Deep divefrom Episode 12Longevity10 November 202622 min

When Your Weight Strays From Your Genes

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: BMI Deviations from Genetic Prediction Are Associated with Distinct Molecular Signatures in Finnish Population Cohorts (medRxiv preprint)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science, where artificial intelligence reads real science, so you don't have to pretend you understood the paper.

Host 2Today is one paper, a preprint posted on medRxiv on October twenty-third by fifteen researchers, most of them at the University of Helsinki, with colleagues in Tampere, Turku, Oulu, Kuopio and Amsterdam. It asks what happens in the blood when a person's weight differs from what their genes would predict.

Host 1Before we get into it, what is the one-sentence answer they came back with?

Host 2Some blood markers behave differently depending on whether you weigh more or less than your genes predict, but the signals are small, most do not survive strict statistical correction, and the paper has not yet been peer reviewed. Inflammation and blood pressure rose more steeply above the prediction, while several metabolites and proteins moved more steeply below it.

Host 1And when you say preprint, does that mean nobody outside the team has checked this yet?

Host 2Yes, exactly. The paper itself carries the warning that it has not been certified by peer review and should not be used to guide clinical practice.

Why genes and weight got tangled

Host 1Let me start at the beginning, then: why did anyone think genes had much to do with how heavy we are?

Host 2Because twins told them so. Twin, family and adoption studies put the heritability of body mass index somewhere between forty and seventy percent, and the paper notes that this share falls with age as life piles on its own influences.

Host 1When people say forty to seventy percent heritable, does that mean forty to seventy percent of my weight comes from my genes?

Host 2No, and that is the most common misreading. It describes a population, not a person: it says how much of the spread in weight between people in that group tracks with the spread in their genes.

Host 1And how do twins let you work that out in the first place?

Host 2The wider picture in the field is that identical twins share essentially all their DNA, while non-identical twins share about half of the variants that differ between people, on average. If identical pairs resemble each other in weight much more than non-identical pairs do, genes are doing a good part of the work.

Host 1So once they knew genes mattered, how did they go looking for the actual genes?

Host 2With genome-wide association studies, which scan hundreds of thousands of people for single-letter differences in DNA that turn up more often in heavier people. Those studies found a great many such spots, each nudging weight by a tiny amount.

Host 1If each spot only nudges things a little, how do you turn that into something useful about one person?

Host 2You add them up into a polygenic risk score, a weighted sum of all those small nudges. Here the score was built from nine hundred and ninety-six thousand, nine hundred and nineteen common variants, using Finnish reference data.

What was known before

Host 1How good are those scores at actually guessing someone's weight?

Host 2Modest. The paper cites the newest multi-ancestry scores explaining up to seventeen point six percent of the variation in body mass index in people of European ancestry, and in these Finnish cohorts the score explained only about five percent in the twins and ten percent in the second cohort.

Host 1Five percent sounds tiny; is that as weak as it sounds?

Host 2It is weak for predicting any one person. Adding age and sex to the score raised the explained share to nineteen and thirteen percent, so the genetic prediction on its own leaves most of the story untold.

Host 1Then why would the gap between your genetic prediction and your real weight be interesting to anyone?

Host 2Because of two earlier findings the authors build on. One reported that people with obesity and a high genetic score had a lower risk of cardiovascular disease than people with obesity and a low score, and another, in the UK Biobank, found that weighing more or less than genetically predicted was linked to higher mortality in both directions.

Host 1So if I have this right, the idea is that each of us has a kind of genetically expected weight, and straying far from it either way might be the problem?

Host 2That is the hypothesis the authors put forward, yes, a genetically set range of tolerated fat. But it is a hypothesis drawn from those earlier studies, and nobody had looked at what such a gap looks like inside the body.

Host 1What did people already know about weight and the molecules in our blood?

Host 2Quite a lot about weight itself. Earlier studies linked body mass index to widespread changes in DNA methylation, in circulating metabolites, in plasma proteins, and in so-called epigenetic clocks that estimate biological age.

Host 1You just said DNA methylation; what is that in plain words?

Host 2They are small chemical tags sitting on the DNA that help switch genes on and off without changing the sequence itself. The wider picture is that patterns of these tags shift with age and with exposures like smoking and weight, which is what the epigenetic clocks are built from.

The open question

Host 1So what exactly was still unknown before this paper?

Host 2Whether weighing more than your genes predict and weighing less than they predict leave different molecular fingerprints. In the authors' words, the molecular correlates of such deviations remained largely unknown.

Host 1Why does it matter whether you separate those two directions instead of just asking about weight?

