Decoding Biothreats With FT‐ICR‐MS: Metabotyping of Bacillus cereus Spores Through Untargeted Metabolomics

ABSTRACT

Genome sequencing is the gold standard for microbial identification. However, it fails to distinguish closely related species with overlapping genetic features. This is the case of Bacillus cereus and Bacillus cytotoxicus, two members of the B. cereus group with distinct pathogenic profiles. While B. cereus , a well‐known opportunistic pathogen involved in intestinal and extraintestinal infections, produces the tissue‐destructive exoenzymes hemolysin BL (HBL), nonhemolytic enterotoxin (NHE), and cytotoxin K (CytK‐2) as well as the plasmid‐encoded emetic toxin cereulide, the thermotolerant species B. cytotoxicus is primarily characterized by producing cytotoxin K‐1 (CytK‐1), implicated in severe diarrheal outbreaks. Despite these differences, genomic data alone often fail to clearly differentiate the two species. Moreover, these bacteria primarily exist as highly resilient spores in environmental samples, further complicating their identification using traditional methods and hindering accurate classification and risk assessment. This distinction becomes particularly relevant in the context of biowarfare. The B. cereus group includes strains that possess genetic elements similar to Bacillus anthracis , the causative agent of anthrax, and some B. cereus strains can acquire virulence plasmids, making them capable of causing anthrax‐like infections. Misidentification or delayed differentiation of these resilient bacteria at spore level is of severe consequences in the event of a biological attack or an accident, where rapid and accurate detection is critical for an effective and timely response. Here, we show that FT‐ICR‐MS–based untargeted metabolomics enables B. cereus and B. cytotoxicus discrimination, while providing critical insights into spore‐specific metabolomic signatures, ultimately contributing to improved microbial forensics and bio‐surveillance strategies.

1. Introduction

The intentional use of microorganisms as bioweapons is as long as humanity itself. Depending on the scale and intent, such events are classified as biowarfare, bioterrorism, or biocrime. In biowarfare, the practice dates to prehistoric times, when corpses were used to contaminate water sources. From 1900 to 1945, national biowarfare programs emerged and biological agents were deployed in both World Wars. After 1945, advances in biotechnology, biochemistry, and genetic engineering expanded biowarfare capabilities and increased lethality through genetic manipulation. Thus, biological agents may be used in their natural form or in genetically modified forms. Regardless of the context, rapid identification and characterization of the agent are essential to an effective response and to reduce morbidity and mortality [1].

Not all pathogenic microorganisms are suitable bioweapon candidates, as pathogenicity or toxicity alone does not determine weaponization potential [2]. Factors such as tissue‐damaging capacity, availability, and efficient delivery to a host are critical when evaluating their suitability for bioterrorism. The distinction between whole microbes and isolated toxins further complicates classification [2]. Historically, the Centres for Disease Control and Prevention (CDC) grouped biological agents into three categories: Category A agents pose the highest risk due to their ease of dissemination, high mortality rates, and potential to cause major public health and societal disruption; Category B agents are moderately easy to disseminate and associated with moderate morbidity and low mortality; and Category C agents comprise emerging pathogens that could be engineered for mass dissemination in the future [1, 3].

Bacillus anthracis and Bacillus cereus biovar anthracis are classified as Category A bioagents and are listed within the Tier 1 subset of the US Select Agents and Toxins List. Although belonging to the same genus, B. cereus has not been recognized as having “well‐earned pathogenic status,” being only mentioned in the B. anthracis context. Even though B. cereus gathers the conditions to be used as a study model for B. anthracis outbreaks, it represents by itself a major threat to human health [4]. Typically known as a foodborne opportunistic pathogen responsible for emetic and diarrheal syndromes, B. cereus can also cause severe, life‐threatening infections, particularly in immunocompromised individuals [4].

B. cereus is closely related to B. anthracis , with both belonging to the B. cereus sensu lato (s.l.) complex, a Bacillus genus that comprises numerous lineages of closely related, gram‐positive facultatively anaerobic bacteria. Members of this complex range from significant pathogens, such as B. cereus and B. anthracis , to strains widely used in agriculture and industry. Accurate species‐level classification of B. cereus s.l. isolates is therefore critical for assessing public health risks and industrial relevance [5]. As spore‐forming bacteria, these species can adapt to extreme environmental conditions, remaining dormant or rapidly growth and multiplying when favorable conditions return [6, 7]. These reasons make bacterial spores the gold standard in bioweapons applications, and their rapid identification in a bioterrorism scenario is of utmost importance.

