Metal cross-resistance across bacterial fitness landscapes

Metal cross-resistance is a broadly positive, gene-level relationship in which bacterial fitness responses to one metal tend to predict responses to another, while the magnitude of that relationship varies with metal chemistry. ₁ This concept connects Gene-Resolution Metal Cross-Resistance Across Diverse Bacteria and Pan-Bacterial Metal Fitness Atlas to Metal-Specific vs General Stress Genes, Metal Resistance Ecology — Phylogenetic Conservation vs. Environmental Selection, Soil Metal Concentrations Drive Functional Gene Shifts, T4SS–CAZy Environmental HGT, cofitness-network-architecture, pangenome-integration, environmental-resistome, and condition-specific-fitness.

Evidence for a universal directional layer

The study analyzed 452 metal experiments across 37 organisms and 14 metals, with 119,561 genes having metal fitness data from 28 organisms tested against at least 3 metals. ₁ Across 317 organism–metal-pair observations involving 28 organisms and 85 unique metal pairs, 98.1% of gene-level fitness correlations were positive (311/317), and 99.1% were statistically significant at p < 0.05. ₁ All 15 metal pairs tested in at least 5 organisms showed greater than 90% sign consistency, and no metal pair showed systematically negative cross-resistance in any organism. ₁ This supports a universal directional layer in which metal conditions tend to share the direction of gene-importance effects rather than producing consistently opposing fitness landscapes. ₁ The newer Pan-Bacterial Metal Fitness Atlas supports the breadth of this interpretation at a larger atlas scale: it assembled 559 metal-related experiments across 31 organisms and 16 metals, with 383,349 gene × metal fitness records. ₂

The report attributes this shared direction to common cellular disruptions involving protein stability, DNA integrity, membrane function, and cofactor insertion. ₁ Because the study did not include non-metal stress controls, the evidence cannot distinguish universal metal cross-resistance from a broader general-stress response to all tested conditions. ₁ The atlas likewise finds that broad metal-important genes include general stress functions, so this refines rather than resolves the causal interpretation. ₂ The new specificity analysis supports the importance of this distinction: among 7,609 metal-important gene records, 4,177 (54.9%) were metal-specific under a <5% sick-rate threshold across 5,945 non-metal experiments, while 2,888 (38.0%) were general sick and 544 (7.2%) were metal+stress. ₃ Thus, the positive cross-metal layer contains both metal-specific determinants and broadly stress-sensitive genes rather than representing one homogeneous response. ₃ The central discoveries digest reports the same partition for 7,609 metal-important gene records across 24 organisms. In that partition, 4,177 (55%) had a <5% sick rate across 5,945 non-metal experiments, 38% were general sick and 7% were metal+stress. The digest reads this as showing that the atlas gene set is not dominated by general stress response. ₄

The MicrobeAtlas analysis refines the ecological interpretation of this directional layer rather than testing gene-level correlation directly: across 606 genera, metal type diversity predicted broader inferred niche breadth after phylogenetic correction, whereas total AMR cluster count and core AMR fraction did not. ₅ This supports the idea that breadth across metal types, rather than resistance-gene burden alone, is the ecologically relevant dimension, but the cross-sectional result cannot establish whether resistance expands niche breadth or broad niches promote resistance acquisition. ₅

Overlap with osmotic and ionic stress

The counter-ion analysis refines the shared-stress ambiguity with an explicit osmotic/ionic comparison. Across 19 organisms and 14 metals (86 organism × metal pairs), 4,304 of 10,821 metal-important gene records (39.8%) were also important under NaCl stress. Overlap was present for every metal tested, ranging from 9.2% for molybdenum to 57.6% for manganese. ₆ The report reads this as showing that a large fraction of the metal fitness signal reflects general cellular vulnerability shared between metal and osmotic/ionic stress. The pooled 39.8% is not a per-metal mean. ₆ The discoveries digest attributes the overlap to shared stress biology (cell envelope, ion homeostasis, DNA repair) rather than to chloride counter-ion contamination. As evidence, it cites zinc sulfate (0 mM Cl⁻), which showed 44.6% NaCl overlap and the highest DvH profile correlation (r=0.715), higher than most chloride-delivered metals. ₄

