Condition-Specific Mutant Fitness

Condition-specific mutant fitness measures how the growth effect of disrupting a gene changes across environmental or nutrient conditions. The corpus spans Acinetobacter baylyi ADP1 mutant growth across 8 carbon sources, cross-organism Fitness Browser comparisons, metal-fitness assays, pathway-completeness analyses, and community-scale metabolomics. _1,2 Across 194,216 protein-coding genes from 43 bacteria, fitness importance and pangenome conservation form a modest gradient rather than a binary distinction: essential genes were 82% core, often-sick genes 78%, mixed genes 70%, sometimes-sick genes 72%, always-neutral genes 66%, and sometimes-beneficial genes 70%; for strongest effects, core fractions ranged from 82.2% among essential genes to 66.4% among genes with min_fit from -1 to 0, and fitness breadth had Spearman rho=0.086, p=8.1e-230. _3,4

The plant-ecotype analysis refines this framing by showing that genomic capability, ecological association, and measured dependence are separable. Across 293,059 GTDB r214 genomes, 7,995 genomes from 1,136 species had plant-associated metadata, but compartment effects in a refined 17-marker functional panel explained only 0.060 of variance by db-RDA location effects and 0.071 by PERMANOVA, despite p = 0.001 for both tests. ₅ Thus environmental labels can predict small shifts in genomic capability without establishing condition-specific activity or mutant fitness. The detailed analysis is available in Plant Microbiome Ecotypes — Compartment-Specific Functional Guilds and Their Genetic Architecture.

The SNIPE analysis supports and extends this separation. SNIPE-bearing and non-SNIPE-bearing species differed significantly in 22 of 64 AlphaEarth environmental-embedding dimensions, with the largest effect in dimension A19 (Cohen’s d = 0.26, p = 5.0e-14), but effect sizes ranged only from |d| = 0.11–0.26. AlphaEarth coverage was 28.4% of genomes and biased toward environmental isolates with geographic metadata, so the result indicates a modest environmental association rather than a defined habitat mechanism or measured fitness dependence. ₆ Environmental association can therefore be statistically reproducible while remaining weakly informative about when a capability becomes fitness-relevant. ₆

The respiratory-chain analysis supports and mechanistically specifies this separation in ADP1. Across carbon sources, the same respiratory inventory is associated with qualitatively different measured requirements: quinate specifically required Complex I, lactate specifically required cytochrome bo3, acetate required Complex I, cytochrome bo3, ACIAD3522, and additional components, glucose had no specifically required respiratory component, and urea was broadly demanding. ₇ Quinate generated 4 NADH, or 0.57 NADH per carbon, whereas glucose generated 9 NADH, or 1.50 NADH per carbon; concentrated pathway-specific NADH flux, rather than total yield, was proposed to determine which encoded respiratory capability becomes fitness-dependent. This stoichiometric explanation is theoretical rather than a measured flux distribution. ₇

Conserved functions, specialized functions, and measured dependence

The metal-fitness atlas refines and strengthens this context-dependent conservation model. Across 559 metal experiments in 31 organisms and 16 metals, 383,349 gene × metal fitness records yielded 12,838 broad metal-important genes, or 3.3%, and 5,667 strict metal-important genes, or 1.5%. Using fit < -1 and |t| > 4, iron and essential-metal limitation produced the highest fractions of important genes (12.3% for Fe and 11.2% for Mo/W), whereas toxic metals clustered around 2.7-4.4%. Organism sensitivity was uneven: DvH (Desulfovibrio vulgaris Hildenborough) had 1,366 metal-important genes, 49.8% of its genome across 13 metals, and the cyanobacterium synechococcus-elongatus (SynE) had 33.6% of genes important across just 2 metals. Across 22 organisms and 14 metals, metal-important genes were 87.4% core versus 76.9% for baseline genes (OR=2.08, p=4.3e-162), with 21 of 22 organisms showing positive core enrichment. ₈ This contrasts with the prior condition-specific heavy-metal result of 71.2% core because broad metal definitions capture general cellular functions vulnerable to metal disruption, not only accessory resistance genes. ₈

Essential-metal responses were more core-enriched than toxic-metal responses, with mean core-fraction delta +0.148 versus +0.081, one-sided Mann–Whitney U=39 and p=0.015. The strongest single-metal enrichment was manganese, with delta +0.198 and all 30 important genes core. The atlas therefore supports a two-tier model in which conserved general-stress functions coexist with specialized efflux, sequestration, and detoxification functions. ₈ Applying an explicit non-metal stress control refines that model: among 7,609 metal-important records from 24 organisms, 4,177 (54.9%) were metal-specific, 2,888 (38.0%) general sick, and 544 (7.2%) metal+stress. Metal-specificity required a significant metal defect and a sick rate below 5% across 5,945 non-metal experiments; the fraction was approximately 41% at a 2% threshold and approximately 67% at a 10% threshold. ₉

Metal-specific genes remained strongly core-associated: pooled core fraction was 84.8% (2,969/3,500), organism-mean core fraction 88.0%, and mean delta versus baseline +6.9%; general sick genes were 90.2% core pooled (2,183/2,420), organism mean 90.2%, and mean delta +9.0%, with a cross-category Cochran–Mantel–Haenszel p=0.011. Metal-specific genes had metal-resistance keyword matches of 12.2% versus 7.8% for general sick genes (Fisher exact OR=1.64, p=2.4e-8). Candidate families included UCP030820, YebC, and DUF1043/YhcB, which were metal-specific in 2/3, 7/12, and 3/6 organisms respectively; the proposed YebC translation-bottleneck mechanism remains a hypothesis. Novel families were less often predominantly metal-specific than annotated families (45.6% versus 58.2%; Fisher OR=0.60, p=0.003). ₉