However, the taxonomic classification of the B. cereus s.l. group has long been challenging and controversial, largely because different approaches yield inconsistent species boundaries. Using phenotypic and classical molecular methods, the group was originally divided in 1998 into six species: B. anthracis , B. cereus sensu stricto (s.s.), B. mycoides , B. thuringiensis , B. pseudomycoides , and B. weihenstephanensis . The adoption of whole‐genome sequencing (WGS) and average nucleotide identity (ANI) led to the recognition of additional species, including B. toyonensis and B. cytotoxicus in 2013, followed by B. wiedmannii in 2016 and nine more species (i.e., B. albus , B. luti , B. mobilis , B. nitratireducens, B. pacificus , B. paramycoides, B. paranthracis, B. proteolyticus, and B. tropicus ) in 2017, expanding the complex to at least 18 recognized species. Yet, ANI‐based classification, which measures the average nucleotide similarity of two genomes, is widely used for species delineation, typically at thresholds of 95%–96%. However, its application remains sensitive to both the selected cutoff and the reference genome, often resulting in overlapping genomospecies clusters and inconsistent classifications. Consequently, the same isolate may be assigned to different species depending on the analytical framework employed [5].

To address these inconsistencies, multiple nomenclatural systems have been proposed, including phenotypic, genomic, and hybrid approaches integrating both data types. Among these, the NCBI Taxonomy provides a widely used, curated hierarchical framework, currently encompassing 21 species within the B. cereus s.l. group, excluding strains lacking genome data or not validly published under the International Code of Nomenclature of Prokaryotes. However, as a dynamic resource that incorporates user‐submitted data, it remains susceptible to misclassification, reflecting the broader lack of standardization across the field [5, 8].

In practice, phenotypic characterization remains the primary step for preliminary classification, commonly following standardized protocols such as the Bacteriological Analytical Manual (BAM), with genomic approaches applied when ambiguity persists [5]. Nevertheless, both strategies have limitations: Phenotypic traits often lack sufficient resolution, while genomic classifications can suffer from variability and reproducibility issues. Moreover, the high frequency of horizontal gene transfer within the B. cereus s.l. group further complicates species identification, as gene and plasmid content alone may not reliably reflect taxonomic boundaries [9, 10].

Alternative and more effective identification methods are required, especially at the spore level. Some spores of Bacillus species ( B. anthracis , B. cereus , and B. thuringiensis ) are very difficult to distinguish through light or electron microscopy, being almost identical within strains of the same species [11]. Also, the core components (dipicolinic acid, small acid‐soluble proteins [SASPs], and calcium ions) and spore coat proteins, despite being difficult and time‐consuming to extract for accurate identification [11], are quite conserved across the genus [12].

Upon sporulation, the vegetative cell undergoes extensive metabolic reprogramming to form a completely different entity—the spore. During this process, several key enzymes involved in the synthesis of metabolic precursors are lost, while the enzymatic machinery required for macromolecular biosynthesis is preserved. Furthermore, the spore retains a highly specific and specialized metabolome, imported from the mother cell [13]. As a result, spores are metabolically dormant but not metabolically inactive. This strategic reorganization supports long‐term survival [14] and ensures sufficient endogenous resources to sustain metabolism during the early stages of germination [15]. However, the characterization of the spore metabolome remains far from fully established and represents a powerful tool for spore biotyping. Subtle genetic variations can translate into significant downstream metabolic alterations, providing a snapshot of the organism's actual physiological state [16].

Hence, an untargeted metabolomics approach could be the perfect strategy for complete metabolome characterization and discrimination between the similar Bacillus spores. Metabolomics has successfully been used to discriminate between organisms, not only at the species level [17, 18, 19] but also between cultivars [17] and mutants [20] from the same species, especially when Fourier transform ion cyclotron resonance mass spectrometry (FT‐ICR‐MS) is employed due to its extreme resolution and mass accuracy.

In this work, the metabolome of five B. cereus s.s. strains (two wild and three reference strains) and one B. cytotoxicus reference strain were characterized using FT‐ICR‐MS, with the purpose of species and strain discrimination.