In DvH, profiled with 13 metals and 6 NaCl experiments, whole-genome metal–NaCl fitness correlations formed a hierarchy. Zinc ranked first (r=0.715), followed by manganese (0.545), copper (0.532) and cobalt (0.498), with tungsten (0.298) and iron (0.086) at the bottom. Zinc ranked first even though it was delivered as sulfate with zero Cl⁻. ₆ Across the 14 metals, zinc (sulfate, zero chloride) ranked 4th in NaCl overlap and uranium (acetate, zero chloride) ranked 8th. ₆ The report and the digest interpret the ordering as following toxicity mechanism. In this reading, metals that broadly displace cofactors (Zn, Mn, Cu, Co) share more genes with ionic stress than pathway-targeted metals (Mo/W for molybdopterin enzymes, Fe for iron-sulfur clusters). This is an interpretation, not a direct test. _6,4 The digest reports the same hierarchy in rounded form, from Zn (r=0.72) to Fe (r=0.09). ₄ Manganese, cadmium, selenium, mercury, iron, molybdenum and tungsten were each tested only in DvH, so their overlap statistics have no cross-organism replication. ₆ The general-stress axis itself appears graded by metal. This suggests the hypothesis that part of the cross-metal correlation among cofactor-displacing metals reflects a shared NaCl-like component, but the cross-resistance study included no non-metal stress controls to test it. _6,1 The cross-resistance report supports the mechanism reading from a different angle. It cites the counter-ion finding that DvH’s metal–NaCl correlation hierarchy follows toxicity mechanism rather than counter-ion identity. It then states that the same mechanistic grouping emerges from its own metal–metal correlations across 28 organisms, and treats this as confirming and extending that result. The two reports agree in interpretation, but this is not an independent test of mechanism. ₁

Chemistry-dependent magnitudes

The strongest consensus associations were Fe–Zn (mean r = 0.61, n = 6 organisms), Co–Ni (r = 0.56, n = 28), Co–Zn (r = 0.52, n = 18), Ni–Zn (r = 0.51, n = 18), Cu–Zn (r = 0.48, n = 16), Cu–Fe (r = 0.45, n = 7), Al–Zn (r = 0.44, n = 13), Co–Fe (r = 0.45, n = 7), Co–Cu (r = 0.43, n = 24), and Cu–Ni (r = 0.43, n = 24). ₁ Al was the most independent metal, with a mean r = 0.34, while Al–Ni and Al–Co were the weakest listed pairs at r = 0.34 and r = 0.30, respectively. ₁ Sign consistency was 100% for Fe–Zn, Co–Ni, Co–Zn, Ni–Zn, Cu–Zn, Cu–Fe, Al–Zn and Co–Fe. It was 96% for Co–Cu and Cu–Ni, 94% for Al–Cu (r = 0.34, n = 17), and 95% for Al–Ni (n = 20) and Al–Co (n = 20). Al–Fe (r = 0.38, n = 5) and Fe–Ni (r = 0.38, n = 7) were each 100% consistent. ₁ The report proposes shared disruption of iron-sulfur cluster proteins as an explanation for the unexpectedly strong Fe–Zn association, but does not demonstrate it. ₁ Cu–U had mean r = 0.51 (n = 5 organisms). The report attributes this to both metals causing membrane/oxidative damage, but that attribution is an interpretation, not a tested mechanism. ₁ The report likewise reads aluminium’s independence (mean r = 0.34) as consistent with its unique trivalent toxicity mechanism. This is also an interpretation. ₁

These results support a second, chemistry-specific layer: divalent cations that compete for binding sites may produce stronger associations, whereas metals with more distinctive toxicity mechanisms may produce weaker associations. ₁ The report separates two layers. The direction of cross-resistance (all positive) is universal, while the quantitative magnitude of the chemistry-specific layer is only moderately conserved across organisms, with leave-one-out consensus prediction yielding r = 0.41 and the Mantel mean yielding r = 0.23. ₁

The cross-resistance direction remained consistent across organisms from Proteobacteria, Bacteroidetes, Firmicutes, and Actinobacteria. ₁ Co–Ni had median r approximately 0.58 across 28 organisms, whereas Al–Co and Al–Ni had median r approximately 0.30. ₁ Individual-organism matrices repeatedly placed Ni–Co and Co–Zn among the strongest blocks, while Mo was the most independent metal in the 13-metal DvH dataset. ₁ The report links the Co–Ni result to classical Co–Ni–Zn cross-resistance mediated by CzcCBA efflux systems. It argues that its genome-wide Co–Ni correlation (r = 0.56, n = 28 organisms) extends that pattern beyond efflux genes to the entire genome. ₁

