Worked example: GRB2

Species: human

Worked example: GRB2 (P62993)

Back to the project page.

GRB2 is the textbook signaling adaptor and a good first worked example because
it shows the problem this project exists to fix in a single gene: its GOA is
dominated by uninformative protein binding, while the function that makes
GRB2 GRB2 — bridging an activated receptor tyrosine kinase to the Ras
exchange factor SOS — is carried by a handful of adaptor rows. GRB2 already has
a COMPLETE review (its review page is linked from the gene symbol GRB2); this
page reads that review through the project's
decision rules rather
than re-deciding anything.

Starting state (GOA)

The table below is the experimental-only view (the sidecar filters to
ECO:0000269 descendants); the full GRB2-goa.tsv also carries an IEA adaptor
row, so there are 7 adaptor rows, not 6 — see Rule 5.

GO term id rows evidence
protein binding GO:0005515 567 IPI (uninformative — not an adaptor term)
protein-macromolecule adaptor activity GO:0030674 4 IDA ×3 + IEA
guanyl-nucleotide exchange factor adaptor activity GO:0005091 2 IDA + IEA
transmembrane receptor protein tyrosine kinase adaptor activity GO:0005068 1 IEA only (mouse Q60631)

567 bare protein binding rows against the adaptor rows. Almost all of the 567
are IPI (ECO:0000353) interactions deposited by IntAct — one row per
partner GRB2 was seen to bind (EGFR P00533, SOS, SHC, CBL, GAB1, …). They
record what GRB2 sticks to, which is exactly the information protein binding
throws away: 567 partners of one adaptor function, not 567 functions.

The informative molecular functions are already present, with direct
experimental evidence:

Applying the decision rules

Rule 1 — bare protein binding → adaptor only when it is bridging. The
existing review removed 472 of the 567 rows under the protein-binding policy and
rewrote 92 to the specific activity each interaction actually demonstrates
(MODIFY). The replacements are instructive:

Replacement term id rows (from MODIFY)
phosphotyrosine residue binding GO:0001784 56
proline-rich region binding GO:0070064 20
GEF adaptor activity GO:0005091 7
transmembrane receptor protein tyrosine kinase adaptor activity GO:0005068 6
protein-macromolecule adaptor activity GO:0030674 4

(Replacement-term frequency; a few rows carry two replacement terms, so the
column sums slightly above the 92 MODIFY rows. The 567 GOA rows collapse to 564
distinct reference+partner pairs in the review — nothing is missing.)

The two biggest buckets are not adaptor terms but the two binding activities
that mechanistically constitute GRB2's adaptor function: the SH2 domain
reading phosphotyrosine (GO:0001784) and the SH3 domains reading proline-rich
PxxP motifs (GO:0070064). This is the key teaching point — an adaptor's
bridging is the sum of its individual recognition activities, and a reviewer
chooses the adaptor MF for the bridging as a whole and the binding MFs for the
individual contacts, rather than leaving either as protein binding.

Rule 2 — most specific descendant. GRB2 uses the specific children the
evidence licenses: GO:0005068 (adaptor for the activated RTK) and GO:0005091
(adaptor for the GEF). It also keeps the parent GO:0030674 — but, per
decision rule 5, not as a redundant general term beside its own descendants.
The review keeps it for a distinct bridging event the children do not cover:
GRB2's organisation of LAT microclusters and recruitment of PLCγ1 at the immune
synapse (PMID:25870599, directly_involved_in GO:0050852 T cell receptor
signaling). Without that distinct event, a parent sitting beside its children is
TOO_GENERAL; the condition is what makes all three legitimate.

Rule 3 — adaptor vs scaffold/sequestering/carrier. GRB2 is an adaptor, not a
scaffold: it is a small, transient, stoichiometric bridge (SH2 to one partner,
SH3 to another), not a platform that nucleates a condensate or holds many
clients. None of the scaffold/sequestering/carrier terms apply.

Rule 4 — name the bridged partners (has input). The adaptor MF is only
meaningful with its two ends named. For GRB2 the canonical pair is the
autophosphorylated receptor (or the LAT/SHC phosphoprotein) on the SH2 side and
SOS1/SOS2 on the SH3 side; these are its has input molecules.

The hard call (and the one the project exists to settle) is the MF-to-MF
relation to the SOS GEF activity in a GO-CAM. The project page's gloss makes
provides input for the default for "brings an enzyme to its substrate", which
fits GRB2 recruiting SOS to membrane Ras. The case for the stronger
directly positively regulates is specific: SOS is autoinhibited in the
cytosol, and GRB2-mediated recruitment to the membrane (plus allosteric Ras-GTP
binding) relieves that autoinhibition — i.e. the adaptor does not merely
deliver SOS to Ras, it raises SOS's exchange activity. That is a genuine
activation, so directly positively regulates is defensible here — but it is a
judgement that needs stating, not a default; a reviewer who cannot show the
activation step should use provides input for.

Rule 5 — treat propagated rows as inference. GRB2 has three IEA adaptor
rows, and they split two ways. GO:0005091 and GO:0030674 sit on top of direct
human IDA for the same terms, so they are ACCEPTed as consistent — not
independent support. GO:0005068, by contrast, has no human experimental row
at all: it is IEA-only (from mouse Q60631) yet is ACCEPTed as a core
function on converging small-scale evidence (with PMID:11533253 as an additional
reference). That is the most instructive rule-5 case in the gene — an IEA-only
term kept because the biology converges, not because an IDA already carries it.

Outcome in the review

The four core functions the review settled on are all adaptor/recognition MFs,
with protein binding gone entirely:

Across the whole review the 567 protein binding rows became 472 REMOVE + 92
MODIFY; every adaptor row was ACCEPTed as core.

Lessons for the project

  1. Scale. The payoff is concentrated — one gene turned 92 protein binding
    rows into informative MF rewrites (plus 472 removed). The core_adaptor
    worklist genes are where this concentration is highest.
  2. Adaptor MF ≠ a single binding MF. Expect to add both the adaptor term(s)
    and the specific domain-level binding terms (phosphotyrosine, proline-rich,
    etc.); they are complementary, not redundant.
  3. IntAct protein binding rows are a partner list. Read them as candidate
    has input partners of one adaptor function, not as separate functions —
    this is the fastest way to pick the right adaptor child and its inputs.
  4. Specific children are usually justified for well-studied adaptors. Do not
    stop at GO:0030674 when the evidence names both ends of the bridge.