BERIL Knowledge Wiki

Project report in Data Infrastructure, Provenance, and Research Practice, Ecotypes and Environmental Gene-Content Differentiation, Functional Dark Matter and Annotation Resolution, Metabolic Traits, Auxotrophy, and Community Dependence, Pangenome Structure and Genome Evolution, Sampling Coverage and Statistical Confounding

Plant Microbiome Ecotypes — Compartment-Specific Functional Guilds and Their Genetic Architecture

AI-generated wiki from AI-conducted research reports. Partially reviewed by scientists. See About This Wiki.

Projects cited
1 project report, listed in the margin
Linked from
74 pages14 topics
Conflicts
11 open
Author
Adam P. Arkin

Raw report: plant_microbiome_ecotypes Feeds into: adaptive-versus-housekeeping-functional-differentiation, adversarial-research-quality-assurance, callability-limited-comparative-inference, community-metabolic-interdependence, computational-pathway-prediction-validation, condition-specific-fitness, confirmatory-exploratory-ecological-association-discordance, core-gene-annotation-paradox, cross-cohort-microbiome-portability, cross-tenant-data-bridging, cultivation-collection-bias-in-ecological-genomics, dual-nature-plant-associated-bacteria, ecotype-clustering-validity, ecotype-environment-gene-content, environment-embedding-geography, evidence-triangulation-for-functional-annotation, experimental-prioritization-of-functional-dark-matter, functional-marker-validation, gene-cooccurrence-ecological-guilds, gene-function-acquisition-depth, genome-wide-versus-locus-specific-ecological-adaptation, homology-search-negative-evidence, horizontal-gene-transfer-driven-innovation, metabolic-model-gapfilling, ontology-and-category-schema-sensitivity, pangenome-core-boundary-and-clade-size-bias, pangenome-integration, pathway-versus-reaction-evidence-resolution, phb-granule-ecology, phenotype-database-coverage-bias, phylogenetic-confounding-of-pangenome-associations, research-attention-inequality, sample-size-aware-phenotype-consensus, sampling-depth-and-downsampling-effects, scale-dependent-mobile-element-associations, selection-on-outcome-leakage, taxonomic-nomenclature-reconciliation, taxonomic-resolution-dependent-functional-inference, two-speed-bacterial-genome

Overview

This report analyzes plant association across 293,059 GTDB r214 genomes, identifying 1,136 plant-associated species and 7,995 plant-associated genomes across root, rhizosphere, phyllosphere, endophyte, and other plant contexts. It combines pangenome gene-cluster architecture, marker-gene surveys, eggNOG enrichment, GapMind pathway predictions, NMDC community data, MGnify metagenomes, and within-species phylogenetic analyses to test compartment effects, beneficial-versus-pathogenic genomic architecture, horizontal gene transfer, functional complementarity, plant-enriched gene families, dual-nature classifications, and host-specific ecotypes. Phase 2b adversarial review substantially revised several conclusions: compartment effects and complementarity effects are small, categorical cohort labels are not species-level discriminators, most marker associations are genus-scale, and phylogenetic-tree coverage limits subclade inference. Three adversarial-review issues were fixable and were resolved in notebooks NB13–NB15: the tautological genus-level validation (C3), the original zero-hit Pfam query (a documented limitation), and the NB12 subclade genome-ID mismatch (I6). Four statistical controls that RESEARCH_PLAN.md had specified but Phase 1 had not executed were also completed: C1, C4, I1 and within-genus label shuffling. 1

Key Findings

Compartment effects are statistically significant but small

Using the refined 17-marker panel and Jaccard distances across 607 root, rhizosphere, and phyllosphere species, PERMANOVA gave pseudo-F = 23.2, p = 0.001, and R² = 0.071 with 999 permutations. PERMDISP detected dispersion heterogeneity with F = 15.6 and p = 0.001: root species were tightest around their centroid (mean distance 0.452), phyllosphere species were intermediate (0.503), and rhizosphere species were most variable (0.528). Distance-based redundancy analysis (db-RDA), which isolates centroid or location shifts, gave location-only R² = 0.060, p = 0.001; this represents 0.060 / 0.071 = 84% of the total PERMANOVA R², with 16% attributable to dispersion. Thus, compartments have real but weak functional separation of approximately 6% of variance. Because PERMDISP was significant, the adversarial-review issue table records this dispersion issue (C2) as documented, with the PERMANOVA R² treated as an upper bound. The analysis was run on all 607 plant-compartment species using the refined 17-marker v2 panel. The report also gives the db-RDA location-only value more precisely as R² = 0.0595 (permutation p = 0.001), representing 84% of the total PERMANOVA R². Its final verdict on the compartment hypothesis (H1) is that it is weakly supported: compartments impose statistically significant but small (~6% of variance) location shifts on plant-associated functional profiles, alongside significant dispersion heterogeneity, with root tightest and rhizosphere most variable. 1

The earlier Phase 1 PERMANOVA on a 25-marker panel reported R² = 0.527 and pseudo-F = 235.1, p = 0.001, but excluding the three genome-rich species per compartment reduced R² to 0.072, an 86% loss. The report attributes most of the original headline effect to taxonomic sampling and dominant rhizobial or Pseudomonas clades rather than a community-wide signal. Per-marker patterns remained directional: in the original framework, root species showed ACC deaminase enrichment with OR = 69.3, T3SS enrichment with OR = 65.6, nitrogen-fixation enrichment with OR = 14.5, and quorum-sensing enrichment with OR = 24.1; Fisher’s exact tests found 69 of 96 marker-by-compartment combinations significant at q<0.05. The report states that these per-function enrichments are not contradicted by the Phase 2b decomposition; only the headline magnitude of the multivariate R² was inflated. The report’s discussion quotes the original result in rounded form as R²=0.53 and root ACC deaminase OR=69, and notes T3SS enrichment in phyllosphere species (OR=10.7) as well as root species (OR=65.6). It links that result to plant compartments acting as ecological filters (Trivedi et al. 2020). A separate PERMDISP test (NB08) found significant dispersion heterogeneity between cohorts (H=33.12, p=3.0e-7). The R²=0.53 may therefore partly reflect variance differences and should be interpreted as an upper bound. The compartment effect is still supported by the 69/96 significant Fisher tests. The limitations section names two of the three RESEARCH_PLAN.md safeguards as executed in NB14. The first was the top-3-species exclusion, which reduced R² from 0.527 to 0.072 (still significant, p = 0.001). The second was within-genus label shuffling, in which only 3/15 markers (nitrogen fixation, ACC deaminase, T3SS) survived. In this passage the report describes the 1,136 plant-associated species as classified across four compartments, listing root 292, rhizosphere 160, phyllosphere 157, and endophyte 29. Those four counts do not by themselves account for all 1,136 species. The genome census below also lists a plant-other category. 1

The genome census classified 292 species and 2,446 genomes as root-associated, 160 species and 941 genomes as rhizosphere-associated, 157 species and 1,334 genomes as phyllosphere-associated, 29 species and 291 genomes as endophytes, and 498 species and 2,983 genomes as plant-other; the total was 1,136 plant-associated species and 7,995 genomes, compared with 25,375 known non-plant species and 285,064 genomes. Root communities were dominated by rhizobia (Rhizobium 47, Mesorhizobium 45, Bradyrhizobium 41 species), whereas the phyllosphere was dominated by Sphingomonas (16), Methylobacterium (13) and Pseudomonas_E (13); Pseudomonas_E was the only genus with significant representation across all compartments. The report’s discussion says the phyllosphere dominance of Sphingomonas and Methylobacterium matches Knief et al. (2012), who identified methylotrophy as the defining phyllosphere function. It describes the root-dominant genera (Rhizobium, Mesorhizobium, Bradyrhizobium) as well-characterized nitrogen fixers and takes this as validating the compartment classification pipeline. 1

A random forest classifier using 25 binary markers predicted compartment (root/rhizosphere/phyllosphere) with 64.4% accuracy. That is above the 33% random baseline but far from deterministic. The report reads it as consistent with compartment being one of multiple factors shaping marker profiles. Cohort prediction reached 99.9% accuracy, which the report calls trivial because cohorts are defined by marker presence. 1

