Rare Disease, Real Kid · computational hypotheses, not a treatment

The fold did not come back. We still have a screen: reverse the BUB1B-knockdown nucleus with FDA drugs.

A faculty–student team at Taipei Medical University asked whether a small molecule could rescue BubR1 p.Asn1002Lys. DynamicBind on an FDA library never pulled residues 1002–1012 toward wild type. We then docked 2,427,793 blood–brain-barrier-filtered ligands on the new lysine, and we built a HeLa BUB1B shRNA imaging assay. If this entry is awarded, the prize funds a phenotype-first FDA screen with the Academia Sinica HTS core — hits are defined by nuclear rescue, not by binding BubR1.

Gene: BUB1B / BubR1 Allele: p.N1002K FDA DynamicBind: 3,619 compounds · no fold rescue CNS docks: 2,427,793 KD nuclei: −51%

3-minute English pitch (voice + burned captions). Sidecar subtitles: pitch_en.srt. Upload this file to YouTube or Vimeo for the Track 2 form.

Who we are

Taipei Medical University — College of Interdisciplinary Studies and College of Pharmacy. Faculty and students working as one Track 2 team.

Code: github.com/hunglinchen2003/BUB1B-N1002K-screening
This site: hunglinchen2003.github.io/bub1b-n1002k

What we actually did, in order

Read the allele as a stability problem. Kinase-domain missense MVA alleles (same helix as L1012P) destabilize BubR1 and accelerate proteasomal turnover. N1002K also creates a solvent-exposed primary amine that wild-type Asn cannot present to an E3 ligase.

Try to reverse the fold with FDA drugs. 3,619 approved/pharmacopeial compounds through DynamicBind. Best WT-similarity scores (~47) stayed below the apo-mutant baseline (49.9). Ligands drifted 25–32 Å from K1002. Structure reversal failed.

Re-score the same FDA poses as a lysine cover. Hydrogen-bond / ion contact to K1002 NZ. Survivors: bicalutamide, hederagenin, glimepiride, infigratinib, cynarin. These remain the approved-drug shortlist.

Open a larger CNS catalog on 1002K only. Wager BBB filter. Uni-Dock / Vina, 22 Å box on mutant NZ. 2,427,793 successful docks → 10 computational hits. Rank 10 is notoamide B, a fungal natural product with anticancer literature — not FDA-approved.

Build the cellular assay we would actually run. HeLa BUB1B shRNA, Hoechst 33342. Nuclear count −51%, micronuclei 1.1% → 4.7%, pyknosis 3.4% → 9.3%. That morphology is the HTS readout.

If awarded: phenotype-first FDA HTS. Academia Sinica HTS core. Screen approved drugs for reversal of the KD nucleus. A hit need not bind BubR1 or cover K1002.

Mechanism

BubR1 is a spindle-assembly checkpoint protein. With Mad2, Bub3 and Cdc20 it forms the mitotic checkpoint complex and holds APC/C. This child is compound-heterozygous: a truncating allele (copy loss) plus p.N1002K at the N-cap of helix αG in the kinase C-lobe — ten residues from the known MVA hotspot L1012P.

Asn → Lys

Neutral N-cap residue replaced by a +1 amine. Volume +48%. Contacts to W973 / W978 / S976 are lost.

~4–5 Å

After aligning the high-pLDDT core (956–995), N/K1002 Cα moves ~4.1 Å and T1008 ~5.0 Å.

82 vs 27 Å²

Lys SASA at 1002. The new NZ is solvent-exposed. Asn cannot be ubiquitylated there; Lys can.

The drug class is not an ATP-site inhibitor, not an HSP90 inhibitor, and not a PLK1 inhibitor. Those would further starve BubR1. The computational ideas were a pharmacological chaperone (failed) and a steric cover of 1002K. The experimental idea is broader: reverse the SAC-failure nucleus, by any mechanism.

AlphaFold N1002K kinase

Mutant kinase only, UniProt 700–1050. Drag to rotate. Brick cartoon is N1002K; orange sticks and the label mark K1002.

