Case studyWHO Critical Priority Pathogen — Tuberculosis

InhA — Enoyl-ACP Reductase

The primary target of isoniazid — the most widely used first-line anti-tuberculosis drug. InhA is a validated, extensively studied druggable target with one of the highest druggability scores in the ResistAI database.

Protein overview

UniProt IDP9WGR1
GeneinhA
OrganismMycobacterium tuberculosis H37Rv
Resistance familyIsoniazid / TB resistance
WHO priorityCritical — drug-resistant tuberculosis (DR-TB)
MechanismNADH-dependent enoyl-ACP reductase; inhibited by isoniazid-NAD adduct

Clinical significance

Tuberculosis kills ~1.6 million people annually. Isoniazid resistance — predominantly driven by mutations in inhA and katG — is the most common mechanism of first-line drug resistance in M. tuberculosis. InhA mutations (e.g. C15T promoter variant, I21V, I47T) confer low-level isoniazid resistance by reducing enzyme-drug binding affinity. Direct InhA inhibitors — compounds that bypass the need for KatG-mediated activation — represent a major strategy in the current TB drug discovery pipeline.

Druggability analysis — fpocket

0.983
Best druggability score
High
Druggability tier
14
Total pockets detected
1
High-druggability pockets

Key finding

InhA scores 0.983 — placing it in the top 1% of all 2,433 analysed resistance proteins. This is consistent with decades of experimental drug discovery: InhA harbours a well-defined, hydrophobic NADH/NAD⁺ binding pocket that has been successfully targeted by both prodrugs (isoniazid) and direct inhibitors (GSK693, triclosan analogues, thiadiazoles).

Interpretation

fpocket identified 14 surface cavities on the AlphaFold-predicted InhA structure. The top-ranked pocket (score 0.983) corresponds to the substrate-binding channel adjacent to the NAD⁺ cofactor — the site occupied by isoniazid-NAD adduct in crystal structures. The high volume, hydrophobicity, and geometric regularity of this cavity explain the exceptional druggability score.

Note: Druggability scores are structural proxies computed by fpocket on AlphaFold-predicted structures (Le Guilloux et al. 2009). Thresholds: high ≥ 0.7, medium ≥ 0.4. Experimental validation is required to confirm binding site tractability.

ESM-2 embedding similarity

480-dim cosine similarity

ESM-2 similarity search against all 2,433 proteins reveals InhA's evolutionary context — and, critically, identifies fabI (E. coli enoyl-ACP reductase) as a close homolog (similarity 0.98). This is biologically validated: InhA and FabI are both members of the short-chain dehydrogenase/reductase (SDR) superfamily and share the same catalytic mechanism. Cross-species druggability comparison enables prioritisation of inhibitor scaffolds with broad-spectrum potential.

Key finding: ESM-2 correctly identifies FabI (E. coli) as a homolog of InhA despite no explicit structural alignment — demonstrating that sequence-level embeddings capture functionally relevant evolutionary relationships. FabI inhibitors such as triclosan have been explored as InhA inhibitor scaffolds, validating this similarity.

Pipeline summary

01

UniProt ID P9WGR1 resolved — InhA sequence from M. tuberculosis H37Rv fetched via UniProt REST API

02

3D structure retrieved from AlphaFold DB v4 — high-confidence predicted model

03

fpocket detected 14 binding cavities — top pocket scored 0.983 (high druggability, top 1% of all 2,433 proteins)

04

ESM-2 generated 480-dim embedding — ChromaDB cosine search identified FabI homologs (similarity 0.98) and InhA orthologues across mycobacteria

05

RAG retrieved literature on isoniazid resistance mechanisms and direct InhA inhibitor development

06

Llama 3.3 70B generated PMID-cited summary confirming InhA as a high-priority validated drug target