Pairwise Spearman correlations of metal-pair rankings across 10 common metal pairs and 12 organisms with at least 5 metals showed mostly positive agreement. ₁ Mantel tests across 351 organism pairs produced a mean r = 0.23, with 62% positive values. ₁ The metal-label permutation test was non-significant (p = 0.42), because all metal pairs were positive and label shuffling did not change the mean. ₁ Leave-one-out consensus prediction had mean r = 0.41, although only 2/28 organisms reached individual significance. ₁

The atlas refines this chemistry-level picture by showing stronger core enrichment for essential metals than toxic metals: the mean core-fraction delta was +0.148 for Fe, Mo, W, Se, and Mn versus +0.081 for toxic metals, with one-sided Mann-Whitney U = 39 and p = 0.015. ₂ This association is conservation-based rather than a replacement for the pairwise correlation analysis, and rare-metal coverage remains uneven. ₂ Specificity also varied by metal, ranging from 60.6% for manganese (20/33) and 60.5% for molybdenum (185/306) to 21.9% for iron (144/659), showing that the chemistry-dependent layer extends to the fraction of determinants classified as metal-specific. ₃ The report also gives metal-specific fractions of 56.5% for tungsten and 46.4% for selenium. It describes iron, at 21.9%, as the lowest and suggests this likely reflects iron’s central role in core metabolism. That explanation is proposed, not tested. ₃

MicrobeAtlas provides a complementary, genus-level result: metal type diversity had a positive PGLS coefficient for Levins’ B_std niche breadth (β = +0.021, SE = 0.0056, p = 1.5×10⁻⁴), while total AMR cluster count had β = −0.002, SE = 0.0065, p = 0.789 and core AMR fraction had β = +0.006, SE = 0.0054, p = 0.288. ₅ The association remained after controlling for genome size (metal type β = +0.0218, SE = 0.0061, p = 3.6×10⁻⁴), but the strict 5% within-environment prevalence analysis reduced the sample from 606 to 379 genera and yielded p = 0.092. ₅ This supports a distributed breadth signal while refining its strength: inferred ecological range is not equivalent to demonstrated metal-specific adaptation. ₅ Statistical significance was lost when individual metals were removed. Each single-metal exclusion at n = 606 left a positive coefficient, but none was significant after BH-FDR (Benjamini–Hochberg false discovery rate) correction: excluding Hg gave β = +0.0082 (SE 0.0069, p = 0.233, q = 0.313); As, β = +0.0039 (SE 0.0061, p = 0.518, q = 0.572); Cu, β = +0.0102 (SE 0.0065, p = 0.115, q = 0.187); Zn, β = +0.0095 (SE 0.0064, p = 0.140, q = 0.212); Cd, β = +0.0084 (SE 0.0064, p = 0.186, q = 0.258); Cr, β = +0.0088 (SE 0.0064, p = 0.173, q = 0.255); and Ni, β = +0.0095 (SE 0.0064, p = 0.140, q = 0.212). ₅ The report explicitly leaves the cause unresolved. The lost significance could reflect a genuinely distributed signal to which each metal type contributes independently. It could also reflect a near-threshold original finding pushed below significance because each exclusion compresses the predictor range and costs statistical power. The positive direction in all seven exclusions favours the first reading, but the two cannot be distinguished without a larger metal-type vocabulary. This limits claims that the association is robust to removing individual metals, without showing that the association is absent. ₅

Three-tier gene architecture

Among 8,162 metal-important genes across 28 organisms, 1,484 (18.2%) were classified as general stress genes, 2,306 (28.3%) as metal-shared genes, and 4,372 (53.6%) as metal-specific genes. ₁ Here, a pangenome means the combined gene repertoire represented across the organisms or strains being compared. ₁

Mean pangenome core fractions were 92.0% for general stress genes (1,484 genes; 18.2% of metal-important genes), 91.0% for metal-shared genes (2,306; 28.3%), and 89.8% for metal-specific genes (4,372; 53.6%). ₁ The percentages fully core at at least 95% were 57.2%, 50.4%, and 45.7%, respectively. ₁ The report describes this fully core gradient as an 11.5 percentage point spread. ₁

This gradient supports, but does not by itself prove, an evolutionary model progressing from ancestral general stress defense to shared metal defense and then to more accessory, faster-evolving specialized metal-specific resistance. ₁ General stress genes were enriched for energy/respiration and cell-envelope functions, whereas metal-specific genes were enriched for transporters/efflux and iron/metal-related functions. ₁

The specificity analysis refines this three-tier model using explicit non-metal controls. Across 22 organisms with pangenome links, metal-specific genes had a pooled core fraction of 84.8% (2,969/3,500) and an organism-mean core fraction of 88.0%, with a mean delta versus baseline of +6.9%; general sick genes had pooled and organism-mean core fractions of 90.2% (2,183/2,420) and a mean delta of +9.0%. ₃ Metal-specific genes were less core-enriched than general sick genes, and the Cochran-Mantel-Haenszel comparison was significant (p=0.011), while both categories remained strongly core-associated. ₃ This supports a modest two-tier interpretation within the broader three-tier architecture: broadly stress-sensitive functions are more core-enriched than specialized metal determinants, but specialized determinants are not predominantly accessory.

