Mobile-element associations can change with the genomic scale at which they are measured: local gene proximity, contig-level co-occurrence, and species-level repertoire breadth do not provide equivalent evidence for mobilization. The Prophage-AMR Co-mobilization Atlas shows a modest and heterogeneous local association between antimicrobial-resistance (AMR) genes and prophage markers, but a substantially stronger species-level association between prophage-marker density and AMR repertoire breadth. 1 The Within-Species AMR Strain Variation refines this framework by showing that AMR genes also form tightly co-inherited, frequently multi-mechanism resistance islands, without establishing which mobile element mobilizes them. 2 The Pan-Bacterial AMR Gene Landscape supports treating accessory AMR as a broad pangenome pattern rather than a prophage-specific signature: across 27,690 pangenome species, AMR was depleted from the core genome and enriched in the auxiliary genome, while clinical AMR was less core than soil or plant AMR. 3 The AMR Co-Fitness Support Networks refines this scale framework by showing that AMR genes also occupy large, organism-shaped fitness-support networks; these are evidence of shared fitness phenotypes, not evidence that a mobile element physically links or transfers the genes. 4
Evidence Across Scales
Gene and contig scale
Across the full pangenome inventory, 83,008 AMR gene clusters and 3,465,244 broad prophage marker clusters were identified, including 1,261,929 strict prophage-marker clusters. 1 Prophage markers were 83.8% accessory and 53.8% singleton, whereas AMR genes were 69.7% accessory and 36.1% singleton. 1
The pan-bacterial AMR atlas independently found 83,008 AMRFinderPlus hits on gene-cluster representatives across 27,690 species, spanning 82,908 distinct clusters and 14,723 species; 53.2% of species carried at least one AMR hit. 3 It supports the existing accessory-genome interpretation: only 30.3% of AMR genes were core versus 46.8% for the pangenome baseline, and the auxiliary genome was 2.2x enriched for AMR (33.6% versus 15.3%). 3 In 4,252 species with at least 5 AMR clusters, 63.7% had AMR less core than their species baseline (Wilcoxon p=1.1e-130, mean difference -0.102). 3 This broad conservation signal does not identify the mobile carrier, so it refines rather than replaces the local prophage evidence.
Among 36,041 AMR gene instances from 1,953 genomes sampled across the top-100 AMR-burdened species, 20,073 (55.7%) occurred on contigs carrying strict prophage markers. 1 The same analysis found that 12,026 (33.4%) were within 50 genes of a prophage marker, 7,137 (19.8%) were within 20 genes, 3,731 (10.4%) were within 10 genes, and 1,991 (5.5%) were within 5 genes; the median distance to the nearest prophage marker among co-localized AMR genes was 34 genes. 1
At a 10-gene threshold, AMR genes near prophage markers were 67.6% accessory (2,523 accessory and 1,208 core), compared with 65.5% for distal AMR genes (21,158 accessory and 11,152 core). 1 Fisher’s exact test gave an odds ratio of 1.10, one-sided p=0.005, and a bootstrap 95% confidence interval of [1.024, 1.185]. 1 The statistically significant association was therefore modest, corresponding to a 2.1 percentage point difference in accessory fraction. 1
The local association was threshold-dependent rather than uniformly positive: the odds ratio was 0.78 at 3 genes, 0.92 at 5 genes, 1.10 at 10 genes, 1.19 at 15 genes, and 1.28 at 50 genes. 1 Only 33 of 74 testable species had species-level odds ratios above 1, and the median species-level odds ratio was 0.85. 1 This refines a simple proximity-based mobilization interpretation because broader windows produced stronger aggregate associations while the species-level pattern remained inconsistent. 1
The reversal at very close range may reflect the presence of core phage components immediately adjacent to phage structural genes, whereas broader windows may capture genomic islands containing both prophage remnants and laterally transferred genes. 1 The aggregate effect was modest, a 2.1 percentage point difference in accessory fraction, and heterogeneous across species; these mechanisms were interpretations rather than direct tests of mobilization. 1
