Transposon Sequencing

What this entity is

Canonical name: Transposon Sequencing (TnSeq). ₁

Known aliases: TnSeq; transposon sequencing; transposon insertion sequencing; RB-TnSeq (random-barcode transposon sequencing). _{1 2}

Stable external identifier: No stable external identifier is specified in the source reports. ₁

TnSeq is a sequencing-based method for measuring gene essentiality and mutant fitness through transposon insertions. The kescience-fitnessbrowser RB-TnSeq compendium measured mutant fitness across 14 organisms and carbon-source conditions, while ADP1 data provide essentiality measurements. _{1 2 3}

Across 48 organisms, 57,011 of 228,709 genes (24.9%) were classified as dark; 7,787 had strong fitness effects (|fitness| ≥ 2 in at least one condition), and 9,557 were essential because no viable transposon mutants were recovered. ₄ This makes essentiality a prioritization signal, but absent insertion mutants require separate CRISPRi validation because RB-TnSeq cannot produce viable knockout fitness profiles for essential genes. ₄

The Truly Dark Genes — What Remains Unknown After Modern Annotation? refines this dark-gene use case by separating annotation-lag genes from genes that remain hypothetical after modern annotation. Of 57,011 dark genes, 39,532 had pangenome links: Bakta reclassified 33,105 (83.7%), while 6,427 remained hypothetical in both pipelines; 17,479 lacked pangenome links. ₅ Truly dark genes were shorter (median 121 versus 194 amino acids), less core-associated (43.1% versus 72.7%), more often essential (18.0% versus 13.4%), less GC-rich (0.542 versus 0.584), and less broadly orthologous (29.3% versus 63.7% with orthologs; median breadth 1 versus 4 organisms) than annotation-lag genes. ₅ These measurements support using TnSeq fitness, essentiality, genomic context, and cofitness to prioritize experimentally testable unknown genes, while short-gene insertion and polar-effect biases limit causal interpretation. ₅

Metabolic capability and condition-specific fitness

The metabolic-capability analysis refines TnSeq interpretation by separating genomic pathway capability from experimentally observed dependency. Across 161 organism–pathway pairs spanning 7 model bacteria and 23 GapMind pathways, 57 (35.4%) were Active Dependency, 66 (41.0%) Latent Capability, 24 (14.9%) Incomplete but Important, and 14 (8.7%) Missing. Active Dependency means a complete pathway containing fitness-important genes; Latent Capability means a complete pathway whose genes showed no significant fitness defects under standard conditions; Incomplete but Important means an incomplete pathway whose mapped genes were fitness-important; and Missing means neither complete nor important. ₆ The composite importance score weighted essentiality at 40%, fitness breadth at 30%, and fitness magnitude at 30%; Fitness Browser genes were mapped to GapMind through the besthitkegg-to-keggmember-to-EC-to-KEGG-map-to-GapMind pathway route. ₆

This supports and extends evidence that TnSeq fitness is condition-dependent. All 66 aggregate Latent Capability pairs became fitness-important under at least one condition type, with nitrogen limitation, stress, and carbon limitation the most frequent triggers. ₆ The result is partly threshold-dependent: applying a median-based importance threshold to condition-specific subsets can cause pathways to cross the threshold by construction, so independent calibration against known essentials, such as the Essential Metabolome: GapMind Pathway Analysis Across Essential-Gene Organisms reference set, is required before treating every reclassification as biological confirmation. ₆

Active Dependencies had mean core gene completeness of 0.986 versus 0.975 for Latent Capabilities. ₆ This small difference is limited by the use of well-studied model organisms with near-complete, well-annotated core genomes: only 7 of 48 Fitness Browser organisms had matching GapMind genome data. The Tier 1 organisms were Desulfovibrio vulgaris Hildenborough, Shewanella oneidensis MR-1, Pseudomonas putida KT2440, Pseudomonas stutzeri RCH2, Caulobacter crescentus, Sinorhizobium meliloti, and Azospirillum brasilense. ₆

The same analysis tested 2,810 GTDB species with at least 10 genomes across approximately 293,000 genomes and 80 GapMind pathways: 18 amino-acid biosynthesis pathways and 62 carbon-source utilization pathways. Variable pathway count correlated with pangenome openness at raw Spearman rho=0.327, p=7.2e-71, and at partial Spearman rho=0.530, p=2.83e-203 after controlling for genome count. This supports pangenome-integration and provides a pathway-level signal, contrasting with null relationships between openness and broader environment or phylogeny effect sizes reported by the pangenome_openness project. ₆

The pathway–openness signal was positive in 13 of 18 genera and significant at p<0.05 in 5 of 18: Clostridium, Eubacterium, Mesorhizobium, Pseudomonas, and Streptomyces. The strongest within-genus correlations were rho=0.534 for Clostridium and rho=0.506 for Eubacterium. ₆

Accessory genes contributed most visibly to amino-acid biosynthesis completeness. Leucine and valine each had all-gene completeness 0.614, core-only completeness 0.468, and a gap of 0.146; arginine had 0.613, 0.472, and 0.141; lysine had 0.804, 0.664, and 0.140; and threonine had 0.803, 0.663, and 0.140. ₆ These differences refine core-versus-accessory interpretations and are consistent with Black Queen dynamics and possible community sharing of leucine, valine, arginine, and lysine biosynthetic capacity, but do not demonstrate metabolite exchange. ₆

Among 225 species with at least 50 genomes and at least 3 variable pathways, binary completeness profiles across 80 pathways were clustered with hierarchical clustering and Jaccard distance. The median number of metabolic ecotypes was 4 per species and the maximum was 8, observed in Alistipes onderdonkii and Barnesiella intestinihominis. Ecotype count correlated with pangenome openness at raw Spearman rho=0.262, p=6.8e-05, and partial rho=0.322, p=8.0e-07 after controlling for genome count. ₆ This supports ecotype-environment-gene-content while genome sampling and the fixed 50% maximum-distance clustering cut remain limitations. ₆