Figure. Local AlphaFold kinase model of the N1002K mutant. Orange sticks = Lys1002. The WT overlay is omitted so the mutated lysine is easy to find.

Aligned on 956–995

Residue	Cα shift (Å)	pLDDT WT / MUT
K995	2.44	96 / 96
N/K1002	4.07	96 / 92
T1008	5.01	98 / 97
L1012	4.39	99 / 99
K795 (ATP)	1.03	94 / 94

What that means

The kinase core is not exploded. αG is displaced enough that a ligand sitting on NZ is a different pose than a ligand sitting on wild-type Asn. Full-length AF3 Cα RMSD of 33.5 Å is dominated by low-pLDDT IDRs and should not be read as a crystal-quality conformational change.

Stage A — FDA library: structure reversal failed

Library: MedChemExpress HY-L066 (FDA-approved and pharmacopeial). DynamicBind job of 3,619 compounds on mutant vs wild-type kinase. Apo mutant vs WT similarity baseline: 49.9 / 100.

47.4

Best ligand WT-similarity (exendin-4). Still below apo. Every Δ is negative.

25–32 Å

DynamicBind poses of pocket hits from K1002. The ligand did not stay on the lysine.

NHS esters or sulfonyl fluorides in the library. A covalent Lys cover was geometrically unavailable.

A CPU Vina pass on the same box (22 Å, center on mutant NZ) scored 2,432 unique ligands. Pocket occupancy plus mutant-vs-WT gap put aprepitant first (−7.66 vs −5.73, 3.03 Å). That ranking did not survive the next gate.

Stage A — same poses, now as a lysine cover

After fold rescue failed, we re-scored existing mutant / wild-type Vina poses: acceptor ≤ 3.4 Å from NZ; carboxylate / sulfonate ≤ 4.0 Å; mutant-specific versus Asn ND2.

#	Compound	Role	MUT Vina	Contact
1	Bicalutamide	approved anti-androgen	−6.36	sulfonate 3.03 Å, geometric H-bond, mutant-selective
2	Hederagenin	pharmacopeial triterpene	−6.75	carboxylate 2.94 Å
3	Glimepiride	approved sulfonylurea	−6.76	sulfonate 3.15 Å; largest WT selectivity among AI seeds
4	Infigratinib	approved FGFR inhibitor	−6.79	geometric H-bond 3.17 Å, selectivity +1.30
5	Cynarin	artichoke polyphenol	−6.56	carboxylate 3.09 Å

Aprepitant remains the occupancy champion and fails the NZ H-bond / ion gate. It is not a cover candidate. A local latent-space generator (115 molecules, four rounds) never beat the seeds; hederagenin still had the best composite, glimepiride the best selectivity.

Stage B — 2.43 million CNS ligands on K1002

In-house catalog (ChEMBL + ZINC), Wager CNS subset (bbb_cns: MW ≤ 450, TPSA ≤ 90, HBD ≤ 3, 1 ≤ clogP ≤ 3.5). Uni-Dock 1.2, Vina scoring, GPU, 22 Å cube on mutant K1002 NZ. Rank: mutant Vina, NZ distance, H-bond / ion flags, then WT selectivity.

2,427,793

Successful docks (cursor.json n_ok).

Hits after MUT affinity, NZ geometry, and WT selectivity.

Notoamide B

Rank 10. Fungal prenylated indole. Natural product. Not FDA-approved.

Rank	ID	Class	MUT	WT	Sel.	NZ Å
1	CHEMBL1445111	quinazoline-dione–piperazine	−8.64	−5.75	+2.90	2.96
2	CHEMBL229651	fused indole–DKP	−8.09	−6.29	+1.80	3.30
3	CHEMBL5816698	quinazolinone–piperazine	−7.96	−6.11	+1.85	3.22
4	CHEMBL4633401	benzofuran–barbiturate-like	−7.95	−6.02	+1.93	3.39
5	CHEMBL1255082	naphthamide–spiro imidazolidinone	−7.85	−6.06	+1.79	2.98
6	CHEMBL3781825	bis-tetrahydro-β-carboline DKP	−7.83	−6.35	+1.48	3.04
7	CHEMBL1365318	quinazoline-dione–piperazine	−7.81	−5.94	+1.87	3.04
8	CHEMBL1537224	anthracene–succinimide	−7.75	−6.31	+1.44	3.00
9	CHEMBL6015444	spiro-phthalide oxazolinone	−7.74	−6.69	+1.05	3.05
10	CHEMBL2042022 · notoamide B	Aspergillus prenylated indole	−7.74	−7.23	+0.51	3.16

