Is the Pcbd2-specific N-terminal extension a mitochondrial presequence?

All numbers below are reproduced by python3 nterm_mts_analysis.py and stored in
results.json. Nothing is hardcoded. Method, reference sets and limits:
see README.md and the module docstring.

The sequences

Mouse Pcbd1 residue 1 aligns to mouse Pcbd2 residue ~34: MAGKAHRLSAEERDQL
versus Pcbd2 MSSDAQWLTAEERDQL (Pcbd2 34–49). So the Pcbd1/Pcbd2 difference in
length is one N-terminal extension on Pcbd2, not scattered insertions.

Three independent observations say the extension is separable from the functional
core, and that the core is what has actually been characterised:

• Both Pcbd2 crystal structures cover 33/34–136 (PDB 1RU0, 4WIL) — i.e. Rose
  et al. (PMID:15182178) crystallised and assayed the protein without the
  extension, and that construct showed dehydratase activity.
• UniProt flags both source mRNAs (AAH28642.1, BAB28260.1) as
  Erroneous initiation — those clones began at the internal Met34; the
  extension was added later (sequence version 2, 2009).
• Mouse residue 34 is a Met, so Met34 is a usable alternative initiation codon.

Feature percentiles

Scored window = first 32 residues. pos = 259 reviewed mouse proteins with a
UniProt TRANSIT "Mitochondrion" feature (the presequence itself is scored);
neg = 300 reviewed mouse cytosolic (SL-0091) proteins with no transit, signal
or transmembrane feature.

On the four classical presequence features the mouse Pcbd2 extension lands
almost exactly at the median of real mitochondrial presequences: net charge
+4 (positives' median is +4), 4 arginines (median 4), zero acidic residues
(median 0), and a maximum mean hydrophobic moment of 0.423 against a positives'
median of 0.416. Against cytosolic N-termini the same values are at the 89th,
89th, 2.3rd (i.e. unusually acidic-free) and 77th percentiles.

Internal negative control. Mouse Pcbd1 — the paralog with no mitochondrial
annotation anywhere — behaves as the opposite case over the same window: net
charge −1 (0.2nd percentile of presequences; essentially no real presequence is
that acidic) and 6 acidic residues (99.8th percentile). The discrimination is
therefore not an artefact of the window or the scales.

The confounder is real and it is large. The extension is 40.6% alanine, which
is off the top of both reference distributions (99.6th and 100th percentile).
A low-complexity Ala tract is acidic-residue-free for reasons that have nothing
to do with targeting, so the single most presequence-like feature in the table —
acidic count = 0 — is substantially explained by composition bias rather than
by selection for import. What survives that caveat is the positive charge (4 Arg
are real residues, not an absence) and the amphipathicity.

Human PCBD2 is directionally the same but weaker: net charge +3, 0 acidic,
3 Arg, 25% Ala, ⟨µH⟩ 0.334 (24th percentile of presequences). The extension is
present and basic in both mammals examined.

Orthologs

Only four reviewed PCBD2 entries exist. Human (130 aa) and mouse (136 aa) carry
the extension; orangutan Q5R7K1 (117 aa) begins mid-extension (LLAALRGQ…), so
its entry is N-terminally incomplete rather than informative. Chicken Q9DG45
(103 aa) starts at MSSQSHWLTAEERTQ…, which aligns to the mouse core at residue
34 — i.e. it appears to lack the extension entirely.

That chicken entry does not establish that the extension is mammal-specific.
It is PE 2: Evidence at transcript level from a single mRNA, and mouse's own
mRNAs were flagged Erroneous initiation for exactly this failure mode — a clone
that began at the internal Met. Treat the phylogenetic distribution of the
extension as unresolved from reviewed UniProt entries alone.

A mitochondrial PCD exists elsewhere in the family

PANTHER's PAINT table for PTHR12599 (interpro/panther/PTHR12599/PTHR12599-paint.tsv)
contains an IBD for GO:0005739 mitochondrion at node PTN000972177, placed at
Embryophyta (taxon:3193, land plants) and seeded by Arabidopsis
AGI_LocusCode:AT1G29810 — UniProt Q6QJ72, whose recommended name is
"Pterin-4-alpha-carbinolamine dehydratase 2, mitochondrial". A second plant node
(PTN000972175) carries GO:0009536 plastid.

