Decoupling between pangenome conservation and fitness cost

Pangenome conservation and measured fitness cost are distinct properties: an AMR gene’s prevalence across sampled genomes does not, in this dataset, predict the relative fitness effect of disrupting it. The evidence comes from integrating kescience-fitnessbrowser measurements with pangenome conservation labels, as described in the Fitness Cost of Antimicrobial Resistance Genes. ₁

Core–accessory AMR genes show similar fitness distributions

Core, or intrinsic, AMR genes (N=638) and accessory, or acquired, AMR genes (N=163) had virtually identical mean fitness values of −0.024, with Cohen’s d = 0.002 and Mann-Whitney U p = 0.33. ₁ This supports a decoupling between pangenome conservation status and baseline AMR fitness cost: being broadly conserved did not correspond to a measurably different disruption phenotype from being accessory in the analyzed data. ₁

The broader stratification analysis likewise found no conservation difference (p = 0.33), no Tier 1-versus-Tier 2 difference (p = 0.26), and no antibiotic-versus-metal resistance-type difference (p = 0.87). ₁ These null results refine environmental-resistome by separating the persistence or distribution of resistance genes from the metabolic burden measured when those genes are disrupted.

A separate Fitness Browser cross-reference qualifies this null rather than contradicting it outright. Using a DIAMOND-based FB-pangenome link table (177,863 links at 100% sequence identity), amr_pangenome_atlas identified 178 AMR genes across 37 Fitness Browser organisms, yielding 29,386 fitness measurements. AMR genes showed slightly less fitness cost than the non-AMR baseline (median fitness -0.007 vs -0.012, Mann-Whitney p=3.7e-6), beta-lactamases were nearly neutral (median -0.001), and singleton AMR genes were costliest (median -0.019). ₂ The report explains this by saying that the AMR genes in these predominantly environmental organisms are well-integrated intrinsic resistance genes rather than recently acquired mobile elements. That is an interpretation of this sample, not a direct measurement of gene history. The report also gives two caveats. First, the 100% identity threshold is conservative: it avoids paralog confusion but may undercount fitness effects for closely related gene variants, such as alleles differing by a single synonymous substitution. Second, the cost of recently acquired mobile resistance elements in clinical pathogens may differ substantially. ₂

A separate pangenome integration provides a refining genome-wide comparison rather than a contradiction: among 33 organisms, 27,693 putative essential genes were 86.1% core versus 81.2% for non-essential genes, with a median odds ratio of 1.56; 18 of 33 organisms showed significant enrichment by Fisher’s exact test after Benjamini-Hochberg false discovery rate (BH-FDR) correction at q < 0.05. ₃ Essentiality was inferred from RB-TnSeq, a random-barcode transposon sequencing approach, under the library construction and growth conditions represented in the Fitness Browser. ₃ Thus, essentiality showed a modest conservation association in this broader comparison, while the AMR subset showed no difference in averaged baseline knockout fitness.

A genome-wide comparison qualifies this AMR-specific null result rather than generalizing it to all genes: across approximately 194,000 genes from 43 diverse bacteria, genes important in more conditions were more likely core, and essential and strongly fitness-affecting genes were more often core than always-neutral genes, although the breadth association was weak (Spearman rho=0.086, p=8.1e-230). ₄ Essential genes were 82% core, whereas always-neutral genes were 66% core. ₄ The new synthesis supports this direction using 194,216 protein-coding genes across 43 bacteria, while emphasizing that even always-neutral genes were 66% core and therefore that conservation is only a modest proxy for measured importance. ₅ It also refines the interpretation by showing that 24.4% of core genes were beneficial when deleted versus 19.9% of accessory genes, and that core genes were 1.78x more likely to have strong condition-specific phenotypes and 1.29x more likely to be important-but-burdensome trade-offs. ₅ The new 33-organism result supports the direction of this association but reports a smaller essential-versus-non-essential contrast, emphasizing that conservation–fitness coupling depends on the definition of fitness category, organism set, and pangenome sampling. ₃ Thus, conservation can track fitness importance when fitness categories and breadth are considered across the genome, while AMR core/accessory labels did not distinguish baseline mean knockout fitness in the narrower analysis. ₄

The fitness–conservation gradient as plotted in the synthesis report. ₅

The synthesis describes a clear, quantitative gradient from essential genes (82% core) to always-neutral genes (66% core): more important genes are more conserved, but the effect is modest. It also notes that even genes with no detectable fitness effect in any experiment are 66% core. ₅ The gradient spans 194,216 protein-coding genes across 43 diverse organisms, and the report’s examples range from archaea (Methanococcus) to plant pathogens (Ralstonia) to gut commensals (Bacteroides). ₅

