Identifying molecular pathways of type 2 diabetes using proteomics, metabolic, and anthropometric profiles in UK and Chinese adults

This study integrates proteomic, metabolic, and anthropometric data from UK and Chinese biobanks to identify five distinct protein clusters—including two novel groups related to adiposity and kidney function—and pinpoints specific causal proteins and pathways that elucidate the heterogeneous molecular architecture of type 2 diabetes, offering new targets for precision medicine.

The Gist

The Big Picture: Diabetes is a "Chameleon"

Imagine Type 2 Diabetes (T2D) not as a single, uniform disease, but as a chameleon. It looks different on different people. For some, it's caused by being overweight; for others, it's about how their liver works; for some, it's their kidneys.

Currently, doctors often treat all chameleons with the same green paint (standard medication). But this paper asks: What if we could see the different colors underneath and paint each one specifically?

To do this, the researchers looked at proteins. Think of proteins as the tiny workers inside your body's factory. They build things, fix leaks, and send messages. The researchers wanted to know: Which specific workers are causing the factory to break down in diabetes?

The Detective Work: Two Cities, One Mystery

The researchers acted like detectives investigating a crime in two different cities:

1. London (UK Biobank): They looked at data from about 33,000 people of European descent.
2. China (China Kadoorie Biobank): They looked at data from about 2,000 people of East Asian descent.

They wanted to see if the "culprits" (the bad proteins) were the same in both cities or if the crime was committed differently depending on the neighborhood.

Step 1: Sorting the Messy Pile (Clustering)

The team started with a massive pile of 1,793 different proteins that seemed linked to diabetes. That's like having a giant box of 1,800 different Lego bricks and trying to figure out which ones build the same house.

They used a smart computer algorithm (called bNMF) to sort these bricks into 5 distinct piles (clusters) based on what else was happening in the body:

• The "Fat" Pile: Proteins linked to body weight and belly fat.
• The "Thin" Pile: Proteins linked to low body fat (a new discovery!).
• The "Oil" Pile: Proteins linked to cholesterol and fats in the blood.
• The "Liver" Pile: Proteins linked to liver function.
• The "Kidney" Pile: Proteins linked to kidney function.

The Big Surprise: They found two new piles that nobody had really noticed before: the "Thin" pile and the "Kidney" pile. This suggests that some people get diabetes not because they are fat, but because of issues with their kidneys or a specific type of fat metabolism that happens even in thinner people.

Step 2: The "DNA Fingerprint" Test (Genetics)

Just because a protein is high in a diabetic person doesn't mean it caused the diabetes. Maybe the diabetes caused the protein to go high (like a smoke alarm going off because of a fire, not causing the fire).

To solve this, they used Mendelian Randomization. Think of this as checking the DNA blueprint.

• They looked at the genes people were born with.
• If a person was born with a gene that naturally makes a specific protein high, and that person also gets diabetes, then that protein is likely the architect of the disease, not just a bystander.

The Verdict: Who is the Culprit?

After cross-referencing the "London" and "China" clues, they identified 20 key proteins that are likely driving the disease.

The "Superstars" (Found in both populations):
Three proteins were the main suspects in both London and China:

1. RTBDN: Linked to body fat.
2. TSPAN8: Linked to the liver and sex hormones.
3. NCR3LG1: Linked to the immune system and kidneys.

The "Local Specialists":

• ENTR1: This protein was a major suspect only in the Chinese population. It seems to be linked to how the pancreas (the insulin factory) works. This explains why diabetes might look slightly different in East Asians compared to Europeans.

Why Does This Matter? (The "Precision Medicine" Dream)

Right now, treating diabetes is like giving everyone the same key to open a door. Sometimes it works; sometimes it doesn't.

This research suggests we should make custom keys:

• If your diabetes is driven by the "Kidney Pile" proteins, you might need a drug that targets the kidneys.
• If it's driven by the "Thin Pile" proteins, you might need a completely different approach than someone who is obese.

The Takeaway

This study is like finding the instruction manual for a broken machine. Instead of guessing which part to fix, the researchers have identified the specific gears (proteins) that are grinding the machine down.

By understanding that diabetes has different "flavors" (clusters) and different "culprits" (proteins) in different people, we can move away from a "one-size-fits-all" treatment. In the future, a doctor might look at your blood, see which "pile" your proteins belong to, and prescribe a treatment that hits the exact target, making diabetes management more effective and personalized.

Technical Summary

1. Problem Statement

Type 2 diabetes (T2D) is a heterogeneous disease with diverse underlying molecular mechanisms, yet it is currently treated as a single entity. Existing therapies often yield variable responses because they do not address this heterogeneity. While previous studies have identified T2D subtypes using clinical or genetic data, there is a lack of integration with proteomic data to elucidate the molecular architecture of T2D. Furthermore, there is a need to understand how circulating proteins interact with metabolic and anthropometric traits to drive T2D risk across different ancestries (European vs. East Asian) to support precision medicine and drug discovery.