The metabolic-capability analyses support and sharpen the distinction between encoded capability and measured dependence. Across 1,695 complete pathway-organism pairs from 48 organisms, 267 (15.8%) were latent capabilities, 547 (32.3%) intermediate, and 881 (51.9%) active dependencies. Carbon-source pathways were latent in 217 of 892 cases (24.3%), compared with 48 of 735 amino-acid biosynthesis cases (6.5%), making carbon pathways 3.7× more likely to be latent. ₁₀ In the more directly mapped 7-organism analysis, 35.4% of 161 pathway-organism pairs were Active Dependencies, 41.0% Latent Capabilities, 14.9% Incomplete but Important, and 8.7% Missing. The mapping joined Fitness Browser-native KEGG annotations to GapMind pathways, combining pathway completeness with gene-level fitness evidence. All 66 aggregate Latent Capability pairs became fitness-important under at least one condition type, especially nitrogen limitation, stress, or carbon limitation. This supports condition dependence while qualifying the earlier interpretation of latent capability as broad laboratory dispensability, because the condition-specific result uses a median-based importance threshold. ₁₁ The central discoveries digest states the same conclusion more broadly: all latent capabilities become fitness-important in condition-specific analyses involving nitrogen limitation, stress or carbon-source shifts. It reads this as evidence that laboratory fitness underestimates natural selection on metabolic genes. That digest passage gives no pair count and does not discuss how the importance threshold was set, so it restates rather than independently supports the 7-organism result. ₁₂

The 48-organism classification qualifies these latent-capability counts because it is itself threshold- and condition-dependent. Latent pathways were genomically complete but showed no detectable fitness importance under tested conditions (mean |t-score| < 1.0, < 5% essential genes). Pathway category strongly predicted dependency class (χ²=163.6, df=4, p=2.5×10⁻³⁴). By category (active / intermediate / latent / total): amino-acid biosynthesis 467 (63.5%) / 220 (29.9%) / 48 (6.5%) / 735; carbon-source utilization 355 (39.8%) / 320 (35.9%) / 217 (24.3%) / 892; “Other” 59 (86.8%) / 7 (10.3%) / 2 (2.9%) / 68; total 881 (51.9%) / 547 (32.3%) / 267 (15.8%) / 1,695. Latent fractions ranged across organisms from 0–31.6% (mean 15.0%), highest in Pseudomonas syringae strains and klebsiella-michiganensis. Varying the active and latent thresholds by ±25% across 16 combinations gave latent fractions from 4.7% to 21.1% (SD = 5.9 percentage points), so the 15.8% default estimate is moderately threshold-sensitive, although the report states that the qualitative conclusion holds across the tested range. Laboratory media may also omit conditions in which pathways matter, potentially inflating the latent fraction. The report turns the 3.7× carbon-versus-amino-acid contrast into a prediction, that organisms in nutrient-rich environments should disproportionately lose carbon-pathway genes over evolutionary time, but this gene-loss prediction was not tested. ₁₀

The 7-organism analysis refines the same point with a standard-versus-stratified contrast. Under standard conditions, genes in 66/161 genomically complete pathway pairs showed no significant fitness defects; condition-type stratification by nitrogen source, carbon source, stress, and other conditions then made all 66 pairs fitness-important under at least one condition type, with nitrogen limitation, stress, and carbon limitation the most frequent triggers (figures fig5_condition_type_shifts.png and tier1_condition_type_shifts.png). The report itself warns of threshold circularity: the median-based importance threshold by construction splits pathways approximately 50/50 into important and not-important categories in each subset, so with enough condition types some pathways will inevitably cross it. The reclassification therefore supports the statement that a complete pathway in a genome does not guarantee dependence on that pathway under any given condition, but given this caveat it does not by itself show that every latent pathway is under condition-specific selection. ₁₁

Active Dependencies had mean core gene completeness of 0.986 versus 0.975 for Latent Capabilities. This small difference refines the gene-level conservation model because the validation set comprised well-studied model organisms with near-complete, well-annotated core genomes, and only 7 of 48 Fitness Browser organisms had matching GapMind data. ₁₁ Conversely, latent complete pathways had mean conservation 0.869 versus 0.829 for active dependencies, with the active-greater-than-latent comparison not significant (p = 0.94). ₁₀ Thus pathway completeness and gene-level essentiality should not be assumed to have the same conservation relationship.

The pathway analysis extends this distinction to pangenome architecture. Across 2,810 GTDB species with at least 10 genomes, variable pathway count correlated with pangenome openness at raw Spearman rho=0.327, p=7.2e-71, and at partial rho=0.530, p=2.83e-203 after controlling for genome count. The positive signal occurred in 13 of 18 genera and was significant in 5 of 18: Clostridium, Eubacterium, Mesorhizobium, Pseudomonas, and Streptomyces; strongest within-genus correlations were rho=0.534 for Clostridium and rho=0.506 for Eubacterium. ₁₁ This provides a pathway-level correlate of genome fluidity, while broader pangenome analyses reported null relationships between openness and broader environment or phylogeny effect sizes. ₁₁

Core-only versus all-gene GapMind completeness supports and specifies a role for accessory genes in amino-acid biosynthesis: exact all-gene minus core-only gaps were 0.146 for leucine and valine, 0.141 for arginine, 0.140 for lysine, and 0.140 for threonine. This is consistent with Black Queen dynamics and community sharing of biosynthetic capacity, but does not demonstrate metabolite exchange. ₁₁ At community scale, GapMind completeness was weighted across 220 samples, 27,690 GTDB species, and 80 pathways; the analysis included 305M GapMind records and NMDC multi-omics data. In 13 testable amino-acid pathways, 11 of 13 (85%) had negative Spearman correlations between community pathway completeness and ambient amino-acid intensity; a binomial sign test gave p = 0.011. Leucine biosynthesis had r = −0.390, q = 0.022, n = 62, and arginine biosynthesis r = −0.297, q = 0.049, n = 80; these were the two pathways significant after Benjamini–Hochberg false-discovery-rate correction. ₂ This supports a weak community-scale Black Queen signal, while not establishing expression, causality, or gene-level essentiality. The predictor was the fraction of community taxa with a GapMind score of 5; a score of at least 4 was retained only for sensitivity comparisons. The full analysis-ready matrix contained 174 samples, but all 33 Freshwater samples lacked paired metabolomics; 125/131 H1 samples came from one NMDC study. Abiotic pH, temperature, and total organic carbon were unavailable, GapMind measured genomic potential rather than expression, and string-based metabolite matching required correction of an isoleucine/leucine collision. ₂

Evidence from ADP1 carbon-source fitness

Pairwise mutant-fitness correlations across 8 carbon sources had a mean of 0.44. Across all 28 condition pairs, deletion phenotypes had median r = 0.25 and maximum r = 0.58 for acetate–butanediol; five principal components captured 82% of variance, with no component explaining more than 37%. _1,13 These results support approximately 5 partly independent phenotypic dimensions rather than a simple demanding-versus-robust split.

