Systematic multi-omics meta-analysis of public chronic pain datasets

Registration status and timing. This is a retrospective registration, not a preregistration. The study was designed, analysed and written before this record was created, and it must not be cited as a prospectively registered review. Analysis completed: 30 October 2026. Registered: [registration date]. No prospective protocol was prepared; the study began as an exploratory integration of public repositories. Study question. Chronic pain is studied across rodent models and human cohorts profiled at several molecular layers, but whether pain models, species and layers implicate the same genes had not been tested systematically. The analysis addresses three questions: (1) how much transcriptomic effect structure is attributable to pain model rather than tissue; (2) whether rodent and human transcriptomes agree beyond chance and beyond a within-species reproducibility floor; and (3) whether genome-wide association, transcriptomic and proteomic evidence converge on the same genes. Data sources and eligibility. Public repositories only: GEO (bulk and single-cell transcriptomics), the GWAS Catalog, PRIDE, MetaboLights and Metabolomics Workbench, last searched between July and August 2026. Eligibility required a case–control or comparable contrast in a pain model or human pain condition with usable deposited data; one analysis unit was retained per independent sample set, with superseded and derived units recorded in a registry. Screening identified 284 transcriptomic candidates, of which 98 were included, 93 excluded with a reason code, and 93 remained pending file or annotation checks. Screening was performed by a single reviewer against deposited metadata and files, and is reported under PRISMA 2020 with a completed checklist. Analysis as executed. Feature identifiers were harmonised to gene symbols. Differential expression used DESeq2 for RNA-sequencing counts and limma for microarrays, with log2 fold change and its standard error; proteomic and metabolomic effects are Hedges' g with its standard error. Effects were pooled by random-effects meta-analysis with the restricted maximum likelihood estimator (metafor). Benjamini–Hochberg correction was applied within the family of features measured in at least three studies, separately for the whole-cohort analysis and each stratum. Between-model heterogeneity was tested by meta-regression with pain model as moderator and a permutation null over 1,000 pain-model label shuffles within species. Tissue dependence of SNI effects was tested by meta-regression over the pre-specified family of features observed in at least two tissues and three studies. Cross-species comparison mapped rodent features to human symbols through explicit ortholog tables, and assessed directional concordance (binomial test against 50%), Spearman and Pearson correlation, and a within-species split-half control. Small-study effects were assessed with Egger's regression test at α = 0.10 for features measured in at least ten studies. Genomic analyses harmonised GWAS summary statistics, pooled them with METAL's sample-size-weighted Z scheme, and ran MAGMA gene-based analysis against a 1000 Genomes European reference. All software versions are pinned in the analysis repository. Deviations from the intended plan. The following changes are disclosed in full: (i) the GEO query was amended on 28 August 2026, after the initial analysis, to add named pain conditions; (ii) inclusion criteria for the genomic and proteomic arms were refined once the availability of usable data became clear, whereas transcriptomic criteria were fixed before analysis; (iii) the tissue meta-regression family was changed from the 200 largest absolute pooled effects to a pre-specified family (at least two tissues and three studies), removing effect-size-dependent selection; (iv) the ortholog cache was rebuilt to stop caching transient service failures as absence, which changed cross-species estimates slightly; (v) no study-level risk-of-bias tool was applied, because none exists for deposited omics contrasts — each contrast was audited against its deposited metadata instead; and (vi) screening was single-reviewer, so selection errors were not caught by a second reader. Materials and provenance. Analysis code is at commit [commit hash]; configuration, PRISMA screening records, the derived results archive and the claim registry are preserved with the project. Every numeric claim in the manuscript is bound by an automated test to the result file that produces it. The archive and code will be deposited with a DOI at publication. Ethics and data availability. No new human or animal participants were recruited and no primary data were generated. All datasets are public; ethical approvals and consent for the primary studies are reported in those publications, and no additional approval was required for this secondary analysis.