supplementary Table 7). Additional enrichments with different genomic proximity cut-offs and utilizing significant eQTL genes (FDR < 0.05, GTEx v8) linked to these BMI risk variants indicated similar strong enrichments at the amyglada and substantia nigra ( Supplementary Fig. 2). No significant enrichments were detected using genes from group 2 (Fig. 3b).
518 genes from group 1 were identified to be significantly down-regulated in the substantia nigra, and these genes were significantly over-represented in a series of GO biological processes (573 processes, P adj 6.38 × 10 −11 −0.050), including overall growth, organ development, metabolic processes, several neuronal and CNS related processes (Supplementary Table 8). We evaluated if these 518 BMI-related regional genes may be differentially enriched during human brain development stages. Significant up-regulation of 61 genes (P adj = 0.005) was observed in late prenatal brain tissue, while down-regulations of gene subsets were enriched in brain tissues from childhood to adulthood, with especially strong associations in middle adulthood (P adj = 7.82 × 10 −4 ) ( Supplementary Fig. 3a, Supplementary  Table 9). Pathway enrichment indicated that regional BMI-related genes, up-regulated in brain tissues during the prenatal developmental stage, were primarily involved in growth, organ development and metabolic processes (Supplementary Table 10). In contrast, BMI-related genes down-regulated in brain tissue after birth indicated on enrichments in numerous neuron related processes (Supplementary Tables 11-13), and the CNS development GO process was observed as the top pathway from genes down-regulated in the middle adulthood brain tissues (P adj = 1.36 × 10 −4 , Supplementary Table 11). Continuous variables were presented as mean ± SD (normally distributed) or median (IQR), and categorical variables were presented as N (%). BMI body mass index, SD standard deviation, TUG timed up-and-go, wGRS weight Genetic Risk Score, IQR interquartile range.
A similar pattern was observed for the BMI-related genes downregulated in the amygdala (Supplementary Fig. 3b, Supplementary  Tables 14, 15). Genes significantly up-regulated in late prenatal brain tissues (P adj = 3.88 × 10 −4 , Supplementary Table 15), showed involvement in growth and metabolism related processes (Supplementary Table 16). In contrast, genes down-regulated in late childhood and middle adulthood brain tissues were strongly CNS related (Supplementary Tables 17, 18).