2. Methodology

The study employed a multi-omics, cross-ancestry design integrating observational and genetic approaches in two large cohorts:

• Cohorts:
  • UK Biobank (UKB-EUR): 33,301 European ancestry participants with plasma proteomics (Olink Explore 3072 panel, ~2,923 proteins) and extensive metabolic/anthropometric phenotyping.
  • China Kadoorie Biobank (CKB-EAS): 2,029 East Asian ancestry participants with similar proteomic data.
• Observational Analysis:
  • Multivariable-adjusted regression (linear/logistic/Cox) identified proteins associated with T2D prevalence (pT2D), incidence (iT2D), and glycemic traits (HbA1c, glucose).
  • Clustering: Bayesian Non-negative Matrix Factorization (bNMF) was applied to the 1,793 T2D-associated proteins (from UKB) and their associations with 43 metabolic/anthropometric traits. This "soft clustering" approach grouped proteins based on shared phenotypic profiles to reveal disease heterogeneity.
• Genetic Analysis (Triangulation):
  • Colocalization: Used HyPrColoc (Bayesian method) to determine if proteins and traits share causal genetic variants in specific genomic regions. A tiered system (CT1/CT2) was used to prioritize robust colocalization.
  • Mendelian Randomization (MR): Two-sample bidirectional MR was performed using cis-pQTLs (protein quantitative trait loci) as instruments. This assessed causal directions: Protein $\to$ Trait, Trait $\to$ Protein, and Protein $\to$ T2D.
  • Tier System: A rigorous tiering system (MRT1/MRT2) incorporating multiple MR methods (IVW, MR-Egger, MR-PRESSO) and Bonferroni correction was used to filter false positives.
• Integration: Observational and genetic evidence were triangulated across both ancestries to identify consistent causal pathways.

3. Key Contributions

• Novel Clustering Methodology: First application of bNMF to cluster T2D-associated proteins based on their phenotypic associations with metabolic/anthropometric traits, revealing distinct biological pathways.
• Discovery of Novel Clusters: Identified five protein clusters, two of which are novel:
  1. Adiposity
  2. Reduced Adiposity (Novel: proteins associated with lower adiposity but higher T2D risk, potentially reflecting lipodystrophy features).
  3. Lipids
  4. Liver
  5. Kidney (Novel: highlighting renal pathways independent of standard glycemic traits).
• Cross-Ancestry Validation: Systematically compared findings between European and East Asian populations, identifying both shared and ancestry-specific mechanisms.
• Causal Inference Framework: Developed a robust framework combining colocalization and bidirectional MR to distinguish between causal drivers, consequences, and bidirectional relationships.

4. Key Results

• Protein Discovery:
  • UKB-EUR: 1,793 proteins were associated with T2D/glycemic traits.
  • CKB-EAS: 484 proteins were associated with T2D traits (limited by sample size and lack of metabolic data for clustering).
• Causal Proteins (Triangulated Evidence):
  • Shared Cross-Ancestry Effects: Three proteins were identified as causally affecting T2D in both populations:
    • RTBDN: Effects mediated through adiposity (BMI).
    • NCR3LG1: Effects mediated through adiposity (potentially via leptin).
    • TSPAN8: Effects mediated through SHBG (Sex Hormone Binding Globulin).
  • Ancestry-Specific Effects:
    • ENTR1: Identified specifically in CKB-EAS, potentially linked to beta-cell dysfunction and lipodystrophy features common in East Asians.
  • UKB-EUR Specific: Five additional proteins (B4GAT1, DNER, ENO3, HMOX2, OMG) showed causal effects on T2D.
• Bidirectional and Consequence Relationships:
  • Bidirectional: Five proteins (GSTA1, GSTA3, MEGF9, NCAN, SHBG) showed bidirectional associations with T2D.
  • Consequences (T2D $\to$ Protein): Six proteins (CD34, FGFBP3, GALNT10, KHK, MENT, MXRA8) appeared to be consequences of T2D rather than causes.
• Mechanistic Pathways:
  • The study elucidated specific pathways, such as RTBDN and NCR3LG1 influencing T2D via body composition changes, and TSPAN8 acting via SHBG.
  • The "Reduced Adiposity" cluster suggests a mechanism where low BMI does not protect against T2D in certain subgroups (e.g., lipodystrophy-like profiles), particularly relevant for East Asian populations.

5. Significance

• Precision Medicine: By identifying distinct protein clusters (e.g., Liver, Kidney, Reduced Adiposity), the study supports the stratification of T2D patients into molecular subtypes, which could lead to more targeted therapeutic interventions.
• Drug Discovery: The 20 identified causal proteins (19 in EUR, 4 in EAS) represent high-priority drug targets. Specifically, RTBDN, TSPAN8, NCR3LG1, and ENTR1 are highlighted as novel candidates for therapeutic development.
• Biomarker Potential: These proteins serve as potential biomarkers for risk stratification, early detection, and predicting disease progression or complications (e.g., kidney/liver disease).
• Ancestry Insights: The findings highlight that while core mechanisms are shared, specific pathways (like ENTR1 in EAS) are ancestry-specific, emphasizing the need for diverse population studies in genomics and proteomics to avoid missing critical therapeutic targets.

Conclusion: This study provides a comprehensive map of the molecular architecture of T2D, moving beyond a "one-size-fits-all" view to a cluster-based understanding. It successfully integrates proteomics, metabolomics, and genetics to identify causal proteins and pathways, offering a roadmap for precision prevention and treatment strategies.