These pathway-level results connect to prior fitness-conservation findings: 87.4% of metal-fitness genes were core, with OR=2.08 for enrichment; 28,017 genes were costly in laboratory conditions but conserved in nature; and field-important genes were 83.6% core versus a 76.3% baseline. ₆

The broader metabolic-capability analysis classified 1,695 complete pathway–organism pairs from 48 organisms as latent, intermediate, or active using mean absolute fitness t-score and essentiality thresholds. Of these, 267 (15.8%) were latent, 547 (32.3%) intermediate, and 881 (51.9%) active; pathway category predicted dependency class (χ²=163.6, df=4, p=2.5×10⁻³⁴). Carbon-source pathways were 217 of 892 (24.3%) latent, 320 (35.9%) intermediate, and 355 (39.8%) active, while amino-acid biosynthesis pathways were 48 of 735 (6.5%) latent, 220 (29.9%) intermediate, and 467 (63.5%) active; carbon pathways were therefore 3.7× more likely to be latent. ₇ Pathway-level conservation did not distinguish latent capabilities from active dependencies: mean conservation was 0.829 for 755 active dependencies, 0.907 for 508 intermediate pathways, and 0.869 for 248 latent pathways; the active-versus-latent Mann–Whitney U test gave p = 0.94 and rank-biserial r = 0.052. At clade level, latent capability rate correlated positively with pangenome openness (Spearman ρ = 0.69, p = 0.0004, n = 22 clades), supporting a Black Queen framework signal at the level of genome dynamics and community context rather than simple per-pathway conservation. ₇

Metal-fitness applications

A pan-bacterial atlas extended RB-TnSeq to 559 metal-related experiments across 31 organisms and 16 metals, yielding 383,349 gene × metal fitness records. ₈ It refines metal-fitness interpretation: across 22 organisms and 14 metals, metal-important genes were 87.4% core versus 76.9% core for baseline genes (OR=2.08, p=4.3e-162), rather than being predominantly accessory resistance genes. ₈ This supports a two-tier model in which TnSeq detects broadly required core stress and cellular functions as well as accessory specialized resistance mechanisms. ₈

Across 6,504 experiments—559 metal and 5,945 non-metal experiments—7,609 metal-important gene records from 24 organisms were classified as 4,177 (54.9%) metal-specific, 2,888 (38.0%) general sick, and 544 (7.2%) metal+stress. ₉ At a 2% sick-rate threshold approximately 41% were metal-specific, and at 10% approximately 67% were metal-specific. ₉ This supports the two-tier interpretation while showing that broad metal-important calls contain a substantial general-stress component. ₉

Metal-specificity varied by metal: manganese 60.6% (20/33), molybdenum 60.5% (185/306), tungsten 56.5% (173/306), selenium 46.4% (64/138), cadmium 55.9% (52/93), copper 51.9% (1,346/2,594), cobalt 50.2% (1,167/2,324), chromium 49.3% (132/268), uranium 48.6% (88/181), zinc 47.2% (843/1,786), nickel 43.7% (993/2,271), aluminum 42.4% (752/1,772), mercury 32.7% (35/107), and iron 21.9% (144/659). ₉

Among 22 organisms with pangenome links, metal-specific genes were 84.8% core (2,969/3,500) pooled and 88.0% by organism mean, with mean delta versus baseline of +6.9%; 19/22 organisms had positive deltas and 12/22 showed significant enrichment. General sick genes were 90.2% core (2,183/2,420), with mean delta +9.0%; 21/21 organisms had positive deltas and 8/21 showed significant enrichment. Metal+stress genes were 94.3% core (467/495) pooled and 93.6% by organism mean, with mean delta +10.9%; 13/13 organisms had positive deltas and 1/13 showed significant enrichment. ₉ All three categories were significantly core-enriched above a 79.8% pooled baseline (73,957/92,650; organism mean 81.1%), but metal-specific genes were less core-enriched than general sick genes; the Cochran-Mantel-Haenszel comparison was significant (p=0.011). ₉

Metal-specific genes matched metal-resistance keywords at 12.2% versus 7.8% for general sick genes (Fisher exact OR=1.64, p=2.4e-8). ₉ Candidate specificity prioritized UCP030820, YebC, and DUF1043/YhcB: respectively metal-specific in 2/3 organisms across 7 metals (67%) with mean sick rate 0.021; 7/12 organisms across 6 metals (58%) with mean sick rate 0.056; and 3/6 organisms across 5 metals (50%) with mean sick rate 0.054. ₉ The proposed YebC mechanism is a hypothesis, not an established function. Novel candidate families were less often predominantly metal-specific than annotated families: 45.6% of 149 novel families versus 58.2% of annotated families (Fisher OR=0.60, p=0.003). ₉

Independent component analysis (ICA), a decomposition method for activity profiles, identified 0 metal-specific modules in raw module-condition scores; per-module z-normalization produced maximum absolute z values <2.0 for most metal experiments because metal experiments were a small fraction of each organism’s experiments. This contrasts with the 600 metal-responsive module records obtained from precomputed z-scored Metal Atlas profiles and indicates that module responsiveness depends on scale and within-organism normalization. _{8 9}

RB-TnSeq is an independent callable-evidence tier in the ENIGMA Carbon Census. Lauric acid had measured fitness in a reference bacterium but no ENIGMA-isolate utilizer rows; an 83-compound funnel produced 9 callable compounds, while xanthine was mis-scored because reaction R02107 represents purine nitrogen acquisition rather than carbon catabolism, leaving an effective carbon-callable set of 8 while committed tables retain xanthine. Fitness therefore requires interpretation of condition, substrate, and phenotype rather than automatic metabolic assignment. ₁₀