The AMR strain-variation analysis supports the importance of genomic linkage beyond prophage-specific proximity: it detected 1,517 resistance islands across 705 species, with a mean size of 6.2 genes, a median size of 4 genes, a maximum size of 43 genes, and a mean phi coefficient of 0.827. 2 Here phi measures pairwise co-occurrence, so the result indicates tight co-inheritance, not demonstrated transfer. 2 Of these islands, 1,343/1,517 (88%) contained genes from multiple resistance mechanisms; efflux pumps occurred in 954 islands and enzymatic inactivation in 698. 2 This refines interpretation of prophage-proximal AMR because tightly linked AMR modules may reside on resistance islands associated with plasmids, chromosomes, or other mobile elements, and co-occurrence does not establish co-selection or functional synergy. 2
The atlas refines the core/accessory comparison by showing that conservation depends on resistance biology: beta-lactamases were 54.9% core, intrinsic emhABC efflux pumps were >95% core, while regulatory genes were 6.5% core and blaTEM, tet(C), and ant(2”)-Ia were 0% core. 3 Thus, an AMR gene’s accessory status can indicate acquired resistance, but accessory status alone cannot establish prophage association or mobilization.
The cofitness analysis supports separating functional association from physical linkage. Across 28 organisms, 801 AMR genes with fitness data, and independent component analysis (ICA), which decomposes condition-dependent fitness profiles into modules, 769 (96%) AMR genes had at least one extra-operon cofitness partner at |r| > 0.3; the mean support-network sizes were 233 genes at |r| > 0.3, 110 at |r| > 0.4, and 71 at |r| > 0.5. 4 AMR-containing ICA modules had a median of 46 genes versus 27 for non-AMR modules (MWU p = 1.7×10⁻⁸), and 208/209 AMR gene-module assignments were in cross-organism conserved module families. 4 These findings identify broad shared-fitness neighborhoods rather than contig adjacency, resistance-island membership, or a demonstrated mobile carrier. 4
Species and repertoire scale
Across 4,770 species, prophage marker density was positively associated with AMR repertoire breadth, with Spearman rho=0.572, p<10^-300, and n=4,770 species. 1 A log-log regression had a slope of 0.823 (SE=0.018), p<10^-300, and R²=0.30; the report states that a 10-fold increase in prophage density predicts a ~6.6-fold increase in AMR breadth. 1 After controlling for genome count, the partial Spearman correlation remained rho=0.464 with p=1.0×10^-253. 1
The species-level association was significant across the five reported major phyla: Pseudomonadota (rho=0.54), Bacillota_A (rho=0.55), Bacillota (rho=0.40), Bacteroidota (rho=0.59), and Actinomycetota (rho=0.29). 1 This cross-phylogenetic consistency supports an association that is not explained solely by the reported phylum groupings, but it does not establish phage-mediated AMR transfer. 1
The environmental-resistome analysis supports the existing alternative that shared ecological or sampling-associated propensity can contribute to species-level AMR breadth: across 14,723 species and 293K genomes, clinical species had a median of 5 AMR gene clusters versus 2 for soil, aquatic, and host-associated species, and environment effects persisted in 5 of 6 tested major phyla and in 20 of 141 testable families. 5 It refines interpretation of the prophage-density correlation because clinical and human-gut species had predominantly accessory AMR, whereas soil and aquatic species had more metal-resistance AMR; these environment-linked repertoires could covary with prophage density without demonstrating phage mobilization. 5
The atlas supports this ecological-confounding alternative: human/Clinical species carried 10.6 AMR clusters per species versus 4.6 for Soil/Terrestrial, 3.9 for Aquatic, and 3.0 for Animal, with Kruskal-Wallis H=440 and p=7.0e-93. 3 Clinical AMR was 30.8% core versus 58.1% for soil and 63.1% for plant AMR. 3 Among 2,684 species with at least 3 genomes and AlphaEarth embeddings, environmental diversity correlated with AMR count (Spearman rho=0.466, p=1.6e-144) and negatively with AMR core fraction (rho=-0.173, p=1.8e-19). 3 These associations strengthen the need to control for environment and sampling before interpreting prophage density as a mobilization driver.