So this enzyme family does have organelle-targeted members, and a mitochondrial
PCD is not an a priori oddity. This is context, not evidence: the plant and
mammalian extensions are not homologous presequences, and the vertebrate IBD
node PTN002650414 — seeded by human PCBD1 and rat Pcbd1 — carries only
nucleoplasm, cytosol and GO:0008124, with no mitochondrial component.

Reading

The extension is compositionally compatible with a cleavable mitochondrial
presequence — basic, arginine-bearing, acidic-free and amphipathic at the
median of 259 real mouse presequences — while its paralog Pcbd1, which has no
mitochondrial annotation, is not. This supplies a mechanism that the single HDA
mitochondrion annotation (PMID:18614015, MitoCarta) otherwise lacks, and it
fits a coherent model in which residues 1–33 are cleaved on import to give the
mature 34–136 species that was crystallised and shown to be an active
dehydratase.

It does not demonstrate that this happens. Three things specifically block
that conclusion:

1. Compositional heuristics are not a targeting predictor. No MitoFates/TargetP
   run is included (none is callable from this environment without a licence or
   web upload), and the 40.6% Ala content means the features are partly driven by
   low complexity.
2. UniProt annotates no TRANSIT feature on Q9CZL5, and its SUBCELLULAR
   LOCATION records only Cytoplasm and Nucleus — the curators saw this sequence
   and did not call a presequence.
3. The best mouse-specific localisation evidence points elsewhere: Kazgan et al.
   (PMID:24389307) report DCoH2 "distributed in both the nucleus and cytoplasm"
   in Hepa1-6 cells with strong nuclear staining, consistent with its
   characterised HNF1A-coactivator role.

Corroboration for a mitochondrial pool is weaker in kind than imaging would be,
but there is more of it than the single HDA row suggests. Three further
independent mitochondrial preparations place a PCD2 there:

• Human PCBD2 is in MitoCoP. QuickGO records Q9H0N5 located_in GO:0005739
  with HTP evidence from PMID:34800366 (Morgenstern et al., Cell Metab
  2021) — a study whose whole purpose was to separate genuine residents from
  contaminants: "We classified >8,000 proteins in mitochondrial preparations of
  human cells and defined a mitochondrial high-confidence proteome of >1,100
  proteins (MitoCoP)." This is a deliberately high-confidence list from 2021, not
  a 2008 MS inventory, and it concerns the ortholog rather than mouse.
• Mouse Pcbd2 was detected in a liver mitochondrial acetylome. UniProt's
  K120/K124/K131 acetylation sites on Q9CZL5 carry
  ECO:0007744|PubMed:23576753, from a SIRT3 study of the liver mitochondrial
  lysine acetylome, and K120/K124/K131 succinylation from PMID:23806337 (SIRT5).
  SIRT3 and SIRT5 are both mitochondrial sirtuins.
• The MitoCarta HDA itself (PMID:18614015), in mouse.

These are all the same class of evidence — detection in an organellar
preparation — so they do not substitute for a localisation experiment. But four
preparations across two species, one of them explicitly contaminant-filtered,
is not the profile of a single co-purification artefact. Note that the cached
text of PMID:34800366 and PMID:23576753 does not name PCBD2; the checkable facts
are the GOA annotation and the UniProt evidence tags, both verified here.

Conclusion for curation: the HDA mitochondrion annotation is neither
confirmed nor refuted, but it is better supported than it first looks. It should
be retained as non-core rather than removed — removing it would discard a claim
that has four independent organellar preparations behind it and that this
analysis gives a plausible import mechanism for. The dehydratase and
HNF1-coactivator functions are cytosolic/nuclear and are unaffected either way.

What would settle it: N-terminal sequencing or Edman/MS of endogenous Pcbd2
from purified mitochondria to test for cleavage at ~residue 33; an
Pcbd2(1–33)-GFP fusion scored for mitochondrial import; and submitochondrial
fractionation plus protease protection to distinguish a matrix pool from surface
association.