The underlying category table supports a graded rather than binary relation. Core-gene frequency was 82% among 27,693 essential genes (no viable mutants), 78% among 15,989 often-sick genes (>10% of experiments), 70% among 20,739 mixed sick-and-beneficial genes, 72% among 25,201 sometimes-sick genes, 70% among 9,705 sometimes-beneficial genes, and 66% among 94,889 always-neutral genes. ₄ Binning by strongest fitness effect gives the same direction: essential genes were 82.2% core, genes with min_fit < -3 were 77.7% core, and genes with min_fit -1 to 0 were 66.4% core. ₄ The mixed (70%) and sometimes-sick (72%) categories are not strictly ordered by severity, so the gradient is not perfectly monotonic across all six categories. ₄

The cross-project digest describes the same result as a 16-percentage-point gradient from essential genes (82% core) to always-neutral genes (66% core) across 194,216 protein-coding genes in 43 bacteria. According to the digest, the pattern also holds when binning by strongest negative fitness effect (min_fit < -3 → 78% core vs min_fit near 0 → 66%) and by fitness breadth (important in 20+ conditions → 79% core vs 0 conditions → 66%). The digest concludes that the essentiality–conservation link from conservation_vs_fitness is quantitative rather than binary. ₆ For the min_fit < -3 bin, the digest’s 78% and the project report’s 77.7% are both reported values at different precision. _6,4

The synthesis further refines the interpretation of this gradient through a joint selection-signature analysis: 28,017 genes were both costly in the laboratory and conserved, whereas 5,526 were costly and dispensable; the former are evidence for, rather than direct measurements of, purifying selection in natural environments, and the latter are candidates for ongoing gene loss. ₅ This distinction is consistent with the existing caveat that laboratory fitness does not directly measure ecological fitness.

The core_gene_tradeoffs selection-signature matrix extends this picture. Trade-off genes, which are important under some conditions but burdensome under others, were 1.29x more likely to be core (OR=1.29, p=1.2e-44). The neutral row of the matrix, which covers genes with no measured burden, contained 86,761 conserved genes and 21,886 dispensable genes. ₇ The report labels the neutral-conserved class as classic housekeeping genes and the neutral-dispensable class as niche-specific genes. Because neither class showed a laboratory burden, these labels are interpretations rather than measured functions or habitats. ₇ The large neutral-and-conserved class is itself evidence of decoupling, since these genes are conserved without a measured burden. ₇

As literature context, the report cites Rosconi et al. (2022), who showed that gene essentiality is strain-dependent across Streptococcus pneumoniae pangenomes. The report uses this to argue that conservation reflects selection across diverse environments rather than universal essentiality. This is outside support for an interpretation, not a result of the corpus analysis. ₇

Graded essentiality and conservation

Essentiality breadth across organisms refines the essential-versus-core contrast described in gene-essentiality. Universally essential genes were 91.7% core versus 80.7% for non-essential genes, and 71% of universally essential families were 100% core across all genomes in their species. _8,6 Orphan essentials, which have no orthologs, were only 49.5% core. The reports describe them as strain-specific essential functions that have not been conserved across species. _8,6 Orphan essentials show that being essential under assayed conditions does not by itself imply conservation, which supports the decoupling interpretation for lineage-restricted genes. ₈

The report’s conservation-hierarchy table gives these core fractions: 91.7% for 6,963 universally essential genes, 88.9% for 90,844 variably essential genes, 81.7% for 47,140 never-essential genes, and 49.5% for 4,683 orphan essential genes. ₈ The report leaves two denominators unreconciled. First, the table’s 81.7% for the never-essential class uses a differently labeled denominator from the key-findings figure of 80.7% for non-essential genes. Second, the table’s 4,683 orphan essentials differ from the report’s total of 7,084 essential genes with no orthologs in any other Fitness Browser organism, and the report does not explain how the two counts relate. The wiki records both values rather than choosing one. ₈

Essentiality penetrance has only a weak positive association with core fraction (rho=0.123, p=1.6e-17). Families essential in more organisms are slightly more core: 97.1% for >80% penetrance versus 92.8% for <20%. Clade size does not predict essentiality rate (rho=-0.13, p=0.36). ₈ This supports the reading that conservation is a modest proxy for fitness importance even among essential families; see pangenome-core-boundary-and-clade-size-bias. ₈

The organism-level essential-core enrichment in conservation_vs_fitness was heterogeneous. The strongest signals were in Methanococcus maripaludis S2 (OR=5.21), Ralstonia syzygii PSI07 (OR=3.41), and Marinobacter adhaerens (OR=3.08). ₃ The report itself calls the overall enrichment modest (OR=1.56), because most genes in well-characterized bacteria are core regardless of essentiality. This ceiling limits how well conservation can separate essential from non-essential genes. ₃