2. Experimental Procedures

2.1. Purification of Bacterial Spores

Five strains of B. cereus and one of Bacillus cytotoxicus were used in this work, belonging to the Laboratório de Defesa Biológica collection, Portuguese Army (Table 1). The two B. cereus wild strains were isolated from human vomiting and food, while the reference strains were purchased from the NCTC culture collections. Bacterial spores were first cultivated in AK sporulation agar (Difco) for 7 days at 37°C. Bacterial colonies were then transferred to a falcon tube containing 20 mL of ultrapure water. Spore suspensions were purified according to a density gradient procedure. Briefly, samples were gently shaken and submitted to a thermal shock at 70°C for 30 min, followed by a 5‐min vortex at maximum speed and a centrifugation step at 3000 g for 15 min. The supernatant was discarded, and the spore pellet was suspended in 20 mL of ultrapure water. Spores were kept overnight at 4°C and observed by phase contrast microscopy to confirm the absence of vegetative cells.

TABLE 1.

Wild and reference B. cereus and B. cytotoxicus strains used in this work.

Species	Strain code	Strain	Description	Toxins
Bacillus cereus	J3	Wild	Spores isolated from human vomiting.	Ces
	J271	Wild	Spores isolated from food. No reports of its involvement in spore outbreaks.	—
	J630	ATCC 10987	Nonlethal strain, isolated from dairy products.	Nhe
	J648	F4810/72	Strain associated with emetic illness.	Ces, Nhe
	J649	Reference, F528/94	Strain associated with diarrheal disease.	Nhe, Hbl, CytK2
Bacillus cytotoxicus	J650	Reference, NVH0391‐98	Strain associated with a lethal form of diarrheal disease. Isolated from food.	CytK1

Abbreviations: Ces, cereulide; CytK1, cytotoxin K1; CytK2, cytotoxin K2; Hbl, hemolysin BL; Nhe, nonhemolytic enterotoxin.

2.2. Spore Inactivation and Metabolite Extraction

For simultaneous spore inactivation and metabolite extraction, spore suspensions were treated with a solvent mixture of 70%:30% (v/v) formic acid: acetonitrile, according to Bruker's protocol [21]. The resulting extract was filtered through a 0.2 μm, 1.5‐mL ultrafiltration filters (Pall). The filtrate was centrifuged at 7000 rpm for 15 s, and the supernatant was collected for metabolomics analysis. Efficient spore inactivation was validated by inoculating 50 μL of the culture media and incubating at 37°C for 48 h. No bacterial growth was observed, confirming complete inactivation. Extracted samples were stored at −80°C until MS analysis.

2.3. FT‐ICR‐MS Untargeted Metabolomics

Extracted metabolites were diluted 1000‐fold in 70%:30% (v/v) acetonitrile: H₂O. The internal standard human leucine enkephalin (YGGFL, Sigma Aldrich) was added to a final concentration of 0.005 μg/mL and was used as calibrant. Formic acid (FA, Sigma Aldrich, HPLC grade) was added at a final concentration of 0.1% (v/v) to improve protonation. Diluted samples were directly infused by electrospray ionization into a 15‐T SolariX XR FT‐ICR mass spectrometer (Bruker Daltonics, Bremen, Germany), at a flow rate of 2.0 μL/h. The nebulizer gas pressure was set at 2.5 bar, the drying gas flow rate at 4.0 L/min at a temperature of 200°C, and the capillary exit voltage at 200 V. Ions were accumulated in the collision cell for 0.2 s. For each analysis, 100 transients were accumulated, using a 4M size, in the 90–1500 m/z range. Resolving power is sufficient to separate two peaks at 340.34850 and 340.34611 m/z (Figure 1d).

FIGURE 1.

FT‐ICR MS–based metabolic fingerprinting. (a) Spore biotyping workflow in which relative mass intensities and abundances are translated into metabolic variation, enabling spore discrimination. (b) Mass profiles of direct‐infusion spore extracts. Despite the presence of common mass peaks (like the m/z values 239.10601 and 338.34166), each strain displays a distinct metabolic fingerprint that forms the basis of the identification method. (c, d) Metabolite identification example (13‐docosenamide). High‐resolution isotopic pattern analysis enables compound identification without the need for tandem mass spectrometry. Isotopes of carbon (¹³C), nitrogen (¹⁵N), hydrogen (²H), and oxygen (¹⁸O) were detected, allowing the unambiguous assignment of erucamide.