Beneficial functions are more core-encoded than pathogenic functions

Beneficial plant-growth-promotion (PGP) gene clusters had a 64.6% core fraction, compared with 45.2% for pathogenic clusters and 46.8% for the genome-wide baseline. The per-species Mann–Whitney U comparison gave U = 83,567,419 and p = 3.38e-125, with a bootstrap 95% confidence interval of [0.089, 0.106] for the beneficial-minus-pathogenic difference. This complementary per-species comparison included only species having ≥3 markers of each type, and the report’s discussion quotes the same result as p=3.4e-125. In the cluster-level table, beneficial-marker clusters were 64.6% core and 20.7% singleton (N = 60,590 clusters), colonization-marker clusters were 66.6% core and 20.8% singleton (N = 153,092), pathogenic-marker clusters were 45.2% core and 31.0% singleton (N = 374,416), and the genome-wide baseline was 46.8% core and 35.3% singleton. The 46.8% baseline is computed at the gene-cluster level across all species in the pangenome, as the fraction of all gene clusters classified as core by their respective species pangenomes, not as a species-level average; all chi-square tests against this baseline were significant (p≈0). Biofilm genes had an 83.3% core fraction, IAA biosynthesis 78.1%, nitrogen fixation 72.3%, and phosphate solubilization 70.9%; T4SS was 48.5% singleton, coronatine toxin 45.2% singleton, and effectors 39.8% singleton. 1

This architecture supports a model in which beneficial functions are more stably inherited after acquisition, while pathogenic functions remain more accessory and dynamic. The report notes a taxonomic-scope tension with a study of ten Pseudomonas strains that found biocontrol traits to be largely accessory. That study is Loper et al. (2012); the report proposes, rather than demonstrates, that the discrepancy reflects taxonomic scope, because its analysis spans 25,660 species whereas Loper et al. studied 10 strains within a single species group. 1

HGT signals are mixed and scale-dependent

The HGT co-occurrence proxy comprised 986,464 transposase/integrase singleton clusters (722,674 transposase and 263,790 integrase). Among 1,136 plant-associated species, species carrying singleton marker clusters were 16 times more likely to carry transposase or integrase singletons, with Fisher OR = 15.95 and p = 8.8e-20. However, marker singletons had an overall enrichment ratio of 0.78 relative to the genomic average, indicating lower rather than higher mobility overall, with Wilcoxon p < 1e-300. Context variation was higher for pathogenic markers than beneficial markers, 0.534 versus 0.450, but the difference was not significant by Kruskal–Wallis testing, p = 0.122. The report reads this mixed signal as suggesting that individual horizontal gene transfer (HGT) events are common while plant-interaction markers as a class are under stronger purifying selection than the genomic average, likely because they are functionally important once acquired. That explanation is an interpretation of indirect proxy evidence, not a direct demonstration of selection. 1

A per-marker transposase co-occurrence analysis found the strongest HGT-proxy signal among PGP markers: DAPG biocontrol (OR=8.75), ACC deaminase (OR=6.43) and nitrogen fixation (OR=3.76). Among pathogenic markers, T4SS (OR=3.23), T6SS (OR=2.66) and effectors (OR=2.67) also showed significant enrichment. Contig co-location analysis found 498,677 marker-transposase pairs on shared contigs across 18,569 species. T4SS had the most co-located pairs (84,411) and the closest median distance (238 genes). Fifteen marker-transposase pairs were at gene-number distance 1 (immediately adjacent). They spanned effectors, T6SS, T3SS, quorum sensing, cellulase and T4SS, in species including Rhizobium ecuadorense, Phytobacter ursingii and Burkholderia puraquae. Like the other proxies, co-location is indirect evidence of mobility, not an observed transfer. 1

The MGnify genus-level analysis found a higher mobilome burden in plant-associated genera than in non-plant genera, with median 3.7 versus 2.8 mobile elements per genome and Mann–Whitney p = 1.49×10⁻⁵. The report explicitly contrasts this with genome-level studies reporting fewer mobile elements in plant microbiota genomes and interprets the discrepancy as potentially arising from genus-versus-genome scale: accessory elements may be diverse across plant-associated genera while individual adapted strains stabilize useful functions into the core genome. The report cites Levy et al. 2018 and Bograd et al. 2025 as the genome-level studies, and Rajabal et al. 2024 for frequent HGT in rhizosphere/soil genera. It suggests that the contrast may reflect the difference between rhizosphere/soil genera, where HGT is frequent, and endosphere-focused genome-level comparisons; this explanation is proposed rather than tested. The cross-validation drew on the MGnify genome catalogue, comprising 20,473 species across 4 biomes. The taxonomy bridge linked these MGnify species to 1,185 pangenome genera, with 9,603 MGnify species in matched genera. Biome composition was soil 19,472 species (4,229 genera), tomato rhizosphere 579 species (309 genera), maize rhizosphere 336 species (212 genera) and barley rhizosphere 86 species (52 genera). The mobilome analysis covered the soil biome only, with 563 pangenome genera having data. ISEScan, ICEfinder and geNomad detected 17,323 mobilome entries there, including IS elements (IS110, IS701, IS1634, IS256, IS3), plasmids, prophages, terminal inverted repeats and viral sequences. The report separately states that HGT was assessed without GeNomad mobile-element annotations, which concerns the transposase/integrase proxy analysis; it does not reconcile that caveat with this MGnify geNomad usage. 1

Co-occurring genera show weak redundancy rather than complementarity

All 69 genera observed in 348 soil/rhizosphere samples had GapMind data across 80 pathways; the 1,048 co-occurring pairs had mean complementarity of 10.72, below the permutation null (1,000 iterations preserving richness) of 12.74±0.27. The corrected complementarity analysis found that co-occurring plant-associated genus pairs in NMDC soil communities were slightly less complementary than random pairs, with Cohen’s d ≈ −0.4 and permutation p < 0.001. Recomputing the original max-aggregated analysis with the correct raw pair-level standard deviation gave d ≈ −0.43, while prevalence-weighted aggregation gave d = −0.39. The report’s discussion still cites the original d = −7.54 as indicating strong redundancy, consistent with Louca et al. (2018), who argued that functional redundancy is an emergent property of microbial systems; that magnitude was later revised. The original NB06 value of d = −7.54 was caused mainly by dividing by the standard deviation of permutation means rather than the raw pair-level standard deviation: approximately 90% of the apparent 20-fold reduction came from the formula correction and approximately 10% from the aggregation change. The report states that the prevalence-weighted re-test (NB14) and the Cohen’s d formula correction reduced the original implausible |d|=7.54 to |d|≈0.4. It adds that the direction, redundancy, is robust at this resolution, meaning the 80-pathway GapMind pathway level used in the analysis. The adversarial-review issue table records the max-aggregation issue (I1) as only a partial fix: prevalence-weighted aggregation with a consistent Cohen’s d formula gave an approximately 10% reduction in effect size, with the direction unchanged, while most of the shift from −7.54 to −0.39 came from the formula change rather than the aggregation change. 1

BacDive metabolite-utilization data agreed with GapMind predictions at 83.1% consistency, supporting the reliability of the pathway completeness scores, although the GapMind core-level completeness score was 0% across compartments and may be too stringent for broad taxonomic comparisons. The initial Phase 1 C-score analysis for PGP–pathogen exclusion was not feasible: only 0 PGP-dominant and 2 pathogen-dominant genera were represented in the NMDC co-occurrence data, with the remaining 67 dual-nature. The refined C-score test remained structurally underpowered: among 69 NMDC co-occurring genera, there were 0 PGP-dominant, 3 pathogen-dominant, and 66 dual-or-mixed genera. The report states that the dual-nature class swamps the others whether the Phase 1 25-marker or the Phase 2 17-marker panel is used, so the underpowering is structural in this dataset rather than a marker-panel artifact. This formal re-test on the Phase 2 refined cohorts showed the same pattern as Phase 1, so the design target of ≥5 PGP-only genera in NMDC for C-score analysis (verification #12) was not met; the report presents this failure as a result and states that it does not change the complementarity hypothesis (H3) verdict. 1

Plant-enriched gene families retain signal after phylum and genome-size controls

Server-side aggregation of 93M eggNOG annotations produced OG-level prevalence for 5,671 eggNOG ortholog groups (OGs) passing the 5% prevalence filter. Of these, 5,341, or 94.2%, were significant in raw plant-versus-non-plant Fisher’s exact tests with Benjamini–Hochberg false-discovery-rate (BH-FDR) correction, and 3,840 had OR > 2. The report treats this near-saturation as evidence of broad taxonomic compositional differences rather than plant-specific enrichment; it states that the test is effectively detecting taxonomic divergence and that the meaningful signal comes from the phylum-controlled subset. Top raw enriched OGs included COG3569 (OR=8.92), COG1764 (OR=9.29), COG5343 (OR=7.74), COG0654 (OR=12.66), and COG1845 (OR=14.47). After phylum-level control, 50 enriched gene families retained significance; COG3569, a COG (Clusters of Orthologous Groups) family, had Fisher OR = 8.92 and phylogenetically controlled OR = 6.01, with q = 7.7e-242, and occurred in 54% of plant-associated species versus 12% of non-plant species. The report describes these 50 as novel OGs, all with p<0.05 and OR>1. The paired adversarial review flagged the “novel OGs” wording as overstated (issue I2), and the report reframes these groups as “enriched gene families”. It interprets the attenuation from Fisher OR to phylogenetically controlled OR for COG3569, 8.92 to 6.01, as partial phylogenetic confounding, because plant-associated species cluster taxonomically. In the report’s reading, a substantial ecological signal persists after correction. However, although all 50 top OGs survived phylum-level logistic regression with 28 large phyla as fixed effects, all of those models showed convergence warnings (likely due to quasi-separation when some phyla are entirely plant-associated or entirely non-plant), so the report states that their p-values and odds ratios, including the controlled OR of 6.01, should be treated as approximate. 1

