Under the Shadow: Old-biased Genes Are Subject to Weak Purifying Selection at Both the Tissue- and Cell Type-Specific Levels.
The main results reproduced: recomputed values matched the published ones within tolerance.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- ✓Same input data as the authors
- ✓Reported values were directly comparable
- ✓No relevant deviation in data/preprocessing
- ✓No authors-side cause for any deviation
- ✓Reported values are derivable from the shared data
- ✓Any deviation was negligible
- ✓The central claim held under reproduction
- ✓Overall, the reproduction was clean
- Every checked point held up.
A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.
▸Reproduction agent’s raw note
Described well enough and reproduced 1:1. Paper: 'Under the Shadow' (GBE 2025), multi-species test of the ADICT hypothesis (old-biased genes under weaker purifying selection). Reproduced the killifish dataset (GSE66712, the brief's named accession) end-to-end from the authors' own repo (NisanYildiz/selection-shadow @ 91b7948): shipped raw counts + shipped dN/dS xlsx + GEO metadata -> DESeq2 size-factor normalization -> per-gene Spearman(expr,age) -> BH adjust -> age-class assignment (|rho|>0.5, padj<0.1) -> expression-conservation-vs-age rho (Fig1C). Compared against the authors' OWN shipped derived tables (supplements gene_class CSVs), a file-level check. RESULT: skin table BYTE-IDENTICAL (same SHA256); liver table identical except 2 of 9851 BH-adjusted p-values differing only in the 15th significant digit (IEEE-754 last bit, BLAS/R minor-version drift) with all gene classes identical; all 6 class counts exact (liver 537/618/8696, skin 1468/1405/5996); Fig1C rho values reproduced (liver -0.76/-0.32, skin -0.84/-0.71), matching the paper's negative-ADICT direction. One reproducibility-critical env gotcha: the authors' scripts only run under R<4.0 (stringsAsFactors=TRUE) -- under R 4.3 they silently produce EMPTY results; faithful re-run needs that default restored. No fabrication concern: every value is fully derivable from shipped data+code. NOT attempted (optional tail): the other species (chicken GSE114129, fly Pacifico, mouse astrocyte GSE99791, naked mole-rat GSE30337), the Tabula Muris Senis single-cell cell-type figures (Fig3-7; h5ad inputs .gitignored), and the Tajima's D population-genomics tracks -- same code idiom on different shipped data, skipped to keep to a few clear data points on the named dataset.
These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.
Assessment versions
Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.
-
v1 current initial assessment Score 93assessed: 2026-06-14 ⛓ 623cc2b65a36
✎ I am an author of this paper
Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.
Provenance — full disclosure
When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.
- Reproduced
- 2026-06-14
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-15no human curator yet
- Last updated
- 2026-08-05
Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.
Deep full-text extraction
Model: opusDoes the age-related decrease in transcriptome conservation (ADICT)—weaker purifying selection on old-biased genes, predicted by Medawar's mutation accumulation theory and the selection shadow—extend across diverse nonmammalian metazoans, and does it arise cell-autonomously within specific cell types rather than merely from age-related shifts in tissue/cell type composition?
- ★ Age-related decrease in transcriptome conservation (ADICT) is commonly found in ageing tissues of nonmammalian species (chicken brain, killifish liver and skin, fruit fly brain). finding
- ★ Old-biased genes show consistently lower average sequence conservation than young-biased genes across the analyzed datasets. finding
- ★ The ADICT trend is detectable at the single-cell-type level in adult mouse tissues, supporting a cell-autonomous component rather than purely composition-driven changes. finding
- ★ Cell type-specific transcriptomes vary dramatically in average conservation levels, with neural cells highest and immune cells lowest. finding
- ★ Gene class (young- vs old-biased) is a significant predictor of conservation independent of gene age and mean expression level (linear models). finding
- An expression-conservation-age correlation approach using age-series data provides a more powerful metric than dividing genes into young/old-biased classes. method
- ★ Results support Medawar's mutation accumulation process / selection shadow as a shaper of metazoan tissue ageing. mechanism
- In long-lived naked-mole rat, old-biased genes showed weaker conservation across brain, kidney, liver, but lack of replicates precludes generalization. finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| bulk RNA-seq (expression-conservation analysis) | Gallus gallus (chicken) brain | none (ageing, 100-1825 days old) | expression-conservation correlation per individual vs age; gene-wise dN/dS conservation | — |
| bulk RNA-seq | Nothobranchius furzeri (turquoise killifish) liver and skin | none (ageing, 5-39 weeks old) | expression-conservation correlation; old- vs young-biased gene conservation | — |
| bulk RNA-seq | Drosophila melanogaster (fruit fly) brain | none (ageing, 5-40 days old) | expression-conservation correlation; old- vs young-biased gene conservation | — |
| bulk RNA-seq | Heterocephalus glaber (naked-mole rat) brain, kidney, liver | none (ageing, 4 vs 20 years old) | old- vs young-biased gene conservation | — |
| single-cell RNA-seq (44 cell types) | Mus musculus lung, skeletal muscle, brain, skin, kidney, liver (Tabula Muris Senis) | none (ageing, 3/18/24 months old) | per-cell-type transcriptome conservation (correlation of conservation scores with mean expression); cell-type-specific ADICT | — |
| bulk RNA-seq with astrocyte ribotagging | Mus musculus astrocyte-enriched cerebellum, hypothalamus, motor cortex, visual cortex | none (ageing, 4 vs 24 months old) | age-related expression changes; cell-type-specific transcriptome conservation | — |
- ▼ Whole-transcriptome analysis revealed moderate ADICT (negative expression-conservation-age correlations) in all four nonmammalian datasets, becoming more conspicuous when limited to differentially expressed genes.
