Integrated functional genomics and safety assessment of plant-growth-promoting Caryophanales from post-maize-cultivation soils

Abstract

This study aimed to evaluate six environmental bacterial strains isolated from post-maize cultivation soils as candidates for agricultural biopreparation development, using an integrated functional genomic and safety assessment framework. Building on experimental validation of plant-growth-promoting activities, the analysis included: plant-growth-promoting traits (PGPT-Pred) using PLABase; carbohydrate-active enzymes (CAZymes) relevant for lignocellulosic crop residue degradation (dbCAN3); secondary metabolite profiles (antiSMASH); and screening for virulence factors and antibiotic resistance genes (ABRicate, BTyper3).

All analyzed strains possess 1,449–1,617 predicted PGPT-encoding genes (24.1–35.9% of total genes), which are strongly shaped by taxonomic relatedness, as confirmed by congruence testing against ANI-based genomic divergence. Paenibacillus amylolyticus 5mez and Priestia megaterium 7psych showed distinct functional profiles compared to Bacillus spp., while Bacillus subtilis sensu lato strains were most similar to each other. Genomic predictions suggest involvement in nutrient acquisition (N, P, K, Fe) and stress mitigation. Secondary metabolite analysis revealed high biosynthetic potential, with non-Bacillus species harbouring a large proportion of unknown gene clusters, indicating underexplored metabolite diversity. CAZyme profiling identified P. amylolyticus 5mez as the most enzyme-rich strain, while B. cereus s.s. zielonkawy showed ligninolytic potential despite low overall CAZyme abundance. The safety assessment identified B. cereus s.s. zielonkawy as toxigenic and unsuitable for use. Of the remaining strains, P. amylolyticus 5mez and Pr. megaterium 7psych demonstrated the most favourable safety profiles, exhibiting no detectable virulence factors or antibiotic resistance genes, justifying their priority use in agricultural biopreparations, pending phenotypic validation. Given the high-dimensional, low-sample-size nature of multi-trait datasets in applied microbial genomics, tailored statistical approaches, including noise-reduction-validated PCA and distance-based congruence testing, were applied; their rationale and limitations are discussed.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12864-026-13058-2.

Background

The world today faces multiple environmental challenges, including ongoing climate change, disruptions to biogeochemical cycles, and various types of pollution. The anticipated grain crop losses due to climate change are estimated to be approximately 3.8% to 5.5% [1]. Simultaneously, the European Union has taken action to achieve climate neutrality, including a radical reduction in the use of pesticides and artificial fertilizers and an increase in the area of ​​organic farming [2]. Taking the above into account, effective and environmentally friendly biopreparations are currently being sought, in particular microbiological preparations for the decomposition of biomass from crop residues [3].

Circa 3.8 billion tons of crop residues are produced annually worldwide [4]. Such residues are often burned, posing a serious threat to both environmental and human health [4]. Those residues are predominantly composed of cellulose, hemicelluloses, and lignins. For example, corn stalk is composed of 40–50% cellulose, 20–30% hemicelluloses, and 10–15% lignins [5]; elements of corn cob (chaff, woody ring, pith) contain 32–36% cellulose, 41–47% hemicelluloses, and 16–19% lignins [6]. The decomposition of crop residues is essential for several reasons: it plays a crucial role in nutrient cycling and soil fertility, and it also affects soil structure, water retention, and the general microbial communities [7].

Literature data indicate that microorganisms with high enzymatic activity (production of extracellular hydrolases) are key to the decomposition of crop residues [3, 8]. Among them, the phylum Bacillota (former Firmicutes) should be mentioned, which is one of the dominant phyla in cultivation soils, ranging from 2 to 20% [8]. This phylum includes the families Bacillaceae and Paenibacillaceae, and therefore the genera Bacillus, Priestia and Paenibacillus. Those genera are also major contributors among the plant-growth-promoting bacteria [9, 10]. Those microorganisms can exhibit various modes of promotion, depending on the species or strain, such as nitrogen transformation, phosphate solubilisation, potassium and zinc uptake, siderophore production, phytohormone production, and synthesis of antimicrobial lipopeptides, some of which exhibit biosurfactant activity as well [1]. At the same time, they possess the ability to produce spores, exhibit metabolic capabilities in both aerobic and relatively anaerobic conditions, and resist multiple environmental stresses, including drought, water stress, UV radiation, low nutrient content, and salinity [1, 8]. They can withstand prolonged storage as spores [3]. Their origin, physiological and biochemical characteristics make bacteria belonging to Bacillota ideal candidates for the production of biopreparations supporting ecological agricultural production.

Developing new biopreparations is a multi-stage and time-consuming task, primarily involving the selection of strains with desired phenotypic properties using classical (plate) culture methods. This approach has limited capacity to assess the full metabolite profile and pathogenic potential [11]. Therefore, an integrated functional genomich approach was applied here to develop more efficient and comprehensive framework for rapidly assessing the potential of environmental bacterial isolates for the production of agricultural and other biopreparations.

This study aimed to evaluate six bacterial strains isolated from post-maize cultivation soils as candidates for agricultural biopreparation development, using an integrated functional genomic and safety assessment framework. Specifically, we sought to characterise:

1. biotechnological potential, including plant-growth-promoting traits, carbohydrate-active enzymes relevant for lignocellulosic crop residue degradation, and secondary metabolite profiles.

2. safety profiles, including virulence factors and antibiotic resistance genes.

Functional trait variation among strains was contextualised against ANI-based genomic divergence to assess whether taxonomic relatedness predicts functional differentiation. Building on experimental validation of plant-growth-promoting activities [3], this study provides a comprehensive in silico characterisation of the functional potential of experimentally tested strains, informing their prioritisation for downstream applied research.

Methods

Bacterial strains and genome sequences

Six bacterial strains were isolated from soil samples collected from post-maize-cultivation fields, on a farm in the Lodz region, Poland. All samples were collected in the autumn season (October 2022). Detailed properties of soil samples, the isolation procedure, and optimal growth conditions of the analysed strains have been described by Rowińska et al. [3].

Genomic DNA was isolated using GeneMATRIX Gram Plus & Yeast DNA Purification Kit (EURX, Gdańsk, Poland). The isolated genomic DNA was sequenced on the DNBSEQ G400 platform (MGI, Shenzhen, China) using 150 bp paired-end reads (PE 150). NGS library preparation and sequencing were performed at BGI-TECH (Wuhan, China). To improve annotation accuracy and functional prediction, the strain zielonkawy was additionally sequenced using Oxford Nanopore technology (genXone S.A., Złotniki, Poland). Whole-genome sequences were deposited in the NCBI database under BioProject PRJNA1129008 – Innovative technology and organisation of biologically supported maize cultivation (Wspomagana Kukurydza) – with the following accession numbers: GCF_040513715.1 (Paenibacillus amylolyticus 5mez), GCF_040513995.1 (Priestia megaterium 7psych), GCF_040513685.1 (Bacillus velezensis Bac2), GCF_040513675.1 (Bacillus subtilis s.s. Bac3), GCF_040513645.1 (Bacillus licheniformis Bac8), and GCF_043099295.1 (Bacillus cereus s.s. zielonkawy). RefSeq annotations generated through the Prokaryotic Genome Annotation Pipeline (PGAP) were used for all functional analyses.

Plant-growth-promoting traits prediction

Plant-Growth-Promoting Traits (PGPT) were predicted using PLABase v.1.0.2 online platform (accessed July 2025) [12]. Protein sequences (.faa files) from NCBI RefSeq annotations were uploaded, and predictions were performed using confirmed hits from both BLASTP and HMMER searches. Results were processed in R v. 4.4.2 using vegan v.2.7–2 [13], pheatmap v. 1.0.13, pvclust v. 2.2–0, ggplot2 v. 3.5.2, dplyr v. 1.1.4, stringr v. 1.5.1, psych v. 2.5.3, and RColorBrewer v. 1.1–3.

Hierarchical clustering was performed using Euclidean distance with average linkage (cophenetic r = 0.987), selected for optimal dendrogram representation. Cluster robustness was assessed via multiscale bootstrap resampling (10,000 iterations). All dendrogram nodes showed strong bootstrap support (Approximately Unbiased p-values ≥ 95%), confirming the statistical significance of hierarchical relationships.

Principal Component Analysis (PCA) was performed to identify functional patterns across strains. Raw PGPT function counts were centred and scaled to unit variance (mean = 0, SD = 1) before analysis. Pre-analysis diagnostics (Bartlett's test for sphericity and the Kaiser–Meyer–Olkin measure) were computed. However, these metrics have limited statistical power at small sample sizes, where their diagnostic properties deteriorate rapidly [14]. Given the high-dimension, low-sample-size (HDLSS) configuration (p = 43, n = 6), PCA robustness was validated using Noise-Reduction Methodology (NRM), which provides mathematically consistent eigenvalue and eigenvector estimates under HDLSS asymptotics [15], implemented via custom R scripts from the HDLSS-Tools repository (https://github.com/Aoshima-Lab/HDLSS-Tools). Agreement between classical PCA and NRM-PCA was assessed using: (i) Pearson correlation of eigenvalues (r = 0.999), (ii) absolute cosine similarity of eigenvectors (to account for sign ambiguity) (1.0), (iii) Pearson correlation of PC scores (r = 1.0), and (iv) Procrustes analysis of the three-dimensional ordination space, reporting m² residual statistic (m² = 0.20, p = 0.001). Component retention was determined using the Kaiser criterion (> 1), cumulative variance explained (92.6%), and scree test plot analysis; additionally, NRM imposes a computational constraint of maximum n-2 components for consistent estimation, which, for our sample size (n = 6), limited validation to four components, concordant with Kaiser criterion results. Components with loadings ≥|0.7| were considered significant for interpretation. Three-dimensional visualization was generated using plotly v.4.11.0 [16].

Nitrogen metabolism pathway visualization was performed using KEGG Mapper–Reconstruct Pathway tool (https://www.genome.jp/kegg/mapper/reconstruct.html, accessed January 2026) [17, 18], and then custom-adapted. Genes detected in PGPT-Pred were mapped onto reference nitrogen transformation pathways. Pathway maps were generated with strain-specific gene presence indicated by split coloring, with each strain assigned a distinct color code.

Secondary metabolite biosynthetic gene cluster analysis

Secondary metabolite biosynthetic gene clusters (BGCs) were identified using the bacterial version of antiSMASH 8.0 web server (accessed October 2025) [19]. The antiSMASH pipeline integrates multiple detection algorithms for identifying diverse BGC types. Genome sequences were submitted in GenBank format, and analysis was performed using strict detection mode with "all on" parameters. BGC detection employed profile hidden Markov models from PFAM, TIGRFAMS, SMART, and BAGEL databases. The comparative analysis included the following databases: MIBiG 4.0, antiSMASH Database v4, and the TFBS analysis (CollecTF). Only high-confidence matches (i.e., a cluster similarity of minimum 75%) were considered for downstream interpretation.

Carbohydrate-active enzymes (CAZymes) analysis

CAZymes were identified using run_dbCAN v.4.2.0 (a standalone CLI version of the dbCAN3 annotation tool) [20] employing three prediction tools: HMMER v. 3.4. (E-value threshold < 1 × 10⁻¹⁵, coverage > 0.35), DIAMOND v. 2.1.12 (E-value threshold < 1 × 10⁻¹⁰²), and dbCAN-sub (E-value threshold < 1 × 10⁻¹⁵, coverage > 0.35), against the CAZyDB (downloaded July 2025). Only enzymes predicted by at least two methods were considered reliable and included in downstream analysis. Results were processed using R v. 4.4.2 with libraries matching those used for PGPT analysis. Hierarchical clustering validation was performed analogously (Euclidean distance, average linkage, 10,000 bootstrap iterations, cophenetic r = 0.984). All dendrogram nodes showed strong bootstrap support (Approximately Unbiased p-values ≥ 95%), confirming the statistical significance of hierarchical relationships.

Congruence of functional traits with ANI-based divergence

To assess whether functional variation reflects genomic relatedness, Average-Nucleotide-Identity-based divergence was calculated from genome-wide average nucleotide identity. ANI is the most well-known and widely used metric for identifying and classifying prokaryotic species, and is also used for species delimitation [21, 22]. Initial attempts using FastANI v.1.33 [21] failed for several inter-genus comparisons, returning no alignment due to the method's detection threshold. Therefore, OrthoANI with OAT.jar software was employed [23], which provides reliable calculations across a broader divergence range. ANI-based distances were computed as (100 – OrthoANI%). Genomic-relatedness structure was visualized using three-dimensional Principal Coordinates Analysis (PCoA). PCoA1-2 captured only 56.2% cumulative variance (30.7% + 25.5%), insufficient for adequate representation. PCoA3 contributed an additional 18.1% variance and was necessary to correctly distinguish Priestia megaterium 7psych from Bacillus cereus s.s. zielonkawy – taxa that appeared artificially proximate in 2D projection despite high ANI-based distance. The congruence was tested using Mantel tests (Pearson correlation, 999 permutations) correlating ANI-based distances with functional distances (Euclidean on annotation counts for PGPT, Jaccard for CAZyme presence/absence), and Procrustes analysis (999 permutations) comparing ANI-based ordination (3D PCoA) with functional ordination (3D PCA for PGPT, 3D PCoA for CAZyme). Per-genome Procrustes residuals were calculated to assess individual genome contributions to overall mismatch. Cluster coherence was evaluated by comparing within-cluster pairwise distances to between-cluster distances; lower within/between ratios indicate tighter clustering relative to non-cluster members. These ratios were calculated from the original distance matrices used in Mantel tests. All analyses used vegan v.2.7–2. Three-dimensional interactive visualizations of ANI-based structure, functional ordinations, and Procrustes superimpositions were generated using plotly v.4.11.0

Antimicrobial resistance and virulence factors analysis

Antimicrobial resistance genes (ARGs) and virulence factors were identified using ABRicate v. 1.0.1 (Seemann, https://github.com/tseemann/abricate) against the NCBI AMRFinderPlus [24], CARD [25], ResFinder [26], MEGARes 2.0 [27], and VFDB [28] databases, respectively (access date: July 2025). For strain B. cereus s.s. zielonkawy, a comprehensive Bacillus cereus sensu lato toxin profile was performed using BTyper3 v. 3.4.0 [29]. Results were processed and analysed using dplyr in R. PlasmidHunter v1.4 [30] was used to assign genome contigs, and therefore ARGs and virulence factors, to either chromosomes or plasmids.

Scientific names at all taxonomic ranks are set in italics, as postulated by Thines et al. [31].

Results

Plant-growth-promoting traits

General profile

Plant-growth-promoting traits (PGPT) were predicted using the PLABase platform with the PGPT-Pred tool. The PGPT analysis revealed functional diversity across strains, with 1,449–1,617 genes per strain involved in plant-growth promotion, representing 24.1%−35.9% of total bacterial genes (Table 1). When considering trait annotations (noting that individual genes may contribute to multiple traits), the analysis yielded 4,550–6,338 trait annotations per strain.

Table 1.

Overview of plant-growth-promoting trait (PGPT) annotations across six bacterial strains. Individual genes may contribute to multiple trait categories, resulting in higher annotation counts than gene counts

Genome	PGPT Annotations	PGPT Genes	PGPT Gene [%]	Total Genes
P. amylolyticus 5mez	5997	1531	24.1	6365
Pr. megaterium 7psych	6338	1617	26	6218
B. velezensis Bac2	4812	1574	33.8	4658
B. subtilis s.s. Bac3	4777	1584	35.9	4416
B. licheniformis Bac8	4550	1521	33.8	4494
B. cereus s.s. zielonkawy	5125	1449	25.7	5637

At PGPT prediction level 2, there are eight major functional categories. The lowest annotation counts were observed in the Putative Functions category, while the highest were observed in the Colonizing Plant System, Competitive Exclusion, and Stress Control/Biocontrol categories. Strain-specific patterns were observed: P. amylolyticus 5mez presented the highest annotation counts in Bio-Remediation (only 6 annotations above the second-highest strain), Colonizing Plant System (exceeding the next highest by over 200 annotations and the lowest by nearly 700), Competitive Exclusion, and Putative Functions. Pr. megaterium 7psych had the highest annotation counts in Bio-Fertilization and in the Phytohormone/Plant Signal Production and Stress Control/Biocontrol categories. For Plant Immune Response Stimulation, B. velezensis Bac2 showed the highest annotation count.

At the more detailed level (level 3) (Fig. 1) annotations displayed the highest counts in the Colonization-Plant Derived Substrate Usage and Neutralizing Abiotic Stress categories, with relatively high scores also observed for Ce-Bacterial Fitness. Other abundant categories included Heavy Metal Detoxification, Quorum Sensing Response/Biofilm Formation, Neutralizing Biotic Stress, Phosphate Solubilization, and Ce-cell Envelope Remodelling.

Fig. 1.