All 50 enriched gene families were annotated using eggNOG descriptions, InterProScan domain hits across 18 databases, GO terms and MetaCyc pathways. Of these, 50/50 had eggNOG descriptions and InterProScan domains, 48/50 had GO terms (699 unique), and 39/50 had MetaCyc pathways. The 1,206,652 gene clusters carrying these OGs span 14,329–16,166 species per OG. Their dominant themes were unknown-function COG category S, 19/50 or 38%; electron transport and energy metabolism, 7/50 or 14%; inorganic ion transport, 6/50 or 12%; carbohydrate metabolism, 4/50 or 8%; and secondary metabolite biosynthesis, 3/50 or 6%. The report highlights cytochrome c oxidase subunits, NADH dehydrogenase, ubiquinol–cytochrome c reductase, and IscA iron-sulfur cluster biosynthesis as candidate contributors to rhizosphere or microaerobic competitiveness. The report states that Phase 2 annotation (NB09) found all 50 OGs to be well-characterized proteins, none hypothetical, and that every OG has a characterized function. Even so, the 19/50 unknown-function groups have characterized domains but no clear plant-interaction mechanism, and the report presents them as candidates for experimental characterization. More broadly, the report frames the 50 enriched OGs as candidates for previously unrecognized plant-interaction functions rather than demonstrated mechanisms. It suggests that the dominant themes (electron transport, iron-sulfur biosynthesis, ion transport) point to enhanced metabolic competitiveness in the rhizosphere rather than plant-specific signaling, and it highlights the enrichment of the high-affinity cytochrome oxidases COG1845 and COG0843 in this context; the competitiveness reading is an interpretation, not a test. It describes the inorganic ion transport theme as consistent with the nutrient-acquisition demands of rhizosphere colonization. 1

Electron transport and energy metabolism accounted for 7/50 (14%) of the plant-enriched OGs. Named hits include cytochrome c oxidase subunit COG1845/ctaE, with OR=14.5 and presence in 93.2% of plant versus 48.7% of non-plant species. They also include COG0843/ctaD, with OR=13.7, NADH dehydrogenase COG0649, and ubiquinol-cytochrome c reductase COG0723. The report links the enrichment of high-affinity terminal oxidases to Preisig et al. (1996), who showed that cbb3-type cytochrome oxidases enable respiration in microaerobic root nodule environments. COG0316/iscA, an iron-sulfur cluster biosynthesis group, occurred in 94.3% of plant-associated species. The report’s future-directions text states that the enrichment of high-affinity cytochrome oxidases (COG1845, COG0843) in plant-associated species suggests microaerobic adaptation as a key niche axis. The microaerobic-adaptation mechanism remains a hypothesis: the report states that experimental measurement of oxygen tolerance ranges in plant-associated versus non-plant strains would test it. All 50 OGs had core gene fractions of 60.1–83.1% against a 46.8% baseline, which the report reads as reinforcing stable core inheritance of plant-enriched functions. Cross-reference with MGnify rhizosphere KEGG modules found 8 modules from novel OGs present in rhizosphere species. 1

An earlier Phase 2b draft used a prevalence-based Bernoulli simulation for the per-OG test, which the paired adversarial review flagged as circular by construction; it was replaced with the real per-species OG matrix from NB03 and re-run. In the re-run, a real per-species logistic regression regressed each gene family’s presence vector on plant status, log₁₀ genome size, and top-10 phylum dummies across 22,384 species with complete covariates (1,134 plant-associated, 21,250 non-plant), and converged for 48/50 gene families. All 48 retained positive plant-association coefficients with absolute coefficient > 0.2 and Benjamini–Hochberg false-discovery-rate (BH-FDR) q < 0.05; controlled odds ratios for the top ten ranged from 4.4 (COG5516) to 6.6 (COG1845, cytochrome oxidase ctaE), and q-values were as small as 10⁻⁹⁶ (COG3569). The report summarizes this as 48 of 50 OGs (96%) retaining a large, significant positive plant-association coefficient after phylum and log₁₀(genome size) control. The genome-size coefficient was strongly positive, with a median of 3.4 on the log₁₀ scale, so plant-associated species do have larger genomes. That genome-size effect was absorbed by the log₁₀(genome size) term and did not explain away the plant-association effect: both coefficients were simultaneously large and positive. The two non-convergent cases, COG2351 and COG5343, had strong raw prevalence gaps, with prevalence 3× higher in plant than in non-plant species (62% vs 23% and 64% vs 21% respectively), and singular design matrices; the report proposes perfect separation by a phylum dummy as the likely cause, which is an explanation rather than an established one. The report therefore interprets them as numerical failures rather than biological rejections. Controlled effects for these two OGs were not estimated. The adversarial-review issue table records C4 (genome-size confound) as fixed by this real per-species test and pairs it with a quartile analysis: the dual-nature rate was 54% in the bottom genome-size quartile versus 87% in the upper quartiles (the table writes “top”; the Caveats below describe this group as the top three quartiles), which the table treats as confirming an observational genome-size gradient. 1

Refined markers confirm widespread dual-nature genomes but not a useful categorical classifier

The Phase 1 marker survey searched a curated set of 91 marker genes, described as 39 PGP, 44 pathogenic and 9 colonization markers (category counts that do not add up to the stated total), across bakta annotations, Pfam domains, KEGG KOs and product descriptions, yielding 588,098 marker gene clusters across 25,660 species. No Pfam domain hits were found, which the report attributes to the query format in bakta_pfam_domains rather than to the separate domain-coverage gap, so classification relied on gene name, KEGG KO and product keyword matches. The most prevalent functional categories were T6SS products (80,324 clusters), chemotaxis (69,986), quorum sensing (57,201), T3SS products (52,247) and T4SS (48,145); among PGP functions, phosphate solubilization (23,876), phenazine (12,727), biofilm (7,512) and nitrogen fixation (6,139) were the most common. The initial 91-marker panel included ubiquitous bacterial functions that classified even Escherichia and Salmonella as 100% dual-nature: flagella, chemotaxis, T6SS, biofilm, quorum sensing, and T2SS. It was reduced to 17 plant-specific markers, 9 PGP and 8 pathogenic, by removing 6 ubiquitous categories that had only 1.2–2.4× enrichment in plant vs non-plant species, and KEGG module gating required at least three genes for T3SS, at least two genes for nitrogen fixation, and three genes for T4SS. Under the T3SS (M00332) gate, T3SS-positive species decreased from 10,300 to 1,661. The report labels this change as −8,855 false positives and an 86% reduction, but those stated figures do not agree with the stated before-and-after counts, so the exact size of the change should be treated as unresolved. Under the nitrogen fixation (M00175) gate, nitrogen-fixation positives decreased by 346, or 14%. The T4SS (M00333) gate had minimal impact because most hits were already multi-gene. Across all 25,660 species, the refined cohorts were dual-nature 12,463 (48.6%), pathogenic 6,272 (24.4%), beneficial 3,394 (13.2%) and neutral 3,531 (13.8%). Despite stricter marker definitions, 878 of 1,115 plant-associated species, or 78.7%, were dual-nature, with 108 pathogenic, or 9.7%; 94 beneficial, or 8.4%; and 35 neutral, or 3.1%. The report describes this as an increase from 60.3% and attributes it to retained markers that genuinely co-occur in plant-associated genomes: phosphate solubilization at 57.9% prevalence, CWDE (cell-wall-degrading enzyme) cellulase at 63%, phenazine at 63%, and T4SS at 45.8%. On that basis it concludes that the high dual-nature rate is biologically real. However, the 60.3% comparator is the NB02 rate of 15,474/25,660 species overall, not a rate among the same plant-associated population. Because the two populations differ, the comparison does not establish a within-population increase caused by marker refinement. 1