Single-organism results from acinetobacter-baylyi-adp1 point in different directions and do not independently test the gradient. The ADP1 explorer reports that essential genes are more likely to be in the core pangenome, consistent with the pattern that conserved genes tend to be essential. ₉ However, the ADP1 triple-essentiality comparison found pangenome status overwhelmingly core across all classes (93–100%), with no significant enrichment among discordant genes (Fisher OR = 0.89, p = 0.80). In this organism, conservation therefore did not distinguish genes whose essentiality calls disagreed; see essentiality-assay-discordance. ₁₀

Mechanism predicts distribution, not measured cost

Resistance mechanism was strongly associated with pangenome conservation status (χ² = 69.3, p = 1.4×10⁻¹³): 44% of metal-resistance genes were accessory, compared with 13% of efflux genes and 16% of enzymatic-inactivation genes. ₁ Thus, mechanism supports prediction of where an AMR gene occurs in the pangenome, but it did not predict baseline fitness cost: efflux (N=254), enzymatic inactivation (N=304), metal resistance (N=144), and unknown (N=74) mechanisms showed no significant baseline-cost difference (Kruskal-Wallis H = 0.65, p = 0.89). ₁

This separation indicates that evolutionary or ecological processes shaping gene prevalence need not produce corresponding differences in laboratory knockout fitness. The report proposes, as an interpretation rather than a directly measured process, that horizontal transfer may preferentially retain cost-optimized genes or that compensation may reduce costs after acquisition. ₁ The genome-wide analysis refines this interpretation by showing that core genes had heavier tails in both negative and positive fitness effects and that 24.4% of core genes were ever beneficial versus 19.9% of auxiliary genes. ₄ Conservation therefore may be associated with broader, condition-dependent functional involvement without implying a uniformly larger baseline resistance burden. ₄

The conservation_vs_fitness integration further supports this distinction: essential-core genes were more enzyme-rich and better annotated than essential-auxiliary genes, whereas essential-auxiliary genes were 38.2% hypothetical and essential-unmapped genes were 44.7% hypothetical. ₃ These categories show that a gene can be classified as essential under the assayed condition without being broadly conserved or functionally characterized, so conservation status remains an imperfect proxy for measured fitness importance. ₃

Metal-fitness genes and shared-stress correction

Metal fitness data provide a case where conservation and fitness importance stay coupled after a confound is removed. The metal-fitness-atlas reported that metal-important genes were 87.4% core (OR=2.08). That result could in principle have been inflated by the shared-stress component, since osmotic stress genes are core cellular machinery. ₁₁ counter_ion_effects removed the ~40% shared-stress genes and restricted the analysis to metal-specific genes. The report states that core enrichment was then fully preserved for 12 of 14 metals. Deltas increased for molybdenum (+0.132 → +0.145), tungsten (+0.129 → +0.134), mercury (+0.116 → +0.133), and selenium (+0.115 → +0.131), and decreased for aluminum (+0.099 → +0.068) and zinc (+0.145 → +0.115). ₁₁ The prediction that core enrichment would weaken after correction (H1c) was rejected. This supports the atlas conclusion that metal-fitness core enrichment is not an artifact of overlap with the shared stress response; see shared-stress-versus-stressor-specific-fitness. ₁₁

The per-metal details contain an internal inconsistency, which the wiki records rather than resolves. The narrative states that only cadmium reverses (delta -0.008 → -0.108). In the corrected-conservation table, however, cadmium’s core fraction moves 0.522 → 0.422 against a baseline of 0.530, so its delta stays negative and does not change sign. Iron moves 0.778 → 1.000 against a baseline of 0.818, with delta -0.040 → +0.182, which does change sign. ₁₁ The narrative counts 7 of 14 metals (Mo, W, Hg, Se, Ni, Cr, U) as showing stronger corrected enrichment, but the table also shows an increased delta for iron, which that list omits. ₁₁ Each of these low-count metals comes from a single organism. The report describes the corrected deltas for cadmium (n=92) and iron (n=9) as statistically unreliable. Manganese (n=30) is also low-powered, although the report calls its 100% core result robust by ceiling effect. ₁₁

A stressor-specific subset refines this coupling. counter_ion_effects attributes to metal_fitness_atlas the finding that condition-specific genes, important only for metals and not for other stresses, are less core, and proposes that its own gene classification could refine that analysis. ₁₁ Taken together, metal-important genes as a whole are core-enriched while metal-only genes are reported to be less core, so the coupling depends on how fitness specificity is defined. ₁₁

The atlas’s own report gives the primary measurement. Across 22 organisms and 14 metals, genes with significant fitness defects under metal stress were 87.4% core versus a 76.9% baseline (OR=2.08, p=4.3e-162). This was the opposite of the initial hypothesis (H1a), which predicted accessory enrichment for toxic-metal genes on the basis of a prior DvH finding. ₁₂ The enrichment was broad across organisms. In 21 of 22 organisms, metal-important genes were more core than baseline, and 14 of these comparisons were significant at p<0.05. Only pseudomonas-fw300-n2e3 (P. fluorescens FW300-N2E3) showed a negligible negative delta (-0.003, p=0.70). ₁₂ The report attributes the enrichment to genome-wide metal fitness defects that predominantly reflect core cellular processes vulnerable to metal disruption (cell envelope, DNA repair, central metabolism), not specialized metal resistance mechanisms. This is the report’s interpretation of the gene sets, not a separate test. ₁₂ For metal stress, this supports coupling between conservation and fitness importance, in contrast to the AMR baseline-cost null above. _12,1