The atlas supports the core-heavy end of this model at genome-wide scale: across 22 organisms and 14 metals, broad metal-important genes were 87.4% core versus 76.9% for baseline genes (OR=2.08, p=4.3e-162). ₂ It also refines the comparison by distinguishing this broad definition from the prior condition-specific heavy-metal result of 71.2% core: the atlas includes general cellular processes vulnerable to metal disruption, whereas specialized resistance is more visible after such functions are excluded. ₂ The discoveries digest supports the atlas result after an explicit shared-stress correction. After the ~40% of metal-important genes that overlap with NaCl stress were removed, core genome enrichment persisted for 12 of 14 metals. Metals the digest groups as essential (Mo, W, Se, Hg) showed stronger enrichment after correction (e.g., Mo delta +0.132 → +0.145), and the digest judged that the original atlas conclusion (OR=2.08) was not an artifact. ₄ The digest’s grouping of Hg with essential metals differs from the atlas’s essential-versus-toxic comparison, which used Fe, Mo, W, Se and Mn as the essential set. Both classifications are retained here rather than reconciled, and the digest’s per-metal deltas should not be merged with the atlas’s group-level means. _4,2

The analysis identified 318 ortholog groups that were metal-shared, meaning important for at least 2 metals, in at least 2 organisms. ₁ The most broadly conserved families spanned up to 14 organisms and included cell-envelope, energy-metabolism, DNA-repair, and ion-homeostasis functions. ₁ These results refine pangenome-integration by linking pangenome conservation to tiers of cross-metal gene importance rather than treating all metal-associated genes as one class. ₁ Consistently, the atlas found 1,182 ortholog groups with metal phenotypes conserved across at least 2 organisms and 601 across at least 3, while 149 conserved candidates lacked full functional annotation. ₂ In the atlas, 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, consistent with fundamental cellular processes. The atlas’s 17-organism span and the cross-resistance report’s span of up to 14 organisms come from different analyses and are not reconciled here. _2,1 On the DNA-repair component, the specificity report cites Wu et al. (2019), which reported that the RuvRCAB operon contributes to resistance against Cr(VI), As(III), Sb(III), and Cd(II). That is cited literature, not a result established in this corpus. ₃

Metal-specific genes matched metal-resistance keywords at 12.2% versus 7.8% for general sick genes (Fisher exact OR=1.64, p=2.4e-8), which supports their interpretation as a distinct resistance-enriched class. ₃ Candidate prioritization further identified UCP030820, YebC, and DUF1043/YhcB as the strongest novel families because they combined 50–67% metal-specificity with lower pleiotropic effects across multiple species; the proposed YebC translation-bottleneck mechanism remains a hypothesis. ₃ The discoveries digest gives per-family values for the three leading candidates: UCP030820/OG01015 (67% metal-specific, oxidoreductase, 7 metals), YebC/OG01383 (58%, transcriptional regulator/translation factor, 6 metals, 11 organisms) and DUF1043-YhcB/OG03264 (50%, cell division protein, 5 metals). The digest contrasts them with pleiotropic YfdZ (15%) and Mla/Yrb (0-25%), and with DUF39 (0%, sick rate 0.64), which is a general fitness factor despite spanning 8 metals. ₄ The digest’s YebC hypothesis is untested. It proposes that metal stress up-regulates proline-rich metal transporters such as the CopA/ZntA P-type ATPases and CDF transporters, creating a translation bottleneck. YebC, recently described as a translation factor for proline-rich proteins, would resolve that bottleneck, which would explain why it is dispensable under non-metal conditions. ₄ The specificity report describes UCP030820 as an oxidoreductase involved in sulfite reduction and important for 7 metals, including Cd and Cr. ₃

Environmental validation and its limits

The study tested whether multi-metal tolerance predicted isolation from metal-associated environments using bacdive and Fitness Browser data. ₁ Multi-metal tolerance scores did not correlate with BacDive isolation from metal environments at the Fitness Browser species scale (Spearman rho approximately -0.02, p > 0.8). ₁

