1 BODY MASS INDEX AND ITS SIGNIFICANCE FOR HEALTH

Body mass index (BMI) is the traditional numerical measurement, calculated by (bodyweight/height²), that enables the detection of over-weight and diagnosis of obesity. Obesity is a complex trait caused by a combination of genetic, environmental and lifestyle factors.¹ The prevalence of obesity varies significantly among different populations.² High prevalence of obesity is seen in Latin Americans,³ Middle East populations⁴ and Pacific Islanders.⁵ In the United States, Hispanic and African Americans have higher rate of obesity than European Americans.⁶ Understanding these disparities is crucial in developing targeted interventions and strategies to address the specific needs of these populations.

Obesity causes adipose tissue dysfunction and leads to chronic low-grade inflammation, which serves as a key contributor to the development of chronic diseases associated with obesity.⁷ Excessive bodyweight also places mechanical stress on heart, contributes to respiratory difficulties and paroxysmal nocturnal dyspnoea (PND), and impacts joint health. As a result, obesity contributes to the risk of a number of serious diseases, including cardiovascular diseases (e.g., hypertension, heart failure and coronary artery disease),⁸ chronic metabolic diseases (e.g., metabolic syndrome, type 2 diabetes and non-alcoholic fatty liver disease [NAFLD]) and respiratory diseases (e.g., asthma and chronic obstructive pulmonary disease [COPD]).^{9, 10} Moreover, obesity is a known risk factor for several types of cancer.¹¹ The chronic inflammation and observed hormonal imbalance associated with obesity can promote the development of certain cancers, including breast, colon, ovarian and pancreatic cancers.¹² Beyond its physical implications, obesity also has a substantial impact on mental health.¹³ The psychological consequences of obesity, such as body image dissatisfaction, low self-esteem and social stigmatization, further contributes to the negative burden experienced by individuals with obesity.¹⁴

2 BMI AND ITS RISK FACTORS

The awareness of BMI and its implications in reference to health extends beyond a simple numerical measurement. Genetic variations have been found to play a significant role in the susceptibility to obesity by influencing various physiological processes such as metabolism, fat storage and distribution, and appetite regulation. The fat mass and obesity-associated gene (FTO) is the first gene identified of association with the genetic susceptibility of obesity.¹⁵ Tews et al. showed that FTO-deficient adipocytes upregulate uncoupling protein 1 (UCP1) expression, resulting in enhanced mitochondrial uncoupling and energy expenditure.¹⁶ Haupt et al. observed significant impact of FTO variation on whole-body fat distribution.¹⁷ Both common¹⁸ and rare^{19, 20} variants of the melanocortin 4 receptor gene (MC4R) have been associated with obesity. MC4R is expressed mainly in hypothalamus, and is the central melanocortin receptor regulating appetite and energy expenditure.^{21, 22}

Genetic factors alone do not account for the entire obesity epidemic. Long-term environmental factors and lifestyle habits also play a significant role. Dietary choices (high sugar, high fat, low fibre) and lack of physical activity have been strongly associated with the development of obesity²³ (Figure 1). It is important to recognize that the prevention of obesity through healthy dietary habits and regular exercise is easier and more effective than treating it once it has been diagnosed. By focusing on prevention and early intervention, individuals can reduce the risk of obesity-related health complications. Furthermore, obesity itself can create a vicious cycle that compounds its presence.²⁴ The mechanical stress and joint injuries resulting from excess weight can limit physical exercise of an obese patient, further exacerbating the problem. This cycle emphasizes the importance of early intervention and prevention efforts to break the cycle and promote better health outcomes. For these reasons, the prediction of BMI has important implications in the prevention/early intervention of obesity and overall health promotion.

3 POLYGENIC RISK SCORE (PRS) FOR BMI

Polygenic risk scores (PRS), which combines information from multiple genetic variants across the genome, have emerged as a promising tool for assessing the genetic contribution to complex traits such as BMI. Successful PRS models have been developed for a number of common diseases, for example, breast cancer,²⁹ type 2 diabetes³⁰ and coronary artery disease.³¹ For type 1 diabetes (T1D), the area under the ROC curve (AUC) for the prediction of the disease is up to 0.900.^{32, 33} Common genetic variation accounts for over 20% of variation in BMI.³⁴ A large number of BMI-associated common variants have been identified by genome-wide association studies, for example, the GWAS meta-analysis of BMI in European population by the GIANT consortium.³⁵ The BMI-associated genetic markers identified in the previous GWASs enable the development of BMI-PRS. In contrast, including rare variants associated with BMI has no obvious improvement of PRS models built using common variants.³⁶

Importantly, a previous report highlights the association of BMI-derived PRS across the lifespan.⁴¹ Currently, the performance of PRS for BMI prediction is relatively modest. Our recent PRS model for extreme BMI, that is, the top 1% and 5% has a moderate AUC of 0.703–0.736 with an acceptable predictive power in two large European populations, the Nurses’ Health Study (NHS) and Nurses’ Health Study II (NHSII).⁴² As further research and advancements continue, the BMI-PRS model is expected to evolve and enhance its predictive power, ultimately contributing to more effective obesity prevention and treatment strategies.

