Loading
What changed between revision 1 and revision 2, compared over the article's markdown source with its frontmatter — so a change to status, evidence or reversibility shows up as the edit it is.
Older · revision 1
Import· initial corpus import, not an edit
Initial import of content/steve-horvath.md — the filesystem corpus, unchanged. Not an edit.
12.9 kB
Revision 1
Revision 2
lines 7–13 → 7–136 unchanged lines not shown
categories: ["people", "longevity"]categories: ["people", "longevity"]tags: ["epigenetics", "biomarkers", "aging", "dna methylation", "biostatistics", "measurement"]tags: ["epigenetics", "biomarkers", "aging", "dna methylation", "biostatistics", "measurement"]summary: "German-American biostatistician who built the first multi-tissue epigenetic clock, giving aging research a quantitative readout it could test and argue about."summary: "German-American biostatistician who built the first multi-tissue epigenetic clock, giving aging research a quantitative readout it could test and argue about."updated: "2026-07-27"updated: "2026-10-29"issues: ["Needs a sourced birth year and education detail; the infobox omits Born and Education rows."]issues: ["Needs a sourced birth year and education detail; the infobox omits Born and Education rows."]------ lines 36–45 → 36–4522 unchanged lines not shown
## Overview## Overview Horvath is a statistician by trade rather than a bench biologist, and the clock reflects thatHorvath is a statistician by trade rather than a bench biologist, and the clock reflects thatorigin. It was not derived from a theory of what ages cells. It was fitted — a penalised regressionorigin. It was fitted, not derived from a theory of what ages cells: a penalised regression was runrun over methylation arrays until a small subset of sites reproduced the calendar. That method isover methylation arrays until a small subset of sites reproduced the calendar. That method accountsthe source of both the clock's reach and its central weakness, and Horvath has generally been morefor both the clock's reach and its central weakness, and Horvath has generally been more explicitexplicit about the second than his field's publicity has been.about the weakness than his field's publicity has been. Before the clock he was known for network methods in genomics. Weighted gene co-expression networkBefore the clock he was known for network methods in genomics. Weighted gene co-expression networkanalysis, developed with colleagues at UCLA in the mid-2000s, groups genes into correlated modulesanalysis, developed with colleagues at UCLA in the mid-2000s, groups genes into correlated moduleslines 52–59 → 52–596 unchanged lines not shown
Horvath trained first in mathematics in Germany and then in biostatistics, and spent most of hisHorvath trained first in mathematics in Germany and then in biostatistics, and spent most of hisacademic career at UCLA, holding appointments in human genetics and in biostatistics. In 2022 heacademic career at UCLA, holding appointments in human genetics and in biostatistics. In 2022 hemoved to [[altos-labs]], the reprogramming company that had launched publicly that January, where hemoved to [[altos-labs]], the reprogramming company that had launched publicly that January, where heworks on measurement for rejuvenation programmes — a natural destination, since a company attemptingworks on measurement for rejuvenation programmes. A company attempting to reset cellular age needsto reset cellular age needs some way to say whether it has.some way to say whether it has succeeded. ```timeline```timeline[[lines 74–97 → 74–9614 unchanged lines not shown
an elastic-net regression against chronological age, and reported that 353 CpG sites sufficed toan elastic-net regression against chronological age, and reported that 353 CpG sites sufficed topredict age with a median error of roughly three and a half years.predict age with a median error of roughly three and a half years. The number was not the point. Single-tissue predictors already existed, and a blood clock is aThe error figure was not the main result. Single-tissue predictors already existed, and a bloodreasonable thing to expect: leukocyte populations shift with age in ways an assay can see. Whatclock is a reasonable thing to expect, since leukocyte populations shift with age in ways an assayHorvath showed was that a single set of coefficients, fitted once, worked on brain, breast, kidney,can see. Horvath showed that a single set of coefficients, fitted once, worked on brain, breast,liver, saliva and buccal cells alike. That implied the methylation changes it tracks are not akidney, liver, saliva and buccal cells alike. That implied the methylation changes it tracks areproperty of one cell lineage's turnover history but something closer to a shared clock running insomething closer to a shared clock running in most of the body than a property of one cellmost of the body.lineage's turnover history. One side result in the same paper did as much work as the main one: the predictor returned close toOne side result in the same paper proved as influential as the main one. The predictor returnedzero for [[induced-pluripotent-stem-cells]], meaning that reprogramming a cell resets the readingclose to zero for [[induced-pluripotent-stem-cells]], meaning that reprogramming a cell resets thealong with everything else. A second oddity came out of his group's later tissue comparisons. Thereading along with everything else. A second oddity came out of his group's later tissue comparisons. Thecerebellum reads consistently younger than the rest of the body, and no settled explanation for thatcerebellum reads consistently younger than the rest of the body, and no settled explanation for thathas emerged.has emerged. > [!key] Why a number changed the field> [!key] Why a number changed the field> Before 2013, arguments about whether an intervention slowed aging in people had no endpoint short> Before 2013, arguments about whether an intervention slowed aging in people had no endpoint short> of waiting for deaths. A cheap molecular readout gave the field something it could put in a trial> of waiting for deaths. A cheap molecular readout gave the field something it could put in a trial> protocol. Whether it deserves that role is a separate question from whether it transformed the> protocol. It transformed the conversation; whether it deserves that role is a separate question.> conversation, and it did. ## Age acceleration## Age acceleration lines 107–113 → 106–1129 unchanged lines not shown