Plant-growth-promoting trait (PGPT) heatmap showing distribution of top 20 functional categories at prediction level 3 across six bacterial strains. Color intensity represents annotation counts. Hierarchical clustering of strains (Euclidean distance, average linkage) reveals functional grouping patterns

PCA was performed based on PGPT prediction level 3 data to identify functional patterns across strains. Pre-analysis diagnostics yielded Bartlett's test of sphericity p = 1.0 and Kaiser–Meyer–Olkin measure (0.50). These metrics have limited statistical power at small sample sizes, where their diagnostic properties deteriorate rapidly below n ≈ 20 [14]. Given the high-dimension, low-sample-size (HDLSS) configuration of our dataset (p = 43 functional categories, n = 6 strains), classical PCA was validated using Noise-Reduction Methodology (NRM), which provides mathematically consistent eigenvalue and eigenvector estimates under HDLSS asymptotics [15]. Agreement between classical PCA and NRM-PCA was assessed across all four mathematically derivable components (maximum n − 2 for NRM). Validation metrics demonstrated very high congruence: Pearson correlation of eigenvalues r = 0.999, absolute cosine similarity of eigenvectors = 1.0 (accounting for sign ambiguity), and Pearson correlation of PC scores r = 1.0, confirming reliability of classical PCA results despite limitations of traditional adequacy metrics designed for large-sample contexts. Additionally, Procrustes analysis with optimal rotation and uniform scaling was performed on the three-dimensional ordination space used for visualization (PC1–PC3), yielding m² = 0.20, p = 0.001, confirming concordance between the classical and noise-corrected score configurations.

Four components exceeded Kaiser criterion (eigenvalue > 1), concordant with the NRM computational constraint of n − 2 components (PC1: 57.2%, PC2: 23.7%, PC3: 11.7%, PC4: 5.4%). The first three accounted for 92.6% of the total variance and were retained for interpretation. PCA ordination revealed distinct functional clustering, with strain differentiation along the first three principal components (Fig. 2).

Fig. 2.

Principal component analysis of plant-growth-promoting traits based on PGPT categories at prediction level 3. PC1 (57.2% variance) separates Pr. megaterium 7psych from B. velezensis Bac2/B. subtilis s.s. Bac3/B. licheniformis Bac8 cluster, with P. amylolyticus 5mez intermediate. PC2 (23.7% variance) differentiates P. amylolyticus 5mez from other strains. PC3 (11.7% variance) separates B. cereus s.s. zielonkawy from remaining strains. The complete 3D interactive plot is available in Supplementary File S1

First principal component (PC1, 57.2% variance)

PC1 separated strains along a functional gradient, with strain Pr. megaterium 7psych positioned at the negative extreme, and strains B. velezensis Bac2, B. subtilis s.s. Bac3, and B. licheniformis Bac8 clustered at the opposite positive end, strain P. amylolyticus 5mez positioned centrally, and strain B. cereus s.s. zielonkawy trending toward positive values.

The negative end of PC1 (associated with strain Pr. megaterium 7psych) was characterized by strong loadings for multiple plant-growth-promoting traits. The strongest negative loadings (|loading|> 0.90) included Nitrogen Acquisition, Neutralizing Abiotic Stress, Colonization-Adaptation to Plant Immune System, Plant Vitamin Production, Plant Signal-Phospholipid Production, Phytohormone-Cytokinins/Derivative Production, Phytohormone-Abscisic Acid Degradation, Xenobiotics Biodegradation, Root Colonization, Phytohormone-Gamma-Aminobutyric Acid/GABA Production, Plant Signal-Branching Stimulation, and Plant Signal-Other Terpenoid/Derivative Production.

Additional functional categories with negative loadings on PC1 are presented in supplementary materials (Table S2-S3). At the positive end of PC1, only one functional category exceeded a loading of 0.70: Triggered Immunity. Additional positive loadings above 0.50 included Induction of Systemic Acquired Resistance/SAR and Plant Signal-Root/Shoot Stimulation.

Second principal component (PC2, 23.7% variance)

PC2 primarily differentiated strain P. amylolyticus 5mez (positive direction) from the remaining strains (Table S2-S3). Colonization-related functions dominated the positive end of PC2. Strong positive loadings (> 0.90) included Colonization-Plant Cell Wall/Membrane Degradation, Ce-Exopolysaccharide Production/EPS, Plant Signal-Ubiquinone/Coenzyme Q Production, and Colonization-Motility/Chemotaxis. The negative end of PC2 exhibited fewer strong loadings. Functions with loadings below −0.70 included Potassium Solubilization, Ce-Spore Production, and Phosphate Solubilization.

Third principal component (PC3, 11.7% variance)

PC3 provided secondary differentiation, separating B. cereus s.s. zielonkawy (positive direction) from other strains. The strongest PC3 loadings were Induction of Systemic Acquired Resistance (0.76) and Fluoride Detoxification (0.72), though several prominent PC3 functions overlapped with PC1-PC2 (Table S2), indicating that PC3 primarily captures residual variance.

PGPT annotations for all strains are detailed in Tables S4-S7, covering hierarchical levels 3–5 (functional categories) and level 6 (individual genes).

NPK and Iron acquisition

Nutrient acquisition capabilities varied among strains (Table 2). Phosphate solubilization demonstrated the highest overall annotation counts among nutrient acquisition categories, ranging from 157 (P. amylolyticus 5mez) to 263 (Pr. megaterium 7psych) (Table 2). This trait was predominantly mediated through organic acid metabolism, with pyruvic acid biosynthesis representing the most frequent mechanism, followed by succinic, propionic, acetic, lactic, and malic acids biosynthesis (Table S5). Pr. megaterium 7psych exhibited the highest capacity (221 annotations) in the organic acids subcategory, at the same time possessing the maximum number of annotations in each of the above-mentioned organic acids. Other acid metabolism mechanisms contributed 10–15 annotations per strain (Table 2). It included inorganic acids (carbonic, nitric, and sulfuric acid metabolism), with the highest number of annotations in B. subtilis s.s. Bac3 genome (Table S6), as well as the phosphonate degradation category, with the highest score in Pr. megaterium 7psych, which also included the phnX gene, present only in Pr. megaterium 7psych and B. cereus s.s. zielonkawy (Table S7).

Table 2.

Distribution of nutrient acquisition trait annotations at prediction level 3 and level 4

	P. amylolyticus 5mez	Pr. megaterium 7psych	B. velezensis Bac2	B. subtilis s.s. Bac3	B. licheniformis Bac8	B. cereus s.s. zielonkawy
Nitrogen acquisition, including	103	126	74	74	68	78
atmoshpheric nitrogen fixation	5	8	7	7	6	6
denitrification|nitrate usage	26	15	16	17	18	13
ammonium assimilation usage	28	43	27	28	21	27
Phosphate solubilization, including	157	263	186	189	178	198
organic acid metabolism	110	221	141	150	142	153
other acid metabolism	12	14	12	13	10	15
phosphatase activity	3	5	5	6	5	4
Potassium solubilization, including	119	228	150	164	153	165
organic acid metabolism	103	210	136	146	136	147
Iron acquisition	147	119	97	109	101	122

Phosphatase activity was represented by 3–6 annotations across strains, comprising alkaline phosphatases, an uncharacterised phosphatase, exopolyphosphatase, and phytase (Table S6). Alkaline phosphatases included phoA – present in all strains, and phoD – present in strains B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8, and Pr. megaterium 7psych (Table S7). Uncharacterised phosphatase phoE is absent only in P. amylolyticus 5mez. The ppx gene (encoding exopolyphosphatase) is present in strains B. cereus s.s. zielonkawy, P. amylolyticus 5mez and Pr. megaterium 7psych, whereas phytase, encoded by phyA, is present only in B. velezensis Bac2, B. subtilis s.s. Bac3, and B. licheniformis Bac8.

For nitrogen acquisition (Table 2), annotation counts ranged from 68 (B. licheniformis Bac8) to 126 (Pr. megaterium 7psych). Atmospheric nitrogen fixation was relatively uniform across strains, with all strains possessing the nitrogenase biosynthesis pathway level 5 annotation encoded by three accessory genes: nifM, nifS, and nifU (Table S7). These genes were present in 1–4 copies depending on the specific gene and strain. However, no core nitrogen fixation genes have been detected in any genome. Ammonium assimilation usage (Table 2) demonstrated the widest variation, with Pr. megaterium 7psych recording the highest count, over twice as much as B. licheniformis Bac8. Denitrification/nitrate usage ranged from 13 (B. cereus s.s. zielonkawy) to 26 (P. amylolyticus 5mez), with the most abundant level 5 annotation being nitrate reduction, possessing genes belonging to the clusters nar, nas, nfr, nir, nor, nos, and nrf, such as narGHI/V, nirBD, nirK. (Table S7). Ammonium assimilation usage contained, among others, crucial genes – amtB, glnA, and gdhA. The first two were present in all studied genomes, but gdhA was only in 5mez, Bac8, and zielonkawy (Table S7). To visualize the putative influence on nitrogen cycling, KEGG Mapper Reconstruct was employed to map detected genes onto nitrogen transformation pathways (Fig. 3). In the case of nitrification and denitrification (Fig. 3a, b) four Bacillus genomes contained the narGHI/V cluster, scientific integrity nitrate NO₃ to nitrite NO₂, the first step in denitrification, as well as the first step in DNRA (dissimilatory nitrate reduction to ammonium) (Fig. 3c). In the case of denitrification, P. amylolyticus 5mez is capable of reducing nitrite to nitric oxide (NO) (nir Kitalics). Bac8 possesses a norB gene, which is involved in the conversion of nitric oxide to nitrous oxide (N₂O). However, it lacks norC, so the enzyme would be incomplete and nonfunctional. When it comes to DNRA, all six strains contain the nirBD cluster, which is responsible for transforming nitrite to ammonium (NH₄⁺).

Fig. 3.

KEGG pathway reconstruction showing nitrogen transformation capabilities with split coloring indicating strain-specific gene presence. Strain color coding (left to right): 5mez (pastel green), 7psych (pastel pink), Bac2 (pastel blue), Bac3 (light pastel green), Bac8 (light pastel pink), zielonkawy (light pastel blue). Custom-added colored circles represent nitrogen compounds: NO₃⁻ (blue), NO₂⁻ (orange), NO (purple), N₂O (black), N₂ (gray), NH₂OH (yellow), N₂H₄ (white), NH₄⁺ (green). White boxes indicate genes absent from all tested strains. Performed using Kegg Mapper - Reconstruct Pathway tool [17, 18]

Potassium solubilization followed a similar pattern to phosphorus, with annotation counts ranging from 119 (P. amylolyticus 5mez) to 228 (Pr. megaterium 7psych (Table 2). Organic acid metabolism, primarily through pyruvic acid biosynthesis, represented the principal mechanism (Table S6). Iron acquisition capabilities ranged from 97 (B. velezensis Bac2) to 147 (P. amylolyticus 5mez) annotations. Siderophore production represented the dominant mechanism, with bacillibactin metabolism being the predominant pathway across all strains (Table S4-S6). Bacillibactin transport systems ranked second. Hemophore-mediated iron acquisition was uniformly distributed.

Stress neutralisation

Stress tolerance mechanisms showed diversity across strains (Table 3; Table S5). For neutralising abiotic stress, total annotation counts ranged from 566 (B. licheniformis Bac8) to 852 (Pr. megaterium 7psych). Salinity stress tolerance was the most abundant category, with all strains achieving high scores (215–322 annotations). Nitrosative/oxidative stress and ROS scavenging mechanisms represented the second most abundant subcategory (166–257 annotations). Osmotic stress tolerance ranged from 78 (B. licheniformis Bac8) to 131 (Pr. megaterium 7psych), while high temperature tolerance ranged from 31 (B. cereus s.s. zielonkawy) to 41 (B. subtilis s.s. Bac3), and low temperature tolerance from 13 (B. velezensis Bac2, B. subtilis s.s. Bac3) to 29 (Pr. megaterium 7psych). Herbicidal stress resistance varied from 27 (P. amylolyticus 5mez) to 58 (Pr. megaterium 7psych), and acidic stress tolerance was relatively variable, ranging from 8 to 16 annotations.

Table 3.

Distribution of stress neutralization trait annotations at prediction level 4 and level 5

	P. amylolyticus 5mez	Pr. megaterium 7psych	B. velezensis Bac2	B. subtilis s.s. Bac3	B. licheniformis Bac8	B. cereus s.s. zielonkawy
Neutralizing abiotic stress	663	852	610	609	566	663
Neutralizing biotic stress, including	262	297	266	215	210	273
bactericidal compounds|antibiotics	126	127	126	84	96	134
fungicidal compounds|antibiotics	44	22	32	25	26	36
insecticidal compounds	1	9	4	2	1	4
antiprotozoan|antiprotistal activity	2	2	2	2	2	0
biotic stress resistance-volatiles	57	105	70	71	63	66
biotic stress resistance-phenazine derivates	14	21	12	14	13	16
biotic stress resistance-phytotoxin degradation	0	0	0	0	1	1
Heavy metal detoxification	347	330	234	238	228	290

Neutralising biotic stress capabilities ranged from 210 (B. licheniformis Bac8) to 297 (Pr. megaterium 7psych) annotations (Table 3). Bactericidal compound/antibiotic resistance mechanisms were most abundant in this category, with three major pathways at level 5 showing the highest representation: bacteriocins/lantibiotics/nisin metabolism (21–48 annotations), prodigiosin metabolism (14–22 annotations), and spermidine/putrescine metabolism (7–33 annotations) (Table S6). Strain-specific patterns were observed, with B. subtilis s.s. Bac3 having the lowest overall annotation count, while B. cereus s.s. zielonkawy – the highest.

Fungicidal compound resistance was more variable (22–44 annotations) (Table 3). General chitinolytic activities represented the most abundant mechanism at levels 5 and 6, with P. amylolyticus 5mez showing the highest capacity (Table S6). Other significant pathways included fungal glycogen degradation, toxoflavin metabolism, and lipopeptide antibiotics: fengycin/plipastatin metabolism was detected exclusively in B. velezensis Bac2 and B. subtilis s.s. Bac3, while bacillimycin D/iturin A/mycosubtilin metabolism was detected only in B. cereus s.s. zielonkawy, B. velezensis Bac2 and P. amylolyticus 5mez. Fusarcidin metabolism was detected only in B. velezensis Bac2. However, lipopeptide biosynthetic gene cluster analysis revealed variable completeness (Table S7). Complete gene sets were present for fengycin/plipastatin (ppsABCDE/fenABCDE) in B. velezensis Bac2 and B. subtilis s.s. Bac3. In contrast, iturin, fusaricidin, and pelgipeptin clustersshowed partial or fragmented distribution. For the iturin cluster (ituABCD), ituA was present in P. amylolyticus 5mez and B. velezensis Bac2; ituB in P. amylolyticus 5mez, B. velezensis Bac2, and B. cereus s.s. zielonkawy; ituC only in B. velezensis Bac2; and ituD was absent from all genomes. The fusaricidin gene fusA was detected only in B. velezensis Bac2. For pelgipeptin, plpE was present in P. amylolyticus 5mez, B. velezensis Bac2, and B. cereus s.s. zielonkawy; plpFGH only in P. amylolyticus 5mez; and plpD was absent from all strains.

Insecticidal compound productionshowed relatively low but variable annotation counts (1–9 annotations) (Table 3). At level 5, this category was represented by two distinct mechanisms. GABA-mediated activity (1–9 annotations), an indirect mechanism, dominated insecticidal capacity (Table S6). Direct insecticidal toxin production was detected exclusively in B. velezensis Bac2 (1 annotation), represented by the TccC insecticidal toxin complex gene (Table S7). Antiprotozoal/antiprotistal activity was detected at low levels in five of six strains, B. cereus s.s. zielonkawy had no annotations in this category (Table 3). This activity was mediated through alkylresorcinol/alkylpyrone biosynthesis at level 5 (Table S6).

Biotic stress resistance mediated by volatile compounds showed variation (57–105 annotations), with Pr. megaterium 7psych demonstrating a noticeable difference from the rest (Table 3). The most frequent subcategories were volatile-related fatty-acid metabolism and 2,3-butanediol biosynthesis (Table S6). Phenazine derivative-mediated resistance ranged from 12 (B. velezensis Bac2) to 21 (Pr. megaterium 7psych). Phytotoxin degradation capacity was detected exclusively in B. licheniformis Bac8 and B. cereus s.s. zielonkawy (1 annotation each) (Table S7).

Heavy metal detoxification capacity varied substantially, with annotation counts ranging from 228 (B. licheniformis Bac8) to 347 (P. amylolyticus 5mez) (Table S6). Among the 17 metal resistance systems detected, iron resistance was the most prevalent. The copper and nickel resistances had the second- and third-highest annotation counts, followed by arsenic, antimony, and zinc resistances. Other heavy metal resistance systems included bismuth, cadmium, chromate, cobalt, lead, manganese, selenium, and tellurium. Tungstate resistance was absent in 5mez. Mercury and gold resistance were detected only in P. amylolyticus 5mez and Pr. megaterium 7psych.

Plant-derived substrate usage

Colonization through plant-derived substrate usage reached the highest overall annotation counts among all analyzed functional categories, ranging from 937 (B. cereus s.s. zielonkawy) to 1,444 (P. amylolyticus 5mez) (Table 4). Within this, plant-derived complex sugar utilization demonstrated strain-specific variation, with annotation counts ranging from 16 (B. cereus s.s. zielonkawy) to 107 (P. amylolyticus 5mez). P. amylolyticus 5mez exceeded the second-highest strain (Pr. megaterium 7psych) by more than twofold. At the more detailed level, cellulose/hemicellulose degradation capacity varied from 6 (B. cereus s.s. zielonkawy) to 37 (P. amylolyticus 5mez), representing the most abundant mechanism within complex sugar utilisation. Glucan/glycan breakdown ranged from 0 (B. cereus s.s. zielonkawy) to 26 (P. amylolyticus 5mez), while pectin degradation showed variation from 0 (B. cereus s.s. zielonkawy) to 11 (P. amylolyticus 5mez). Starch/glycogen degradation and fructan/mannan breakdown were at lower absolute counts.