The counter-ion analysis used RB-TnSeq across 71 NaCl experiments to compare 10,821 metal-important gene records from 86 organism × metal pairs spanning 14 metals and 19 organisms; 4,304 records (39.8%) were also NaCl-important. ₁₁ Zinc sulfate produced 44.6% overlap despite delivering 0 mM chloride, compared with cobalt at 41.3%, copper at 41.0%, and nickel at 39.3%, so chloride dose was not the primary driver. ₁₁ Shared signals may reflect cell-envelope, DNA-repair, and ion-homeostasis responses, but these remain hypotheses because formal enrichment tests were not performed. ₁₁ This supports the atlas’s caution that general stress responses can dominate broad metal-fitness calls. ₈

The counter-ion comparison in the metal-specificity analysis reported 39.8% overlap between metal-important and NaCl-stress genes, whereas the stricter analysis found 14.7% of metal-important genes sick under osmotic stress, a 2.7x discrepancy. The direction of overlap supports both analyses; most of the difference was attributed to the stricter fit < -1 and |t| > 4 threshold and partially different organism sets. ₉

Conservation and condition-specific fitness

In one synthesis of 194,216 protein-coding genes from 43 bacteria, essential genes were 82% core versus 66% core among genes always neutral in tested experiments. ₁₂ A later integration mapped 44 of 48 Fitness Browser organisms to pangenome species clades, retained 33 for essentiality analysis, and produced 177,863 gene-to-cluster links with 100.0% median protein identity and 94.2% median gene coverage. It identified 27,693 putative essential genes among 148,826 protein-coding genes; essential genes were 86.1% core versus 81.2% of non-essential genes, with median odds ratio 1.56 and 18 of 33 organisms significant at BH-FDR q < 0.05. BH-FDR is the Benjamini–Hochberg false-discovery-rate adjustment. ₁₃ These findings refine binary essentiality: laboratory fitness can be compared with evolutionary retention, but neither measurement alone establishes natural-environment fitness. _{12 13}

Fitness importance showed a graded but weak conservation relationship across approximately 194,000 genes from 43 bacteria. Essential genes with no viable mutants were 82% core (n=27,693), always-neutral genes 66% core (n=94,889), often-sick genes 78% (n=15,989), mixed genes 70% (n=20,739), sometimes-sick genes 72% (n=25,201), and sometimes-beneficial genes 70% (n=9,705). Binning by strongest effect gave 82.2% core for essential genes, 77.7% for min_fit < -3, and 66.4% for min_fit from -1 to 0. Fitness breadth was positively but weakly associated with core status (Spearman rho=0.086, p=8.1e-230): genes affecting 20+ experiments were 79% core, 6-20 were 73%, 1-5 were 71%, and 0 were 66%. ₁₄ Thus TnSeq fitness is statistically informative but only a weak predictor of conservation. ₁₄

The metal atlas contradicts a simple generalization from the prior DvH heavy-metal result: heavy-metal genes in that condition-specific analysis were 71.2% core, whereas the atlas’s genome-wide metal-important set was 87.4% core. _{15 8} The difference is attributed to scope: the broad definition captures core cellular processes vulnerable to metal disruption, including cell envelope functions, DNA repair, central metabolism, protein quality control, and general stress response, while condition-specific analyses more clearly expose specialized resistance functions. ₈ Essential-metal tolerance genes for Fe, Mo, W, Se, and Mn had a mean core-fraction delta of +0.148, versus +0.081 for toxic metals; the difference was significant by one-sided Mann-Whitney U testing (U=39, p=0.015). ₈

Core genes were more likely than auxiliary genes to show a positive fitness effect when deleted: 24.4% versus 19.9%, with OR=0.77 for auxiliary versus core; their fitness distributions had heavier tails in both negative and positive directions. ₁₄ This contradicts the expectation that accessory genes are generally costly to carry and supports condition-dependent trade-offs. Strong condition-specific phenotypes were 77.3% core versus 70.3% without such annotations (OR=1.78, p=1.8e-97), and 4,450 genes (2.7%) fit an ephemeral-niche pattern of overall neutrality but criticality in one condition; these were 3.0% of core genes, 1.7% of auxiliary genes, and 1.6% of singleton genes. ₁₄

The Desulfovibrio vulgaris Hildenborough analysis classified 757 experiments into lab-nutrient (237, 31.3%), field-core (204, 26.9%), lab-other (140, 18.5%), field-stress (78, 10.3%), heavy-metals (55, 7.3%), and lab-antibiotic (43, 5.7%); 337 were field (44.5%) and 420 lab (55.5%). Among 2,725 non-essential genes with fitness and pangenome links, 76.3% were core; 678 essential genes were 80.1% core but lacked fitness data because no transposon mutants were recovered. ₁₅

At fitness < -2, field-stress genes were 83.6% core (298 genes, OR=1.58, FDR q=0.026), field-core 82.4% (376, OR=1.46, q=0.026), lab-other 81.5% (292, OR=1.37, q=0.073), lab-nutrient 81.4% (452, OR=1.36, q=0.037), lab-antibiotic 73.4% (109, OR=0.86, q=0.49), and heavy-metals 71.2% (198, OR=0.77, q=0.14). ₁₅ Fitness importance was more informative than the field-versus-lab label, while antibiotic and heavy-metal resistance functions were disproportionately accessory. Across thresholds from -1 to -3, field-stress genes were 89.4%, 83.6%, 84.0%, and 82.1% core; field-core 84.4%, 82.4%, 81.9%, and 79.0%; lab-nutrient 83.5%, 81.4%, 81.0%, and 79.5%; lab-other 85.2%, 81.5%, 80.1%, and 77.5%; lab-antibiotic 82.2%, 73.4%, 77.3%, and 76.2%; and heavy-metals 70.6%, 71.2%, 76.7%, and 76.7%. ₁₅

