The Bacillus spp. strains were routinely grown in Lysogeny Broth (LB) and incubated at 37°C overnight while shaking at 200 rpm. Avian pathogenic Escherichia coli (APEC) serotypes O2, O18, O78, and Clostridium perfringens NAH 1314-JP1011 were obtained from the Elanco pathogen library. Salmonella enterica serovar Typhimurium ATCC 14028 was purchased from the American Type Culture Collection (ATCC, Manassas, VA, United States). E. coli strains and S. Typhimurium, were routinely grown in LB, and C. perfringens was grown in anaerobic Brain Heart Infusion (BHI) broth supplemented with yeast extract (5.0 g/L) and L-cysteine (0.5 g/L). For growth in liquid culture, a colony from the respective agar plate was inoculated into a 10 mL tube containing liquid media and the tube was incubated in a shaker incubator at 37°C and 200 rpm for E. coli and S. Typhimurium, and statically at 39°C for C. perfringens inside a Bactron anaerobic chamber (Sheldon Manufacturing, Inc., Cornelius, OR, United States). The anaerobic chamber contained a mixture of N₂:CO₂:H₂ (87.5:10:2.5, v/v/v).

Vero cells were obtained from Elanco cell culture collection and were maintained in Opti-MEM^® I reduced serum media containing 5% Fetal Bovine Serum (FBS) (Cytiva, Marlborough, MA, United States) and Gentamicin (Opti-5-Gent) (Life Technologies, Carlsbad, CA, United States). The serum-free cell culture medium was similarly prepared with Minimal Essential Medium with Earle’s Balanced Salt Solution (MEM/EBSS), 10% fetal bovine serum (FBS), 1% non-essential amino acids and 1% L-glutamine in place of FBS. To generate wells containing 100% confluent cells for the cytotoxicity assay, Vero cells grown for 2–3 days were divided into a 96-well flat bottom tissue culture plate (Fisher Scientific, Waltham, MA, United States) where each well contained 1 × 10⁴ cells. The cells were then incubated on the plate for 48–72 h inside a CO₂ incubator (37^oC; %CO₂ was maintained at 5 ± 1%).

Eight Bacillus spp. strains were screened for their antimicrobial activity against five microorganisms, namely APEC serotypes O2, O18, O78, Salmonella Typhimurium ATCC 14028, and C. perfringens NAH 1314-JP1011. The assays were modified from a protocol described in Latorre et al. (2016) and performed in duplicate. The detailed protocol is presented in Supplementary Materials.

The β-mannanase assay was adapted from a protocol as described by McCleary (1978). Assays for amylase and protease activities were done following protocols in Latorre et al. (2016). These protocols are presented in Supplementary Materials.

Cytotoxicity assays of eight Bacillus culture supernatants were performed following the protocol described in EFSA guidelines (EFSA Panel on Additives and Products or Substances used in Animal Feed [FEEDAP] et al., 2018). A detailed protocol is described in Supplementary Materials. Culture supernatant of B. cereus ATCC 14579 and B. licheniformis ATCC14580 were used as positive and negative controls, respectively.

Antibiotic susceptibility assays of Bacillus spp. for tetracycline, chloramphenicol, streptomycin, kanamycin, erythromycin, vancomycin, gentamycin, ampicillin, and clindamycin were performed and assessed according to an EFSA guideline for Antimicrobial resistance of the Bacillus spp. as direct fed microbials (DFM) (EFSA Panel on Additives and Products or Substances used in Animal Feed [FEEDAP] et al., 2018). Bacillus spp. strains on LB agar plates were sent to Microbial Research, Inc. (Fort Collins, CO, United States) for analysis following protocols in compliance with Clinical Laboratory Standard Institute (CLSI) document VET01 (World, 2021). Briefly, minimum inhibitory concentration (MIC) plates were prepared using cation-adjusted Mueller Hinton Broth (MHB) and the antimicrobials were twofold serially diluted to obtain a final concentration range of 0.06–32 μg/mL. Growth of Bacillus spp. in the presence of each of nine antimicrobials with different dilutions was monitored. Susceptibility was interpreted as the lack of Bacillus spp. growth in the presence of antimicrobial at a concentration that was lower that the cut-off values of the respective antimicrobials described in the EFSA guideline. For quality control, the following organisms were used as controls, E. coli ATCC 25922, Enterococcus faecalis ATCC 29212, Pseudomonas aeruginosa ATCC 27853 and Staphylococcus aureus ATCC 29213.