Table 4.

Distribution of plant-derived substrate utilization trait annotations at prediction levels 3, 4, and 5

	P. amylolyticus 5mez	Pr. megaterium 7psych	B. velezensis Bac2	B. subtilis s.s. Bac3	B. licheniformis Bac8	B. cereus s.s. zielonkawy
Colonization – Plant-derived substrate usage, including	1444	1326	991	1055	1002	937
Plant-derived complex sugar utilization, including	107	43	41	33	40	16
cellulose|hemicellulose degradation	37	13	17	9	9	6
glucan|glycan breakdown	26	2	7	2	2	0
pectin degradation	11	2	1	5	7	0
starch|glycogen degradation	7	6	1	3	6	4
fructan|mannan breakdown	6	1	0	1	1	0
Plant-derived aromatic|phenolic compound utilization, including	39	57	24	25	27	24
lignin utilization	1	1	1	2	2	1
beta-ketoadipate pathway	6	12	4	3	1	4
protocatechuic acid degradation	6	12	4	3	1	4
HCA degradation|vanillin intermediate	0	1	1	1	2	2

Plant-derived aromatic and phenolic compound utilization presented annotation counts ranging from 24 (B. velezensis Bac2, B. cereus s.s. zielonkawy) to 57 (Pr. megaterium 7psych). Lignin utilization genes were detected at low levels across all strains (1–2 annotations). The beta-ketoadipate pathway, representing a central route for aromatic compound catabolism, ranged from 1 (B. licheniformis Bac8) to 12 (Pr. megaterium 7psych), with identical annotation counts observed for protocatechuic acid degradation. HCA degradation/vanillin intermediate processing was detected at low levels (0–2 annotations) across strains, with P. amylolyticus 5mez being the only strain lacking this pathway.

Congruence between ANI-based divergence and PGP traits

To assess congruence between ANI-based divergence and PGPT level 3 functional variation, ANI-based distances were compared with PGPT functional distances using Mantel and Procrustes analyses. The ANI-based structure (Supplementary File S2) showed separation between genera and different sensu lato groups, with the three Bacillus sensu lato strains (B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8) clustering together, while P. amylolyticus 5mez, Pr. megaterium 7psych, and B. cereus s.s. zielonkawy (belonging to the B. cereus sensu lato group) occupied distinct positions reflecting their greater taxonomic divergence. The pairwise OrthoANI matrix is reported in Table S8.

The Mantel test revealed a significant correlation between ANI-based and functional distance matrices (r = 0.703, p = 0.019), indicating a significant association between genome-wide divergence and PGPT functional variation (Table 5a). Procrustes analysis, which directly superimposes the two three-dimensional ordinations after optimal rotation and scaling, confirmed significant congruence (m² = 0.490, p = 0.036), with 51% of spatial structure preserved between ANI-based (Supplementary File S2) and functional (Fig. 2; Supplementary File S1) configurations (Fig. 4; Supplementary File S3; Table 5a).

Table 5.

Phylogenetic congruence of PGPT profiles: (a) global tests, (b) Bac2/Bac3/Bac8 cluster coherence

a)
test	statistic	p-value
Mantel	r = 0.703	0.019
Procrustes (3D)	m2 = 0.490	0.036

b)
metric	phylogenetic	PGPT functional
within/between ratio	0.769	0.230

Fig. 4.

Procrustes superimposition of phylogenetic (ANI-based) and PGPT functional ordinations. Blue points represent phylogenetic positions; green points represent PGPT functional positions after optimal rotation and scaling (m.² = 0.490, p = 0.036, spatial preservation = 51%). Connecting lines indicate per-genome residuals. The complete 3D interactive plot is available in Supplementary File S3

Per-genome Procrustes residuals quantified individual contributions to overall mismatch (Table S9a). Strains P. amylolyticus 5mez, Pr. megaterium 7psych, and B. cereus s.s. zielonkawy showed low residuals, indicating close correspondence between their ANI-based and functional positions. Strain B. subtilis s.s. Bac3 showed intermediate residuals. In contrast, B. velezensis Bac2 and B. licheniformis Bac8 exhibited elevated residuals, suggesting their functional positions deviated from expectations based on ANI-based divergence alone. To evaluate whether these elevated residuals indicated inconsistent ANI-functional association within the Bac2/Bac3/Bac8 cluster, or merely differential scaling, within-cluster and between-cluster distances were compared (Table S9b). Within-cluster distances (mean pairwise distance among Bac2, Bac3, Bac8) were smaller than between-cluster distances (mean distance from cluster members to remaining strains) in both spaces. The within/between ratio was 0.77 for ANI-based and 0.23 for functional distances (Table 5b). Both ratios below 1.0 confirm that these strains remain tightly clustered in both ordinations. Despite elevated residuals, Bac2 and Bac8 maintained tight clustering with Bac3 in both ordinations.

Secondary metabolites

Secondary metabolite biosynthetic potential was evaluated using antiSMASH analysis in strict mode across all six bacterial strains. The analysis displayed variation in biosynthetic gene cluster (BGC) content, with total BGC counts ranging from 7 (Pr. megaterium 7psych) to 15 (P. amylolyticus 5mez and B. velezensis Bac2) per genome (Table 6).

Table 6.

Distribution of BGCs detected by antiSMASH (strict mode) Cluster types: NRPS (Non-Ribosomal Peptide Synthetase), T3PKS (Type III Polyketide Synthase), transAT-PKS (trans-Acyltransferase Polyketide Synthase), PKS (Polyketide Synthase), CDPS (Cyclodipeptide Synthase), RiPP (Ribosomally synthesized and Post-translationally modified Peptide), NI-siderophore (Non-ribosomal peptide synthetase-independent siderophore). Hybrid clusters combine multiple biosynthetic pathways

Strain	Total BGCs	Cluster Types	Known Compounds (high similarity)	Percentage of identified metabolites
P. amylolyticus 5mez	15	2 × NRPS, 2 × T3PKS, 3 × hybrid, 1 × terpene, 1 × NI-siderophore, 1 × lassopeptide, 1 × lanthipeptide class IV, 1 × proteusine, 1 × opine-like metallophore, 1 × triceptide, 1 × cyclic-lactone-autoinducer	bacillopaline	6.7%
Pr. megaterium 7psych	7	3 × terpene, 1 × T3PKS, 1 × NI-siderophore, 1 × lassopeptide, 1 × phosphonate		0%
B. velezensis Bac2	15	1 × NRPS, 1 × T3PKS, 2 × transAT-PKS, 2 × transAT-PKS-like, 1 × PKS-like, 3 × hybrid, 2 × terpene, 1 × terpene-precursor, 1 × azole-containing RiPP, 1 × other	fengycin, bacillaene, macrolactin H, surfactin, bacillibactin, bacilysin	40%
B. subtilis s.s. Bac3	11	1 × NRPS, 1 × T3PKS, 4 × hybrid, 2 × terpene, 1 × CDPS, 1 × sactipeptide, 1 × other	bacillaene, fengycin, bacillibactin, pulcherriminic acid, subtilosin A, bacilysin, sporulation killing factor, surfactin	72.7%
B. licheniformis Bac8	12	3 × NRPS, 1 × T3PKS, 1 × terpene, 1 × hybrid, 1 × NI-siderophore, 1 × CDPS, 1 × betalactone, 1 × lanthipeptide class II, 1 × lassopeptide, 1 × azole-containing RiPPV	lichenicidin VK21 A1/lichenicidin VK21 A2, bacillibactin/bacillibactin E/bacillibactin F	16.7%
B. cereus s.s. zielonkawy	10	2 × NRPS, 2 × hybrid, 1 × terpene, 1 × NI-siderophore, 1 × betalactone, 1 × lanthipeptide class I, 1 × sactipeptide, 1 × azole-containing RiPP	Petrobactin, bacillibactin, zwittermicin A	30%

Terpene biosynthesis clusters were universally present across all six strains, representing the only cluster type detected in every genome, with 1–3 clusters per strain. Type III polyketide synthase (T3PKS) clusters were detected in five strains (absent only in B. cereus s.s. zielonkawy), as well as NRPS clusters (but absent in Pr. megaterium 7psych). NI-siderophore clusters were detected in four strains (P. amylolyticus 5mez, Pr. megaterium 7psych, B. licheniformis Bac8, and B. cereus s.s. zielonkawy). Hybrid clusters, combining multiple biosynthetic pathways, were present in five strains (absent in Pr. megaterium 7psych) and were highly abundant in B. subtilis s.s. Bac3 (4 clusters), followed by P. amylolyticus 5mez and B. velezensis Bac2 (3 clusters each).

Less commonly distributed cluster types included lassopeptides (detected in P. amylolyticus 5mez, Pr. megaterium 7psych, and B. licheniformis Bac8), CDPS clusters (B. subtilis s.s. Bac3 and B. licheniformis Bac8 only), and betalactone clusters (B. licheniformis Bac8 and B. cereus s.s. zielonkawy only) (Table 6). Lanthipeptide clusters showed strain-specific class distribution: class IV in P. amylolyticus 5mez, class II in B. licheniformis Bac8, and class I in B. cereus s.s. zielonkawy. Azole-containing RiPP clusters were restricted to B. velezensis Bac2, B. licheniformis Bac8, and B. cereus s.s. zielonkawy, while sactipeptide clusters were present only in B. subtilis s.s. Bac3 and B. cereus s.s. zielonkawy. TransAT-PKS, transAT-PKS-like, and PKS-like clusters (5 clusters total) were exclusively detected in B. velezensis Bac2. Strain Pr. megaterium 7psych was the only one to harbour a phosphonate cluster, while P. amylolyticus 5mez possessed several unique cluster types: proteusine, opine-like metallophore, triceptide, cyclic-lactone-autoinducer, and lanthipeptide class IV.

High-confidence matches to known secondary metabolites were obtained for a limited subset of the detected BGCs. Across all strains, only 23 compound matches were identified from a total of 70 BGCs, indicating that the majority of biosynthetic clusters (67%) showed no high similarity to characterised compounds in the database. The identification rate varied between strains: B. subtilis s.s. Bac3 exhibited the highest proportion with 8 matches from 11 BGCs (72.7%), followed by B. velezensis Bac2 with 6 matches from 15 BGCs (40%), B. licheniformis Bac8 with 2 out of 12 BGCs (16.7%), while P. amylolyticus 5mez yielded only 1 match from 15 BGCs (6.7%), and Pr. megaterium 7psych showed no matches despite possessing 7 BGCs (Table 6).

Bacillibactin was the only compound identified across multiple strains, with high similarity matches in B. velezensis Bac2 and B. subtilis s.s. Bac3, B. licheniformis Bac8, and B. cereus s.s. zielonkawy. B. licheniformis Bac8 additionally harboured bacillibactin structural variants (bacillibactin E and F). Other siderophore systems included petrobactin in B. cereus s.s. zielonkawy.

Several compounds were shared between B. velezensis Bac2 and B. subtilis s.s. Bac3, including the lipopeptide antibiotics fengycin and surfactin, the polyketide bacillaene, and the dipeptide bacilysin. B. velezensis Bac2 uniquely possessed macrolactin H, while B. subtilis s.s. Bac3 harboured three strain-specific compounds: pulcherriminic acid, subtilosin A, and sporulation killing factor.

The remaining strains showed highly restricted compound identification. Strain P. amylolyticus 5mez matched only bacillopaline despite possessing the highest BGC count (15 clusters). B. licheniformis Bac8 yielded lichenicidin VK21 A1/A2 beyond the shared bacillibactin variants. B. cereus s.s. zielonkawy showed matches to zwittermicin A in addition to the shared petrobactin and bacillibactin. Strain Pr. megaterium 7psych produced no high-confidence matches to known compounds.

Carbohydrate-active enzymes

General profile

The carbohydrate metabolism potential of the bacterial strains was comprehensively investigated using the dbCAN3 annotation pipeline. P. amylolyticus 5mez demonstrated the highest CAZyme representation both in percentage (4.70% of total genes) and absolute numbers (299 genes, 1,064 domains), exceeding all other strains (Table 7). The remaining strains showed more moderate CAZyme content, ranging from 116 to 149 genes (2.04–3.32% of total genes) and 369 to 461 domains. Throughout this study, "CAZyme genes" refers to individual protein-coding sequences, each of which may contain multiple catalytic or ancillary domains (e.g., 299 genes yielded 1,064 domains in P. amylolyticus 5mez); domains are classified into families based on sequence similarity, and families are grouped into functional classes (GH, GT, PL, CE, CBM, AA, SLH).

Table 7.

Carbohydrate-active enzyme (CAZyme) distribution across six functional classes. Total CAZyme gene counts, domain counts, and genome percentages are shown. Domains classified as: GH (Glycoside Hydrolases), GT (Glycosyltransferases), PL (Polysaccharide Lyases), CE (Carbohydrate Esterases), CBM (Carbohydrate-Binding Modules), AA (Auxiliary Activities), and SLH (Surface Layer Homology domains)

Genome	CAZyme Genes	CAZyme Domains	Total Genes	CAZyme Genes [%]	GH	GT	PL	CE	CBM	AA	SLH
P. amylolyticus 5mez	299	1064	6365	4.70	563	156	37	84	220	2	2
Pr. megaterium 7psych	127	397	6218	2.04	143	106	3	63	74	8	0
B. velezensis Bac2	140	418	4658	3.01	178	106	13	43	67	11	0
B. subtilis s.s. Bac3	144	443	4416	3.26	166	123	21	50	75	8	0
B. licheniformis Bac8	149	461	4494	3.32	187	117	24	51	73	9	0
B. cereus s.s. zielonkawy	116	369	5637	2.06	101	161	6	43	41	17	0

P. amylolyticus 5mez also prevailed at the functional domain level, showing numerous annotations (563 domains, over half of all its CAZyme domains) in the GH class – Glycoside Hydrolases. Notably, every strain exhibited the highest abundance in GH domains, followed by CBM (Carbohydrate-Binding Modules) domains in P. amylolyticus 5mez or GT (Glycosyltransferases) domains in all other strains. Although the number of different CAZyme families varied only modestly between the strains (for instance, P. amylolyticus 5mez possesses nine cellulose-related GH/CBM families, whereas B. licheniformis Bac8 encodes eight), the total number of domains per family highly differed – P. amylolyticus 5mez harbours 93 cellulose GH domains compared to only 33 in B. licheniformis Bac8.

Analysis by substrate and families

Substrate-level analysis revealed that P. amylolyticus 5mez exhibited numerous annotations in Cellulose GH (93), Xylan GH (110), Xylan CE (53), Xylan CBM (50), Pectin CBM (41), and Starch GH (39) families, indicating a high potential for cellulose and hemicellulose degradation (Fig. 5). Pr. megaterium 7psych recorded the highest performance in Xylan CE (49 domains), similarly to B. cereus s.s. zielonkawy (35 domains). B. velezensis Bac2 and B. licheniformis Bac8 demonstrated capabilities in Cellulose GH (38 and 33 domains, respectively), though lower than P. amylolyticus 5mez. The matching of individual families to substrates is shown in Table S10.

Fig. 5.

Biomass component processing capacity by CAZyme class. Heatmap shows domain counts per strain organized by substrate specificity (Cellulose, Xylan, Starch, Pectin, Mannan, Chitin, and associated substrates) and CAZyme class (GH, PL, CE, AA, CBM). Hierarchical clustering of strains (Euclidean distance, average linkage) reveals functional grouping patterns

Due to the specific composition of maize crop residues (cellulose, hemicelluloses, lignins), detailed analysis was performed to examine the occurrence of enzyme families directly involved in the degradation of these major lignocellulosic components.

For cellulose degradation (Table 8), the highest family diversity occurred in P. amylolyticus 5mez (9 families), with B. licheniformis Bac8 having only slightly fewer (8 families). Only P. amylolyticus 5mez possessed the GH6 family and CBM4 family, while B. licheniformis Bac8 was the sole strain harboring the GH12 family. No strains possessed the cellulase families GH7, GH44, GH45, or the lytic polysaccharide monooxygenase family AA9. However, the difference in domain copy number is more noticeable – P. amylolyticus 5mez encodes 93 GH domains related to cellulose, whereas B. licheniformis Bac8 encodes only 33.

Table 8.

Presence/absence of GH, CBM, and AA gene families involved in cellulose degradation. + indicates presence; - indicates absence. Total – total of different domain types

Strain	GH1	GH3	GH6	GH7	GH9	GH12	GH44	GH45	GH48	GH5	GH8	CBM1	CBM2	CBM3	CBM4	AA9	AA10	Total
P. amylolyticus 5mez	+	+	+	-	+	-	-	-	+	+	+	-	-	+	+	-	-	9
Pr. megaterium 7psych	+	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	1
B. velezensis Bac2	+	+	-	-	-	-	-	-	-	+	-	-	-	+	-	-	+	5
B. subtilis s.s. Bac3	+	+	-	-	-	-	-	-	-	+	-	-	-	+	-	-	-	4
B. licheniformis Bac8	+	+	-	-	+	+	-	-	+	+	-	-	-	+	-	-	+	8
B. cereus s.s. zielonkawy	+	-	-	-	-	-	-	-	-	+	+	-	+	-	-	-	+	5

The GH1 family was present in all strains. Five strains possessed the GH4 family, while four strains harboured both the GH3 and CBM3 families.