The AMR strain-variation analysis supports retaining lineage and ecological confounding as alternatives: among 37,444 AMR gene-species records, 51.3% were rare, 41.3% variable, and 7.5% fixed, while 701/1,261 species (55.6%) showed significant AMR–ANI distance correlation at FDR < 0.05. 2 Its host-associated species carried more AMR genes per genome than terrestrial or aquatic species under both approximate environment-classification approaches, with Kruskal-Wallis p < 0.05. 2 These results refine the species-level prophage interpretation: lineage structure and host-associated sampling can organize AMR breadth independently of direct phage-mediated transfer.
The cofitness analysis supports retaining condition-dependent shared dispensability as another alternative to mobilization. Among AMR-containing support neighborhoods, flagellar motility, flagellum assembly, chemotaxis, and amino-acid biosynthesis were enriched in multiple organisms, but the analysis could not distinguish shared condition-dependent dispensability from shared regulation. 4 Its support-network organization was more organism-specific than mechanism-specific: mean GO-term Jaccard similarity was 0.375 for different AMR mechanisms within an organism versus 0.207 for the same mechanism across organisms (MWU p = 4.3×10⁻¹³). 4 This refines the interpretation of species-level prophage associations by showing that organismal regulatory and metabolic architecture can structure AMR-related functional associations without identifying a transfer route. 4
The species-level result is compatible with two non-exclusive explanations: prophages may mobilize resistance genes through specialized or generalized transduction, and species with high recombination potential may independently acquire genes from multiple mobile elements. 1 Because the analysis is correlational, the result cannot distinguish direct phage-mediated transfer from shared ecological, genomic, or recombination-related propensity. 1 The environmental-resistome evidence supports retaining this distinction rather than treating accessory AMR as a specific signature of prophage transfer. 5 The resistance-island evidence further supports this caution because co-inherited multi-mechanism islands were widespread, but their genomic carrier and transfer mechanism were not resolved. 2 The cofitness evidence supports the same caution at a functional scale: cofitness indicates shared fitness phenotypes, not direct transcriptional control or physical mobilization. 4
What Each Scale Can Support
Contig-level co-occurrence establishes that AMR genes and strict prophage markers can occupy the same contig in the analyzed genomes, but it does not by itself establish that a prophage carries, excises with, or transfers an AMR gene. 1 Gene-window results provide a scale-sensitive association whose magnitude changes from an odds ratio of 0.78 at 3 genes to 1.28 at 50 genes, so the selected distance threshold materially affects the inferred relationship. 1
Species-level density and repertoire correlations capture broader accumulation patterns than local neighborhoods and produced the strongest association in this report. 1 However, those correlations cannot separate phage mobilization from plasmid or integrative-conjugative-element mobilization, shared recombination capacity, or other causes of joint AMR and prophage accumulation. 1 This complements the environmental-resistome finding that ecological environment is associated with AMR diversity and core/accessory composition, while its observational design likewise cannot establish causation. 5
The atlas supports interpreting species-level breadth as a composite of resistance biology, taxonomy, and environment: AMR clusters were enriched in defense mechanisms and inorganic-ion transport, and the five most abundant families included mercury- and arsenic-resistance genes as well as beta-lactamases. 3 Because AMRFinderPlus includes stress-response genes alongside classical antibiotic-resistance genes, broad AMR counts may not measure antibiotic resistance narrowly. 3
Resistance-island co-inheritance provides an intermediate module-level signal: it identifies linked AMR mechanisms but does not identify the carrier element or prove co-selection. 2 This refines the interpretation of contig and species-scale associations by separating evidence for linkage from evidence for mobilization.
