🧬 Aging Wiki

symphony-age

Properties1
aliasesSYMPHONYAge, SYMPHONY Age, Systems Age clock, organ-system aging clock, TruDiagnostic organ clock

16 min read

SYMPHONY Age

A DNA-methylation organ-system biological age clock that returns a separate biological age estimate for each of 11 distinct physiological systems from a single blood draw, plus a composite “SYMPHONY Age” summary score. SYMPHONY Age is TruDiagnostic’s commercial implementation of the Systems Age framework developed by Sehgal et al. at Yale University and published in Nature Aging in September 2025 1. It differs conceptually from whole-body second- and third-generation clocks (GrimAge, PhenoAge, DunedinPACE) by decomposing aging heterogeneity: a person may show accelerated aging in one organ system while others appear concordant with or younger than chronological age. The approach is the DNAm analogue of the proteomic organ-clock work by Oh et al. 2023 2, but uses DNA methylation from blood rather than plasma protein levels — a distinct modality and distinct clock.

Identity and Provenance

  • Academic primary source: Sehgal R, Markov Y, Qin C et al. (2025) “Systems Age: a single blood methylation test to quantify aging heterogeneity across 11 physiological systems.” Nature Aging 5(9):1880–1896 — doi:10.1038/s43587-025-00958-3
  • Preprint (2023): bioRxiv doi:10.1101/2023.07.13.548904 (PMC10370047) — precedes the peer-reviewed paper
  • Commercial product: SYMPHONY Age (TruDiagnostic); acronym = System Methylation Proxy of Heterogeneous Organ Years; marketed as “Developed with Yale”
  • IP: Subject to an invention declaration at Yale University and a provisional patent application (Sehgal 2025 competing interests)
  • Key authors (academic): Raghav Sehgal (Yale, Computational Biology and Bioinformatics), Albert T. Higgins-Chen (Yale, Departments of Psychiatry and Pathology), Margarita Meer (Altos Labs, co-inventor); Morgan Levine (Altos Labs, corresponding author). Per preprint competing interests: A.H.C. received consulting fees from TruDiagnostic; R.S. received fees from LongevyTech.fund and Cambrian BioPharma (not TruDiagnostic in the Sehgal preprint). Both A.H.C. and R.S. are listed as TruDiagnostic scientific advisors per the Harvanek 2024 conflicts statement, and are named co-inventors on the licensed Systems Age patent
  • November 2025 retrain: TruDiagnostic states SYMPHONY Age was retrained in November 2025 on ~8,000 participants with 133 clinical biomarkers. No peer-reviewed publication accompanies this version; product-report figures should be interpreted accordingly and comparability with the 2025 academic paper is uncertain.

The 11 Organ Systems

The clock partitions biological aging into 11 system-specific scores, each trained against clinical biomarkers relevant to that system 1:

SystemRepresentative clinical biomarkers / functional measures
BloodCBC indices (RBC, WBC, hemoglobin, MCV, RDW, platelets, lymphocyte %)
BrainCognitive function assessments (WAIS-type), neurological clinical measures
HeartCardiovascular measures (blood pressure, ECG-derived, cardiac biomarkers)
HormoneEndocrine biomarkers (thyroid, insulin, sex hormones)
ImmuneImmune-cell-count-based markers, lymphocyte subsets
InflammationCRP, IL-6, ESR, and related inflammatory biomarkers
KidneyCreatinine, cystatin C, BUN, electrolytes, GFR-based measures
LiverALT, AST, albumin, ALP, bilirubin, GGT
LungFEV1/FVC ratio and spirometry-derived pulmonary function
MetabolicBMI, waist-hip ratio, HbA1c, cholesterol, triglycerides, glucose
MusculoskeletalGrip strength, gait speed, physical performance measures

Note on biomarker lists: The academic paper describes the mapping conceptually; full per-system biomarker lists with counts are in the Methods section of Sehgal 2025. The TruDiagnostic product report states 133 total biomarkers across all 11 systems but does not break this down per system publicly. Exact system-specific biomarker assignments may differ between the published model and the November 2025 retrain. gap/needs-replication

Deep-dive: the Musculoskeletal sub-clock — what it measures, and how to move it

The Musculoskeletal score is the sub-clock most users ask about, and the one most often misread. Two separate things are routinely conflated:

1. What the assay physically measures from blood = nothing but CpG methylation. The only data a dried-blood-spot yields is methylation beta-values at the array’s CpG sites; the Musculoskeletal score is a sparse elastic-net blood-methylation surrogate of those values 1. TruDiagnostic does not publicly disclose which CpGs load on the Musculoskeletal system, and — as for every clock — those CpGs were selected for predictive power, not because they sit in muscle-causal genes. There is therefore no published “muscle gene panel” to read off this score; mapping its CpGs to mechanism is interpretively fraught.

