AIGR Deep Research Report — PIK3R4 / VPS15 (Q99570)

Hypothesis under review

Human PIK3R4/VPS15 (Q99570) localizes to a nucleus-vacuole junction (GO:0071561).
Distinguish the defined organelle contact site from generic nuclear contact or
lysosomal trafficking, and evaluate whether the human VPS34/VPS15 complex
participates in a relevant contact-site structure, without treating a
predominant endosomal/autophagic location as proof of exclusivity.

• Focus type: function_assignment
• Term in question: GO:0071561 — nucleus-vacuole junction (cellular_component)
• Source review: genes/human/PIK3R4/PIK3R4-ai-review.yaml

Summary

The nucleus-vacuole junction annotation on human PIK3R4 (GO:0071561) is not
supported by any human experimental evidence. It is an IBA-only,
phylogenetically propagated (PAINT) annotation carried over from experimentally
annotated Saccharomyces cerevisiae orthologs into a taxon that does not possess
a vacuole and has no described nucleus-vacuole junction. Every piece of positive,
human-specific evidence places the PIK3R4-containing class III PI3K complex at
Rab7-positive late endosomes and autophagosomes, not at the nuclear envelope
or a defined nuclear–lysosomal contact site.

Three independent lines of reasoning converge on this conclusion. First, the term
is structurally organism-specific: GO:0071561 is explicitly defined as the contact
formed between the vacuole membrane and the outer nuclear membrane, mediated in
S. cerevisiae by the direct interaction of Vac8p and Nvj1p — two proteins
that build the junction and are unrelated to VPS15/VPS34. Second, the human
annotation has no experimental basis: the only GO:0071561 annotation on Q99570 is
evidence code IBA (ECO:0000318, GO_REF:0000033), inferred from the yeast ortholog
via PANTHER node PTN000426471. Third, the annotation is internally inconsistent —
only PIK3R4 carries the human NVJ term while its obligate catalytic partner
PIK3C3/VPS34 (Q8NEB9) does not.

Verdict: over-annotated / refuted for human. The most important caveat is that
absence of a curated human NVJ annotation is not absolute proof that human class
III PI3K never visits a nuclear-envelope contact site — human nuclear-lysosomal /
nuclear-envelope membrane-contact biology is an active field and remains
incompletely mapped. However, the specific GO:0071561 term denotes a defined yeast
organelle, and there is currently zero positive human evidence for it.
Curation should reflect the state of evidence, not a phylogenetic guess.

Executive Judgment

Verdict: Over-annotated / refuted for human.

Three converging lines of evidence:

1. The term is structurally organism-specific. GO:0071561 is the contact
   between the vacuole membrane and the outer nuclear membrane, built in yeast by
   Vac8p and Nvj1p. It is not architecturally a "Vps15 structure."
2. The human annotation has no experimental basis. The only GO:0071561
   annotation on Q99570 is IBA, inferred from the yeast ortholog via PANTHER
   PTN000426471. All human experimental CC annotations are
   endosomal/autophagosomal.
3. The annotation is internally inconsistent. Only PIK3R4 carries the human
   NVJ term; PIK3C3/VPS34 does not. A genuine complex-resident localization would
   annotate both subunits. The single-subunit, IBA-only pattern is the signature
   of an isolated PAINT propagation artifact.

Evidence Matrix

Key Findings

F001 — GO:0071561 is a structurally yeast-defined organelle contact site

The QuickGO record for GO:0071561 (aspect cellular_component) defines the term as
"an organelle membrane contact site formed between the vacuole membrane and the
outer nuclear membrane. In S. cerevisiae these contacts are mediated through
direct physical interaction between Vac8p and Nvj1p." The synonyms are
"nucleus-vacuole membrane contact site" and "NVJ."

The crystal structure of the Vac8p–Nvj1p interface
(PMID: 28533415) confirms that the
junction is physically built by these two proteins — "Formation of the
nucleus-vacuole junction (NVJ) is mediated by direct interaction between the
vacuolar protein Vac8p and the outer nuclear endoplasmic reticulum membrane
protein Nvj1p." Disrupting this interaction abolishes NVJs and piecemeal
microautophagy of the nucleus (PMN). The "at a glance" review
(PMID: 42375028) reinforces that the
NVJ is "a central membrane contact site in yeast that connects the nuclear
endoplasmic reticulum and the vacuole."

Implication: The junction is architecturally defined by Vac8p/Nvj1p, not by
Vps15/Vps34. Even in yeast, Vps15/Vps34 are residents at the junction, not its
structural scaffold. The term therefore carries strong organism-specific and
organelle-specific meaning that does not transfer cleanly to a vacuole-less taxon.