4 POTENTIAL HEALTH DISPARITIES

The PRS for BMI was facilitated by the comprehensive identification of SNPs associated with BMI by the previous large-scale GWAS by the GIANT consortium.³⁴ The GWAS meta-analysis by the GIANT consortium, including 339 224 individuals of European ancestry,³⁴ was the basis of our PRS model.⁴² The updated GWAS by the GIANT consortium and the UK Biobank included approximately 700 000 individuals of European ancestry.³⁵ The GWAS by Heilbron et al. included 2 433 011 European.⁴³ In contrast, GWASs on populations other than European are heavily under-represented (Table 1). The large sample GWAS by Huang et al. included 509 429 Europeans with 16 962 individuals of other ethnicities, including African American or Afro-Caribbean, East Asian, Hispanic or Latin American, and South Asian.⁴⁴ Notably, the exome study by Akbari et al. included 549 780 European and 95 846 Hispanic or Latin American.⁴⁵ However, as an exome study cannot capture variants in non-coding regions, its contribution to PRS is modest.

With the high prevalence of obesity in under-represented diversity populations, the discrepancy of lack of genomic data in populations other than European raises potential inequalities in the application of PRS models developed solely based on European data. A PRS model created by European data may not be applied to other populations. The marker SNPs in a European PRS may have different linkage disequilibrium (LD) in other populations. Two variants in LD are inherited together more often than expected by chance, while the LD may vary by populations related to founder effects, population bottlenecks, natural selection, genetic drift and admixture.⁴⁶ Due to LD, two causal SNPs in BMI do not have completely independent effects.

Furthermore, environmental factors and lifestyle habits may have varied interactions with genetic factors across populations (Figure 1). Corella et al. showed that a high intake of saturated fatty acids strengthens the association between FTO variants and BMI.²⁵ Conversely, physical activity attenuates association between FTO variants and BMI.²⁶ Children carrying the protective FTO genotype may be more protected by a favourable social environment.²⁷ Given the significant difference of environmental factors and lifestyle habits across human populations, a European PRS may not perform adequately in other human populations. Consequently, a PRS for BMI only available for European may lead to healthcare disparities, resulting in less effective prevention and intervention of obesity in under-represented populations. To address this issue and promote equitable healthcare, there is a pressing need for the development of a trans-ethnic (TE) PRS for BMI and obesity, requiring proactive action from the research community to ensure diverse representation and the collection of genomic data from different populations. A concerted effort to acquire and incorporate genomic information from under-represented populations in the development of PRS models is essential to overcome potential health disparities.

5 GLOBAL GENOMIC DATA SHARING

The global sharing of genomic data is a crucial aspect of developing and testing a TE PRS, which requires genomic data of diverse populations. However, there are numerous challenges in facilitating the international sharing of genomic data, encompassing both ethical and legal considerations. Consent agreements from participants in genomic studies may not explicitly encompass global data sharing. Consent agreements are typically obtained within the context of a specific research project and may not explicitly cover the broad sharing of genomic data with researchers from different countries and institutions. To address this issue, efforts are needed to enhance consent processes and develop standardized frameworks that encompass global data sharing while upholding ethical principles and respecting participant autonomy. Moreover, legal and regulatory frameworks differ across countries. Certain countries have regulations that prohibit the cross-border transfer of genomic data. These regulations aim to safeguard privacy and protect sensitive information. However, they can hinder international collaborations and impede the progress of genomic research. Overcoming these barriers necessitates international collaborations aimed at addressing these challenges and facilitating responsible and secure data sharing.

To tackle these challenges and promote global genomic data sharing, various initiatives and organizations have emerged. The Global Alliance for Genomics and Health (GA4GH)⁴⁷ and the Global Genomic Medicine Collaborative (G2MC)⁴⁸ are leading efforts to develop frameworks and standards for responsible data sharing across international borders. These initiatives aim to establish guidelines, tools and best practices that enable secure and ethical sharing of genomic and clinical data. By fostering collaboration and building consensus among researchers, policymakers and stakeholders, these organizations facilitate the sharing of genomic data while addressing ethical, legal and technical considerations. The International HundredK+ Cohorts Consortium (IHCC, https://ihccglobal.org/) was established in 2019 as a collaborative effort involving GA4GH and G2MC. The IHCC aims to accelerate genomic research by leveraging large-scale genomic and phenotypic data from diverse population cohorts worldwide, comprising 103 cohorts in 43 countries involving nearly 50 million participants.⁴⁹

