Antimicrobial-resistance (AMR) genes can impose a measurable baseline burden while becoming beneficial under antibiotic exposure. The evidence supports a relative AMR-associated fitness difference, but compensation remains an interpretation rather than a directly measured process. This distinction is essential because RB-TnSeq (random barcode transposon sequencing) measures competitive fitness of insertion mutants relative to the pool average, rather than an absolute selection coefficient in an isogenic strain. 1 Consequently, a difference between mutant classes is a relative perturbation-fitness shift, not automatically beneficial or harmful absolute growth. 1
Baseline relative fitness cost
Using genome-wide transposon fitness data from the kescience-fitnessbrowser compendium, the analysis identified 1,352 AMR genes across 43 organisms and tested 25 organisms with sufficient data for per-organism effects. 1 Under non-antibiotic conditions, AMR-gene knockouts had systematically higher fitness than non-AMR-gene knockouts, indicating a relative metabolic burden associated with AMR genes in the measured strains. 1
Every organism showed a positive AMR-versus-background shift, and a DerSimonian-Laird random-effects meta-analysis estimated a pooled shift of +0.086 [95% CI: +0.074, +0.098], z = 14.3, p ~ 0. 1 The median per-organism Cohen’s d was 0.18, heterogeneity was I² = 54.3%, and Cochran’s Q = 52.54, p = 0.0007. 1 Six organisms were nominally significant at p<0.05, and two were significant after false-discovery-rate correction at q<0.05: acidovorax_3H11 and Koxy. 1 The report compares the median per-organism Cohen’s d of 0.18 with a literature prediction of +0.05 to +0.20 for fitness costs after compensatory evolution; this is a consistency comparison, not a direct measurement of compensation. 1 The cross-project digest states the same result as a pooled fitness shift of +0.086 [+0.074, +0.098] across 801 AMR genes and 25 organisms, with all 25/25 organisms positive; that statement does not name the comparator, so on its own it does not establish an absolute benefit of deleting an AMR gene. 2
The perturbation scale refines the interpretation of this result. AMR-gene knockouts had average fitness −0.024, whereas the non-AMR knockout background averaged approximately −0.11, producing the relative difference of +0.086. 1 The positive difference means that AMR knockouts were less impaired than the comparison knockout background; it does not mean that they grew faster than an unperturbed reference. 1 Absolute AMR-gene knockout fitness averaged −0.024, and a one-sample Wilcoxon test against zero gave p = 0.999. 1 Thus, removing AMR genes reduced burden relative to the comparison background, not that AMR knockouts had positive absolute fitness. 1
The report describes this relative shift as a baseline AMR burden, but the inference is indirect because it compares knockout classes rather than directly measuring the cost of acquiring or expressing an intact resistance gene. 1 A relative fitness effect is meaningful only with an explicitly defined comparison background: here, AMR and non-AMR knockouts were compared under non-antibiotic conditions, not against an unperturbed wild type. 1 This makes fitness-matched-null-models and condition matching central to deciding whether an apparent cost reflects resistance biology or differences in experiment composition. 1
Conditional benefit under antibiotic exposure
The baseline burden was partially decoupled from antibiotic-dependent importance. 1 Across any-antibiotic experiments, 57.0% of AMR genes showed a fitness flip toward greater importance under antibiotic exposure, with N = 797, mean flip +0.045 defined as non-antibiotic fitness minus antibiotic fitness, and Wilcoxon signed-rank p = 0.0001. 1 The unit of that denominator is ambiguous within the source: the prose describes 57% of AMR genes at N = 797, whereas the report’s NB03 table labels the same 797 denominator as gene-antibiotic pairs, so the flip rate should not be read as a per-gene rate without resolving which unit was counted. 1 A positive flip therefore indicates greater importance of the gene during antibiotic exposure, rather than a positive absolute growth rate. 1 The report proposes that the 43% of AMR genes that did not flip likely reflect narrow-spectrum resistance being tested against non-matching antibiotics rather than non-functional genes; this antibiotic-mismatch explanation is an interpretation, not a tested result. 1 This supports condition-specific-fitness: the same resistance system can carry a modest relative burden without antibiotics yet become important when the relevant environmental challenge is present. 1