Cofitness provides a distinct functional scale between individual genes and organism-wide repertoires. Its large support neighborhoods and conserved ICA modules support condition-linked functional coupling, but the enrichment for flagellar and biosynthetic functions may reflect genes that are jointly dispensable in shaken laboratory cultures or supplemented media rather than genes that are co-regulated or co-mobilized. 4 Network size was not associated with AMR fitness cost (Spearman rho = −0.006, p = 0.87, N = 769), so functional-network breadth does not provide a proxy for the cost or transferability of resistance. 4
This distinction complements environmental-resistome, where AMR distributions are interpreted across genomic and ecological contexts, and it refines pangenome-integration by showing that integrated pangenome signals depend on whether evidence is summarized at the neighborhood, contig, or species level. 1 It also connects to phage-defense-syndromes-and-arms-race because keyword and Pfam-based prophage-marker identification may include phage-defense systems and may miss divergent prophages. 1
Measurement and Inference Limits
Distances were calculated from ordinal gene positions parsed from gene_id formats rather than base-pair coordinates, so the reported gene distances may differ from true genomic distances. 1 Prophage markers were identified by keyword and Pfam matching in bakta_annotations rather than by dedicated prophage-prediction tools such as PHASTER or geNomad, which may cause false positives, including phage-defense systems, and may miss divergent prophages. 1
The co-localization analysis sampled 20 genomes per species for the 100-species analysis rather than examining all 293K genomes, so the local association may change under exhaustive genome-level analysis. 1 The co-localization analysis also did not partition AMR mobilization among prophages, plasmids, and integrative conjugative elements (ICEs). 1 Core and accessory labels depended on species-level pangenome calling with motupan, meaning that the same gene can have different conservation status in different species. 1
The environmental-resistome analysis further cautions that NCBI sampling overrepresents clinical isolates, majority-vote environment labels collapse within-species variation, 18.7% of AMR clusters lacked mechanism assignments, and its clinical-environment association was observational. 5 These limitations refine the interpretation of species-level mobile-element associations by adding environment classification, coverage, and annotation as possible confounders. 5 The AMR strain-variation analysis likewise found that 52.7% of genomes had no classifiable isolation_source and that dedicated environment curation was needed for ecotype analysis. 2
The atlas adds that only 7,838 of 14,723 AMR-carrying species (53.2%) received a non-Other/Unknown environment classification, while the Other/Unknown bin contained 46.8% of species; AlphaEarth embeddings covered only 28% of genomes. 3 This supports treating environmental AMR–prophage comparisons as coverage-limited. Its mechanism classification used keyword matching rather than CARD ARO terms, leaving 18,448 hits (22.2%) Other/Unclassified. 3
AMR detection in the strain-variation analysis relied on the AMRFinderPlus database, so novel resistance mechanisms absent from that database were missed. 2 This refines the interpretation of resistance-island counts and AMR breadth because apparent absence or rarity can reflect annotation scope as well as biology. The atlas likewise reported that 51.5% of hits were detected by HMM, 22.7% by BLASTP, 13.0% by EXACTP, 9.7% by PARTIALP, and 3.0% by ALLELEP, while singleton inflation may reflect annotation artifacts. 3
Cofitness interpretation has parallel limitations: cofitness is not equivalent to co-regulation, and the analysis treated missing fitness values as zero in z-score space through np.nan_to_num, which approximates but does not equal pairwise-complete Pearson correlation. 4 At |r| > 0.3, the mean support network contained 233 genes and therefore included many weak associations; confirmation at |r| > 0.4 was needed. 4 The 28 organisms were lab-adapted, phylogenetically biased, included many Pseudomonas organisms, and had limited ecological diversity. 4 These limitations support treating cofitness as functional context rather than evidence for a mobile carrier or ecological mobilization mechanism. 4