F002 — The human PIK3R4 NVJ annotation is IBA-only; human localization is endosomal/autophagosomal

In the QuickGO annotation set for UniProtKB:Q99570 (human PIK3R4), the only
nucleus-vacuole junction annotation is IBA (ECO:0000318, GO_REF:0000033 /
PAINT), with withFrom = PANTHER:PTN000426471 and SGD:S000000301 — i.e.,
inferred phylogenetically from the yeast ortholog, with no human experimental
support.

By contrast, every human experimental CC annotation is non-nuclear:

• Late endosome — IDA (ECO:0000314), PMID: 14617358
• Membrane — HDA, PMID:19946888
• Autophagosome and cytosol — Reactome TAS

UniProt Q99570 lists subcellular location = Late endosome; Cytoplasmic
vesicle / autophagosome; Membrane (keywords: Endosome, Cytoplasmic vesicle; no
nuclear or vacuolar terms). The primary human study
(PMID: 14617358) reports that "the
hVPS34/p150 complex colocalized with rab7 on late endosomes and hVPS34 activity
was dependent on nucleotide cycling of rab7." Here p150 is PIK3R4.

Across all organisms, GO:0071561 has ~5,093 gene-product annotations,
overwhelmingly IEA (GO_REF:0000118) and IBA (GO_REF:0000033), spanning
plants, nematodes, birds, and fungi — a classic signature of automated /
phylogenetic over-propagation of a yeast-specific structural term.

F003 — Yeast Vps15/Vps34 have genuine NVJ localization; the human paralog does not (lineage-specific architecture)

In S. cerevisiae (taxon 559292), both VPS15 (P22219) and VPS34 (P22543)
carry IDA (ECO:0000314) NVJ annotations from
PMID: 23335340, alongside the
canonical NVJ builders NVJ1 (P38881) and VAC8 (P39968) (IDA,
PMID:10888680). So the ancestral source that propagated to PANTHER node
PTN000426471 is experimentally grounded in yeast.

The crucial asymmetry: the equivalent human evidence is absent. No human
PIK3R4/VPS34 dataset places the complex at the nuclear envelope or a defined
nucleus–lysosome junction. This is exactly the pattern the seed hypothesis asked
us to distinguish — positive evidence of lineage-specific architecture versus a
yeast-only assumption based on the word "vacuole." The evidence favors the
former: the junction is a real yeast structure that simply has no demonstrated
human counterpart for this complex.

F004 — No human literature or database record places PIK3R4 at the nuclear envelope, vacuole, or a nucleus–lysosome junction

Targeted NCBI PubMed searches returned:

• PIK3R4 nuclear envelope → 0 hits
• nucleus lysosome contact site VPS34 → 0 hits
• PIK3R4 nucleus → 1 hit (PMID:33906557, an autophagy/intestinal-barrier
  review — not a localization study)

The three human PIK3R4 "localization" hits concern autophagy initiation via ULK1
(PMID:37992308), a VPS15/PIK3R4 ciliopathy affecting IFT20 release from the
cis-Golgi (PMID:27882921), and Vps34/PIK3C3 complex-subunit characterization
(PMID:27630019) — all non-nuclear, non-vacuolar. Combined with the curated
UniProt/QuickGO human locations (late endosome, autophagosome, membrane), there is
no positive human evidence for the seed hypothesis.

F005 — The human NVJ annotation is internally inconsistent (single-subunit, IBA-only)

QuickGO shows that human PIK3C3/VPS34 (Q8NEB9) has 0 annotations to
GO:0071561, whereas in yeast both Vps15 (P22219) and Vps34 (P22543) carry
experimental IDA NVJ annotations (PMID:23335340). Across all human proteins
(taxon 9606), the only GO:0071561 annotation is PIK3R4 (IBA, GO_REF:0000033);
no human protein has any experimental (IDA/IPI/IMP/HDA) NVJ annotation.

A genuine complex-resident localization would annotate both obligate subunits.
The single-subunit, IBA-only pattern is diagnostic of an isolated phylogenetic
(PAINT) propagation rather than a real human cellular structure.

Mechanistic Model / Interpretation

The class III PI3K (VPS34/PIK3C3 catalytic + VPS15/PIK3R4 regulatory pseudokinase
scaffold) is deeply conserved, and its core biochemistry (PI3P production for
endosomal sorting and autophagy) transfers across eukaryotes. What does not
transfer is the specific yeast nucleus-vacuole junction organelle, which is a
lineage-specific architecture built by Vac8p–Nvj1p.