After excluding 2 organisms without tier-classification data and collapsing multiple Fitness Browser strains from the same species, the effective sample size was 20 independent species, which the report considers too small for a meaningful correlation test. ₁ Genus-plus-species substring matching was imprecise for organisms identified only to genus level, including Acidovorax sp., so those fuzzy matches require caution. ₁ This result limits the environmental inference in environmental-resistome rather than disproving an association between cross-resistance gene signatures and metal-associated habitats. ₁

The MicrobeAtlas study supports this cautious interpretation with a larger environmental comparison, but not with a direct test of isolation phenotype. Across 767 genera detected in 1,624 groundwater samples, metal type diversity correlated with groundwater prevalence (Spearman ρ = +0.112, p = 0.0019), while groundwater-specific fold-enrichment was null (ρ = +0.042, p = 0.242). ₅ Thus, broader metal repertoires track occurrence in groundwater at a weak prevalence level, but the result does not demonstrate groundwater-specific enrichment or a contamination mechanism. ₅

A second MicrobeAtlas validation found community-weighted mean metal-type diversity differed across 8 wells (Kruskal–Wallis H = 29.10, p = 0.0001), increased with time point (Spearman ρ = +0.383, p < 0.0001), and increased within FW216 (ρ = +0.576, p = 0.0001). ₅ However, only 16.8% of reads were joinable to genus-level metal-diversity data, and the report notes that the dominant Sulfurimonas trend decreased with time, illustrating that ecological dominance can be driven by electron-acceptor availability rather than metal resistance. ₅ This refines the environmental-resistome interpretation: community-level metal-diversity gradients are detectable, but incomplete taxonomic coverage and competing environmental controls prevent treating them as resistance-driven community assembly.

The atlas extends the environmental question to pangenome-scale prediction but does not establish a simple gene-presence predictor: a 1,286-KEGG-KO metal signature scored 27,702 pangenome species, and genome-size-normalized scores placed Leptospirillum at the 91st percentile, Acidithiobacillus at the 77th, Marinobacter at the 75th, and Sulfobacillus at the 71st; bioleaching genera were not significantly enriched over background (Mann-Whitney p=0.17). ₂ The simple repertoire score failed to predict metal fitness, supporting the need to distinguish cross-resistance fitness effects from gene presence alone. ₂

The new soil functional-genomics analysis refines these environmental validations by measuring associations between metal concentrations and community functional gene content rather than isolation or prevalence directly. Across 51,748 soil samples and nine metals—copper, cobalt, chromium, nickel, zinc, lead, arsenic, cadmium, and mercury—it identified 2,355 significant COG–metal associations at FDR < 0.05; COG means a Cluster of Orthologous Groups functional category and FDR means false discovery rate. Chromium and lead produced the strongest signals, with transporters including ABC and RND systems and biosynthesis genes dominating the top hits. ₇ A db-RDA model, where db-RDA is distance-based redundancy analysis, yielded R² = 0.799 and p = 0.005 using 999 permutations; after conditioning on batch and project effects, metal concentrations explained 80% of variance in community COG profiles. ₇ Biome-stratified PGLS, or phylogenetic generalized least squares, found distinct metal–COG relationships in soil, marine, and wastewater environments, supporting environment-specific functional responses rather than a universal resistance programme. ₇ A copper-specific analysis found 116 COGs at FDR < 0.05 among 7,566 samples with nearby KBase genomes at 10 km. Its top positives included cell division and nucleotide transport (BQ, FQ, FK, CF), and its top negatives included energy production (DI, CE). The report reads this as suggesting, not establishing, energetic trade-offs under copper stress. It treats the energetic trade-offs and membrane-transport enrichment as consistent with known bacterial copper-toxicity mechanisms. ₇

This soil result supports the environmental-resistome concept by showing that environmental metal gradients coincide with functional shifts, but it does not validate the fitness-based cross-resistance architecture or establish metal-specific resistance. The analysis is observational; chromium, copper, lead, and zinc co-vary in many industrial soils, so individual associations may reflect generic multi-metal contamination. The conditional db-RDA R² = 0.799 describes residual variance after project accession was removed and may be substantially higher than the unconditional metal-only R²; the latter was not reported. ₇

The T4SS–CAZy analysis extends this environmental evidence from community functional profiles to gene-mobility neighborhoods. In 376 genomes, GT2-neighborhood MAGs had a mean of 0.045 metal-resistance types versus 0.004 for non-GT2 MAGs (n=260,276), with Mann–Whitney p=8.6e-27; genomes with GT2 in T4SS-proximal neighborhoods carried 11× more metal-resistance genes. ₈ This supports a link between environmental horizontal-transfer hubs and resistance breadth, but the association is observational and does not establish that T4SS transfer causes metal-resistance acquisition. ₈