The proposed fitness comparison could not be tested because the KBase Data Lakehouse fitness browser contained RB-TnSeq, random barcode transposon sequencing, data for only 48 model organisms, with poor overlap with the GTDB pangenome species analyzed here. 1 Whether prophage-proximal AMR genes have distinct fitness costs therefore remains unresolved. 1 The atlas refines this gap rather than closing it: among 37 Fitness Browser organisms with AMR genes, AMR genes had median fitness -0.007 versus -0.012 for the non-AMR baseline, but the predominantly environmental organisms and laboratory conditions do not test recently acquired mobile resistance in clinical pathogens. 3
Mobile-Element Proxies Beyond Prophage–AMR Pairs
Accessory status and pending carrier mapping
The environmental-resistome report interprets core AMR as chromosomally encoded (intrinsic) and accessory AMR as mobile (acquired) by analogy to the pattern Jiang et al. (2024) observed in river metagenomes, and describes clinical and human gut species as having overwhelmingly acquired their AMR through horizontal gene transfer. 5 That mechanism was not established for individual clusters: whether accessory AMR clusters in clinical species are preferentially associated with plasmids, transposons, or integrative elements is proposed for testing rather than reported as a finding. 5 This supports the page’s caution that accessory status is evidence compatible with mobility, not evidence of a carrier. 5
Singleton or accessory enrichment is used as a mobility proxy elsewhere in the corpus with the same limitation. In the phage-defense analysis, accessory-enriched defense systems had singleton fractions 1.5–1.8× above the background 37.9 % baseline, consistent with, but not a direct demonstration of, mobile-element-mediated horizontal transfer of defense loci. 6 A DUF4041 (PF13250) survey of the 293K-genome KBase Data Lakehouse pangenome found 4,572 gene clusters across 1,696 species and 33 phyla, compared with >500 homologues reported in Saxton et al. 2026; 86.7% were accessory or singleton genes, consistent with, but not direct proof of, mobile defense-island carriage. 7 These results support reading accessory and singleton status as compatible with, rather than diagnostic of, mobile carriage, which links this page to phage-defense-syndromes-and-arms-race. 6,7
Several projects leave the carrier question explicitly open. The pan-bacterial atlas lists, as proposed rather than completed analyses, whether AMR genes cluster in genomic islands and co-localize with mobile-element markers such as IS elements and integrons. 3 The strain-variation analysis lists mapping detected resistance islands to their mobile genetic element context (plasmid versus chromosome, integron boundaries, IS elements) as future work, so resistance-island-coinheritance remains a linkage rather than a mobilization result. 2 The field-versus-lab fitness analysis likewise leaves unresolved whether accessory-genome antibiotic and heavy-metal resistance genes are on mobile elements, recently acquired, or shared with co-occurring species at Oak Ridge. 8
Protein annotation versus gene neighborhood
The ecological-agora analysis refines this framework by separating two independent measurement substrates for mobile genetic elements (MGEs): per-cluster MGE-machinery, which asks whether a KEGG ortholog (KO) itself encodes an MGE protein, and gene-neighborhood MGE-cargo, which asks whether the KO sits ±5kb from MGE machinery in actual genome assemblies; these are distinct measurements. 9 Because the genomad mobile-elements table was not ingested in the KBase Data Lakehouse, the available signal was Bakta product keyword matching for phage, transposase, integrase, plasmid, IS-element, recombinase, and conjugation terms across all 132M gene clusters. 9