The strongest correlations were butanediol–acetate and butanediol–lactate, at r = 0.58 and r = 0.53. Fitness on urea was nearly uncorrelated with quinate, at r = 0.11, and weakly correlated with other conditions, at r = 0.12-0.28. Urea was the most demanding condition, with 97.9% of genes showing severe defects at growth ratio < 0.5; pathway analysis linked urea specificity to the urease complex genes. _1,13

Across the 8 carbon sources, 333 of 478 genes (70%) showed condition-specific growth defects, 10 (2%) showed defects across all 8, and 135 (28%) showed no defect on any condition; mean pairwise defect correlation was 0.38, ranging from -0.03 to 1.0. Continuous fitness predicted knockout lethality better than essentiality fraction, with AUC = 0.700 in rich media and AUC = 0.725 in minimal media, versus 0.344 and 0.403 for essentiality fraction. ₁₄

These counts depend on how defects were defined. The original study defined a growth defect separately within each condition as a rate in the bottom 25% (Q25) of that condition’s growth-rate distribution. It then flagged a gene if it had a defect on at least one of the 8 conditions. This “any defect” aggregation inflates background prevalence by design: 72% of the 478 genes had a defect on ≥1 condition, versus 90% expected if conditions were independent. The report attributes the shortfall to positive inter-condition correlation (mean inter-condition correlation 0.38). The 70% condition-specific figure therefore rests on a high-background binary flag and should not be read as an independent count of condition-specific genes. ₁₄ Essentiality calls also shift with the growth medium: 499 genes were called essential on minimal media versus 346 on LB. The report attributes this to the additional biosynthetic burden of minimal media. ₁

The respiratory-chain study supports and explains this architecture. ADP1 has 62 respiratory-chain genes across 8 subsystems, but quinate, acetate, lactate, glucose, and urea select different required components rather than merely different magnitudes of one common demand. Complex I growth ratios were 0.37 on quinate, 1.44 on glucose, and 0.49 on acetate; ACIAD3522 had growth ratios of 1.39 on quinate, 1.39 on glucose, and 0.013 on acetate. ACIAD3522 is therefore an acetate-specific fitness signal, although its annotation as an NADH-FMN oxidoreductase may represent another metabolic function rather than a respiratory NADH dehydrogenase. ₇ The aromatic-catabolism network added a 51-gene support network around the beta-ketoadipate-pathway; co-fitness assigned 44/51 genes (86%) to aromatic degradation, Complex I / NADH dehydrogenase, iron acquisition, or pqq-biosynthesis. ₁₅ Quinate produced 0.57 NADH per carbon versus 1.50 for glucose, yet Complex I was more essential; the proposed concentrated NADH burst remains theoretical. _12,7

The SNIPE/ManXYZ case supports the same principle through a non-metabolic defense trade-off. In 168 RB-TnSeq (random barcode transposon sequencing) experiments per ManXYZ gene in E. coli K-12 (Keio collection; Price et al. 2018), manX, manY, and manZ showed average fitness values of -0.249, -0.171, and -0.082, worst fitness values of -3.93, -3.82, and -4.14, fitness < -1 in 6, 10, and 7 conditions, and fitness < -2 in 4 conditions each, with strongest defects on D-glucosamine (−3.79, −2.79, and −3.63) and D-mannose (−2.75, −3.00, and −2.74); smaller listed defects occurred on D-trehalose (-1.27 and -1.49 for manX and manY, with no value listed for manZ) and sodium chlorite (-1.47, -1.32, and -1.08). Cofitness correlations were 0.851 for manX↔manZ, 0.705 for manX↔manY, and 0.725 for manY↔manZ. The data contradicted a fructose-specific ManX annotation: on D-fructose, manX, manY, and manZ were -0.22, +0.15, and -0.07, while fruA was -1.44; on mannose and glucosamine, fruA was -0.06 and -1.02, respectively. Retention of a phage-entry transporter can therefore be metabolically important in the same conditions in which its loss is costly, while SNIPE-mediated phage resistance remains a hypothesis. ₆

Core genes were 1.78x more likely to have strong condition-specific phenotypes and 1.29x more likely to be trade-offs important and burdensome in different conditions. The cross-organism result found 77.3% of strongly specific-phenotype genes were core versus 70.3% without such phenotypes (OR=1.78, p=1.8e-97); 24.4% of core genes showed positive fitness versus 19.9% of accessory genes. _3,4 The central digest reports the same 77.3% versus 70.3% contrast from the Fitness Browser specificphenotype table. It argues that this contradicts the intuition that condition-specific fitness marks niche-specific accessory genes. The trade-off report likewise states that genes with strong condition-specific effects are more likely core, though it gives no effect size in that passage. _12,16 The fitness-effects report itself notes that the 77.3% versus 70.3% result contradicts the naive expectation that condition-specific genes would be accessory, and suggests instead that core genes are more likely to have measurable condition-specific effects, perhaps because they participate in well-characterized pathways; this too is offered as a possible explanation rather than a demonstrated one. Overall, 4,450 genes (2.7%) fit the ephemeral-niche pattern, neutral overall but critical in one condition; this pattern was also more common in core genes (3.0%) than in auxiliary (1.7%) or singleton (1.6%) genes. The report proposes, but does not demonstrate, that core genes may simply have more detectable conditional effects because they participate in more pathways, so this enrichment may partly reflect detectability rather than niche function. ₄ This contradicts the expectation that accessory genes are generally costly to carry. Core-gene excesses were +6.2 percentage points in Protein Metabolism, +7.8 in Motility, and +12.9 in RNA Metabolism, whereas non-core Cell Wall genes were more burdensome by -14.1 percentage points. There were 25,271 true trade-off genes (17.8% of genes examined), defined as fitness below -1 in some conditions and above 1 in others; they were 1.29 times more likely to be core, with odds ratio 1.29 and p=1.2e-44. ₁₆

