BMI-adjusted adipose tissue volumes exhibit depot-specific and divergent associations with cardiometabolic diseases

Saaket Agrawal, Marcus D. R. Klarqvist, Nathaniel Diamant, Takara L. Stanley, Patrick T. Ellinor, Nehal N. Mehta, Anthony Philippakis, Kenney Ng, Melina Claussnitzer, Steven Grinspoon, Puneet Batra, Amit V. KheraNature Communications 2023 · DOI

TL;DR

This study leverages deep learning on MRI data from over 40,000 individuals to precisely quantify three key fat depots (visceral, abdominal subcutaneous, and gluteofemoral). Crucially, they developed BMI-adjusted metrics to isolate the impact of fat distribution independent of overall body weight, revealing that visceral fat is linked to increased cardiometabolic risk, while gluteofemoral fat appears protective.

What Problem Does This Paper Address?

Obesity, often defined by Body Mass Index (BMI), is a major public health concern. However, individuals with the same BMI can have vastly different health outcomes. This heterogeneity is partly attributed to variations in how and where fat is stored in the body – known as fat distribution. While visceral adipose tissue (VAT) has long been associated with increased cardiometabolic risk (like type 2 diabetes and heart disease), previous studies have been limited by small sample sizes, cross-sectional designs, and the difficulty of precisely measuring different fat depots, especially when accounting for overall BMI. This paper aims to overcome these limitations by using advanced machine learning to accurately quantify fat depots in a large cohort and then disentangle the specific health impacts of each depot, independent of BMI.

Methods

The core methodological innovation lies in the application of deep learning, specifically convolutional neural networks (CNNs), to estimate fat depot volumes from body MRI scans. To handle the massive dataset (over 18 terabytes), they ingeniously reduced the input data by transforming 3D MRI scans into 2D coronal and sagittal projections. This allowed their CNNs, trained on a subset of participants with pre-existing fat depot measurements, to achieve near-perfect prediction accuracy (R² > 0.978) on unseen data.

To address the challenge of isolating the effects of fat distribution from overall BMI, they derived "BMI-adjusted" metrics for each fat depot (VATadjBMI, ASATadjBMI, GFATadjBMI). These metrics represent the residual fat depot volume after accounting for an individual's BMI, effectively capturing BMI-independent variation in local adiposity. These adjusted metrics were then used in logistic and Cox proportional-hazard regression models to assess their associations with prevalent and incident cardiometabolic diseases, respectively.

Datasets

The primary dataset used is the UK Biobank, a large-scale observational study. The analysis included MRI data from 40,032 participants. A subset of these participants (around 9,000 for each depot) had pre-existing, manually quantified fat depot volumes (VAT, ASAT, and GFAT) which were used to train and validate the deep learning models. The remaining participants' fat depot volumes were then estimated using these trained models.

Key Insights and Ideas

  • Deep Learning for Precise Fat Quantification at Scale: The study demonstrates the power of deep learning to accurately and efficiently quantify complex biological structures like fat depots from medical imaging, opening doors for large-scale phenotyping.
  • Decoupling Fat Distribution from BMI is Crucial: The concept of BMI-adjusted fat depots is a key conceptual advance. It allows researchers to move beyond simply correlating obesity with disease and instead understand the specific metabolic roles of different fat storage sites.
  • Visceral Fat: The "Bad" Fat: The findings strongly reinforce the detrimental role of visceral fat (VAT) in cardiometabolic health, linking higher BMI-independent VAT to increased risk of type 2 diabetes and coronary artery disease.
  • Gluteofemoral Fat: The "Good" Fat: Conversely, higher BMI-independent gluteofemoral fat (GFAT) appears to be protective, associated with a reduced risk of these diseases. This supports the idea of GFAT as an "adaptive energy storage depot."
  • Abdominal Subcutaneous Fat: Largely Neutral: ASAT, while a significant fat depot, showed a largely neutral association with cardiometabolic risk in this BMI-adjusted analysis.
  • Metabolically Distinct Fat Depots: The study provides strong evidence for three distinct metabolic roles of these fat depots, moving beyond a one-size-fits-all view of adipose tissue.