Taxonomic labeling of the assembled microbial genomes was carried out using CAMITAX (Bremges et al., 2020). CAMITAX is a scalable workflow that combines genome distance–, 16S ribosomal RNA gene–, and gene homology–based taxonomic assignments with phylogenetic placement. OrthoFinder v2.3.1 (Emms and Kelly, 2019) was used to determine orthologous relationships (Sayers et al., 2021).

Phylogenetic relationships of the genomes were explored with UBCG v3.0 using default settings (Na et al., 2018). This software tool employs a set of 92 single-copy core genes commonly present in all bacterial genomes. These genes then were aligned and concatenated within UBCG using default parameters. The estimation of robustness of the nodes is done through the gene support index (GSI), defined as the number of individual gene trees, out of the total genes used, that present the same node. A maximum-likelihood phylogenetic tree was inferred using FastTree v.2.1.10 with the GTR + CAT model (Price et al., 2010).

Bacillus amyloliquefaciens ATCC PTA-126784 and PTA-126785, and B. subtilis ATCC PTA-126786 strains were deposited in the ATCC culture collection (Manassas, VA, United States). For simplicity, Bacillus amyloliquefaciens ATCC PTA-126784 and PTA-126785, and B. subtilis ATCC PTA-126786 strains are referred to as Ba PTA84 and Ba PTA85, and Bs PTA86, respectively, throughout the manuscript.

Bacillus strains Bs PTA86, Ba PTA84, and Ba PTA85 were grown as three single strain cultures, and then were analyzed as a two-strain (Ba-PTA84 and PTA85) or three-strain (Bs-PTA86, Ba-PTA84, and Ba-PTA85) consortia in 5 mL of minimal or rich liquid media. For growth in minimal media, medium containing 1X M9 salts, and glucose at a final concentration of 0.5% (w/v) was used. Rich medium contained the following entities (g/L): peptone 30; sucrose 30; yeast extract 8; KH₂PO₄ 4; MgSO₄ 1; and MnSO₄ 0.025. The culture was grown at 37°C overnight. Bacillus cells were pelleted by centrifugation at 10,000 × g for 10 min, cell pellets were washed three times with ice-cold PBS. The resulting cell pellets and cell-free supernatants were stored at −80°C and sent to metabolon Inc. (Durham, NC, United States) for global untargeted metabolomic profiling. The experiment was performed in one biological replicate. Detailed description of metabolomic analysis is presented in Supplementary Materials.

Bacillus spp. strains were isolated from the cecal contents and fecal materials of healthy chickens. The taxonomic identities of the isolates were determined by 16S-rRNA amplicon sequencing. These isolates belonged to 30 different Bacillus species with the top hits of B. velezensis, B. amyloliquefaciens, Bacillus haynesii, Bacillus pumilus, B. subtilis, and B. licheniformis, Supplementary Table 2.

Due to safety considerations, Bacillus spp. isolates chosen for further screening included only those that belong to the species listed as DFMs in the Association of American Feed Control Officials, Inc. (AAFCO) Official Publication since they “were reviewed by FDA Center for Veterinary Medicine and found to present no safety concerns when used in direct-fed microbial products” (AAFCO, 2018), and to the species listed as Qualified Presumption of Safety (QPS) status according to the European Food Safety Authority (EFSA) BIOHAZ Panel (Koutsoumanis et al., 2020). These were B. subtilis, B. amyloliquefaciens, B. pumilus, and B. licheniformis.