The broader NB02 binary-presence scheme classified 15,474 of 25,660 species, or 60.3%, as dual-nature because they carried at least one PGP and one pathogenic marker. The more assumption-laden NB07 composite scoring scheme classified 38.0% as neutral, 29.5% as pathogenic, 25.2% as dual-nature, and 7.3% as beneficial. The difference reflects the distinction between carrying both marker types and scoring highly on both genomic and metabolic axes. The report calls this a 35 percentage point difference between schemes. NB07 weighted 40% PGP markers, 20% core fraction, 15% complementarity, 10% metabolic breadth, and a 15% pathogen penalty, with adaptive thresholds set at the median of non-zero scores. The report describes NB02 as more inclusive and NB07 as more discriminating but also more assumption-laden. The NB07 passage does not state its denominator. Elsewhere the report states that 65–85% of plant-associated species per compartment carry both PGP and pathogenic marker genes. The report’s 30 genus dossiers showed that plant-associated genera nearly universally carry both PGP and pathogenic markers. In the dossier table, PGP ratio = sum(n_pgp) / sum(n_pgp + n_pathogen) across plant-associated species in the genus. Values < 0.5 indicate a genus that is more pathogen-marker-rich than PGP-marker-rich (e.g., Xanthomonas at 0.28, Chryseobacterium at 0.23). Cluster-robust markers are those that survive the cluster-robust GLM at q < 0.05 BH-FDR (NB14, item 18). MGnify BGC counts come from the soil biome only, and ”—” means the genus is not in MGnify soil. The full 30-genus dossier, with all 23 columns, is in data/genus_dossiers_v2.csv. Dossier rows include Chryseobacterium (11 plant-associated species, dual-nature cohort, PGP ratio 0.23, markers cellulase, pectinase, effector and phenazine, and 8 MGnify soil-biome BGCs), Frankia (10 plant-associated species, dual-nature, PGP ratio 0.60, markers cellulase, pectinase, N-fix, phenazine and phosphate-sol), Agrobacterium (10 plant-associated species, dual-nature, PGP ratio 0.32, markers cellulase, pectinase, effector, phenazine and phosphate-sol, with A. pusense listed in its row) and Xanthomonas (12 plant-associated species, dual-nature, PGP ratio 0.28, markers cellulase, pectinase, effector, phenazine, phosphate-sol and T3SS, with H7-significant X. campestris and X. vasicola and the corresponding X. campestris × Brassica and X. vasicola × Zea mays H6 host associations). The report’s discussion also calls the finding that 60–85% of plant-associated bacteria carry both PGP and pathogenic markers perhaps its most striking; that range differs from the 65–85% per-compartment statement, and marker carriage alone does not establish phenotypic duality. It names three key dual-nature genera. 1

  • Pseudomonas_E, 358 species, with PGP markers HCN, phenazine, and phosphate solubilization and pathogenic markers T2SS, T3SS, T6SS, and effectors.
  • Bradyrhizobium, 73 species, with nitrogen fixation and ACC deaminase alongside T3SS, T4SS, and CWDE.
  • Streptomyces, 261 species, with phenazine and phosphate solubilization alongside CWDE and T3SS. 1

A curated validation panel of 18 model organisms with experimentally confirmed phenotypes (7 beneficial, 7 pathogenic, 4 neutral) was used, and all 18 were found in the refined cohort data. The report’s primary finding was that categorical cohort assignment is uninformative at the species level: all 14 known beneficial or pathogenic species were assigned to the dual-nature class, while all four neutral controls were misclassified: Escherichia coli K-12 and Mycobacterium tuberculosis as dual-nature, Deinococcus radiodurans as beneficial, and Thermus thermophilus as pathogenic. The continuous pathogenicity ratio, n_pathogen / (n_pgp + n_pathogen), was lower in known beneficial species than in known pathogens, with medians of 0.50 and 0.60, respectively; Mann–Whitney U = 9 and p = 0.027 for N = 7 versus 7. The report therefore treats categorical cohorts as a coarse screen and the continuous pathogen ratio as the discriminating signal. The adversarial-review issue table nonetheless records this C3 fix (circular genus-level validation) as partial: the continuous pathogen-ratio Mann–Whitney test (p=0.027, N=7+7) replaces the tautological relaxed-accuracy figure, while categorical cohort assignment remains uninformative at species level. The seven known beneficial species were P. simiae WCS417, P. protegens CHA0, B. velezensis FZB42, A. brasilense Sp245, R. leguminosarum, P. phytofirmans PsJN, and S. meliloti. The seven known pathogens were P. syringae, R. solanacearum, X. campestris, A. tumefaciens, E. amylovora, P. carotovorum, and C. michiganensis. The report calls the continuous-ratio difference a small effect at small sample size. It also states that the earlier Phase 1 figure of 92.7% genus-level agreement was tautological, because cohorts are defined from marker presence. An earlier Phase 2b draft reported a “77.8% relaxed accuracy” figure, which counted dual-nature as correct for both beneficial and pathogenic ground truth. The paired adversarial review noted that this rule cannot fail for any species assigned to dual-nature. The figure is therefore arithmetically 14/18 but methodologically tautological, contains no discriminative information, and is withdrawn. The report does not treat the neutral-control misclassifications as failures of the panel to detect plant adaptation. It attributes them to the generality of the marker set: flagella, secretion, chemotaxis and biofilm functions overlap with “beneficial” or “pathogenic” markers in all bacteria. The report proposes a stronger follow-up validation that would add near-pure beneficial organisms (Methylobacterium extorquens) and near-pure pathogens (Xylella fastidiosa) with known minimal PGP or pathogenicity machinery. 1

The marker association results were also scale-dependent. The report states that NB10’s statsmodels logit with genus fixed effects failed to converge, in its wording for 0/14 markers (a wording the report’s own data table contradicts; see Caveats), and attributes this to plant association being heavily confounded with genus-level taxonomy. Phase 2b ran three increasingly strict tests of the plant-association signal after phylogenetic control. First came an L1-regularized logit (scikit-learn, saga, C=1.0) with top-20 genus dummies plus log₁₀(genome_size) and a 100-replicate bootstrap. It found 9/14 markers with bootstrap 95% CIs excluding zero. The report cautions that L1 bootstrap CIs are biased near zero (Chatterjee & Lahiri 2011) and that n_bootstrap=100 is small. The paired adversarial review flagged this L1-bootstrap version as suggestive rather than confirmatory. 1

Second, item 18 of the open-issues list closed C1 with a cluster-robust generalized linear model (GLM). This was an unpenalized statsmodels binomial GLM with top-10 phylum fixed-effect dummies plus log₁₀(genome_size) and cluster-robust covariance with 7,555 genus clusters across 24,554 species. It gives proper Wald p-values and 95% Wald CIs with no L1 shrinkage. The report calls it the pragmatic frequentist analogue of a phylogenetic generalized linear mixed model (PGLMM) at species scale. It found 8/14 markers significant at BH-FDR q < 0.05 with absolute coefficient > 0.2, including ACC deaminase OR = 8.01, T3SS OR = 2.71, nitrogen fixation OR = 2.71, phenazine OR = 2.12, CWDE cellulase OR = 1.87, phosphate solubilization OR = 1.68, CWDE pectinase OR = 1.59, and effector OR = 1.35. 13 of 14 markers agreed with the L1 verdict. The one disagreement was iaa_biosynthesis, with a significant L1 coef = 0.61 but a cluster-robust coef = −0.09, p = 0.82, not significant. The report reads this as the kind of L1-shrinkage-near-zero artifact the adversarial review had flagged. The report presents the 8/14 result as stricter than the suggestive 9/14 L1-bootstrap result. It regards a full PGLMM with GTDB-tree-derived random-effect covariance as theoretically stronger, but intractable because the 25K x 25K dense covariance matrix is too large. The genus-level cluster-robust GLM is therefore the practical analogue. 1

  • Cluster-robust GLM results for all 14 markers (coefficient; OR; 95% CI; BH-FDR q):
  • Robust significant: acc_deaminase 2.08; 8.01; [4.19, 15.32]; 2.1×10⁻⁹.
  • Robust significant: t3ss 1.00; 2.71; [1.94, 3.80]; 2.6×10⁻⁸.
  • Robust significant: nitrogen_fixation 1.00; 2.71; [1.54, 4.78]; 1.3×10⁻³.
  • Robust significant: phenazine 0.75; 2.12; [1.69, 2.64]; 5.9×10⁻¹⁰.
  • Robust significant: cwde_cellulase 0.63; 1.87; [1.46, 2.40]; 2.5×10⁻⁶.
  • Robust significant: phosphate_solubilization 0.52; 1.68; [1.20, 2.35]; 4.8×10⁻³.
  • Robust significant: cwde_pectinase 0.46; 1.59; [1.25, 2.01]; 3.4×10⁻⁴.
  • Robust significant: effector 0.30; 1.35; [1.09, 1.67]; 1.2×10⁻².
  • Not significant: dapg_biocontrol −0.02; 0.98; [0.41, 2.33]; 0.96.
  • Not significant: t4ss −0.02; 0.98; [0.72, 1.34]; 0.96.
  • Not significant: hydrogen_cyanide −0.04; 0.96; [0.52, 1.76]; 0.96.
  • Not significant, labelled an L1 false positive: iaa_biosynthesis −0.09; 0.91; [0.40, 2.06]; 0.96.
  • Not significant: siderophore −0.33; 0.72; [0.44, 1.17]; 0.26.
  • Not significant: acetoin_butanediol −0.38; 0.69; [0.46, 1.03]; 0.11. 1