A distinct costly-accessory subset qualifies this general core pattern. Costly+dispensable genes were 7.45x more likely than costly+conserved genes to contain mobile-element keywords (OR=7.45, p=4.6e-71); phage, prophage, transposable-element, and plasmid keywords were 11.7x enriched (FDR=1.3e-17), and virulence was 26.7x enriched (FDR=5.6e-14). They had lower annotation, more orphan genes, narrower ortholog breadth, more singletons, and shorter median length: SEED 50.8% versus 74.9%, KEGG 20.0% versus 42.7%, orphan status 44.5% versus 13.1%, median ortholog breadth 15 versus 31 organisms, singleton fraction 24.2% versus none, and median length 615 bp versus 765 bp. ₁₇

Functional dark matter and condition-specific evidence

The new truly-dark-gene census extends and sharpens the functional-dark-matter interpretation. Of 57,011 Fitness Browser dark genes, 33,105 were reclassified by Bakta v1.12.0, 6,427 remained hypothetical in both pipelines, and 17,479 lacked pangenome links and could not be assessed. Among linked genes, truly dark genes had median length 121 amino acids versus 194 for annotation-lag genes, core-genome fraction 43.1% versus 72.7%, essential fraction 18.0% versus 13.4%, mean GC content 0.542 versus 0.584, ortholog presence 29.3% versus 63.7%, and median ortholog breadth 1 versus 4 organisms. ₁₈ These differences support persistent annotation resistance and biological novelty, but short genes are harder both to annotate and to measure with transposon insertion, so length may partly explain the associated effects. ₁₈

Sequence databases recognize many truly dark genes without assigning function: 79.4% have UniRef50 links and 84.7% have database cross-references, but only 4.0% have Pfam hits and 4.6% have KEGG KOs. eggNOG-mapper provides partial signal for 43.5% of truly dark clusters; only 17.0% have COG categories, and 55.4% of COG assignments are category S, function unknown. A clue matrix found 246 genes (3.8%) with no clues, 3,867 (60.2%) with minimal sequence identifiers, 711 (11.1%) with partial functional clues, and 1,603 (24.9%) with phenotype-only clues; Tier 3 and Tier 4 together contain 2,314 promising candidates. ₁₈

The truly-dark result supports the existing link between unknown genes, accessory architecture, and mobile context: only 3 of 65 dark-gene ortholog groups with cross-organism concordance data contained truly dark genes; mean absolute GC deviation was 0.047 versus 0.038 for annotation-lag genes, strong GC deviation affected 9.2% versus 4.0%, and 12.0% of truly dark genes were within 2 genes of a mobile genetic element. These patterns are consistent with horizontal gene transfer, but GC deviation and mobile proximity do not establish transfer or function. ₁₈

The report contradicts a simple stress-enrichment hypothesis for unknown genes. Among truly dark genes with strong fitness phenotypes (|f| ≥ 2), one comparison reported stress conditions at 28.7% versus 43.2% for annotation-lag genes (OR = 0.53, p < 0.001); the results section separately reported stress at 43.3% versus 54.7%, carbon-source conditions at 13.7% versus 21.7%, and motility at 1.2% versus 3.2%. It reported enrichment in mixed-community conditions (7.5% versus 0%), iron conditions (0.7% versus 0%), nutrient time-series conditions, and rich media. These observations suggest, but do not establish, inter-species interaction functions, novel metabolic capabilities, or community-specific behaviors poorly represented in single-organism annotation databases. ₁₈

The functional-dark-matter analysis qualifies how far condition matching can go for unknown genes. For 2,038 T1 Void or T2 Twilight genes across all 42 organisms, detectable fitness phenotypes came with no converging evidence for a specific condition; the report recommends a broad RB-TnSeq phenotypic screen across a diverse condition panel (the standard Fitness Browser protocol of ~50-100 conditions), and 109 such genes are in escherichia-coli K-12 alone. The same report identified stress-responsive dark genes in Pseudomonas species and cites Vaccaro et al. (2016), in which fitness profiling of pseudomonas-stutzeri-rch2 identified novel metal-resistance genes among hypotheticals, as a precedent. These Pseudomonas observations concern individual genes, whereas the truly-dark stress depletion above is a class-level comparison among strong-phenotype genes, so the two are not directly contradictory. _19,18

Dark genes also cluster in genomic neighborhoods and fitness modules: among 4,394 truly dark genes in organisms with integrated-condition data, 41% of neighboring genes were hypothetical, 12.0% were within 2 genes of a mobile 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, with candidate links to phage integrases, metal transporters, chemotaxis systems, and iron regulation. The report ranked 100 candidates across 19 organisms; 34 were essential, 53 were in operons, and 30 were in ICA modules. ₁₈ This supports the use of cofitness and genomic context as hypothesis-generating evidence, not as direct functional proof. Detailed results are available in Truly Dark Genes — What Remains Unknown After Modern Annotation?.

Universally essential genes were 91.7% core, variably essential genes 88.9%, never-essential genes 81.7%, and orphan essential genes 49.5%; essentiality penetrance and core fraction had weak positive correlation (rho=0.123, p=1.6e-17). Across 48 organisms, 859 of 17,222 ortholog families (5.0%) were universally essential, 4,799 (27.9%) variably essential, and 11,564 (67.1%) never essential; 7,084 orphan essential genes were 58.7% hypothetical versus 8.2% of universally essential genes. ₂₀ Truly dark genes add a direct comparison: their essential fraction was 18.0% versus 13.4% for annotation-lag genes, and 34 of the top 100 candidates were essential. This supports the view that annotation uncertainty does not imply dispensability, while not proving that individual dark-gene fitness effects are autonomous because polar effects can occur. ₁₈

Single-condition essentiality calls qualify all of these conservation comparisons. RB-TnSeq (random barcode transposon sequencing) essentiality is defined under the library-construction conditions, typically rich media, so genes essential only under stress are missed. _21,20 In rich media, nutrient supplementation can make biosynthetic genes appear non-essential. These data also cannot distinguish essential for biosynthesis from essential for viability, so pathway completeness should not be equated with viability essentiality. ₂₂ Single-condition designs similarly limit regulatory inference. In the Caulobacter study, transcriptomes were measured only in routine PYE rich medium, so the Fur signal reflects constitutive Δfur derepression rather than a measured iron-limitation response. ₂₃