Bacillus spp. strains were tested to determine their effect on selected microorganisms and their ability to secrete selected enzymes (Latorre et al., 2016). For the former, Gram-negative and Gram-positive microorganisms (E. coli O2, O18, O78, C. perfringens NAH 1314-JP1011, and S. enterica serovar Typhimurium ATCC 14028) were used. For the latter, plate-based assays for determining the secretion of amylase, protease, and β-mannanase were performed (Figure 1).

A total of 266 Bacillus strains were first screened against E. coli O2, and 71% of the strains showing positive E. coli O2 inhibition were selected for a second-round of assays targeting E. coli O18, then E. coli O78, S. Typhimurium and lastly C. perfringens JP1011. The top 8 Bacillus strain candidates were selected according to their cumulative inhibition scores, Supplementary Table 3. These included 6 B. amyloliquefaciens (Ba): Ba ELA006, Ba ELA071, Ba ELA151, Ba ELA175, Ba PTA84, Ba PTA85, and 2 B. subtilis (Bs) isolates; Bs ELA082 and Bs PTA86.

The cumulative inhibition score was calculated as the sum of the inhibition score values of a Bacillus strain against the five microorganisms tested. The data showed that six Ba strains had better cumulative microbial inhibition scores compared to two Bs strains. The average cumulative inhibition scores of 8.5 and 5.5 for six of Ba and two of Bs, respectively.

The eight Bacillus strain candidates were also evaluated for their ability to secrete enzymes. Bacillus strains are known to produce a variety of enzymes (Contesini et al., 2018; Danilova and Sharipova, 2020). In vitro plate-based assays for protease, amylase, and β-mannanase activities showed that six Ba and two Bs showed comparable amylase, protease, and β-mannanase activities, Supplementary Table 3 Ba ELA071 and Ba PTA85 showed the lowest and highest cumulative REA values of 4.73 and of 6.1, respectively.

To evaluate the safety of Bacillus spp. as microbial feed ingredients, the Bacillus candidates were tested for antimicrobial susceptibility to medically relevant antimicrobials.

The results from antimicrobial susceptibility tests for 8 Bacillus strains showed that all strains were susceptible to the tested antibiotics. The only exception was Bs ELA082 which exhibited a borderline resistance to streptomycin. The MIC of streptomycin for this strain was 16 μg/mL, which is twofold higher than the EFSA threshold cut-off of streptomycin for Bacillus as DFM, Figure 2A.

To determine the potential toxicity of Bacillus strains on the host cells, culture supernatants of Bacillus spp. were tested for cytotoxicity toward Vero cells according to EFSA Panel on Additives and Products or Substances used in Animal Feed [FEEDAP] et al. (2018). The cytotoxicity assay was performed by monitoring the lactate dehydrogenase (LDH) enzyme originated from compromised Vero cells as described in Fagerlund et al. (2008). Figure 2B shows the cytotoxicity levels of each Bacillus strain tested. The results suggested that all 8 Bacillus strains were non-cytotoxic with toxicity levels far below 20%, a percentage that is considered cytotoxic according to the EFSA guidelines. Of all isolates, the cytotoxicity level of Ba PTA84 was closest to that of the negative control. In general, Ba strains exhibited toxicity levels that were lower than those of Bs strains. Within the B. subtilis group, the cytotoxicity level of Bs PTA86 was the lowest, 5%.

Based on their performance on microorganism inhibition, enzymatic activities, antimicrobial susceptibility, and low toxicity against Vero cells, the strains Ba PTA84, Ba PTA85, and Bs PTA86 were chosen for more detailed characterization employing genomic and metabolomic approaches described in the following sections.

Untargeted metabolomics analysis of cell pellets and culture supernatants of the three candidate strains was performed to assess metabolite profiles of each strain and strains-in consortia when grown in two different media. Cells were cultured in both rich and minimal media as individual strains, as well as a consortium of Ba PTA84 and Ba PTA85, and a consortium including all three strains. In total, 436 and 457 named metabolites were identified in the supernatant and pellet samples, respectively (Supplementary Tables 4–7).