Fitness modules, dark genes, and pangenome conservation

The ICA module analysis supports the view that TnSeq modules capture process-level network structure. It analyzed 1,116 ICA modules across 32 organisms, module families, gene-to-cluster conservation mappings, and essentiality classifications. Among 974 modules with at least 3 mapped genes, module genes were 86.0% core versus 81.5% for all genes, an enrichment of +4.5 percentage points (OR=1.46, p=1.6e-87). ₁₆

Of the 974 mapped modules, 577 (59%) were core modules (>90% core genes), 349 (36%) mixed modules (50-90% core), and 48 (5%) accessory modules (<50% core); the median module was 93.4% core. ₁₆ This refines the broader module architecture: most co-regulated fitness-response units are embedded in the conserved genome, but accessory module families remain. Family breadth did not predict conservation: Spearman rho=-0.01 with p=0.914. ₁₆ Zero essential genes appeared in any ICA module because essential genes lack usable insertion-based fitness data. ₁₆

The underlying fitness-modules analysis identified 1,116 stable modules across 32 organisms, each with at least 100 experiments; 94.2% had significantly elevated within-module cofitness by Mann-Whitney U test at p < 0.05, with mean within-module |r| of 0.34 versus 0.12 background, a 2.8x enrichment, and 22.7x genomic-adjacency enrichment. A strict |weight| >= 0.3 threshold and maximum of 50 genes produced 94% enrichment and 2.8x correlation enrichment; the initial D’Agostino K-squared approach produced 100-280-gene modules with 59% enrichment and 1-17x correlation. ₁₇

Held-out benchmarking found ortholog transfer at 95.8% strict precision, 91.2% coverage, and 0.934 F1; domain-based prediction 29.1%, 66.6%, and 0.401; Module-ICA <1%, 23.3%; and cofitness voting <1%, 73.0%. ₁₇ TnSeq modules provide process context but do not replace sequence-based orthology: KEGG KO groups averaged approximately 1.2 genes per unique KO, so a module with 20 annotated members typically contained 20 different KOs. ₁₇

Cross-organism alignment produced 1.15M BBH pairs across 32 organisms and 13,402 ortholog groups, identifying 156 module families spanning 2+ organisms, including 28 spanning 5+, 7 spanning 10+, and 1 spanning 21 organisms; 145 families had consensus labels (93%). It generated 6,691 predictions for hypothetical proteins: 2,455 family-backed (37%) and 4,236 module-only. These are conserved process-level contexts, not proof of identical molecular functions. ₁₇

The new dark-gene analysis extends this cofitness evidence: among 4,394 truly dark genes in organisms with integrated condition analysis data, 41% of neighboring genes were also hypothetical, 12.0% were within 2 genes of a mobile genetic element, and 25.9% showed operon-like cofitness with adjacent genes using |r| ≥ 0.3. A total of 594 genes belonged to ICA fitness modules, whose guilt-by-association signals included phage integrases, metal transporters, chemotaxis systems, and iron regulation. ₅ These module and neighborhood assignments are candidate functions rather than direct molecular-function demonstrations. ₅

Using z-scored module activity profiles standardized across experiments per organism, 600 metal-responsive module records were found among 19,453 module × metal-experiment records (3.1%), and the 183 responsive modules with conservation data had mean core fraction 0.826 and median 0.929. An initial raw-score analysis found zero responsive modules because raw scores were on a different scale; per-experiment z-normalization resolved the issue. ₈

Metabolic capability, deletion, and annotation evidence

All 10 target species with at least 50 genomes and at least 15 variable pathways formed metabolic clusters with silhouette scores >0.2, ranging from 0.35 for PALSA-747 sp. to 0.89 for Salmonella enterica: Salmonella enterica, 11,396 genomes, k=6, silhouette 0.894; Stutzerimonas stutzeri, 149, k=2, 0.780; Alteromonas macleodii, 56, k=2, 0.738; Acetatifactor intestinalis, 59, k=6, 0.657; Ruminococcus E sp., 53, k=2, 0.609; Phenylobacterium sp., 51, k=2, 0.544; Limivicinus sp., 50, k=2, 0.528; Pelagibacter sp., 79, k=2, 0.430; Prochlorococcus A sp., 74, k=2, 0.429; and PALSA-747 sp., 51, k=2, 0.349. ₇ Clusters were associated with isolation environment in Salmonella enterica (χ²=1570.2, df=25, p<0.0001) and Phenylobacterium sp. (χ²=12.2, df=1, p=0.0005), but not in Stutzerimonas, Alteromonas, Pelagibacter, or Prochlorococcus. ₇

The ADP1 genome_features table contains TnSeq essentiality data for 58% of 5,852 genes. TnSeq calls were compared with FBA (flux-balance analysis, a computational method for predicting metabolic flux) for 866 genes: 639 (73.8%) were concordant and 227 discordant. TnSeq essentiality was condition-specific: 499 genes were essential on minimal media versus 346 on LB. FBA flux data cover only 15% of genes and no gene has measurements across all six integrated data modalities. ₁

The respiratory-chain wiring analysis supports and sharpens this condition-specific interpretation. In ADP1, 62 respiratory-chain genes across 8 subsystems showed qualitatively different requirements by carbon source: quinate required Complex I, acetate required Complex I, cytochrome bo3, ACIAD3522, and additional components, lactate specifically required cytochrome bo3, glucose had no specifically required respiratory component, and urea was generally demanding. Complex I growth ratios were 0.37 on quinate, 1.44 on glucose, and 0.49 on acetate; ACIAD3522 had a growth ratio of 0.013 on acetate. ₁₈