YEAST (S. cerevisiae)                        HUMAN (H. sapiens)
---------------------                        ------------------
Nucleus (outer NE)                           Nucleus
   | Nvj1p  <-- junction scaffold            (no Nvj1p / Vac8p orthologous NVJ)
   | Vac8p  <-- junction scaffold
Vacuole membrane                             Lysosome (no defined NVJ described
   +-- Vps15/Vps34 RESIDE at NVJ (IDA,        for class III PI3K)
       PMID:23335340)
                             Vps15(PIK3R4)/Vps34(PIK3C3)
                                +-- Late endosome (Rab7+, IDA
                                    PMID:14617358)
                                +-- Autophagosome, cytosol,
                                    membrane (experimental)

PANTHER node PTN000426471 (Vps15 family)
   |  IBA / PAINT propagation
   v
Human PIK3R4 gets GO:0071561 -- WITHOUT any human experimental support,
and WITHOUT the partner PIK3C3 receiving the same term.

The annotation flow is transparent: yeast Vps15/Vps34 were experimentally seen at
the NVJ → this became the ancestral state at PANTHER node PTN000426471 →
PAINT/IBA propagated the CC term GO:0071561 down to human PIK3R4. The propagation
crossed an organelle boundary that does not exist in human (no vacuole; no
demonstrated NVJ), and it landed on only one of the two complex subunits. Both the
biology (organism-specific organelle) and the annotation metadata
(IBA-only, single-subunit, cross-taxon frequency bias) point to over-annotation.

Evidence Base

Database orientation used (labeled as database-level support): QuickGO term and
annotation records for GO:0071561, Q99570 (PIK3R4), Q8NEB9 (PIK3C3); UniProt
Q99570 subcellular location; PANTHER node PTN000426471.

GO Curation Implications

Lead (requires curator verification): Remove or NOT-qualify GO:0071561 on human
PIK3R4 (Q99570).

• The term is a CC term denoting a yeast-specific organelle contact site.
• The current human annotation is IBA-only (GO_REF:0000033, PANTHER
  PTN000426471) with no human experimental support and no partner-subunit
  corroboration (PIK3C3 lacks it).
• Recommended action: remove the human GO:0071561 annotation, or apply a
  NOT qualifier if the curator prefers to preserve the phylogenetic trail
  while signaling absence of human evidence. As PAINT annotations are generated by
  the PAINT curation team, the cleaner route may be a PAINT curation note at
  node PTN000426471 flagging that the NVJ CC term should not be propagated beyond
  fungi (organism-specific organelle), since human/metazoa lack the vacuole and a
  described NVJ.
• Retain the well-supported human CC terms: late endosome (GO:0005770,
  IDA, PMID:14617358), autophagosome (GO:0005776), cytosol, membrane,
  and phosphatidylinositol 3-kinase complex, class III (GO:0005942) / its
  autophagy-specific variants.
• The yeast NVJ annotations on Vps15/Vps34 (P22219/P22543, IDA, PMID:23335340)
  remain valid and should not be touched.

This avoids a "protein binding" fallback — the informative replacement is the
experimentally supported endosomal/autophagosomal localization already present.

Mechanistic Scope

The claim being tested is strictly a cellular-component / localization
assertion: does the gene product physically reside at a defined nucleus-vacuole
junction? This is distinct from PIK3R4's molecular function (regulatory
pseudokinase scaffold of the class III PI3K / VPS34 complex) and its biological
processes (autophagy initiation, endosomal PI3P production, vesicular
trafficking), all of which are well supported and unaffected by this review.

The refutation applies only to the direct localization at GO:0071561. It does
not challenge:

• PIK3R4's role as the VPS15 regulatory subunit of class III PI3K.
• Its established late-endosomal, autophagosomal, and membrane localization.
• Any downstream autophagy or trafficking phenotype.

We are separating direct gene-product localization (the term under review)
from pathway function (intact) and from loss-of-function phenotypes
(irrelevant to this CC assertion).

Conflicts and Alternatives

1. Paralog/ortholog carry-over (primary alternative, and the most likely
   explanation): The human term derives from yeast orthologs via PANTHER
   PTN000426471. This is not paralog confusion within human, but cross-species
   phylogenetic carry-over of an organism-specific organelle term. The
   ~5,093-annotation, IEA/IBA-dominated spread of GO:0071561 across plants,
   nematodes, and birds confirms systematic over-propagation.

2. Organism-specific organelle absence: Humans have no vacuole; the lysosome
   is the functional analog, but no nucleus-lysosome junction resembling the
   yeast NVJ has been described for the class III PI3K complex. Applying a
   vacuole-defined term to human cells is a category mismatch unless a human
   structure is demonstrated.