For hemicellulose degradation (Table 9), P. amylolyticus 5mez again demonstrated the highest family diversity, including GH10, GH11, GH30, GH43 (xylanases), CE1, CE2, CE4 (acetylxylan esterases), GH26 and GH76 (mannanases), and multi-substrate GHs (GH5, GH8, GH16), including GH16, involved also in mixed-linkage β-glucans hydrolysis. B. velezensis Bac2 was the closest competitor, possessing a similar range of families but far fewer domain copies (e.g., 38 vs 110 xylan GH domains), illustrating the difference between family diversity and overall gene content. P. amylolyticus 5mez lacked only three families (GH17, CE3, AA14) that were absent across all strains. The CE4 family was present in all strains. GH5 and GH43 families were present in five strains, while four strains possessed both GH43 and CBM6 families.

Table 9.

Presence/absence of GH, CE, CBM, and AA gene families involved in hemicellulose degradation. + indicates presence; - indicates absence. Total – total of different domain types

Strain	GH10	GH11	GH17	GH26	GH30	GH43	GH76	GH5	GH8	GH16	CE1	CE2	CE3	CE4	CBM6	CBM9	CBM22	CBM35	AA14	Total
P. amylolyticus 5mez	+	+	-	+	+	+	-	+	+	+	+	+	-	+	+	+	+	+	-	15
Pr. megaterium 7psych	-	-	-	-	-	-	-	-	-	-	+	-	-	+	-	-	-	-	-	2
B. velezensis Bac2	+	+	-	+	+	+	-	+	-	+	-	+	-	+	+	+	+	-	-	12
B. subtilis s.s. Bac3	-	+	-	+	+	+	-	+	-	+	-	-	-	+	+	-	-	-	-	8
B. licheniformis Bac8	-	-	-	+	-	+	-	+	-	-	-	-	-	+	+	-	-	-	-	5
B. cereus s.s. zielonkawy	-	-	-	-	-	-	-	+	+	-	-	-	-	+	-	-	-	-	-	3

For lignin-related compound processing (Table 10), P. amylolyticus 5mez was the only strain lacking lignin-degrading enzymes. All other strains possessed exactly 2 auxiliary activity families each: Pr. megaterium 7psych, B. velezensis Bac2, B. subtilis s.s. Bac3 and B. licheniformis Bac8 harboured both AA4 and AA6, while B. cereus s.s. zielonkawy possessed AA1 and AA4. The families AA2, AA3, and AA5 were absent across all strains.

Table 10.

Presence/absence of AA gene families involved in lignin modification. + indicates presence; - indicates absence. Total – total of different domain types

Strain	AA1	AA2	AA3	AA4	AA5	AA6	Total
P. amylolyticus 5mez	-	-	-	-	-	-	0
Pr. megaterium 7psych	-	-	-	+	-	+	2
B. velezensis Bac2	-	-	-	+	-	+	2
B. subtilis s.s. Bac3	-	-	-	+	-	+	2
B. licheniformis Bac8	-	-	-	+	-	+	2
B. cereus s.s. zielonkawy	+	-	-	+	-	-	2

Complete CAZyme annotation data, including gene identifiers, EC numbers, and multi-tool validation scores, are provided in Table S11. The absolute abundance of all enzyme families is presented in Table S12.

Congruence between ANI-based divergence and CAZyme profiles

To assess congruence between ANI-based divergence and CAZyme functional variation, ANI-based distances were compared with CAZyme distances (Jaccard dissimilarity on family-level presence/absence; Table S13) using Mantel and Procrustes analyses. The CAZyme ordination (Supplementary File S4) showed clustering patterns similar to those of PGPT profiles, with Bac2/Bac3/Bac8 grouping together while P. amylolyticus 5mez, Pr. megaterium 7psych, and B. cereus s.s. zielonkawy occupied distinct positions.

The Mantel test revealed a significant correlation between the ANI-based and CAZyme distance matrices (r = 0.800, p = 0.018), indicating a strong association between genome-wide divergence and CAZyme functional variation (Table 11a). This correlation was stronger than observed for PGPT profiles (r = 0.703; Table 6a). Procrustes analysis confirmed significant congruence (m² = 0.783, p = 0.004), though with lower topological preservation (22%) compared to PGPT (51%) (Fig. 6; Supplementary File S5; Table 11a).

Table 11.

Phylogenetic congruence of CAZyme profiles: (a) global tests, (b) Bac2/Bac3/Bac8 cluster coherence

a)
test	statistic	p-value
Mantel	r = 0.800	0.018
Procrustes (3D)	m2 = 0.783	0.004

b)
metric	phylogenetic	CAZyme functional
within/between ratio	0.769	0.603

Fig. 6.

Procrustes superimposition of phylogenetic (ANI-based) and CAZyme functional ordinations. Blue points represent phylogenetic positions; orange points represent CAZyme functional positions after optimal rotation and scaling (m.² = 0.783, p = 0.004, spatial preservation = 22%). Connecting lines indicate per-genome residuals. The complete 3D interactive plot is available in Supplementary File S5

Per-genome Procrustes residuals quantified individual contributions to overall mismatch (Table S14a). Strains P. amylolyticus 5mez and B. cereus s.s. zielonkawy showed low residuals. Strains Pr. megaterium 7psych and B. subtilis s.s. Bac3 showed intermediate residuals. In contrast, B. licheniformis Bac8 and B. velezensis Bac2 exhibited elevated residuals, consistent with the pattern observed for PGPT profiles.

Within-cluster and between-cluster distance comparison (Table S14b) confirmed cluster coherence. The within/between ratio was 0.77 for ANI-based and 0.60 for CAZyme distances (Table 11b). Both ratios below 1.0 confirm tight clustering. Despite elevated residuals, Bac2 and Bac8 maintained tight clustering with Bac3 in both ordinations.

Virulence factors and antibiotic resistance genes

Toxins and non-toxin virulence factors were analysed using the VFDB database screening for all strains. Virulence factors were detected in only two strains: B. subtilis s.s. Bac3 and B. cereus s.s. zielonkawy. Since B. cereus s.s. zielonkawy belongs to Bacillus cereus s.l. clade, its genome was additionally analysed using BTyper3.

Virulence potential

Virulence factors were detected in only two of six strains tested (Table 12). P. amylolyticus 5mez, Pr. megaterium 7psych, B. velezensis Bac2, and B. licheniformis Bac8 showed no detectable virulence factors or toxin-encoding genes. B. subtilis s.s. Bac3 harboured a single non-toxin virulence factor: bslA/yuaB encoding hydrophobin BslA, detected with 100% coverage and 98.90% identity to the reference sequence. This protein functions as a component of the biofilm matrix but lacks direct toxicity. No toxin-encoding genes were identified in this strain.

Table 12.

Virulence factors detected in strains B. subtilis s.s. Bac3 and B. cereus s.s. zielonkawy using VFDB and BTyper3 databases. Gene name, product description, coverage, and identity percentages, source database, and toxin classification are shown

Sample	gene	product	database	toxin
B. subtilis s.s. Bac3	bslA/yuaB	(BslA/YuaB) hydrophobin BslA [BslA (VF0411)]	VFDB	-
B. cereus s.s. zielonkawy	BAS3109	(BAS3109) thiol-activated cytolysin [ALO (VF0534)]	VFDB	ALO cytolysin
	bpsE	(BpsE) exopolysaccharide capsule protein	bTyper3	-
	cytK	(CytK) cytotoxin K [CytK (VF0535)]	bTyper3;VFDB	cytotoxin K
	hblA	(HblA) hemolysin BL binding component precursor [HBL (VF0532)]	bTyper3;VFDB	hemolysin BL (HBL)
	hblB	(HblB) hemolysin BL binding subunit HblB	bTyper3
	hblC	(HblC) hemolysin BL lytic component L2 [HBL (VF0532)]	bTyper3;VFDB
	hblD	(HblD) hemolysin BL lytic component L1 [HBL (VF0532)]	bTyper3;VFDB
	inhA	(InhA) immune inhibitor A metalloprotease [InhA (VF0536)]	VFDB	-
	nheA	(NheA) non-hemolytic enterotoxin A [Nhe (VF0533)]	bTyper3;VFDB	non-hemolytic enterotoxin (Nhe)
	nheB	(NheB) non-hemolytic enterotoxin B [Nhe (VF0533)]	bTyper3;VFDB
	nheC	(NheC) non-hemolytic enterotoxin C [Nhe (VF0533)]	bTyper3;VFDB
	sph	(Sph) sphingomyelinase	bTyper3	sphingomyelinase

B. cereus s.s. zielonkawy harboured genes for five distinct toxin systems (Table 12). The strain possessed the following genes: BAS3109 encoding ALO cytolysin, cytK encoding cytotoxin K, a complete hemolysin BL (HBL) system, comprising four genes: hblA (binding component precursor), hblB (binding subunit), hblC (lytic component L2), and hblD (lytic component L1), as well as a complete non-hemolytic enterotoxin (Nhe) system consisting of nheA, nheB, and nheC, and the sph gene encoding sphingomyelinase. Non-toxin virulence factors in B. cereus s.s. zielonkawy included bpsE (exopolysaccharide capsule protein) and inhA (immune inhibitor A metalloprotease). Interestingly, the genes hblAB were located on the chromosome, while hblCD were located on the plasmid; all other genes were located on the chromosome.

Complete results are available in Tables S15 and S16.

Antibiotic resistance potential

Antibiotic resistance genes (ARGs) were detected in four of six strains using Abricate (Table 13). Only P. amylolyticus 5mez and Pr. megaterium 7psych contained no detectable ARGs.

Table 13.

Antibiotic resistance genes (ARGs) detected in strains B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8, and B. cereus s.s. zielonkawy using ABRicate against multiple databases (NCBI AMRFinderPlus, CARD, ResFinder, MEGARes, VFDB)

sample	Gene	no. of databases	mechanism
B. velezensis Bac2	cfr/clbA	4	23S rRNA methylation
	rphC	2	rifamycin phosphorylation
	satA	2	streptothricin acetylation
B. subtilis s.s. Bac3	aadK	4	aminoglycoside adenylation
	blt	2	efflux pump
	bmr (2 loci)	2	efflux pump
	lmrB	2	efflux pump
	mphK	4	macrolide phosphorylation
	mprF	1	membrane modification
	rphB/rphC	1/2	rifamycin phosphorylation
	satA	2	streptothricin acetylation
	tet(L)	3	tetracycline efflux pump
	tmrB	2	efflux pump
	vmlR	3	ribosomal protection
	ykkC (1 locus) and ykkC/ykkD (1 locus)	2/1	efflux pump
B. licheniformis Bac8	blaP	2	beta lactamase
	rphB/rphC	1/2	rifamycin phosphorylation
B. cereus s.s. zielonkawy	bla1/BcI	3	beta lactamase
	bla2/BcII	3	beta lactamase
	fosB	4	fosfomycin transferase
	tet(45)	4	tetracycline efflux pump
	vanZ_F	3	glycopeptide resistance

B. velezensis Bac2 harboured three antibiotic resistance genes with high sequence identity, including cfr/clbA (23S rRNA methyltransferase conferring resistance to multiple antibiotic classes), detected across all four databases.

B. subtilis s.s. Bac3 demonstrated the highest ARG diversity, with 14 resistance determinants, including duplicated efflux pump genes (bmr at 2 loci and the ykkC/ykkD complex). The strain exhibited multiple resistance mechanisms: efflux pumps (blt, bmr, lmrB, tmrB, ykkC/ykkD), antibiotic modification enzymes (aadK, mphK, satA), ribosomal protection (vmlR), membrane modification (mprF), and tetracycline efflux (tetL).

B. licheniformis Bac8 contained two antibiotic resistance genes: blaP (β-lactamase with 100% identity) and rifamycin phosphorylation enzymes (rphB/rphC).

B. cereus s.s. zielonkawy harbored five antibiotic resistance genes, including two distinct β-lactamases (bla1/BcI, bla2/BcII), a fosfomycin transferase (fosB), a tetracycline efflux pump (tet45), and a glycopeptide resistance determinant (vanZ_F).

All ARGs besides tet(45) were assigned to chromosome regions; tet(45) was assigned to the plasmid region.

Complete results are available in Table S15.

Discussion

Functional potential

General profile

PGPT gene distribution and genome specialization

The total number of PGPT genes varied substantially among strains, with Pr. megaterium 7psych harboring the highest number, followed by B. velezensis Bac2 and B. subtilis s.s. Bac3, then P. amylolyticus 5mez and B. licheniformis Bac8, and finally B. cereus s.s. zielonkawy with the lowest count. However, this ranking did not correspond to the percentage of PGPT genes relative to the total genome size, revealing a different pattern of genome specialization. When normalized by genome size, B. subtilis s.s. Bac3, B. velezensis Bac2, and B. licheniformis Bac8 exhibited the highest proportions of PGPT genes, indicating specialized genomic composition oriented toward plant – growth-promoting traits.

Furthermore, the absolute number of genes and their genomic proportion did not directly correlate with annotation abundance, as individual genes often received multiple annotations across different functional categories. In this respect, Pr. megaterium 7psych and P. amylolyticus 5mez dominated the annotation landscape. These patterns collectively indicate that no single strain emerges as an unequivocal "leader" across all metrics; rather, distinct functional profiles characterize each strain.

Strain Pr. megaterium 7psych exhibited a profile of PGPT dominance, with the highest absolute gene count and annotation abundance, suggesting a comprehensive multi-functional profile. Priestia megaterium is a species with high biotechnological potential, including plant-growth promotion and biocontrol properties [32]. Recently, Escalante-Beltrán et al. described Pr. megaterium from maize-associated soil, possessing multiple plant-growth-promoting traits, as evidenced by both genomic predictions and in vitro-documented PGP properties [33].

Strain P. amylolyticus 5mez displayed a moderate gene count but a high annotation density, indicating the presence of multifunctional genes involved in multiple processes and receiving annotations across diverse functional categories. As reported by Lal et al., plant-associated Paenibacillus species may possess biocontrol and PGP properties [34]. Moreover, these species can be readily cultured from both bulk and rhizosphere soil, which enhances their practical applicability [34].

Strains B. velezensis Bac2 and B. subtilis s.s. Bac3 possessed substantial gene counts with high genome specialization, while B. licheniformis Bac8 presented moderate gene numbers with considerable genome specialization. B. velezensis is a species commonly found in the soil, especially in the rhizosphere, which has been considered a model organism for both plant-growth-promoting and biological control agent properties [35]. B. subtilis, a type species, a model for studies on physiology and metabolism, is at the same time a species possessing properties that can be applied not only in agriculture, but in industry and medicine as well [36]. B. licheniformis is also known to have PGP properties; moreover, it has been applied as a mixture of B. subtilis and B. licheniformis, successfully stimulating tomato growth [37]. All three Bacillus subtilis s.l. strains (B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8) exhibited relatively fewer annotations, suggesting enrichment in genes dedicated to specific, singular processes. In this regard, the pattern would suggest some consistency with phylogenetic relationships; this will be discussed in the following sections.

Strain B. cereus s.s. zielonkawy presented the smallest gene count with moderate annotation abundance. B. cereus s.s. is a serious, though often forgotten, food pathogen. In 2018, it was estimated that 1.4%- 12% of foodborne outbreaks worldwide could be attributed to B. cereus [38]. However, it is often found in soil and exhibits valuable PGP traits, directly stimulating the growth of crops and even medicinal plants [39]. Therefore, the application of non-pathogenic B. cereus strains is being tested.

Functional PGPT differentiation along principal components

The PCA performed on PGPT functional profiles primarily differentiated strain Pr. megaterium 7psych from other strains along PC1, with the most visible separation observed between Pr. megaterium 7psych and the Bacillus subtilis clade (B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8). Strain P. amylolyticus 5mez occupied an intermediate position along PC1, while B. cereus s.s. zielonkawy was positioned between the Bacillus subtilis s.l. cluster and P. amylolyticus 5mez. PC2 primarily separated P. amylolyticus 5mez from the remaining strains. Given that Pr. megaterium 7psych possessed both the highest gene count and annotation abundance, its distinct positioning along PC1 was expected. Concluding, the most considerable similarities were observed within Bacillus subtilis s.l., and the highest differences were observed between Pr. megaterium 7psych, P. amylolyticus 5mez, and the rest of the strains, while B. cereus s.s. zielonkawy falls somewhere in the middle, but closer to Bacillus subtilis s.l., confirming that phylogenetic relationships affect PGPT profiles. These similarities further reinforce the observation that Bacillus subtilis s.l. may form a coherent functional unit with conserved plant-growth-promoting traits, which is tested in this research.

The functional basis of Pr. megaterium 7psych differentiation was evident in the strong negative loadings on PC1, which included Nitrogen Acquisition, Neutralizing Abiotic Stress, Plant Vitamin Production, multiple Phytohormone production categories (Cytokinins/Derivatives, GABA, Abscisic Acid Degradation), Sulfur Assimilation/Mineralization, Heavy Metal Detoxification, and Induction of Systemic Resistance/ISR. In contrast, the positive end of PC1, associated with the Bacillus strains, was characterized primarily by Triggered Immunity. The differentiation of strain P. amylolyticus 5mez along PC2 was driven by strong positive loadings in colonization-related functions, particularly Colonization-Plant Cell Wall/Membrane Degradation, Ce-Exopolysaccharide Production/EPS, and Iron Acquisition.