2.4. Data Analysis

FTMS peak finder algorithm in Data Analysis 5.0 software package (Bruker Daltonics, Bremen, Germany) served as the initial quality control filter for the MS raw data. Default thresholds for spectral quality were set as follows: signal‐to‐noise ratio (S/N) of 4, relative intensity at 0.1%, and absolute intensity at 10⁶. Subsequently, the S/N and relative intensity thresholds were adjusted based on sample variability, ranging from 5 to 6 for S/N and 0.2% to 0.3% for relative intensity. An in‐house software built in Python and based on Metabolinks [22] was used to align the m/z and peak intensities list, for formula assignment, compound annotation, and statistical analysis. The software uses the modules pandas [23], Scikit‐learn [24], SciPy [25], NumPy [26], as well as matplotlib [27], seaborn [28], and UpSetPlot [29] for data visualization. Initial data filtering was performed considering m/z peaks present in more than one replicate. Missing values' imputation was achieved using one‐fifth of the smallest intensity value in the dataset, followed by sum normalization, generalized log (glog) transformation, and a mean centering scaling. Compound annotation was performed using the Human Metabolome Database (HMDB, [30]) and Microbial Metabolome Database (MiMeD, [31]). Due to the possibility of several annotations to the same m/z value, manual filtration was performed for metabolite selection within a reasonable bacterial/spore biological framework.

2.4.1. Enrichment Analysis

Overrepresentation analysis (ORA) was performed to identify the most enriched HMDB chemical subclasses among the commonly detected metabolites in the dataset. All annotated metabolites were used to define the background universe. Metabolites were grouped according to their HMDB chemical subclass, and enrichment analysis was carried out using a hypergeometric test to evaluate whether the number of detected metabolites within each subclass was higher than expected by chance. Subclasses containing fewer than two detected metabolites were excluded from the analysis. The resulting p‐values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) correction.

2.4.2. Unsupervised and Supervised Statistical Analysis

Principal component analysis (PCA) and hierarchical clustering analysis (HCA) were performed, with sample projection based on the first three principal components and HCA conducted using Bray–Curtis distance with single linkage. Key metabolites contributing to the metabolic segregation of the spore strains were identified using a random forest (RF) model with 500 trees and a random seed of 42 for reproducibility. Gini importance values were used to rank all metabolites, and the top 2% highest scoring compounds were retained for further analysis. Statistical significance of these metabolites was assessed by one‐way ANOVA, with Benjamini–Hochberg correction for multiple testing.

3. Results and Discussion

To gain insights into the metabolic nature of spores, we performed an untargeted metabolomics analysis using FT‐ICR‐MS (Figure 1a). As illustrated in the mass spectra (Figure 1b), the overall spectral profile clearly differs among strains, indicating pronounced metabolic specificity. Therefore, we aimed to determine whether spores retain a basal metabolic signature, and their metabolomic profiles exhibit sufficient to enable discrimination at the species and strain levels.

3.1. Metabolic Profiling of Spores

Common and exclusive compounds detected across all bacterial strains were used to provide an overview of spores' metabolic profiles (Figure 2a). Due to the distinct sources from which they were directly isolated, the B. cereus wild strains J3 and J271 exhibited the highest number of detected m/z masses, with 502 and 178 peaks, respectively. Moreover, the B. cytotoxicus reference strain (J650) displayed greater metabolic diversity compared to the B. cereus reference strains. The source of isolation may explain this difference. Similar to strain J271, the J650 was isolated from food, specifically vegetable purée [32]. This is consistent with the high number of shared m/z masses between these two strains, which exhibited the greatest overlap (212).

FIGURE 2.

Global assessment of B. cereus and B. cytotoxicus metabolic profiles. (a) UpSet plot of common and strain‐specific compounds. Values correspond to the detected m/z ratios for each strain and for the combinations indicated by the connected black circles. (b) Van Krevelen metabolic maps in which the x‐ and y‐axes represent the hydrogen‐to‐carbon (H/C) and oxygen‐to‐carbon (O/C) ratios, respectively. Each point corresponds to a detected peak, while shaded regions indicate typical locations of chemical classes. Point density reflects the relative abundance of compounds within each class, providing a global overview of the spore metabolome.