For our IHCC study on a TE BMI-PRS, the Center for Applied Genomics (CAG) at the Children's Hospital of Philadelphia (CHOP) with one of the world's largest paediatric biobank⁵⁰ created the PRS model. In 57 613 randomly selected CAG samples from different populations, AUC = 0.735 was acquired with the fraction 0.1 of causal markers to predict top 1% BMI. Without the need for data transfer and time-consuming data sharing agreements, the detailed protocols and all required files to generate the PRS were shared with the participating cohorts. The participating IHCC cohorts in this study included the Nurses’ Health Study I (over 17 000 participants of European ancestry), the Nurses’ Health Study II (over 12 000 participants of European ancestry), the Shanghai Women's Health Study (SWHS, ∼75 000 Chinese women),⁵¹ the Shanghai Men's Health Study (SMHS, 61 480 Chinese men),⁵² the Norwegian Mother, Father and Child Cohort Study (MoBA, 50 290 Northern European adult males and females analyzed in this study), the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil, 9333 civil servants in Brazil analyzed in this study)^{53, 54} and the INTERVAL BioResource (38 319 Europeans included in this study).⁵⁵ As the lead group, we shared the protocol of the PRS models, as well as the SNPs and weights, with the IHCC collaborators for assessment by the international sites. Harmonization of SNP alleles in the PRS model was by comparing with the reference alleles of the TOPMed imputation.

6 TE PRS FOR BMI

While population-specific PRS models tend to have optimal performance, a TE PRS has the advantage of broader applicability, particularly in populations with limited genomic data. Ideally, a TE PRS needs to incorporate disease-associated genetic variants identified from diverse populations.⁵⁶ In the case of the TE PRS for BMI, with limited data available from TE GWAS, we made efforts on developing a TE PRS based on the GWAS results in Europeans, while accounting for population-specific linkage disequilibrium (LD) patterns using a Bayesian approach as implemented in LDpred.⁵⁷ Compared to the PRS approaches of LD pruning followed by p-value thresholding (P+T), LDpred demonstrated its improved performance by taking into account of LD information between genetic markers,⁵⁷ thereby retaining genetic information that might be lost through LD pruning. For example, two genetic markers in LD each may have causal effect. While different populations have different genetic risk of obesity, the underlying assumption of our approach is that causal SNPs of BMI in proximity tends to be in LD in different populations.

Besides natural selection, other evolutionary forces, such as genetic drift, migration patterns, admixture events, recombination rates and assortative mating, can also contribute to population-specific LD patterns. Genetic drift can be a dominant force shaping LD patterns in small isolated populations.⁶⁹ Conversely, migration, or gene flow, facilitates the movement of genetic material between populations, thereby reducing population-specific LD patterns.⁷⁰ Admixture events can introduce new alleles into a population or dilute existing LD patterns.⁷¹ Recombination breaks down LD, while lower recombination rates can maintain LD over longer stretches of DNA.⁷² Assortative mating, which involves individuals preferring mates with similar characteristics, leads to increased homozygosity and the preservation of LD.⁷³

Due to significant differences of environmental factors and evolutionary forces across populations, LD patterns of BMI-causal SNPs may vary. Shown in our study, our TE PRS models were based on the meta-LD matrices, including four major human populations, that is, African American (AA), Hispanic American (HAMR), East Asian (ASN) and Northern European (EUR). Our data showed that the TE PRS outperformed the ancestry-specific models in the Chinese and Brazilian populations, whereas there is no appreciable loss in predictive power in European ancestry individuals when using a TE score. However, as shown in our study, the TE PRS for BMI has a poor performance in the admixed Brazilian population, although it outperforms the European PRS.⁴² Increased genetic diversity and unique genetic drift and selection pressures contribute different LD patterns. In particular, genetic admixture generates LD blocks inherited from each ancestral population. Over time, as recombination events occur in subsequent generations, LD tends to decrease. The meta-LD matrices of non-admixed ancestral populations may not be able to account for the LDs properly in the admixed Brazilian population. It is worth mentioning that in addition to LD, population-specific mutations may also contribute to the limited portability of a TE PRS.⁷⁴ This highlights the importance of increasing diversity in genetic studies by including under-represented populations.

7 PERSPECTIVE

The challenges and opportunities require extensive collaborations and innovative mechanisms for data sharing among international research consortia, such as the IHCC, to pool resources and expertise. Beyond genetic factors, taking into account other risk factors of BMI may enable more accurate prediction of BMI. For example, Welten et al.⁸³ took into account parental BMIs, birthweight, sex and a number of socioeconomic and environmental factors, and acquired AUC = 0.845 without PRS to predict sex/age-specific overweight in the Netherland population. Family history is a valuable and cost-effective tool that should not be neglected in disease risk prediction. It is based on observable patterns of diseases among relatives, and relies on the accuracy and completeness of the information provided by individuals. However, there may be cases where family history is not evident. In such situations, genetic testing may be necessary to complement the information obtained from family history. Genomic informed risk assessment (GIRA)⁸⁴ including major environmental and lifestyle factors in addition to PRS may have improved accuracy in the prediction of BMI.

AUTHOR CONTRIBUTIONS

Conceptualization, H.H. and H-Q.Q.; writing—original draft preparation, H-Q.Q.; writing—review and editing, H-Q.Q., J.J.C., and H.H.; supervision, H.H. All authors have read and agreed to the submitted version of the manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

FUNDING INFORMATION

International HundredK+ Cohorts Consortium (IHCC); Institutional Development Funds from Children's Hospital of Philadelphia to the Center for Applied Genomics; Children's Hospital of Philadelphia Endowed Chair in Genomic Research to HH.

ETHICS APPROVAL

Not applicable.