The report identified three parallel NADH dehydrogenases: 13-subunit proton-pumping Complex I, single-subunit non-proton-pumping NDH-2, and single-subunit ACIAD3522, an NADH-FMN oxidoreductase. Complex I pumps 4 H⁺/NADH. NDH-2 was TnSeq-dispensable, had no growth data, and FBA predicted zero flux through it on all standard carbon sources; ACIAD3522 was dispensable on quinate and glucose but nearly lethal on acetate, although it may have another metabolic function. ₁₈ These findings refine interpretation of TnSeq–FBA discordance: growth-optimizing FBA routes NADH through the ATP-favorable Complex I and can miss alternative, capacity-limited or suboptimal respiratory pathways. ₁₈

The aromatic-catabolism analysis defined a 51-gene quinate-catabolism support network around the beta-ketoadipate-pathway; 10/13 complex-i operon subunits independently produced quinate-specific growth defects despite FBA predicting 0% essentiality. FBA captured 1.76× higher Complex I flux on aromatic substrates, with fluxes of 0.55 versus 0.31, but 30/51 quinate-specific genes had no FBA reaction mappings. ₁₉ Cross-species Fitness Browser data covered 2,005 genes and 13 conditions in 12,241 entries; Complex I ortholog mean fitness was -1.35 on aromatic conditions versus -0.77 on comparison conditions (Mann-Whitney p < 0.0001), with largest relative defects on acetate (-1.55) and succinate (-1.39), while Complex I was dispensable on glucose and lactate. This suggests, but does not establish, a role for NADH flux rather than aromatic chemistry alone. ₁₉

The ADP1 deletion analysis found 625 of 2,034 genes (31%) had condition-specific growth importance across 8 carbon sources. Among 2,593 TnSeq-dispensable genes, 272 (10.5%) lacked deletion-collection growth data and 2,321 had data; missing genes were 76.5% pangenome-core versus 93.3% for present genes (p = 1.4×10⁻²⁰). ₂₀

Among 478 genes with TnSeq, FBA, and mutant-growth data, all were TnSeq-dispensable, and FBA class was not associated with growth-defect status (chi-squared = 0.93, p = 0.63, 2 df). ₂₁ FBA concordance with complete-gene knockout lethality was moderate: rich media covered 724 genes (recall = 60.8%, precision = 64.0%, specificity = 79.7%, F1 = 0.624, Cohen’s kappa = 0.486), and minimal media 833 (recall = 65.6%, precision = 69.2%, specificity = 78.9%, F1 = 0.673, Cohen’s kappa = 0.493). ₂₁

RB-TnSeq essentiality contradicted knockout essentiality in rich media at thresholds 0.01, 0.025, 0.05, 0.10, and 0.20, with Cohen’s kappa -0.139, -0.122, -0.081, -0.024, and -0.014. At 0.05, 1,933 genes were compared (recall = 7.9%, precision = 5.8%, specificity = 82.8%, F1 = 0.067, kappa = -0.081). Continuous fitness performed better: inverted fitness AUC = 0.700 in rich media and 0.725 in minimal media, versus essentiality-fraction AUC = 0.344 and 0.403. AUC is Area Under the ROC Curve, a threshold-independent classification measure. ₂₁

The annotation-gap pipeline resolved 96 of 201 gapfilled enzymatic reaction-organism pairs (47.8%) with candidate genes; BLAST alone resolved 70 (34.8%), and the full pipeline added 13 percentage points. ₃ The Carbon Census found all 83 enrichment compounds had PubChem-resolved structures with InChIKeys, 54 linked to KEGG, and 9 callable before xanthine correction; 74/83 (89%) were organism-dark. ₁₀

The knockout validation inserted 23 gene-protein-reaction rules, but simulations produced zero wildtype growth because models required the gapfilled reactions themselves; this validation was circular. ₃ In aromatic catabolism, co-fitness assigned 16 of 23 initially Other or Unknown genes to support subsystems with medium or high confidence; ACIAD3137 and ACIAD2176 correlated with Complex I genes at r > 0.98, but the matrix used only 8 conditions and 8-dimensional growth vectors, and 11 non-core assignments relied on correlation rather than direct physical evidence. ₁₉

Essentiality and cross-project fitness

The essential-genome analysis identified 41,059 essential genes (18.6%) among 221,005 genes across 48 bacteria, with essentiality rates from 12.2% in Pedo557 to 29.7% in Magneto; 2,838,750 BBH pairs yielded 17,222 ortholog groups. ₂₂ Of these, 859 (5.0%) were universally essential, 4,799 (27.9%) variably essential, and 11,564 (67.1%) never essential. Fifteen families were essential in all 48 organisms: rpsC, rplW, rplK, rplB, rplA, rplF, rps11, rpsJ, rpsI, rpsM, groEL, pyrG, fusA, valS, and SelGGPS. ₂₂

Universally essential genes were 91.7% core, variably essential 88.9%, never-essential 81.7%, and orphan essential 49.5%; counts were 6,963, 90,844, 47,140, and 4,683. Essentiality penetrance and core fraction had weak positive correlation (rho=0.123, p=1.6e-17), while clade size did not predict essentiality rate (rho=-0.13, p=0.36). ₂₂ The 7,084 essential genes lacking detectable orthologs were 58.7% hypothetical versus 8.2% of universally essential genes. Of 8,297 hypothetical essential genes, 3,912 had orthologs and 4,385 were orphans. These lower core fractions suggest, but do not prove, strain-specific functions, recent acquisition, rapid evolution, or divergent homologs. ₂₂

ICA-based transfer from non-essential orthologs generated 1,382 family-backed function predictions (35.3% of predictable targets), including TIGR00254 signal transduction, PF00460 flagellar basal body rod, and PF00356 lactoylglutathione lyase. These are indirect predictions because lethal essential genes cannot be directly profiled by disruption. ₂₂

