📊 epidemiology

Connecting diet and disease: Using Mendelian randomisation to bridge the gap

This paper proposes and validates a two-step framework that integrates short-term dietary randomised controlled trials with Mendelian randomisation to identify molecular mediators and infer the long-term causal effects of dietary interventions on disease risk, demonstrating its utility through the DiRECT diabetes remission study.

Original authors: Deslandes, B., Corbin, L. J., Goudswaard, L. J., Sandu, M. R., Lee, M. A., Beynon, R. A., McGeagh, L., Smith, G. D., Sattar, N., Lean, M. E., Taylor, R., Lane, J. A., Timpson, N. J., Martin, R. M., Ri

Published 2026-09-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Deslandes, B., Corbin, L. J., Goudswaard, L. J., Sandu, M. R., Lee, M. A., Beynon, R. A., McGeagh, L., Smith, G. D., Sattar, N., Lean, M. E., Taylor, R., Lane, J. A., Timpson, N. J., Martin, R. M., Richenberg, G., Gunter, M. J., Yarmolinsky, J., Koumanov, F., Gonzalez, J. T., Richmond, R. C., Vincent, E. E.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Food is the most powerful medicine we have, yet proving exactly how a specific meal changes our long-term health is one of the hardest puzzles in science. We know that what we eat matters, but watching a person eat a certain diet for twenty years to see if they develop a disease is often impossible. It is too expensive, too difficult to control, and ethically risky to ask people to stick to a strict plan for decades. Scientists have two main tools to solve this. One is the randomized controlled trial, where people are assigned to a diet for a short time to see immediate changes. The other is a method called Mendelian randomization, which uses a person's genetic code as a natural experiment to guess what would happen if they had different levels of a trait, like a specific protein in their blood, over a lifetime. The problem is that the first tool only sees the short term, while the second tool struggles to handle the messy complexity of a whole diet.

A team of researchers has now built a bridge between these two worlds. They created a two-step approach that uses the immediate biological changes seen in a short diet trial to predict long-term disease risks. Instead of trying to track a diet for twenty years, they looked at how a diet changed the body's chemistry over one year, and then used genetic data to see if those specific chemical changes were linked to the risk of developing disease later in life. This method allows scientists to take a snapshot of a short-term intervention and extend its meaning into the future, offering a clearer path to understanding how food prevents or causes illness.

The researchers tested this new framework using data from a famous study called the Diabetes Remission Clinical Trial, or DiRECT. In that original study, people with type 2 diabetes were put on a strict, low-calorie diet to see if they could reverse their condition. The trial was a success, with many participants achieving remission, but the researchers wanted to understand the biological machinery behind that success. They looked at the blood of 149 people in the intervention group and 149 in the control group. They measured 4,601 different proteins, which are tiny molecules that act as the body's workers and messengers. After one year, they found that 216 of these proteins had changed significantly in the people who lost weight compared to those who did not. These changes were the first step: identifying which parts of the body responded to the diet.

The second step was to ask if those specific changes mattered for the future. The researchers took the list of 216 proteins that changed and asked a different question: if a person naturally had higher or lower levels of these proteins because of their genes, would that change their risk of getting type 2 diabetes? They used genetic data from hundreds of thousands of people to answer this. This is where the method shines. While the diet trial showed what happened in one year, the genetic analysis showed what would likely happen over a lifetime. The results were striking. Out of the 216 proteins that changed, ten showed a clear link to the risk of diabetes. For six of these, the direction of change made sense. For example, the diet caused levels of a protein called NCAN to go up, and the genetic data showed that having higher levels of NCAN naturally lowers the risk of diabetes. Similarly, the diet lowered levels of a protein called BDH2, and genetics suggested that having lower levels of this protein also lowers risk.

To check if their new method was actually working, the researchers compared their genetic predictions against the real-world results of the original trial. They looked at whether the proteins that changed in the trial were the same ones that predicted who would go into remission. The two sets of results matched up well. The proteins that the genetic method said were good for health were the same ones that increased in the people who successfully reversed their diabetes. The correlation was strong enough to suggest that the framework is a reliable way to connect short-term diet changes to long-term health outcomes. The researchers calculated a combined score for all the proteins, which showed that the overall effect of the diet-induced changes was to lower the risk of diabetes, matching the real-world success of the trial.

However, the study also found that the method is not perfect. Two of the proteins, ADH1B and CCDC126, did not behave as expected. The diet changed them in a way that the genetic data suggested would be harmful, but the trial showed they were associated with success. The researchers noted that this could happen because the body reacts differently to a sudden diet change than it does to a lifetime of genetic differences, or because the specific proteins behave differently in the blood than in the tissues where they matter most. These discrepancies do not mean the method failed; rather, they highlight the complexity of biology and show that this approach is a tool for prioritizing which biological signals to study further, not a magic wand that solves everything instantly.

The value of this work lies in its ability to turn a short-term observation into a long-term insight. By combining the immediate evidence of a diet trial with the lifetime perspective of genetic analysis, scientists can now identify which molecular changes are truly important for preventing disease. This does not replace the need for long-term studies, but it offers a practical way to bridge the gap when those studies are impossible to run. The findings suggest that the changes in the body's chemistry seen after just one year of a healthy diet are indeed signals of a healthier future, providing a scientific basis for why these interventions work and pointing the way toward better treatments for diabetes and other chronic conditions.