Third, within-genus label shuffling retained only 3 of 15 testable markers: nitrogen fixation (perm p < 0.001), ACC deaminase (perm p < 0.001) and T3SS (perm p = 0.005). The other 12 markers failed the test: phenazine, IAA biosynthesis, CWDE cellulase, CWDE pectinase, effector, DAPG biocontrol, hydrogen cyanide, siderophore, acetoin/butanediol, phosphate solubilization, T4SS and phytotoxin. The report identifies three markers carried by specific plant-associated species within genera where non-plant species do not reliably have them: the nif nitrogen-fixation operon, the ACC deaminase enzyme, and the hrp/hrc T3SS structural cluster. They are the only such markers. The test used n = 200 permutations, which is under-resolved for p ≤ 0.005, so T3SS is marginal; no multiple-testing correction was applied across the 15 markers, and Bonferroni α = 0.0033 would demote T3SS while nitrogen fixation and ACC deaminase remain significant. The report’s combined three-tier verdict is 3 markers species-level robust under any test: nitrogen fixation, ACC deaminase and T3SS, which have large coefficients, proper Wald CIs not crossing zero, and survive within-genus permutation. Five markers are cassette-level: phenazine, CWDE cellulase, CWDE pectinase, phosphate solubilization and effector are robust under the cluster-robust GLM but fail the within-genus shuffle, meaning that when a genus is plant-associated the cassette tends to be acquired but is not species-specific within the genus; the report calls this genus-acquired plant adaptation still ecologically meaningful but not a species-level distinguisher. A further 6 markers were not significant in any test: DAPG biocontrol, T4SS, hydrogen cyanide, IAA biosynthesis, siderophore and acetoin/butanediol, which had been suggested by L1-bootstrap (IAA) or were known weak signals. The cluster-robust GLM is reported over 14 markers, whereas the within-genus shuffle is reported as 3/15. The report concludes that the headline dual-nature rate (78.7%) should be read as a genus-level genomic background associated with plant-associated lineages. The same applies to the compartment-specific marker enrichments. Neither is species-level functional evidence of plant adaptation. The adversarial-review issue table records C1 (phylogenetic confounding) as fixed under this three-tier scheme: 8/14 markers survive the genus-clustered GLM at q<0.05 with proper Wald CIs, with iaa_biosynthesis as the one L1 false positive, and combined with the within-genus shuffle (3/15) the verdict is 3 species-level, 5 cassette-level and 6 not robust markers. 1

The report’s mechanism-hypothesis table assigns High confidence to four mechanisms: nitrogen fixation (nifHDK) and siderophore iron acquisition as beneficial, and T3SS effector injection and cell wall degradation (CWDE) as pathogenic. It assigns Medium confidence to four more: ACC deaminase stress relief and biocontrol antimicrobials (DAPG/phenazine) as beneficial, and context-dependent T6SS and PGP + secretion system co-occurrence as dual-nature. These ratings concern proposed mechanisms, not tested ones, and they address a different question from the statistical plant-association tests above. 1

A Phase 1 sensitivity analysis reclassified T3SS, T6SS and T2SS from “pathogenic” to “colonization” markers. This changed 16.4% of dual-nature species to PGP-only, but plant-associated species remained 86.0% dual-nature. The report takes this as showing that the dual-nature finding is not driven solely by secretion-system annotations. In dual-nature species, the most prevalent pathogenic markers were T6SS products (64%), T2SS (55%), T3SS products (50%) and chemotaxis (52%). The most prevalent PGP markers were quorum sensing (49%), phenazine (34%) and flagella (40%). The report notes that chemotaxis and flagella, which are general motility functions, contribute substantially to the dual-nature classification. These analyses used the Phase 1 panel, and several of its ubiquitous categories were later removed. 1

A PGP-versus-pathogen composite-score scatter placed B. subtilis, R. leguminosarum, B. japonicum and S. meliloti in the expected quadrants. After marker refinement, positive controls were Rhizobium 97% beneficial+dual, Bradyrhizobium 100%, Azospirillum 100% and Mesorhizobium 99%. Negative controls improved: Escherichia was PGP-only 25% (down from ubiquitous dual-nature in Phase 1), Salmonella 0% PGP-only, Staphylococcus 18% and Mycobacterium 6%. The positive-control rates count dual-nature as consistent with a beneficial control, a rule similar to that of the withdrawn relaxed-accuracy figure. They also contrast with the species-level panel above, in which all four neutral controls were misclassified. 1

Host and subclade structure is concentrated in a minority of species

Host extraction parsed 11,852 genomes with identifiable plant hosts from NCBI isolation_source and identified 1,307 species across 19 testable hosts with at least 10 species each. The most represented hosts were rice with 286 species, Arabidopsis thaliana with 184, wheat with 122, and maize with 98. The remaining testable hosts were soybean (84), potato (58), tomato (57), sugarcane (47), lettuce (34), banana (30), pepper (25), cotton (24), grape (22), strawberry (18), peanut (17), citrus (14), tea (13), tobacco (12) and apple (10). MGnify detected 17 genera across all three crop rhizospheres, including Pseudomonas_E, Streptomyces, Variovorax, Telluria and Acidovorax. The report calls these 17 genera a candidate core rhizosphere microbiome that could inform crop-agnostic biocontrol strategies; it is a candidate, not a validated consortium. Another 117 genera were unique to tomato, 54 to maize, and 5 to barley. The report proposes these host-specific genera as candidates for crop-specific biocontrol formulations; such formulations have not been validated. In MGnify, T3SS prevalence was roughly 2× higher in rhizosphere biomes than in bulk soil (12.3%): tomato 24.0%, maize 21.6% and barley 22.3%. The report treats this as independently confirming T3SS as a marker of plant association. In the KEGG module comparison, M00332 (T3SS) was enriched in all 3 rhizosphere biomes (22–24%) over soil (12.3%). Modules M00238, M00769, M00136, M00615 and M00438 were 2–3× enriched in rhizosphere. Its discussion summarizes the MGnify result as a 2× T3SS enrichment in rhizosphere versus soil, consistent with Gonzalez-Vallina et al. (2024) and Tampakaki (2022), who found T3SS widespread in both beneficial and pathogenic plant bacteria. In the soil biome, 17,608 BGCs from 579 pangenome genera with data were classified into MiBIG categories: Polyketide (23.1%), NRP (13.9%), Saccharide (13.2%), RiPP (12.6%) and Terpene (12.7%). Average BGC-to-MiBIG similarity scores ranged from 0.38–0.89. MGnify BGC profiles showed 84 plant-associated genera producing NRP/siderophore BGCs. The top producers were Mycobacterium at 406, Pseudomonas_E at 125 and Streptomyces at 113. The report reads this as consistent with the known importance of siderophore-mediated iron acquisition in plant colonization. Only 11.7% Jaccard overlap existed between 487 pangenome-derived plant genera and 438 MGnify rhizosphere genera, reflecting different isolation-metadata and metagenomic-detection processes. 1

Among 65 plant-associated species with at least 20 genomes, only 18, or 28%, had any phylogenetic-tree distance data in kbase_ke_pangenome.phylogenetic_tree_distance_pairs. The report documents this database-coverage limitation in docs/pitfalls.md. The Phase 2b 65-species scan therefore closed the genome-ID linkage and then ran the available 18 species; 47 lacked coverage, including Bradyrhizobium japonicum, Bradyrhizobium diazoefficiens, Mesorhizobium ciceri, Sinorhizobium medicae, Burkholderia glumae, Methylobacterium extorquens, Streptomyces scabiei, Xylella taiwanensis, and Clavibacter michiganensis. Seventeen species met the testability criteria after the genome-ID fix: at least 3 plant genomes, at least 3 non-plant genomes and at least 2 subclades. Of these, 5, or 29%, passed Bonferroni-corrected Fisher tests at α = 0.05/17 = 0.0029: Xanthomonas vasicola, p = 1.6×10⁻⁴; Mesorhizobium sp002294985, p = 1.1×10⁻⁴; Agrobacterium pusense, p = 4.5×10⁻⁴; Pseudomonas_E avellanae, p = 7.3×10⁻⁴; and Xanthomonas campestris, p = 1.0×10⁻³. The report’s issue list summarizes this as 5/17 testable species passing Bonferroni-Fisher (3 also Cochran-valid), while 47/65 candidate species lack phylo tree coverage. The per-species table reports genome totals, plant-genome counts, best-subclade plant share and BH-FDR q-values, listed below. 1

