In-silico structural and evolutionary characterisation of NKG7 — a four-transmembrane granule-membrane protein of cytotoxic lymphocytes (NK and CD8⁺ T cells) — using structure prediction, structural-homology search, and pore analysis on a GPU cluster.
Headline finding. NKG7 is a genuine member of the claudin / PMP-22 / Ca_V-γ (CACNG) tetraspan superfamily, and it self-associates — but it does not form a self-contained ion channel. Across three independent structure predictors and five oligomeric states, no confident assembly encloses a conducting pore; the reproducible self-interface it does form is a lateral helix-packing contact, the hallmark of a scaffold / auxiliary-modulator, not a pore. This matches its closest relatives (non-conducting Ca_V/TARP-γ auxiliary subunits and GSG1L) and its known role inhibiting the v-ATPase to restrain mTORC1.
NKG7 (Natural Killer cell Granule protein 7; UniProt Q16617 human / Q99PA5 mouse; gene ENSG00000105374, chr19q13.4; RefSeq NM_005601) is a 165-aa, four-transmembrane protein of cytotoxic-lymphocyte lysosomal/granule membranes. It has no documented channel activity, yet its loss reduces Ca²⁺ influx in CD8⁺ T cells, and its four-TM architecture resembles the Ca_V-γ (CACNG) auxiliary subunit of L-type Ca²⁺ channels — prompting the hypothesis that NKG7 might itself be an ion channel (most plausibly Ca²⁺-permeable).
This project tests that hypothesis computationally, with an explicit, pre-registered counter-hypothesis: that NKG7 is a channel-family lookalike that modulates Ca²⁺ indirectly (scaffold/auxiliary role), the way its γ-subunit relatives do.
Each tier is cheap-to-expensive and gates the next: a negative result stops the spend.
| Tier | Question | Methods | Verdict |
|---|---|---|---|
| 0 — Triage | What is NKG7? Fold, relatives, gene/transcripts, domains; does the monomer conduct? | Foldseek, BLAST/Compara/phmmer, Ensembl+GTEx, InterProScan, monomer pore scan | ESCALATE — bona-fide tetraspan; every analog non-conducting; monomer occluded (expected) |
| 1 — Oligomer | Does an oligomer enclose a pore? | Boltz-2 + AlphaFold3 + Chai-1 (C2–C6) + reference HOLE | STOP channel track — no confident oligomer; every confident model occluded |
| (pivot) — Hetero-complex | What does NKG7 dock onto? | Boltz-2 + AlphaFold3 + Chai-1 (NKG7 + ATP6AP2/ORAI1/STIM1) | INCONCLUSIVE — no confident interface; positive control (ATP6AP2) also fails the bar → binary co-folding under-powered |
Two criteria, both negative, and they reinforce each other:
- No confident oligomer. Best interface confidence anywhere is ipTM 0.63 (Boltz-2 dimer), below the ~0.8 confident bar. The three predictors disagree on both preferred stoichiometry and confidence — the signature of no well-defined assembly.
- Pore-opening anti-correlates with confidence. Every model any predictor is confident about is occluded (sub-water). The only "open" models are the lowest-confidence ones, with 60–115 Å pore axes — splayed, non-membrane-like aggregates, not a compact ~40 Å transmembrane channel. No model is both confident and pore-enclosing.
Wet-lab context (western-blot multimer bands, NKG7 self-coprecipitation) confirms NKG7 does self-associate — and the predictors corroborate a reproducible dimer interface. But that interface is a lateral TM1–TM1 helix-packing contact, exactly why the pore stays shut. Oligomerisation and pore-formation are separate claims: the first is supported, the second is not.
Since NKG7's whole family (Ca_V/TARP-γ, GSG1L) works by docking onto a partner channel and regulating it, the better-posed question after the negative pore result is what NKG7 binds. NKG7 was co-folded with three candidate partners — ATP6AP2 (a known binder, used as a positive control), ORAI1 (the CRAC channel pore), and STIM1 (its ER sensor) — across all three predictors.
No complex reached a confident interface (best cross-chain ipTM 0.59, AlphaFold3·STIM1;
confident bar ~0.8). Decisively, the ATP6AP2 positive control also failed the confidence bar
(0.10–0.33) — so the method is under-powered for NKG7's partnerships and the ORAI1/STIM1 negatives
are inconclusive, not evidence of no interaction. Full analysis:
reports/nkg7_hetero_complex_report.html. Recommended
follow-up: higher-order co-folds (NKG7 + ORAI1 hexamer) and experimental partner discovery
(co-IP / proximity-labelling mass-spec), given the independent western-blot/pulldown evidence that
NKG7 self-associates.
docs/ CAMPAIGN_PLAN.md (full self-contained plan), DESIGN_DECISIONS.md (decision log)
pipeline/ tier0/ (workstream command references), tier1/ (predictor inputs + HOLE scan script)
envs/ compute-environment specifications (marvin cluster; Boltz-2 / AF3 / Chai-1 / HOLE)
targets/ NKG7 human + mouse sequences
reports/ self-contained HTML reports (Tier 0, Tier 1) + data/ (all underlying tables/figures)
results/ 15 Tier-1 oligomer models (PDB) + cross-predictor dimer-interface analysis
Structure prediction and pore analysis ran on the marvin SLURM cluster (University of Bonn),
one A40 GPU per predictor job. See envs/README.md for environment specs and
pipeline/README.md for the per-tier commands. Tier 0 used public web
services (NCBI, Ensembl, EBI/InterProScan, GTEx, Foldseek); Tier 1 used Boltz-2, AlphaFold3 3.0.1,
and Chai-1 0.6.1, with reference HOLE (mdahole2/MDAnalysis) for the pore scan.
Heavy data (full prediction output trees, MSAs, raw Foldseek TSVs) stays on the cluster; this repository holds the scripts, the curated small results, the models scored in the reports, and the reports themselves — enough to trace and re-run every conclusion.
nkg7-binder-design— de-novo mini-binder design campaign against NKG7 (same target, complementary approach).
All results are computational. Channel function is defined electrophysiologically; any positive would require wet-lab confirmation (patch clamp / bilayer reconstitution / Ca²⁺-flux imaging). Predicted interface confidences (ipTM) are model-confidence proxies, not free energies.
Cytotoxic-lymphocyte biology · structural bioinformatics · GPU structure prediction. Part of ongoing work on NK/T-cell effector-molecule structure–function.