Condition-dependent architecture and modeling limits

The complete growth matrix of the ADP1 deletion collection comprises 2,034 genes with mutant/wild-type growth ratios measured on all 8 carbon sources. Ratios below 1.0 indicate growth defects. Ratios above 1.0 indicate no defect, but can reflect experimental normalization rather than a true growth advantage. Mean ratios span a 3.3-fold range, from 0.41 on urea to 1.36 on quinate. In this matrix, 625 genes (31% of the complete matrix) had condition-specificity score ≥ 1.0, meaning their growth importance was concentrated on one carbon source. The conditions partition into tiers based on the fraction of genes with defects at ratio < 0.8; the report describes this tier structure as consistent across multiple thresholds:

• Demanding: urea, acetate, and butanediol, with mean growth ratios of 0.41–0.65 and defects at ratio < 0.8 for 95–100% of genes.
• Moderate: asparagine and lactate, with mean growth ratios of 0.80–0.82 and defects at ratio < 0.8 for 37–45% of genes.
• Robust: glucarate, glucose, and quinate, with mean growth ratios of 1.25–1.36 and defects at ratio < 0.8 for 0.5–2.4% of genes. Urea had severe defects at ratio < 0.5 for 97.9% of genes, whereas only 1.6% of genes were defective on quinate. Profiles were predominantly a continuous sensitivity gradient, with a 24-gene discrete module showing extreme quinate-specific defects; no specific functional enrichment survived FDR correction in the two large modules. These are single-timepoint measurements with unknown technical noise, and the specificity analysis assumes that cross-condition variation reflects biology rather than measurement error. The collection uses single-gene deletions rather than transposon sequencing, although the report compares its 625 condition-specific genes with RB-TnSeq findings from Wetmore et al. (2015). ₁₃ The discoveries digest describes the 31% as a share of dispensable genes. In the deletion-phenotype report, however, the denominator is the complete growth matrix, not all TnSeq-dispensable genes, so the digest wording should not be reused as stated. _12,13

The top condition-specific genes match the biochemically expected pathways, which the digest reads as validating the matrix as a functional-genomics dataset. ₁₂ Named hits, by condition:

• urea: urease-complex genes ureA, ureB, ureC, ureD, ureE, ureF, and ureG (all 7 subunits/accessory proteins).
• quinate: pcaC, pcaG, pcaH, pcaB, quiA, quiB, pqqC, and pqqD, for protocatechuate/quinate degradation and pqq-biosynthesis.
• Asparagine: aspartate ammonia-lyase and L-asparaginase, for asparagine catabolism.
• Acetate: fadB, malate synthase G, and citB, for fatty-acid β-oxidation and the glyoxylate-shunt.
• Glucarate: gudD, a D-glucarate transporter, and 2,5-dioxovalerate dehydrogenase.
• Glucose: eda, gntT, gluconokinase, and PQQ-dependent glucose dehydrogenase, for the entner-doudoroff-pathway.
• Butanediol: 2,3-butanediol dehydrogenase and E2 acetyltransferase, for butanediol catabolism and the acetoin pathway.
• Lactate: lldR and cyoC/cyoD, for lactate regulation and cytochrome oxidase. ₁₃ The 51-gene quinate-specific set at spec > 0.5 and z < -1 extends beyond the core degradation pathway to complex-i subunits. PQQ-biosynthesis genes are condition-specific for both quinate and glucose, consistent with PQQ-dependent dehydrogenases catalyzing the first step of both pathways. The proposed unique electron-transport demand of aromatic catabolism is an inference. ₁₃ This supports the network account above, which groups the 51 quinate-specific genes around the β-ketoadipate pathway. ₁₅ At the single-gene extreme, aciad3522 had a growth ratio of 0.013 on acetate versus 1.39 on quinate and glucose. The digest describes this as the largest condition-specific effect of any single gene in the 2,034-gene matrix; FBA predicts zero flux through it, and its biological role is unknown. ₁₂

The counter-ion analysis supports and broadens this architecture: across 71 NaCl experiments, 4,304 of 10,821 metal-important gene records (39.8%) were also NaCl-important; excluding the Synechococcus elongatus outlier gave 36.7% (3,739/10,183). Shared-stress genes were important for a mean of 4.1 metals per gene versus 2.5 for metal-specific genes. The metal-specificity analysis found 14.7% overlap under its stricter threshold, compared with 39.8% in the counter-ion analysis. _24,9 The only within-metal counter-ion comparison was copper in psRCH2, tested as CuCl₂ (anaerobic, 3 replicates) and CuSO₄ (aerobic, 3 replicates). Cross-salt fitness correlation was r=0.439, versus within-replicate correlations of r=0.720 for CuCl₂ and r=0.859 for CuSO₄. This comparison is severely confounded by aerobic versus anaerobic growth, however, so it cannot isolate counter-ion effects. ₂₄

The broader fitness-module analysis supports and extends this architecture. Independent component analysis (ICA), a method for decomposing activity profiles into modules, identified 1,116 stable modules across 32 organisms; 94.2% had significantly elevated within-module cofitness, mean within-module |r| was 0.34 versus 0.12 in the background, and genomic-adjacency enrichment was 22.7x. Cross-organism alignment identified 156 module families spanning 2+ organisms, including 28 spanning 5+, 7 spanning 10+, and 1 spanning 21 organisms; 145 families had consensus functional labels. ₂₅ PFam domains increased module annotation from 92/1,116 (8.2%) to 890/1,116 (79.7%), increased predictions from 878 to 6,691, and raised annotated families from 32 to 145 of 156. Ortholog transfer achieved 95.8% strict precision, 91.2% coverage, and 0.934 F1; domain-based prediction achieved 29.1% precision, 66.6% coverage, and 0.401 F1. _12,25

