EIF2AK3/PERK Nuclear Localization (GO:0005634) — Focused Curation Report OpenScientist openscientist-autonomous 6 citations 2 artifacts 2026-09-20T17:59:17.952843 citations file

OpenScientist final report (text/html)

OpenScientist final report (application/pdf)

EIF2AK3/PERK Nuclear Localization (GO:0005634) — Focused Curation Report

Gene: EIF2AK3 (PERK) · Human · UniProt Q9NZJ5
Hypothesis (seed): Human EIF2AK3/PERK localizes to the nucleus (GO:0005634).
Focus type: function_assignment (does the gene product directly have GO:0005634?)
Date: 2026-09-21 · Iteration 1

Executive Judgment

Verdict: Over-annotated / weakly supported (non-core at best).

The "nucleus" (GO:0005634) assignment for PERK is not supported by any direct primary
experimental localization evidence. In UniProt/QuickGO (Q9NZJ5) the single GO:0005634 term
carries the evidence code IBA / GO_REF:0000033 (GO_Central/PANTHER PTN000113601) — a
phylogenetically inferred prediction, not an experimental result. Its withFrom set is seeded by
soluble family paralogs that genuinely enter the nucleus — notably PKR/EIF2AK2 (P19525), which
carries EXP nucleus annotations (PMID:21029237, 21072047) and IDA (PMID:26705305), and
HRI/EIF2AK1 (Q9BQI3) with IDA. PERK itself (and GCN2/Q9P2K8) have IBA-only nucleus calls.
Every experimentally- or manually-supported cellular-component annotation for PERK is ER-centric,
and UniProt's manually curated Subcellular location comment is exclusively
"Endoplasmic reticulum membrane; Single-pass type I membrane protein."

The one quasi-experimental nuclear signal for PERK itself is Human Protein Atlas immunofluorescence
(Nucleoplasm/Cytosol main; Mitochondria additional). However this is single-source, only "Approved"
reliability, and — tellingly — HPA fails to detect the ER, PERK's biochemically established
compartment, which points to antibody-specificity limitations rather than a robust nuclear pool.

Two independent lines of reasoning argue the term is a paralog/family over-annotation:

Topology. PERK is an integral single-pass type-I ER-membrane protein (luminal sensor 30–514,
TM 515–535, cytoplasmic kinase domain 536–1116). A membrane-anchored protein cannot reside free
in the nucleoplasm. Any nuclear-adjacent pool would be the nuclear envelope (outer nuclear
membrane, continuous with ER → GO:0005635), with the kinase domain facing the cytoplasm — not
the nucleoplasmic interior that GO:0005634 denotes.
Family inference. The soluble eIF2α-kinase paralogs do have genuine nuclear pools
(PKR/EIF2AK2 in nucleus/nucleoli; GCN2). The GO_Central family tree therefore propagates a
"nucleus" ancestral state that fits the soluble kinases but is inappropriate for the uniquely
membrane-bound PERK.

Caveat / what would change the call: ER is continuous with the nuclear envelope, and PERK
functionally interacts with nuclear-envelope-associated factors (e.g., PARP16). If a curator wishes
to capture that, the correct term is nuclear envelope (GO:0005635) or ER, not nucleus
(GO:0005634). I found no primary paper demonstrating nucleoplasmic PERK.