3. Genuine-but-undiscovered human contact site (the seed's steelman): Nuclear
   envelope–lysosome and nuclear-envelope membrane contact biology is an active
   frontier. It remains possible that human class III PI3K transiently visits a
   nuclear-envelope contact site. However, (a) there is currently no positive
   human data, and (b) even if found, it would likely warrant a new/more
   generic contact-site term rather than the yeast-specific GO:0071561. This
   keeps the door open scientifically while still supporting removal of the
   specific term now.

4. Single-subunit inconsistency as artifact signal: The fact that only PIK3R4
   (not PIK3C3) carries the term is best explained by annotation mechanics (which
   ortholog groups were PAINT-annotated), not by a biological scenario in which
   only the regulatory subunit visits the junction without its catalytic partner.

Limitations and Knowledge Gaps

Discriminating Tests

1. Proximity labeling of endogenous PIK3R4 (BioID/TurboID or APEX2) with
   nuclear-envelope, lysosomal, and endosomal reference markers. If PIK3R4
   proximity partners are dominated by Rab7/late-endosome/autophagosome proteins
   and lack nuclear-envelope/INM proteins, this refutes a human NVJ localization.

2. Two-subunit co-localization requirement: Simultaneous imaging of PIK3R4 and
   PIK3C3. A real complex-resident junction localization must show both
   subunits at the site; PIK3R4-only signal argues against it.

3. PANTHER/PAINT node audit at PTN000426471: Confirm the propagation path and
   evaluate whether GO:0071561 should be restricted to fungal descendants.

4. Ortholog conservation check for the junction scaffold: Test whether human
   has functional orthologs of Nvj1p/Vac8p forming an NVJ-like structure. Absence
   of the scaffold strengthens the "no human NVJ" conclusion.

5. Systematic GO:0071561 annotation audit across non-fungal taxa to quantify
   the over-propagation and inform a term-scope fix.

Proposed Follow-up Actions / Curation Leads (require curator verification)

Proposed action change
- Remove (or NOT-qualify) GO:0071561 (nucleus-vacuole junction) on human
PIK3R4 (Q99570); prefer fixing at PAINT node PTN000426471 so the term is
not propagated beyond fungi.
- Retain: GO:0005770 late endosome (IDA, PMID:14617358), GO:0005776
autophagosome, GO:0005942 class III PI3K complex, cytosol/membrane.
- Leave yeast Vps15/Vps34 (P22219/P22543) NVJ IDA annotations unchanged.

Candidate references with snippets to verify
- PMID: 28533415: "Formation of the
nucleus-vacuole junction (NVJ) is mediated by direct interaction between the
vacuolar protein Vac8p and the outer nuclear endoplasmic reticulum membrane
protein Nvj1p." → NVJ is a Vac8p/Nvj1p structure, not a Vps15 structure.
- PMID: 42375028: "The
nucleus-vacuole junction (NVJ) is a central membrane contact site in yeast..."
→ yeast-specific organelle.
- PMID: 14617358: "The hVPS34/p150
complex colocalized with rab7 on late endosomes..." → human localization is
endosomal.

Suggested curator questions
- Should organism-specific CC terms like GO:0071561 be blocked from IBA
propagation into taxa lacking the organelle (vacuole)?
- Is a NOT qualifier or outright removal preferred for IBA-only, single-subunit,
cross-taxon over-propagations?

Suggested experiments — see Discriminating Tests above (PIK3R4 proximity
labeling; two-subunit co-localization; PAINT node audit).

Conclusion

The seed hypothesis is refuted for human as stated. Human PIK3R4/VPS15
(Q99570) has no direct evidence of localizing to a nucleus-vacuole junction.
GO:0071561 is a structurally yeast-defined contact site built by Vac8p–Nvj1p
(PMID: 28533415); the human
annotation is IBA-only, phylogenetically propagated (PANTHER PTN000426471 /
PAINT) from experimentally annotated yeast Vps15/Vps34
(PMID: 23335340) into a taxon that
lacks a vacuole and any described NVJ. Human PIK3R4 is instead experimentally
localized to Rab7+ late endosomes and autophagosomes
(PMID: 14617358), and only PIK3R4 —
not its obligate partner PIK3C3 — carries the term, marking it an isolated
propagation artifact. Curation lead: remove or NOT-qualify GO:0071561 on human
PIK3R4 while retaining its late-endosome, autophagosome, and class III PI3K
complex CC terms; yeast NVJ annotations remain valid.

Artifacts

• OpenScientist PIK3R4 GO decision table
• OpenScientist PIK3R4 NVJ evidence matrix
• OpenScientist final report
• OpenScientist final report