Active-site-constrained structure-guided hypotheses for six uncharacterised enzymes of Mycobacterium tuberculosis

Roughly a quarter of the Mycobacterium tuberculosis H37Rv proteome is still annotated as "hypothetical", and many of these conserved-hypothetical genes are predicted enzymes whose substrate and chemistry remain unassigned. Transferring function from structural matches has become routine: proteome-wide search assigns candidate folds cheaply, and several bacterial proteomes, including a mycobacterial one, have been re-annotated this way. Two things such transfers rarely test are whether the candidate can actually perform the chemistry, and whether the function proposed is not already carried elsewhere in the genome. Here we make both conditions of acceptance rather than remarks. A structure-guided pipeline (ESMFold modelling, Foldseek/TM-align template search, profile–profile validation with HHpred, three-dimensional active-site mapping cross-validated with AlphaFold3, transposon-derived essentiality, and purifying-selection analysis across ~250,000 sequenced Mycobacterium tuberculosis genomes) is applied to six conserved hypothetical proteins flagged as putative novel enzymes under two rejection criteria: the catalytic constellation must be geometrically competent, measured atom by atom on the superposed template, and the proposed function must not already be attributed to another locus (a "fold-paralogue" safeguard). Three targets emerge as robust novel enzymes with a structurally competent active site: Rv1118c, a circularly permuted NlpC/P60 amidase with an assembled Cys234–His97–Glu115 triad acting on an N-acyl-fatty-amide bond; Rv3577, a binuclear metallo-β-lactamase-fold hydrolase of the UPF0173/UlaG family with an unassigned substrate; and Rv2492, a thymidylate-synthase-superfamily (hydroxy)methyltransferase whose specificity residues point to a deoxyuridylate-type substrate rather than the cytidylate substrate suggested by its closest structural template. A fourth, Rv2558, is a cofactor-independent ABM monooxygenase with a single competent pocket but no demonstrable substrate (weaker candidate). A fifth, Rv3196, is rejected and reported as a false positive: its template match covers only the N-terminal scaffold of a YcaO heterocyclase, none of the catalytic domain, an instance of the normalised-TM-score pitfall a fold-level transfer would have propagated. The sixth, Rv3839, is a heme-homeostasis regulator independently characterised while this work was in progress, an a-posteriori confirmation of one structure-guided hypothesis. All six genes are non-essential in vitro yet their catalytic residues are essentially invariant across the sampled genomes, the signature of maintained, functional enzymes. The single loss of function we detect supports that reading: Rv1118c is disrupted by a fixed frameshift throughout M. africanum lineage 6 and is a pseudogene in M. leprae, both undergoing reductive evolution. Four of the six proteins are detected in public mass-spectrometry compendia. We provide testable biochemical predictions for each target, including the one we reject.Preprint of a manuscript submitted to a peer-reviewed journal. All structural models are predictions; no experimental structure or activity is reported. Supplementary data and code: 10.5281/zenodo.20571949.