MGE-context products occurred in 248,908 of 18,204,007 KO-bearing gene clusters, an atlas-wide MGE-machinery baseline of 1.37%, and the per-(KO × genus) MGE-machinery output has 3.94M rows. 9 All three pre-registered hypothesis KO sets showed ≤0.57% (near-zero), which the report reads as confirming that these gene families are not phage, transposase, or integrase products themselves; biome-stratified MGE rates were similar (~1.1–2.0%), with no strong biome-specific elevation, and the MGE-context figure compares per-category, per-hypothesis, and per-biome rates against the 1.37% baseline. 9 This null concerns protein annotation, not genomic cargo status. 9
At the neighborhood scale, the photosystem II (PSII) scan (Cyanobacteriia × PSII, 27,148 focal features, 218,321 neighbor pairs) found a mean MGE-neighbor fraction of 1.65%, and 10.91% of focal features had at least one MGE neighbor within ±5kb, approximately the stated 10.6% Poisson expectation derived from the atlas MGE rate p = 0.014 and mean neighbors n ≈ 8; no MGE-cargo enrichment was detected, which is an absence of detected enrichment rather than proof that PSII genes are never MGE cargo. 9 The per-feature PSII output table lists 26,864 rows, a count that differs from the 27,148 focal features stated in the analysis text; the report does not identify the row count as the analysis denominator, and the difference is left unreconciled here. 9 Unlike the prophage–AMR gene-window results, this neighborhood test was compared explicitly against a random baseline. 9,1
Individual PSII subunits varied in MGE-neighbor frequency from 3.5% to 22.1%: PsbZ (K02720; 1,769) 22.1%, PsbX/M (K02718; 1,500) 15.5%, PsbB (CP47; K02704; 1,794) 14.2%, PsbA (D1, RC; K02703; 4,217) 13.7%, PsbE (cyt b559α; K02710; 1,188) 12.6%, and PsbV (cyt c550; K02716; 1,386) 3.5%. 9 The report treats this subunit variation as suggesting differential mobility within the PSII complex, which remains a hypothesis because the ensemble mean (10.91%) sits at the random-baseline expectation. 9 This supports the page’s central point that proximity signals for the same gene set depend on the level at which they are aggregated, here subunit versus ensemble. 9
The pangenome-to-genome cross-walk (pangenome.gene_genecluster_junction → pangenome.gene → kbase_genomes.name → kbase_genomes.feature + contig_x_feature) combined with bakta_annotations.product keyword matching is presented as a reusable pipeline for cargo-on-MGE measurement at gene-neighborhood scale. 9 It was demonstrated on PSII, but polysaccharide utilization locus (PUL) and mycolic-acid neighborhood analyses were deferred after a pandas spatial-merge out-of-memory failure at 723K Bacteroidota focal features × 210K contigs; the report states that per-cluster MGE-machinery rates (0% / 0.57%) suffice for its headline non-phage-borne conclusion, so PUL and mycolic cargo status remains untested at neighborhood scale. 9
Plant-associated markers and genus-level mobilome
The plant-microbiome analysis evaluated three horizontal gene transfer (HGT) proxies with mixed results: among 1,136 plant-associated species, those carrying singleton marker gene clusters were 16× more likely to also carry transposase/integrase singletons (Fisher’s exact OR=15.95, p=8.8e-20), but the overall singleton enrichment ratio for markers was 0.78 (markers less mobile than the genomic average; Wilcoxon p<1e-300), and cross-species context variation was higher for pathogenic markers (0.534) than beneficial markers (0.450) but not significantly so (Kruskal-Wallis p=0.122). 10 The co-occurrence proxy comprised 986,464 transposase/integrase singleton clusters (722,674 transposase + 263,790 integrase). 10 The central discoveries digest records the same co-occurrence (OR=15.95, p=8.8e-20, described as 16x) and treats horizontal transfer as suggested, not directly shown. 7
Per-marker transposase co-occurrence odds ratios were DAPG (2,4-diacetylphloroglucinol) biocontrol (OR=8.75), ACC (1-aminocyclopropane-1-carboxylate) deaminase (OR=6.43), and nitrogen fixation (OR=3.76) among plant-growth-promoting markers, and type IV secretion system (T4SS; OR=3.23), type VI secretion system (T6SS; OR=2.66), and effectors (OR=2.67) among pathogenic markers. 10 At contig scale, 498,677 marker–transposase pairs shared contigs across 18,569 species; T4SS had the most co-located pairs (84,411) with the closest median distance (238 genes), and fifteen pairs were at gene-number distance 1 (immediately adjacent), spanning effectors, T6SS, type III secretion system (T3SS), quorum sensing, cellulase, and T4SS in species including Rhizobium ecuadorense, Phytobacter ursingii, and Burkholderia puraquae. 10 This parallels the prophage–AMR contig and gene-window evidence: shared contigs and gene-number proximity establish co-location, not demonstrated transfer. 10,1