Host 2Because a plain straight line forces one slope across the whole range. If being above your prediction strains the body in one way and being below it shows up in another, a single line would blur both.

How they did it

Host 1Who were the people they studied?

Host 2Two long-running Finnish studies. Two thousand five hundred and eight adults from the Finnish Twin Cohort and one thousand seven hundred and fifteen from the Young Finns Study, which began in nineteen eighty and has followed people from childhood into adulthood.

Host 1Was everyone included, or did they leave some people out?

Host 2They kept adults over twenty, without diabetes, with a body mass index between sixteen and forty-five, and with genetic data plus at least one molecular data set. About two thirds of the twin cohort were women, average age forty, while the Young Finns averaged about forty-two with a much narrower age spread.

Host 1How did they actually get people's weight; did someone put them on a scale?

Host 2In the twins it was mostly self-reported height and weight at the time of the blood draw, which earlier work found reliable, while a metabolic sub-group was measured clinically. The paper does not go into how the Young Finns values were taken in the main text.

Host 1So how do they turn all that into one number for the gap between genes and reality?

Host 2They fit a straight line predicting body mass index from the genetic score, age and sex, and what is left over for each person is the residual. A positive residual means you weigh more than predicted, a negative one means less, and they did this separately in each cohort.

Host 1How big are those residuals in practice; are we talking a pound or two?

Host 2Larger than that. In the twin cohort the standard deviation was four point two units of body mass index, which for someone of average height is very roughly a dozen kilograms.

Host 1And what did they measure in the blood to compare against that number?

Host 2Four layers, depending on who had what. Clinical measures like blood pressure and C-reactive protein in three hundred and ninety-two people from the metabolic sub-group, a standard panel of metabolites in over a thousand twins and all the Young Finns, DNA methylation at about three hundred and forty-four thousand sites, and two thousand three hundred and twenty-one plasma proteins in three hundred and eighty-one twins.

Host 1You mentioned C-reactive protein; what does that tell a doctor?

Host 2It is a protein the liver releases when there is inflammation in the body. Long, low-grade rises in it are a common marker in heart and metabolic research.

Host 1How does their statistical method separate the above-prediction people from the below-prediction people?

Host 2They used what is called hinge regression, which lets the line bend at a residual of zero. The model estimates one slope for people below their prediction and another for people above, and a test on the bend tells you whether the two slopes really differ.

Host 1Picture it for me: is that like a road that is flat and then suddenly climbs?

Host 2That is a fair picture. A marker might barely change as the residual rises below zero and then climb steeply once you pass zero, or the reverse, and the bend test asks whether that kink is more than noise.

Host 1With so many proteins and methylation sites, wouldn't some of them look significant just by chance?

Host 2Yes, and that is the right worry. They applied a false discovery rate correction within each data layer, but they say up front that the main results report the uncorrected p values unless stated otherwise.

Host 1Where do the twins come in, if they already had thousands of people?

Host 2Twins let them hold genes and upbringing constant. They looked for same-sex pairs whose body mass index differed by more than three units, then sorted them by which twin had strayed from their genetic prediction.

Host 1How did they decide when someone counted as having strayed?

Host 2Being at least half a standard deviation from the prediction, which is two point one units of body mass index. They tried thresholds from a tenth of a standard deviation up to one and a half, and chose a half because it gave two groups of balanced, usable size.

Host 1And what were those two groups of twins?

Host 2Group A was one hundred and twenty-two pairs where the heavier twin was above prediction and the lighter twin was on target. Group B was one hundred and nine pairs where the heavier twin was on target and the lighter twin was below prediction.

Host 1Were those all identical twins?

Host 2No, which matters. Group A had fifty-eight identical and sixty-four non-identical pairs, Group B thirty-nine identical and seventy non-identical, so the genetic control is complete only in the identical pairs.

Host 1Once you have the pairs, what does the comparison actually look like?

Host 2Each pair gets its own baseline, so anything the twins share drops out, and the model asks whether differences in the residual inside a pair go along with differences in a marker. The key test is whether that within-pair slope differs between Group A and Group B.

A word from our sponsor

Host 1This episode is brought to you by Predicted Coat, the winter coat tailored not to the body you have, but to the body your genes predicted.

Host 2It fits about five percent of me, which the brochure calls a good fit.

Host 1Each coat is cut from nearly one million common variants, adjusted for age and sex, so it is never wrong about you, only about your residuals.

Host 2Mine has a hinge at zero. Above it, the buttons get inflamed.