A principal component analysis (PCA) of the normalized metabolite abundances showed a clear separation between samples from rich and minimal media along the first principal component (∼70% explained variance), both in the supernatant and pellet (Figures 3A,B). Additionally, the separation between samples along the two first principal components suggested that under the tested growth conditions the strains and consortia differ in their profiles of secreted/consumed metabolites as well as their cell-pellet small-molecule content.

Interestingly, looking at the metabolite profiles in the culture supernatant (Figure 3A), all strains and strain combinations tightly clustered in the rich media samples while they separated under minimal media conditions. Indeed, looking at the number of secreted metabolites under each condition (abundance > 1.5-fold above media controls, Figure 3C), cells in all cultures secreted approximately 100 named metabolites in rich media with a median Jaccard similarity of 0.8, compared to a range of 134–250 secreted metabolites in minimal media with only 0.54 median overlap (Mann–Whitney p-value = 0.01). This shows that especially under minimal media conditions, each strain or consortia secretes distinct sets of small molecules.

In rich media, Ba PTA84 secreted the largest number of metabolites (104 metabolites) and it had the fewest uniquely secreted metabolites (abundance > 1.5-fold higher than other individual strains) across both media. Thirteen unique metabolites related to amino acid, central carbon, nucleotide, and lipid metabolisms as well as some xenobiotic compounds (i.e., quinate, 4-hydroxybenzyl alcohol, and 3-dehydroshikimate) were detected.

As expected, the three-strain consortium was found to have the largest number of secreted metabolites in minimal media, a total of 250 metabolites. These include potential host beneficial metabolites such as betaine, kynurenine, indolactate, tyrosol, citrulline, tricarballylate, vitamins B5 and B6, hippurate, and kestose. Of all three single strains, the culture supernatant of Bs PTA86 carried the highest number of metabolites in minimal media (219 metabolites), followed by Ba PTA85 and Ba PTA84 with 195 and 130 metabolites, respectively. Ba PTA85 had the greatest number of unique secreted metabolites, a total 78 metabolites. The larger number of secreted metabolites under minimal media conditions was partly due to a higher number of amino acid metabolism intermediates, followed by nucleotide and carbohydrate metabolites (Figure 3D). Thus, our results suggest that the different strains may use different metabolic strategies to synthesize amino acids/proteins, nucleotide, and carbohydrate molecules given a limited nutrient availability.

The genomes of Ba PTA84, Ba PTA85, and Bs PTA86 were sequenced by PacBio sequencing. Assembly of Ba PTA84 and Bs PTA86 genomes yielded 1 contig each while the Ba PTA85 assembly contained 2 contigs – a large 4,084,681 bp contig and a smaller 231,132 bp long contig. The genome properties and annotation of different features are summarized in Table 1. The whole-genome sequences were deposited at DDBJ/ENA/GenBank under BioProject numbers PRJNA701126 and PRJNA701127.

Ortholog analysis was performed to identify paralogous and/or orthologous relationships between the genomes of Ba PTA84, Ba PTA85, and Bs PTA86. Ba PTA84 shared the highest number of genes with 99.4% gene presentation in orthogroups while Ba PTA85, and Bs PTA86 shared 93.6% and 90.1%, respectively (Supplementary Table 8). A total of 3,024 orthologs were shared among all three strains while 586 orthologs were shared only between Ba PTA84 and Ba PTA85, 60 orthologs between Ba PTA84 and Bs PTA86, and 34 orthologs were shared only between Ba PTA85, and Bs PTA86 (Figures 4A,B).