DISCUSSION
We demonstrated a key role for obesity in development of late life mobility dysfunction among a community-dwelling population of Chinese adults. In a MR framework, we extended these findings and report a causal effect between increased BMI and poorer performance on the TUG test. The TUG has been suggested as an indicator of subtle motor deficits that may be used as a prodromal marker for Parkinson's disease (PD) development [34]. Taken  together, our results indicate early interventions to reduce excess weight may be critical to preserving mobility function in old age.
Consistent results with similar effect sizes were achieved by the majority of the MR methods employed, except for MR-Egger. However, it is typical for the SE of the causal estimate obtained from MR-Egger to be larger than that obtained from the fixedeffect IVW since MR-Egger requires genetic variants to have a wide range of associations with the risk factor for a precise estimate [35]. Reassuringly, the insignificant result from the MR-Egger intercept indicated minimal pleiotropic effects, and the large I 2 GX indicated minimal weak instrument bias in the analysis.
Control of human body weight is regulated in a complex manner in the CNS [36]. Severe gene dysfunctions at the hypothalamic leptin-melanocortin pathway results in monogenic Rucker's Q. Significant associations in bold (p < 0.05). Fig. 3 Top tissue enrichments observed with gene expression data in tissues from GTEx database (both side). a Top 10 tissues enriched using regional genes (within 50 kb) at body mass index (BMI) loci correlated with slower timed up-and-go (TUG). Red threshold line indicates P Adj = 0.05. b Top 10 tissues enriched using regional genes (within 50 kb) at BMI loci not correlated with slower TUG. None of the tissue enrichments were statistically significant were significant (P Adj > 0.05). Full results of all 54 GTEx tissue enrichments (both-sided, downregulated and up-regulated) available in Supplementary Table S7. forms of obesity and affected individuals display constant hyperphagia [13]. Whether similar pathways are affected in common forms of obesity is controversial, and conversely, studies indicate that cognitive and affective processes of the CNS may be implicated in the dysregulation of eating behavior in humans [37]. Besides the hypothalamus, the insula and the substantia nigra brain regions, which are involved with processing addiction and reward behaviors, have been highlighted as potential neuronal regions enriched among top genetic findings from BMI GWAS studies [14].
Our downstream gene enrichment analyses through the use of established human brain databases, suggested the possibility of brain regions that co-regulate both obesity and mobility function. Genes at obesity risk loci that were correlated with slower TUG, indicated on potential enrichments in lymphocytes and brain regions that control mobility, the substantia nigra and amygdala. The enrichment observed in lymphocytes may be due to a known effect of obesity that impairs immune function, altering leukocyte counts as well as cell-mediated immune responses [38]. Substantia nigra is a midbrain dopaminergic nucleus that modulates motor movement and reward functions as part of the basal ganglia circuitry [39]. Increased dopamine release in the substantia nigra due to ingestion of highly palatable food may result in a positive hedonic state linked to overfeeding and obesity [39,40]. Pathological changes to the amygdala are among the earliest in movement disorders and have been associated with depression among PD patients [41][42][43]. At the same time, the amygdala activates food cues and may control appetite in relation to emotions [44].
Additionally, our analyses suggest on varied enrichments of some of the BMI-related genes during human brain development. Pathways observed to be up-regulated in prenatal brain, indicated on processes regulating overall growth, perhaps contributing to obesity effects early in life. In contrast, down-regulated pathways, especially in adulthood brain tissues, indicated on overwhelming neuronal dysfunctions that may contribute to CNS control of motor functions. Several genes enriched in the CNS biological process, such as NFIB [45], KDM1A [46], XRN2 [47], and ABL1 [48], that was down-regulated in adulthood brain tissues have known regulatory roles in neurodegenerative diseases. Certain genes such as SEMA3F and NRP1 perform critical axonal guidance functions that are essential for appropriate synaptic connections between different brain regions [49] and may display temporal regulations during CNS development [50].
A major strength of our cohort study was the detailed longitudinal follow-up of participants over a mean period of 21 years. This made our study less prone to issues due to reverse causation, allowing a reliable causal relationship between BMI and subsequent mobility dysfunction to be identified. However, some limitations have to be acknowledged. Firstly, in the MR, the significance level of the simple median estimate was just above the significance threshold (P = 0.051). As the robustness of different loci on BMI susceptibility may vary, the simple median estimator may be unsuitable in such instances. This variability can be accounted for, more effectively, using the weighted median method that indicated on a significant association [29]. Secondly, genetic variants for BMI evaluated were originally reported in GWAS where participants were mainly of European ancestry; in contrast, the participants in our study were of East-Asian descent. However, these BMI risk loci demonstrate good transferability to East-Asian populations [36,51,52]. Similar analysis in other ethnicity groups would be warranted to robustly investigate the link between increased BMI and mobility. Additionally, BMI values were derived through self-reported height and weight and while, robust associations were observed between the BMI wGRS and BMI values, there may be inaccuracies in effect estimates risk between BMI at midlife and subsequent TUG at late life. Lastly, a caveat of the study was that precise functional genes at each of these BMI risk loci have not been systematically elucidated. The analyses may have missed long range interactions and the inclusion of non-causal proximal genes in these analyses may have reduced power in the study to identify all significant tissue and pathway enrichments.

CONCLUSION
Our study highlights that higher BMI was causally associated with poorer mobility function in late life and suggests that expression patterns of BMI related genes in specific brain regions and during developmental stages may regulate obesity and subsequent movement function.

DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding authors upon reasonable request.