Field genus abundance adds a separate, observational uranium axis. Herbaspirillum (Spearman rho +0.336; p=3.8e-4; FDR q=0.001) and Bacteroides (rho +0.264; p=0.006; FDR q=0.013) were more abundant at high-uranium sites, whereas Caulobacter was less abundant (rho -0.411; p=1.0e-5; FDR q=1.1e-4). ₉ These correlations concern genus relative abundance along a uranium gradient, not gene-level cross-metal fitness. They therefore neither directly support nor contradict the cross-resistance architecture. ₉

Tensions

The species-scale BacDive validation was null, with Spearman rho approximately -0.02 and p > 0.8, whereas the prior BacDive Isolation Environment × Metal Tolerance Prediction project reported Cohen’s d = +1.0 from pangenome-scale analysis of 42K strains. ₁ The report argues that a proper test would apply the KEGG/PFAM mapping approach from the Metal Fitness Atlas to the cross-resistance gene signatures across 27K species. ₁ The results therefore leave unresolved whether the discrepancy reflects insufficient power and imprecise matching in the species-scale test or a genuine difference between multi-metal fitness signatures and environmental isolation. ₁

The MicrobeAtlas groundwater prevalence association (ρ = +0.112, p = 0.0019) supports a weak environmental occurrence relationship, but its null groundwater-specific fold-enrichment test (ρ = +0.042, p = 0.242) does not resolve the BacDive tension. ₅ The datasets differ in environmental definition, taxonomic aggregation, and outcome, so they should not be combined into a single validation estimate. ₅

The soil analysis adds a related tension: its conditional db-RDA model reports R² = 0.799 and p = 0.005 after conditioning on batch and project effects, whereas the BacDive and groundwater tests address environmental occurrence and isolation rather than residual community functional variance. ₇ These results cannot be treated as convergent effect estimates; the soil signal may partly reflect project structure and has not yet been spatially or effect-size validated. ₇

The T4SS result adds a related observational association rather than resolving this tension: GT2-neighborhood MAGs showed 0.045 versus 0.004 metal-resistance types and an 11× difference in metal-resistance genes, but these data concern syntenic gene neighborhoods and not fitness, isolation, or prevalence outcomes. ₈ Thus, the new result supports environmental coupling between mobility and resistance breadth while remaining non-comparable to the BacDive, groundwater, and soil effect estimates. ₈

The report also identifies a tension between strong universal positivity and uncertainty about its cause: all tested metal pairs were positive, but no non-metal stress controls were included. ₁ The Counter Ion Effects on Metal Fitness Measurements project is identified as partially addressing this issue, but the present study cannot independently separate metal-specific cross-resistance from a general-stress response. ₁ The specificity analysis supports the existence of a substantial metal-specific component, but also found 38.0% general-sick records and 7.2% metal+stress records, so it does not eliminate the causal ambiguity. ₃

The soil study reinforces this causal tension: its 2,355 discoveries among 3,915 implied tests represent a 60% discovery rate, but positive correlation among co-contaminating metals may make Benjamini–Hochberg FDR correction anti-conservative and the true FDR higher than reported. ₇ Effect sizes were not systematically reported, and planned audits will flag associations with rho < 0.05; therefore statistical significance does not yet demonstrate substantial biological effect. ₇

A further definitional tension is that the three-tier analysis found metal-specific genes to be 89.8% core, while the atlas found broad metal-important genes to be 87.4% core versus 76.9% baseline and explicitly attributes much of its signal to general cellular vulnerability. _1,2 These values should not be reconciled as a single estimate: they use different gene classes, organism sets, and analytical definitions. _1,2 The specificity analysis adds a third estimate—84.8% pooled core for metal-specific genes and 90.2% for general sick genes—but likewise uses different inclusion and classification rules, so it refines rather than resolves this tension. ₃ A single-organism field-versus-lab analysis of DvH adds a further contrast. Heavy-metal-important genes were 71.2% core, whereas genes important in the broader field-stress class were 83.6% core. That class includes uranium and mercury alongside other field stresses such as nitrate, so 83.6% is not a uranium/mercury-only estimate. ₁₀ The report’s explanation is tentative. It proposes that specific metal resistance mechanisms such as efflux pumps and metal-binding proteins are accessory traits, while uranium and mercury responses involve more fundamental core stress pathways such as DNA repair and sulfate reduction. ₁₀