2. What the score was trained to predict = a composite of musculoskeletal clinical + functional biomarkers measured in the training cohorts (primarily the Health and Retirement Study, HRS), not in the customer’s blood draw. This is the “blood markers used for muscle” most people are actually asking about. The academic paper does not publish the exact per-system list, but combining (a) the TruDiagnostic SYMPHONY report’s stated Musculoskeletal inputs, (b) the paper’s HRS clinical-chemistry + functional-measure basis 1, and (c) TruDiagnostic’s related “Fitness Age” panel (VO2max, FEV1, grip strength, gait speed), the Musculoskeletal training target reconstructs to:

InputTypeWhy it indexes musculoskeletal aging
Grip strength (hand dynamometer)physical-functionDirect strength readout; best-validated muscle-aging functional biomarker; defines sarcopenia (EWGSOP2, sarcopenia)
Gait / walking speedphysical-functionWhole-body function + lower-limb power; strong independent mortality predictor
Balance / standing balancephysical-functionNeuromuscular integration; falls risk
Mobility-difficulty (self-report)self-reportFunctional limitation
BMI / body weightanthropometricBody composition; high adiposity → myosteatosis + anabolic resistance
IGF-1 (insulin-like growth factor 1)serum analyteGH→IGF-1 anabolic axis; declines with age, lower in sarcopenia (igf-1-biomarker)
DHEA-S (dehydroepiandrosterone sulfate)serum analyteAdrenal androgen precursor; “adrenopause” decline; supports muscle anabolism
25-OH vitamin Dserum analyteVitamin D receptor in muscle; neuromuscular function; deficiency → weakness + falls

So the blood/serum markers behind the muscle score are the endocrine-anabolic trio IGF-1, DHEA-S, and 25-OH vitamin D, riding alongside the physical-function tests (grip/gait/balance) and body composition (BMI). All are established skeletal-muscle-aging biomarkers in the Aging Biomarker Consortium 2024 consensus framework 3. Provenance caveat: the exact per-system assignment is not peer-reviewed-published and may differ in the Nov-2025 retrain; treat this table as a best-reconstruction, not a disclosed spec. gap/needs-replication

The interpretive consequence (important). A customer’s own IGF-1, grip, gait, and vitamin D are not plugged into their score — only their methylation is. A high Musculoskeletal age means the blood methylation pattern resembles that of people who have lower grip/gait/IGF-1/etc. It is a methylation estimate of musculoskeletal status, not a measurement of it. This is also why it can disagree with actual function — so ground-truthing a “bad” Musculoskeletal score against a real grip-strength + gait-speed test (grip-strength-biomarker) is the correct first response.

How to move it (and the honest caveat). The levers that improve every reconstructed input converge on one program — full ranked table on skeletal-muscle:

  • Resistance + power training — raises grip/gait/balance, drives local IGF-1 and the mTORC1 anabolic response, preserves Type II fibers (myofibers).
  • Protein 1.6–2.0 g/kg/day, ≥2.5–3 g leucine/meal — overcomes anabolic resistance.
  • Vitamin D repletion if insufficient — restores the receptor-dependent neuromuscular component.
  • Fat loss / body-composition — improves the BMI/adiposity input and anabolic sensitivity.
  • Sleep — supports the nocturnal growth-hormone / IGF-1 pulse.

Caveat: these reliably move the functional markers and true musculoskeletal age — but whether they move the DNAm Musculoskeletal sub-score specifically is unproven (no RCT has used an organ sub-clock as an endpoint; see § What Organ-Score Asymmetry Means Clinically). Treat a high score as a screening flag to verify with a grip/gait test, then act on the function. For the muscle-tissue methylation clock (different modality, needs a biopsy) see mskage-2025; for the individual blood methylation loci correlated with grip/gait see dnam-muscle-function-markers.

