celA2a Family assigned · low auto-curated · to review

Annotation: from legacy to revised

Curation note: Added P16.5d: Mycobrowser-annotated gene absent from RefSeq NC_000962.3 (the atlas' original 3906-CDS reference). Foundation record from the Mycobrowser release + translated H37Rv sequence; heavy enrichment layers (structure, orthology, conservation, interaction) pending.

Auto-curated: this verdict and function were generated by rules from PGAP + Pfam + Foldseek and have not been hand-reviewed (flagged for human review).

In the literature (TB corpus sweep)

This distinguishes a gene that is dark because nobody has looked from one that is dark despite having been studied. Source: .

CRISPRi vulnerability

Vulnerability index -2.95 (95% CI -14.12 to 8.86). A more negative index = more vulnerable to knockdown (better drug-target quality); indicative threshold VI ≤ -6 = highly vulnerable.

Quantitative CRISPRi knockdown, graded (finer than binary Tn-seq essentiality). Source: CRISPRi vulnerability index (Bosch 2021, supplementary Table S2 'mmc3.xlsx', bulk import via Europe PMC -- supersedes the per-gene pebble.rockefeller.edu scrape of phase46_crispri_vulnerability.py).

Caveat on this estimate. Bosch et al. flag this gene's vulnerability estimate as NOT certain, and here that call has a mechanical cause visible in the library itself: every guide targeting this gene is >=21 nt (str_span=0.018 on a 0-1 scale, near the genome floor), and guide repression strength saturates from 21 nt on -- so the library never sampled a partial knockdown for this gene, only the near-maximal repression plateau. The dose-response fit (and this VI) is therefore essentially unconstrained by data, not merely uncertain -- a limitation of the CRISPRi library design for the non-essential class (Bosch et al. 2021), not a defect in this specific measurement. n_guides=2 is descriptive only and does not predict this. Do not cite this VI as a standalone vulnerability claim.

Legacy record & comparison (Mycobrowser)

The legacy Mycobrowser record is shown for verification. Mycobrowser is no longer maintained; its EC numbers predate recent nomenclature revisions, so a class change usually reflects re-numbering, not a conflict.

Curated reference (UniProt)

Functional vocabulary (eggNOG-mapper, orthology transfer)

Orthology-based transfer (eggNOG 5.0.2, diamond). EC/KO/GO/CAZy are computed annotations, not manual curation; cross-check against the primary literature before treating a specific reaction as established.

Conservation & selection (intra-MTBC, 142 601 strains, snapshot of 2026-09-29)

pN/pS from segregating SNPs (singletons removed) normalised by possible sites. Low pN/pS = purifying selection (a strong signal that a "hypothetical" is a real, constrained gene). A high pN/pS is ambiguous: relaxed constraint or positive selection (drug resistance, antigenic variation) inflate it; e.g. rpoB/katG/pncA score high here for resistance, not loss of function. A clonal disruption (one allele over a clade) suggests lineage pseudogenisation; a convergent one (many independent alleles) is typical of resistance loss-of-function.

Essentiality (transposon mutagenesis)

Genome-wide Himar1 transposon essentiality in H37Rv (DeJesus 2017). An essential call (ES/ESD/GD) is strong, independent evidence that a "hypothetical" locus encodes a functional, selectively required gene — orthogonal to intra-species conservation.

Mutant phenotypes (conditional Tn-seq, MtbTnDB) in-vivo phenotype

Conditional fitness of transposon-disruption mutants across 2 significant condition(s) (|log2FC|≥1, q≤0.05), from the standardized MtbTnDB compendium. A negative log2FC means the mutant is depleted — the gene contributes to fitness in that condition. An in-vivo defect for a "hypothetical" is strong evidence it matters for infection, even without a known molecular function. Disruption (Tn insertion), not a clean deletion; genetic-interaction screens excluded.

Proteomics (mass spectrometry) not detected

Not detected in the mass-spectrometry datasets integrated by PaxDb. This is not evidence of absence, and this atlas can say which bias dominates, because it measured it on its own 482 undetected proteins. Size is the main factor: undetected proteins have a median length of 145 aa against 295 aa for detected ones, and 25.3 % of them are shorter than 100 aa against 6.0 %. Predicted localisation matters far less than usually assumed: detection rates are 89.3 % for globular, 84.9 % for transmembrane, 88.0 % for lipoproteins and 77.8 % for signal-peptide proteins — so a short protein is under-sampled whatever its predicted compartment, and hydrophobicity alone is a weak excuse. Condition-specific expression and few tryptic peptides (PE/PPE repeats) add to this.