These results indicate that while Pr. megaterium 7psych exhibits broad metabolic versatility, including stress mitigation, nutrient acquisition, and phytohormone modulation, concordant with the literature information [32]. Strain P. amylolyticus 5mez is functionally specialized toward plant tissue colonization through degradative enzymes and biofilm formation, as well as iron acquisition. This species has been reported as rich in carbohydrate-active enzymes [40]. Moreover, siderophore production has been reported in the genus [34].

General secondary metabolite profile

Biosynthetic gene cluster analysis using antiSMASH revealed substantial heterogeneity in secondary metabolite potential across the six strains. Total BGC counts ranged from 7 (Pr. megaterium 7psych) to 15 (P. amylolyticus 5mez, B. velezensis Bac2), with intermediate values in B. subtilis s.s. Bac3, B. licheniformis Bac8, and B. cereus s.s. zielonkawy. This variation suggests different metabolic strategies, with strains P. amylolyticus 5mez and B. velezensis Bac2 possessing broader biosynthetic capacity than, e.g., Pr. megaterium 7psych.

Despite detecting 70 BGCs across all strains, only 33% presented high-similarity matches to characterized compounds in reference databases. Many identified matches correspond to well-studied Bacillus metabolites. Among predicted compounds, the majority were compounds with antibacterial and/or antifungal activity. The second major functional group was siderophores and metallophores. The low identification rate is not uncommon, suggesting the potential novelty of the metabolites associated with these clusters [41, 42]. The identification rate varied highly between strains: B. subtilis s.s. Bac3 achieved 73%, B. velezensis Bac2 40%, while P. amylolyticus 5mez showed only 7%, and Pr. megaterium 7psych yielded no matches despite possessing seven BGCs.

The visibly lower identification rates in Priestia and Paenibacillus than in Bacillus strains may reflect the current state of secondary metabolite research. The unique cluster types detected in P. amylolyticus 5mez (proteusine, triceptide, cyclic-lactone-autoinducer) and Pr. megaterium 7psych (phosphonate) suggest distinct biosynthetic capabilities. Paenibacillus species have increasingly been recognised as sources of novel natural products [43]. The low identification rate of BGCs has already been studied for the Paenibacillus genus – Kim et al. recently analysed 554 Paenibacillus genomes with antiSMASH and identified 848 BGCs, of which 716 (84.4%) were classified as unknown [44].

CAZyme profile distribution

Analysis of CAZyme profiles using the dbCAN database revealed a different distribution pattern. Strain P. amylolyticus 5mez dominated across all CAZyme metrics: it harboured the highest number of CAZyme genes (approximately twice that of the second-ranked strain), the largest proportion of CAZyme genes relative to genome size, and the greatest number of CAZyme domains (more than double that of the runner-up). This clear dominance positioned P. amylolyticus 5mez as the unequivocal leader in carbohydrate-active enzyme spectrum. A strain of this species – P. amylolyticus 27C64 – has already been described as having many or more putative CAZymes than most of the well-studied polysaccharide-degrading organisms [40].

Strains B. velezensis Bac2, B. subtilis s.s. Bac3, and B. licheniformis Bac8 ranked considerably lower, followed by Pr. megaterium 7psych, with B. cereus s.s. zielonkawy exhibiting the smallest CAZyme spectrum. The CAZyme profile (supported by PCoA) confirmed the PGPT data (supported by PCA) showing the highest functional similarities within B. subtilis s.l.

NPK and Iron acquisition

Nitrogen fixation

The capacity for biological nitrogen fixation is restricted to certain members within the Bacteria and Archaea domains. It necessitates the nitrogenase enzyme complex, whose subunits are typically encoded by nifDHK genes within the nif cluster [45]. Nevertheless, the synthesis of a functional nitrogenase complex in bacteria appears to require a minimum of nine genes – the conserved in Paenibacillus polymyxa WLY78 cluster nifBHDKENXhesAnifV transduced into Escherichia coli cells enabled nitrogen fixation with nifB or T7 promoter [45, 46]. In contrast, the introduction of merely six core genes (nifBHDKEN) proved insufficient for functional nitrogenase synthesis; seven genes (nifBHDKENnifV) resulted in only minimal nitrogenase activity, while eight genes (lacking either nifX or hesA) achieved nitrogen fixation at approximately 50% reduced efficiency [45, 46].

Nitrogen-fixing bacteria can be categorised into three primary groups: symbiotic nitrogen fixers (SNF), which function as endophytic mutualists; associative nitrogen fixers (ANF); and free-living nitrogen fixers (FLNF). Free-living diazotrophs fix nitrogen independently of a plant host [45]. For plants, the key advantage of FLNF bacteria is that the nitrogen they fix becomes available to all plant species. Free-living diazotrophs play a substantial role in global terrestrial BNF, contributing to approximately one-third of it [47]. FLNF bacteria are distributed across multiple phyla, including Bacillota.

In all six strains, some nif cluster genes have been detected – nifM, nifS, and nifU – varying in copy number. The nifU and nifS genes are responsible for the assembly of the Fe-S cluster, whereas nifM is involved in the proper folding of the Fe protein [46]. However, those are not core genes (which are nifBHDKEN), and mutations in those genes do not entirely eliminate nitrogenase activity; evidence suggests that homologues of these nif genes present elsewhere in the genome may substitute for their function [46]. No genes connected to catalytic subunits (nifHDK) of the nitrogenase complex, nor genes connected to alternative nitrogenases (vnfHDGK, anfHDGK [45]) have been found in the genomes.

Denitrification and dissimilatory nitrate reduction to ammonium (DNRA)

Denitrification is a process of reduction of NO₃⁻ to NO₂⁻, then to NO, N₂O, and finally to N₂. Therefore, denitrification results in a loss of bioavailable nitrogen and is the major form of nitrogen loss in most soils [48, 49]. This consequence also underlies the phenomenon of biological denitrification inhibition (BDI), in which plants block these enzymatic processes via secondary metabolites [49, 50].

DNRA is another process within the nitrogen cycle. DNRA competes with denitrification for NO₃ [51]. DNRA, on the other hand, leads to NO₃⁻ reduction to NO₂⁻, but then to ammonium – NH₄⁺ [51]. Ammonium is a nitrogen source for plants, far less costly than NO₃⁻, whose transport requires significant energy consumption [51]. As a result, DNRA does not lead to a loss of bioavailable nitrogen but to its recycling, potentially enhancing plant nitrogen use efficiency. DNRA utilizes distinct nitrite reductase enzymes (nirBD/nrfAH) specific to the DNRA pathway [52, 53].

The denitrification pathway is incomplete in each strain and when treated collectively. No strain is capable of transforming nitrate to atmospheric nitrogen or nitrous oxide; most strains are capable of only the transformation of nitrate to nitrite, which is the first step of the denitrification and DNRA pathways. Concerning nitrification, no strain is capable of transforming ammonia to hydroxylamine, nor of hydroxylamine to nitrite. Pr. megaterium 7psych is not capable of neither in nitrification, nor in denitrification.

In the denitrification pathway, P. amylolyticus 5mez is the only one capable of transforming nitrite to nitric oxide (but is unable to transform nitrate to nitrite). 5mez cannot transform nitric oxide to nitrous oxide, or further to nitrogen. However, NO is also a gaseous form, leading to nitrogen loss in both nitrification and denitrification processes [54, 55].

nirBD genes are the only ones present in all six strains. It is consistent with the literature data concerning the studied species. DNRA activity has been observed in Bacillus species for years; however, knowledge at the genomic level is a relatively recent one [56]. Hao et al. have experimentally proven the DNRA potential of B. velezensis BV-1, at the same time describing upregulated expression of the key gene nirD [57]. Ahn et al. have experimentally proven DNRA processes in Bacillus and closely related Neobacillus strains [58]. Heo et al. have described DNRA-capable Bacillus sp. DNRA2, which does not possess the nirB gene, but nfrA, absent in the strains studied by us [59]. Zhou et al. have described nirBD genes in 14 strains belonging to the genus Paenibacillus, though not P. amylolyticus [60]. Finally, Sun et al. studied three non-denitrifying DNRA-capable Bacillus licheniformis strains (LMG 6934, 7559, 17,339) that possess the nirBD genes [61].

Ammonium assimilation represents potential competition with plants for NH₄⁺ uptake, which may suggest detrimental effects. However, Månsson et al. have shown experimentally that, under standard moisture conditions, they do not specifically compete for NH₄⁺ [62]. Moreover, a recent study by Gao et al. shows that bacteria assimilating ammonium via GDH (glutamate dehydrogenase) and GS-GOGAT (glutamine synthetase-glutamine-2-oxoglutarate aminotransferase) pathways may actually stimulate plant growth, as an agent for nitrogen accumulation and immobilisation [63]. The gdhA gene, responsible for the GDH pathway, was detected only in 5mez, Bac8, and zielonkawy, but glnA, responsible for the GS-GOGAT pathway, was detected in all studied strains.

Nitrogen – experimental data comparison

Rowińska et al. have shown experimentally that all five strains tested – B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8, Pr. megaterium 7psych, P. amylolyticus 5mez (i.e., WK-14, WK-15, WK-20, WK-5, WK-9 in the article’s labels) are capable of atmospheric nitrogen fixation [3]. The strain B. cereus s.s. zielonkawy (WK-21), due to its pathogenic risk, has not been tested. One must note that nitrogen fixation tests performed by culturing on nitrogen-free media (NFMs) have been controversial [64]. Koirala et al. have demonstrated that only 17% of putative nitrogen-fixers grown on NFMs possessed the nifH marker gene [64]. Furthermore, NFMs prepared with even Milipore-grade pure water contained 6.42 μmol/L of total nitrogen (TN) – for Escherichia coli, it would be enough to grow 4.28 × 10⁶ cells per mL [64]. What is more, as early as in 1969, it was noted that not all isolates growing on nitrogen-free media were able to fix nitrogen [65]. Since then, various reports of false-positive results for diazotrophic bacteria and fungi have emerged due to their ability to grow in nitrogen-free media [64]. Giller et al. suggest that, among others, to acknowledge a strain as a diazotroph, the minimum set of six core genes needs to be present, and there should be a clear mechanism to explain the oxygen paradox [66].

As for today, under current classification, there are no Bacillus or Priestia members known to possess nif core genes; it has been reported in closely-related genera Paenibacillus, Evansella, and Niallia, and diazotrophy among aerobic endospore-forming Bacillota (as tested by Rowińska et al. [3]) is restricted to the genus Paenibacillus only [64]. There are, however, multiple reports of diazotrophic Bacillus isolates classified as B. subtilis, B. licheniformis, and B. cereus, supported only by culture evidence for growth on nitrogen-free media [64]. None of these claims has been supported by genetic evidence, and, as noted by Koirala et al., understanding the mechanisms underlying the growth of Bacillus and closely related genera on nitrogen-free media is urgently needed [64]. To finally verify whether query strains do fix nitrogen, the ¹⁵N-stable isotope probing should be applied [67].

Phosphate solubilisation

After nitrogen, phosphorus represents the most critical nutrient for plant development, participating in numerous fundamental metabolic pathways. Plant roots take up phosphorus as orthophosphate ions (H₂PO₄⁻ or HPO₄²⁻), which typically occur at micromolar concentrations in soil [45]. Soil phosphorus exists in both mineral and organic forms. Phosphate-solubilizing bacteria (PSBs) can mobilise inorganic phosphate and/or mineralise organic phosphate compounds through various mechanisms.

Inorganic phosphorus

Mineral phosphate solubilization occurs through organic acid secretion, inorganic acid production, proton release, exopolysaccharide synthesis, and – in the case of inorganic polyphosphates – enzymatic cleavage [45]. Among these strategies, organic acid production appears to be the predominant mechanism. In our dataset, Pr. megaterium 7psych dominates overall, as evident in the PCA analysis, and particularly in organic acid production pathways. The highest gene counts are found in organic acid-related functions, consistent with published literature. While PSBs can also produce hydrochloric, sulfuric, nitric, and carbonic acids to mobilise phosphate, these inorganic acids are considerably less efficient than their organic counterparts [45]. Inorganic acid production, therefore, represents an alternative strategy, and B. subtilis s.s. Bac3 shows dominance in this context. Finally, inorganic polyphosphates – though less studied and underexplored – may also be present; those are degraded enzymatically by exopolyphosphatase (ppx), and this gene is present only in Pr. megaterium 7psych, P. amylolyticus 5mez, and B. cereus s.s. zielonkawy [68]. ppx gene presence has recently been described in Priestia megaterium LAMA1607 [69] and in P. polymyxa strains [70], but we found no literature data on P. amylolyticus. In B. cereus ATCC 14579, the ppx gene was described already in 2004, and the obtained protein has been studied [71].

Organic phosphorus

Three enzymatic systems mobilise organic phosphates: (i) non-specific acid phosphatases (NSAPs), (ii) phytases, and (iii) phosphonatases and C–P lyases [45]. NSAPs are classified as acidic or alkaline based on their pH optima. In our strains, alkaline NSAPs are represented by phoA and phoD, with phoD serving as the most universal marker gene. The phoA gene is present in all strains, whereas phoD is found in B. velezensis Bac2 and B. subtilis s.s. Bac3, B. licheniformis Bac8, and Pr. megaterium 7psych. Additionally, the poorly characterised phoE gene, which is likely also an NSAP, is present in our dataset [72].

Phosphonatases and C–P lyases mediate phosphonate degradation. Within this category, Pr. megaterium 7psych dominates, and the key gene is phnX (phosphonatase: phosphonoacetaldehyde hydrolase), which catalyzes the cleavage of stable C–P bonds [73]. This gene is present exclusively in Pr. megaterium 7psych and B. cereus s.s. zielonkawy, demonstrating the remarkable specialization of those strains for phosphonate metabolism. It has been recently described in Pr. megaterium LAMA1607, as well as four B. cereus strains (905, AR156, LCR12, UW85) [69, 74]. Phytate (inositol hexaphosphate) comprises 30–50% of total organic phosphorus [75]. However, it remains inaccessible to plant roots, therefore bacterial phytases are of high importance [75]. Phytases, encoded by phy genes, are present only in B. velezensis Bac2, B. subtilis s.s. Bac3 and B. licheniformis Bac8, reflecting their adaptation to this specific phosphorus source, are consistent with phylogenetic closeness.

Phosphorus – experimental data comparison

Rowińska et al. have also tested phosphate solubilisation experimentally using Pikovskaya medium (PVK) [3]. All strains, but B. licheniformis Bac8, demonstrated the phosphate solubilisation potential. However, this does not mean that B. licheniformis Bac8 is not a PSB, as Pikovskaya (PVK) medium may yield false-negative results. As noted by Nautiyal [76], many isolates that did not show clear halos in PVK agar nevertheless solubilized phosphate in liquid culture, in NBRIP medium. However, both PVK and NBRIP media contain only one phosphate source: tricalcium phosphate (TCP). Bashan et al. have shown that TCP is not an appropriate universal screening medium for PSBs and suggested that a combination of two or three insoluble metal-P compounds should replace TCP as an initial screening method [77]. In conclusion, most often only inorganic phosphate solubilisation tests are performed, organic phosphorus sources are not included, and the only inorganic source is TCP. To fully verify the phosphate solubilising potential, different methods for inorganic phosphates, as well as organic phosphorus sources, should be used.

Potassium solubilisation

Potassium constitutes the third essential nutrient in NPK fertilizers alongside nitrogen and phosphorus. Potassium-solubilizing bacteria (KSBs) function primarily through organic acid production [78]. Our results for potassium solubilization closely mirror those for phosphate solubilization, reflecting similar organic acid profiles. The annotation count for potassium-related organic acids is only slightly smaller than that for phosphate solubilization. Potassium-solubilising species described in the literature include B. cereus, B. subtilis, B. licheniformis, Pr. megaterium, and P. amylolyticus [78].

Iron acquisition

Iron functions as a crucial cofactor in numerous enzymatic processes in both plants and bacteria [79]. Despite iron's abundance in the Earth's crust, its bioavailability is severely limited because it predominantly exists as oxidised Fe³⁺, which exhibits poor solubility at neutral and alkaline pH [79, 80]. Bacteria commonly acquire iron through siderophore production. There are five main groups of siderophores [45]. Different organisms may produce identical siderophores, and individual organisms can synthesise multiple siderophore types.

The former scenario is present in our dataset, as siderophore-related annotations in all strains are linked to bacillibactin metabolism and transport, a catecholate-type siderophore, known to be produced by B. subtilis, B. velezensis, B. licheniformis, and B. cereus [81–84]. It is also produced by the Paenibacillus genus, but it has been proven only for P. larvae and not for P. amylolyticus or even the type strain P. polymyxa [43, 85]. Due to the incorporated threonine residue, bacillibactin exhibits exceptional ferric iron binding affinity among natural siderophores and serves as the primary extracellular iron scavenger in B. subtilis under iron-limiting conditions [81]. Siderophore annotations are relatively evenly distributed across strains, though with a slight predominance in P. amylolyticus 5mez.

Secondary metabolite prediction

In the antiSMASH analysis, bacillibactin BGCs were detected in four Bacillus strains (B. velezensis Bac2, B. subtilis s.s. Bac3, B. licheniformis Bac8, B. cereus s.s. zielonkawy), but neither in P. amylolyticus 5mez, nor Pr. megaterium 7psych. As for siderophores, petrobactin has also been detected in B. cereus s.s. zielonkawy. It is not surprising, as this siderophore is predominantly studied in closely related B. anthracis [86]. In P. amylolyticus 5mez BGC of bacillopaline, a recently discovered opine-type metallophore, has been detected. It is understandable, as bacillopaline has primarily described in other Paenibacillus species, i.e., P. mucilaginosus [87].

