G4 Paradox v2 Reproducibility Package: Compartment-dependent selection regimes maintain the G-quadruplex paradox across the eukaryotic kingdoms

Reproducibility package for the v2 manuscript analysing theG-quadruplex (G4) paradox across 198 reference genomes spanning thefour eukaryotic kingdoms (Animalia n=86, Fungi n=49, Plantae n=33including the gymnosperm Ginkgo biloba, Protozoa n=30). Includes: 43 analysis scripts in total (29 main + 4 upstreamG4-detection/dG-stability/feature-compile + 10 experimental-validation for BG4 ChIP-seq, G4P-ChIP and rG4-seq overlap; Python3.9 + bash), curated sample metadata (species_features_v2.tsv,species trees, NCBI/Ensembl/WormBase/Ginkgo accessions), fullanalytical outputs (helicase counts, PGLS results on curated andOpen Tree of Life synthetic trees, polytomy-free 146-speciessensitivity, per-pair CAI control for 7 unsaturated dN/dS pairs,Nei-Gojobori dN/dS for 9 pairs with parametric 95% CI, motif agestratification, BG4/G4P/rG4-seq overlap statistics), and thecomplete Supplementary Information PDF plus six supplementarytables S1-S6 (TSV). Key findings reproducible from this deposit: (i) Cross-kingdom architecture. CDS depletion and intergenicenrichment of stable G4 motifs are strictly sign-consistent acrossall four sampled eukaryotic kingdoms under an aggregation rule thatrequires all four kingdom medians to be non-zero and share the samesign (Rule B; joint binomial P = 0.049). Intronic and promoterenrichment additionally become sign-concordant under atie-permissive rule that treats a kingdom median of exactly zeroas compatible with same-signed non-zero medians (Rule A; jointP = 2 x 10^-4), yielding a four-region cross-kingdom architecture.The tie kingdoms are Protozoa intron (15/30 species with zero stableintronic G4) and Fungi promoter. See CORRECTION_NOTES.md. (ii) Two-regime selection architecture (central synthesis).In coding sequence, codon-adaptation selection takes priority: anaive Nei-Gojobori comparison reproduces 5-16% dS suppression atG4-overlapping codons in 6/7 unsaturated species pairs, but thesignal is fully absorbed by per-gene CAI adjustment (0/7 pairsretain a G4 odds ratio significantly below 1; 5/7 reverse toOR > 1). In introns and promoters, selection acts directly on G4structure: PGLS yields five Benjamini-Hochberg-significantkingdom-specific helicase associations at q < 0.10, all at intronicor promoter G4s, including context-dependent FANCJ/BRIP1 (positivelywith intron G4 in Plantae, beta = +1.54, q = 0.057; negatively withpromoter G4 in Protozoa, beta = -2.49, q = 0.004); three of thefive remain BH-significant on the independent Open Tree of Lifebackbone. (iii) Intronic G4 are concentrated in deeply conserved genefamilies (root- or LECA-level orthologue groups in >= 87.5% ofcases across all four kingdoms), consistent with the host-genecontext of intronic G4 having been broadly maintained since earlyeukaryotic evolution. The published figures (Figure 1 G4 paradox overview, Figure 2sample composition, Figure 3 naive dN/dS, Figure 4 PGLS forest,Figure 5 motif age) and the manuscript itself are not included inthis deposit (subject to publishing journal copyright); figuresare regenerable from the deposited data using the includedplotting scripts. Companion paper: Tanigawa & Iwaki (2026) "G4-quadruplex analysisof coronaviruses reveals functional constraints and antiviraltherapeutic opportunities", Virus Research 364:199692,DOI: 10.1016/j.virusres.2026.199692. In accordance with the CSHL Press policy on AI tools, Anthropic'sClaude (model: claude-opus-4-7, 1M-context configuration; accessedvia the Claude Code CLI in agentic-coding mode between 2026-04-15and 2026-07-14) was used for script drafting, language editing,and post-submission reproducibility audit under explicitversion-controlled human review. The AI assistant did not generateresearch data, run statistical analyses autonomously, proposescientific interpretations independently, or produce figurecontent. All analytical decisions, code, computed values, figures,interpretations, and conclusions were independently verified by theauthors, who take full responsibility for the accuracy andintegrity of this deposit. ## Version 2 (2026-07-14): Post-submission corrections This version contains three post-submission corrections to theinitial deposit, disclosed to the Genome Research editorial officeon 2026-07-14. See CORRECTION_NOTES.md at the root of the depositfor the full change log. Summary: 1. Table 1 of the manuscript body: Fungi and Protozoa intron medians corrected against the deposited species_features_v2.tsv.2. Sign-consistency test: aggregation rule made explicit; two rules (strict Rule B and tie-permissive Rule A) reported.3. PIF1 KEGG identifier: Methods corrected from K11385 (which returns zero copies) to K15255 (the operative identifier used in the analysis). The central two-regime architecture finding is unchanged. ## Version 7 (2026-08-03): Interim documentation correction This version accompanies bioRxiv v3 update and applies documentation-level corrections identified in a coauthor audit (attached atdeposit root as AUDIT_REPORT.md). Central analytical claims of themain manuscript remain under clean-environment re-verification;readers are advised to consult AUDIT_REPORT.md and the manuscript'sInterim-correction notice (page 2) for the current list of openmethodology-level issues. This is an interim documentationcorrection, not a fully verified revision.