# Scientific results and limits — PWWP-EVO-1.0.0

The data support a **partly resolved, unrooted domain relationship analysis**. They do not support a complete evolutionary history of the PWWP family. The default display therefore retains supported local relationships and leaves the remaining branching unresolved.

## Data and model

All 39 Atlas canonical sequences exactly matched the retrieved UniProt records. The analysis contains 39 annotation-supported domain instances: 30 human domains from 27 proteins, with the two NSD domains treated separately, and nine nonhuman experimental comparators. Three predicted PWWP-like candidates remain separate.

Source cores were extended or replaced with traceable domain constructs, published homology intervals, CDD spans or explicitly documented homolog transfers. The resulting intervals are 83–178 residues. They are not uniformly experimentally certified isolated folds. Two significant historical numbering shifts were corrected by unique exact full-sequence matching: ZMYND8 +5 and ZMYND11 +94. A one-residue printed-endpoint discrepancy in the published NSD1 PWWP2 sequence is retained in the provenance.

MAFFT L-INS-i produced 257 columns. The deterministic occupancy/concordance mask retained 78 and excluded 179; all excluded positions remain available. A broader mask retained 108 columns. Both analyses selected LG+G4 under BIC from the specified 32-model search. Each used 1,000 SH-aLRT and 1,000 UFBoot replicates with bootstrap NNI refinement.

## What is resolved

The full maximum-likelihood topologies share all 36 unrooted internal splits (Robinson–Foulds distance 0), despite differences in support. **13 of 36** primary branches meet SH-aLRT ≥80 and UFBoot ≥95. **12** meet both thresholds in both masks. The default cladogram contracts the other **24** branches to polytomies and omits branch lengths. Agreement of the two masks is a sensitivity check, not independent replication.

| Published relationship tested | Primary SH-aLRT / UFBoot | Broader-mask support | Assessment |
|---|---:|---:|---|
| BRPF1 / BRD1(BRPF2) / BRPF3 | 99.7 / 100 | 99.9 / 100 | Supported local grouping |
| Three NSD PWWP1 domains | 97.1 / 99 | 98.3 / 99 | Supported local grouping |
| Three NSD PWWP2 domains | 99.8 / 100 | 100 / 100 | Supported local grouping, distinct from first domains |
| ZMYND subgroup including mouse comparator | 99.5 / 100 | 98.8 / 99 | Supported local grouping |
| HDGF-related subgroup including stated rodent comparators | 99.3 / 97 | 99.1 / 100 | Supported local grouping |
| NSD PWWP1 domains plus MSH6 | 77.7 / 52 | 84 / 47 | Unresolved |
| NSD PWWP2 domains plus GLYR1 | Exact split not recovered | Exact split not recovered | No support for that exact sampled split in this analysis |

These comparisons concern explicit full-sample bipartitions. The sampling differs substantially from Alvarez-Venegas and Avramova (2012); non-recovery is not proof that a broader published evolutionary interpretation is wrong. The related NSD first/second-domain and other subgroup relationships are consistency checks, not newly claimed evolutionary discoveries.

## Remaining limitations

1. The retained alignment is short and the dataset is human centered. Twenty-three primary internal branches fail at least one support threshold. Deep relationships, ancestral order, duplication timing and family origin remain unresolved.
2. There is no justified outgroup. The traversal origin used to draw the unrooted tree has no ancestral meaning. No branch length is a calendar date.
3. The nine nonhuman comparators are sparse and selected for available experimental structures. They do not provide balanced taxon coverage or one-to-one orthology assignments. No extra taxa were added for claims this release does not make.
4. Human HDGF and rat Hdgf have identical retained sequences. Their separate labels preserve provenance; the alignment cannot resolve their divergence or internal ordering.
5. The pdp2 comparator has an uncorrected IQ-TREE composition diagnostic p≈0.0136 (broader mask p≈0.0133), among 39 such tests. This diagnostic is recorded without interpreting it as proof of model failure or silently excluding the sequence.
6. Overall MSA agreement with the frozen structural correspondence is 85.06% across 8,492 pairs. Among the 7,575 source correspondence pairs for which both residues remain in retained columns, agreement is 93.81%. Structural equivalence is a diagnostic, not proof of homology or binding equivalence.
7. Entropy-based conservation describes this uneven sample. Neither conservation nor evolutionary proximity establishes pocket equivalence, histone specificity, affinity, chemical selectivity or therapeutic interchangeability.
8. The current homology intervals still need independent structural boundary work for proteins without an isolated experimental fold. Family-wide evolutionary claims require balanced per-subfamily sampling and more extensive model/alignment adequacy analysis.

## Manuscript assessment

The work supports an additional **resource-methods figure and analytical section**: a support-collapsed domain relationship panel linked to the canonical alignment, source boundaries, selected domain architectures and experimental-structure availability. The section can report the independently versioned derivation, the recovered established subgroup relationships, and the explicitly unresolved portion of the family.

It does **not** justify a claim of newly discovered PWWP evolutionary origins, a resolved family chronology, conserved ligand pharmacology or transfer of target selectivity between neighboring tree tips. A figure should identify the conservative tree as an unrooted cladogram and use the full maximum-likelihood phylogram only when its support qualifications and branch-length units are shown.

Sources: Wu et al. (2011), DOI 10.1371/journal.pone.0018919; Alvarez-Venegas and Avramova (2012), DOI 10.1186/1471-2148-12-101. Exact methods and software citations are in METHODS.md; raw model rankings and support output are retained in trees/.