The specificity report is internally inconsistent on toxic-metal specificity. Its prose states that toxic metals range from 42-56% metal-specific and that iron is lowest at 21.9%, which it suggests likely reflects iron’s central role in core metabolism (a proposed, untested explanation). However, its per-metal table lists mercury, which it classes as toxic, at 35/107 (32.7%). ₃ Both the prose range and the table entry are retained here without correction. Until the source is rechecked, the 42-56% range should not be quoted as covering all toxic metals. ₃

Limitations

Metal concentrations differed among experiments, so dose-response effects could influence cross-resistance estimates. ₁ Organisms ranged from 3 to 112 metal experiments, which affected the reliability of per-organism matrices. ₁ The 28 organisms were not phylogenetically independent, and formal PGLS, or phylogenetic generalized least-squares, and independent-contrast analyses would strengthen the conservation claim. ₁

The report gives n = 26 at the Fitness Browser organism scale in its limitations section but reports 20 independent species after matching and collapsing in the validation results. ₁ This discrepancy should be retained rather than reconciled silently, because both values describe different stages or definitions of the validation sample in the source report.

The atlas adds that metal coverage was uneven, concentrations were not normalized by dose or tolerance threshold, and its primary conservation analysis covered 22 of 31 metal-tested organisms; excluding four duplicate P. fluorescens FW300 strains still gave OR=2.065 and p=5.9e-141. ₂ Its broad definition, fit < -1 OR n_sick ≥ 1, captures approximately 3.3% of genes and includes general stress genes, while putatively essential genes—approximately 14.3% of protein-coding genes and approximately 82% core—lack transposon insertions and are absent from fitness data. ₂ Dose-response normalization is also missing: nickel concentrations, for example, varied from 0.01-2.0 mM across organisms, and fitness effects depend on concentration relative to each organism’s tolerance threshold. ₂

The specificity analysis excluded 5,229 of 12,838 metal-important gene records because seven of 31 metal-tested organisms had locusId mismatches; 24 included organisms contributed 7,609 records. ₃ Its 5% sick-rate threshold was arbitrary, although results were qualitatively stable across 1–20%, and validation against Fitness Browser specificphenotype annotations was not performed. ₃ ICA (independent component analysis) module-level specificity analysis using z-scored activity profiles found 0 metal-specific modules, and the report calls this result inconclusive. Its per-module z-normalization gave max |z| values < 2.0 for most metal experiments, because metal experiments are a small fraction of each organism’s total experiments. The raw activity scores from the module condition files are on a different scale from the z-scored module profiles used in Metal Atlas NB05, which did identify 600 metal-responsive module records. The report recommends using the atlas’s precomputed z-scores directly. ₃ The pitfalls digest records the failure mode behind the excluded organisms: a locusId type mismatch silently removed DvH (495 metal-important genes) and Btheta (276 genes) from the analysis. No warning was raised, and the processing loop reported “SKIPPED (no fitness matrix or no non-metal experiments)” even though the matrix existed. ₁₁ The specificity report’s per-metal results describe DvH as “now included”, so which organisms remain affected should be checked against the final organism list rather than assumed. _{3 3}

MicrobeAtlas adds that niche breadth is a sequencing-effort proxy rather than confirmed ecological range; genus aggregation can combine different species occupying different habitats; pangenome coverage correlated with inferred metal type diversity at r = 0.35, p = 2.6×10⁻¹⁹; and 83% of Track B reads could not be joined to genus-level AMR annotations. ₅ The strict prevalence analysis, heterogeneous environment categories, possible AMRFinderPlus cross-reactivity, and lack of manual validation further limit causal environmental interpretation. ₅

The soil analysis adds unresolved confounding and validation limits: co-contamination prevents separating individual-metal effects from generic multi-metal stress; unconditional db-RDA R² was not reported; systematic effect sizes are unavailable; spatial residual autocorrelation has not yet been tested; and the copper-specific analysis matched soil samples to nearby KBase genomes using a 10 km criterion whose attribution requires 5 km and 20 km sensitivity analyses. ₇ Planned tests include Moran’s I, with SEVM if spatial autocorrelation is significant, partial correlation of COG ~ Cr | Cu + Zn + Pb (intended to distinguish metal-specific from general responses, and still pending), mechanistic classification of COGs, and reruns from the kescience_mgnify and kbase_ke_pangenome Spark tables. ₇