The essential-metabolome pilot selected 7 organisms from a 45-organism collection: 17 of 18 amino-acid biosynthesis pathways were complete in all 7, while serine biosynthesis was complete in 6 of 7. The computational DvH gap may indicate auxotrophy, but divergent or unannotated genes remain alternatives. ₂₃ The report included 305M predictions across 293K genomes in kbase-ke-pangenome, 859 universally essential families across 45 organisms, 80 pathway-completeness records, 18 amino-acid records, 62 carbon-pathway records, 7,389 raw predictions, and 8 manual organism-to-genome mappings. ₂₃

The metal atlas supports and extends this essentiality-conservation pattern at metal scale: among 2,891 ortholog groups with metal phenotypes, 1,182 were conserved across at least two organisms and 601 across at least three organisms; the broadest family, OG00128, spanned 17 organisms and nine metals. ₈ Families with metal phenotypes in more organisms tended to have higher pangenome conservation, but the simple gene-presence repertoire score failed to predict metal fitness, reinforcing that essentiality and fitness are context-dependent. ₈

The truly dark-gene study supports and qualifies the association between essentiality and unknown genes: 18.0% of truly dark genes were essential versus 13.4% of annotation-lag genes, and 34 of the top 100 candidates were essential. ₅ However, short genes are harder both to annotate and to measure by transposon insertion, and some strong phenotypes may reflect polar effects on downstream genes. ₅

The AMR analysis linked 1,352 resistance genes across 43 organisms to Fitness Browser measurements; 25 organisms qualified for per-organism tests. Under non-antibiotic conditions AMR knockouts had higher relative fitness than non-AMR knockouts in all 25 organisms, with pooled random-effects effect +0.086 [95% CI: +0.074, +0.098]. This compares AMR knockout fitness (-0.024) with a non-AMR background of approximately -0.11, not beneficial absolute growth or an absolute selection coefficient. Among AMR genes, 57.0% flipped toward greater importance under antibiotics, although class-matched validation covered 157 gene-antibiotic pairs and was non-significant (p = 0.14). ₂

Truly dark genes as a TnSeq prioritization target

The truly dark-gene census found that sequence databases often recognize genes without assigning function: 79.4% had UniRef50 links and 84.7% had database cross-references, but only 4.0% had Pfam hits and 4.6% had KEGG KOs. eggNOG-mapper provided partial signal for 43.5% of truly dark clusters; only 17.0% had COG categories, and 55.4% of those assignments were category S (function unknown). ₅ A clue matrix classified 246 genes (3.8%) with no clues, 3,867 (60.2%) with minimal sequence identifiers, 711 (11.1%) with partial Pfam/COG/eggNOG function, and 1,603 (24.9%) with phenotype-only evidence; Tier 3 and Tier 4 together contained 2,314 promising candidates. ₅

Truly dark genes were 4.2 times less likely to have cross-organism orthologs (OR = 0.236), with median ortholog breadth 1 versus 4 organisms. Only 3 of 65 dark-gene ortholog groups with cross-organism concordance data contained truly dark genes. Mean absolute GC deviation from the host-genome mean was 0.047 versus 0.038 for annotation-lag genes (d = 0.247, p = 1.3e-43); strong GC deviation affected 9.2% versus 4.0%, and 12.0% were within 2 genes of a mobile genetic element. These patterns are consistent with horizontal gene transfer, but GC deviation and mobile-element proximity do not establish acquisition history. ₅

The report contradicts a general stress-enrichment hypothesis for truly dark genes. Among genes with strong fitness phenotypes (|f| ≥ 2), stress conditions were reported at 28.7% versus 43.2% for annotation-lag genes (OR = 0.53, p < 0.001); the results section separately reports stress at 43.3% versus 54.7%, carbon-source conditions at 13.7% versus 21.7%, and motility conditions at 1.2% versus 3.2%. It instead reports enrichment in mixed-community conditions (7.5% versus 0%), iron conditions (0.7% versus 0%), nutrient time-series conditions, and rich media. ₅ The possible interpretation that these genes encode novel metabolic or community-interaction functions rather than stress responses remains a hypothesis. ₅

A multi-criteria score combining fitness importance, annotation clues, ortholog breadth, genomic context, and experimental tractability ranked all 6,427 truly dark genes on a maximum scale of 12 and identified 100 candidates scoring 8–10 across 19 organisms. PV4/5210953 scored 10, with a motility phenotype of |f| = 5.5, an operon association with TatC, and ICA module M016. ANA3/7026383 scored 9, with a nitrogen-source phenotype of |f| = 8.6, an ABC-transporter operon association, and ICA module M018. DvH/206658 scored 9, with a stress phenotype of |f| = 5.4 and an eggNOG “trehalose synthase” signal despite its hypothetical annotation. Methanococcus_S2/MMP_RS06570 scored 9, with 8 annotation clues including DUF190 and COG-T signal-transduction evidence, plus an operon association with fluoride-efflux transporter CrcB. ₅

The top 100 span 19 organisms; Methanococcus_S2 contributes 29, DvH 13, Methanococcus_JJ 13, and MR-1 8. Of these candidates, 34 are essential, 53 are in operons, and 30 are in ICA fitness modules. Approximately 2,841 additional truly dark genes may exist among the 17,479 unlinked dark genes at the 16.3% truly-dark rate, including 2,208 with strong fitness phenotypes; the ranked list covers approximately 69% of the estimated total truly dark gene population. ₅

Limitations and directions

TnSeq essentiality is an upper-bound estimate: absent insertions may reflect small size, AT-rich or low-complexity sequence, or scaffold edges. RB-TnSeq measures essentiality under specific library-construction and medium conditions, typically rich media; essential genes were shorter than non-essential genes, with median lengths of 675 bp and 885 bp, and 17.8% of essential genes were shorter than 300 bp. _{13 22} The fitness-conservation gradient was robust but weak, and singleton neutrality may reflect poor coverage. ₁₄