At genus scale, the MGnify mobilome analysis of 563 pangenome genera with soil-biome data detected 17,323 mobilome entries with ISEScan, ICEfinder, and geNomad, including IS elements (IS110, IS701, IS1634, IS256, IS3), plasmids, prophages, terminal inverted repeats, and viral sequences. 10 Plant-associated genera carried significantly more mobile elements per genome than non-plant genera (median 3.7 vs 2.8, Mann-Whitney U p=1.49×10⁻⁵). 10 MGnify biosynthetic gene cluster (BGC), mobilome, and defense annotations were available only for the soil biome, not for the tomato, maize, or barley rhizosphere biomes, so the comparison was between soil-dwelling plant-associated and soil-dwelling non-plant genera rather than rhizosphere versus bulk soil. 10
Type IV secretion neighbourhoods and metal resistance
In the metal-resistance biogeography analysis, the set of 22,356 MAGs (metagenome-assembled genomes) with coordinates showed 2.8% global metal resistance prevalence, which the report contrasts with a 21.8% T4SS prevalence and takes as confirming that AMRFinderPlus-detected metal resistance genes and T4SS HGT machinery are distinct features not uniformly co-distributed. 11 A contrast of marginal prevalences does not by itself test whether the two features co-occur within the same genomes, so this decoupling interpretation is weakly supported. 11
The T4SS–CAZy analysis reports 92 CAZy families with elevated co-occurrence with T4SS loci at ≤10 kb, with GT2 glycosyltransferases as the top hit (767 genomes, avg 5,041 bp), but threshold validation is pending. 12 In list-format neighbourhoods, T4SS co-occurred in 503 neighbourhood entries and GT2 in 495, which the report interprets as contig-level syntenic co-localisation; GH23 (murein lytic transglycosylase) was the second most common CAZy family in GT2 neighbourhoods (106 occurrences), suggesting rather than establishing that cell-wall remodelling genes cluster with GT2–T4SS loci. 12 GT2-neighbourhood MAGs (n=376) had mean 0.045 metal resistance types versus 0.004 for non-GT2 MAGs (n=260,276; Mann-Whitney p=8.6e-27), which the report separately describes as 11× more metal resistance genes. 12 These neighbourhoods feed chromosomal-and-integrative-gene-transfer as co-location evidence, not demonstrated transfer. 12
The T4SS–CAZy analysis lists pending checks: a synteny-threshold permutation test (requiring Spark unfiltered data), Node_4915 BLAST validation against NCBI nr, a housekeeping-gene null baseline, and biome-enrichment factorisation θ = OR(T4SS-CAZy) / [OR(T4SS) × OR(CAZy)]. 12 Until the housekeeping-gene null is run, its neighbourhood signal lacks the random-baseline comparison against which the PSII neighbourhood scan showed no detected enrichment. 12,9
Tensions
The plant-microbiome report’s genus-level mobilome enrichment (median 3.7 versus 2.8 mobile elements per genome; p=1.49×10⁻⁵) appears to contradict genome-level studies by Levy et al. (2018) and Bograd et al. (2025) that report fewer mobile elements in plant-associated genomes. 10 The report proposes a scale-dependent resolution, rhizosphere/soil genera where HGT is frequent versus endosphere-focused genome-level comparisons, and its own issue list documents scale-dependent mobilome signals (genus versus genome level) as a discussed caveat; this resolution is proposed, not established, and the soil-only MGnify coverage means the biome contrast was not directly tested. 10
Within the same project the HGT proxies also disagree: species-level co-occurrence of marker and transposase/integrase singletons was strongly positive (OR=15.95), whereas the marker singleton enrichment ratio of 0.78 indicated markers were less mobile than the genomic average. 10 Neither side is preferred here; the report proposes replacing the proxy-based HGT analysis (transposase/integrase singletons, MGnify mobilome aggregates) with direct geNomad mobile-element predictions to strengthen its mobilome conclusions and resolve the genus-versus-genome scale-dependent signal. 10