  • Xanthomonas vasicola: N = 101, 30 plant, best subclade 71% plant, q = 0.0013.
  • Mesorhizobium sp002294985: N = 84, 78 plant, best subclade 97% plant, q = 0.0013; small-cell, Fisher only.
  • Agrobacterium pusense: N = 80, 13 plant, best subclade 100% plant, q = 0.0026; small-cell, Fisher only.
  • Pseudomonas_E avellanae: N = 319, 68 plant, best subclade 31% plant, q = 0.0031.
  • Xanthomonas campestris: N = 128, 42 plant, best subclade 42% plant, q = 0.0034. 1

Three of those five species—Xanthomonas vasicola, Pseudomonas_E avellanae, and Xanthomonas campestris—also satisfied Cochran’s rule for chi-square validity and are the most robust. The other two passed only via Fisher’s exact test, which is exact and not subject to the small-cell constraint. A host-by-subclade analysis was layered on the same scan. It was restricted to the 9 species with sufficient plant-host annotation: at least 5 host-annotated genomes, at least 2 hosts and at least 2 subclades. It found two species passing both within-species and across-species Bonferroni corrections: Xanthomonas campestris segregated toward Brassica hosts, with 46/47 genomes in subclade 2 on Brassica and p = 2.7×10⁻¹² raw, 3.3×10⁻¹¹ after correction; Xanthomonas vasicola segregated toward Zea mays, with 47/52 genomes in subclade 0 and p = 1.4×10⁻¹¹ raw, 1.7×10⁻¹⁰ after correction. Ralstonia solanacearum passed BH-FDR but not across-species Bonferroni, with p = 5.9×10⁻³ within species and BH q = 1.8×10⁻². The report labels the X. campestris × Brassica and X. vasicola × maize patterns canonical pathovar-host segregations confirmed at the genomic level. X. campestris pv. campestris is the textbook Brassica pathogen, and X. vasicola pv. vasculorum is a known maize/sugarcane pathogen. It pairs them with the NB11 host-specificity matrix, which identified 117 tomato-specific, 54 maize-specific, and 5 barley-specific genera. 1

The original within-species subclade analysis (NB12) selected five species with ≥175 genomes and high plant-associated fractions and grouped their genomes by phylogenetic distance-based agglomerative clustering: X. euvesicatoria, 332 genomes, 169 (51%) plant-associated, 3 subclades; P. avellanae, 319 genomes, 68 (21%) plant-associated, 3 subclades; S. meliloti, 241 genomes, 191 (79%) plant-associated, 2 subclades; P. amygdali, 239 genomes, 32 (13%) plant-associated, 3 subclades; and R. laguerreae, 175 genomes, 139 (79%) plant-associated, 2 subclades. Pairwise phylogenetic distances from phylogenetic_tree_distance_pairs (22.6M rows) were used to build distance matrices for MDS visualization and clustering. In the Phase 2 pass all 5 species showed 0 testable subclade × plant-association enrichments, because a genome-ID mismatch (bare NCBI accessions versus GTDB-prefixed IDs) dropped the plant-associated subset to zero. That is a data-plumbing null, not biological evidence against segregation. 1

The original NB12 result for H7 was null: 5 species were tested and 0/5 were significant. The report attributes it to a genome-ID mismatch: phylogenetic tables used bare NCBI accessions, while environmental metadata used GTDB-prefixed identifiers. Adding GB_ to GCA_ accessions and RS_ to GCF_ accessions recovered 1,306/1,306 genomes and 599 plant-associated genomes that NB12 had silently dropped; after the faulty merge, the is_plant_associated column had been entirely NaN. With the fix, the original 5-species set showed 1/5 significant under all metrics. The full 18-species scan raised this to 5/17 by including the broader candidate pool. The issue list records the plumbing as resolved in NB13 (100% overlap) and the downstream H7 as extended in item 19. In that extension, the H6 host × subclade analysis found 2/9 species with robust pathovar-host segregation. P. amygdali had a borderline chi-square p = 0.031 in the 5-species subset. It did not survive Bonferroni in the larger scan and remained invalid under Cochran’s rule. The report adds that this earlier chi-square-only flag also fails the Fisher 2×2 collapse, and it points to the corrected Phase 2b verdict as canonical. Twelve testable species showed no such clustering. The final interpretation is weak support for within-species plant-association segregation in 5/17 testable species, with strongest evidence in Xanthomonas and Pseudomonas, but the true effect across all 65 candidates cannot be estimated until KBase Data Lakehouse phylogenetic-tree coverage improves. The strongest signal across both Fisher’s exact and Cochran-valid chi-square came from X. vasicola (Fisher p = 1.6×10⁻⁴), P. avellanae (Fisher p = 7.3×10⁻⁴) and X. campestris (Fisher p = 1.0×10⁻³). The report compares the 5/17 result with two partial-support patterns in the literature. Melnyk et al. (2019) found that plant-lifestyle transitions in Pseudomonas are driven by convergent gain or loss of accessory genomic islands rather than by core-genome phylogeny. Satterwhite & Bergelson (2025) found local host adaptation in 3/5 P. syringae pathovars but only 1/5 locally dominant. The report infers that within-species plant adaptation is predominantly accessory-driven. It also proposes that in specific genera, notably Xanthomonas, accessory cassettes have stratified onto core-phylogenetic subclades, so that pathovar designations have a real within-species phylogenetic basis. This is an interpretation rather than a direct test, because accessory-gene profiling was not run. Among the 30 genus dossiers, 4 genera contain at least one species with significant subclade × plant-association segregation (Pseudomonas_E via P. avellanae, Mesorhizobium via sp002294985, Xanthomonas via both X. campestris and X. vasicola, and Agrobacterium via A. pusense), and only Xanthomonas also has species with H6 host segregation (X. campestris × Brassica, X. vasicola × Zea mays); the report presents Xanthomonas as the genus where strain-level plant adaptation is both phylogenetically structured and host-specialized at the genomic level. A discussion passage of the report still describes H7 as a null result, with no subclade-level segregation of plant-associated genomes, consistent with Melnyk et al. (2019); that passage is not reconciled with the corrected 5/17 result, so it should not be read as the settled H7 verdict. 1

Annotation-table limitations were identified and operationally resolved

An audit of 22 marker Pfams found that 12 were absent from bakta_pfam_domains despite being abundant in interproscan_domains. The audit (pfam_bakta_ips_audit.csv, 2026-04-25) found that InterProScan contained all 22 project marker Pfams. The missing set was dominated by secretion-system components, with InterProScan hit counts of PF00771 18.6K, PF01313 13.6K, PF01514 15.0K, PF02579 16.8K, PF04610 15.6K, PF07916 8.4K, PF09599 0.3K and PF01312 22.0K for T3SS; PF03135 11.0K and PF03743 17.5K for T4SS; and PF05936 7.5K (Hcp tube) and PF04183 7.5K for T6SS. The 10 Pfams that are present in bakta, mostly nitrogen fixation, CWDEs and a few others, still returned only 10–35% of the InterProScan hit count, so even those are systematically undercounted but non-zero. The report proposes, as the likely cause, that bakta uses a reduced “core” Pfam HMM profile set while InterProScan uses the full Pfam-A release. This coverage gap is distinct from the NB02 zero-hit query-format failure. The report’s issue list (item 21) records the project as unaffected because NB10 uses InterProScan. 1

The Phase 1 NB02 query used exact matching (WHERE pfam_id IN (‘PF00771’, …)) and returned zero rows because bakta_pfam_domains stores versioned identifiers such as PF00771.22; that bakta-only step fell back to gene-name keyword matching. Re-running with LIKE prefix patterns on the 10 Phase 2b marker Pfams recovered 19,364 domain hits across 7,962 species. Only five of those 10 Pfam IDs produced hits: nitrogenase NifH PF00142 (6,632 hits, 4,040 species, 32.4% core), cellulase GH5 PF00150 (6,898 / 3,234 / 25.9%), pectate lyase 3 PF12708 (4,149 / 2,283 / 24.6%), pectate lyase PF00544 (1,115 / 546 / 20.8%) and T6SS VgrG PF05943 (570 / 343 / 10.9%); the remaining five returned 0 hits. The recovery confirmed rather than expanded InterProScan numbers for the secretion-system Pfams, but for nitrogen fixation and CWDEs it added 4,217 species gaining at least one new marker: 2,872 gained nitrogen fixation, 1,082 CWDE pectinase and 811 CWDE cellulase, cases where InterProScan missed hits that bakta picked up. The refined cohort pipeline (NB10) used the InterProScan signature_acc field for T3SS, T4SS, T6SS and CWDE detection, recovering 280,193 InterProScan hits across 24,595 species, so its biological assignments were unaffected. These hits included 2,804 additional species with CWDE domains (cellulase/pectinase Pfams PF00150 and PF00544) not detected in Phase 1’s bakta-only search; the report recommends treating bakta Pfam data as a supplement rather than a substitute for InterProScan. The report states that the query-format issue was resolved in NB13 Cell 3, where a LIKE ‘PF00771%’ pattern recovers the versioned IDs (its example of a stored versioned ID is PF13629.12), while the marker set itself remains literature-curated. It also notes that the higher core-genome fraction of NifH (32.4%) than of the CWDEs (21–26%) is consistent with the H2 pattern of more core-encoded beneficial functions. 1