The antibiotic-dependent shift varied by resistance mechanism. 1 Broad-spectrum efflux genes had a mean flip of +0.094, compared with −0.001 for enzymatic-inactivation genes, with Mann-Whitney U p = 0.007. 1 The cross-project digest records the same contrast, stating that efflux genes (broad-spectrum) show a stronger antibiotic-dependent fitness flip than enzymatic inactivation genes (narrow-spectrum), +0.094 vs −0.001, MWU p=0.007. 2 This supports the hypothesis that broad-spectrum efflux systems can protect against multiple antibiotic conditions, whereas enzymatic inactivation is more dependent on exposure to a matching antibiotic. 1
A class-matched validation included 157 gene-antibiotic pairs across four resistance classes, with a mean flip of +0.113, a flip rate of 54.8%, and Wilcoxon p = 0.14. 1 Chloramphenicol resistance genes showed the strongest validation, with 6/6 (100%) showing the expected flip. 1 Beta-lactam resistance contributed 105 pairs across 10 organisms and had a 50% flip rate. 1 The report suggests that this beta-lactam rate is consistent with many genes being tested against non-carbenicillin beta-lactams, and attributes the non-significant class-matched result to small per-class samples after matching and to heterogeneity across organisms, some showing strong flips and others near zero; both are proposed rather than established explanations. 1 These results support separating condition-specific-fitness from a pooled baseline contrast: mechanisms had similar modest relative costs without antibiotics, while broad-spectrum efflux systems became important across many antibiotic conditions and narrow-spectrum enzymes generally required matching antibiotics. 1
Matched-antibiotic coverage limits how far that class-matched validation can reach. Only 4 AMR classes — beta-lactam, aminoglycoside, chloramphenicol, and tetracycline — have matched antibiotic experiments. 1 Other classes, including macrolide, glycopeptide, and polymyxin, could not be validated at all, so the class-matched evidence speaks only to the four tested classes and the pooled any-antibiotic signal cannot be assumed to hold for untested ones. 1
The strain-variation atlas refines this condition-dependent interpretation by showing that AMR repertoires are also structured within species: among 37,444 AMR gene-species records, and across 1,305 species and 180,025 genomes, 51.3% of AMR gene-species occurrences were rare, meaning present in <=5% of strains, 41.3% were variable, meaning 5-95% of strains, and only 7.5% were fixed, meaning >=95% of strains, with median variability index 0.526 and median pairwise Jaccard distance 0.435, so strains within the same species share less than 60% of their AMR repertoire. 3 These prevalence patterns do not measure fitness costs, but they support the hypothesis that conditional benefits and lineage or ecological context can produce substantial within-species AMR heterogeneity rather than a uniform species-level phenotype. 3
The broader conservation analysis supports treating conditional fitness as compatible with conservation rather than assuming that condition-specific genes are mainly accessory: across approximately 194,000 genes from 43 bacteria, genes tagged with strong condition-specific effects were 77.3% core versus 70.3% without such phenotypes, and the analysis identified 4,450 neutral-overall but condition-critical genes, 3.0% of core genes versus 1.7% of auxiliary and 1.6% of singleton genes. 4 This is not an AMR-specific estimate, but it supports testing whether antibiotic-conditioned AMR effects are more detectable for conserved genes instead of interpreting conservation as evidence of constitutive, condition-independent importance. 4
Literature cited by a separate project refines the gene-centric framing of resistance used throughout this page. Colistin resistance in Acinetobacter baumannii is mediated by complete loss of lipopolysaccharide production (Moffatt et al., 2010). 5 Septal class A penicillin-binding-protein activity and ld-transpeptidases mediate selection of colistin-resistant, LOS-deficient A. baumannii (Kang et al., 2021). 5 Both are external findings rather than measurements in this corpus; both describe resistance arising from the loss of a cell-envelope structure and the enzymes that compensate for it rather than from acquiring a resistance determinant, which suggests the hypothesis that such loss-of-structure routes fall outside the gene-presence AMR inventories analysed here and connects the conditional-benefit argument to outer-membrane-lipid-homeostasis. 5
Mechanism, conservation, and compensation
Baseline fitness cost did not differ significantly among efflux (N=254), enzymatic inactivation (N=304), metal resistance (N=144), and unknown (N=74) mechanisms. 1 The Kruskal-Wallis test gave H = 0.65, p = 0.89, and the Jonckheere-Terpstra test for the predicted ordering efflux > enzymatic > metal > unknown gave z = 0.23, p = 0.41; no pairwise comparison survived false-discovery-rate correction. 1 These results contradict the prediction that resistance mechanism alone determines baseline fitness cost in this dataset. 1
