# PWWP Evolution and Discovery: analysis 1.0.0 **Analysis identifier:** PWWP-EVO-1.0.0. **Prepared:** 26 October 2026. **Scope:** independently derived, support-qualified, unrooted domain relationships for the curated Atlas. The original v1.0, v1.1 and v1.2 data, coordinates, results and identifiers remain frozen. This module does not infer chemical selectivity or evolutionary novelty. ## 1. Source sequences, membership and provenance The source is the frozen Atlas `data/proteins.csv` and `data/domains.csv`, not a newly assembled web sequence collection. All 39 canonical protein sequences were verified against newly retrieved UniProt JSON; all 39 matched exactly. Original sequence SHA-256 digests are checked again during derivation. The full canonical inputs and the exact extracted domain sequences are distributed in FASTA. Coordinates are 1-based inclusive canonical UniProt amino-acid positions. The analysis retains 39 annotation-supported domain instances: 30 human domains from 27 human proteins and 9 nonhuman domains. NSD1, NSD2 and NSD3 each contribute separate PWWP1 and PWWP2 leaves. JMJD1C, KDM3A and KDM3B remain separately displayed predicted PWWP-like family candidates. Their frozen annotation intervals and sequences are supplied separately; they are not included in this alignment, conservation estimates or phylogenetic inference. Sequence review status and family evidence are distinct: the Jasper and trypanosome UniProt entries are unreviewed TrEMBL, but their Atlas canonical sequences have experimentally deposited, exactly mapped PWWP constructs. They are retained with this explicit qualification. Per-domain provenance, canonical and extracted-sequence hashes, interval types, all original source annotations, selected experimental constructs and source-file hashes are in `domains.json` and `domain_boundaries.tsv`. `sources/Wu2011_verified_intervals.json` retains complete published sequences, unique exact canonical matches, source numbering and discrepancies. Newly retrieved UniProt JSON and retrieval receipts are in the companion `analysis/discovery_v1/sources/uniprot/` directory of the application package. ## 2. Defensible comparative intervals The frozen 25–99-residue source cores were not automatically treated as complete domains. PWWP structures include a beta-barrel, a variable insertion and C-terminal helices. Comparison intervals were resolved by evidence in the following categories, reported individually rather than conflated: 1. **Deposited domain constructs.** Twenty-four intervals use continuous canonical mappings from experimentally deposited domain constructs, generally 83–178 residues. Source mapping files, PDB and entity identifiers and any engineered substitutions are retained. The extracted sequence is always the canonical UniProt sequence, not an engineered construct or tag. Construct termini may contain flexible residues; this category does not assert that every residue in the construct is part of the fold. 2. **Published structure-assisted alignments.** For several proteins without an isolated experimental domain construct, longer sequences from Figure 1 of Wu et al. (2011) were transcribed and required to match uniquely and exactly within the frozen canonical sequence. These are supported homology intervals, not new experimental fold validations. Particularly important mappings are MBD5 1381–1494, GLYR1 4–110, HDGFL1 7–108, ZCWPW2 94–202, ZMYND8 273–370 and ZMYND11 276–374. Historical ZMYND8 and ZMYND11 numbering differs by +5 and +94 respectively. The NSD1 PWWP2 printed endpoint 1863 differs by one from the actual published sequence, which maps 1752–1862; the exact sequence controls this analysis. 3. **Curated CDD homologous spans.** The independently sourced CDD spans for NSD1 PWWP1, PWWP2A and ZCWPW1 are used as broader comparative regions and remain labeled as homology-model intervals, not experimentally certified complete folds. 4. **Explicit homolog transfers.