Predicted localisation (DeepTMHMM + lipobox) signal peptide

Transmembrane topology and signal peptide from DeepTMHMM (deep-learning reference predictor); lipoproteins from a (myco)bacterial lipobox motif. A sequence-based prediction of subcellular context.

Physico-chemical properties (computed, ProtParam)

Computed from the ancestral MTBC0 sequence with the ExPASy ProtParam method (Biopython). Descriptive biophysical context: a positive GRAVY flags a hydrophobic (often membrane) protein, a high instability index (>40) predicts a short in-vitro half-life, an extreme pI hints at compartment or binding partner.

Domains (Pfam, hmmscan --cut_ga)

No Pfam-A domain above the gathering threshold (hmmscan --cut_ga scan completed).

3D structure (interactive — Mol*/pdbe-molstar)

No experimental structure is deposited for this gene: showing the AlphaFold DB predicted model (mean pLDDT 71.7, confident), coloured by per-residue confidence (dark blue = very high, orange/red = low — treat low-confidence regions as unreliable, especially termini and loops). Drag to rotate, scroll to zoom.

Transcriptional regulation (signed TRN: ChIP-seq + TFOE)

Regulatory edges from the ISB signed transcriptional regulatory network (TF ChIP-seq binding, Minch 2015 + TF-overexpression response, Rustad 2014). An edge is regulatory evidence (binding and/or expression change), not necessarily direct. For a "hypothetical", membership in a known regulon (e.g. DosR dormancy, PhoP virulence) is a physiological-context lead, not a function.

Condition: Condition-blind by construction. These edges come from transcription-factor OVEREXPRESSION in a single laboratory condition, so a listed regulator is not a claim that it governs this gene in every circumstance. Measured counter-examples exist within this atlas (see Rv2516c, where the DosR regulon collapses under moxifloxacin while the gene itself rises). Read a regulator as 'capable of shifting this gene', and look for the condition before building a hypothesis on it.

Node colour = verdict, dashed = still uncharacterised; edge colour = evidence (green experimental, orange genomic-context, grey co-expression), width ∝ score. Click a partner to open its page; "Explore full network" to walk the graph.

Closest characterised functional partner: celA2b (Rv1090, (MTV017.43), len: 151 aa. Probable celA2b,second part of cellulase (endoglucanase), similar to C-terminus of others e.g. O08468 cell), high confidence (score 978, all STRING channels except text-mining; the partner is chosen on its genomic-context evidence). The pair is recorded in a curated database (pathway or complex membership) for M. tuberculosis genes themselves (score ≥ 200): a curated association, not a measurement in this organism.

Showing the 15 strongest of this gene's 32 partners at medium confidence (15 at high confidence). The table is an extract, not a neighbourhood: across the proteome the real median degree is 34 while the served median is 15, so no hub, degree distribution or centrality may be read on this table. None of the 32 partners rests on an interaction measured in M. tuberculosis itself.

STRING combines evidence channels (neighborhood, fusion, cooccurrence, coexpression, experimental, database, text-mining) into a 0–1000 score. The ctx badge marks edges carried by the genomic-context channels (conserved neighborhood, fusion, phylogenetic co-occurrence), which are independent of orthology and structure and the strongest signal for an unknown gene. The no text-mining column recomputes the score from data alone, so a link that does not depend on the literature is visible. Association is a function hypothesis, not proof: corroborate with the operon context and the primary literature before assigning a function.

Where the experimental evidence comes from, and why it matters here. STRING's experimental channel fuses interactions measured in M. tuberculosis with interactions measured in another organism and transferred to this pair by orthology. The two are not equivalent for this atlas, whose standing rule is that better studied elsewhere is not a function established in M. tuberculosis. Measured on the served network: of the 38 229 partner pairs shown across the atlas, 13 483 partner rows carry a non-zero experimental channel, and only 106 pairs rest on an interaction measured in M. tuberculosis itself — 0.3 % of the network. Those 106 are long-established complexes (ESX-1/ESAT-6, RNA polymerase with CarD and RbpA, the Pup–Mpa proteasome, gyrase, Emb, GarA with PknB and PknG, the Rel modules); no still-uncharacterised gene is among them. The reason is not a gap in this atlas but in the literature: there is almost no experimental interactome of this bacterium, IntAct holding 134 curated protein pairs for a ~4 000-gene proteome. A exp (transferred) badge is therefore a lead, never a measurement, and an absence of badge is not evidence that the pair does not interact.

Ancestral MTBC0 protein sequence

>|Rv1089A|celA2a
MNGAAPTNGAPLSYPSICEGVHWGHLVGGHQPAY