The same measurement across samples and laboratories
Three studies of epigenetic age, brain heritability, and catalyst performance show how one measurement can yield different answers when the sample, calculation, or laboratory changes.
Science··Evening
The clock shifts with the sample
A peer-reviewed eLife study tested six widely used epigenetic age clocks in the 621-person MAGENTA sample and in three replication cohorts totaling more than 2,500 people. These clocks estimate biological age from methylation patterns at CpG sites on DNA. The correlation between age from the Horvath clock and chronological age was 0.72 among White participants, but fell to 0.51 among African American participants and 0.45 among Puerto Rican participants. The authors connect part of that shift to genetic structure: 24 percent of the CpG sites used by the clock are differentially methylated between African and European ancestry, while 77 percent lie under genetic variants that alter methylation. Age acceleration associated with Alzheimer's disease was significant only in the White sample. The study is observational, cannot separate environmental contributions from genetic ones, and does not represent every ancestry combination. Even with those bounds, its comparison identifies a concrete sample effect: the same clock does not reproduce the same strength of relationship with chronological age across genetic compositions. The resulting age therefore depends both on the selected methylation sites and on the population in which that selection is developed and applied.[1]
One scan, three heritability estimates
A second eLife study examined 178 people from the Human Connectome Project's 7-tesla imaging release while they watched short film clips; the participants included 51 monozygotic and 34 dizygotic twin pairs. Researchers extracted three different measures from the same brain-imaging sessions. Heritability of the raw brain signal stayed around 0.064, whereas heritability of the connectivity profile—how brain regions vary together—was 0.36. The estimate for the neural timescale was 0.38 on the first day and 0.75 on the second. That wide change between sessions indicates that this measure has not yet yielded a stable answer. Alignment methods that superimpose each person's functional brain map also reduced raw-signal heritability by 33 percent on the first day and connectivity-profile heritability by 39 percent. Images from the same people engaged in the same activity therefore yielded different heritability values depending on whether researchers selected the raw signal, connectivity, or timescale, and on how they aligned the maps. The sample places another boundary around those numbers: close to 90 percent of participants identified as White, and everyone was between 22 and 36 years old. Values obtained in that narrow group cannot be transferred directly to other populations.[2]
The relationship disappears across laboratories
A four-institution round-robin made the same issue visible at the level of an experimental setup. SLAC National Accelerator Laboratory, Pennsylvania State University, Stanford, and UC Santa Barbara ran the same rhodium-on-titania catalyst batches for carbon-dioxide hydrogenation in four separate laboratories under a shared testing protocol. In the peer-reviewed Nature Catalysis study, reaction temperature, rhodium loading, and synthesis method were inputs; conversion, selectivity, and carbon-monoxide and methane production rates were outputs. Input-output relationships that looked clear in data from one laboratory lost statistical significance when interlaboratory variability was included. The authors identify heat management as the largest contributor to that variation. The three developments operate at different scales: one compares human populations with different genetic compositions, another compares brain measures derived from the same scans, and the third compares laboratories following a shared protocol. Read together, they show how difficult it is to separate a measurement from its context. A number presented as age, heritability, or catalyst performance also carries its sample, calculation choice, and experimental environment. Changing those contexts can make a measurement with the same name return a different answer.[3], [1], [2]
Related columns
For more information on this topic, you can read the related columns.