** ARID4B and the PWWP3B/PWWP4 paralogs use a logged local BLOSUM62 alignment to the source-supported ARID4A or PWWP3A construct. Biopython PairwiseAligner uses gap-open−10 and gap-extension−0.5, with a window from 45 residues before the native core through 100 residues after it. Included transfers must align ≥95% of donor residues with ≥40% identity. The first and last aligned positions determine the interval; no arbitrary terminal padding is added. Any unaligned donor terminal residues remain recorded. These endpoints are comparative inference, not isolated-fold experiments. 5. **Subdomain within a cassette.** Mouse Zmynd11 uses a source-supported PWWP subdomain inside its deposited bromodomain–zinc-finger–PWWP construct. It is not presented as an independently folded experimental construct. This procedure provides traceable homologous comparison regions. It does not certify all domains as independently folded, and it does not overwrite the original Atlas annotation bounds. Remaining terminal and insertion uncertainty is handled explicitly by alignment masking and sensitivity analysis. ## 3. Alignment and every excluded position MAFFT 7.526 was run on the 39 source-supported intervals in fixed Atlas order: ```sh mafft --amino --localpair --maxiterate 1000 --thread 2 domain_sequences.fasta > alignment_linsi.fasta mafft --amino --genafpair --maxiterate 1000 --thread 2 domain_sequences.fasta > alignment_einsi.fasta ``` The primary alignment uses L-INS-i; E-INS-i provides a separate alignment-strategy sensitivity check. Both must preserve every input sequence exactly after gap removal. The final L-INS-i alignment has 257 columns. A deterministic, predeclared mask retains a column only when (1) at least 70% of sequences contain a canonical amino acid there and (2) at least 80% of its nongap residue pairs occupy the same corresponding column in the E-INS-i alignment. This retains 78 columns and excludes 179. The minimum row coverage in retained columns is 80.77%. The 78 retained sites include 76 parsimony-informative sites, one singleton site and one constant site according to IQ-TREE. The 80% cross-alignment criterion is an analyst-selected concordance rule, not a probability of correct alignment. The broader sensitivity alignment retains every primary column with ≥50% occupancy, giving 108 columns, without the cross-alignment concordance requirement. Neither threshold was tuned to maximize branch support. `alignment_columns.tsv` gives the occupancy, concordance, descriptive conservation and exact exclusion reason for every column. `alignment_residue_mapping.tsv` maps each nongap position to its native canonical residue. Gaps, uncertain columns, mismatches and mask membership remain visible rather than silently removed from the alignment explorer. An independent diagnostic compares the MSA with the 8,492 frozen US-align core residue correspondence pairs across 153 structural comparisons. Overall 85.06% of these pairs coincide in the primary MSA. The complete per-pair results are in `structure_alignment_diagnostic.tsv`. Structural correspondence does not by itself establish evolutionary homology, and the frozen source correspondences were never modified to improve this result. ## 4. Maximum-likelihood model and support IQ-TREE 2.4.0 ModelFinder compared 32 combinations of the general protein substitution matrices LG, WAG, JTT and VT; matrix-defined or empirical (+F) amino-acid frequencies; and equal rates, invariant sites, gamma-distributed rates with four categories, or invariant sites plus gamma. This specified set avoids treating mitochondrial or chloroplast matrices as defaults for these nuclear proteins. Bayesian information criterion (BIC) selected **LG+G4** for both primary and broader-mask analyses. Model rankings and likelihoods are retained in the IQ-TREE reports. ```sh iqtree2 -s alignment_masked.fasta -st AA -m MFP \ --mset LG,WAG,JTT,VT --mrate E,I,G,I+G \ -B 1000 -alrt 1000 -bnni -T 2 -keep-ident \ -seed 20261026 -pre trees/primary iqtree2 -s alignment_occupancy50.fasta -st AA -m MFP \ --mset LG,WAG,JTT,VT --mrate E,I,G,I+G \ -B 1000 -alrt 1000 -bnni -T 2 -keep-ident \ -seed 20261027 -pre trees/sensitivity50 ``` Each inference performs 1,000 SH-aLRT replicates and 1,000 ultrafast bootstrap replicates. `-bnni` optimizes bootstrap trees with nearest-neighbor interchange to reduce the risk of inflated support under model violation. `-keep-ident` retains