Experimental Samples
Animals from ITP Cohorts
Liver samples were collected from UM-HET3 mice of the ITP61 study, comprising animals from the 2015, 2016, and 2017 cohorts. Both female and male mice (22–23 months old) were euthanized following various interventions, including control (untreated) groups. All mice were fed Purina 5LG6 diet ad libitum and housed under similar environmental conditions. Details are provided in the original text.
Klotho-KO Mice
Eight-week-old male wild-type (C57BL/6J) and homozygous Klotho knockout (Klotho-/-) mice were obtained from CLEA Japan. Housing followed specific pathogen-free standards with a 12:12 light:dark cycle, temperature 22±2°C, humidity 50±10%, and free access to water and chow.
RNA Sequencing
Bulk RNA-seq Profiling of Mouse Tissues
- ITP cohorts: 182 liver samples (three biological replicates per sex per intervention, except canagliflozin with seven per sex; plus old and young controls).
- Klotho-KO: 24 samples from kidney and gastrocnemius muscle (six replicates per genotype per tissue).
- Sequencing: Illumina NovaSeq 6000, paired-end 150 bp.
snRNA-seq Profiling of Klotho-KO Mouse Tissues
Nuclei were extracted from brain cortex and kidney radial sections using commercial kits. Single-nucleus RNA-seq was performed using 10x Genomics Chromium Single Cell 5' Reagent Kits.
Datasets and External Resources
Transcriptomic Data for Rodent Meta-dataset
Mouse and rat gene expression data were obtained from the current study (ITP cohorts) and numerous public repositories (GEO, ArrayExpress, SRA) via listed accession numbers. See Supplementary Table 1b for complete annotation.
Transcriptomic Data for Primate Meta-dataset
Rodent data were augmented with macaque and human gene expression datasets from public repositories, including MESA (dbGaP phs001416.v3.p1). Full annotation in Supplementary Table 1c.
Bulk Gene Expression Datasets for Clock Applications
Datasets from various models (e.g., FACS-sorted cells, LPS injection, caloric restriction, chronic diseases) were used for validation. Specific accessions are provided in the text.
Single-cell/snRNA-seq Datasets
- Tabula Muris Senis (GSE149590)
- Early mouse organogenesis (E6.5–E8.5)
- Kidney and brain snRNA-seq from Klotho-KO mice (current study)
Human Cohorts
- Framingham Heart Study: Blood RNA-seq (n=3709) and DNA methylation (n=1796) with mortality and CVD data (dbGaP phs000974.v6.p5).
- UK Biobank: Plasma proteomics for GPNMB, CDKN1A, LGALS3 (Olink platform).
Data Preprocessing
Processing of Generated Bulk RNA-seq Data
Reads were mapped to GRCm39 with STAR and counted with featureCounts. Low-expression genes filtered; count matrices normalized with RLE. Outliers removed by sex-specific PCA (1.5× IQR criterion).
Processing of External Gene Expression Data
All transcriptomic data were preprocessed in R following a pipeline: soft-filtering, normalization, log-transformation, scaling. Microarray data were log-transformed and scaled. Features harmonized to mouse Entrez gene identifiers via biomaRt. YuGene normalization applied as alternative scaling method.
Integration of Multi-study Gene Expression Data
Genes present in <20% of datasets were removed. Sample quality control: Spearman correlation with tissue median profile >0.5. Final rodent meta-dataset: 3876 mouse and 663 rat samples; multi-species: additional 2623 macaque and 4003 human samples (total 11165 samples).
Relative Expression Adjustment for Batch Correction
Within each dataset-tissue combination, expression values were centred by subtracting median expression of a reference control group. Chronological age, normalized age, mortality rate, and maximum lifespan were similarly centred. For signature analysis, reference groups were matched also by sex.