Training Architecture

The published Systems Age framework uses a six-step supervised + unsupervised pipeline 1:

  1. Biomarker grouping — clinical measures from population cohorts (HRS primary training, n=3,593) are grouped into the 11 systems by biological relevance and clinical convention.
  2. PCA within systems — principal component analysis condenses each system’s biomarkers into composite system-level principal components, capturing correlated within-system variation.
  3. DNAm surrogate training — elastic net regression (α=0.5, λ via tenfold cross-validation) maps DNA methylation patterns to the system PCs, drawing on 125,175 CpGs present in all training and validation datasets. One sparse CpG model per system is the result.
  4. System-specific mortality calibration — Cox elastic net models (trained in FHS Offspring) combine DNAm system PCs with mortality risk to score each system on a “biological years” scale relative to chronological age.
  5. Composite Systems Age — a final Cox elastic net integrates all 11 system-score PCs (plus an age-prediction score) into the composite summary index.
  6. Scaling to age range — all 11 system scores and the Systems Age composite are standardized to mean 0 / SD 1, then re-scaled to match the mean and SD of chronological age in the FHS Offspring + Third Generation combined dataset (n=3,935).

This architecture shares structural logic with GrimAge (DNAm surrogates of biological signals) but extends it from a handful of plasma proteins to a multi-system panel of clinical biomarkers. The training target is not mortality directly, but the clinical-biomarker composites that are then calibrated to mortality — hence training-target: organ-clinical-biomarker.

Validation Performance

System scores showed domain specificity in independent validation cohorts (WHI, BLSA, SATSA, ADNI) 1:

  • Heart score: strongest association with coronary heart disease (meta z-score = 8.29)
  • Musculoskeletal score: strongest association with physical function decline (z = 8.53)
  • Brain score: strongest association with cognitive function (z = 3.51)
  • Systems Age (composite): “had the strongest associations of all clocks for four conditions, was the second best for eight conditions, and was the third best for two conditions” across a wide panel of aging phenotypes

Across all conditions tested, second- and third-generation clocks outperformed first-generation clocks 4; systems-based approaches such as Systems Age are motivated by this finding even though Systems Age was not included in the Mavrommatis 2025 comparison panel. System-specific accuracy for domain-relevant outcomes is the primary performance claim of the Sehgal 2025 paper — not that Systems Age outperforms existing clocks on global mortality (where GrimAge and DunedinPACE remain benchmarks).

Clustering identifies aging subtypes

An unsupervised analysis of system-score patterns identified distinct biological aging subtypes — subgroups characterized by different combinations of accelerated and decelerated systems — each associated with unique patterns of health decline and disease risk. This result suggests that aging heterogeneity is not just quantitative (“aging faster or slower overall”) but qualitative: different people follow different multi-system trajectories. gap/needs-replication — the subtype analysis was performed in the same cohorts used for clock development; independent replication in a new prospective cohort is needed.

External application: schizophrenia

Harvanek et al. (2024 preprint) applied Systems Age clocks in a meta-analysis of schizophrenia patients (1,891 patients vs 1,881 controls across 7 cross-sectional datasets) and found that 10 of the 11 system sub-clocks showed accelerated aging, with the Heart and Lung systems showing the largest effects, followed by Metabolic and Brain 5. The paper notes that clozapine use was independently associated with increased Heart and Inflammation aging. This is an early external-application result; the paper was a preprint as of 2026-06-12. gap/needs-replication

What Organ-Score Asymmetry Means Clinically

The primary interpretive claim of Systems Age is that aging occurs at different rates across physiological systems within individuals, and that this heterogeneity contains clinically actionable information beyond what a single composite score captures. Approximately 20% of the general population show markedly accelerated aging in a single organ system relative to others (analogous to the proteomic-organ-clock finding in Oh 2023 2).

Key caveat on actionability (2026): As of 2026-06-12, no validated organ-system-clock-targeted interventions exist. An elevated Kidney score on SYMPHONY Age does not imply that interventions shown to reduce renal biomarkers (creatinine, cystatin C) will also reduce the DNAm-derived Kidney score, nor that acting on the Kidney score will alter hard renal outcomes. The organ scores are informational and prognostic — not yet prescriptive. Translating an organ-asymmetric aging profile into a modified intervention strategy remains an open empirical question. gap/no-mechanism

Relationship to Proteomic Organ Clocks (Oh 2023)

Oh et al. 2023 (Nature, Wyss-Coray lab) measured aging signatures across 11 major organs using plasma proteomics (SomaScan) in ~5,600 individuals and similarly found that organs age at different rates within individuals 2. The conceptual parallel is strong — both frameworks identify organ-specific biological age from a blood sample and reveal multi-system aging heterogeneity — but these are distinct clocks using distinct modalities:

FeatureSystems Age / SYMPHONY AgeOh 2023 organ clocks
ModalityDNA methylation (blood)Plasma proteomics (SomaScan)
Input125,175 CpGs (EPIC/450K array)~5,000 proteins (aptamer-based)
Systems11 (matching list)11 major organs
TrainingClinical biomarker PCs → elastic net DNAm surrogatesOrgan-enriched protein signatures → PCA
Commercial availabilitySYMPHONY Age (TruDiagnostic)Not directly commercialized as of 2026

Citing Oh 2023 as a basis for interpreting SYMPHONY Age scores would be a modality conflation error. The proteomic literature is useful conceptual context for why organ-specific aging scores carry biological meaning; it is not a validation of the DNAm-based implementation.