Host 1Every Predicted Coat ships with a discordant twin coat, one size apart, to control for shared upbringing.

Host 2After a few weeks the two coats stopped needing us and now go for walks alone, matched on everything.

Host 1Predicted Coat. It follows prediction, so you don't have to.

What they found

Host 1Before the main results, did they check that the genetic score was measuring anything real?

Host 2They did. In the twins, a higher score went with higher C-reactive protein, blood pressure, belly fat, waist size, fat mass and the liver enzyme gamma-glutamyl transferase, plus higher triglycerides and lower good cholesterol, which is what you would expect from a score for heavier build.

Host 1So in the clinical group, what changed depending on the direction of the gap?

Host 2In those three hundred and ninety-two people, C-reactive protein and systolic blood pressure rose with the residual above zero but barely moved below zero, with bend tests of p equals zero point zero three seven and zero point zero two six. Going the other way, body fat percentage and the Matsuda index changed more steeply below zero.

Host 1What is the Matsuda index, and should I want it high or low?

Host 2It is a measure of insulin sensitivity from a glucose tolerance test, and higher is better. It fell as the residual rose, more steeply below zero, and that was by far the strongest signal in the clinical layer, a p value of about one in a hundred million.

Host 1Those two p values for inflammation and blood pressure, are they convincing on their own?

Host 2Not very. Values just under zero point zero five in a panel of around twenty clinical measures are the kind that can appear by chance, which is why the twin results, coming later, matter as a second look.

Host 1What about the metabolites, where they had two separate cohorts?

Host 2Five metabolites bent the same way in both: creatinine, histidine, omega-three fatty acids, the omega-three called docosahexaenoic acid, and cholesterol in very-low-density lipoprotein. All five rose with the residual when people were below prediction and barely changed above it.

Host 1Does turning up in two separate cohorts make those five more believable?

Host 2It helps, because a chance hit rarely repeats in the same direction in an independent group. But each still only reached nominal significance, with p values from about zero point zero zero one to zero point zero four.

Host 1Did the DNA methylation tags show anything?

Host 2Not at the strict bar. None of the roughly three hundred and forty-four thousand sites reached genome-wide significance or survived the false discovery correction, and six reached only a looser suggestive level.

Host 1Those six sites, do they point at anything meaningful?

Host 2They sit in or near genes with plausible roles, including one in leptin signalling and one linked to fat tissue. But the authors themselves stress the effect sizes were around one thousandth, and say they should be read with caution.

Host 1And the epigenetic clocks, the biological age estimates you mentioned earlier?

Host 2Here the result was the opposite of a bend. Higher residual went with faster epigenetic ageing on every clock in the twins and most in the Young Finns, and the slope was about the same on both sides of zero.

Host 1So if I have this right, extra weight seems to age the clocks whether or not it matches your genes?

Host 2That is how the authors read it, with one correction: it is an association, not proof that weight causes the ageing. The one exception was a clock called DunedinPACE, which climbed slightly more steeply above prediction in the Young Finns, at p equals zero point zero four.

Host 1Then what turned up in the two thousand or so proteins?

Host 2No single protein survived the false discovery correction. Three hundred and forty-two passed the uncorrected threshold, and the clearest was leptin, which rose with the residual at roughly twice the slope below zero as above it.

Host 1What does leptin do, and why might it flatten out like that?

Host 2Leptin is a hormone made by fat tissue that tells the brain how much energy is stored. The authors suggest the flattening fits leptin resistance at higher fat levels, but that is their interpretation, not something this study tested.

Host 1Three hundred and forty-two out of two thousand three hundred sounds like a lot; is it?

Host 2At a five percent threshold you would expect around one hundred and sixteen by chance alone, so three hundred and forty-two is more than chance, but it does not tell you which ones are real. The pathways they over-represented were inflammation and growth signalling, including cytokine and JAK-STAT signalling.

Host 1Did the twin pairs tell the same story as the big population analysis?

Host 2Partly. Systolic blood pressure and the liver enzyme gamma-glutamyl transferase rose more steeply with the residual in Group A, where the heavier twin was above prediction, at p equals zero point zero zero four and zero point zero three, but none survived the correction.

Host 1How many twin pairs actually went into that clinical comparison?

Host 2Far fewer than the headline numbers. The figure shows thirty-six pairs in Group A and thirty-two in Group B with clinical data, and about sixty-seven and sixty-six with metabolite data.

Host 1And the metabolites and proteins inside the twin pairs?