Processing of Klotho-KO snRNA-seq Data
Raw data processed with Cell Ranger and Seurat. Quality filtering: mitochondrial content >10%, <200 or >4000 genes excluded. Cells annotated using marker gene panels. Nuclei aggregated into metacells (≥200,000 reads per metacell) and processed with bulk RNA-seq pipeline.
Processing of scRNA-seq Data from Tabula Muris Senis
Tissues with ≥7 mice and age range ≥15 months retained (9 tissues). Individual cells aggregated into metacells per tissue/mouse (varying cell numbers). Clock performance assessed for metacell size from 1 to 750 cells.
Processing of Early Organogenesis scRNA-seq Data
Cells filtered for >5000 UMIs, >1000 genes, mitochondrial content <2.37%. Doublets and droplets without nuclear RNA signal removed. Cells aggregated into whole-embryo metacells (minimum 200,000 reads). Lineage-specific metacells created using WOT-inferred ancestral probabilities.
Analytical Methods
Estimation of Maximum Lifespan, Expected Mortality, and Normalized Age
Maximum lifespans obtained from AnAge database. Survival curves collected from published studies. Gompertz mortality function fitted using flexsurvreg. Aggregated Gompertz parameters for experimental models estimated via mixed-effects meta-analysis. Expected hazard rate calculated for each sample. Normalized age = chronological age / expected maximum lifespan (99.9th percentile).
Transcriptomic Signatures in ITP Cohort and Aggregated Meta-dataset
Signatures of chronological age, normalized age, mortality rate, and maximum lifespan were identified using generalized linear models (edgeR for ITP) or linear mixed-effects models (lme4 for meta-dataset), with covariates as appropriate. Significance threshold: Benjamini–Hochberg adjusted P < 0.05. Signature concordance assessed by Spearman correlation of top 1000 genes.
Co-regulated Transcriptomic Modules
WGCNA applied to relative scaled rodent and multi-species meta-datasets (SoftThreshold=4, unsigned network, deepSplit=2, minClusterSize=20). Modules refined by within-module connectivity, yielding 28 rodent and 15 multi-species modules. Annotations based on top enriched pathways (MSigDB, KEGG, Reactome, ChEA).
Development of Transcriptomic Clocks
Clocks (chronological age, normalized age, expected mortality, maximum lifespan) were trained on aggregated meta-datasets using elastic net and Bayesian ridge regression. Features: gene expression deviations (for relative clocks) or scaled expression (for absolute clocks). Hyperparameters optimized via grid search with 5-fold cross-validation. Final models selected by grouped 5-fold cross-validation. Performance evaluated on held-out test sets and using leave-one-dataset/tissue/species-out cross-validation.
Application of Transcriptomic Clocks
Pre-trained clocks applied to independent datasets (including single-cell metacells). tAge estimated for each sample. Group comparisons using linear models, ANOVA, or mixed-effects models as appropriate. Gene and module contribution analyses performed by multiplying differential expression effects by clock coefficients.
Additional Analyses
- Aging dynamics across cell types (Tabula Muris Senis, FACS-sorted cells)
- Klotho-KO mouse models
- LPS neuroinflammation, caloric restriction, in vitro culturing, cellular reprogramming, heterochronic parabiosis, chronic disease models
- Association with human mortality (Framingham Heart Study) and plasma proteins (UK Biobank)
- Aging trajectory during embryogenesis (bulk microarray and scRNA-seq)
- Software development: TACO web app and tAge R package
Software and Data Availability
- TACO web app: https://app.gladyshevlab.org/TACO/
- tAge R package: Zenodo https://doi.org/10.5281/zenodo.19039290 and GitHub https://github.com/Gladyshev-Lab/tAge
- Trained clock models: Zenodo https://doi.org/10.5281/zenodo.18763485
Reporting Summary
Further information on research design is available in the Nature Portfolio Reporting Summary.
Key Takeaway: This study integrates over 11,000 transcriptomic samples across species to construct and validate biological age clocks, with applications spanning from development to chronic disease.