Condition-specific FBA (flux balance analysis, a constraint-based metabolic modeling method) correlated weakly and inconsistently with measured growth: glucose rho = -0.021 (p = 0.677; n = 387), acetate rho = -0.153 (p = 0.004; n = 352), asparagine rho = -0.257 (p < 0.001; n = 286), butanediol rho = -0.145 (p = 0.092; n = 137), glucarate rho = +0.246 (p = 0.005; n = 127), and lactate rho = -0.160 (p = 0.065; n = 135). ₁₄ In the aromatic example, FBA captured 1.76× higher Complex I flux on aromatic substrates, with fluxes 0.55 versus 0.31, but predicted 0% essentiality; 30/51 network genes had no FBA mappings, and 10/13 Complex I operon subunits had quinate-specific defects. _15,12 FBA predicted zero flux through NDH-2 and ACIAD3522 on all standard media because growth optimization routes NADH through Complex I; this misses alternative, suboptimal pathways that may be required under capacity constraints. ₇ The triple-essentiality study summarizes this mismatch directly: 70% of genes showed condition-specific growth defects, yet correlations with FBA flux were weak (ρ = -0.26 to +0.25), and FBA essentiality class did not capture the variation (p=0.63). The study tested defect thresholds from Q10 to Q40 and reports that its null result held from Q10 through Q35. ₁₄

The ENIGMA study supports and qualifies the distinction. Its corpus contained 46,389 genome × condition pairs from 727 genomes and 363 conditions, with 4,293 shared kegg orthologs. Adding KO × condition interactions increased mean genus-blocked AUC from 0.620 to 0.653, an improvement of 0.032 in 80 of 106 held-out genera; 95 of 343 individually testable conditions achieved AUC > 0.75. Gene presence can encode whether an organism can use a condition, whereas growth rate also depends on kinetics, expression, regulation, and ribosome efficiency; continuous growth metrics had negative R² under genus-blocked holdout. ₂₆

Condition-matched Fitness Browser concordance qualifies the mechanistic reading of the ENIGMA model. SHAP (a feature-attribution method) features were mechanistically coherent, with K10440, a ribose transporter, predicting ribose growth and protocatechuate cycloisomerase predicting aromatic catabolism. Yet after expanding the top SHAP KOs to correlated gene blocks at |r|>0.8 (57 KOs → 335 Fitness Browser loci), 18.7% showed significant fitness effects (|t|>4) in matched experiments versus a 16.3% random baseline, which the report labels 1.19x enrichment and characterizes as a weak positive. The stated percentages and the stated ratio do not correspond exactly; both are retained here as reported. Per-strain enrichment ranged from 1.72x for Cup4G11 to 0.83x for pseudo1_N1B4, which showed no enrichment. The model predicted growth correctly (AUC 0.78 for amino acids) through combinations of prevalence-variable KOs acting as genus-level proxies rather than through individually causal genes. The report interprets the weak concordance as a difference in question rather than evidence against mechanism: gene presence across genera and gene essentiality within one strain are different biological questions, a strain may have redundant pathways or be tested under conditions that differ from the growth assay, and SHAP distributes credit across correlated features, so the causal gene may be a correlated neighbor of the highlighted one. ₂₆

Cross-organism pangenome profiling adds a smaller, enzyme-level example. A presence/absence matrix of 57 ECs across 14 organisms, scored with fitness-specificity z-scores, identified 11 strong co-occurrence cases and four carbon-source-specific fitness defects. ₂₇ Two antimicrobial-resistance (AMR) analyses treat condition specificity as a missing test rather than a measured result. One proposes testing AMR fitness costs under antibiotic stress rather than only standard laboratory media, and reports no antibiotic-stress result. The other proposes computing cofitness separately under antibiotic stress and standard growth, to distinguish regulatory coupling from shared dispensability. _28,29

Metabolite-matched fitness in pseudomonas-fw300-n2e3 shows that much of a metabolite-condition fitness landscape is housekeeping rather than substrate-specific. Across 21 web-of-microbes (WoM) metabolites with matching Fitness Browser experiments, 601 unique genes had significant fitness effects (|fit| > 1, |t| > 4), totaling 4,764 significant gene-condition hits; the most genetically complex metabolisms were carnitine (283 genes), alanine (295 genes across D/L forms), arginine (270 genes), and tryptophan (231 genes). However, 231 genes were significant in 3+ metabolite conditions, and the top 18 genes, significant in all 21 conditions, are amino-acid biosynthesis enzymes; the report reads these as “housekeeping fitness” required on any minimal medium rather than substrate-specific catabolism, and identifies the ~370 genes significant in only 1-2 conditions as the truly substrate-specific ones. Emerged metabolites, produced de novo, showed slightly more fitness-important genes than Increased metabolites, already present but amplified, though with high variance; both categories had detrimental:beneficial ratios of ~10:1, which the report reads as most fitness-significant genes being required for growth on these substrates rather than inhibitory. The figure fitness_hits_per_metabolite.png shows significant fitness-gene counts per WoM-produced metabolite, separating detrimental and beneficial genes. ₃₀

Community and ecological context

The pathway analysis extends condition-specific interpretation to within-species metabolic ecotypes. Among 225 species with at least 50 genomes and at least 3 variable pathways, binary completeness profiles across 80 pathways yielded a median of 4 metabolic ecotypes per species, with a maximum of 8 in Alistipes onderdonkii and Barnesiella intestinihominis. Ecotype count correlated with pangenome openness at raw rho=0.262, p=6.8e-05, and partial rho=0.322, p=8.0e-07 after controlling for genome count. ₁₁

The plant study refines this model. Host extraction identified 1,307 species across 19 testable hosts; rice contributed 286 species, arabidopsis thaliana 184, wheat 122, and maize 98. Among 65 plant-associated species with at least 20 genomes, only 18 had phylogenetic-tree distance data, and only 5 of 17 testable species passed Bonferroni-corrected tests. Strongest host-by-subclade evidence occurred in xanthomonas campestris, where 46/47 genomes occupied one Brassica-associated subclade (corrected p = 3.3×10⁻¹¹), and xanthomonas vasicola, where 47/52 occupied one Zea mays-associated subclade (corrected p = 1.7×10⁻¹⁰). ₅