Key Results

  • High Prediction Accuracy: The CNN models achieved R² values of 0.991 for VAT and ASAT, and 0.978 for GFAT in the held-out dataset, indicating near-perfect estimation.
  • VATadjBMI: Associated with increased risk of type 2 diabetes (OR/SD 1.49) and coronary artery disease (OR/SD 1.17).
  • ASATadjBMI: Showed a largely neutral association (OR/SD 1.08 for T2D, 1.00 for CAD).
  • GFATadjBMI: Associated with decreased risk of type 2 diabetes (OR/SD 0.75) and coronary artery disease (OR/SD 0.89).
  • Incident Disease: Over a median follow-up of 2.8 years, VATadjBMI was associated with increased risk of incident type 2 diabetes (HR/SD 1.45) and coronary artery disease (HR/SD 1.17), while GFATadjBMI was associated with reduced risk of incident type 2 diabetes (HR/SD 0.84).
  • Lifestyle Associations: Ideal diet and physical activity were associated with reduced VATadjBMI, and to a lesser extent ASATadjBMI and GFATadjBMI.

Limitations and Future Work

  • Cross-Sectional Data: The primary analysis is cross-sectional, limiting definitive causal inferences. Longitudinal studies are needed to confirm these associations and explore therapeutic interventions.
  • Underpowered for Subgroups: While the models performed similarly across self-reported ethnicities, the study was underpowered to detect disease associations in non-White subgroups, which is a critical area for future research given known ethnic variations in fat distribution and cardiometabolic risk.
  • Simplification of MRI Data: While effective for volume estimation, the 2D projection approach might not be suitable for quantifying "density-like quantities" (e.g., liver fat).
  • Future Directions: The authors suggest that BMI-adjusted fat depot measures could serve as valuable endpoints for evaluating the efficacy of obesity interventions. They also highlight the need to explore the genetic underpinnings of these BMI-independent fat distribution patterns and the role of sex differences. Further work could also explore using these deep learning models on other imaging modalities like DEXA or CT scans, or even static images, to broaden accessibility.

Summary updated

Abstract

For any given body mass index (BMI), individuals vary substantially in fat distribution, and this variation may have important implications for cardiometabolic risk. Here, we study disease associations with BMI-independent variation in visceral (VAT), abdominal subcutaneous (ASAT), and gluteofemoral (GFAT) fat depots in 40,032 individuals of the UK Biobank with body MRI. We apply deep learning models based on two-dimensional body MRI projections to enable near-perfect estimation of fat depot volumes (R<sup>2</sup> in heldout dataset = 0.978-0.991 for VAT, ASAT, and GFAT). Next, we derive BMI-adjusted metrics for each fat depot (e.g. VAT adjusted for BMI, VATadjBMI) to quantify local adiposity burden. VATadjBMI is associated with increased risk of type 2 diabetes and coronary artery disease, ASATadjBMI is largely neutral, and GFATadjBMI is associated with reduced risk. These results - describing three metabolically distinct fat depots at scale - clarify the cardiometabolic impact of BMI-independent differences in body fat distribution.
BibTeX
@inproceedings{agrawal2023bmi,
  author = {Saaket Agrawal and Marcus D. R. Klarqvist and Nathaniel Diamant and Takara L. Stanley and Patrick T. Ellinor and Nehal N. Mehta and Anthony Philippakis and Kenney Ng and Melina Claussnitzer and Steven Grinspoon and Puneet Batra and Amit V. Khera},
  title = {{BMI-adjusted adipose tissue volumes exhibit depot-specific and divergent associations with cardiometabolic diseases}},
  booktitle = {Nature Communications},
  year = {2023},
  doi = {10.1038/s41467-022-35704-5},
}
Figure