Phylogenetic relationships of the three genomes were explored with UBCG v3.0 which employs a set of 92 single-copy core genes commonly present in all bacterial genomes. Ba PTA84, Ba PTA85 and Bs PTA86 genomes were compared against the genomes of B. amyloliquefaciens, B. velezensis, and B. subtilis strains along with Lactobacillus reuteri as an outgroup (Accession numbers: AL009126, CP000560, CP002627, CP002634, CP002927, HE617159, HG514499, JMEF01000001, CP005997, CP009748, CP009749, CP011115, LHCC01000001, CP014471, and QVMX01000001). As shown in Figure 4, both strains Ba PTA84 and Ba PTA85 showed closest relationship to B. amyloliquefaciens B4 while Bs PTA86 showed closest relationship to B. subtilis subsp. subtilis 168.

The assembled genome sequences of 3 Bacillus strains were annotated for the following potential probiotic properties, enzymes, antioxidants, bacteriocins, and secondary metabolites, and for the presence of genes of potential safety concerns such as genes encoding toxins, virulence factors, and antimicrobial resistance genes. A detailed description of each of the above-mentioned features is described below.

All three strains were scanned for the presence of mobile genetic elements such as prophages, insertion sequences (IS) and transposases. Both Ba PTA84 and Ba PTA85 have 3 transposases while Bs PTA86 has 4 transposases. Ba PTA84, Ba PTA85 and Bs PTA86 share 2 copies of IS21 insertion sequence.

As a proof of principle, Ba PTA84 was selected in an in vivo pilot study as a DFM to evaluate its probiotic efficacy in supporting improved broiler growth performance. To accomplish this, 2,500 1-day old broiler chickens (Cobb 500) were randomly assigned to 50 pens of 50 birds each and split between two treatment groups; an untreated Control group and a group receiving 1.5 × 10⁵ CFU of Ba PTA84 per gram of feed for the full 42-day production period. Despite similar feed intake, birds fed the DFM had 3.5% higher final body weight compared to the control (2.16 vs. 2.23 kg for Control and DFM, respectively; p = 0.0018). This translated to a 3.3% improvement in feed conversion ratio (FCR) for the DFM-fed group compared to Control (1.50 vs. 1.45 for Control and DFM, respectively; p = 0.011) and a 6.2% increase in the European Broiler Index (EBI), a metric of overall production efficiency (337 vs. 358 for Control and DFM, respectively; p < 0.0001), Figure 6. These results indicated that use of Ba PTA84 as a DFM can significantly improve weight gain and feed efficiency in broiler chickens.

To gather insights into the in vivo effects of in-feed administration of B. amyloliquefaciens on broiler chickens, we analyzed the cecal microbiomes of animals in the control and treatment groups of our clinical study. Samples from twenty 42-day old animals from each treatment group were used to build 16S rRNA amplicon libraries for sequencing, corresponding to one randomly chosen animal per pen. The median coverage was ∼36,000 read pairs per sample, and 1,945 ASVs were identified across samples.

Samples in the control and treatment groups showed similar values for ASV richness and diversity (Figures 7A,B, p-values = 0.07, 0.44–0.36, respectively). Additionally, ANOSIM and PERMANOVA analyses based on the Bray–Curtis dissimilarity between samples did not support significant differences in community structures between treatment groups (ANOSIM p-value = 0.66. PERMANOVA p-value = 0.44). In addition, using ANCOM, we did not observe differentially abundant ASVs between treatment groups, including those classified to the genera of pathogens used for in vitro screening (Mandal et al., 2015). These results, which are evident by the lack of clustering of samples by treatment following principal components analysis (Figure 7C), indicate that supplementation of Ba PTA84 at a dose sufficient to induce a positive effect on performance (Figure 7) had minor effects on the diversity and structure of the cecal microbiome.

We expected that host-adapted Bacillus strains to exert better probiotic effects in the host environment than those isolated from other sources, thus, we targeted our isolation to those Bacillus spp. from animal GIT content or fecal samples of healthy animals (Barbosa et al., 2005). A higher diversity of isolates was obtained from the ethanol-treated samples compared to heat-treated samples, as reported previously (Koransky et al., 1978; Barbosa et al., 2005). Despite the general heat resistance feature of Bacillus spores, spore core, cortex, coat, and membrane composition determines the degree of the spores’ heat resistance (Setlow, 2006; Brul et al., 2011; Berendsen et al., 2015) resulting in different responses of spores toward heat stresses.