Data Sources and Generated Outputs

The report draws on these collections and tables:

  • kbase_ke_pangenome (genome, gene_cluster, pangenome, gtdb_taxonomy_r214v1, gtdb_metadata, bakta_annotations, bakta_pfam_domains, eggnog_mapper_annotations, gapmind_pathways): core pangenome data, functional annotations and pathway predictions.
  • kbase_ke_pangenome (phylogenetic_tree, phylogenetic_tree_distance_pairs): within-species phylogenetic distances for Phase 2 subclade clustering.
  • kescience_mgnify (species, genome, gene_mobilome, gene_bgc, genome_kegg_module, genome_cog): the MGnify rhizosphere/soil genome catalogue with mobilome, BGC and KEGG profiles (Phase 2).
  • kescience_bacdive (isolation, sequence_info, metabolite_utilization): BacDive strain-level isolation source and metabolic phenotypes.
  • nmdc_arkin (taxonomy_features, study_table): community-ecology co-occurrence data. 1

Generated output files include:

  • data/genome_environment.csv, 293,059 rows: per-genome compartment classification.
  • data/species_compartment.csv, 26,511 rows: species-level compartment assignments by majority vote.
  • data/pfam_marker_hits.csv, 0 rows: Pfam domain hits from the initial query (none found).
  • data/novel_plant_markers.csv, 50 rows: plant-enriched OGs (labelled “novel” in the file).
  • data/phylo_control_results.csv, 15 rows: NB10 phylogenetic-control logistic regression, recorded as 0/14 converged.
  • data/subclade_host_mapping.csv, 0 rows: NB12 host × subclade mapping, insufficient data (the later H6 extension is described above).
  • data/subclade_full_scan.csv, 18 rows: full 65-species subclade × plant scan with Bonferroni and BH-FDR corrections (2026-04-25).
  • data/cscore_refined_cohorts.csv, 69 rows: C-score PGP–pathogen mutual-exclusion test on refined cohorts (verification #12). The result was 0 PGP-only, 3 pathogen-only and 66 dual-or-mixed, the same pattern as Phase 1. 1

Figures

  • Compartment census showing species distribution across plant compartments (figures/nb01_compartment_census.png).
  • Heatmap of marker enrichment odds ratios across plant compartments (figures/compartment_heatmap.png).
  • Boxplot comparing core fractions of beneficial, pathogenic and colonization categories (figures/core_vs_pathogenic.png).
  • Multi-panel mobility proxy analysis covering singleton ratio, transposase OR and context variation (figures/mobility_proxies.png).
  • Volcano plot of eggNOG OG enrichment (log2 OR vs −log10 q) with known marker annotations (figures/volcano_enrichment.png).
  • Stacked bar of cohort proportions by compartment (figures/nb04_cohort_compartment_bar.png).
  • Synthesis overview of genus profiles, compartment distribution, and genomic architecture (figures/synthesis_overview.png).
  • Novel OG annotation covering functional classification, core fractions, and domain architecture (figures/novel_og_annotation.png).
  • Refined cohort comparison of Phase 1 versus Phase 2 marker panels (figures/refined_cohort_comparison.png).
  • MGnify integration covering the taxonomy bridge, host specificity, mobilome, and biosynthetic gene cluster (BGC) profiles (figures/mgnify_integration.png).
  • Phylogenetic subclade clustering and plant-association mapping from NB12 (figures/subclade_analysis.png).
  • Subclade MDS colored by plant association after the genome-ID fix, NB13 (figures/subclade_corrected.png).
  • Six-panel synthesis of Phase 2b results across H0–H7, NB15 (figures/final_synthesis.png).
  • Complementarity comparison of max-aggregated versus prevalence-weighted aggregation (figures/complementarity_comparison.png).
  • Marker gene survey overview with functional category counts and cohort distribution (figures/nb02_marker_survey.png).
  • Species-level validation panel showing pathogen ratio by ground-truth class for 18 model organisms (figures/species_validation.png).
  • Guild network of genus–genus metabolic complementarity (figures/guild_network.png).
  • Complementarity heatmap (figures/complementarity_heatmap.png).
  • Complementarity heatmap detail of top complementarity pairs (figures/complementarity_heatmap_detail.png).
  • Secretion-system sensitivity analysis, original versus revised cohort distribution (figures/sensitivity_t3ss_t6ss.png).
  • Marker prevalence in dual-nature species (figures/dual_nature_marker_drivers.png).
  • PGP versus pathogen composite scores with known organism annotations (figures/pgp_vs_pathogen_scatter.png).
  • Gene cluster count versus total marker count by cohort (figures/genome_size_vs_markers.png).
  • Random Forest feature importance for compartment classification (figures/feature_importance_compartment.png).
  • Dot plot of transposase co-occurrence odds ratio per functional category (figures/hgt_per_marker_transposase.png).
  • Contig co-location distances of marker–transposase pairs (figures/hgt_contig_distance.png). 1

Caveats and Limitations

Plant-compartment labels were inferred primarily from NCBI isolation_source metadata, whose quality and coverage vary; only 7,995 of 293,059 genomes, or 2.7%, had plant-associated annotations, and the endophyte category contained only 29 species, below the 30-species threshold. The report states that more targeted isolation could expand the endophyte compartment to enable a full four-way compartment comparison. Plant association was determined for 293,059 GTDB r214 genomes from NCBI isolation_source regex matching as the primary source, cross-validated against the ncbi_env EAV table (695 genomes upgraded) and BacDive isolation records; BacDive identified 2,482 plant-related strains but contributed 0 new genome upgrades beyond ncbi_env, which the report reads as complete overlap between these metadata sources for the plant-associated genome set. Species-level classification used majority vote, with mean confidence 0.88. 1

The marker set is literature-curated and gene presence does not establish expression or phenotype. T3SS, T6SS, flagella, chemotaxis, biofilm, quorum sensing, and secretion functions can support colonization, interbacterial competition, beneficial symbiosis, or pathogenicity, and negative controls were frequently classified as dual-nature. In the NB08 negative controls, Escherichia, Salmonella and Clostridioides were 100% classified as dual-nature, and Staphylococcus was 98%. Combining this with the genome-size confound, the report concludes that the NB02 dual-nature rate of 60.3% substantially overestimates genuine plant-adapted duality. It adds that the NB07 composite scoring (25.2% dual-nature) only partially addresses this, because it depends on arbitrary scoring weights. The report documents T3SS classification as context-dependent (adversarial-review issue I3). It cautions that classification as beneficial or pathogenic from genomic markers alone is insufficient: the same T3SS that delivers pathogenic effectors in Pseudomonas syringae facilitates beneficial nodulation signaling in Rhizobium, and context (host genotype, environmental conditions, community composition) determines the outcome (Osayande et al. 2025). RNA-seq or experimental validation is required to determine whether PGP and pathogenic marker sets are co-expressed under the same conditions. The report adds that presence of both marker types does not confirm simultaneous expression. Such validation is needed to tell genuine lifestyle flexibility from annotation artifacts (e.g., T6SS serving inter-bacterial competition rather than pathogenicity). 1

Genome size correlates with total marker count at r = 0.44, p < 1e-300. Among plant-associated species, the dual-nature rate was 54% in the lowest genome-size quartile versus 87% in the top three quartiles; approximately 33% of dual-nature species shifted to neutral after genome-size normalization in an earlier sensitivity analysis. The report states that this gradient was measured directly on refined cohort data and is observational, not model-dependent; the 78.7% refined dual-nature value therefore contains a real genome-size gradient and remains dependent on marker definitions and scoring assumptions. The earlier sensitivity analysis described the normalization effect as meaningful but not eliminating the dual-nature pattern. The limitations recap states that all 50 novel OGs retain positive plant-association coefficients with genome size as a covariate (resolved in NB14). That statement is superseded by the report’s per-species analysis, in which 48/50 converged and were significant, while COG2351 and COG5343 had singular design matrices and do not count as successful adjusted fits. 1