Host 2Very-low-density lipoprotein cholesterol, saturated fat share, alanine, histidine and polyunsaturated fats tracked the residual more strongly in Group B, while two omega-three measures went the other way, all nominal only. Among proteins, midkine and an adipokine called NAMPT rose in the heavier twin in Group A but fell in Group B, which the authors read as both being lowest near the predicted weight.

Host 1So the overall shape of the findings, as briefly as you can put it?

Host 2Below the prediction, many molecules track weight more steeply; above it, inflammation, blood pressure and a liver enzyme rise more steeply; and epigenetic ageing tracks weight on both sides. The authors call these patterns subtle but consistent, and subtle is the right word.

Why it matters

Host 1What is the bigger idea here, if it holds up?

Host 2That the same body mass index might not mean the same thing for two people whose genes point to different weights. Weighing thirty with a genetic expectation of thirty may differ biologically from weighing thirty with an expectation of twenty-five.

Host 1Is there a simple reason the slopes might be steeper below the prediction that has nothing to do with biology?

Host 2Yes, and to their credit the authors raise it themselves. People can only fall so far below their prediction before reaching a lower physical limit, while people above it can go much higher, so steeper slopes below zero may partly reflect that squeezed range.

Host 1Doesn't that same logic make the above-prediction results harder to dismiss?

Host 2That is their argument. A steeper slope above zero cannot easily come from the range alone, which is why they treat the inflammation and blood pressure findings as the more telling ones, though those were also the weaker p values.

Host 1For someone listening, does this mean they should get a genetic test to see if their weight is right for them?

Host 2No. The score here explained about five percent of weight, the paper says different scores can give very different estimates for the same individual, and nothing was tested against actual illness, so there is no clinical use yet.

Host 1Then who should actually pay attention to this paper?

Host 2Researchers who study why people with similar weight have very different metabolic health. The paper offers a way to frame that question, using the gap from genetic expectation instead of the usual high-score versus low-score split.

Honest caveats

Host 1What are the biggest reasons to be careful with these results?

Host 2First, in the methylation, protein and twin analyses nothing survived the correction for many tests, and elsewhere the paper reports uncorrected p values, so most findings are leads rather than results. Second, the work is a preprint, not yet peer reviewed.

Host 1If the genetic score explains so little, isn't the residual basically just body mass index under another name?

Host 2Largely, yes, and the authors say so plainly. With the score explaining about five percent, the residual stays strongly correlated with weight itself, so some of what looks like an effect of the gap may simply be an effect of weight.

Host 1Can we tell from this whether being above your prediction causes inflammation?

Host 2No. These are snapshots in time, so the study shows things moving together, not what drives what, and the authors list longitudinal data as a future step.

Host 1Who does this apply to; would it hold for someone outside Finland?

Host 2We cannot say from this paper. Everyone was Finnish, people with diabetes were excluded, and the authors themselves call for testing across diverse populations.

Host 1Is there anything about how the twin groups were built that might skew things?

Host 2A few points. The groups mix identical and non-identical pairs because identical pairs alone were too few, smoking exposure was marginally higher in Group B, and the clinical twin comparisons rest on only a few dozen pairs per group.

Host 1Anything smaller that a careful reader might notice?

Host 2The text gives about three hundred and forty-four thousand methylation sites while a figure legend says about three hundred and forty-seven thousand, a minor inconsistency. And in the twin cohort, weight was mostly self-reported rather than measured.

What to watch next

Host 1What would you want to see next before taking this idea seriously?

Host 2Whether the gap predicts actual disease, such as heart disease or liver problems, which the authors name as a key priority. Without outcomes, these are molecular patterns waiting for a reason to matter.

Host 1Could better genetic scores change the picture?

Host 2Quite possibly. The authors point to whole-genome sequencing, which can capture up to thirty percent of the heritability of body mass index, and to scores built for harmful kinds of fat such as visceral fat or waist-to-height ratio.

Host 1And beyond better scores, what kind of study would settle the cause-and-effect question?

Host 2Following the same people over years, with detailed records of their environment, to see whether moving away from the genetic expectation comes before the changes in blood markers. The authors also say the mechanisms need targeted experiments and replication in independent samples.

Sign-off

Host 1So, to wrap up, what should someone take away from this paper?

Host 2That weighing more or less than your genes predict seems to go with somewhat different molecular patterns, especially inflammation and blood pressure above the prediction. The signals are small, mostly uncorrected, and not yet linked to illness, so it is a framework to test, not a finding to act on.

Host 1That's the deep dive. Every weekend this same feed carries the roundup, the week's new papers in one short episode, genetically predicted length not guaranteed.

Host 2Thanks for listening to the Hastrup Review of Science.