Stress neutralisation

Abiotic stresses

Genomic annotations indicate that strain Pr. megaterium 7psych possesses the greatest potential for abiotic stress tolerance, with particularly high annotation counts for salinity, oxidative/nitrosative/ROS scavenging, and herbicide tolerance. Additionally, Pr. megaterium 7psych showed elevated low-temperature tolerance annotations alongside robust high-temperature tolerance. It should be noted that – as in all cases – these annotation counts represent predicted genetic capacity rather than confirmed phenotypic traits.

The high salinity and oxidative stress annotation counts in Pr. megaterium 7psych suggest strong potential for agricultural applications. Salt-tolerant PGPR reduce oxidative damage in plant tissues by upregulating ROS-scavenging enzymes and help maintain K⁺/Na⁺ homeostasis under saline conditions [88]. ROS-scavenging bacteria can induce systemic tolerance in plants, priming stress responses through PGPR-plant interactions that enhance the plant's own antioxidant defence capacity [89].

The elevated low-temperature tolerance annotations, combined with solid high-temperature capacity, suggest Pr. megaterium 7psych may function across broad thermal ranges. This combination was demonstrated in Bacillus sp. IHBT-705, which retained growth-promoting activity from 4 °C to 50 °C and tolerated high salinity (8% NaCl), as well as a wide pH range (5–11) [90]. Herbicide tolerance annotations enable predicted persistence during agricultural weed management. Various herbicide-resistant PGP bacteria have been described, promoting plant fitness while maintaining herbicide tolerance [90]. Such an effect may be a two-way solution – promoting plant fitness and simultaneously degrading problematic synthetic pesticides [91].

Together, these genomic features suggest Pr. megaterium 7psych has strong potential for agricultural applications in stress-prone, thermally variable, or chemically managed environments, though experimental confirmation is necessary. Other strains have moderately high counts as well, indicating that multiple isolates (e.g., B. subtilis, B. velezensis) could have useful but less extreme stress‐tolerance profiles. In all cases, however, these genomic predictions must be validated by phenotypic assays under stress conditions.

Heavy metal tolerance

Plant-growth-promoting bacteria commonly have many strategies to cope with heavy metal toxicity [92]. Heavy metal-resistant PGP bacteria can be used to improve the growth of plants in heavy metal-contaminated soils, to remediate the soil, and to neutralise plant heavy metal stress. The best-described species in terms of the bioremediation potential of Bacillus spp. are B. subtilis and B. cereus. Biosorption, bioaccumulation, and bioprecipitation are the most common heavy metal removal strategies in the genus Bacillus [92]. Heavy metal detoxification capacity shows marked heterogeneity across query strains, with P. amylolyticus 5mez demonstrating the highest overall capacity (337 total annotations), followed by Pr. megaterium 7psych (320 annotations).

Besides the already-discussed iron, the second- and third-highest annotation counts are for resistance to copper and nickel. The number of annotations in those metals is very similar for P. amylolyticus 5mez and Pr. megaterium 7psych, distinct from the other strains. Like many other heavy metals, Cu is considered essential for Bacillus spp.; however, concentrations exceeding normal levels can be toxic to cells [93]. Species including B. subtilis and B. cereus are involved in the removal of Cu when their concentrations exceed the required limit [93]. Nickel is essential for the growth of plants, microorganisms, and animals, as Ni-based enzymes and cofactors serve key roles. B. thuringiensis and Pr. megaterium are known to uptake Ni from contaminated soils [93]. Arsenic is of a soluble nature, which makes it difficult to remove from the environment. Arsenic removal has been described, for instance, in Pr. megaterium, Pr. aryabhattai, and B. cereus [93]. In the studies strains, the most annotations were ascribed to B. cereus s.s. zielonkawy and P. amylolyticus 5mez. Bacillus spp. The ability to uptake zinc (high results for P. amylolyticus 5mez, Pr. megaterium 7psych, and B. cereus s.s. zielonkawy) is well studied in the context of bioremediation [93]. In conclusion, it seems that P. amylolyticus 5mez, Pr. megaterium 7psych, and B. cereus s.s. zielonkawy may possess the highest heavy metal tolerance potential in all strains.

Biotic stresses

The top three annotation categories included bactericidal, volatile compounds, and fungicidal compounds. The highest annotation numbers for bactericidal mechanisms primarily reflect genes for bacteriocin/lantibiotic/nisin biosynthesis, prodigiosin metabolism, and spermidine/putrescine metabolism. Bacteriocins are ubiquitous in Bacillaceae, and species such as B. subtilis, B. licheniformis, B. cereus, Pr. megaterium. B. velezensis have been reported multiple times to produce those [94, 95]. Plant-associated members of Bacillus cereus s.l. are perceived as a rich source of antimicrobial compounds [96]. Aside from Bacillaceae, the genus Paenibacillus is also known for producing at least two or three classes of bacteriocins [43].

Volatile organic compounds (VOCs)

VOCs biosynthesis annotations were dominated by volatile-related fatty acid metabolism and acetoin/2,3-butanediol biosynthesis, with strain Pr. megaterium 7psych showing the highest capacity in both categories and overall. Acetoin and 2,3-butanediol (2,3-BDO) are key biocontrol volatiles produced by Bacillus spp., exhibiting direct antifungal activity and indirect effects through induction of plant systemic resistance [97, 98]. For instance, B. licheniformis DSM13 culture broth produced meso-2,3-BDO that significantly reduced R. solanacearum-induced disease severity [97]. Moreover, VOCs produced by B. amyloliquefaciens L3 played important roles in biocontrol, as shown experimentally against F. oxysporum f.sp. niveum [98]. Furthermore, acetoin and 2,3-BDO from B. amyloliquefaciens FZB42 induced stomatal closure in Arabidopsis thaliana and Nicotiana benthamiana [99].

Fungicidal annotations

Total fungicidal annotations show P. amylolyticus 5mez dominance, followed by B. cereus s.s. zielonkawy and B. velezensis Bac2. The primary category was chitinolytic activities. It is consistent with the literature, which shows that, e.g., B. licheniformis chitinases have potential for cell wall lysis of various tested phytopathogenic fungi [100]. There are also pathways connected to specific compounds; however, various clusters remain incomplete, such as in the case of antifungal lipopeptides. Several strains encoded lipopeptide fungicidal antibiotics.

In the case of the fengycin/plipastatin gene clusters, genes are found only in B. velezensis Bac2 and B. subtilis s.s. Bac3 and both possess the complete ppsABCDE/fenABCDE cluster. B. velezensis Bac2 and B. subtilis s.s. Bac3 have the full potential to produce fengycin/plipastatin. Fengycin is a cyclic lipopeptide with remarkable antifungal, antitumor, and antiadhesion properties, described predominantly in Bacillus species [101].

Yet, in the case of the iturin gene cluster (ituABCD), ituA is present in the genomes of P. amylolyticus 5mez and B. velezensis Bac2, ituB is present in P. amylolyticus 5mez, B. velezensis Bac2, and B. cereus s.s. zielonkawy, ituC – only in B. velezensis Bac2, and ituD is absent. It has been experimentally demonstrated that the absence of ituD results in specific iturin A deficiency [102]. Of course, it is always technically possible that ituD is missing due to scaffold-level genome assembly, and B. velezensis Bac2 operon is complete and functional. Fusaricidin gene fusA is present only in B. velezensis Bac2 and is absent in P. amylolyticus 5mez, which is interesting considering the fact that the fusaricidin has been described in the genus Paenibacillus only (yet in P. polymyxa) [103, 104]. However, it has been shown that even a single deletion in the fusABCDEFG gene cluster (excluding fusTE) reduces antifungal activity; it is unlikely that a single gene confers antifungal activity [105]. Pelgipeptin genes are also present – plpE in P. amylolyticus 5mez, B. velezensis Bac2, and B. cereus s.s. zielonkawy, plpFGH – in P. amylolyticus 5mez. Pelgipeptin is another antibacterial and antifungal lipopeptide [106]. The analysis, therefore, revealed a highly fragmented distribution of the core NRPS genes (plpE, plpF) across the tested strains, with no single strain harbouring the complete set of genes required for functional pelgipeptin synthesis, and none harbouring plpD [106].

The incomplete biosynthetic gene clusters found here raise questions about their origin. Francis and Tanaka [107] argue that variation in bacterial pathway genes typically results from the decay of existing complete pathways. According to this model, horizontal gene transfer mainly introduces new complete pathways rather than building up variation through gradual gene acquisition. Moreover, Steinke et al. [108], while studying Bacillus subtilis s.l., found that partial BGC deletions, incomplete BGCs, and frameshift mutations in biosynthetic genes are conserved across phylogenetically related isolates from different geographic locations.

Insecticidal annotations

Insecticidal compound production presented relatively low but variable annotation counts, from P. amylolyticus 5mez, B. licheniformis Bac8 to Pr. megaterium 7psych. At level 5, this category was represented by two distinct mechanisms. A γ-amino-butyric-acid (GABA)-mediated activity dominated insecticidal capacity. This pathway appears to function through indirect biocontrol mechanisms. It has been proven that GABA treatment strongly increased rice resistance to Sogatella furcifera, most probably by regulating the antioxidant defense system, TCA cycle regulation, phytohormonal signaling, and PR gene regulation [109]. Direct insecticidal toxin production was detected exclusively in B. velezensis Bac2, represented by the TccC protein, a subunit of the insecticidal toxin complex, which can be insecticidal on its own [110].

Secondary metabolite prediction

In the antiSMASH analysis, many antimicrobial compounds have been detected. For instance, BGC of fengycin has been discovered in B. velezensis Bac2 and B. subtilis s.s. Bac3. Bacillaene, which is a compound with broad antibacterial and antifungal activity, has been predicted in B. velezensis Bac2 and B. subtilis s.s. Bac3 as well [111]. Surfactin, a powerful biosurfactant, also exhibits antibacterial properties, and has also been predicted in B. velezensis Bac2 and B. subtilis s.s. Bac3 [112]. Both in B. velezensis Bac2 and B. subtilis s.s. Bac3 bacilysin is present, which is a common Bacillus-related antibacterial and antifungal dipeptide, though very variable in the producer species or strain affiliation [113]. Solely in B. velezensis Bac2 macrolactin H, another antibacterial compound has been detected [114]. On the other hand, only B. subtilis s.s. Bac3 produces pulcherriminic acid – an antibacterial dipeptide with iron chelating properties, and subtilosin – another antibacterial peptide [115, 116]. Furthermore, B. licheniformis Bac8 is the only potential producer of lichenicidin VK21, a bactericidal dipeptide lantibiotic, very characteristic of B. licheniformis [83].

Biotic stress – experimental data comparison

Rowińska et al. have tested 5 strains (excluding B. cereus s.s. zielonkawy) against eight fungal phytopathogens and used the post-culture tryptic soy broth (TSB) medium filtrates to assess toxicity against the insect cell line (ovarian pupa from Spodoptera frugiperda Sf-9) [3]. Strains B. velezensis Bac2, B. subtilis s.s. Bac3 and B. licheniformis Bac8 exhibited antifungal activity against all the tested phytopathogens, with B. velezensis Bac2 and B. subtilis s.s. Bac3 exhibiting the largest zones of inhibition. The species to which those strains belong (B. velezensis, B. subtilis, B. licheniformis) possess well-documented antifungal activities [3]. P. amylolyticus 5mez displayed high antifungal activity against three maize-related phytopathogens, but not against five others. Pr. megaterium 7psych exhibited the lowest level of antimicrobial activity. This is inconsistent with general annotation values, as P. amylolyticus 5mez had the most annotations. However, Pr. megaterium 7psych indeed possessed the fewest annotations, though not much different from B. subtilis s.s. Bac3 and B. licheniformis Bac8. The highest antifungal activity was present in B. velezensis Bac2 and B. subtilis s.s. Bac3, the only strains possessing a complete lantopeptide cluster (fengycin). Moreover, the results are consistent with identified secondary metabolites. B. velezensis Bac2 and B. subtilis s.s. Bac3 were both predicted to produce the same three antifungal compounds, that is, fengycin, bacillaene, and bacilysin.

Insect cell line toxicity results are inconsistent with the number of insecticidal annotations, as the highest toxicity was assessed in B. licheniformis Bac8 post-culture TSB filtrate, followed by B. subtilis s.s. Bac3. B. licheniformis Bac8 possessed only two annotations, and B. subtilis s.s. Bac3 four (7psych possessed nine). It is not surprising, as most annotations are connected to GABA pathways, which, although with documented potential against Sogatella furcifera, act more indirectly [109]. It is, though, surprising that B. velezensis Bac2 possessed weak activity, even being the only strain predicted to be able to produce proper insecticidal toxin, that is, TccC. However, one must notice that there was no exposure to insects or insect cells/tissues, as the tested material was just a filtrate from rich medium; therefore, there was no potentially necessary stimulation and pathway induction.

Plant-derived substrate usage and carbohydrate-active enzymes

All discussions and interpretations below are based on domain descriptions in the following sources: Carbohydrate Active Enzymes database [117], CAZypedia [118], and PROSITE [119].

Genomic predictions

The studied strains differ substantially in their metabolic capacities. P. amylolyticus 5mez outnumbers other strains in cellulose GHs, cellulose CBMs, xylan GHs, xylan CEs, xylan CBMs, mannan CBMs, pectin GHs, pectin PLs, pectin CBMs, starch GHs, chitin GHs, and chitin CBMs. Importantly, even if the difference in family diversity is modest, the number of domains per family in P. amylolyticus 5mez is 2–3 times higher, pointing to possible gene duplication and suggesting higher catalytic output rather than fundamentally new enzymatic functions. P. amylolyticus 5mez functions as a generalist degrader with extensive CAZymes for degrading cellulose, hemicelluloses, pectin, and starch. GH6 processive cellobiohydrolase and GH9 endoglucanase domains indicate the capacity to degrade both crystalline and amorphous cellulose. Its hemicellulose domains allow the degradation of xylan (via GH10/11/30/43 and CE1/2/4), mannans (via GH26 with CBM35 for binding), and mixed-linkage β-glucans (via GH16). High PGPT-Pred annotations for complex sugar utilisation support a role in crop residue decomposition.

This CAZyme profile is consistent with literature data for P. amylolyticus 27C64, which was also annotated using dbCAN [40]. Direct comparison reveals striking conservation of the polysaccharide-degrading machinery despite fundamentally different ecological niches – 27C64 originates from the hindgut of Tipula abdominalis larvae functioning in anaerobic leaf litter decomposition, whereas 5mez derives from aerobic post-agricultural soil [40]. The presence-absence patterns of GH and PL families are nearly identical: with one exception (GH76 mannanase absent in 5mez but present in 27C64), both genomes encode the same enzymatic toolkit. Where families are shared, single-copy families invariably show 1:1 correspondence between strains. For multi-copy families, domain count ratios range from 0.75 to 2.0, with the majority clustering around 1:1. This within-species conservation across ecologically disparate environments – anaerobic aquatic insect gut versus aerobic agricultural soil – could potentially indicate a stable species-characteristic CAZyme profile in P. amylolyticus. However, with only two genomes available, broader comparative genomic analyses across additional P. amylolyticus isolates from diverse environments would be necessary to determine whether this represents evolutionary conservation at the species level. Nevertheless, as P. amylolyticus 27C64 is perceived as a valuable resource for industrial applications, such as food processing or natural product extraction [120], and potentially, from our perspective, post-harvest residues decomposition, and with 5mez mimicking its putative potential, there is a visible potential of 5mez that has to be, however, verified experimentally under various conditions.

Strains B. velezensis Bac2, B. subtilis s.s. Bac3 and B. licheniformis Bac8 have intermediate total CAZyme profiles. B. licheniformis Bac8 shows cellulose family diversity close to P. amylolyticus 5mez, and B. velezensis Bac2 shows hemicellulose family diversity similar to P. amylolyticus 5mez, although total domain counts in these categories remain much lower. These strains are the only ones besides P. amylolyticus 5mez to possess GH26 mannanase, and B. velezensis Bac2 and B. subtilis s.s. Bac3 are the only ones besides P. amylolyticus 5mez to possess GH16 for degrading mixed-linkage β-glucans. B. licheniformis Bac8 is unique in having GH12 endoglucanase and AA10 lytic monooxygenase domains. All three share several core domain types for cellulose and hemicellulose degradation. This shared enzymatic core within Bacillus subtilis s.l. strains mirrors the PGPT profile similarities observed in PCA analysis, indicating conserved functional genotype within this clade.

Strains Pr. megaterium 7psych and B. cereus s.s. zielonkawy show more specialised profiles. B. cereus s.s. zielonkawy has average cellulose family diversity but low hemicellulose family diversity, while Pr. megaterium 7psych has low diversity in both categories. Notably, Pr. megaterium 7psych has high CE1 and CE4 domain content, suggesting a focus on xylan deacetylation rather than extensive hydrolysis. B. cereus s.s. zielonkawy has the lowest CAZyme content but is the only strain to possess AA1 laccase, indicating involvement in phenolic compound oxidation and detoxification of lignin-derived intermediates rather than polysaccharide breakdown.