Phylogenetic control remains incomplete. The original genus-level fixed-effect logistic regression is described in the report’s results text as having “failed to converge for 0/14 markers”, whereas the report’s data table lists the same NB10 output as “0/14 converged”. The two wordings are inconsistent. The report’s limitations recap (item 13) states it explicitly: NB10 attempted genus-level fixed-effects logistic regression for all 14 refined markers, and all models failed to converge (0/14). That recap and the data table’s “0/14 converged” both say no model converged. The results-text phrasing is the outlier, and the report does not reconcile it. A third passage, in the Phase 1 compartment results, says the planned genus-level logistic regression was not completed because of a code error (an undefined logit import), so the H1 compartment enrichments are not phylogenetically controlled beyond the PERMANOVA test itself. The report explains the failure as plant association being heavily confounded with genus-level taxonomy, so that standard maximum-likelihood logistic regression cannot fit; the cluster-robust GLM is a practical genus-level analogue rather than a full phylogenetic generalized linear mixed model. The limitations section adds that RESEARCH_PLAN.md specified genus-level fixed effects, but Phase 1 used phylum-level control because of computational cost. It says NB10’s genus-level logit “failed for 0/14 markers”. It describes the partial resolution in NB14: a scikit-learn L1-regularized logit with top-20 genus dummies and log₁₀(genome size) as covariates produced convergent fits, with “9/17 markers” having bootstrap 95% CIs excluding zero. The results text reports this result as 9/14, so its denominator is inconsistent within the report. The canonical 9/14 result supersedes the recap’s 9/17 denominator. The 9/14 figure rests on 100-replicate L1-bootstrap CIs. These are biased near zero and only suggestive. The later genus-cluster-robust GLM found 8/14 markers significant. Phylogeny-versus-ecology variance decomposition remains future work, and compartment profiling uses PERMANOVA rather than logit as the primary test. A full GTDB-tree covariance analysis was not feasible because a dense 25K × 25K covariance matrix would be intractable in memory. The report names sparse approximations or species subsampling as the practical path. The within-genus shuffle used only 200 permutations, which is under-resolved for p ≤ 0.005, and a full permulation-style analysis remains future work. 1

The raw OG enrichment was saturated at 5,341/5,671 significant tests (94.2%). The report cautions that this mostly reflects taxonomic divergence, meaning genome-wide compositional differences, rather than plant-specific functional enrichment. Only the 50 OGs surviving phylogenetic control represent candidate plant-interaction genes. Although 48/50 gene families survived phylum and genome-size control, two models were singular, and the earlier phylum-level models had convergence warnings, which the report attributes likely to quasi-separation when some phyla are entirely plant-associated or entirely non-plant. The report states that these warnings mean the reported p-values and odds ratios should be treated as approximate. Finer-grained control proved intractable. Family-level logistic regression (NB08) found insufficient within-family variation for all 10 OGs tested, indicating that plant association is taxonomically clustered at the family level. The planned genus-level fixed effects were not executed in Phase 1 because of computational constraints. The report’s practical reading is that plant-associated functional enrichment is real but partially confounded by phylogeny. It is difficult to disentangle whether the genes cause plant association or whether plant-associated clades simply retain them. The causal direction between plant association and enriched functions therefore remains unresolved. 1

The complementarity analysis was limited by 80 GapMind pathways, a potentially over-stringent core-level completeness threshold, and loss of 62/322 genera, or 19.3%, in the NMDC taxonomy bridge, which matched 260 of 322 genera (80.7%) to GTDB. The corrected effect is small, with Cohen’s d = −0.39 for prevalence weighting, but unmatched genera could include complementary partners. The report states that the impact of the 62 lost genera on the H3 complementarity analysis is unknown. If they include key complementary partners, the observed redundancy signal could be artifactual. The adversarial-review issue table records the NMDC taxonomy bridge as 80.7% matched (issue I5), with sensitivity to unmatched genera deferred to future work. The report states that the 80-pathway GapMind level may be too coarse, and that its redundancy result does not rule out a complementarity signal at finer-grained metabolic resolution than the 80 GapMind pathway categories can detect. It adds that reaction-level complementarity or substrate-specific analysis could reveal patterns masked by the current pathway-level aggregation. The report records the max-aggregation inflation (d = −7.54) as partially resolved in NB14 by the prevalence-weighted re-test (d = −0.39, same direction). The core-level completeness threshold, which scored 0% across all compartments, remains a separate open question. 1

HGT evidence is indirect because GeNomad mobile-element annotations were unavailable. The transposase/integrase co-occurrence signal, lower overall marker singleton ratio, and MGnify genus-level mobilome enrichment do not provide a single genome-level estimate of horizontal transfer. MGnify BGC, mobilome, and defense data were available only for the soil biome, so those comparisons do not directly contrast rhizosphere with bulk soil. The current proxies are transposase/integrase singletons and MGnify mobilome aggregates. The report proposes replacing them with direct mobile-element annotations from GeNomad mobile-element predictions, to strengthen the mobilome conclusions and resolve the genus-versus-genome scale-dependent signal. 1

The MGnify and pangenome classifications have only 11.7% Jaccard overlap because they measure different phenomena: isolation metadata versus metagenomic detection. The report notes that this low concordance limits the strength of cross-validation conclusions. The report attributes the low concordance to data-source and sampling differences rather than classification error. It notes that 93 pangenome plant genera were found in MGnify soil but not rhizosphere, which suggests broader environmental associations. It also counts 390 pangenome-only plant genera not found in MGnify rhizosphere, which it suggests likely include opportunistic soil bacteria isolated from plants but not rhizosphere specialists, and 341 MGnify-only rhizosphere genera that may include uncultured lineages absent from GTDB; both explanations are tentative. The subclade analysis is additionally limited by phylogenetic-tree coverage, with only 18/65 candidate species represented. Expanded tree coverage, accessory-gene profiling, gene-content trees, and direct mobile-element annotations are needed to resolve these uncertainties. The current subclade method is core-phylogeny-based. Two analyses remain outstanding: accessory-gene presence/absence profiling as an alternative subclade definition, and gene-content-versus-species-tree comparisons to identify HGT-driven incongruences. Both require larger gene-cluster-by-genome data pulls than the project executed. The report also notes that expanding tree coverage would directly increase the power of any within-species phylogenetic analysis. 1

Future Directions

The report prioritizes a sparse or subsampled full phylogenetic generalized linear mixed model (PGLMM) using GTDB-tree covariance; transcriptomic validation of dual-nature organisms; AlphaFold and pathway-based mechanistic analysis of COG3569, COG1845, COG0843, and COG0316; oxygen-tolerance experiments targeting high-affinity cytochrome oxidases; accessory-genome and gene-content-tree analyses of subclade adaptation; expanded KBase Data Lakehouse phylogenetic-tree coverage; finer-grained compartment sampling; reaction-level complementarity; GeNomad integration; SynCom design; and crop-specific biocontrol studies focused on host-specialized lineages. For synthetic-community design, the report proposes, rather than demonstrates, using genus dossiers, core rhizosphere genera, and BGC profiles, following the approach of Song et al. (2026) for Bacillus SynComs. It names the 84 NRP/siderophore-producing plant genera identified in NB11 as prime candidates. It also names the 17-genera core rhizosphere set (NB11) and the species-level Tier-1 markers (nitrogen fixation, ACC deaminase, T3SS) as candidate organisms and gene targets for targeted experiments. All 50 OGs have annotated function from NB09 but lack mechanism-level proposals. For the top OGs (COG3569, COG1845, COG0843, COG0316), AlphaFold structural prediction or systematic cross-referencing with biochemical pathway databases would generate testable mechanistic hypotheses about plant-interaction roles. These would go beyond the current “energy metabolism / Fe-S cluster / inorganic ion transport” classifications. The report also argues that targeted isolation of crop-specific protective genera, optionally screened against the corresponding H6 pathovars, would yield more effective biocontrol agents than generalist approaches. That advantage is prospective, not a measured result. The originally planned NB13 co-occurrence network expansion (Louvain modules, hub centrality, partial Mantel, and NMDC abiotic_features and metabolomics_gold integration) was re-scoped to validation, Pfam and subclade fixes when the paired adversarial review surfaced higher-priority methodological issues; its complementarity sub-item was executed in NB14, and the remaining sub-items are deferred as exploratory enhancements on which the H3 verdict does not depend. 1

Slots Into

Projects cited

  1. Plant Microbiome Ecotypes — Compartment-Specific Functional Guilds and Their Genetic Architectureplant_microbiome_ecotypes · cited 48× · primary source