Despite the challenges associated with metabolite annotation, primarily due to the presence of constitutional isomers and the limited availability of dedicated databases for bacterial spores, molecular formula assignment can be robustly achieved by assessing isotopic pattern distributions, which are readily resolved using extreme‐resolution FT‐ICR mass spectrometry (Figure 1b,c). As expected, a remarkably rich array of compounds was detected. To collect information from the large number of assigned formulas, all identified compounds were clustered into individual regions based on hydrogen‐to‐carbon (H/C) and oxygen‐to‐carbon (O/C) molecular ratios in Van Krevelen diagrams (Figure 2b).

Although the overall Van Krevelen scatter patterns were broadly comparable across all samples, the B. cereus wild strains exhibited markedly greater metabolomic diversity than the reference strains, particularly within the lipid‐, amino acid/peptide‐, and terpenoid‐like regions. This expanded molecular richness suggests a more complex chemical architecture in wild isolates, potentially reflecting adaptive metabolic plasticity shaped by environmental pressures. In contrast, the carbohydrate‐like domain remained sparsely populated across all strains, indicating that highly oxygenated carbohydrate‐type compounds contribute only marginally to the spore metabolome.

Among the reference strains, B. cereus J648 showed the lowest degree of molecular diversity, especially within the amino acid/peptide‐like region. This reduced compositional breadth is consistent with the limited number of detected m/z values (538) and the absence of strain‐exclusive metabolites (Figure 2a), suggesting a comparatively streamlined or less chemically diverse spore metabolome.

Overall, distinct metabolomic profiles enabled species‐ and strain‐level discrimination. The B. cereus wild isolates (J3 and J271) showed greater metabolic richness and diversity than reference strains, while strain J648 exhibited the lowest diversity and no unique metabolites. The B. cytotoxicus reference strain (J650) displayed higher diversity than B. cereus reference strains and shared the greatest overlap with the food‐derived strain J271, suggesting an influence of isolation source on metabolomic composition.

While Van Krevelen diagrams provide an empirical compositional framework and do not impose strict chemical boundaries, the clustering patterns consistently reveal five predominant chemical classes within the spore metabolome. Independent of relative abundance differences, lipids, amino acids/peptides, carbohydrates, terpenoids, and aromatic oxygenated compounds were systematically detected across all strains. This conserved compositional architecture suggests that these chemical classes constitute a core molecular signature of the B. cereus spore metabolome, potentially reflecting fundamental structural and functional requirements for spore resistance, dormancy, and germination efficiency.

It is well established that lipids, amino acids/peptides, and carbohydrates constitute the major chemical classes of spore structural layers [33]. Nevertheless, considering the previously described inactivation procedure, which included a filtration step after chemical inactivation, we propose that the detected compounds represent a mixture of residual structural components and, predominantly, core‐associated metabolic content.

For instance, SASPs are highly abundant low‐molecular–weight proteins (6–9 kDa) that protect DNA during dormancy and facilitate rapid repair during outgrowth. In addition, they serve as a readily available source of free amino acids during early germination [33]. Therefore, SASPs‐derived peptides from core extraction likely contribute substantially to the pool of amino acid/peptide‐like compounds detected.

However, the considerable diversity and abundance of amino acid/peptide‐like metabolites observed across strains strongly indicate the presence of additional protein sources beyond SASPs, including environmental peptides co‐extracted during spore purification. Their relative contribution likely depends on the metabolic state of the mother cell and the environmental conditions during sporulation, which may explain the broader amino acid/peptide‐like diversity observed in wild strains.

Similarly, lipid‐ and terpenoid‐like compounds were abundantly and consistently detected across strains. These biomolecular subclasses are most plausibly associated with structural membrane components rather than freely dispersed within the core. In contrast to fungal spores, where lipid droplets have been described [34], no clear evidence of lipid droplet accumulation has been reported in bacterial spores.

Nonetheless, to mechanistically explain the ~1.3‐ to 1.6‐fold expansion of the core membrane surface area during germination, Laue and co‐workers reported intracellular membrane structures located beneath the inner (core) membrane in spores of B. subtilis , B. thuringiensis , and B. anthracis , to name a few [35]. These structures, termed sub‐core membrane membranes (sCMMs), appear as compressed vesicles or tubules and disappear upon core expansion. They exhibit a composition like the inner membrane and are proposed to function as a membrane reservoir [35]. The abundance of lipid species detected in our dataset is consistent with the presence of such membrane‐derived structures.