Mechanism did, however, predict pangenome conservation status. 1 A chi-squared test found a strong association between mechanism and conservation status (χ² = 69.3, p = 1.4×10⁻¹³). 1 Metal-resistance genes were accessory in 44% of cases, compared with 13% of efflux genes and 16% of enzymatic-inactivation genes. 1 This refines the relationship between mechanism and persistence: mechanism predicted where AMR genes occurred in the pangenome, but not their baseline fitness cost. 1 The cross-project digest records the same association in the same terms: mechanism strongly predicts conservation status, metal resistance 44% accessory vs efflux 13%, χ²=69.3, p=1.4e-13. 2
The strain-variation analysis supports this separation between prevalence and measured cost. Atlas Core AMR genes were fixed in 77.3% of cases, whereas 78.7% of Singleton genes were rare; Auxiliary genes were 57.3% variable and 42.7% rare. 3 Thus, conservation class strongly organizes within-species prevalence, while the Fitness Browser comparison found no baseline fitness difference between core or intrinsic and accessory or acquired AMR genes. 1 The two results are compatible because prevalence and knockout fitness are different observables, but neither establishes that conserved genes are intrinsically less costly.
A pangenome-scale AMR atlas refines the mechanism-to-conservation link at finer resolution and shows that mechanism labels aggregate genes with opposite pangenome positions. Not all AMR mechanisms behave the same: beta-lactamases are 54.9% core (p=7.7e-74 for enrichment vs baseline), which the atlas reads as intrinsic resistance genes present in the species’ chromosomal backbone, whereas regulatory genes are only 6.5% core, and known mobile elements such as blaTEM, tet(C), and ant(2”)-Ia are 0% core, that is fully accessory or singleton. 6 Efflux pumps split along the same axis: intrinsic pumps such as emhABC are >95% core, while acquired efflux genes are accessory. 6 This does not contradict the efflux-versus-metal conservation contrast above; it refines it by showing that a single mechanism label can contain both core and accessory members. 6
Linking that atlas to the kescience-fitnessbrowser through a DIAMOND-based link table (177,863 links at 100% sequence identity) identified 178 AMR genes across 37 Fitness Browser organisms, yielding 29,386 fitness measurements; in that comparison 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). 6 This independently supports the direction of the pooled knockout contrast — AMR knockouts less impaired than the non-AMR comparison background — through a different gene-linking route and with medians rather than means. 6 It also refines the core-versus-accessory null, because singleton AMR genes carried the most negative median here while the Fitness Browser AMR stratification found no core/accessory difference (p = 0.33); the analyses used different gene sets, linking thresholds, and summary statistics, so the difference should be investigated rather than averaged away. 6 1 The atlas explains its own result by proposing that the AMR genes present in these predominantly environmental organisms are well-integrated intrinsic resistance genes rather than recently acquired mobile elements — an interpretation of a biased sample rather than a test, and one that bears directly on pangenome-conservation-fitness-decoupling. 6
The atlas reads the core-resident AMR genes as a species’ baseline defensive repertoire: intrinsic resistance genes — beta-lactamases such as ampC, efflux systems such as emhABC, rifampin monooxygenases such as rox — are core-genome residents, vertically inherited, present in >95% of genomes within a species, and, in the report’s wording, impose negligible fitness costs. 6 The negligible-cost component of that reading is an interpretation rather than a per-gene cost measurement, and it sits beside the Fitness Browser result that core or intrinsic and accessory or acquired AMR genes had indistinguishable baseline fitness; together they feed the two-tier genome picture in two-speed-bacterial-genome without establishing that core AMR genes are cheaper. 6 1
The all-gene conservation analysis refines this AMR-specific null result rather than overturning it: essential genes were 82% core versus 66% for always-neutral genes, and core genes were more likely than auxiliary genes to show both negative and positive fitness tails, including 24.4% ever beneficial versus 19.9% for auxiliary genes. 4 Therefore, conservation is weakly associated with broader and stronger measured fitness effects across genes, whereas the AMR subset showed no detectable core/accessory difference in its baseline knockout contrast. 4 1 The distinction may reflect subset-specific power, resistance-mechanism composition, or the fact that the AMR result compares perturbation classes rather than intact-gene costs; it should not be resolved by importing the all-gene gradient into the AMR subset.