Essential-metal tolerance genes (Fe, Mo, W, Se, Mn) showed a mean core-fraction delta of +0.148, which the report describes as nearly double the toxic-metal delta of +0.081 (Mann-Whitney U=39, p=0.015, one-sided). The strongest enrichments were manganese (+0.198, with all 30 important genes core), zinc (+0.151), molybdenum (+0.148), tungsten (+0.145) and iron (+0.116). ₁₂ Of the 14 metals, 12 were individually significant at p<0.05. Only cadmium (1 organism, delta=-0.010, p=0.92) and uranium (2 organisms, delta=+0.035, p=0.34) were not significant, which the report attributes to likely limited organism coverage. ₁₂

Per-metal core enrichment in the atlas, listed as important genes; core fraction; delta; OR; p ₁₂:

• Manganese (essential): 30; 1.000; +0.198; OR inf; p=2.0e-03 ₁₂
• Zinc (toxic): 1,517; 0.890; +0.151; OR 2.94; p=2.0e-49 ₁₂
• Molybdenum (essential): 302; 0.950; +0.148; OR 5.32; p=1.3e-14 ₁₂
• Tungsten (essential): 303; 0.947; +0.145; OR 4.98; p=5.3e-14 ₁₂
• Mercury (toxic): 106; 0.934; +0.132; OR 3.62; p=1.6e-04 ₁₂
• Selenium (essential): 134; 0.933; +0.131; OR 3.59; p=2.9e-05 ₁₂
• Iron (essential): 651; 0.919; +0.116; OR 3.07; p=8.5e-18 ₁₂
• Aluminum (toxic): 1,381; 0.898; +0.107; OR 2.37; p=1.2e-26 ₁₂
• Copper (toxic): 2,139; 0.867; +0.090; OR 1.91; p=3.8e-27 ₁₂
• Nickel (toxic): 1,760; 0.877; +0.088; OR 1.93; p=3.0e-22 ₁₂
• Cobalt (toxic): 1,859; 0.859; +0.072; OR 1.67; p=8.1e-16 ₁₂
• Chromium (toxic): 262; 0.710; +0.060; OR 1.33; p=4.0e-02 ₁₂
• Uranium (toxic): 178; 0.685; +0.035; OR 1.18; p=3.4e-01, not significant ₁₂
• Cadmium (toxic): 92; 0.522; -0.010; OR 0.96; p=9.2e-01, not significant ₁₂

Ortholog-level breadth gives a further signal. Of 2,891 ortholog groups with metal phenotypes, 1,182 were conserved across ≥2 organisms and 601 across ≥3. The most broadly conserved family (OG00128) spans 17 organisms and 9 metals. Families with metal phenotypes in more organisms tend to have higher pangenome conservation (% core), which the report reads as consistent with fundamental cellular processes. ₁₂ This trend contrasts with the module_conservation finding that module-family breadth did not predict conservation (rho=-0.01, p=0.914). Metal ortholog families and ICA module families are different units, however, so the two results are not a direct contradiction. _12,13

The atlas tries to reconcile its result with the prior DvH analysis in field_vs_lab_fitness, which found heavy-metal genes to be the least conserved condition class (71.2% core). According to the atlas, that analysis isolated condition-specific genes, important only for metals and not for other stresses, whereas the atlas includes all genes with any metal fitness defect, and these are dominated by core stress-response functions. ₁₂ The field_vs_lab_fitness report does not describe its gene set that way. Its heavy-metals class is 198 genes important (fitness < -2) under heavy-metal conditions, with no stated exclusion of genes important under other stresses. These genes were 71.2% core (OR 0.77 versus baseline, FDR q=0.14), a trend below the baseline that was not significant after BH-FDR correction. ₁₄ The atlas’s exclusivity explanation is therefore an interpretation that the original report does not confirm, and the wiki records the disagreement rather than resolving it. The two analyses do use different gene-set definitions, and the apparent direction of the metal–conservation relationship depends on that definition; see ontology-and-category-schema-sensitivity. _12,14 metal_cross_resistance refines the atlas’s 87.4% core estimate with a finer gradient within metal genes: general stress (92%) > shared (91%) > specific (90%). That report interprets the layered core enrichment as reflecting the evolutionary history of metal tolerance. ₁₅