source identities even when the masked sequences are identical. Newick internal labels are **SH-aLRT/UFBoot**, in that order. An edge is classified as passing the primary reporting thresholds only when SH-aLRT ≥80 and UFBoot ≥95. These values are not probabilities that a history is true. Support is conditional on the chosen alignment, substitution model and sampling. The human HDGF and rat Hdgf masked sequences are identical. Their provenance remains separate, but the retained sequences do not resolve divergence between them. IQ-TREE's uncorrected composition chi-square diagnostic flags the yeast pdp2 sequence (O94312) at p≈0.0136 for the primary mask and p≈0.0133 for the broader mask. This is a diagnostic across 39 comparisons, not by itself proof of compositional misspecification or a reason to delete a comparator. The flag and its limitations must remain visible. No outgroup, molecular clock, ancestral reconstruction or root was imposed. The IQ-TREE Newick traversal origin is an arbitrary display origin. The default rectangular and circular views show an unrooted support-collapsed cladogram: branch lengths are omitted and the spacing is topological. Only the optional full maximum-likelihood phylogram retains expected amino-acid substitutions per retained site. Neither view encodes ancestral direction or elapsed years. ## 5. Stability and comparison with previous PWWP work `tree_sensitivity.json` records shared unrooted splits and the Robinson–Foulds distance between the 78- and 108-site analyses. `branch_support.json` reports both support measures for every primary internal split, whether it is recovered in the broader-mask tree and that tree's support where available. Low-support splits remain unresolved; high primary support does not erase alignment sensitivity. The default support-collapsed tree retains only edges passing SH-aLRT ≥80 and UFBoot ≥95 in BOTH masks; all other internal edges are contracted to polytomies. Its Newick file omits branch lengths, and its JSON uses unit layout edges only. It does not redistribute discarded lengths or imply preserved evolutionary distances. The exploratory full best tree retains every original estimated length and support value. Alvarez-Venegas and Avramova (2012) used separate N-terminal/a and C-terminal/b NSD domains and a broader evolutionary sampling design, including cnidarian and choanoflagellate proteins. Their animal tree used neighbor joining and 5,000 bootstrap replicates; nodes below 50% were collapsed. The present analysis does not claim to reproduce their entire sampling or method. `published_relationship_comparison.json` explicitly tests BRPF, NSD first/second domains, the proposed broader associations with MSH6/GLYR1, ZMYND and HDGF-related groupings as unrooted splits in the available sample. Each row reports actual recovery and support rather than imposing the published topology. Non-recovery under different short-domain sampling is not automatically a contradiction of the 2012 history. All 9 preselected nonhuman comparators are assessed in `nonhuman_sampling_assessment.json`. They are useful experimental homology comparators; 4 are close rodent representatives while the remaining entries span fly, fungi and a trypanosome. This sparse structure-selected set is not taxonomically balanced and supplies no defensible universal outgroup. No extra taxa were added because this release addresses relationships within the curated resource and makes no family-origin, species-tree or duplication-timing claim. Those questions require balanced per-subfamily taxon sampling, explicit orthology/paralogy reconciliation, substantially stronger alignment/model checks and an independently justified rooting strategy. Simply adding unrelated distant sequences would not fix the limited alignment information. ## 6. Conservation, contacts and interpretation boundaries Column conservation is the descriptive score **1−H/ln(20)**, where H is Shannon entropy of unweighted observed amino-acid frequencies among nongap canonical residues. Gaps are excluded from entropy and shown separately through occupancy. Scores describe this uneven sample; they are not site-specific evolutionary rates or evidence of