Strains have limited lignin-modifying capacity. AA2 peroxidase and AA5 radical copper oxidase domains are absent. The detected AA4 and AA6 enzyme domains can oxidise para-substituted phenolics and reduce quinones, but primarily modify or detoxify lignin fragments rather than cleave the polymer. Interestingly, P. amylolyticus 5mez is the only strain lacking any lignin-related AA domains. Complete lignin degradation likely requires interaction with other microorganisms, such as ligninolytic fungi or bacteria with robust peroxidases. In general, even if present, bacterial lignin breakdown is typically more limited and slower than fungal [121].

CAZymes – experimental data comparison

Rowińska et al. performed experimental analyses concerning the enzymatic potential of the query strains [3]. Aside from zielonkawy, the genomic predictions suggested a descending order of hydrolytic capacity: 5mez, followed by Bac3/Bac2/Bac8, followed by 7psych.

The genome of P. amylolyticus 5mez encodes multiple processive cellobiohydrolases and abundant cellulose-binding modules, so P. amylolyticus 5mez was expected to be the most potent cellulase producer. Based on family diversity, we would expect cellulase activity in the order 5mez, followed by Bac8, followed by Bac2/Bac3, followed by 7psych. Based on domain count: 5mez, followed by Bac2/Bac8, followed by Bac3, followed by 7psych. Rowińska's assays, however, demonstrated B. subtilis (WK-15 = B. subtilis s.s. Bac3) as the most consistent cellulase performer across substrates and pH values, while P. amylolyticus (WK-9 = P. amylolyticus 5mez) and B. licheniformis (WK-20 = B. licheniformis Bac8) produced strong activity only under specific conditions. Pr. megaterium (WK-5 = Pr. megaterium 7psych) genome predicted minimal cellulose degradation, yet it produced an unexpected spike of cellulase activity on triticale at neutral pH [3]. These differences likely reflect gene regulation and substrate induction rather than gene content alone. The gene regulation aspect has been discussed in "Limitations of genomic predictions" section.

For hemicellulases, genomic data indicated the following patterns. Family diversity generally follows: 5mez followed by Bac2 followed by Bac3 followed by Bac8 followed by 7psych. Domain count in xylan GH shows 5mez followed by Bac2 followed by Bac3/Bac8 followed by 7psych (zero), while xylan CE domain count shows 5mez/7psych followed by Bac2 followed by Bac3 followed by Bac8. Experimentally, B. licheniformis Bac8 was the most robust xylanase producer (particularly under acidic conditions), and B. subtilis s.s. Bac3 showed moderate activity [3]. P. amylolyticus and B. velezensis exhibited little to no xylanase activity at pH 5.5 and only weak activity at pH 7.0, suggesting that their hemicellulase genes were poorly expressed under the tested conditions.

For pectin degradation, pectin GH domain count follows: 5mez followed by 7psych followed by Bac8, with the rest showing zero activity. Pectin lyase domain count shows: 5mez, followed by Bac8, followed by Bac2/Bac3, followed by 7psych. Pectin CE domain count follows: Bac8, followed by 5mez/Bac3, followed by 7psych, followed by Bac2 (zero). P. amylolyticus genome contains several pectin lyase families, while Bacillus genomes harbor fewer such modules. In the laboratory, pectinase activity was highest in B. subtilis and B. licheniformis, especially at neutral pH, whereas P. amylolyticus produced only low pectinase activity [3]. This may be linked to the pH optima of individual enzymes; pectate lyases from P. amylolyticus strain 27C64 are known to possess a highly alkaline optimum pH and may not be fully active under neutral or slightly acidic conditions [122].

For starch degradation, the GH domain count is: 5mez, then 7psych/Bac8, then Bac3, then Bac2. Gene predictions suggested substantial amylase capability in all strains. Experimentally, B. subtilis again topped the ranking with very high amylase activity across all substrates and pH values, followed by B. velezensis and B. licheniformis. P. amylolyticus and Pr. megaterium produced low amylase activity. Despite moderate enzyme activities, P. amylolyticus released the highest concentrations of reducing sugars from maize, while B. velezensis generated the most sugars from barley and triticale [3]. These results indicate that sugar release does not always correlate directly with individual enzyme assays; synergistic action and non-CAZy enzymes may contribute. All strains accelerated maize residue decomposition relative to the water control.

Rowińska's on-plate hydrolytic activity index assays, conducted at each strain's optimal growth temperature over seven days, showed no detectable ligninolytic activity in any of the isolates. However, the B. cereus strain (WK-21 = B. cereus s.s. zielonkawy) gave a strong positive reaction in an ABTS plate assay, turning the medium a greenish hue (data unpublished). The ABTS assay detects laccase activity; laccases oxidize the nonphenolic dye ABTS to a stable blue-green cation radical. This suggests that the strain may possess laccase-like enzymes involved in phenolic oxidation, even though it was excluded from further study due to pathogenicity.

Congruence of ANI-based divergence and functional traits

The observed variation in PGPT and CAZyme profiles across the six strains, particularly the functional similarity within the Bacillus subtilis sensu lato cluster (Bac2, Bac3, Bac8), led to an investigation into whether genomic relatedness constrains functional trait distributions. Comparative genomic studies have shown that no single PGPT gene is universal across all plant-growth-promoting bacteria; rather, different taxonomic groups harbour characteristic gene combinations that reflect lineage-specific evolutionary trajectories, with PGPT genes showing phylogenetic signal and being more likely to be conserved in closely related species [123]. For example, in Paenibacillus species, genes for IAA production, phosphate solubilization, and antimicrobial compounds are highly conserved across related strains, while biosynthetic gene clusters for secondary metabolites such as non-ribosomal peptides and polyketides show substantial variation and are frequently horizontally transferred [124]. It was already discussed that, at least within the Bacillus subtilis clade, even the profile of partial BGC deletions and frameshift mutations in biosynthetic genes is conserved phylogenetically [108]. Regarding CAZymes, at the highest taxonomic level, López-Mondéjar et al. demonstrate that these enzymes are phylogenetically conserved and differ even among microbial phyla [125]. When it comes to glycoside hydrolases (GHs), there are perceived to be conserved at the genus or species level [126, 127]. Of course, horizontal gene transfer (HGT) could cause many traits to be unrelated to vertical transmission; however, Martiny et al. point out that most traits tend to be at least somewhat conserved [126]. The effect of gene-environment interactions and biotic interactions is hard to assess.

ANI-based genomic divergence was used as a proxy for genomic relatedness because it serves as the primary species delimitation threshold and provides robust genome-wide similarity estimates [21, 22]. While ANI is not a phylogenetic or phylogenomic measure in the strict sense, it effectively quantifies taxonomic relatedness. Initial attempts with FastANI failed for inter-genus comparisons due to sequence similarity falling below the method's detection threshold. OrthoANI successfully quantified divergence across all pairwise comparisons, including inter-generic distances.

Euclidean distances were calculated on PGPT annotation counts (preserving annotation counts for overall profile similarity) and Jaccard dissimilarity on CAZyme family presence/absence (emphasizing enzymatic capability differences rather than copy number variation). Three-dimensional ordinations were necessary for ANI-based structure; lower-dimensional projections artificially collapsed distantly related taxa. For ordination comparisons, PGPT used PCA on count data, while both CAZymes and ANI-based distances used PCoA.

Mantel tests revealed significant correlations between ANI-based and functional distance matrices for both trait categories. The correlation was stronger for CAZymes (r = 0.800) than for PGPT (r = 0.703). This pattern suggests that CAZyme repertoires are more evolutionarily constrained than PGPT profiles, consistent with observations that carbohydrate-active enzymes show strong phylogenetic conservation across phyla [125] and with those showing glycoside hydrolase conservation at the genus or species level [127].

Procrustes analysis confirmed significant congruence between the two datasets. PGPT showed moderate topological preservation (51% of structure maintained), while CAZyme functional profiles presented lower preservation (22%) despite stronger distance correlation. This difference reflects distinct aspects of congruence – distance proportionality versus spatial positioning – and suggests that while CAZyme functional distances scale predictably with genomic divergence, the specific spatial arrangement differs more between genomic and CAZyme functional spaces.

The most noticeable convergence across results was observed for the Bacillus subtilis s.l. cluster (Bac2, Bac3, Bac8), which was therefore selected for cluster coherence testing. The same cluster was studied concerning evolutionary preservation of partial BGCs [108]. Within-cluster versus between-cluster distance comparisons were performed as a descriptive statistic to assess cluster coherence. Within/between ratios below 1.0 for ANI-based (0.77), PGPT (0.23), and CAZyme (0.60) distances confirmed that the Bac2/Bac3/Bac8 cluster exhibits consistent internal coherence. The functional ratios were even lower than the genomic ratio, indicating even tighter clustering in functional space.

These results demonstrate substantial ANI-functional congruence. The remaining variation likely reflects ecological adaptation, horizontal gene transfer, and adaptive evolution. The tight clustering of B. subtilis s.l. strains in both genomic and functional spaces confirms that closely related taxa share similar functional repertoires. The remaining taxa (P. amylolyticus 5mez, Pr. megaterium 7psych, B. cereus s.s. zielonkawy) were more isolated in ordination space, making within-cluster coherence testing less informative for these outliers.

Safety considerations

Virulence genes

Genomic analysis revealed virulence genes in only two of the six tested strains. The B. cereus s.s. zielonkawy strain, classified within the Bacillus cereus s.l. clade demonstrated an extensive toxigenic profile encompassing five independent toxin systems: ALO cytolysin, cytotoxin K, a complete hemolysin BL (HBL) system, a complete non-hemolytic enterotoxin (Nhe) system, and sphingomyelinase. Such a toxigenic profile, even without cereulide, classifies this strain as potentially hazardous to human and animal health [38]. One has to remember, though, that the actual toxigenic phenotype depends on gene expression regulation, which is proven to be affected by various environmental factors, as discussed in detail in "Limitations of genomic predictions" section [128].

The B. subtilis s.s. Bac3 strain contained only the bslA/yuaB gene encoding hydrophobin BslA. It is classified as a virulence factor in the VFDB database, but this protein exhibits no direct toxicity. BslA functions as a structural component of the biofilm matrix, forming a hydrophobic surface layer [129]. In the context of virulence, BslA presence alone does not constitute a significant pathogenic risk factor and should not disqualify the strain from potential biotechnological applications.

Antibiotic resistance genes (ARGs)

ARGs were detected in four of the six tested strains, with significant variation in the identified determinants. Antimicrobial resistance (AMR) poses a major threat to human health around the world [130]. In the studied strains, diverse resistance mechanisms were identified, whose clinical and epidemiological significance varies substantially.

PhLOPS_A phenotype

The most concerning is the cfr-like/clbA gene (23S rRNA methyltransferase, detected in strain B. velezensis Bac2), which confers multidrug resistance via the PhLOPS_A phenotype – resistance to phenicols, lincosamides, oxazolidinones, pleuromutilins, and streptogramin A. It is increasingly common in Gram-positive pathogens [131]. Its location in B. velezensis Bac2 has been predicted to be in the chromosome. Cfr is not only an antibiotic resistance enzyme that inhibits five clinically important antibiotic classes, but it is also genetically mobile and has a minimal fitness cost [131]. Some of those antibiotics are last-resort antibiotics [132, 133]. Over a decade ago, it was detected that the order Caryophanales naturally harbours functional homologs of the worrisome cfr ARG in their chromosomes [134]. Those cfr-like genes and Cfr-like proteins were proven to be functionally capable of decreasing susceptibility to five classes of antibiotics when heterologously expressed in E. coli [134]. It has been described in B. amyloliquefaciens, B. clausii, Brevibacillus brevis, and later in Paenibacillus sp. Y412MC10 [135]. They can provide antibiotic resistance, and some cfr-like genes have been found in pathogenic strains [135]. Moreover, there are rising suggestions that these genes may have been inserted into the chromosomes via transposons or some other mechanism of insertion and therefore possess the capability to easily spread further [135, 136]. What is especially concerning is that Bacillus sp. containing cfr/cfr-like genes in the plasmids has been described [137, 138]. Sun et al. recently identified Lysinibacillus sp. strain containing the cfr/clbA gene on its chromosome, and showed that this strain is phenotypically clindamycin-resistant [139].

Vancomycin resistance

The presence of vanZ, a part of the VanA-type gene cluster involved in vancomycin resistance, in the B. cereus s.s. zielonkawy chromosome could be seen as problematic. Vancomycin is yet another last-resort antibiotic. However, vanZ is only an auxiliary protein, which seems to be dispensable for vancomycin resistance [140]. It is obviously not harmless, as it is required for conferring resistance to teicoplanin, as well as dalbavancin and oritavancin [140, 141]. Sun et al. have shown that B. anthracis, B. paranthracis, and B. cereus strains isolated from biogas digestates and possessing this gene in their chromosomes were shown to be vancomycin-susceptible [139].

β-lactam resistance

The β-lactam antibiotics are the most frequently prescribed antibacterial agents, as they are safe, effective, and have a broad spectrum of activity [142, 143]. The four most common β-lactam antibiotics are penicillins, cephalosporins, carbapenems, and monobactams [144]. Penicillins are the most widely used antibacterials in both community consumption and the hospital sector in the European Union [145]. The major resistance method is β-lactamases, which initially emerged from environmental sources, most likely to protect producing bacteria from attack by naturally occurring β-lactams [146]. β-lactam-resistance genes have been found in B. licheniformis Bac8 (blaP) and B. cereus s.s. zielonkawy (bla1/BcI, bla2/BCII) chromosomes. blaP is a class A β-lactamase-encoding gene commonly detected in B. licheniformis and B. paralicheniformis [147, 148]. bla1/BcI (class A), bla2/BcII (class B) β-lactamase genes have been predominantly described within Bacillus cereus s.l. taxonomic group [144, 149, 150]. blaP-containing B. paralicheniformis strain, as well as bla1- and bla2-containing B. anthracis, B. paranthracis, and B. cereus strains, all isolated from biogas digestates, were proven to be resistant to ampicillin and ceftazidime [139]. All 234 Bacillus cereus group strains isolated from food products in Southern China, as well as all 12 strains isolated from an Algerian dairy plant, and all 49 strains isolated from ready-to-eat (RTE) food in Poland, possessed at least one β-lactamase-encoding gene [151–153]. Moreover, among all B. cereus isolates from food products in Southern China, 98.5% were resistant to penicillin and 98.9% to ampicillin [153].

Tetracycline resistance

Tetracycline-class antibiotics have a broad spectrum of antibacterial activity and many clinical applications; however, their utility has declined over time due to the emergence of antibiotic resistance [154–157]. Tetracyclines have continued to be used in livestock, agriculture, aquaculture, and animal and human medicine [157]. Most commonly, tetracycline resistance is based on efflux pumps and ribosomal protection proteins. In Bacillus spp. at least five types of efflux pumps linked to tetracycline resistance have been described, including proteins encoded by tet(K), tet(L), tet(39), tet(42), and tet(45) [156, 157]. The most common anti-tetracycline efflux pumps in Gram-positive bacteria are Tet(K) and Tet(L) [154]. In our strains, tet(L) has been detected in strain B. subtilis s.s. Bac3, and tet(45) – in strain B. cereus s.s. zielonkawy. Similarly to our study, Sun et al. have found tet(L) in B. subtilis strain, as well as in H. oleronia [139]. In both cases, tet(L) genotypes were linked to tetracycline-resistance phenotype. Ma et al. have studied various Lactocaseibacillus spp. and Lactobacillus spp., finding, among others, that tet(L) and tet(45) are key resistance genes for the tetracycline-resistance phenotype [158]. tet(L) in B. subtilis s.s. Bac3 is predicted to be located in the chromosome; however, in the case of B. cereus s.s. zielonkawy, tet(45) is located in the plasmid, which is particularly concerning.

Rifampin resistance

B. velezensis Bac2, B. subtilis s.s. Bac3 and B. licheniformis Bac8 possess rphC or rphB/rphC genes in their chromosomes. rph is a gene encoding rifampin phosphotransferase (RPH) [159]. Rifampin is the most important tuberculosis drug [160]. RPH orthologs are widespread in rifampin-sensitive bacteria, such as Bacilli, including various Bacillaceae, for instance, B. cereus, B. anthracis, and Bacillus sp. FW1 and Paenibacillaceae, such as Paenibacillus sp. LC231 and Brevibacillus brevis VM4 [161, 162]. Pawlowski et al. have shown that heterologously expressed Paenibacillaceae rphC gene in E. coli results in rifampin resistance [163]. On the other hand, Bacillus sp. FW1 showed no rifampin resistance despite carrying the rphC gene [164]. Recently, Sun et al. have isolated rphC-containing B. paralicheniformis, Heyndrickxia oleronia, and B. subtilis. All those strains were susceptible to rifampin in experimental validation [139].

Fosfomycin resistance

Fosfomycin is a broad-spectrum antibiotic that is back due to the increasing prevalence of antibiotic resistance [165]. It has re-emerged for the treatment of e.g., methicillin-resistant Staphylococcus aureus, vancomycin-resistant Enterococcus, and extended-spectrum of β-lactamase-producing bacteria [165, 166]. Currently, fosfomycin resistance is perceived as low to moderate, yet there are significant discrepancies between in vitro and in vivo studies [165, 166]. The emergence and spread of fosfomycin resistance, primarily through fosfomycin-modifying enzymes (such as fosA and fosB), poses a critical threat to global public health. fosA is predominantly found in Gram-negative bacteria, and fosB is found in Gram-positive bacteria. fosB may be encoded either in plasmids (Staphylococcus spp., Enterococcus spp.) or in chromosomes (B. subtilis) [165, 166]. However, there have been reports on the chromosome-encoded fosB gene in Staphylococcus aureus MRSA, which acquired fosfomycin resistance due to the overexpression of that gene [167]. In our study, we have found the chromosome-encoded fosB gene in strain B. cereus s.s. zielonkawy. It seems that the fosB gene is widely spread in environmental and industrial B. cereus s.l. group strains. For instance, fosB has been found in all 234 strains isolated from food products in Southern China, in all 12 strains isolated from an Algerian dairy plant, and in 32 out of 49 strains isolated from RTE food in Poland [151–153]. The FosB protein from B. cereus has also been well-characterised [168].