metal_specificity provides the most direct test of the stressor-specific subset. It covered the 22 organisms with pangenome links and reported both pooled core fractions (total core / total genes) and organism-mean core fractions (mean of per-organism core fractions). ₁₆ Metal-specific genes were 84.8% core pooled (2,969/3,500) and 88.0% by organism mean, with a +6.9% mean delta; 19/22 organisms had positive deltas and 12/22 were significant. Metal+stress genes were 94.3% core pooled (467/495) and 93.6% by organism mean, with a +10.9% mean delta; 13/13 organisms had positive deltas but only 1/13 was significant. General-sick genes were 90.2% core both pooled (2,183/2,420) and by organism mean, with a +9.0% mean delta; 21/21 organisms had positive deltas and 8/21 were significant. The baseline was 79.8% core pooled (73,957/92,650) and 81.1% by organism mean. ₁₆ The pooled metal-specific fraction is lower than the organism mean because organisms with large gene sets and lower core fractions carry more weight in the pooled estimate. ₁₆ A Cochran-Mantel-Haenszel (CMH) test, which combines comparisons across organism strata, found metal-specific genes less core-enriched than general-sick genes across organisms (p=0.011). ₁₆ This refines the stressor-specific claim attributed to metal_fitness_atlas above. Metal-specific genes are less core-enriched than general-sick genes but remain above the baseline in both pooled and organism-mean terms, so in this dataset specificity weakens the coupling rather than reversing it. _16,11

The metal_specificity estimates carry a measurement caveat. About 14% of protein-coding genes (approximately 82% core) are putatively essential and absent from fitness data. The report says this biases all categories toward core enrichment, that the true baseline core fraction is likely lower than 81%, and that the reported deltas are therefore conservative. ₁₆ The ICA module analysis below has the same essential-gene exclusion; see transposon-callability-bias. _16,13

Several atlas estimates rest on very few organisms. Manganese coverage is 1 organism (desulfovibrio-vulgaris-hildenborough, DvH). The report describes its result as 100% core enrichment (delta=+0.198) and asks whether this is universal. Tungsten coverage is also 1 organism (DvH), with strong core enrichment (delta=+0.145); whether the pattern extends beyond sulfate reducers is untested. Uranium coverage is 2 organisms (DvH and psRCH2), with delta=+0.035 (not significant, p=0.34). The report states that more organisms would be needed to determine whether uranium fitness genes are truly neutral or the effect was underpowered. ₁₂ These metal results should be treated as organism-specific until replicated; see fitness-condition-coverage-prioritization-bias. ₁₂

Pathway-level conservation and fitness dependency

Pathway-level analyses extend the question from genes to metabolic capabilities; see community-metabolic-interdependence. metabolic_capability_dependency tested whether pathway-level conservation differs between latent capabilities and active dependencies (H2a), and this hypothesis was not supported. Conservation rates, defined as the fraction of species genomes carrying a complete pathway, did not differ between the two classes. ₁₇ Active dependencies had n=755, mean 0.829 and median 1.000; intermediates had n=508, mean 0.907 and median 1.000; latent capabilities had n=248, mean 0.869 and median 1.000. Latent capabilities showed slightly higher mean conservation than active dependencies (Mann-Whitney U, p = 0.94 for active > latent; rank-biserial r = 0.052). ₁₇ This null supports decoupling at the pathway level: in this analysis, the report’s active-dependency class did not show more frequent complete-pathway carriage than latent capabilities. ₁₇

The null is limited by measurement granularity. The conservation rate records whether a pathway is complete in each genome of the species clade. This is a coarse measure that does not detect partial gene erosion, meaning individual gene deletions within an otherwise complete pathway. ₁₇ Median conservation was 1.000 in all three classes, so the measure may leave little room to separate them. ₁₇

Conservation by pathway category as plotted in the pathway_capability_dependency report. ₁₈

pathway_capability_dependency used a different measure and reports a small difference in the opposite direction to the metabolic_capability_dependency means: Active Dependencies had a mean core-gene completeness of 0.986 versus 0.975 for Latent Capabilities. ₁₈ This comparison covers only 7 Fitness Browser organisms, all well-studied model organisms (Desulfovibrio vulgaris Hildenborough, Shewanella oneidensis MR-1, Pseudomonas putida, Pseudomonas stutzeri, Caulobacter crescentus, Sinorhizobium meliloti, Azospirillum brasilense) with near-complete, well-annotated core genomes. The report gives this as the reason the gap between categories is small. The narrow sample and the near-ceiling completeness compress the conservation signal and limit comparison with the broader metabolic_capability_dependency means. ₁₈ The report describes the difference as a small but consistent enrichment, aligned with the metal_fitness_atlas finding that fitness-important genes are enriched in the core genome (OR=2.08). It also sets that atlas finding against its own observation of variable accessory pathways. ₁₈ The report adopted the atlas result that 87.4% of metal-fitness genes are core (OR=2.08) as a methodological template for OR-based conservation validation and cross-species aggregation. It also cites the 16pp fitness-conservation gradient as establishing that conserved genes are functionally active, not inert. ₁₈ The 0.986 versus 0.975 difference is reported here without a test statistic, so it is weak evidence. The cross-project citations reuse other projects’ results and are not independent replication. ₁₈

