Transcriptomic noise accumulates within tissue identity across human aging

A systemic signature of mammalian aging, distinct from cell-composition drift — analysis pipeline, intermediate data, figures, and manuscript.

Spiro T. (2026) Transcriptomic noise accumulates within tissue identity across human aging: a systemic signature distinct from cell-composition drift. Preprint, Vaika Inc. Zenodo DOI: 10.5281/zenodo.19944444.

📄 Preprint PDF: paper/pi_tissue_paper_v4.pdf 🔬 Main analysis script: src/step10_variance_conservation.py (three-level ANOVA on GTEx v8) 📦 Build PDF from scratch: src/step23_v4_pdf.py

---

Brief paper summary

We performed a three-level variance decomposition (V_total = V_tissue + V_donor + V_residual) on bulk transcriptomes from 263 GTEx v8 donors (ages 20–79) with matched samples in six tissues, combined with single-cell data from two Tabula Muris Senis platforms (mouse, ages 1–30 months), the Calico rat caloric-restriction atlas, and a rhesus macaque cross-species atlas, to ask where in variance space age-related transcriptomic change is located.

Five findings:

1. Tissue identity is preserved across forty years of human aging. π_tissue declines from 0.764 to 0.733 (Δ = −0.031); π_donor is flat (Δ < 0.005); the change is absorbed almost entirely by within-tissue, within-donor residual variance (π_residual: 0.168 → 0.194).
2. The signature is systemic noise, not selective accumulation. Aging adds within-cell-type stochastic variance, not between-population shifts — directly adjudicating between senescent-cell-accumulation and generalized-regulatory-erosion views of aging.
3. Per-tissue rates differ by an order of magnitude and do not track a simple dividing-vs-post-mitotic axis. Whole blood (Δvar = +0.079, hematopoietic) and left-ventricular myocardium (+0.121, post-mitotic but fibrosis/infiltration-driven) accumulate noise fastest.
4. Caloric restriction acts as a noise filter, not a structure restorer. In rat bone marrow, CR reverses 87% of the aging π loss (95% CI 82–91%) by reducing V_residual, not by restoring V_tissue. The mechanistic distinction predicts a specific signature for CR-mimetics that differs from partial-reprogramming-style structure restorers.
5. Cross-species scaling. Across mouse, rat, macaque, and human (~30× lifespan range), π erosion rate scales inversely with maximum lifespan (α = −1.02 ± 0.24, R² = 0.90, Spearman ρ = −1.0).

The framework is complementary to DNA-methylation clocks: where methylation clocks track chronological age with tissue-invariant CpG panels, transcriptomic residual variance exposes tissue-specific rates of regulatory reserve erosion that methylation clocks cannot see. Joint application is predicted to stratify individuals of the same biological age into functionally distinct subgroups.

---

Repository layout

pi_tissue_paper/
├── paper/
│   └── pi_tissue_paper_v4.pdf       # Final preprint PDF (open me)
├── manuscript/
│   ├── pi_tissue_paper_v4.md        # Source markdown
│   └── figures/                     # Publication figures (PNG + PDF, 300 DPI)
├── src/                             # Analysis pipeline (numbered step00–step23)
├── results/
│   ├── step10_variance_conservation/   # Main GTEx three-level ANOVA results
│   ├── step11_per_tissue/              # Per-tissue Δvar (6-tissue matched panel)
│   ├── step12_rat/                     # Calico rat CR atlas
│   ├── step15_three_tests/             # Scaling law + CR mechanism + gene leakage
│   ├── step18_vp/                      # variancePartition REML (R)
│   ├── step20_verification/            # Macaque + Kimmel independent validation
│   ├── step22_dividing/                # 15-tissue dividing vs post-mitotic
│   └── step39_sc_pi/                   # Single-cell π_tissue (TMS FACS, Droplet)
├── data/                            # NOT in git; obtain from upstream sources
├── README.md
├── LICENSE
└── .gitignore

