💻 bioinformatics

Fasting Status and Epigenetic Clock Stability: Implications for Aging Research

This study demonstrates that acute fasting induces small, systematic shifts in specific epigenetic clocks—particularly PC-based and SystemsAge variants—driven by immune cell redistribution, whereas mortality-trained clocks remain stable, highlighting that clock reliability is context-dependent and requires perturbation-specific assessment beyond standard ICC metrics.

Original authors: Seale, K. B., Dwaraka, V. B., Giosan, I., Mendez, T., Smith, R.

Published 2026-06-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Seale, K. B., Dwaraka, V. B., Giosan, I., Mendez, T., Smith, R.

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

Imagine your body has a built-in "biological clock" made of tiny chemical tags on your DNA. Scientists use these tags to estimate how old you are, similar to checking a watch to see the time. Recently, researchers were worried because these "DNA watches" seemed a bit shaky—they didn't always give the same reading if you tested them a few times, leading to doubts about whether they could be trusted for medical use.

This paper investigates a specific question: Does skipping breakfast (fasting) and then eating again (refeeding) make these DNA clocks jump around?

Here is the breakdown of what they found, using simple analogies:

1. The "Noise" vs. The "Signal"

Think of the DNA clock as a radio. Sometimes the signal is clear, and sometimes there is static.

The Finding: The researchers found that the "static" (measurement error) is actually very low. The DNA tags are stable and don't change just because of machine errors.
The Twist: However, when people fasted and then ate, the "signal" (the actual age reading) did shift slightly for some clocks. It wasn't that the clock was broken; it was that the clock was reacting to the body's change in state.

2. Not All Clocks Are Built the Same

The study looked at 24 different types of these clocks. They found that different clocks react to food in very different ways, just like different types of thermometers react to the weather.

The "Sensitive" Clocks (The PC-based and SystemsAge clocks):
Imagine these clocks are like a sensitive weather vane. When the wind (food) changes, the vane spins.
- What happened: After eating, these clocks said people looked significantly "younger" (by about 1 to 3 years) compared to when they were fasting.
- Why: These clocks are heavily influenced by your immune cells (the body's defense team). When you eat, your immune cells move around and change their activity. Since these clocks listen closely to the immune cells, they get "confused" and change their age reading.
The "Sturdy" Clocks (The Mortality-trained clocks like GrimAge):
Imagine these clocks are like a heavy stone anchor. They are built to ignore the wind.
- What happened: These clocks didn't budge at all. Whether the person was fasting or had just eaten, the age reading stayed exactly the same.
- Why: These clocks were trained specifically to predict death and long-term health, so they ignore the short-term "noise" of immune cell movements.

3. The "Immune Cell" Filter

The researchers realized that the reason some clocks jumped around was because they were paying too much attention to the immune system.

The Experiment: They created a mathematical "filter" to remove the immune cell influence from the data.
The Result: Once they filtered out the immune cells, the "sensitive" clocks stopped jumping around. The "younger" reading disappeared, and the clocks went back to their normal, stable state.
The Takeaway: The change wasn't a glitch; it was the clock correctly detecting a change in the immune system, which happens naturally after eating.

4. The "Group Average" Trap

Here is the most important lesson from the paper:

The Problem: If you look at a group of people and calculate an "average" reliability score (like a grade for the clock), you might miss the fact that the clock is actually reacting to the food. It's like saying a thermometer is "reliable" because it always reads 98.6°F, even if it's actually reading 98.6°F for a healthy person and 98.6°F for a person with a fever (because the fever is small compared to the normal range).
The Solution: You can't just ask, "Is this clock reliable?" You have to ask, "Is this clock reliable specifically when someone has just eaten?"

Summary

The paper concludes that fasting does cause small, temporary shifts in some DNA age clocks, but only the ones that are sensitive to immune cells. The "sturdy" clocks designed for long-term health predictions don't care about a single meal.

The main message is: Don't judge a clock by a single score. To know if a clock is good, you have to test it under specific conditions (like fasting) and understand why it changes. If a clock changes because your immune system is reacting to food, that's actually a sign it's working correctly, not that it's broken.