DynamicBind on the Stage B top 100 failed on this Windows build (torch_cluster DLL). These ten ranks are Vina + NZ geometry, not induced-fit confirmation. They are a docking hypothesis, not the prize library.

BUB1B knockdown: the imaging assay

HeLa, pLKO.1-puro shRNA, Hoechst 33342, 40×. 14 October 2026. This is a loss-of-function phenocopy of SAC failure — not an N1002K patient line.

Control versus BUB1B knockdown Hoechst fields and nuclear segmentation

Figure. Control (N = 88 nuclei) versus BUB1B KD (N = 43). Same field split, cropped originals, and instance segmentation used for the table below.

Metric	Control	BUB1B KD	Change
Nuclei counted	88	43	−51%
Mean nuclear area (px)	288.8	233.7	−19%
Median area (px)	273.0	232.0	shift to smaller nuclei
Area < 160 px	11.4% (10/88)	20.9% (9/43)	~2×
Micronucleus-like	1.1% (1/88)	4.7% (2/43)	~4×
Pyknosis (bright small nuclei)	3.4% (3/88)	9.3% (4/43)	~3×
Mean intensity	140.0 ± 17.9	130.0 ± 17.6	KD more heterogeneous

Quantitative comparison table of control versus BUB1B knockdown

Source quantification table from the imaging notebook.

Notes on nuclear heterogeneity, micronuclei, and pyknosis

How micronuclei and pyknosis were scored (area and intensity gates).

HTS definition of a hit: a compound that moves KD nuclear count, size heterogeneity, micronuclei, or pyknosis toward control. Binding BubR1 is not required. A downstream or bypass suppressor is a valid Track 2 output.

If we win: Academia Sinica HTS, FDA drugs, phenotype reversal

Docking will not treat MVA. The bottleneck is a wet-lab screen on a library that can actually be given to a child.

Platform

HeLa BUB1B shRNA. Readout = nuclear morphology already quantified.

Library

FDA-approved drugs. Not the ten unapproved Stage B poses.

Partner

Academia Sinica high-throughput screening core. Prize covers plates, imaging, core time.

Transfer the KD nuclear assay to the Academia Sinica HTS core; lock plate QC (Z′, DMSO).
Screen an FDA-approved collection for reversal of KD nuclear morphology.
Confirm actives in dose–response on the same imaging metrics. Mechanism (BubR1 vs bypass) is secondary.

Limits

AlphaFold kinase fragments, not a crystal structure of full-length BubR1.
Vina / Uni-Dock scores are not Kd.
Stage B top 10 are not DynamicBind-validated.
Most Stage B IDs are not approved drugs. Track 2’s approved-drug request is carried by Stage A plus the planned HTS library.
Whether 1002K is a real ubiquitin acceptor was not measured.
KD is HeLa wild-type knockdown, not patient cells and not an N1002K knock-in.
No human dosing, no pediatric PK, no claim of efficacy.

Selected references

Suijkerbuijk et al. Cancer Res 2010 — MVA kinase-domain missense → instability, HSP90, proteasome.
Suijkerbuijk et al. Dev Cell 2012 — BubR1 as an unusual pseudokinase; KARD recruits PP2A-B56.
Bolanos-Garcia & Blundell. FEBS J 2011 — L1012 on αG.
Ding et al. JACS 2010 — stephacidin / notoamide anticancer alkaloids.
ChEMBL 34, CHEMBL2042022 (notoamide B).
Wager et al. — CNS multiparameter filters used for bbb_cns.