Technical Summary: Connecting Diet and Disease via a Two-Step RCT-MR Framework

Problem Statement
Establishing causality in nutrition research is inherently difficult due to the complex, time-varying, and intercorrelated nature of dietary exposures. While Randomised Controlled Trials (RCTs) offer robust causal inference, long-term dietary intervention studies with hard disease endpoints are often impractical due to cost, duration, and ethical constraints. Consequently, many dietary RCTs are limited to short durations and intermediate outcomes, which may not adequately capture long-term disease relationships. Conversely, Mendelian Randomisation (MR) uses genetic variants as proxies for exposures to estimate lifelong effects but is often limited in nutrition research. This is because robust genetic instruments for complex dietary patterns are difficult to identify, and genetic predisposition to a behavior may not recapitulate the biological effects of an environmental dietary change (violating the gene-environment equivalence assumption).

Methodology
The authors propose a two-step framework that integrates short-term dietary intervention RCTs with MR to infer potential lifetime effects of dietary interventions. The framework operates as follows:

Step 1 (RCT Identification): Data from a short-term dietary intervention RCT are used to identify intermediate molecular traits (e.g., circulating proteins, metabolites) that change in response to the intervention relative to a control group.
Step 2 (MR Causal Inference): These intervention-responsive intermediate traits are treated as exposures in MR analyses. Using genetic variants (typically cis-protein quantitative trait loci, or pQTLs) as instrumental variables, the framework estimates the causal effect of lifelong differences in these specific molecular traits on long-term disease risk.

The authors demonstrate this framework using data from the Diabetes Remission Clinical Trial (DiRECT).

Step 1 Application: Using DiRECT data (Lean et al., 2018), they identified 216 of 4,601 circulating proteins that significantly changed following a structured weight-loss intervention compared to standard care (Goudswaard et al., 2023).
Step 2 Application: They performed summary-level MR analyses using genetic instruments for these proteins to estimate their effect on Type 2 Diabetes (T2D) risk, utilizing data from the DIAGRAM consortium.
Validation: To validate the framework, the authors compared the MR-derived effect estimates against observed protein-remission associations within the DiRECT trial using a Prognostic Association Analysis (PAA). They also calculated an "integrated score" by multiplying the intervention effect on protein levels (Step 1) by the MR effect of that protein on disease risk (Step 2) to summarize the overall direction of the intervention's molecular impact.

Key Contributions

Methodological Integration: The paper introduces a novel methodological approach that bridges the gap between short-term intervention data and long-term causal inference, addressing the complementary limitations of RCTs (short duration) and MR (difficulty with complex dietary patterns).
Empirical Demonstration: The study provides a worked example using DiRECT proteomic data, successfully identifying specific molecular intermediates that link a dietary intervention to T2D risk.
Validation Strategy: The authors validate their framework by demonstrating concordance between MR estimates (T2D risk) and trial-based prognostic associations (T2D remission), despite these being distinct clinical outcomes.

Results

Identification of Intermediates: Of the 216 proteins altered by the DiRECT intervention, 146 had available genetic instruments. MR analysis identified 10 proteins with evidence of a causal effect on T2D risk (FDR-P < 0.05).
Concordance: Six of these proteins showed intervention-induced changes consistent with a reduction in T2D risk (e.g., increased NCAN, NEGR1, DUSP13, SHBG, CDON; decreased ADH1B). Conversely, changes in BDH2, APOA4, HTRA1, and CCDC126 were associated with higher risk.
Correlation: A comparison between the two-step framework (MR) and the trial-based PAA revealed a moderate inverse correlation ( $r \approx -0.604$ ). This inverse relationship is expected because MR estimates T2D risk (where higher risk is negative), while PAA estimates T2D remission (where higher remission is positive).
Concordance Rate: Of the 10 proteins identified by the framework, 8 showed concordant directional effects in both the MR and PAA analyses.
Integrated Score: The integrated score across the ten proteins was negative (-0.0155), indicating that the combined intervention-induced changes in circulating proteins were overall associated with a predicted reduction in T2D risk, consistent with the observed high rates of remission in the DiRECT trial.

Significance and Claims
The paper claims that this two-step framework provides a practical strategy for strengthening causal inference in nutrition research. By linking short-term biological responses observed in dietary trials with long-term disease risk estimated via genetic instruments, the approach allows researchers to:

Prioritise candidate subclinical endpoints (molecular intermediates) for further investigation.
Infer potential lifetime effects of dietary interventions without requiring impractical long-term RCTs.
Overcome the difficulty of performing MR on complex, whole-diet patterns by focusing on well-defined molecular traits.

The authors emphasize that the framework is intended to complement, not replace, independent RCTs or MR studies. They acknowledge limitations, including the assumption that short-term molecular changes reflect long-term biological adaptations, the potential for non-linear relationships, and the distinction between T2D risk and remission as outcomes. The study positions the DiRECT analysis as a "positive control" example, suggesting that robustness requires application across multiple interventions and disease outcomes in future work.

Technical Summary: Connecting Diet and Disease via a Two-Step RCT-MR Framework

More like this