Sourcing Caveat and Validation Status

A central uncertainty on this page is the relationship between the academic Systems Age paper and the commercial SYMPHONY Age product:

  1. The Sehgal 2025 Nature Aging paper is peer-reviewed and provides the methodological foundation. DOI confirmed via Crossref and PubMed (PMID 40954326). It is not open-access as of 2026-06-12. gap/no-fulltext-access
  2. The TruDiagnostic product report (a consumer-facing artifact, not peer-reviewed) is the source for: the 133-biomarker count, the ~8,000-participant training figure, and the November 2025 retrain.
  3. The November 2025 retrain may incorporate additional proprietary data, altered biomarker assignments, or revised model architecture relative to the published paper. Without a separate methods paper, the retrained SYMPHONY Age is not independently verifiable.
  4. Quantitative specifics in the product report (11 organ systems, 133 biomarkers, ~8,000 participants) should be treated as commercial claims pending peer-review, not as validated performance benchmarks.

Residual verification gap: the published Sehgal 2025 PDF (doi:10.1038/s43587-025-00958-3) remains closed-access. Full per-system biomarker tables (Supplementary Table 1 of the published paper), exact BLSA/SATSA/ADNI cohort sizes, and any pipeline changes between preprint and published version cannot be confirmed until the full text is accessible. gap/no-fulltext-access

Limitations and Gaps

  1. Commercial-academic gap — the November 2025 retrained SYMPHONY Age is not described in any peer-reviewed publication. Product-report claims cannot be verified by external researchers. gap/unsourced for the retrain-specific figures.
  2. No intervention-response data — No published RCT has used any Systems Age sub-clock as a primary endpoint. Responsiveness to known geroprotectors (caloric restriction, rapamycin, metformin, senolytics) is unknown. intervention-responsive: partial reflects the theoretical expectation that lifestyle changes moving clinical biomarkers would propagate to DNAm surrogates, not demonstrated trial data. gap/needs-replication
  3. Blood specificity — Like all DNAm clocks trained in blood, Systems Age is not validated in other tissues. The organ-system scores are inferred from blood methylation — they are surrogate estimates of organ status, not direct tissue measurements.
  4. No Mendelian randomization — Systems Age is a composite multi-system score; GWAS instruments for the composite index are not available. mendelian-randomization: not-applicable reflects this structural feature (not a germline-instrumentable single-locus exposure).
  5. Aging-subtype replication — The clustering into distinct aging subtypes was performed within the development cohorts; independent prospective replication is needed before subtypes can be used clinically. gap/needs-replication
  6. November 2025 retrain comparability — Scores produced by the retrained SYMPHONY commercial version may not be directly comparable to scores published in the Sehgal 2025 paper or to prior SYMPHONY results, which is relevant for longitudinal tracking.
  7. Human-evidence level: limited — Despite the well-powered validation in the Sehgal 2025 paper, the commercial product has limited independent clinical validation, no RCT endpoints, and no MR evidence. The limited grading reflects the gap between academic development and clinical proof of utility.

Cross-references

  • omicmage — TruDiagnostic’s composite DNAm clock (OMICmAge); different architecture (EBPs), global not organ-specific
  • dunedinpace-2022 — pace clock; currently the only DNAm clock to show RCT responsiveness to caloric restriction
  • mskage-2025 — muscle/cartilage/bone/tendon-tissue MSK methylation clock; the tissue-level counterpart to this clock’s blood-trained Musculoskeletal sub-score
  • dnam-muscle-function-markers — blood FGF2/CXCL12/FGF21 methylation loci correlated with grip strength + gait speed (the functional measures the MSK sub-clock proxies)
  • grimage-2019 — gold standard for mortality prediction; single composite score
  • phenoage-2018 — second-generation mortality-trained clock; partly feeds the biomarker logic underlying organ-clock approaches
  • telomere-length-leukocyte — blood-based aging biomarker; different modality
  • epigenetic-alterations — hallmark context for DNAm clocks
  • dna-methylation — the molecular substrate measured
  • liver, kidney, heart, brain, lung — relevant organ-tissue pages
  • biological-age-measurement — cross-clock comparison MOC