Technical Summary: Fasting Status and Epigenetic Clock Stability

Problem Statement
Epigenetic clocks, which utilize DNA methylation patterns as biomarkers for aging, are increasingly central to aging research and clinical trials. However, a recent assessment of 18 clocks across short-term perturbations suggested that most demonstrate only moderate biological reliability, raising concerns regarding their translational utility. A critical limitation in current reliability assessments is the tendency to pool metrics across heterogeneous populations and array platforms. This approach may obscure the specific mechanisms driving variability, as different clocks are constructed to capture distinct biological properties that respond uniquely to specific perturbations. Furthermore, relying solely on reliability metrics like the Intraclass Correlation Coefficient (ICC) without accounting for systematic group-level effects can lead to misleading conclusions about a clock's stability.

Methodology
To address these gaps, the authors evaluated 24 epigenetic clocks spanning five construction categories:

First and second-generation classical clocks (e.g., Horvath, Hannum, PhenoAge).
Principal Component (PC) versions of the classical clocks.
SystemsAge organ-system clocks.
Mortality-trained clocks (GrimAge, PCGrimAge, OMICmAge).
Pace of aging clocks (DunedinPACE) and the IntrinClock.

The study utilized three distinct datasets:

A within-person paired fasting design ( $n=15$ pairs) to assess acute effects of fasting followed by refeeding.
A cross-sectional cohort comparing fasted vs. non-fasted individuals ( $n=2,895$ ).
EPICv2custom technical replicates ( $n=96$ samples from 4 individuals) to establish the technical measurement floor.

For each clock, the authors quantified acute fasting effects with and without immune cell adjustment. They decomposed variance into between-person and within-person components at successive adjustment levels (Raw, EAA, and IAA) and benchmarked biological variability against technical noise.

Key Results

Fasting Effects and Variance: Acute refeeding after an overnight fast induced group-level shifts of 0.5 to 3 years in immune-sensitive clocks. Despite these shifts, within-person reliability remained high (median Raw clock ICC $\approx$ 0.96). This apparent contradiction is explained by the fact that fasting-induced shifts are small relative to the age-driven between-person variance that dominates the ICC denominator.
Clock-Specific Responses:
- PC-based Clocks: Showed larger effects than their classical counterparts (e.g., PC Hannum shifted -2.03 years vs. Hannum -1.37 years).
- SystemsAge: Exhibited the largest effects, appearing 1.15 to 2.9 years younger when fasted.
- Mortality-Trained & Pace Clocks: GrimAge (V1/V2), OMICmAge, and DunedinPACE showed no detectable acute fasting effect (all FDR $p > 0.10$ ).
- IntrinClock: As a negative control designed to exclude immune-variable CpGs, it showed no fasting effect in either cohort.
Role of Immune Cell Adjustment: Adjusting for immune cell composition significantly attenuated or eliminated fasting effects in sensitive clocks (e.g., 88% attenuation for PC Hannum; 99.7% for SystemsAge Blood). No clock retained a significant fasting effect after FDR-corrected immune adjustment.
Mechanistic Correlation: In the cross-sectional cohort, a clock's "immune content" (the fraction of age-independent variance explained by immune cell composition) correlated strongly with the degree to which immune adjustment attenuated its fasting effect ( $r = 0.68, p = 0.003$ ).
Technical vs. Biological Variance: Technical replicates confirmed near-perfect measurement reproducibility (median Raw ICC > 0.97), confirming that variance observed in fasting pairs reflects biology rather than noise.
ICC Dynamics: Immune-adjusted ICCs behaved differently depending on the clock's composition. For clocks where fasting generated within-person variance, immune adjustment removed this noise, increasing the ICC (e.g., SystemsAge EAA 0.768 to IAA 0.913). Conversely, for clocks unaffected by fasting, immune adjustment removed between-person structure, causing the ICC to fall substantially (e.g., OMICmAge 0.922 to 0.160).

Significance and Conclusions
The paper argues that acute refeeding elicits small, systematic shifts in specific epigenetic clocks, particularly those that concentrate correlated CpG variance associated with immune cell composition (such as PC-based and SystemsAge clocks). In contrast, mortality-trained clocks appear resistant to these acute perturbations.

The authors conclude that a reliability-only framework, which reports ICC without testing for systematic group-level effects, is insufficient and can miss structured biological variation. They emphasize that ICC is not a fixed property of a clock; rather, it is shaped by study design, population heterogeneity, the specific perturbation, and the adjustment applied. Consequently, the paper recommends that clock reliability be assessed on a perturbation-specific, clock-by-clock basis, incorporating variance decomposition at each adjustment level and explicit benchmarking against technical replicates.