Although the lipid composition of the spore inner membrane resembles that of vegetative plasma membranes, it exhibits remarkably low permeability [33]. In this context, the detection of terpenoid‐like compounds may be relevant to membrane rigidity. Hopanoids, pentacyclic triterpenoids that function as sterol analogs in bacteria, are known to enhance membrane stability and reduce permeability [36, 37]. Additionally, they have been implicated in oxidative stress mitigation [38], like sporulenes previously reported in B. subtilis spores [39]. The detection of squalene, a triterpene and direct precursor of hopanoids, among the commonly annotated compounds further supports the potential presence of triterpenoid‐derived structures within the spore metabolome.

3.2. Functional Implications of Lipids and Terpenoids

Chemical classes encompass a broad diversity of molecular structures. To gain a more detailed understanding of the predominant metabolites and their potential roles in spore physiology, an ORA was performed at the subclass level using the 463 commonly annotated m/z peaks (Figure 3a). Due to multiple annotations per peak, 1028 metabolites were initially identified as potentially present in the spore samples. After manual curation, approximately 12% of these were retained for enrichment analysis.

FIGURE 3.

Fatty acid–associated metabolite enrichment in bacterial spores. (a) Overrepresentation analysis of common metabolites according to HMDB chemical subclasses. Circle size represents the enrichment ratio (number of hits per subclass), and color scale indicates p‐values (blue, significant; red, nonsignificant). Fatty acids and conjugates, fatty amides, and fatty acid esters were significantly enriched, with fatty acids and conjugates showing the highest enrichment ratio. (b) Distribution of compounds within the fatty acids and conjugates subclass across strains (HMDB subclass weight). Wild strains show the greatest diversity, followed by the dairy‐isolated strain J630; strains J649 and J650 show similar levels, while the emetic B. cereus strain J648 (rice isolate) shows the lowest diversity. (c) Schematic representation of the influence of the sporulation environment on spore properties through incorporation of exogenous lipids. (d, e) Normalized intensities of 13‐docosenamide and methyltetradecanoylcarnitines across strains. Error bars represent the standard deviation of technical replicates. Asterisks (*) indicate statistical significance (p < 0.05, pairwise t‐test).

Fatty acids and conjugates, fatty amides, and fatty acid esters emerged as three significantly enriched subclasses, underscoring the central role of lipids in spore physiology. Nevertheless, other metabolite subclasses were also consistently detected among the 143 curated metabolites, including amino acids, peptides and analogs, diterpenoids, triterpenoids, and sesquiterpenoids. These last three subclasses correspond to the repetition of four, six, and three isoprene units (C₅H₈), respectively, and constitute part of the spore terpenome, whose function was already described.

Within the fatty acids and conjugates subclass, most annotations corresponded to saturated, branched, and non‐branched long‐chain fatty acids, namely, pentadecanoic acid (C15:0), palmitic acid (C16:0), heptadecanoic acid (C17:0), and stearic acid (C18:0), as well as their respective iso‐ and anteiso‐branched isomers. These fatty acids are widely associated with membrane structure and rigidity, ultimately contributing to spore resistance [40, 41, 42, 43].

In addition to this classical spore lipid profile, first described in the mid‐20th century, several unusual fatty acids were also detected. Notably, vaccenic acid, commonly abundant in dairy products, was among them, suggesting a strong influence of the sporulation environment on the spore metabolome. Indeed, previous studies have demonstrated that dormant spores exhibit distinct properties depending on the growth medium composition and food components present during sporulation, and exogenous lipids can be incorporated into the spore membrane during endospore formation [44, 45, 46]. In both cases, the resulting spore phenotype can be substantially altered (Figure 3c).

Accordingly, the metabolome of spores isolated from human vomiting is expected to differ markedly from that of spores isolated from food matrices, as well as between wild and reference strains, capturing their unique environmental background. In this context, recently isolated strains (wild strains) retain metabolic configurations shaped by fluctuating environmental constraints (like nutrient limitation, low pH, and gastric and intestinal contents), whereas reference strains, ultimately derive from historical wild isolates, have undergone long‐term propagation under standardized laboratory conditions, a phenomenon analogous to “metabolic domestication” described in other microorganisms [47, 48]. This observation is further supported by our data when examining the HMDB subclass weight distribution of the overall spore metabolome (Figure 3b). The relative weight of fatty acids and conjugates was substantially higher in wild strains compared to reference strains, particularly in the food‐isolated J271 spores. Furthermore, beyond discriminating wild‐type from reference strains via fatty acid and conjugate weight distributions (HMDB subclasses), diarrheal‐associated strains are similarly distinguishable. Notably, B. cereus J649 and B. cytotoxicus strains exhibit similar HMDB weights, suggesting comparable lipid abundance that may be linked to host infection processes. Although fatty acids and conjugates account for the largest proportion of the commonly detected metabolites, the additional identification of fatty amides and fatty acid esters raises intriguing questions about the biochemical nature and structural complexity of spores. In particular, the annotation of 13‐docosenamide (erucamide) and multiple (3–13)‐methyltetradecanoylcarnitine isomers.