association with obesity, chronic infection, socioeconomic position and several genetic conditions,association with obesity, chronic infection, socioeconomic position and several genetic conditions,findings that gave the measure clinical face validity well before anyone could say what it wasfindings that gave the measure clinical face validity well before anyone could say what it wasmeasuring. It sits alongside older candidates such as [[telomeres-and-telomerase|telomere length]]measuring. It sits alongside older candidates such as [[telomeres-and-telomerase|telomere length]]in the search for usable [[aging-biomarkers]], and outperforms them on precision by a wide margin.in the search for usable [[aging-biomarkers]], and is far more precise than they are. > [!note] What "the Horvath clock" refers to> [!note] What "the Horvath clock" refers to> The phrase is used for the 2013 multi-tissue model and, loosely, for any clock from his group.> The phrase is used for the 2013 multi-tissue model and, loosely, for any clock from his group.lines 117–124 → 116–1233 unchanged lines not shown
## The correlate problem## The correlate problem The limitation is structural, not a matter of better data. A model fitted to reproduce chronologicalBetter data would not remove the limitation, which is structural. A model fitted to reproduceage is by construction a correlate of chronological age. Nothing in the fitting procedure canchronological age is by construction a correlate of chronological age. Nothing in the fitting procedure candistinguish a site that drives aging from one that merely records it, and no epigenetic clock hasdistinguish a site that drives aging from one that merely records it, and no epigenetic clock hasbeen shown to track a causal driver rather than a downstream consequence. Horvath and Kenneth Rajbeen shown to track a causal driver rather than a downstream consequence. Horvath and Kenneth Rajhave said as much in review, describing the clock as plausibly reading out an epigenetic maintenancehave said as much in review, describing the clock as plausibly reading out an epigenetic maintenancelines 128–136 → 127–1353 unchanged lines not shown
built the main ones. DNAm PhenoAge, led by Morgan Levine, was trained on a composite of clinicalbuilt the main ones. DNAm PhenoAge, led by Morgan Levine, was trained on a composite of clinicallaboratory measures rather than on age alone.[^levine2018] GrimAge went further still, constructinglaboratory measures rather than on age alone.[^levine2018] GrimAge went further still, constructingmethylation surrogates for plasma proteins and for smoking pack-years and regressing those on timemethylation surrogates for plasma proteins and for smoking pack-years and regressing those on timeto death.[^lu2019] Both predict mortality substantially better than the 2013 clock. Both are alsoto death.[^lu2019] Both predict mortality substantially better than the 2013 clock, and both arefurther from being mechanistic: GrimAge carries a large share of its weight in a smoking surrogate,further from being mechanistic. GrimAge carries a large share of its weight in a smoking surrogate,which is a strong predictor of dying and a weak claim about aging.which strongly predicts dying but says little about aging. > [!caution] What a moved clock does not establish> [!caution] What a moved clock does not establish> Showing that a treatment lowers a methylation age reading does not show it slowed aging. The> Showing that a treatment lowers a methylation age reading does not show it slowed aging. Thelines 144–151 → 143–1507 unchanged lines not shown
The Mammalian Methylation Consortium, which Horvath helped organise, profiled tissues from dozens ofThe Mammalian Methylation Consortium, which Horvath helped organise, profiled tissues from dozens ofmammalian species on a shared array targeting CpG sites conserved across mammals, and built clocksmammalian species on a shared array targeting CpG sites conserved across mammals, and built clocksthat estimate age in a mouse, a bat and a bowhead whale from one set of coefficients.[^lu2023] Thethat estimate age in a mouse, a bat and a bowhead whale from one set of coefficients.[^lu2023] Thiscomparative programme is the closest thing the clock literature has to a test of whether methylationcomparative programme is the clock literature's closest approach to testing whether methylationchange is fundamental to mammalian aging or an artefact of how any one species' cells divide. Itchange is fundamental to mammalian aging or an artefact of how any one species' cells divide. Italso supplies a practical instrument for animals whose lifespans can actually be measured, a usealso supplies a practical instrument for animals whose lifespans can actually be measured, a usedescribed under [[epigenetic-clock]]; whether the same relationships hold in animals ofdescribed under [[epigenetic-clock]]; whether the same relationships hold in animals oflines 160–174 → 159–1738 unchanged lines not shown
deliberately chose functional measures instead.deliberately chose functional measures instead. Horvath's own public posture has been unusually restrained for the field. He has repeatedly framedHorvath's own public posture has been unusually restrained for the field. He has repeatedly framedthe clock as a measurement in search of a mechanism rather than as a diagnosis, which distinguishesthe clock as a measurement in search of a mechanism rather than as a diagnosis. That sets him aparthim from the consumer market that grew up around his work — direct-to-consumer methylation agefrom the consumer market that grew up around his work: direct-to-consumer methylation age tests,tests, discussed under [[epigenetic-clock]] and [[consumer-blood-testing]], routinely report adiscussed under [[epigenetic-clock]] and [[consumer-blood-testing]], routinely report a single numbersingle number with a confidence the underlying assay reliability does not support.with a confidence the underlying assay reliability does not support. The open question his work sets up is whether the methylation changes he found are part of theHis work leaves open whether the methylation changes he found are part of the machinery of aging ormachinery of aging or its exhaust. Both readings are consistent with everything measured so far, andits exhaust. Both readings are consistent with everything measured so far, and they imply completelythey imply completely different research programmes: one in which [[epigenetic-reprogramming]] is adifferent research programmes. In one, [[epigenetic-reprogramming]] is a therapy; in the other, ittherapy and one in which it is cosmetics for a biomarker.is cosmetics for a biomarker. ## See also## See also removed, struck through added, underlinedLine numbers count the serialised markdown of each revision, frontmatter included.
161 lines are the same in both revisions, and 99 of them are not shown. Neither revision here is necessarily the current article — read Steve Horvath for the text this wiki stands behind today.