The T4SS–CAZy analysis has corresponding validation limits: its ≤10 kb co-occurrence threshold remains pending validation; CAZy genes were not detected on plasmids by ICEfinder; and the proposed chromosomal or integrative transfer route is consistent with, but does not prove, T4SS-mediated dissemination. ₈ The GT2 neighborhood–metal-resistance association is likewise observational and may reflect shared ecological or genomic backgrounds rather than transfer-mediated resistance acquisition. ₈

Open Directions

• Apply KEGG/PFAM mapping to cross-resistance gene signatures across 27K species, then test whether the signatures predict metal-associated isolation at pangenome scale and whether the result reproduces Cohen’s d = +1.0. ₁
• Normalize fitness effects by metal concentration relative to MIC, as both the cross-resistance and atlas reports propose, to control dose-response confounds and allow fairer cross-species comparison. Then retest the 85 metal-pair associations to determine how much of the chemistry-specific magnitude layer is dose-driven. _1,2
• Add non-metal stress controls and use ICA, or independent component analysis, of metal-condition modules to test whether the 98.1% positive correlation pattern is metal-specific or a general-stress axis. ₁
• Use phylogenetic independent contrasts or PGLS across the 28 organisms to test whether cross-resistance direction and metal-pair ranking remain conserved after accounting for shared ancestry. ₁
• Use AlphaFold-based structural analysis of the 318 conserved metal-shared ortholog groups to ask whether structural features explain their cross-metal importance. Metal binding sites and membrane interfaces are proposed, not demonstrated, explanations for multi-metal sensitivity. ₁
• Rebuild the BacDive validation with explicit species-level matching, the reported sample definitions, and independent species counts to determine whether environmental isolation is associated with multi-metal gene signatures. ₁
• Reanalyze atlas genes using a metal-specific definition that excludes genes important for other stresses, then compare core enrichment and cross-metal correlations with the three-tier classes. ₂
• Test the atlas’s 149 uncharacterized candidates with PaperBLAST, InterPro, and structural prediction, and evaluate whether regulatory or expression features outperform gene-presence scores for predicting metal fitness. ₂
• Resolve the specificity-analysis locusId mismatches, validate classifications against Fitness Browser specificphenotype annotations, and repeat the counter-ion comparison using fit < -1 without the |t| > 4 requirement. ₃
• Use precomputed z-scored Metal Atlas module activities (from Metal Atlas NB05) and AlphaFold predictions for UCP030820, YebC, and DUF1043/YhcB to test whether candidate-specificity signals reflect coherent modules or plausible metal-binding structures. ₃
• Reanalyze MicrobeAtlas with finer marine, freshwater, and saline categories, test alternative phylogenetic models, and relate metal type diversity to site-level contamination metadata to distinguish ecological breadth from sampling and habitat effects. ₅
• Test whether covered-read fraction biases Track B community-weighted metal diversity and whether the groundwater prevalence signal persists after species-level rarefaction and explicit genome-coverage correction. ₅
• Rerun the soil COG–metal models with partial correlation, including COG ~ Cr | Cu + Zn + Pb, to test whether apparent chromium associations remain after co-contaminating metals are controlled. ₇
• Report unconditional and conditional db-RDA R² values, audit Spearman rho effect sizes including rho < 0.05, and test residual Moran’s I with SEVM if needed to separate metal, project, and spatial structure. ₇
• Repeat copper genome–soil attribution at 5 km, 10 km, and 20 km, then compare mechanistically classified resistance, stress, membrane, energy, and unknown COGs across biomes. ₇
• Validate the ≤10 kb T4SS–CAZy synteny threshold with permutation tests, test Node_4915 against NCBI nr and a housekeeping-gene null, and determine whether GT2-neighborhood metal-resistance enrichment persists after controlling for genome background and biome. ₈
• Cross-reference the 144 metal-resistance genes that have fitness data in the AMR fitness-cost project against Pan-Bacterial Metal Fitness Atlas to test whether genes costly under standard conditions are protective under metal stress; this is proposed, not completed. ₁₂
• Run RB-TnSeq (random barcode transposon sequencing) on matched urease-positive and urease-negative strains under nickel versus other metals to test directly whether urease enables nickel-specific rather than cross-metal tolerance. ₁₃
• Recompute cross-metal gene-fitness correlations after removing genes important under NaCl stress, and replicate the DvH-only metals in additional organisms, to test whether the strongest pairs stay strong once the shared ionic-stress axis is removed. _6,1
• Group metal types by genetic co-occurrence, such as Cu-Zn co-resistance versus Hg-As, and test whether specific co-resistance combinations predict niche breadth better than the raw metal-type count. This would help separate a distributed signal from a power loss in the single-metal exclusions. ₅