NF041162 (family 2A encapsulin shell) — viral/prophage contamination check

Question. Aguilar et al. 2026 (Nat Commun 17:10571,
PMID:42875451, "Small viruses reveal bidirectional evolution
between HK97-fold viruses and encapsulins via procapsids") show that the
SrpI-like (family 2) encapsulin domain is shared by encapsulins and by
HK97-fold major capsid proteins of phages infecting Xanthomonas,
Janthinobacterium and Pseudomonas (their structural group 1), and that
~19% of capsid proteins from small (5–11.6 kbp) metagenomic viral genomes
carry encapsulin Pfam domains. Does the proposed
NCBIFAM:NF041162 skos:exactMatch GO:0140737 encapsulin nanocompartment
(897 net-new UniProt entries) therefore reach viral capsids?

Answer: no — NF041162 is clean. The equivalog model separates the two
lineages that the Pfam domain does not.

1. Signature level (nf041162_signature_check.py)

Live UniProtKB, InterPro and NCBI queries; full output, including the 19 viral
accessions with organism and NCBIFAM assignment, is committed as
nf041162_signature_check.yaml (first run 2026-10-17, re-run 2026-10-20 with
identical counts).

query	UniProtKB entries
xref:ncbifam-NF041162	933 (all Bacteria; 0 viral taxonomy)
xref:pfam-PF19307 (SrpI-like)	3,917
xref:pfam-PF19307 AND taxonomy_id:10239 (viruses)	19
xref:pfam-PF04454 (Linocin_M18) AND viruses	0
xref:ncbifam-NF045672	701
xref:ncbifam-NF045672 AND xref:ncbifam-NF041162	0
xref:ncbifam-NF041162 AND xref:ncbifam-NF041163	84

UniProt lists every NCBIFAM signature an entry matches, not only the best
hit: 84 entries carry both the 2A and 2B shell models. The empty
NF045672 ∩ NF041162 intersection is therefore real disjointness of the two
models' memberships, which is what the argument below rests on.

Of the 19 viral PF19307 proteins, 17 are major head/capsid proteins of
Xanthomonas, Stenotrophomonas, Pseudomonas and Janthinobacterium
phages — exactly the phage group the paper places with family II encapsulins.
NCBIFAM assigns all 17 to NF045672 (MCP_gp7_epsi_15, "major capsid
protein", equivalog, no GO term), never to NF041162. The remaining two are
Tupanvirus (giant virus) proteins with no NCBIFAM assignment.

So a Pfam/InterPro-level rule (PF19307 / IPR045641 → GO:0140737) would
mis-annotate phage capsids; the NCBIFAM equivalog does not. InterPro currently
has no GO mapping on IPR045641 or IPR049822 (InterPro API
/entry/interpro/<id>; endpoints recorded in the YAML).

2. Genomic context (nf041162_context.py)

For each NF041162 entry: EMBL protein id → NCBI coded_by CDS → ±8 kb of the
nucleotide record → neighbouring CDS products classified by regex. Per-entry
results: nf041162_context.tsv; counts: nf041162_context_summary.yaml.

• 932/933 neighbourhoods retrieved (1 entry has no EMBL protein id).
• 829/932 have a cysteine desulfurase (the family 2A cargo) within ±8 kb,
  in the canonical layout: shell gene immediately upstream of, and usually
  overlapping, the cysteine desulfurase gene, often with serine
  O-acetyltransferase. The 103 without a match are a lower bound artefact of
  product-name matching (older annotations name the cargo e.g.
  "hypothetical protein"/"aminotransferase class V"), not evidence of
  absence.
• 5 neighbourhoods contain a phage-structural gene name (portal, terminase,
  tail, baseplate, holin, ...). "holin-like" is excluded from that pattern
  because it names the bacterial LrgA/CidA regulators; the one entry it
  removed, A0ABU1IFV9, is listed below for the record. Inspected by hand:

UniProt	organism	phage gene(s) in window	verdict
A0A285D2J8	Cereibacter ovatus	portal, terminase GpA, prohead protease 1–7 kb upstream	shell + cysteine-desulfurase operon intact, adjacent to a prophage remnant
A0A4Q9VKP0	Siculibacillus lacustris	terminase ~4 kb away	operon intact; not phage
A0ABU1IFV9	Paracidovorax wautersii	"holin-like protein" + TIGR00659 murein hydrolase	LrgA/LrgB pair, not phage; now excluded by the pattern
A0ACG4YM36	Cupriavidus necator	tail fibre ~7 kb away	operon intact (shell + annotated encapsulin cargo); not phage
G0ABQ1	Collimonas fungivorans	tail sheath FI ~6 kb away	operon intact; not phage
H6SQ81	Pararhodospirillum photometricum	"portal protein-like" ~7 kb away, transposase, plasmid stability protein	only ambiguous case: no adjacent cargo, mobile-element-rich region; 305 aa, typical shell length

• 15 neighbourhoods contain an integrase/recombinase and 3 contain the word
  "phage" (integrase_neighbour_entries / phage_word_neighbour_entries in
  the summary). The phage-word hits are A0A285D2J8 and G0ABQ1 (above) and
  B1MEQ4. 14 of the 15 integrase neighbourhoods also carry a named
  cysteine-desulfurase cargo; these integrases are the usual genomic-island
  and XerD-type recombinases and say nothing against encapsulin identity.
  Two entries are worth naming:

UniProt	organism	mobile-element gene(s) in window	verdict
B1MEQ4	Mycobacteroides abscessus ATCC 19977	"Putative phage integrase" ~3 kb downstream	shell + "Probable cysteine desulfurase" operon intact; adjacent to an island, not inside a phage
A0A6M9Q0Z8	Polynucleobacter tropicus	integrase + AlpA-family regulator 6–8 kb upstream	no named cargo, but the shell is flanked by an unnamed ~545-aa ORF (cysteine-desulfurase sized), serine acetyltransferase and cysteine synthase A: the sulfur-metabolism layout of the other members, with a genomic island next door

No NF041162 member is itself a phage gene. Two entries
(mobile_element_neighbour_without_cargo: H6SQ81, A0A6M9Q0Z8) sit next to a
mobile element and lack a named cargo in the window; of these, only H6SQ81
has no sulfur-metabolism context at all. This is separate from the 103 entries
without a cargo name match; like A0A6M9Q0Z8, these may carry an unnamed
cargo ORF, but they were not checked individually.

3. Side observations

• 84 entries carry both NF041162 (2A) and NF041163 (2B); 71 of them have the
  cNMP-binding insertion (PF00027) typical of family 2B shells. Both models map
  to the same CC term, so this does not affect GO:0140737, but NF041162 is not
  strictly 2A-specific.
• NF016349 (Linocin_M18 PfamEq) carries GO:0008233 peptidase activity and
  GO:0042742 defense response to bacterium in NCBI's hmm_PGAP.tsv (release hmm_PGAP/20.0) — a
  holdover from the original misidentification of Linocin M18 as a bacteriocin.
  It must not be adopted into ncbifam2go.

Conclusion

The NF041162 → GO:0140737 exactMatch stands. The paper's virus–encapsulin
mixing is real at the Pfam-domain level, and NCBIFAM already resolves it with a
separate phage capsid equivalog (NF045672). The caution to record is about
domain-level rules: any future PF19307/IPR045641 (or Linocin_M18) → GO:0140737
mapping would sweep in phage major capsid proteins and should be rejected.

Limitations: product-name regexes on heterogeneous GenBank annotations;
±8 kb window; at most the first two EMBL protein ids tried per entry; only UniProtKB members (metagenome-only viral contigs of the kind
the paper studies are not in UniProtKB and were not tested).