Core or intrinsic AMR genes (N=638) and accessory or acquired AMR genes (N=163) had virtually identical fitness distributions, with both means at −0.024, Cohen’s d = 0.002, and Mann-Whitney U p = 0.33. 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 The cross-project digest records the same three nulls, describing the cost as mechanism-independent (KW p=0.89), conservation-independent (core = accessory, p=0.33), and tier-independent (bakta_amr = keyword, p=0.26). 2 This null result supports the possibility that horizontal transfer preferentially retains cost-optimized genes or that compensation occurs rapidly after acquisition, but the evidence is strongest for well-sampled species. 1
The approximately +0.086 common cost may therefore reflect residual metabolic overhead after compensatory evolution in lab-adapted RB-TnSeq strains rather than the unmitigated cost of newly acquired resistance. 1 This is an explanation consistent with the data, not a direct measurement of compensatory mutations, their timing, or their causal effect. 1 All 25 tested organisms were lab-adapted strains, so laboratory history may have reduced measurable AMR costs relative to natural populations. 1 The core-versus-accessory null result is also consistent with cost optimization or post-acquisition compensation, but does not distinguish those explanations. 1 The report also raises, speculatively, that the uniform cost may represent a “floor”: the irreducible metabolic overhead of maintaining any extra gene, regardless of its function. 1
External cost estimates sit on a different measurement scale and should not be read as replications. Melnyk et al. (2015) meta-analyzed ~600 resistance cost measurements and found costs in ~70% of cases with a mean relative fitness of 0.91–0.95 (5–9% cost); those are absolute fitness differences between isogenic resistant and sensitive strain pairs, whereas the +0.086 reported here is a relative difference between AMR and non-AMR transposon knockouts within pooled libraries. 1 The report is explicit that comparisons to literature fitness costs measured by isogenic strain competition are reassuring but not direct equivalences. 1
Three further literature links are offered as explanations of the modest, mechanism-independent cost, and none of them is tested in this corpus. Vanacker, Lenuzza & Rasigade (2023) meta-analyzed fitness costs in E. coli and found that horizontally transferred, plasmid-borne resistance genes are less costly than chromosomal mutations; the report applies that result to argue that many Fitness Browser AMR genes are acquired determinants such as beta-lactamases and acetyltransferases rather than costly target modifications, which is an interpretation of its own gene set rather than a measurement of acquisition mode. 1 Olivares Pacheco & Alvarez (2017) showed that efflux pump costs in P. aeruginosa are metabolic, arising from proton motive force drain, and are rapidly compensated through metabolic rewiring; the report treats compensation as a general outcome to explain why its efflux genes show the same modest cost as other mechanisms, and that generalization across the 25 tested organisms remains interpretive. 1 Levin, Perrot & Walker (2000) demonstrated that compensatory mutations arise faster than reversion to susceptibility, predicting that organisms maintain resistance at reduced cost; the report’s reading of +0.086 as the irreducible cost after compensatory evolution — the minimum overhead of expressing a functional protein that provides no benefit without antibiotics — is speculative. 1
The bias may also run in the opposite direction. Roux et al. (2015) showed that resistance genes, and efflux pumps in particular, can be beneficial in vivo by exporting host antimicrobials, so this in vitro cost measurement may overestimate the true ecological cost. 1
Two later reports interpret the pattern in which mechanism predicts conservation but not cost, and both interpretations remain untested here. The cofitness report cites the contrast of metal-resistance genes at 44% accessory versus efflux genes at 13%, with mechanism not predicting cost, and proposes that cost is determined by the organism’s gene content and growth context whereas conservation is determined by the mechanism’s acquisition history. 7 It places this against prior literature: Martinez & Rojo (2011) reviewed how global metabolic regulators modulate antibiotic susceptibility, which would support the AMR-neighborhood enrichment only if the co-regulation interpretation is correct; Olivares Pacheco et al. (2017) showed that efflux-pump costs in P. aeruginosa are metabolic, arising from proton-motive-force (PMF) drain, and are compensated by metabolic rewiring, a model that could explain efflux–flagellar cofitness because both use PMF, although shared dispensability could produce the same signal; and Eckartt et al. (2024) identified compensatory mutations targeting the same functional pathway as the resistance mutation, which the report treats as consistent with compensation integrating AMR genes into the organism’s existing programs. 7 These literature links support the plausibility of the compensation hypothesis but do not measure compensation in the Fitness Browser strains. 7