binding function. All mapped source residues retain canonical numbering. PWWP-associated motifs have family-specific variants, so absence of an exact PWWP tetrapeptide does not imply absence of a PWWP fold. Experimental-coordinate contacts and curated aromatic-cage annotations are independent evidence supplied by the companion residue-annotation module. Geometry, assay observations and chemical identity remain distinct. A conserved column, nearby tree tip or similar backbone does not prove identical binding pockets, histone specificity, chemical selectivity, functional interchangeability or therapeutic interchangeability. No docking, molecular dynamics, affinity prediction or new chemical activity assignment was performed. ## 7. Reproduction and software provenance Install `requirements.txt` in this analysis directory, MAFFT 7.526 and IQ-TREE 2.4.0. From the application root run: ```sh python analysis/evolution_v1/scripts/run_pipeline.py --mafft /path/to/mafft --iqtree2 /path/to/iqtree2 ``` The pipeline is offline once inputs and executables are available. It rebuilds the extracted sequences, alignment, masks, inference and JSON outputs, and verifies source sequences. `--resume` reuses IQ-TREE checkpoints only with unchanged inputs/parameters. Reproductions on other CPUs may differ in search details despite identical seeds; the supplied raw trees, full reports, logs, parameters and hashes are the authoritative results for this release. Python 3.12 and Biopython 1.88 were used. Official source archives used for the computation: - MAFFT: https://mafft.cbrc.jp/alignment/software/mafft-7.526-linux.tgz — SHA-256 `bf59d016f1b2030bc7fc83b7715ae1f0823570de3bd059c26eb522a72f9c2952`. - IQ-TREE 2: https://github.com/iqtree/iqtree2/releases/download/v2.4.0/iqtree-2.4.0-Linux-intel.tar.gz — SHA-256 `49c58e26f192dd5e0045110cee0af151203dfa0bd79fa03d01fcfc3e44ca2f1b`. The application requires no phylogenetic executable on a visitor request; it serves precomputed read-only outputs. Source scientific results and the original molecular-viewer transformations are unchanged. ## References and source attribution - Alvarez-Venegas R, Avramova Z. (2012). Evolution of the PWWP-domain encoding genes in the plant and animal lineages. *BMC Evolutionary Biology* 12:101. https://doi.org/10.1186/1471-2148-12-101. - Wu H, Zeng H, Lam R, et al. (2011). Structural and histone binding ability characterizations of human PWWP domains. *PLoS ONE* 6:e18919. https://doi.org/10.1371/journal.pone.0018919. Figure 1 retained as provenance: ©2011 Wu et al.; Creative Commons Attribution License as stated in the article. - Katoh K, Standley DM. (2013). MAFFT multiple sequence alignment software version 7: improvements in performance and usability. *Molecular Biology and Evolution* 30:772–780. https://doi.org/10.1093/molbev/mst010. - Minh BQ, Schmidt HA, Chernomor O, et al. (2020). IQ-TREE 2: new models and efficient methods for phylogenetic inference in the genomic era. *Molecular Biology and Evolution* 37:1530–1534. https://doi.org/10.1093/molbev/msaa015. - Kalyaanamoorthy S, Minh BQ, Wong TKF, von Haeseler A, Jermiin LS. (2017). ModelFinder: fast model selection for accurate phylogenetic estimates. *Nature Methods* 14:587–589. https://doi.org/10.1038/nmeth.4285. - Hoang DT, Chernomor O, von Haeseler A, Minh BQ, Vinh LS. (2018). UFBoot 2: improving the ultrafast bootstrap approximation. *Molecular Biology and Evolution* 35:518–522. https://doi.org/10.1093/molbev/msx281. - Guindon S, Dufayard JF, Lefort V, et al. (2010). New algorithms and methods to estimate maximum-likelihood phylogenies: assessing the performance of PhyML 3.0. *Systematic Biology* 59:307–321. https://doi.org/10.1093/sysbio/syq010. - Official algorithm/support documentation: https://mafft.cbrc.jp/alignment/software/algorithms/algorithms.html; https://iqtree.github.io/doc/Assessing-phylogenetic-assumptions; https://iqtree.github.io/doc/Assessing-branch-supports; https://iqtree.github.io/doc/Frequently-Asked-Questions. - UniProt sequences and annotations: UniProt, CC BY 4.0, accession-specific receipts in the companion discovery sources. CDD model identifiers and RCSB source mappings are retained per domain.