The respiratory-chain analysis refines these limitations: NDH-2 has no growth data, the stoichiometric explanation uses theoretical rather than measured fluxes, the cross-species comparison included only 4 organisms lacking NDH-2, and annotation-based searches may miss orthologs. Text matching produced false-positive NDH-2 calls; KO-based identification using K03885 is more reliable. ACIAD3522 may not be a true respiratory NADH dehydrogenase. ₁₈

The pathway-capability comparison is biased toward laboratory conditions; its intermediate zone makes latent fractions threshold-sensitive; SEED-proxy mapping can produce false positives; only 48 organisms were fitness-tested among 293,000 genomes with pathway predictions; and pathway-level conservation cannot detect partial erosion or progressive loss within otherwise complete pathways. ₇ GapMind covered 80 pathways—18 amino-acid biosynthesis and 62 carbon-source utilization—but not cofactor biosynthesis, lipid metabolism, or secondary metabolism. KEGG-based mapping can miss genes lacking KEGG annotations or carrying incorrect annotations. ₆

Metabolic ecotype counts depend on hierarchical clustering with a fixed 50% maximum-distance cut; the median of 4 should therefore be treated as an order-of-magnitude estimate. Correlations were controlled for genome count and checked within taxonomic groups, but full phylogenetic independent contrasts were not computed. The related ecotype_analysis project found that phylogeny dominates gene content in 60.5% of species. Species with more sequenced genomes are more likely to show pathway variation and more ecotypes; partial correlations reduce but do not eliminate this sampling bias. ₆ The proposed AlphaEarth environmental niche-breadth analysis was not executed because AlphaEarth embeddings covered 28% of genomes, or 83K/293K. ₆

The module-conservation analysis used |Pearson r| >= 0.3, a maximum of 50 genes per module, and convenient rather than biologically motivated >90% core and <50% core boundaries. Essential genes are excluded from all modules because they lack transposon insertions; future analyses should incorporate orthology- or CRISPRi-based evidence and test whether the 38 accessory module families show direct transfer or niche associations. ₁₆

Metal-atlas coverage was uneven: cobalt and nickel were tested in 27 organisms but uranium, chromium, mercury, cadmium, selenium, and manganese in only one or two organisms; concentrations were not normalized by dose or tolerance threshold, with nickel ranging from 0.01-2.0 mM; and multiple Pseudomonas fluorescens strains could inflate apparent conservation. Excluding four duplicate FW300 strains left the core-enrichment result robust (OR=2.065, p=5.9e-141, versus OR=2.083, p=4.3e-162 with all organisms). The broad metal-important definition, fit < -1 OR n_sick ≥ 1, captures approximately 3.3% of genes and includes general stress genes. ₈

The metal-specificity analysis had 40.7% gene attrition: locusId-format mismatches excluded ANA3, Dino, Keio, MR1, Miya, PV4, and SB2B; the 24 included organisms accounted for 7,609 of 12,838 metal-important records (59.3%), while 5,229 records were excluded. The 5% sick-rate threshold was arbitrary, although results were qualitatively stable across 1–20%, and validation against Fitness Browser specificphenotype annotations was not performed. ₉

The new truly dark-gene analysis adds several limitations: 17,479 dark genes lacked pangenome linkage; Bakta may produce false-negative functional calls; BBH orthology covered only 32 of 48 Fitness Browser organisms; GC deviation is an imperfect HGT proxy; short genes confound annotation and fitness; and polar effects may generate some strong phenotypes. ₅ Detailed coverage among 5,870 unique truly dark clusters included Pfam hits for 235 clusters (4.0%), eggNOG annotations for 2,551 clusters (43.5%), and orthology for 1,885 genes (29.3%) across 48 Fitness Browser organisms. ₅

Other limitations include uneven taxonomic coverage: 37/48 organisms in functional-dark-matter analysis were Pseudomonadota, and its top 500 candidates included 417 Gammaproteobacteria and 0 Archaea, Actinobacteria, or Epsilonproteobacteria. ₄ The DvH analysis was limited to one organism; 678 essential genes lacked fitness data; field/lab sets were small (n=50-52); field and lab effects correlated at r ~ 0.7; and short genes were confounded with fitness quality and core status. ₁₅

BBH orthology can miss paralogs, fusions, and distant homologs, while connected components can over-merge through transitive links. Ten organisms were excluded from conservation analysis because they had <90% DIAMOND coverage; Dyella79 had a 0% join rate between N515DRAFT_* and ABZR86_RS*; Cola, Kang, Magneto, and SB2B were unmatched because their species had too few GTDB genomes; E. coli was excluded because its clade had too many genomes; and Keio mapped to s__Escherichia_coli_E at only 26.1% coverage. _{22 13}

NaCl supplies both Na⁺ and Cl⁻ and produces osmotic effects, so it is not a pure chloride control; the CuCl₂–CuSO₄ comparison in psRCH2 was confounded by anaerobic versus aerobic growth. CuSO₄ had higher NaCl correlation than CuCl₂, r=0.450 versus r=0.212, despite supplying no chloride. Approximately 14.3% of protein-coding genes classified as putative essential lacked insertions and were absent from both NaCl and metal-fitness data; these genes were 82% core. ₁₁

The fitness-modules analysis was weaker with fewer experiments; a 40% component cap may omit modules, PFam associations can overcount functions, and module families represent conserved co-regulation rather than identical molecular functions. The FW300 comparison tested only 21/58 metabolites (36%) outside WoM, only 3 had four-way coverage, BacDive counts ranged from 1 to 51 strains, and excluding two approximate Fitness Browser matches reduced the set from 21 to 19 and mean concordance from 0.937 to 0.930. The planned mapping of fitness genes to GapMind steps was deferred. _{17 24}

Future work should apply DvH classification to other ENIGMA organisms; replace binary core/auxiliary status with quantitative gene-cluster prevalence; link CORAL community composition to geochemistry; characterize acquisition history of accessory resistance genes; use continuous fitness scores in prediction models; calibrate pathway-dependency thresholds against known essentials; directly map essential genes to GapMind steps; expand taxonomic coverage; test pathway variation and ecotype associations with environmental metadata; and evaluate the unexecuted AlphaEarth niche-breadth analysis on its available 83K/293K genomes. _{15 6}