The environmental-resistome report extends the same contrast to ecology, proposing that soil and aquatic bacteria invest heavily in metal resistance while clinical and gut bacteria favor target modification (ribosomal protection, PBP alterations) and, to a lesser extent, efflux pumps; it argues that efflux genes are core because they serve host-associated organisms across conditions and that metal resistance is accessory because it is needed only in metal-rich environments. 8 This ecological cause is an interpretation rather than a direct test, and the report describes the cited contrast as “metal 44% accessory vs efflux 13% core”, whereas the cost report gives 13% as the accessory fraction of efflux genes, which it describes as overwhelmingly core. 8 1 Whether clinically enriched AMR genes are costlier, because they were recently acquired, or cheaper, because they have been under strong selection, remains untested; integrating environment-AMR profiles with the fitness-cost data was proposed. 8
Network breadth and fitness cost
The cofitness analysis supports a separation between mechanism, conservation, network breadth, and cost. Across 28 organisms, 769 of the 801 AMR genes with fitness data (96%) had at least one extra-operon cofitness partner at |r| > 0.3. 7 Support-network size was not correlated with AMR fitness cost: Spearman’s rho, a rank-based correlation coefficient, was −0.006, p = 0.87, N = 769. 7 Within mechanisms, rho values were −0.049 for efflux, +0.038 for enzymatic resistance, and −0.031 for metal resistance, all with p > 0.4. 7 The uniform resistance cost of +0.086 was therefore not explained by co-regulatory-neighborhood size, although the cost analysis did not measure network breadth directly. 7 This refines cofitness-network-architecture: the number of genes sharing a fitness-profile pattern and the magnitude of the resistance-associated fitness phenotype were distinct in this dataset, rather than network breadth necessarily imposing a larger cost. 7
The AMR strain atlas adds a different kind of network evidence and refines this conclusion: it detected 1,517 resistance islands across 705 species, with mean size 6.2 genes, median size 4, maximum size 43, and mean phi coefficient 0.827; phi measures pairwise co-occurrence. 3 Of these islands, 1,343/1,517 (88%) contained genes from multiple resistance mechanisms. 3 These tightly co-inherited genomic modules show that physical or occurrence linkage can be extensive even when cofitness-network breadth is unrelated to measured cost. 3 Co-occurrence does not establish co-selection, functional synergy, or a larger fitness burden. 3
The null association does not show that network context is biologically irrelevant, because insufficient variance in fitness cost could limit detection of a relationship. 7 At |r| > 0.3, the mean support network contained 233 genes, while means at |r| > 0.4 and |r| > 0.5 were 110 and 71 genes, respectively. 7 These broad networks may include many weak associations, and confirmation at |r| > 0.4 was identified as necessary. 7
Cofitness measures shared fitness phenotypes rather than direct transcriptional control, so a large network can reflect shared condition-dependent dispensability without implying a proportionally greater cost. 7 AMR-containing ICA (independent component analysis) modules had a median of 46 genes versus 27 for non-AMR modules (MWU p = 1.7×10⁻⁸). 7 Enrichment for flagellar motility and amino acid biosynthesis may therefore represent shared dispensability in shaken, often nutrient-rich laboratory cultures rather than AMR-specific regulation; a mean-fitness-matched permutation is required to distinguish these explanations. 7 The absence of a network-size–cost relationship is a direct null result within this dataset, not a general demonstration that network architecture can never influence fitness cost. 7
The cross-project digest restates the same neighborhood structure — it attributes these results to the cofitness project rather than corroborating them independently — and adds two qualifications. Cofitness analysis of 801 AMR genes across 28 organisms found AMR cofitness neighborhoods enriched for flagellar motility (GO:0071973, 5 organisms FDR<0.05), flagellum assembly (GO:0044780, 5 orgs), histidine biosynthesis (GO:0000105, 3 orgs), and tryptophan biosynthesis (GO:0000162, 3 orgs); enrichment alone does not demonstrate co-regulation. 2 Support networks were organism-specific rather than mechanism-specific, with cross-mechanism Jaccard similarity 0.375 exceeding within-mechanism similarity 0.207 (MWU p=4.3e-13), which supports reading the neighborhoods as a property of genetic background rather than of resistance mechanism. 2 The digest also reports that AMR genes sit in larger-than-average ICA modules (median 46 vs 27, p=1.7e-8) without stating how many AMR genes were assigned to modules at all. 2 That omission is the digest’s, not the underlying study’s: the cofitness report states that only 24% of AMR genes (192/801) are assigned to ICA fitness modules, so the module-size comparison describes that assigned minority rather than the full AMR set. 7