Footnotes

Footnotes

  1. doi:10.1038/s43587-025-00958-3 · Sehgal R, Markov Y, Qin C, Meer M et al. (Morgan Levine corresponding author) · Nature Aging 2025 · 5(9):1880–1896 · PMID: 40954326 · PMC: PMC13222069 · n=3,593 primary training (HRS EPIC array; age 51–100) + FHS n=3,935 scaling/scoring (Offspring + Third Generation, 450K array); validation WHI n=5,129 preprint Table 2 (BAA23: 2,107; AS311: 855; EMPC: 2,167) — published paper may report additional validation cohorts (BLSA, SATSA, ADNI) not in the preprint; full-text not available (closed-access) · observational, longitudinal validation · 6-step pipeline: biomarker grouping → PCA within systems → elastic-net DNAm surrogates (125,175 CpGs, α=0.5, λ tenfold CV) → FHS mortality calibration → composite Cox Systems Age → age-range scaling; Heart meta z=8.29 CHD, Musculoskeletal z=8.53 physical function, Brain z=3.51 cognition · model: human adults, whole blood, Illumina EPIC/450K array · competing interests (preprint): Yale invention disclosure + provisional patent (M.E.L., R.S., A.H.C., M.M. as inventors); A.H.C. consulting fees from TruDiagnostic; R.S. consulting fees from LongevyTech.fund and Cambrian BioPharma; M.E.L. employee of Altos Labs · gap/no-fulltext-access (published version) ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  2. doi:10.1038/s41586-023-06802-1 · Oh HS et al. (Tony Wyss-Coray corresponding author) · Nature 2023 · n>5,600 individuals · observational (SomaScan plasma proteomics) · 11 organ-specific aging signatures from plasma proteins; ~20% show single-organ accelerated aging; plasma organ scores predict organ-specific disease and mortality · model: human adults; plasma proteomics (distinct modality from DNAm) · conceptual background only — not a DNAm clock ↩ ↩2 ↩3

  3. doi:10.1093/lifemedi/lnaf001 · Aging Biomarker Consortium · Life Medicine 2024;3(6):lnaf001 · PMC11851484 · consensus statement / review · consensus framework of biomarkers for skeletal-muscle aging spanning functional (grip strength, gait speed, chair-stand), imaging/body-composition (DXA, CT myosteatosis), and circulating/serum domains (including IGF-1, the GH/IGF-1 anabolic axis, and other endocrine-anabolic markers) · model: human · supports the identification of grip/gait + IGF-1 / endocrine-anabolic analytes as the canonical musculoskeletal-aging biomarker set the SYMPHONY MSK sub-clock is trained to predict · gap/no-fulltext-access (consensus framework cited at abstract/section level for the biomarker-domain taxonomy) ↩

  4. doi:10.1038/s41467-025-66106-y · Mavrommatis E, Belsky DW et al. · Nature Communications 2025 · n=18,859 (Generation Scotland) · observational, cross-clock comparison · 14 epigenetic clocks (Hannum, Horvath, Lin, Zhang10, PhenoAge, Horvath Skin+Blood, GrimAge v1, DNAmTL, Dunedin PoAm38, Dunedin PACE, GrimAge v2, YingCausAge, YingDamAge, YingAdaptAge) evaluated against 174 incident disease outcomes over 10-year follow-up; second- and third-generation clocks significantly outperform first-generation; strongest performance for respiratory and liver-based conditions; GrimAge v2 had the most Bonferroni-significant associations (37 of 174 diseases) · model: human adults, whole blood (EPIC850K array), Scotland · Note: Systems Age is NOT one of the 14 clocks in this paper — it is cited here as contextual evidence that second/third-generation clocks outperform first-generation, which motivates the systems-based approach of Systems Age ↩

  5. doi:10.1101/2024.10.28.24316295 · Harvanek ZM, Sehgal R, Borrus D, Kasamoto J et al. (Higgins-Chen corresponding author) · medRxiv preprint 2024 · n=1,891 schizophrenia patients + 1,881 controls (7 cross-sectional datasets) · observational meta-analysis (fixed-effect models) · Systems Age sub-clocks in schizophrenia: 10 of 11 system clocks showed accelerated aging; largest effects in Heart and Lung (not Metabolic/Musculoskeletal as previously stated), followed by Metabolic and Brain; clozapine associated with Heart and Inflammation aging · model: human adults with schizophrenia-spectrum disorders vs controls, whole blood · preprint — not peer-reviewed as of 2026-06-12 · competing interests: A.H.C. and R. Sehgal listed as scientific advisors to TruDiagnostic Inc. · gap/needs-replication ↩

Graph View

Backlinks