Genotype–phenotype disassociations

Yet, the bare presence of ARGs is not a guarantee of a resistant phenotype, as genotype–phenotype disassociations are relatively common [169]. The phenomenon of silent ARGs has been described and studied in recent years, with multiple mechanisms underlying this silencing [170, 171]. Mutation-based silencing (SARM – silencing of antibiotic resistance by mutation) is a primary one. Frameshift mutations resulting from insertions or deletions, as well as missense mutations due to nucleotide substitutions, are examples of this mechanism [170]. Promoter region effects represent another possibility: deletions or mutations within the promoter region can lead to gene silencing [170]. Other factors include the insertion of mobile elements such as integrons, regulatory factors that repress gene expression, and epigenetic modifications [170, 172]. Epistatic interactions between resistance genes have also been described.

Furthermore, environmental and metabolic factors, including exposure to antibiotics or other metabolites, may alter the phenotype [169]. These mechanisms can produce resistant genotypes with susceptible phenotypes, or conversely, susceptible genotypes with resistant phenotypes, suggesting the involvement of additional unknown factors that are not always of genetic origin [169, 170]. Aside from simple resistance or susceptibility, intermediate resistance, variable resistance, and heteroresistance further complicate the landscape [171].

To fully assess antibiotic resistance, phenotypic testing must be performed while accounting for these various types of resistance. Culturing on antibiotic-containing media represents one testing approach. When Sun et al. tested Bacillus-related bacteria, they observed all four possible combinations: resistant genotype + resistant phenotype, resistant genotype + susceptible phenotype, susceptible genotype + susceptible phenotype, and susceptible genotype + resistant phenotype [139].

Heterologous protein expression may provide more precise testing of specific ARGs. However, it is a known fact that expression levels and resulting resistance can differ substantially between model organisms like E. coli and environmental strains, or between clinical and non-clinical isolates [173, 174]. It is well demonstrated by Chen et al., who proved that the bla1 and bla2 genes from B. anthracis Sterne encode functional β-lactamases. When these genes were heterologously expressed in E. coli and B. subtilis, they conferred ampicillin resistance. However, B. anthracis Sterne itself remained susceptible to these antibiotics, with too low gene expression level, despite harbouring functional ARGs [173, 174].

The question of how harmful environmental ARGs truly are remains complex. Some of these genes are the products of eons of evolution, while others likely emerged due to unprecedented anthropogenic antibiotic pressure [175, 176]. Several critical questions remain unresolved: To what extent can functional ARGs located in chromosomes, for instance, be transferred to clinical strains [134, 135, 176]? In our genomes, all but the tet(45) gene are located on chromosomes. What is the danger of silent ARGs in the clinical context [162, 171, 172]? Which genes should be classified as “truly” ARGs [177]? There is an ongoing discussion regarding those topics.

Study limitations

Limitations of genomic predictions

As with any study that relies predominantly on in silico approaches, the present work has limitations that should be considered when interpreting the results. The most fundamental of these is the well-recognised discrepancy between genotype and phenotype: the detection of a gene or a biosynthetic gene cluster (BGC) within a genome does not, by itself, guarantee the expression of the corresponding trait under ecologically relevant conditions. As demonstrated by Hone et al., key canonical genes such as the pqq operon and gcd did not reliably predict the phosphate solubilisation phenotype in plant-associated bacteria, underscoring that genomic presence alone is an insufficient proxy for functional capacity [178]. The genotype–phenotype dissociations in antibiotic resistance gene expression, discussed in detail above, extend analogously to all other functional categories analysed in this study – plant-growth-promoting traits, secondary metabolite biosynthesis, carbohydrate-active enzymes, virulence factors, and toxins.

The expression of predicted PGPTs is tightly coupled to specific environmental inducers, and the complexity of their interactions is high. Transcriptomic analysis of Burkholderia multivorans WS-FJ9 revealed that phosphate-solubilising activity is characterised by low-soluble-phosphate induction and high-soluble-phosphate inhibition, with 446 differentially expressed genes [179]. This challenges the assumption that the genomic detection of individual PGP traits can reliably predict ecological function. The condition-dependence of gene expression also applies to siderophore production. After bacteria accumulate a sufficient amount of iron, Fur – the ferric uptake regulator – acts either directly as a transcriptional repressor of Fe-uptake genes (and uses Fe²⁺ as a co-repressor) or indirectly as a protein regulatory activator [45]. Moreover, to produce bacillibactin, detected both in PGPT-Pred and antiSMASH analyses, besides the dhb operon, an Sfp phosphopantetheinyl transferase is necessary, and most laboratory strains of B. subtilis str. 168 are sfp⁰ mutants and are consequently unable to produce bacillibactin despite harbouring the intact biosynthetic operon [180]. These findings show that detecting siderophore-related loci in genomes does not equate to actual siderophore production, which depends on iron availability and the functional status of accessory genes.

The dbCAN-based annotation of carbohydrate-active enzymes (CAZymes) faces distinct challenges related to the inherent polyspecificity of CAZyme families. A single family can encompass enzymes acting on multiple substrates: GH5, for instance, contains experimentally characterised proteins with over 20 different EC numbers targeting more than 10 different carbohydrate substrates [181]. For this reason, in our study, neither GH5 nor GH8 was assigned to a single-substrate category; instead, they were classified as multi-substrate families. CAZyme expression is highly context-dependent. This is especially well described for industrial fungi. Gruben et al. demonstrated that gene activation depends on the type of plant biomass substrate, with highly complex patterns of CAZyme expression; many of those genes were under the control of a few regulators [182]. Llanos et al. have highlighted the substantial plasticity of transcriptional responses to changes in nutrient status, with highly diverse, at times counterintuitive, transcriptional patterns [183].

The gap between genomic prediction and realised phenotype is even more complicated by the multi-layered character of bacterial gene regulation, where transcriptional, post-transcriptional, and translational mechanisms are intrinsically intertwined and integrate multiple environmental signals simultaneously [184]. At the transcriptional level alone, promoter activity is shaped by competition among sigma (σ) factors for a limited pool of core RNA polymerase – a process in which the relative concentration of active alternative σ factors shifts in response to specific environmental inducers, redirecting transcription towards condition-dependent regulons. The availability of alternative σ factors is itself tightly controlled post-translationally by anti-σ factors [184]. Notably, bacteria inhabiting heterogeneous environments such as soil – the six Caryophanales strains analysed here likewise – encode a considerably larger set of transcriptional regulators, two-component systems, and alternative σ factors than organisms occupying more stable niches; B. subtilis alone encodes 19 σ factors including seven of the extracytoplasmic function (ECF) subfamily [185]. Beyond σ factor competition, the transcription initiation frequency of most bacterial genes is modulated by DNA-binding transcription factors (TFs) whose affinity for target sequences is frequently modulated by reversible interaction with small-molecule effectors, phosphorylation cascades, or redox modifications, making their regulatory output inherently condition-dependent. Individual TFs may act as activators at one promoter and repressors at another, or switch function depending on the bound ligand [184]. Additional genome-wide modulation is exerted by nucleoid-associated proteins (NAPs) that combine structural and gene-silencing roles, and by mutations or sequence variation within promoter elements – including variable-length nucleotide repeats near the –35 region – that can drastically alter transcription initiation frequency [184]. None of these regulatory inputs is captured by annotation pipelines operating on genomic sequence alone.

Post-transcriptional layers add further regulatory dimensions that are invisible to in silico prediction. mRNA stability, which directly determines transcript abundance, is modulated by, among others, RNA-binding proteins (RBPs) that compete with ribonucleases for overlapping binding sites, and by environmental factors such as temperature, pH, and salt concentration that alter mRNA secondary structure [184]. Riboswitches function as metabolite sensors – responding to amino acids, vitamins, and ions – and control downstream expression by switching between structures that either sequester or expose the ribosome-binding site, or that promote versus prevent premature transcription termination [184]. Trans-acting small regulatory RNAs (sRNAs), which share only limited complementarity with their targets and thus may regulate multiple mRNAs simultaneously, add yet another condition-responsive layer that modulates translation efficiency and mRNA decay in response to environmental cues, including pH, temperature, and nutrient depletion [184]. Critically, even after decades of investigation, the regulatory logic governing approximately half of all genes remains unknown in Escherichia coli, the best-studied bacterial model; the situation is considerably less resolved in B. subtilis and related soil organisms [184]. Taken together, these overlapping and intertwined regulatory mechanisms mean that the functional potential predicted from genomic sequence represents an upper theoretical boundary that is unlikely to be fully realised under any single set of environmental conditions.

Then, there is a continuous problem of orphan, cryptic, and silent BGCs [186–189]. A number of BGCs are not expressed at all or are expressed at a weak level. In fact, silent and cryptic BGCs outnumber those constitutively expressed by 5–10 times [189]. These BGCs are still highly valuable, as there are multiple methods to activate them and collect potentially unknown secondary metabolites with desired properties [186, 187, 189]. However, in native organisms, those BGCs will not result in any product, even if predicted with software such as antiSMASH, and the phenotype would differ from the genotype. Moreover, as in our study, multiple detected BGCs cannot be reliably assigned to a known compound, which makes it impossible to determine the genotype. However, this may also indicate the potential to obtain novel secondary metabolites.

This does not mean that the genomic predictions are of low value. These in silico analyses should be treated as tools for discovering the general putative potential of microorganisms. The predicted genes may or may not be expressed, but the hypothetical potential remains. As the silent/cryptic BCGs demonstrate, those unexpressed genes or clusters can be ‘hacked’. There is also, however, the possibility of phenotypic capabilities beyond the predicted genotype, with unknown mechanisms, as approximately one-third of protein-encoding genes lack functional annotation [188]. Hoskisson et al. call them ‘Unknown Unknowns, where the real hidden treasures likely lie’ [188].

Statistical challenges

The six strains analysed in this study were selected for their biotechnological potential from a larger screening programme. While this number is a practical outcome of strain discovery and represents a meaningful set from an application perspective, n = 6 imposes substantial constraints on inferential statistical analyses.

Functional differentiation was explored primarily through descriptive approaches. PCA on level-3 PGPT-Pred annotations produced a high-dimension, low-sample-size (HDLSS) configuration (p = 43 functional categories, n = 6 strains). Pre-analysis diagnostics designed for large-sample contexts – Bartlett's test of sphericity and Kaiser–Meyer–Olkin measure – lacked statistical power at this sample size, where their diagnostic properties deteriorate rapidly below approximately n ≈ 20 [14]. PCA robustness was therefore validated using Noise-Reduction Methodology (NRM), which provides mathematically consistent eigenvalue and eigenvector estimates under HDLSS asymptotics [15]. The high agreement between classical PCA and NRM-PCA confirmed that the ordination captured genuine structure rather than sampling noise.

Nevertheless, n = 6 effectively precluded several standard multivariate approaches. Formal cluster validation – such as testing whether the three Bacillus subtilis sensu lato strains (Bac2, Bac3, Bac8) form a statistically distinct group – was not possible: with only C(6,3) = 20 possible three-genome groupings, permutation-based tests such as PERMANOVA lack sufficient permutational diversity to generate robust p-values. Meaningful group-level inference (e.g., differential analysis) would require substantially larger strain panels per taxonomic group. The Mantel and Procrustes tests applied to assess congruence between ANI-based genomic relatedness and functional profiles provided sufficient permutational space for significance testing. However, the subsequent within-cluster versus between-cluster distance ratios for the B. subtilis s.l. clade were reported as descriptive statistics only.

We note that this situation – a small set of genomically characterised strains examined for functional potential – is not uncommon in applied microbiology, where strain collections are assembled based on phenotypic screening rather than statistical design. We approached the analysis as a semi-exploratory framework: using PCA and taxonomic congruence tests to identify patterns and generate hypotheses, while being transparent about the boundaries of what n = 6 can support inferentially. We hope that the methodological solutions adopted here – particularly the HDLSS validation via NRM and the combined Mantel/Procrustes approach for ANI-based signal – may serve as a practical reference for researchers facing similar constraints.

Conclusions

Plant-growth-promoting-traits genome prediction revealed that all strains possess multiple PGPT genes and annotations, and therefore, predicted clear potential. Pr. megaterium 7psych possessed the highest number of genes and annotations, while B. velezensis Bac2, B. subtilis s.s. Bac3 and B. licheniformis Bac8 shared the highest PGPT gene percentage. PGPT PCA analysis revealed profile differentiation, with the biggest similarities within Bacillus subtilis sensu lato, and the highest differences between Pr. megaterium 7psych, P. amylolyticus 5mez, and the rest of the strains, while B. cereus s.s. zielonkawy being in the middle, confirming that phylogenetic relationship affects PGPT profile. PGPT genome prediction suggests strain potential involvement in nitrogen, phosphorus, potassium, and iron acquisition as well as abiotic and biotic stress neutralisation.

The number of antiSMASH-predicted BGCs was the highest in P. amylolyticus 5mez and B. velezensis Bac2, but the highest number of known compounds was identified in B. subtilis s.s. Bac3, followed by B. velezensis Bac2. Species outside the genus Bacillus had the highest percentage of unknown BCGs, suggesting less-studied biosynthetic potential. Fungicidal compounds detected by antiSMASH were a better prediction of published fungicidal activity in vitro than cumulative PGPT-Pred annotations, which often identified incomplete gene clusters insufficient for functional metabolite synthesis.

Carbohydrate-active enzyme profiling was dominated by P. amylolyticus 5mez, which possessed the highest number of CAZyme genes, the highest proportion of CAZyme genes, and the highest number of annotations. These genomic predictions require systematic phenotypic validation across diverse substrate types and environmental conditions to establish reliable genotype–phenotype correlations. B. cereus s.s. zielonkawy, on the other hand, had the fewest CAZyme annotations yet was the strain with the highest ligninolytic potential, as supported by preliminary data. The highest between-strain CAZyme profile similarities were once again detected within Bacillus subtilis s.l.

Virulence profile indicates that B. cereus s.s. zielonkawy may pose a serious threat to human and animal health due to multiple toxin systems, and therefore must be excluded from any applied studies. It is the only strain presenting toxigenic potential.

Antibiotic resistance genes were detected in four of six strains. The only ARG-free strains were P. amylolyticus 5mez and Pr. megaterium 7psych. Among applicable strains (excluding B. cereus s.s. zielonkawy), several ARGs raise concern: the cfr-like gene (PhLOPS_A phenotype) in B. velezensis Bac2, β-lactamase in B. licheniformis Bac8, tet(L) in B. subtilis s.s. Bac3, and rifampin phosphotransferase genes (rphB/C) in all three Bacillus strains. Some of these genes are naturally occurring in Caryophanales chromosomes and may exhibit insufficient expression levels to confer resistant phenotypes, and genotype–phenotype discrepancies are well documented; therefore, genomic detection alone does not reliably predict actual resistance, and phenotypic testing remains essential.

The integrated genomic framework applied here – combining functional trait prediction, secondary metabolite profiling, CAZyme annotation, and safety screening – offers a replicable approach for the in silico pre-selection of microbial candidates intended for agricultural bioformulations, provided the inherent limitations of sequence-based prediction are acknowledged. The small sample size and high-dimensional nature of multi-trait genomic datasets pose statistical challenges that were addressed through tailored approaches, including NRM-validated PCA and distance-based congruence testing; these methods may serve as a practical reference for similar studies operating under comparable constraints.

In summary, the genomic analysis of environmental isolates used in our research is a valuable tool for their selection for biotechnological aims such as the production of biopreparations. Tested strains (excluding B. cereus s.s. zielonkawy) demonstrate high potential for agricultural applications, with P. amylolyticus 5mez and Pr. megaterium 7psych presenting the most favourable safety profiles due to the absence of detectable ARGs and virulence factors. Phenotypic validation of antibiotic resistance, plant-growth-promoting activities, and cellulolytic properties under field-relevant conditions is recommended before practical implementations.

Authors’ contributions

Conceptualization, T.G. and J.S.; Methodology, T.G.; Formal analysis, T.G.; Investigation, T.G.; Resources, J.S.; Data curation, T.G.; Writing – original draft preparation, T.G. and J.S.; Writing – review and editing, J.S. and T.G.; Visualization, T.G.; Supervision, J.S.; Project administration, JS.; Funding acquisition, J.S.

Funding

This work was supported by the Agency for Restructuring and Modernisation of Agriculture, Poland [grant number 00077.DDD.6509.000167.2022.05].

Data availability

All analysed nalysed genomes are available in the NCBI database under BioProject PRJNA1129008 with the following accession numbers: GCF_040513715.1 (Paenibacillus amylolyticus 5mez), GCF_040513995.1 (Priestia megaterium 7psych), GCF_040513685.1 (Bacillus velezensis Bac2), GCF_040513675.1 (Bacillus subtilis s.s. Bac3), GCF_040513645.1 (Bacillus licheniformis Bac8), and GCF_043099295.1 (Bacillus cereus s.s. zielonkawy). All other data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.