Essentiality and measurement limits
Only 4.6% of AMR genes were absent from the fitness matrices and therefore putatively essential, compared with an approximately 14% background essential rate estimated in a prior analysis using a different organism set. 1 These omitted genes cannot contribute ordinary knockout fitness values and could alter the observed contrast if they are disproportionately costly. 1 These findings connect the AMR-cost result to gene-essentiality while leaving open the possibility that essential AMR genes, if systematically more costly, are underrepresented in the measured distributions. 1 The report treats the low essential rate as an argument against strong right-censoring bias in these distributions. 1 It also states the opposite-facing consequence: if the ~4.6% putatively essential AMR genes are the most costly ones, the +0.086 estimate is a lower bound. 1
Tier 1 genes from bakta_amr (N=110) and Tier 2 keyword-annotated genes (N=691) had indistinguishable fitness distributions (KS p = 0.17), although the broader assembly contained 178 Tier 1 and 1,174 Tier 2 genes. 1 Tier 2 contained 86% of the AMR genes and relied on keyword matching from Bakta annotations, which may include non-AMR genes such as general efflux transporters; the Tier 1 sensitivity analysis gave consistent results. 1 This connects relative-fitness interpretation to annotation-dependent resistome inference. 1
Transposon insertions can have polar effects on downstream genes in operons, potentially confounding attribution of a fitness phenotype to the targeted AMR gene itself. 1 The cofitness analysis used fitness profiles rather than a direct measure of metabolic burden. 7 Missing fitness values were treated as zero in z-score space through np.nan_to_num, which approximates but does not equal pairwise-complete Pearson correlation, although the report judged the dense Fitness Browser matrices unlikely to substantially alter the conclusions. 7
Core/accessory labels used a ≥95% prevalence threshold, while most Fitness Browser species had few GTDB genomes: the median was 9, with a range of 2–399. 1 A gene present in all 9 sampled genomes could therefore be labeled core at the available sampling depth but become accessory with broader sampling, making the core-versus-accessory null result especially cautious for species with fewer than 20 genomes. 1 The product classifier did not handle fosfomycin or tellurite resistance annotations, leaving approximately 25 genes in the unknown mechanism category; reclassification would move them to enzymatic inactivation and metal resistance, respectively. 1
The AMR atlas provides a complementary sampling warning: its collection included 180,025 genomes, but the GTDB/NCBI set was biased toward clinical and human-associated isolates, and 52.7% of genomes had no classifiable isolation source. 3 Host-associated species carried more AMR genes per genome than terrestrial or aquatic species under both approximate environment classifiers, with human-clinical isolates highest; this is ecological association, not evidence of a larger per-gene fitness cost. 3
The pangenome atlas adds its own coverage warnings to the same list. Only 37/48 Fitness Browser organisms had AMR genes, and these are predominantly environmental strains, so the fitness analysis does not capture the cost of recently acquired mobile resistance in pathogens. 6 The 100% DIAMOND identity threshold used for Fitness Browser-to-pangenome linking is conservative: it avoids paralog confusion but may miss closely related allelic variants, such as alleles differing by a single synonymous substitution, and so may undercount AMR gene fitness effects. 6 Because intrinsic resistance predominates in environmental isolates, the atlas cautions that the fitness cost of recently acquired mobile resistance elements in clinical pathogens may differ substantially, which places this evidence inside laboratory-fitness-versus-natural-selection. 6
A prophage co-mobilization analysis could not test AMR fitness costs at all, so its silence is a coverage null rather than a biological one. Its fitness dataset covered 48 model organisms with RB-TnSeq data, which do not overlap well with the GTDB pangenome species analyzed there. 9 Its hypothesis H3 — whether prophage-proximal AMR genes have distinct fitness costs — could not be tested and remains an open question for future work with expanded fitness datasets. 9 The report attributes that gap explicitly to limited overlap between fitness browser organisms and GTDB species, an instance of genetic-perturbation-coverage-bias rather than evidence that prophage context is irrelevant to cost. 9