The NMDC analysis supports and extends this interpretation from mutant assays to ecosystem differentiation. PCA of 220 samples × 80 pathway-completeness features explained 49.4% of variance in PC1 and 16.6% in PC2, with PC1–5 explaining 83.0%; Soil and Freshwater communities were nearly non-overlapping, with Soil median PC1 +3.86 and Freshwater −6.28, Mann–Whitney U = 3,674, p < 0.0001. Per-pathway tests found ecosystem differences for 17 of 18 amino-acid pathways after correction, strongest for glycine (H = 92.98, q = 1.2×10⁻¹⁹), asparagine (H = 66.19, q = 3.8×10⁻¹⁴), and cysteine (H = 62.66, q = 1.5×10⁻¹³); tyrosine was the only non-significant pathway (q = 0.71). This supports ecosystem-associated metabolic configurations but does not distinguish physicochemical from taxonomic causes. ₂

The field-versus-lab analysis supports and extends the framework in Desulfovibrio vulgaris Hildenborough. Of 757 Fitness Browser RB-TnSeq (random barcode transposon sequencing) experiments, 337 were field and 420 lab; field-stress, field-core, and lab-nutrient-important genes were enriched in the core genome, whereas heavy-metal- and lab-antibiotic-important genes had lower core fractions. Fitness importance predicted conservation more consistently than the field-versus-lab label: field-plus-lab fitness CV-AUC was 0.548 and a full model including gene length reached 0.645. ₃₁

The Oak Ridge comparison refines laboratory-to-field transfer. Across 108 oak-ridge-field-research-center groundwater sites, 14 of 26 Fitness Browser genera were detected by 16s-amplicon-sequencing, and five of 11 sufficiently prevalent genera had uranium associations passing BH-FDR; Herbaspirillum and Bacteroides increased, whereas Caulobacter, Sphingomonas, and Pedobacter decreased. Aggregate laboratory tolerance had a positive but non-significant association with high-uranium/low-uranium abundance ratio (Spearman rho=0.503, p=0.095, n=12 genera). ₃₂

The lignin-enrichment experiment supports and extends context dependence to community selection and ecological memory. Across 21 samples in 7 groups, lignin enrichment restructured bacterial communities: after one round, pseudomonas-aeruginosa reached 39.3% and Acinetobacter baylyi ADP1 25.2% of 16S reads, while Shannon diversity fell from 6.46 to 3.16 and observed OTUs from 1,594 to 163. Global bacterial PERMANOVA attributed 97.9% of community variance to treatment (R²=0.979, p=0.001); labile-carbon co-supplementation changed ADP1 from 25.2% to 41.7%, Pseudomonas from 39.3% to 23.0%, and Aeromonas from 0.1% to 20.2%. Round-1 history explained 58.9% of Round-2 variance (F=14.31, R²=0.589, p=0.002), while current carbon source explained 32.7% (F=4.85, R²=0.327, p=0.018); the reported memory index was ~0.50. These are community-composition statistics rather than direct gene-level fitness measurements. ₃₃

The IBD pathobiont analysis supports and extends context dependence to disease-associated ecological states. Four ecotypes were selected from 8,489 curatedMetagenomicData MetaPhlAn3 samples; among 23 Crohn’s disease patients, assignments were E0: 7 samples (27%), E1: 11 (42%), E2: 0, and E3: 8 (31%), with chi²(3)=10.0 and p=0.019. External replication had mean leave-one-substudy-out ARI=0.113, and six actionable pathobionts reached total prioritization score ≥2.5. These results support ecological intervention hypotheses but do not equate abundance or co-occurrence with mutant essentiality or causal fitness. ₃₄

Tensions

The pathway-level result introduces a tension with simple conservation expectations. Latent complete pathways had mean conservation 0.869 versus 0.829 for active dependencies, and active-greater-than-latent comparison was not significant (p = 0.94), whereas gene-level analyses associate essentiality and strong fitness effects with higher core fractions. _10,3,4,20 The new 7-organism analysis instead found mean core gene completeness of 0.986 for Active Dependencies versus 0.975 for Latent Capabilities, but its model-organism coverage and near-complete genomes compress the comparison. ₁₁ The NMDC negative leucine and arginine associations do not resolve this tension because GapMind measures genomic potential, not expression or essentiality, and abiotic controls were unavailable. ₂

Variable pathway count was positively associated with pangenome openness after genome-count control (rho=0.530, p=2.83e-203), while broader pangenome analyses reported null relationships between openness and environment or phylogeny effect sizes. This refines rather than resolves the conflict because pathway variability is a narrower predictor and the new analysis did not compute full phylogenetic independent contrasts. _11,35

The metal atlas found 87.4% core among broad metal-important genes versus 76.9% baseline, whereas the DvH condition-specific heavy-metal result found 71.2% core. The metal-specificity analysis found 84.8% pooled core for metal-specific genes and 90.2% for general sick genes, but definitions, organism coverage, general-stress capture, and exclusion of putatively essential genes differ. _8,31,9 Resolving work requires a shared metal-specific definition, non-metal controls, dose-relative-to-MIC normalization, and phylogenetic contrasts.

The truly-dark analysis introduces a related tension with any general claim that unknown genes are stress-enriched: stress represented 28.7% versus 43.2% of strong phenotypes for truly dark versus annotation-lag genes (OR = 0.53, p < 0.001), while the truly-dark analysis also identified genes that became important in selected nutrient, stress, iron, community, or rich-media contexts. ₁₈ This is not a contradiction in measurement: the census compares condition distributions among dark-gene classes, whereas the other analyses test whether individual genes or pathways matter under selected conditions. Resolving work requires matched dark and annotation-lag genes tested across the same condition matrix with controls for length, insertion density, organism, and polar effects.

The plant study adds a related tension: 94.2% of 5,671 eggNOG ortholog groups were significant in raw plant-versus-non-plant tests, but only 50 retained enrichment after phylum-level control, and compartment separation explained only 0.060 of variance by db-RDA. The 78.7% refined dual-nature classification also conflicted with its four-of-four neutral-control failures. ₅ The SNIPE result similarly found 22 of 64 AlphaEarth dimensions significant but a largest Cohen’s d of 0.26 and geographic metadata for only 28.4% of genomes. ₆ These findings support statistically detectable but modest environmental structure rather than direct habitat or causal-fitness measurement.