Coordinated architecture and scope of decoupling

The new synthesis refines a gene-by-gene view of conservation–fitness coupling: independent component analysis (ICA), a method for identifying coordinated fitness modules, found 1,116 co-regulated modules across 32 organisms. ₅ The dedicated module analysis supports this extension: module genes were 86.0% core versus 81.5% of all genes (odds ratio 1.46; p=1.6e-87), and among 974 modules with at least three mapped genes, 577 (59%) were more than 90% core. ₁₃ This supports cofitness-network-architecture while showing that conservation can be organized into coordinated units rather than only individual essential genes. The 86.0% versus 81.5% difference is statistically strong but modest in absolute terms because the baseline core rate is already high. ₁₃

However, module-family breadth did not predict conservation (rho=-0.01, p=0.914), so broader taxonomic distribution of a family is not by itself evidence of stronger conservation. ₁₃ This supports the existing interpretation that conservation is more a property of individual genes than of module-family scope. The 38 accessory module families with fewer than 50% core genes identify a flexible coordinated component, but their possible origin through horizontal transfer or niche-specific regulation remains a hypothesis rather than a demonstrated mechanism. ₁₃

Scope and limitations of the decoupling

The comparison used Fitness Browser data from 25 lab-adapted organisms, so the null result may not represent newly acquired resistance or wild populations. ₁ The analysis also classified a gene as core using a ≥95% prevalence threshold, while the median number of GTDB genomes per Fitness Browser species was 9, with a range of 2–399. ₁ Consequently, a gene present in all 9 sampled genomes could be labeled core despite having lower prevalence under broader sampling, making the core–accessory null result especially cautious for species with fewer than 20 genomes. ₁

The genome-wide comparison provides a related but non-identical test: it used fitness-effect magnitude, breadth, and condition-specific annotations rather than only the AMR baseline mean, and it found that genes with strong condition-specific effects were 77.3% core versus 70.3% without such annotations (OR=1.78 and p=1.8e-97). ₄ This refines the scope of the decoupling: broad conservation categories may fail to explain averaged AMR knockout cost while still tracking breadth or conditional functional importance. ₄ The synthesis qualifies that interpretation: condition-specific fitness did not establish niche-specific fitness, because the stronger effects among core genes may reflect greater detectability under the tested conditions rather than ecological specialization. ₅ The cross-project digest records the same contrast. Genes with strong condition-specific phenotypes in the Fitness Browser specificphenotype table were 77.3% core versus 70.3% for genes without them (OR=1.78, p=1.8e-97). The digest presents this as contradicting the intuition that condition-specific fitness implies niche-specific, accessory genes, and attributes the pattern to core genes being embedded in well-characterized, essential pathways. ₆

The new 33-organism integration supports this caution but adds specific sampling and assay limitations: 44 of 48 Fitness Browser organisms mapped to pangenome clades, only 33 were used downstream, and clades with only 2 genomes can have trivially high core fractions. ₃ Its essentiality calls were made under library construction and growth conditions, and the report notes that short genes, low-complexity regions, or scaffold edges can lack insertions without being essential. ₃ The main Escherichia coli clade was also absent because it contained too many genomes, while Keio mapped to a small clade at only 26.1% coverage. ₃

A single-organism prediction test supports the decoupling at the level of individual genes; see laboratory-fitness-versus-natural-selection. The field_vs_lab_fitness analysis concerns only Desulfovibrio vulgaris Hildenborough (DvH), and its conservation analysis used 2,725 non-essential genes with both fitness data and pangenome links. In this analysis, “field” fitness is not measured in the field. It means laboratory experiments that the report classified as field-relevant, 337 of 757 DvH experiments, compared with 420 lab experiments. ₁₄ Logistic regression predicting core status had a 10-fold cross-validated AUC (CV-AUC) of 0.517 (SD 0.052) with field fitness only, 0.531 (SD 0.052) with lab fitness only, 0.548 (SD 0.053) with field plus lab fitness, and 0.645 (SD 0.068) after adding gene length. ₁₄ The report concludes that gene length predicts core status much better than fitness effects from either field or lab conditions, and that neither fitness dimension alone is informative (CV-AUC near 0.5). ₁₄ Because the result comes from a single organism and excludes essential genes, it is a hypothesis-generating counterpoint to the multi-organism gradients above, not a general refutation. An enrichment that is statistically strong across many genes can still give near-chance prediction for individual genes. _14,4