The 28 cofitness organisms were lab-adapted, phylogenetically biased, included many Pseudomonas organisms, and had limited ecological diversity, restricting ecological generalization. 7 The independent AMR cost estimate likewise covered 25 lab-adapted strains and cautions that compensation may have reduced measurable costs relative to wild populations. 1 These limitations connect the result to condition-space-dimensionality and to perturbation-modality-dependent phenotypic architecture. 1
Tensions
The main tension is between the strong any-antibiotic signal and the weaker class-matched validation. 1 The any-antibiotic analysis found a 57.0% flip rate for N = 797 records — a denominator the source labels as genes in its prose and as gene-antibiotic pairs in its NB03 table, as noted above — with p = 0.0001, whereas the class-matched analysis found a 54.8% flip rate for 157 pairs with p = 0.14. 1 This does not establish opposing biological effects, because the analyses differ in sample size and matching design; it identifies a need to determine whether broad condition coverage or antibiotic-class specificity best explains the stronger result. 1 The class-matched analysis was less powered than the any-antibiotic analysis, so the positive pooled signal should not be treated as equally strong for every drug-resistance class. 1
A second tension concerns the absence of baseline cost differences among mechanisms alongside the strong mechanism-by-conservation association. 1 Mechanism was not associated with baseline cost (H = 0.65, p = 0.89) but was strongly associated with conservation status (χ² = 69.3, p = 1.4×10⁻¹³). 1 The evidence therefore supports different roles for mechanism in fitness and pangenome distribution rather than a single mechanism-to-cost-to-retention pathway. 1
A third tension concerns the AMR-specific conservation null versus the all-gene conservation gradient. The all-gene analysis found higher core representation among essential and broadly fitness-active genes, plus stronger positive and negative fitness tails among core genes. 4 In contrast, core or intrinsic and accessory or acquired AMR genes had indistinguishable baseline fitness distributions (d = 0.002, p = 0.33). 4 1 These results address different scopes and observables, but resolving whether AMR genes are an exception requires a larger, condition-matched AMR subset with comparable coverage of core, accessory, and singleton genes.
A fourth tension concerns whether AMR-associated cofitness enrichment is mechanistic or reflects shared dispensability. Flagellar motility, chemotaxis, and amino acid biosynthesis were enriched in AMR support neighborhoods, but the cofitness analysis identifies shared dispensability under laboratory conditions as an alternative explanation and reports no network-size relationship with cost. 7 This does not contradict the measured AMR fitness burden or antibiotic-dependent benefit; it limits the inference that cofitness-network structure explains that burden. 7
The resistance-island result introduces a related measurement tension rather than resolving the third one: mean phi 0.827 and 88% multi-mechanism islands indicate strong co-inheritance, but the atlas explicitly cautions that linkage on mobile genetic elements does not prove co-selection or functional synergy. 3 Thus, genomic co-inheritance can coexist with a null cofitness-size–cost relationship; the two analyses should not be treated as interchangeable measures of network architecture. 3 7
Open Directions
- Use isogenic strains with direct growth-rate or competition assays to test whether the +0.086 knockout-class contrast corresponds to an absolute cost of intact AMR genes rather than only a relative perturbation shift. 1
- Use organism-stratified models on Cup4G11 (77 AMR genes) and BFirm (50 AMR genes) to test mechanism effects while controlling for genetic background. 1
- Use matched genomes from the 25 lab-adapted organisms, comparative genomics, and fitness association analysis to ask whether specific compensatory mutations explain the residual +0.086 shift. 1
- Reconstruct newly acquired and compensated AMR genotypes in isogenic strains and measure growth and competition fitness to distinguish original from post-compensation costs. 1
- Model insertion position and operon context, using gene-neighborhood information to test how much of the measured AMR phenotype could arise from polar effects on downstream genes. 1
- Incorporate the approximately 4.6% putatively essential AMR genes through targeted essentiality assays or alternative perturbations to test whether their exclusion makes +0.086 a lower bound. 1