With the continuing reduction in use of antibiotics in poultry farms, as driven by regulations, and some customer preferences, the development of microbial feed additives that support maintenance of poultry health in the face of undesirable organisms would be beneficial. Our screening results showed that Bacillus spp. controlled the growth of undesirable E. coli O2, O18, and O78, C. perfringens- and Salmonella Typhimurium. APEC strains cause colibacillosis, which is a major problem in commercial production (Dziva and Stevens, 2008; Mellata, 2013). Colibacillosis occurs when APEC originating from fecal materials translocate into the lung epithelium during fecal aerosolization. Thus, reducing the APEC load in feces as a potential effect of Bacillus spp. in the feed could help reduce the incidence of colibacillosis (Duan et al., 2008; Chien et al., 2011). C. perfringens is a pathogen that causes necrotic enteritis in poultry (Cooper and Songer, 2009) by the production of alpha toxin and NetB (Keyburn et al., 2006, 2008). Necrotic enteritis is a multi-factorial disease that cost poultry farmers 6 billion dollar annually (Wade and Keyburn, 2015). Salmonella Typhimurium, a poultry gut commensal, is the major cause of salmonellosis in humans. This infection is facilitated by the consumption of Salmonella-containing poultry products (Foley and Lynne, 2008; Batz et al., 2012). The ability of Bacillus spp. to suppress the growth of these undesirable organisms might be due to the production of AMPs (bacteriocins). Genome analysis of Ba PTA84, Ba PTA85, and Bs PTA86 suggested that Ba and Bs genomes encoded distinct AMPs (Figure 5B). Ba PTA84 and Ba PTA85 genomes encoded for ribosomally-synthesized lichenicidin A, circularin, LCI, and salicylate containing AMPs that were not found in Bs PTA86 genome. The latter possessed subtilosin A, a cyclic antimicrobial peptide that are potent against some Gram-positive and Gram-negative bacteria such as Listeria monocytogenes, Enterococcus faecalis, Porphyromonas gingivalis, Klebsiella rhizophila, Streptococcus pyogenes and Shigella sonnei, Pseudomonas aeruginosa and Staphylococcus aureus (Babasaki et al., 1985; Stein et al., 2004; Shelburne et al., 2007). For non-ribosomally synthesized AMPs, Ba PTA84 and Ba PTA85 carried plipastatin, surfactin, bacillibactin, bacilysin, and gramicidin. Only the latter was absent in Bs PTA 86.

Bacillus species are known to secrete host beneficial enzymes such as cellulase, xylanase, amylase, protease, β-mannanase, phytase (Latorre et al., 2016; Singh and Bajaj, 2016; Danilova and Sharipova, 2020). These enzymes, when fed to animals, improve digestion of low-calorie diets or reduce intestinal inflammation by breaking down non-starch polysaccharides (NSPs). Some NSPs are anti-nutritional factors, and increase the gut content viscosity, slow down feed retention time in the gut, and thus reduce nutrient absorption (Raza et al., 2019). An accumulation of undigested NSPs can lead to the growth of pathogens that cause subclinical infection challenges (Kaldhusdal, 2000; Van Immerseel et al., 2004). Production of pro-inflammatory cytokines as a response to NSPs demands a significant amount of energy, which otherwise could be preserved for growth, lowering food efficiency and growth performance [reviewed in Cardoso Dal Pont et al. (2020)]. Ba and Bs showed comparable protease, amylase, and β-mannanase activities. These activities were supported by our genomic analysis showing that Ba and Bs possess genes encoding for amylase, protease, β-mannanase, and phytase. Bs PTA86 genome contained more genes for enzymes compared to those of Ba PTA84 and PTA 85.