Gene length confounds both sides of the comparison. Short genes get fewer transposon insertions, which affects fitness measurement quality, and core genes tend to be longer. Length-related callability can therefore create or mask conservation–fitness associations; see transposon-callability-bias. ₁₄ The same report interprets antibiotic and metal resistance as accessory-enriched in line with ecological expectations, but its class comparisons against baseline were not statistically significant. ₁₄ This interpretation is therefore weaker evidence than the mechanism–conservation association reported in amr_fitness_cost (χ² = 69.3). _14,1

The module-linked analysis adds a related coverage limitation: pangenome-linked module conservation covered 29 of 32 organisms because Cola, Kang, and SB2B lacked pangenome links after too few GTDB genomes were available for construction. ₁₃ It also excluded essential genes from ICA modules because essential genes lack usable transposon-insertion fitness variation, so module conservation profiles describe the measurable non-essential genome rather than essential-gene modules. ₁₃ ICA membership used |Pearson r| >= 0.3 and a maximum of 50 genes per module, while the >90% and <50% core boundaries were convenient classification thresholds; both choices can affect module composition. ₁₃

The finding therefore does not establish that conservation is unrelated to fitness under every ecological or genetic context; it shows that the available pangenome labels did not explain the measured baseline knockout-fitness distributions in this dataset. ₁ This qualification connects the concept to pangenome-integration and condition-specific-fitness, where denser sampling and condition-specific measurements could reveal relationships hidden by broad conservation categories or averaged fitness values. ₁ The detailed cross-dataset integration is documented in Conservation vs Fitness — Linking FB Genes to Pangenome Clusters, Gene Conservation, Fitness, and the Architecture of Bacterial Genomes, and Fitness Modules × Pangenome Conservation.

Tensions

The data contain an apparent tension: mechanism strongly predicted conservation status, yet mechanism did not predict baseline fitness cost. ₁ This is not a statistical contradiction because conservation status and knockout fitness are different response variables, but it means that a mechanism-associated distribution pattern cannot be treated as evidence for a mechanism-associated fitness burden. ₁

A second scope-dependent tension is that the AMR analysis found virtually identical baseline fitness distributions for core and accessory AMR genes (mean fitness −0.024 for both; p = 0.33), whereas the genome-wide analysis found a positive, though weak, conservation gradient from essential or broadly fitness-affecting genes to always-neutral genes (82% versus 66% core). _1,4 The AMR result supports decoupling for the analyzed resistance subset and baseline measure; the genome-wide result supports an association with fitness importance when effect breadth and conditional phenotypes are included. _1,4

The new 33-organism integration presents the same directional association with a smaller effect—86.1% core essential genes versus 81.2% core non-essential genes and median odds ratio 1.56—rather than resolving the tension. ₃ The synthesis likewise supports a modest genome-wide gradient but shows that core genes can be more burdensome in the laboratory, so conservation, essentiality, and laboratory cost should not be treated as interchangeable. ₅ The module result is directionally consistent with enrichment of conserved, coordinated fitness units (86.0% versus 81.5%), but it does not resolve the AMR-specific null because ICA modules exclude essential genes and summarize module membership rather than AMR baseline means. ₁₃ The difference may reflect AMR biology, averaging, organism composition, essentiality definitions, module-callability, or pangenome sampling; resolving it requires matched, condition-specific comparisons.

A further tension concerns the two AMR fitness analyses. amr_fitness_cost found no core–accessory difference in baseline fitness (p = 0.33). amr_pangenome_atlas found singleton AMR genes to be the costliest (median -0.019) and AMR genes overall slightly less costly than non-AMR genes (median -0.007 vs -0.012). _1,2 The analyses differ in organism set (25 versus 37 Fitness Browser organisms), in gene-linking method, and in prevalence categories (a ≥95% core threshold versus singleton status). The singleton result may therefore reflect the rarest genes rather than the accessory class as a whole. Neither side is preferred here. _1,2

Within counter_ion_effects, the narrative claim that only cadmium reverses conflicts with the report’s own corrected-conservation table, in which cadmium’s delta stays negative and iron’s delta changes sign. Each metal rests on a single organism, so the direction of corrected metal-specific conservation for these two metals is unresolved. ₁₁

The atlas’s per-metal deltas also do not match the pre-correction deltas reported in counter_ion_effects. For example, cadmium is -0.010 in the atlas versus -0.008 in counter_ion_effects, and molybdenum is +0.148 versus +0.132. _12,11 Neither report, as cited here, explains whether different gene sets or baselines account for the difference, so the wiki records both values. _12,11

The reported core fractions for metal genes defined as specific to metals differ, and so do accounts of what the earlier gene set contained. The atlas describes the prior DvH result (71.2% core) as concerning condition-specific heavy-metal genes. ₁₂ The field_vs_lab_fitness report defines that class as 198 genes with fitness < -2 under heavy-metal conditions, with no stated exclusion of genes important under other stresses, and its below-baseline trend was not significant (FDR q=0.14). ₁₄ metal_specificity, which explicitly separates metal-specific genes, found them 84.8% core pooled across 22 organisms. ₁₆ The DvH figure comes from a single organism, and the analyses define their metal gene sets differently. The wiki does not prefer either value or the atlas’s characterization of the DvH gene set. _12,14,16

