A new study in the journal Nature Medicine analyzes longitudinal and cross-sectional changes in blood analytes associated with variations in body mass index (BMI).
Study: Multiomic signatures of body mass index identify heterogeneous health phenotypes and responses to a lifestyle intervention. Image Credit: jivacore / Shutterstock.com
The health impacts of obesity
The prevalence of obesity has been increasing over the past four decades among adolescents, adults, and children throughout the world. Several studies have reported obesity to be a major risk factor for multiple chronic diseases such as metabolic syndrome (MetS), type 2 diabetes mellitus (T2DM), cardiovascular disease (CVD), and certain types of cancer.
Even 5% weight loss among obese individuals can improve metabolic and cardiovascular health, as well as reduce the risk for obesity-related chronic diseases. However, the physiological manifestations of obesity have been reported to vary considerably across individuals.
How is obesity measured?
Quantification of obesity takes place using the anthropometric BMI, which is body weight divided by body height squared. BMI is typically used for the primary diagnosis of obesity, as well as to assess the effectiveness of lifestyle interventions.
However, there are certain limitations to using BMI as a measurement of health. For example, BMI can cause misclassification of people with a high muscle-to-fat ratio as those with obesity and misjudge metabolic improvements in health post-exercise.
Omics studies have indicated that blood omic profiles can provide information on several human health conditions. A machine learning model that was trained to predict BMI through 49 BMI-associated blood metabolites was reported to provide better obesity-related clinical measurements as compared to genetic predisposition for high BMI or observed BMI.
Another blood metabolomics-based model of BMI also reported differences among individuals with or without acute coronary syndrome. This suggests that multi-omic blood profiling can help bridge the gap between BMI and heterogeneous physiological states.
About the study
The current study involved the recruitment of people who participated in a wellness program by a commercial company between 2015 and 2019. Individuals were included in the current study if they were over 18 years of age, residents of any U.S. state except New York, and not pregnant.
Participants were included if their datasets contained all main omic measurements, genetic information, and a BMI measurement within 1.5 months from the first blood draw. The external cohort was obtained from participants who participated in the TwinsUK Registry and underwent two or more clinical visits for biological sampling between 1992 and 2022. Only participants whose datasets contained all measurements for metabolomics, obesity-related standard clinical measures and BMI were included in the current study.
Peripheral blood, saliva, and stool samples were collected from participants for analysis of genetic ancestry, measurement of blood omics, and generation of gut microbiome data. Information on height, weight, blood pressure, waist circumference, and daily physical activity was also collected.
The analysis of blood metabolomics, BMI, gut microbiome data, and BMI of baseline visits took place for the TwinsUK participants. Machine learning models were trained to predict baseline BMI for each of the omics platforms including proteomics, metabolomics, and clinical lab, or in combination with clinical labs (chemistries)-based BMI (ChemBMI), proteomics-based BMI (ProtBMI), metabolomics-based BMI (MetBMI), and combined omics-based BMI (CombiBMI) models. Another ten fitted sparse models were generated using the least absolute shrinkage and selection operator (LASSO) algorithm for each omics category.
This was followed by the health classification of each participant based on the World Health Organization (WHO) international standards for BMI cutoffs. Gut microbiome models were also generated for the classification of obesity. Assessments of longitudinal changes took place in the measured and omics-inferred BMIs. Finally, an analysis of the plasma analyte correlation network was performed.
Study findings
A total of 1,277 adults participated in the study, most of whom were White, female, and middle-aged. The BMI prevalence at baseline was similar among the normal, overweight, and obese classes.
The models retained 30 proteins, 62 metabolites, 20 clinical laboratory tests, as well as 132 analytes. The CombiBMI model was found to be the best in BMI prediction.
Investigation of the TwinsUK cohort indicated that blood metabolomics better captures BMI as compared to standard clinical measures. Notably, omics-inferred BMI maintained the characteristics of classical BMI.
Proteins were the strongest predictors in the CombiBMI model. More specifically, fatty acid-binding protein 4 (FABP4), adrenomedullin (ADM), and leptin (LEP) were positive regulators, while advanced glycosylation end-product-specific receptor (AGER) and insulin-like growth factor-binding protein 1 (IGFBP1) were negative regulators.
The misclassification rate of omics-inferred BMI was about 30% across all BMI classes and omics categories. The mismatched groups of the normal BMI class showed higher values of the markers positively associated with BMI and lower values of the markers negatively associated with BMI, while the opposite was observed for the mismatched group of the obese BMI class. The omics-based BMI model also captured obesity characteristics, including abdominal obesity.
The MetBMI class reflected bacterial diversity better than the standard BMI class and had stronger associations with gut microbiome features. Lifestyle interventions decreased the overall BMI estimate of the entire cohort, where a decrease of MetBMI was the highest and ProtBMI was the least.
A total of 100 analyte–analyte correlation pairs were significantly modified by the baseline MetBMI. Among them, 27 analyte-analyte correlation pairs were significantly modified by days in the program and were mainly derived from metabolites.
One such time-varying pair was homoarginine and phenyllactate (PLA). A positive association between homoarginine and PLA was observed in the obese MetBMI class at baseline, which became weaker during the intervention.
Conclusions
The current study demonstrates the importance of blood multi-omic profiling for preventive and predictive medicine. Furthermore, these findings demonstrate that multi-omic characterization of obesity can be useful for the characterization of metabolic health, as well as identifying targets for health management.
Limitations
The study has certain limitations. First, the analytes retained in the omics-based models might not have causal relationships with obesity phenotypes. Second, the study did not measure all biomolecules in blood.
An additional limitation is that the current study was unable to evaluate the effectiveness of the lifestyle intervention. The study findings are also not generalizable.
Journal reference:
- Watanabe, K., Wilmanski, T., Diener, C., et al. (2023). Multiomic signatures of body mass index identify heterogeneous health phenotypes and responses to a lifestyle intervention. Nature Medicine. doi:10.1038/s41591-023-02248-0.
Early genomic profiling improves cancer treatment outcomes