It is noteworthy that genome analyses revealed other potential benefits of top three Bacillus candidates for animals. Genes encoding a wide array of antioxidant proteins were identified, superoxide dismutase, catalase, thioredoxin, and methionine sulfoxide, and bacillithiol. These proteins when expressed and secreted in the GIT could provide protection toward oxidative stress (Abudabos et al., 2016; Wang et al., 2017; Inatomi and Otomaru, 2018). Oxidative stress occurs in the GIT when the level of free radicals generated by reactive oxygen/nitrogen species (RO/NS) is much higher than the level of antioxidant proteins for neutralization of these toxic compounds (Sherif, 2009). This event is triggered by various factors including nutritional or environmental heat stress, or pathological factors which ultimately decrease growth performance and quality of meat and eggs (Sherif, 2009). Another key desirable traits in a probiotic candidate is the ability to adhere to epithelial cells. The two genes identified in all three strains encode proteins potentially involved in adhesion to mucus, epithelial cells and are known to be involved in host immunomodulation and unwanted microorganism aggregation, providing stability to the strains and the ability to compete with other undesirable resident gut bacteria, thereby enabling effective colonization of the gut and exclusion of pathogens (Antelmann et al., 2000; Sánchez et al., 2008).

Among proposed functions of probiotic bacteria are the reduction of potential pathogenic bacteria, immune modulation, removal of harmful metabolites in the intestine and/or providing bioactive or otherwise regulatory metabolites. Folate-producing probiotic bacteria enable better nutrient digestion and energy recovery. Folate-producing probiotic strains could potentially confer protection against cancer, inflammation, stress, and digestive disturbances (White et al., 1997; Fuchs et al., 2002; Zhu et al., 2005; Rossi et al., 2011; Zhang et al., 2020). Several studies exploring the commercial utility of probiotic strains for folate production have been reported (Sauer et al., 1998; Schallmey et al., 2004; Rossi et al., 2011). Genes encoding essential enzymes in the biosynthetic pathways of folate were also found in the genome of three Bacillus strains. The products of these pathways supply important cofactors which once secreted would be absorbed by the host improving health status (Sauer et al., 1998; Schallmey et al., 2004; Rossi et al., 2011).

In addition, Bacillus DFM candidates must have acceptable safety profiles as expected by regulatory authorities. Some Bacillus spp. are known to produce AMPs and enterotoxins that might exert deleterious effects on the host cells (EFSA Panel on Additives and Products or Substances used in Animal Feed [FEEDAP] et al., 2018). Lack of detailed characterizations of probiotic strains resulted in the use of B. cereus (Bactisubtil, Biosubtyl, and Subtyl) probiotic strains harboring structural genes of known enterotoxins (Duc et al., 2004). Cytotoxicity assessment of our Bacillus spp. strains suggested that Bacillus spp. did not cause cytotoxicity of Vero cells. Moreover, genome analysis of Ba PTA84, Ba PTA85, and Bs PTA86 suggested that enterotoxins and other known virulence factors were absent in the subject Bacillus spp. Another important safety criterion is that Bacillus DFM genomes must be devoid of transferable antimicrobial resistance genotypes (Ciorba et al., 2015). The data suggested that almost all of the tested Bacillus isolates were sensitive to the antimicrobials tested and the apparent MIC values were below the recommended cut-off values. Genomic analysis of three Bacillus spp. identified putative genes for antimicrobial resistance to tetracycline, lincosamide, and streptothricin. In the genome of Bs PTA86, putative genes conferring resistance to rifampicin and macrolides were found. However, these genes have been reported present in Ba and Bs isolates from the environment (Phelan et al., 2011; Adimpong et al., 2012), suggesting these genes may be intrinsic properties of Ba and Bs strains. Furthermore, transferable mobile genetic elements such as transposons, insertion sequences were absent in the proximity of these genes indicating the very low risk of these genes being horizontally transferred to other gut microbes pose little to no risk to public health safety.