- Reanalyze the 801 non-antibiotic per-gene fitness records with matched metal, osmotic, and carbon-limitation conditions to ask whether the baseline contrast changes across stress environments. 1
- Subclassify efflux pumps into narrow-spectrum drug pumps and general RND systems such as AcrAB-TolC, then test whether constitutively expressed systems have lower baseline costs and whether breadth predicts antibiotic-dependent importance using condition-matched assays. 1
- Replace averages across non-antibiotic experiments with condition-specific analyses of metal, osmotic, and carbon-limitation stresses to test whether baseline cost changes by environment. 1
- Cross-reference the 144 metal-resistance genes with the metal fitness atlas and matched metal-stress measurements to test whether genes costly under standard conditions are protective under metal stress. 1
- Expand GTDB sampling beyond the median of 9 genomes per Fitness Browser species and recompute the ≥95% core/accessory labels to ask whether the cost null result persists with deeper pangenomes. 1
- Use the 1,517 resistance-island records to map islands to plasmids, chromosomes, integrons, and insertion sequences, then test whether physical linkage predicts antibiotic-conditioned fitness or persistence beyond cofitness and prevalence effects. 3
- Use mean-fitness-matched permutations, including random non-AMR genes in the −0.05 to +0.05 range, to test whether AMR-neighborhood structure exceeds what is expected from fitness level alone. 7
- Recompute cofitness separately for antibiotic-treatment and standard-growth conditions, and repeat network-size–cost analyses at |r| > 0.4 and |r| > 0.5. 7
- Quantify variance in AMR fitness costs and use power analysis or a hierarchical model to ask whether the null result reflects insufficient variation across genes, mechanisms, or organisms. 7
- Compare cofitness network breadth with direct regulatory evidence and Pfam-domain enrichment to test whether broad networks reflect shared cofitness, co-regulation, or common functional architecture. 7
- Within organisms with dense AMR and cofitness data, test network breadth, mechanism, and core/accessory status jointly under matched conditions; directly measure flagellar-knockout fitness to distinguish shared dispensability from co-regulation. 1 7
- Test whether the AMR ecotypes identified in 190/974 species predict condition-specific fitness, virulence profiles, or metabolic variation after controlling for clinical and sampling bias. 3
- Extend the analysis from 25 Fitness Browser organisms to all 293K KBase Data Lakehouse genomes by predicting AMR cost from gene-cluster conservation patterns and testing whether the laboratory-derived relationship generalizes across the broader genome collection. 1
- Reanalyze AMR genes against the all-gene conservation framework, stratifying by core, auxiliary, and singleton status, fitness breadth, and antibiotic condition, to test whether the AMR-specific null differs from the broader core-gene fitness gradient. 4
- Combine environment-AMR profiles from the environmental-resistome analysis with per-gene fitness-cost estimates to test whether AMR genes enriched in clinical settings are costlier (recent acquisition) or cheaper (strong selection). 8
- Use the full Fitness Browser-pangenome link table to test AMR fitness costs specifically under antibiotic stress conditions rather than standard lab media alone, closing the gap left by the atlas’s standard-media median comparison. 6
- Revisit the untested H3 hypothesis as fitness browser coverage expands, testing whether prophage-proximal AMR genes have distinct fitness costs once organism overlap with GTDB pangenome species is sufficient. 9
A reported positive or negative transposon fitness contrast should state its perturbation, reference population, condition set, and scale before being called a biological cost or benefit. 1 For this dataset, the most defensible wording is that AMR knockouts showed a positive relative shift against the non-AMR knockout background under non-antibiotic conditions, while AMR genes often became more important under antibiotic exposure. 1 The all-gene conservation analysis refines this interpretation by showing that conservation can accompany both broad and conditional fitness effects, without establishing that the AMR-specific baseline cost differs by conservation class. 4 Within-species atlas data further indicate that the ecological and lineage distribution of AMR repertoires is heterogeneous, but do not convert prevalence or co-inheritance into a direct fitness estimate. 3 The source reports and their detailed evidence are available in Fitness Cost of Antimicrobial Resistance Genes, AMR Co-Fitness Support Networks, Within-Species AMR Strain Variation, and Fitness Effects vs Conservation — Quantitative Analysis. 1 7 3 4