- ▼ Old-biased gene sets showed consistently lower average gene-wise conservation than young-biased genes in all four datasets. Welch's t-test P < 0.003 across all four tests
- – Significant cell type effect and tissue effect on transcriptome conservation in two-way ANOVA. F_tissue=89.1 (d.f.=4), F_celltype=17.8 (d.f.=43), P < 1e-16
- ▼ Immune status had a significant negative effect on transcriptome conservation (mixed model ANOVA). F_immune status=95.79, d.f.=1, P < 0.0001
- – Gene class remained a statistically significant explanatory variable of conservation when gene age and expression were accounted for.
- ▼ In all three naked-mole rat tissues, old-biased genes showed weaker conservation than young-biased genes, but without biological replicates.
- – Neural cells, oligodendrocytes and astrocytes showed highest transcriptome conservation; immune cells (B-cells, T-cells, neutrophils, myeloid dendritic cells) on the lowest end.
- ▼ ADICT signature (cell-type-specific decrease in transcriptome conservation with age) detected within astrocyte transcriptomes of cerebellum and hypothalamus.
- pvalue P < 0.003 (Welch's t-test, lower conservation of old- vs young-biased genes across all four nonmammalian datasets)
- other F_tissue=89.1, d.f.=4; F_celltype=17.8, d.f.=43; P < 1e-16 (two-way ANOVA of transcriptome conservation across cell types and tissues)
- other F_immune status=95.79, d.f.=1, P < 0.0001 (mixed model ANOVA, negative effect of immune status on transcriptome conservation)
- count 13,989 expressed genes (combined set of expressed genes across cell types for mouse single-cell conservation analysis)
- count 10,986 whole transcriptome; 1,920 differentially expressed (G. gallus brain dataset gene counts; n=13 individuals)
- count old-b=1,066, young-b=864 (G. gallus brain old- vs young-biased gene counts)
- count old-b=792, young-b=831 (D. melanogaster brain old- vs young-biased gene counts)
- count ADICT found in 76% of 66 datasets (prior work, Turan et al. 2019) (previously reported prevalence of ADICT in mammalian bulk-tissue datasets)
Statistical methods review
Model: sonnetA neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.
