EPS8L2 (human, Q9H6S3) notes
2026-10-21 — Full re-review, independent of the ProtNLM2 predictions
New review (no prior review existed). ProtNLM2 output was not consulted.
Deep research
just deep-research-falcon human EPS8L2 --fallback perplexity-lite failed for both
providers: falcon returned 403 Forbidden at the Edison auth endpoint ("Failed to
authenticate"), and the perplexity-lite fallback returned HTTP 401 insufficient_quota.
No deep-research file was produced. The review is built from the primary literature
cached in publications/ (PubMed search EPS8L2[tiab], 28 hits, triaged by title).
Evolutionary grounding
- PANTHER PTHR12287 (EPS8 family) splits the four mammalian paralogs into separate
subfamilies: EPS8 = SF21, EPS8L1 = SF19, EPS8L2 = SF20, EPS8L3 = SF22
(interpro/panther/PTHR12287/PTHR12287-entries.csv). Human Q9H6S3, mouse Q99K30 and
orangutan Q5RC07 are all SF20, so mouse Eps8l2 is a same-subfamily ortholog and its
experimental findings can be transferred to human EPS8L2. Eps8 (SF21) data cannot be
transferred as subfamily-specific function (e.g. Eps8's IRSp53-dependent bundling,
MyoXVa-dependent tip targeting, stereocilia elongation).
- IBA rows come from PAINT nodes PTN000958212 (actin binding, plasma membrane, Rho
signal transduction, regulation of Rho signalling) and PTN000264860 (ruffle membrane,
positive regulation of ruffle assembly); donors are mouse Eps8 (MGI:104684), human
EPS8L1 (Q8TE68) and EPS8L2 itself (Q9H6S3). EPS8L2 appearing in its own WITH/FROM is
expected: it has its own experimental annotations from PMID:14565974.
Experimental evidence in human EPS8L2 (PMID:14565974, Offenhauser et al. 2004)
Human EPS8L2 cDNA cloned from placenta; experiments used flag-tagged human EPS8L2 in
293T cells and in mouse fibroblasts / eps8-/- MEFs.
- Co-IPs with endogenous ABI1; under triple transfection forms ABI1-dependent complex
with SOS1 PMID:14565974.
- Immunocomplexes show Rac-GEF activity (Abi1-dependent) PMID:14565974.
- C-terminal fragment binds native SOS1 in vitro.
- C-terminal fragment binds F-actin directly in co-sedimentation PMID:14565974.
- Localises to PDGF-induced F-actin-rich ruffles; rescues PDGF ruffling in eps8-/- MEFs,
as efficiently as eps8 PMID:14565974.
EPS8L2 itself is not a GEF: SOS1 is the catalytic subunit. GOA correctly uses
contributes_to for GO:0005085.
Mouse Eps8l2 (Q99K30, same SF20) — PMID:23918390 (Furness et al. 2013)
- Localised at tips of cochlear and vestibular stereocilia, mostly middle/short rows;
tip localisation independent of Eps8 and of MyoXVa.
- KO: late-onset progressive hearing loss, progressive loss of the hair bundle staircase,
thinner intermediate stereocilia, reduced cuticular plate thickness. MET currents
normal in immature hair cells. Double Eps8/Eps8l2 KO strongly enhances disorganisation.
- Mechanism not established: the paper speculates barbed-end capping or bundling, but
notes Eps8L2 lacks IRSp53 binding and its hair-cell partners were unknown.
- BAIAP2L2 binds EPS8L2 as part of the "row 2" tip complex (PMID:34346063, abstract only).
Human genetics
- DFNB106: homozygous frameshift c.1014delC (p.Ser339Alafs*15) in an Algerian family with
childhood-onset progressive ARNSHL (PMID:26282398); novel frameshift in a Pakistani
family (PMID:28281779); additional families, progressive / U-shaped audiograms, zebrafish
eps8l2 expression in inner ear and pronephros (PMID:41514136). Dog EOAD in Rhodesian
Ridgebacks with an in-frame EPS8L2 deletion (PMID:35385474).
- These support necessity of EPS8L2 for hearing; they do not show which step EPS8L2
performs. Sensory perception of sound is retained as a non-core, organism-level term
(it is an existing annotation, ISS/IEA from mouse); I did not add stereocilium
maintenance (GO:0120045) as NEW: MGI curators annotated PMID:23918390 to localisation
and sensory perception of sound but not to stereocilium maintenance/organisation, and
the KO evidence is necessity evidence. Raised as a suggested question instead.
Other reports (not used for annotation)
- Ctdnep1-Eps8L2 direct interaction and dorsal actin cables for nuclear positioning in
migrating fibroblasts (PMID:33567288; species of construct not checked in detail;
NIH3T3/U2OS). Not added as NEW.
- Colorectal cancer driver via YBX1/S6K1 (PMID:40783393) — cancer-cell overexpression
phenotype, not a normal-function annotation.
Decisions summary
- actin binding (IBA, IDA) -> MODIFY to actin filament binding (GO:0051015); co-sedimentation
with F-actin. MGI already carries GO:0051015 ISO from the same paper on mouse.
- Rho protein signal transduction / regulation thereof -> MODIFY laterally to Rac protein
signal transduction (GO:0016601) / positive regulation of Rac protein signal transduction
(GO:0035022). Verified with OLS that GO:0016601 is not a descendant of GO:0007266 (the
Rho term is now defined as Rho subfamily); the complex activates Rac.
- plasma membrane IBA -> MODIFY to ruffle membrane.
- stereocilium, stereocilium bundle -> MODIFY to stereocilium tip (part_of chain verified).
- 8x protein binding -> REMOVE (policy); ABI1 interaction is captured by GEF contributes_to.
- cadherin binding HDA -> MARK_AS_OVER_ANNOTATED.
- exosome/vesicle HDA, cytosol (HPA), cytoplasm, protein-containing complex -> KEEP_AS_NON_CORE.