To investigate the potential host beneficial metabolites secreted by the three Bacillus strains, we performed global untargeted metabolomic analyses of Ba PTA84 and PTA85, and Bs PTA86. The use of rich and minimal media, as well as the co-culture experiments, allowed us to identify a collection of global secreted molecules of potential value. A metabolite of particular interest was 1-kestose that was identified in the culture supernatants of all strains. 1-Kestose, the smallest fructooligosaccharide (FOS), is a trisaccharide molecule composed of a glucose and two fructose residues linked by glycosidic bonds. Kestose is a prebiotic that, when consumed, enriches the growth of gut commensals such as Bifidobacteria, Lactobacillus, and Faecalibacterium prausnitzii promoting gut health (Patterson et al., 1997). Of the three strains, Ba PTA84 produced the highest amount of 1-kestose. Thioproline, an antioxidant molecule, was identified in the culture supernatant of Ba PTA84 and Bs PTA86. Thioproline was reported to inhibit carcinogenesis in humans, and is expected to act as a nitrite scavenger (Patterson et al., 2020). Pantothenate (Vitamin B5) and pyridoxine (Vitamin B6) were found in the culture supernatants of Ba PTA85 and Bs PTA86, respectively. Betaine and choline were possibly secreted by Bs PTA86. These molecules are methyl donors required for the biosynthesis of acetylcholine and phosphatidylcholine, for neural transmission and cell membrane integrity, respectively (Zempleni et al., 2007). Betaine, when supplemented in feed, has shown improved growth performance of birds during heat stresses (He et al., 2015; Liu et al., 2019). Inclusion of choline has been associated with reduced FCR in broiler chickens (Santiago et al., 2020).

Genomic analysis of Ba PTA84, PTA85, and Bs PTA86 suggested that these three strains harbor genes with complimentary activities (i.e., AMPs and enzymes). Thus, we hypothesize that inclusion of these three strains in a consortium would exert a greater benefit to the animal than that of any single strain. It is noteworthy that the subject Bacillus spp. generated more unique metabolites when grown in consortia of two or three Bacillus strains, suggesting that a combination of strains would generate distinct outcomes compared to that of single isolate (Figure 5). Indoleacetate was detected only in the culture supernatant of consortia of Ba PTA84-Ba PTA85, and Ba PTA84-Ba PTA85-Bs PTA86, but not in the individual strains. Indolacetate is an intermediate of microbial tryptophan biosynthesis that serves as a ligand for aryl hydrocarbon receptors (AhRs) enhances intestinal integrity and modulates host immune systems by exerting anti-inflammatory activities (Nawrocki et al., 2009; Tjaden, 2020). A higher abundance of tryptophan metabolites was also observed in animals treated with sub-therapeutic level of antibiotic Bacitracin Methylene Disalicylate (BMD) (Xie and Tang, 2017).

An in vivo efficacy study employing daily feed inclusion of Ba PTA84 resulted in significantly improved overall growth performance of broiler chickens as shown by a significant increase of ADG, production efficiency, and a reduction in feed conversion ratio. To better understand the mechanism of action underlying the effect of Ba PTA84 supplementation on animal health, we investigated the effects of supplementation of Ba PTA84 on the modulation of intestinal microbiota. The chicken gut microbiota plays prominent roles in bioavailability of nutrients, immune system development, intestinal integrity, and exclusion of unwanted microorganisms (Shang et al., 2018). Growth promotion effects of probiotics have been linked to changes in cecal microbiome structure and function (Al-Fataftah and Abdelqader, 2014; Ma et al., 2018). Interestingly, microbiome taxonomic profiling analyses of cecal contents from a control group and that with dietary supplement of Ba PTA84 suggested no significant differences between the cecal microbiome structures of the two groups according to both alpha and β-diversity parameters. Thus, it is likely that Ba PTA84 supports growth performance without altering the normal cecum microbiome. It is still possible that microbial communities in other organs are affected by supplementation of this strain. It is noteworthy that metabolomic analysis of culture supernatant of Ba PTA84 showed that it had a potential to produce 1-kestose, a microbiome modulator (Patterson et al., 1997). In the future, it would be interesting to test whether 1-kestose is indeed produced in vivo.