The study investigated whether genes with higher expression in old versus young adults (old-biased genes) show weaker evolutionary sequence conservation than young-biased genes, using bulk-tissue and single-cell RNA-seq datasets from five species. The primary analytical framework paired per-sample Spearman correlations between gene expression levels and dN/dS-based conservation scores, then correlated those per-sample scores with individual age to detect age-related decrease in transcriptome conservation (ADICT). Comparisons between old-biased and young-biased gene sets used Welch's t-tests, and cell-type variation in conservation was assessed with two-way and mixed-model ANOVAs. Results were reported as correlation coefficients, F-statistics, and group mean differences with 95% CIs derived from 1,000 bootstraps.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Spearman's rank correlation (within-sample: expression level vs. dN/dS conservation score across genes) | Calculation of per-individual transcriptome conservation metric in all bulk-tissue and cell-type datasets | Varies by dataset; e.g., n = 10,986 genes (whole transcriptome) or n = 1,920 differentially expressed genes for G. gallus brain | not stated |
| Spearman's rank correlation (expression-conservation score vs. individual age) | Detection of ADICT across age series in each species/tissue dataset (Fig. 1b, c) | Varies by dataset; e.g., n = 13 individuals for G. gallus brain | not stated |
| Welch's two-sided t-test | Comparison of gene-wise conservation score distributions between old-biased and young-biased gene sets in each of four non-mammalian datasets (Fig. 2a) | Varies by dataset; e.g., D. melanogaster brain: n_old-biased = 792 genes, n_young-biased = 831 genes | not stated |
| Multiple linear regression (OLS) | Testing whether gene class (old- vs. young-biased) predicts dN/dS independently of gene age and mean/maximum expression level (Table S8) | null | not stated |
| Two-way ANOVA | Testing tissue and cell type effects on transcriptome conservation levels across 44 cell types in young-adult mice (Fig. 3, Table S6) | 44 cell types across 5 tissues; F_tissue = 89.1 df = 4, F_celltype = 17.8 df = 43 | not stated |
| Mixed-model ANOVA (immune status as fixed effect, tissue as random effect) | Testing effect of immune cell status on transcriptome conservation (Fig. 3) | null | not stated |
| Bootstrap resampling (1,000 iterations) for 95% confidence intervals | Mean conservation scores of old-biased and young-biased genes relative to constantly expressed genes (Fig. 2b) | Gene-level; bootstrapped within each gene class per dataset | na |
-
Conservation score distributions between old-biased and young-biased gene sets were compared with Welch's t-test↳ Could also: Mann-Whitney U (Wilcoxon rank-sum) test — dN/dS ratios are typically right-skewed and bounded at zero; a rank-based nonparametric test makes no distributional assumptions and would also be a standard choice for comparing two independent groups of gene conservation scores
-
Separate Welch's t-tests were run for each of four species/tissue datasets independently↳ Could also: A linear mixed-effects model with dataset as a random effect, testing the overall old- vs. young-biased gene class effect across all datasets simultaneously — A mixed-effects framework would test the global effect in a single model, provide an estimate of between-dataset variability, and implicitly address the multiple-testing issue that arises from running parallel tests
-
Genes were dichotomized into old-biased (rho > 0.5) and young-biased (rho < −0.5) classes using a fixed threshold, with the remaining genes treated as constantly expressed↳ Could also: Treating the continuous expression-age correlation (rho) as a predictor of gene conservation in a regression model — A continuous approach retains information lost by thresholding and avoids sensitivity to the chosen cutoff value; the paper acknowledged this by also showing robustness across alternative cutoffs (Fig. S3)
-
A two-way ANOVA was applied to test tissue and cell type effects on mean transcriptome conservation in young-adult mice↳ Could also: A linear mixed-effects model with individual mouse as a random effect nested within tissue — Because multiple cell types are measured from the same individual, a mixed model would explicitly account for within-individual correlation and the unbalanced cell-type-per-individual design, which standard ANOVA assumes away
-
Per-sample expression-conservation Spearman correlations were computed and then correlated with age as the primary ADICT metric↳ Could also: A partial correlation or regression approach that simultaneously controls for known confounders (e.g., mean expression level, gene age) within the same model — The two-step correlation approach estimates the expression-conservation-age relationship indirectly; a single model controlling for confounders at the gene level would provide a more direct test of the age effect on the conservation-expression relationship while adjusting for confounders in one step
-
95% confidence intervals for mean conservation scores were estimated via 1,000 bootstrap resamples↳ Could also: Permutation-based confidence intervals or parametric CIs (e.g., t-distribution-based) where normality holds — Both approaches are standard alternatives; permutation tests make no distributional assumptions and directly address the null hypothesis of no group difference, while parametric CIs are computationally simpler when the central limit theorem applies to the large gene sets used here
Citation network
Where this publication sits in the reproducibility-weighted citation graph — what it is built on, and what is built on it. Citation data from OpenAlex.
No assessed neighbours yet — the network grows as more papers are assessed.
Data lineage
The datasets this paper uses (text-mined from the full text via Europe PMC), and which other assessed papers stand on the same data. A shared dataset is a factual link — not a judgement.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Assessments & scoring basis
Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.
An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
Reproduction of the brief's named killifish dataset (GSE66712) is essentially perfect: the skin per-gene table is byte-identical (same SHA256), all six age-class counts match exactly (liver 537/618/8696, skin 1468/1405/5996), and the liver table matches 9849/9851 rows with the remaining two differing only in the 15th significant digit of a BH-adjusted p-value (IEEE-754 last-bit, BLAS/R version drift). The central ADICT conclusion — negative expression-conservation~age correlations — is confirmed (liver -0.76/-0.32, skin -0.84/-0.71). Every value is fully derivable from shipped data+code; no fabrication concern. Caveat (not a defect): only the killifish dataset was reproduced, not the optional other-species/single-cell tail, but that is the accession named in the brief.
Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.
Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.
🚩 Report an error in this record
Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.
Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.
Reproduction footprint
claude-opus-4-8Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.