The two pathway-level analyses point in different directions. metabolic_capability_dependency found latent capabilities slightly more conserved than active dependencies (mean 0.869 versus 0.829; p = 0.94 for active > latent). pathway_capability_dependency found Active Dependencies slightly higher in mean core-gene completeness (0.986 versus 0.975), in only 7 model organisms with near-complete core genomes. _17,18 The metrics differ, complete-pathway carriage across species genomes in one case and core-gene completeness in the other, and both differences are small. Neither side is preferred here. _17,18

Open Directions

• Use the 144 metal-resistance genes with fitness measurements and compare their standard-condition costs with responses in the metal fitness atlas, using metal-specific fitness analyses to ask whether genes that appear costly under standard conditions are protective under metal stress. ₁
• Recalculate core/accessory labels from expanded GTDB sampling and test the association with fitness using prevalence thresholds and species-stratified models, asking whether the null result persists when the median sample of 9 genomes is replaced by broader within-species coverage. ₁
• Analyze mechanism effects within organisms with many AMR genes, including Cup4G11 (77) and BFirm (50), using genetic-background-controlled comparisons to ask whether conservation–cost relationships emerge within species. ₁
• Extend the analysis from 25 Fitness Browser organisms to all 293K KBase Data Lakehouse genomes by predicting AMR cost from gene-cluster conservation patterns, then test whether predicted cost varies with pangenome prevalence outside the lab-adapted strains. ₁
• Reanalyze the AMR subset using fitness-effect breadth and condition-specific annotations, asking whether the genome-wide conservation gradient appears when baseline means are replaced by conditional importance. ₄
• Recompute the essential-versus-core association with expanded within-species pangenomes and matched condition-specific fitness effects, asking whether the 1.56 median odds ratio persists after correcting for clade size, gene-length bias, and assay coverage. ₃
• Test the 5,526 costly-and-dispensable genes for mobile-element, recent-acquisition, or gene-loss signatures using expanded pangenomes and transfer-aware analyses. ₅
• Compare core trade-off frequencies across organisms sampled from variable versus stable environments using AlphaEarth environmental data, testing whether environmental variability predicts the architecture of conservation–fitness coupling. ₅
• Analyze the 48 accessory co-regulated modules with module-level conservation and functional-enrichment tests to determine whether flexible-genome coordination reflects shared mobility, recent acquisition, or ecological specialization. _5,13
• Reanalyze the 974 mapped modules with insertion-callability masks and organism-stratified models, asking whether the 86.0% module-gene core fraction and 59% core-module fraction persist after accounting for the exclusion of essential genes and the 29/32 organism linkage coverage. ₁₃
• Re-run the shared-stress correction for cadmium and iron with additional organisms that have metal fitness data, asking whether iron’s sign change and cadmium’s more negative delta persist beyond the n=92 and n=9 single-organism samples. ₁₁
• Fit models of core status on fitness category that adjust for gene length and insertion density across multiple organisms, asking whether the 82% versus 66% essential-to-neutral gradient survives once the gene-length signal behind the single-organism 0.645 CV-AUC is controlled. _14,4
• Compare singleton, accessory and core AMR genes within the same 37-organism linked set, using relaxed identity thresholds to recover closely related variants, asking whether the singleton median of -0.019 reflects rarity itself or organism composition. _2,1
• Test the 49.5%-core orphan essentials for recent-acquisition or mobile-element signatures with transfer-aware pangenome analyses, asking whether lineage-restricted essentiality explains the weakest conservation–essentiality coupling. ₈
• Recompute pathway conservation for active dependencies and latent capabilities in one pipeline across more than the 7 model organisms, with gene-level erosion counts alongside complete-pathway carriage and core-gene completeness, asking whether the opposite-signed small differences reflect metric choice, organism sampling or real pathway-level decoupling. _17,18
• Re-derive the DvH heavy-metal gene set from field_vs_lab_fitness with and without genes important under other stress classes, asking whether the 71.2% core figure reflects metal-exclusive genes, as the atlas states, or a broader set. _14,12
• Add organisms with manganese, tungsten and uranium fitness data to the atlas comparison. The report’s suggested candidates are MR-1, Keio, Cup4G11, Putida and Caulo for manganese; Methanococcus_S2, Methanococcus_JJ and Keio for tungsten; and MR-1, ANA3, PV4 and Cup4G11 for uranium. The question is whether the single-organism DvH deltas (+0.198, +0.145) and the uranium null (+0.035, p=0.34) generalize. ₁₂
• Rerun the metal_specificity category comparison with putatively essential genes added as an explicit category, asking whether the CMH p=0.011 gap between metal-specific and general-sick genes persists once the ~14% of essential genes (~82% core) are included. ₁₆