The respiratory-chain result introduces a modeling tension rather than a contradiction in measured fitness. Aromatic analysis and respiratory-chain studies associate quinate with greater Complex I dependence despite lower total NADH yield, but the interpretation relies on theoretical stoichiometry and inferred capacity limits, while FBA predicts zero flux through NDH-2 and ACIAD3522 on standard media. The cross-species test found 5 of 14 organisms with validated NDH-2, whose mean Complex I aromatic deficit was −0.297 versus −0.156 without validated NDH-2; p = 0.52, with only 4 organisms lacking NDH-2. _15,12,7 ADP1 wiring may therefore be species-specific, and the comparison is underpowered and vulnerable to annotation errors.

Only 15 essential families were essential in all 48 organisms, whereas 859 ortholog families were universally essential within every organism in which they occurred. ₂₀ This refines rather than contradicts condition-specific interpretation because genomic context, paralogs, alternative pathways, and compensation affect essentiality. The IBD within-pilot taxonomy–metabolite CCA axis had r=0.964, whereas pooled metabolomics had cross-cohort LOSO ARI=0.000 and the ecotype framework had mean LOSO ARI=0.113; these measure within-pilot association versus cross-cohort reproducibility. ₃₄

Open Directions

• Use the 8-carbon-source mutant-fitness matrix with pathway enrichment to test weak urea correlations, and compare ADP1 profiles with comparable organisms. ₁
• Add measured trace aromatic compounds to condition-specific FBA media; construct an ADP1 NDH-2 deletion mutant and measure NADH/NAD⁺ ratios and growth on quinate, glucose, acetate, lactate, and urea. _14,7
• Directly map GapMind per-step genes to Fitness Browser fitness and essentiality, test whether the 15.8% latent fraction persists under independent thresholds, and extend the comparison beyond the 7 organisms with matching genomes. _10,11
• Test the untested prediction that organisms in nutrient-rich environments disproportionately lose carbon-pathway genes, using pangenome presence/absence with the latent/active classification, and re-derive the 7-organism reclassification with condition-independent rather than median-based importance thresholds. _10,11
• Test variable-pathway–openness relationships with full phylogenetic independent contrasts and matched genome-count sampling within the 18 genera. ₁₁
• Test whether accessory-dependent leucine, valine, arginine, lysine, and threonine pathways predict measured metabolite exchange rather than genomic potential; obtain paired metabolomics for the 33 Freshwater samples and metatranscriptomics for expressed pathway completeness. _11,2
• Test metal cross-resistance with non-metal controls, MIC-relative dose normalization, and PGLS or independent contrasts; recompute metal-specificity fractions across 1–20% sick-rate thresholds. _36,9
• Match Fitness Browser organisms to ENIGMA CORAL genomes at species or strain level, and fit multivariate models to Oak Ridge samples while controlling for pH, redox, and carbon sources. ₃₂
• Reanalyze the lignin experiment with DADA2 ASVs, larger replication, UNITE-based ITS taxonomy, phylogenetic diversity, and strain-resolved functional assays. ₃₃
• Use plant-associated genome sets to test whether compartment markers predict matched condition-specific fitness with genus-blocked validation and transcriptomics; expand phylogenetic coverage and apply sparse PGLMMs. ₅
• Test SNIPE–ManXYZ coupling directly with matched strains, ManXYZ fitness assays on D-mannose and D-glucosamine, and phage challenge. ₆
• For truly dark genes, extend pangenome linkage to the 17,479 unlinked genes; use AlphaFold2 or ESMFold and Foldseek for the top 100 candidates; test PV4/5210953 on motility, ANA3/7026383 under nitrogen limitation, and selected candidates with mobile-CRISPRi; characterize dark islands and perform a Methanococcus-focused analysis using cofitness signals reported as r > 0.97. ₁₈
• Reanalyze dark-gene condition distributions in a matched matrix controlling for gene length, insertion density, organism, and polar effects, and test whether cofitness-module predictions improve experimental prioritization. ₁₈
• Replicate the single-timepoint ADP1 growth ratios to estimate technical noise, then recompute condition-specificity scores and tier assignments. ₁₃
• Compute AMR-gene cofitness separately under antibiotic-stress and standard-growth experiments. ₂₉

The detailed analyses are available in ADP1 Deletion Collection Phenotype Analysis, Triple Essentiality Concordance Analysis for *Acinetobacter baylyi* ADP1, Aromatic Catabolism Support Network in ADP1, Gene Conservation, Fitness, and the Architecture of Bacterial Genomes, Core Gene Paradox — Why Are Core Genes More Burdensome?, The 5,526 Costly + Dispensable Genes, Counter Ion Effects on Metal Fitness Measurements, The Pan-Bacterial Essential Genome, Essential Metabolome: GapMind Pathway Analysis Across Essential-Gene Organisms, Field vs Lab Gene Importance in *Desulfovibrio vulgaris* Hildenborough, Fitness Effects vs Conservation — Quantitative Analysis, Pan-bacterial Fitness Modules via Independent Component Analysis, Functional Dark Matter — Experimentally Prioritized Novel Genetic Systems, Metabolic Consistency of Pseudomonas FW300-N2E3 Across Four KBase Data Lakehouse Databases, Genotype × Condition to Phenotype Prediction from ENIGMA Growth Curves, Metagenome-Prioritized Phage Cocktails for Crohn’s Disease and IBD, Lab Fitness Predicts Field Ecology at Oak Ridge, Lignin Enrichment and Ecological Memory in Microbial Communities, Metabolic Capability vs Metabolic Dependency, Gene-Resolution Metal Cross-Resistance Across Diverse Bacteria, Pan-Bacterial Metal Fitness Atlas, Metal-Specific vs General Stress Genes, Community Metabolic Ecology via NMDC × Pangenome Integration, Metabolic Capability vs Dependency, Plant Microbiome Ecotypes — Compartment-Specific Functional Guilds and Their Genetic Architecture, Condition-Specific Respiratory Chain Wiring in ADP1, SNIPE Defense System in the KBase Data Lakehouse Pangenome, and Truly Dark Genes — What Remains Unknown After Modern Annotation?.