---

Pipeline (numerical order)

Step	File	Purpose
10	step10_variance_conservation.py	Main analysis. Three-level ANOVA on GTEx (V_tissue + V_donor + V_residual).
11	step11_per_tissue_decay.py	Per-tissue Δvar on 6 matched-donor tissues.
12	step12_rat_variance.py	Calico rat CR atlas variance decomposition.
15	step15_three_tests.py	CR mechanism, scaling law, per-gene leakage.
17	step17_tier1_validation.py	Permutation nulls + validation battery.
18	step18_variancePartition.R	REML variance components (R).
20	step20_verification.py	Macaque + Kimmel independent validation.
22	step22_dividing_vs_nondividing.py	15-tissue noise rates by turnover class.
23	src/step23_v4_pdf.py	Compile manuscript markdown to PDF.

---

Reproducing the analysis

Data dependencies (not committed; obtain separately)

• GTEx v8 — dbGaP phs000424.v8 (requires approval). Files needed:
  • GTEx_Analysis_2017-06-05_v8_RNASeQCv1.1.9_gene_tpm.gct.gz
  • GTEx_Analysis_v8_Annotations_SampleAttributesDS.txt
  • GTEx_Analysis_v8_Annotations_SubjectPhenotypesDS.txt Place in data/gtex/.
• Tabula Muris Senis (FACS and Droplet h5ad) — figshare doi:10.6084/m9.figshare.12654728. Place in data/tms_facs/ and data/tms_droplet/.
• Calico rat aging atlas — GEO GSE141784. Place in results/h010_rat_cr/data/.
• Macaque cross-species atlas (Li & Kong 2025) — figshare 26963386. Place in data/macaque/extracted/.

Environment

Python 3.12 and R 4.3+. Key packages:

pip install scanpy anndata numpy pandas scipy matplotlib statsmodels fpdf2
# R:  install variancePartition + BiocParallel from Bioconductor

Run

After updating data paths in scripts as needed:

cd pi_tissue_paper
python src/step10_variance_conservation.py
python src/step11_per_tissue_decay.py
python src/step12_rat_variance.py
python src/step15_three_tests.py
python src/step17_tier1_validation.py
Rscript src/step18_variancePartition.R
python src/step20_verification.py
python src/step22_dividing_vs_nondividing.py
python src/step23_v4_pdf.py            # Build PDF

Total runtime ≈ 4 hours on a workstation; peak memory ≈ 16 GB (GTEx TPM matrix loaded).

---

Citation

Please cite the preprint and the Zenodo release together. BibTeX:

@article{spiro2026pitissue,
  author  = {Spiro, Theodor},
  title   = {Transcriptomic noise accumulates within tissue identity across
             human aging: a systemic signature distinct from cell-composition drift},
  journal = {bioRxiv},
  year    = {2026},
  doi     = {10.5281/zenodo.19944444},
  url     = {https://zenodo.org/records/19944444}
}

Please also cite the underlying data sources per their own policies (GTEx Consortium 2020; Tabula Muris Consortium / Schaum et al. 2020; Zou et al. 2022; Li & Kong 2025).

---

License

• Code (src/): MIT License — see LICENSE
• Intermediate data (results/**/*.csv): CC-BY 4.0, with upstream attribution honored
• Figures (manuscript/figures/): CC-BY 4.0
• Manuscript (paper/pi_tissue_paper_v4.*, manuscript/pi_tissue_paper_v4.md): CC-BY 4.0

---

Contact

Theodor Spiro Vaika, Inc., 1933 Sweet Rd., East Aurora, NY 14052-3016, USA tspiro@vaika.org

Acknowledgements

Andrei V. Gudkov (Roswell Park Comprehensive Cancer Center) provided scientific input on the biological framing, the dividing-vs-non-dividing architectural hypothesis, and the positioning of this work in the systemic-noise versus selective-accumulation debate. Katerina Andrianova (Vaika, Inc.) provided administrative support. All analyses were performed by the author.