Evidence Matrix

Citation	Evidence type	Supports/Refutes/Qualifies	Claim tested	Key finding	Context	Confidence & limitations
UniProt Q9NZJ5 / QuickGO (database record; retrieved 2026-09-21)	Database/annotation	Refutes (as direct evidence)	Is GO:0005634 experimentally supported for PERK?	GO:0005634 = IBA / GO_REF:0000033 (GO_Central) only; `withFrom` = PKR(P19525), HRI(Q9BQI3), GCN2 orthologs; curated CC comment = ER membrane, single-pass type I	Human, Swiss-Prot	High that basis is inferential; IBA ≠ experiment
QuickGO family check (P19525, Q9BQI3, Q9P2K8, Q9NZJ5)	Structural/evolutionary	Competing / explains IBA	Which family members drive the nucleus IBA?	PKR has EXP nucleus (PMID:21029237, 21072047) + IDA (PMID:26705305); HRI IDA; PERK & GCN2 IBA-only	Human paralogs	High; shows paralog-seeded propagation
Human Protein Atlas (ENSG00000172071; IF)	Localization	Qualifies / weak-support	Does PERK show a nuclear pool?	Nucleoplasm+Cytosol (main); reliability "Approved"; no ER detected	Human cell lines, antibody IF	Low–medium; single-source, misses ER (specificity concern)
UniProt Q9NZJ5 topology features	Structural/topology	Qualifies/Refutes	Can full-length PERK be nucleoplasmic?	Signal 1–29; lumenal 30–514; TM 515–535; cytoplasmic kinase 536–1116 → membrane-anchored	Human	High; topology well established
PMID 41406153 (2025)	Localization (paralog)	Competing / explains IBA	Do family paralogs localize to nucleus?	"a fraction of PKR maps to the nucleoli"	Human PKR/EIF2AK2	Medium; about paralog, not PERK
PMID 29352251 (2018)	Localization (paralog)	Competing / explains IBA	PKR nuclear/cytoplasmic partition	"repartition of PKR in the cytoplasm and the nucleus"	PKR/EIF2AK2	Medium; paralog
PMID 34094832 (2021)	Interaction/localization	Qualifies	PERK near nuclear envelope?	PARP16 "correlated with the nuclear envelope and the ER"; interacts with PERK/IRE1	Vascular cells	Medium; supports NE/ER, not nucleoplasm
PMID 16352659 (2006)	Mutant phenotype	Qualifies (ER role)	PERK's cellular compartment of action	PERK−/− causes "distention and fragmentation of the ER"; ER Ca²⁺ signalling defects	Mouse secretory/muscle cells	High for ER function
PMID 26268696 (2015)	Interaction/localization	Qualifies (ER role)	PERK compartment	PERK is an "ER transmembrane protein"; binds ER-membrane TMEM33	Breast cancer cells	High for ER membrane
PMID 16432136 (2006, review)	Review	Qualifies	ER–NE continuity	ER "is contiguous with the nuclear envelope"	Review	Orientation only

GO Decision Table

GO term	Aspect	Current evidence for PERK	Recommendation (lead)
GO:0005634 nucleus	CC	IBA-only (GO_REF:0000033), paralog-seeded; HPA IF nucleoplasm at "Approved", misses ER	Remove / down-weight — non-core, not directly supported
GO:0005635 nuclear envelope	CC	Not currently annotated; ER is continuous with NE; PARP16 interaction (PMID:34094832)	Optional add only if a specificity-validated primary source appears
GO:0005789 ER membrane	CC	ISS + NAS (PMID:11907036) + TAS(Reactome)	Retain (core)
GO:0005783 endoplasmic reticulum	CC	IDA (PMID:9930704), IC (PMID:11907036), TAS	Retain (core)
GO:0044233 MAM contact site	CC	IDA (PMID:39116259)	Retain
GO:0005829 cytosol	CC	TAS(Reactome) — cytosolic kinase-domain face	Retain (supportive)

Isoform / fragment check (Iteration 3)

UniProt Q9NZJ5 annotates a single isoform, one mature chain (30–1116), no released
cytosolic fragment, and a crude scan found no classical monopartite NLS (≥4 consecutive K/R).
The seed hypothesis's "relevant isoforms/fragments" route to the nucleus is therefore not
supported by curated sequence features.

GO Curation Implications (leads — require curator verification)

GO:0005634 (nucleus, CC): Recommend remove or down-weight as a stand-alone localization
claim. It is IBA-only and conflicts with PERK's curated ER-membrane topology; it most likely
reflects phylogenetic carry-over from the soluble paralogs (PKR/GCN2). Do not promote it to an
experimental CC term.
If a nuclear-adjacent location must be captured: the biologically defensible term is
GO:0005635 (nuclear envelope) or retention of the existing ER terms — but only if a curator
finds a primary localization paper; I did not.
Retain (well-supported): GO:0005789 (ER membrane), GO:0005783 (ER, IDA),
GO:0044233 (MAM contact site, IDA), GO:0005829/0005737 (cytosol/cytoplasm — the kinase domain's
cytosolic face). These represent the gene product's primary location.