Erucamide, an antimicrobial long‐chain fatty acid amide (C22:1), is secreted into the medium by many bacteria when growing with glucose. As such, its production closely follows growth and glucose utilization [49, 50]. The detection of this metabolite in dormant spores might have three main implications: Firstly, the metabolite is prestored as a secondary metabolite in the spore (probably in the outer layers), produced during sporulation when glucose was available. Alternatively, the spores may be “primed” with a preformed signal of antimicrobial, potentially to influence nearby microbes or host cells immediately upon germination or rehydration, or it could serve as a nutritional biomarker of the sporulating culture. Overall, erucamide can be functionally involved in defensive coating, early signal pulse during germination, or metabolic remnant as a diagnostic marker for glucose‐associated sporulation. When analyzing its relative normalized intensity across the strains, it is possible to observe its abundance on the pathogenic J648 and J649 B. cereus reference strains, strongly suggesting its role as an early signalling molecule modulating host–pathogen interactions (Figure 3d).

Acylcarnitines, such as (3–13)‐methyltetradecanoylcarnitine, result from esterification of a fatty acid moiety (e.g., anteiso‐C15:0) to carnitine's hydroxyl group and represent a potential source of carnitine. Carnitine functions as a compatible solute in vegetative B. subtilis cells via OpuC uptake [51], though it is not among documented spore osmolytes like proline or glycine betaine [52]. Detection in spore samples likely reflects uptake from gastric secretions and food matrices, both established carnitine sources, during growth and sporulation rather than a defined germination role. Notably, the B. cereus wild strain, J3, isolated from human vomiting, exhibited the highest methyltetradecanoylcarnitine relative intensity, consistent with elevated carnitine availability from salivary/gastric secretions [53] and meat/dairy digestion [54] in the vomiting matrix (Figure 3e). The released fatty acids (such as 13‐methyltetradecanoic acid) match those found in spore membranes, but no evidence confirms their utilization during core membrane expansion or early β‐oxidation post‐germination. Further studies will be necessary to support this theory.

Taken together, wild isolates exhibit a markedly higher relative abundance of fatty acids and conjugates compared to reference strains, consistent with metabolomic adaptation to variable environmental conditions. Diarrheal‐associated strains, including B. cereus J649 and B. cytotoxicus, further display similar lipid abundance profiles, suggesting conserved metabolic features. In addition, food‐ and vomiting‐derived strains show distinct lipid and acylcarnitine signatures, reinforcing the influence of isolation source on spore metabolic composition and enabling robust strain discrimination.

3.3. Metabolic Signatures Underlying Bacillus Spore Discrimination

Independent of the particularity of raising unanswered questions, untargeted metabolomic data interpretation clearly highlights the existence of a specific metabolic signature supporting strain‐level segregation. Accordingly, multivariate statistical analyses were applied to the metabolomic datasets. A PCA and HCA revealed analytical reproducibility and metabolic profile similarities within the different bacterial samples (Figure 4a,b). While B. cereus reference strains share a common metabolic background, clustering closely in both PCA and HCA, the wild strain J3 exhibits the most distinct metabolomic profile. Furthermore, the food‐isolated J271 strain shows the greatest similarity to B. cytotoxicus.

FIGURE 4.

Metabolomic segregation of Bacillus strains. (a, b) HCA and PCA analyses reveal three main clusters: the wild human vomiting B. cereus J3 strain, which shows the highest metabolic differentiation; a group of closely clustered B. cereus reference strains; and a third group including B. cytotoxicus and the wild food B. cereus J271 strain. The results indicate that the isolation source is a major driver of metabolic segregation. (c) Inclusion of the third PCA component suggests an additional contribution of genetic or epigenetic factors, separating the reference strains from the wild isolates, while B. cytotoxicus appears distinct in the lower right region of the plot. Overall, metabolomic profiles discriminate both closely related Bacillus species and strains within the same species.