The metal atlas additionally recommends regulatory or expression-based prediction, concentration-relative-to-MIC normalization, phylogenetic independent contrasts, enrichment-based pangenome scoring, and functional annotation of its 149 candidates using PaperBLAST, InterPro, and structural prediction. ₈ Priorities also include CRISPRi testing of module-transfer predictions; matched RB-TnSeq tests of CuCl₂/CuSO₄, ZnCl₂/ZnSO₄, and CoCl₂/CoSO₄ with choline chloride or KCl controls; formal COG, KEGG, and PFAM enrichment tests; and experimental testing of Module-ICA predictions against ortholog assignments. _{22 23 11 17}

For the 100 highest-ranked truly dark genes, proposed work includes AlphaFold2 or ESMFold structure prediction followed by Foldseek searches; growth assays under predicted conditions, including motility testing for PV4/5210953 and nitrogen limitation for ANA3/7026383; mobile-CRISPRi; extension of pangenome linkage to the 17,479 unlinked genes; analysis of contiguous dark islands as possible prophage remnants or acquired metabolic cassettes; and Methanococcus-focused cofitness analysis using signals reported as r > 0.97. ₅ The respiratory-chain model adds construction of an NDH-2 deletion mutant, NADH/NAD⁺ measurements on each carbon source, expanded K03885 and K00330–K00343 searches across 27K species, ACIAD3522 characterization, pangenome KO co-occurrence analysis, and reanalysis of quinate-versus-succinate proteomics. ₁₈

Related wiki pages

• gene-essentiality — conserved, variable, orphan, condition-specific, and truly dark-gene essentiality.
• genomic-under-representation — experimentally prioritized dark-gene systems and residual annotation gaps.
• condition-specific-fitness — condition-dependent fitness and stress responses.
• metal-cross-resistance — cross-metal fitness, core stress functions, and specialized resistance. ₈
• metabolic-model-gapfilling — fitness-guided metabolic annotation and GapMind.
• pangenome-integration — conservation and pangenome integration.
• gene-function-acquisition-depth — ortholog breadth, accessory localization, and acquisition hypotheses for dark genes.
• ecotype-environment-gene-content — metabolic ecotypes and environment-associated gene content.
• multi-omics-integration — TnSeq within integrated data modalities and compound evidence.
• environmental-resistome — AMR fitness costs and antibiotic-dependent importance.
• cofitness-network-architecture — cofitness networks, ICA modules, and fitness architecture.
• flux-balance-analysis — metabolic-flux modelling compared with TnSeq.
• Truly Dark Genes — What Remains Unknown After Modern Annotation? — residual unknown genes after Bakta reannotation and a ranked experimental-prioritization framework.
• Metabolic Capability vs Dependency — metabolic capability versus experimentally measured dependency.
• Metabolic Capability vs Metabolic Dependency — earlier metabolic capability versus dependency analysis.
• Pan-bacterial Fitness Modules via Independent Component Analysis — pan-bacterial ICA fitness modules and cross-organism process-level predictions.
• Fitness Modules × Pangenome Conservation — fitness-module genes and pangenome conservation.
• Fitness Effects vs Conservation — Quantitative Analysis — fitness effects and pangenome conservation.
• Field vs Lab Gene Importance in *Desulfovibrio vulgaris* Hildenborough — field-versus-lab gene importance in DvH. ₁₅
• Acinetobacter baylyi ADP1 Data Explorer — ADP1 Data Explorer.
• ADP1 Deletion Collection Phenotype Analysis — deletion-collection analysis. ₂₀
• Triple Essentiality Concordance Analysis for *Acinetobacter baylyi* ADP1 — triple-essentiality concordance. ₂₁
• Fitness Cost of Antimicrobial Resistance Genes — AMR fitness-cost analysis. ₂
• Annotation-Gap Discovery via Phenotype-Fitness-Pangenome-Gapfilling Integration — annotation-gap analysis. ₃
• Aromatic Catabolism Support Network in ADP1 — aromatic-catabolism analysis. ₁₉
• Regulatory and proteomic architecture of Δfur-permitted lipid A loss in *Caulobacter crescentus* — Caulobacter analysis. ₂₅
• Gene Conservation, Fitness, and the Architecture of Bacterial Genomes — conservation-fitness synthesis.
• Conservation vs Fitness — Linking FB Genes to Pangenome Clusters — Fitness Browser–pangenome integration.
• Core Gene Paradox — Why Are Core Genes More Burdensome? — core-gene burden and trade-offs. ₂₆
• Counter Ion Effects on Metal Fitness Measurements — counter-ion and metal-fitness analysis. ₁₁
• ENIGMA Carbon Census — Tiered Knowledge Census of 83 Enrichment Compounds — ENIGMA Carbon Census. ₁₀
• The Pan-Bacterial Essential Genome — pan-bacterial essential-genome analysis. ₂₂
• Essential Metabolome: GapMind Pathway Analysis Across Essential-Gene Organisms — GapMind pathway-completeness pilot. ₂₃
• Functional Dark Matter — Experimentally Prioritized Novel Genetic Systems — experimentally prioritized dark-gene systems and organism-level TnSeq campaigns. ₄
• Metabolic Consistency of Pseudomonas FW300-N2E3 Across Four KBase Data Lakehouse Databases — cross-database metabolic consistency of Pseudomonas FW300-N2E3. ₂₄
• Metal-Specific vs General Stress Genes — metal-specific versus general-stress RB-TnSeq fitness determinants. ₉
• Condition-Specific Respiratory Chain Wiring in ADP1 — condition-specific respiratory-chain wiring in ADP1. ₁₈