Mechanistic Scope

The immediate molecular function of PERK is an ER-membrane–resident eIF2α protein kinase: its
luminal domain senses ER unfolded-protein load (via BiP release / direct binding), it
oligomerizes/autophosphorylates, and its cytosolic kinase domain phosphorylates eIF2α (Ser51) to
attenuate translation and induce ATF4. Nuclear consequences of the pathway (ATF4/CHOP
transcription, NRF2 activation) are executed by downstream effectors that translocate to the
nucleus — not by PERK itself. Attributing "nucleus" to PERK conflates the sensor with its
downstream transcriptional output.

Conflicts and Alternatives

Paralog confusion (most likely explanation): PKR/EIF2AK2 has documented nuclear/nucleolar
pools; GO_Central's family-level inference plausibly propagated a "nucleus" state to PERK.
ER–NE continuity artifact: Immunofluorescence perinuclear ER signal can be misread as
"nuclear." GO:0048471 (perinuclear region) is already annotated (IEA) and captures this better.
Topology constraint: For membrane-anchored PERK, nucleoplasmic residence is topologically
incoherent; only the nuclear envelope (ER-continuous) is physically available.

Knowledge Gaps

Is there any specific primary paper showing nucleoplasmic PERK? Checked UniProt CC, QuickGO,
and PubMed; none found. HPA IF does report Nucleoplasm but at "Approved" reliability and without
detecting the ER, so its antibody specificity is uncertain. Matters because HPA is currently the
only experimental-type signal that could upgrade GO:0005634.
Resolve: validate with a knockout-verified antibody (IF + subcellular fractionation), or
confirm HPA antibody specificity against PERK-null cells.
Does a proteolytically released PERK cytosolic fragment ever enter the nucleus? Not
established here. Resolve: look for cleavage/processing reports and fragment localization.
Exact GO_Central IBA ancestor supporting the nucleus call. Confirming it derives from PKR/GCN2
would formally document the over-annotation. Resolve: inspect the PAINT/GO_Central family tree.

Discriminating Tests

Subcellular fractionation + Western (nuclear vs. ER/microsomal vs. cytosol) for endogenous PERK.
Confocal co-localization of PERK with ER (calnexin), nuclear envelope (lamin B/emerin), and
nucleoplasm markers, ± ER stress (thapsigargin/tunicamycin).
Human Protein Atlas subcellular immunofluorescence review for EIF2AK3.
GO_Central/PAINT tree inspection to confirm the IBA ancestor is a soluble nuclear paralog.

Curation Leads (verify before acting)

Action change: Mark GO:0005634 (nucleus) as not supported by direct evidence (IBA-only);
do not treat as core localization. Prefer removal or replacement with GO:0005635 (nuclear
envelope) only if a primary source is found.
Candidate references to cite for the ER core location: PMID 26268696 (PERK an ER transmembrane
protein), PMID 16352659 (PERK ER function/phenotype).
Candidate references explaining the over-annotation: PMID 41406153, PMID 29352251 (nuclear
paralog PKR).
Suggested question for curator: "Is the GO:0005634 term meant to capture nuclear-envelope/ER
continuity? If so, GO:0005635 is more accurate; if nucleoplasm, no primary evidence exists."
Suggested experiment: endogenous PERK fractionation + confocal, ± ER stress.

Provenance (computed, live queries 2026-09-21)

UniProt REST (Q9NZJ5): curated CC = "Endoplasmic reticulum membrane; Single-pass type I
membrane protein"; topology signal 1–29, lumenal 30–514, TM 515–535, cytoplasmic 536–1116;
single isoform; single chain 30–1116; no NLS by crude K/R scan.
QuickGO annotation API (Q9NZJ5): GO:0005634 = IBA / GO_REF:0000033 (GO_Central) only;
withFrom includes PKR (P19525), HRI (Q9BQI3), GCN2 orthologs (PANTHER PTN000113601).
Family check: PKR has EXP nucleus (PMID:21029237, 21072047) + IDA (PMID:26705305);
HRI IDA; PERK & GCN2 IBA-only.
Human Protein Atlas API (ENSG00000172071): Subcellular = Nucleoplasm/Mitochondria/Cytosol;
main = Nucleoplasm/Cytosol; Reliability(IF) = "Approved"; ER not reported (specificity concern).
PubMed: no primary evidence for nucleoplasmic PERK; nuclear pools confirmed for paralog PKR.