The metal-resistance biogeography report contrasts 2.8% metal resistance prevalence in 22,356 geolocated MAGs with a 21.8% T4SS prevalence and reads the difference as non-uniform co-distribution, whereas the T4SS–CAZy analysis found mean 0.045 metal resistance types in GT2-neighbourhood MAGs (n=376) versus 0.004 in non-GT2 MAGs. 11,12 The evidence here does not establish that the two prevalences share a MAG denominator, and different marginal prevalences do not establish a joint-distribution relationship, so the global contrast neither demonstrates decoupling nor directly contradicts the conditional local enrichment; neither has been tested against a shared null baseline, and the local enrichment awaits the T4SS–CAZy threshold and housekeeping-gene checks. 11,12
Open Directions
- Apply geNomad or PHASTER to the KBase Data Lakehouse genomes and compare dedicated prophage calls with keyword/Pfam calls to quantify false positives, missed divergent prophages, and the contribution of phage-defense systems to the observed associations. 1
- Recompute AMR–prophage distances from scaffold sequences (scaffoldseq contig sequences) at base-pair resolution and test whether the threshold-dependent odds ratios persist when ordinal gene-position errors are removed. 1
- Analyze all available genomes from Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Escherichia coli to determine whether the local and species-level associations are robust to the 20-genomes-per-species sampling design. 1
- Partition prophage, plasmid, and integrative-conjugative-element signals and fit comparative models that test whether prophage density predicts AMR breadth independently of other mobile elements. 1
- Map the 1,517 resistance islands to plasmids, chromosomes, integron boundaries, insertion sequences, and prophage calls to determine which carriers explain tight AMR co-inheritance. 2
- Revisit the fitness comparison as fitness-browser coverage expands, testing whether prophage-proximal and distal AMR genes differ in RB-TnSeq fitness effects. 1
- Stratify the mobile-element analysis by environment and clinical representation, then test whether prophage density predicts AMR breadth after accounting for environment, phylogeny, genome count, and other mobile elements. This would address the environmental confounding identified in the 14,723-species resistome analysis. 5
- Replace keyword-based AMR mechanism labels with CARD ARO mappings and test whether prophage-density and resistance-island associations persist after separating antibiotic resistance from metal and other stress-response genes. 3
- Use the full Fitness Browser pangenome link table and antibiotic-stress conditions to test whether the apparently low laboratory cost of AMR genes generalizes to mobile, clinical-associated resistance and differs by prophage proximity. 3
- Perform fitness-matched permutations of the AMR cofitness neighborhoods, recompute cofitness separately in antibiotic and standard-growth conditions, and test flagellar knockouts and other conditionally dispensable gene classes; this would distinguish shared dispensability from regulatory coupling before using cofitness as functional support for mobilization hypotheses. 4
- Apply the PSII-style Poisson baseline (atlas MGE-machinery rate and mean neighbor count) to the prophage–AMR gene windows and to the T4SS–CAZy ≤10 kb neighbourhoods, testing whether their associations exceed random MGE-neighbor expectation. 9,1,12
- Rerun the deferred PUL and mycolic-acid neighborhood scans with a chunked or partitioned spatial join to test whether those cargo classes, like PSII, show no enrichment over the random baseline. 9
- Replace transposase/integrase singleton and MGnify aggregate proxies with direct geNomad predictions at genome level for plant-associated and non-plant genomes, testing whether the genus-level mobilome enrichment survives at genome scale. 10
- Run the housekeeping-gene null and synteny-threshold permutation for the T4SS–CAZy neighbourhoods, then test whether metal-resistance enrichment in GT2-neighbourhood MAGs persists against that null; separately test within-genome co-occurrence of metal resistance and T4SS in the geolocated MAGs rather than comparing marginal prevalences. 12,11