These observations are consistent with previous findings (Figure 2a), indicating that the metabolic background is primarily shaped by the sporulation environment and to a lesser extent by genomic factors. Notably, wild B. cereus strains can be discriminated from both reference B. cereus strains and B. cytotoxicus along the third principal component (PC3; Figure 4c). However, this genetic contribution accounts for only a small proportion of the observed variance (PC3: 12.6%), whereas the isolation origin, reflected in PC1 and PC2, explains the majority (55.8%).

Together, these results suggest that genomic differences alone contribute modestly to the metabolic discrimination between B. cereus strains and B. cytotoxicus, with environmental factors playing a dominant role. This may explain the challenges associated with relying solely on genomic traits to distinguish Bacillus spore isolates from different sources, highlighting the necessity of integrating phenotypic data for accurate characterization [55].

To assess the metabolic specificity of Bacillus spores, enabling species‐ and strain‐level segregation, a supervised multivariate statistical analysis was conducted to identify key discriminatory metabolites. Among the 50 Gini importance‐ranked m/z masses, only 15 were annotated, of which just three seemed to fit the bacterial/spore biological framework (Table 2). Fatty acid esters and diterpenoids were the chemical subclasses of the metabolites annotated.

TABLE 2.

Random forest–based identification and annotation of discriminant metabolites.

Bucket label	Formula	Name	Database IDs	Gini importance	p‐adjusted a
310.31041	C20H39NO	N‐Hexadecanoylpyrrolidine	[HMDB0032740]	0.00161	1.19E‐21
191.14304	C13H18O	β‐Damascenone; 2‐(3‐Phenylpropyl)tetrahydrofuran; 2,2,6,7‐Tetramethylbicyclo[4.3.0]nona‐1(9),4‐dien‐8‐one; xi‐3‐(4‐Isopropylphenyl)‐2‐methylpropanal; dehydro‐β‐Ionone; Megastigmatrienone; 1‐[2,6,6‐Trimethyl‐1,3‐cyclohexadien‐1‐yl]‐2‐buten‐1‐one; trans‐β‐damascenone	[HMDB0013804], [HMDB0036178], [HMDB0036685], [HMDB0037763], [HMDB0059904], [HMDB0059906], [MMDBc0033423]	0.00152	1.13E‐08
341.19586	C18H28O6	11‐Dehydro‐2,3‐dinor‐txb2; α‐O‐Methyldihydrobotrydialone; 5,6‐dihydro‐pinolidoxin; β‐O‐Methyldihydrobotrydialone	[HMDB0244346], [MMDBc0012339], [MMDBc0013526], [MMDBc0016287]	0.00151	4.04E‐09
428.37344	C25H49NO4	5‐; 11‐; 14‐; 9‐; 8‐; 15‐; 10‐; 7‐; 13‐; 3‐; 16; 4‐; 6‐; 12‐Methylheptadecanoylcarnitines	[HMDB0240867], [HMDB0240868], [HMDB0240869], [HMDB0240870], [HMDB0240871], [HMDB0240872], [HMDB0240873], [HMDB0240874], [HMDB0240875], [HMDB0240876], [HMDB0240877], [HMDB0240878], [HMDB0240879], [HMDB0240880].	0.00149	2.76E‐19
391.2843	C24H38O4	12‐Ketodeoxycholic acid; 7‐Hydroxy‐3‐oxocholanoic acid; Nutriacholic acid; 7α‐Hydroxy‐3‐oxo‐5β‐cholan‐24‐oic acid; 9′‐Carboxy‐α‐chromanol; Δ8‐Merulinic acid A; 12α‐hydroxy‐3‐oxo‐5β‐cholan‐24‐oic acid; 3β,7α‐Dihydroxychol‐5‐en‐24‐oic acid; 3α‐Hydroxy‐7‐oxo‐5β‐cholanic acid; Antroquinonol; Bis(2‐ethylhexyl) phthalate; Bis(2‐ethylhexyl) terephthalate; Diisooctyl phthalate; Dioctyl phthalate; 3,4‐Bis(2‐ethylhexyl)phthalic acid; Acetoxy‐[10]‐gingerol	[HMDB0000328], [HMDB0000460], [HMDB0000467], [HMDB0000503], [HMDB0012866], [HMDB0041454], [HMDB0062742], [HMDB0246261], [HMDB0246682], [HMDB0248495], [HMDB0249243], [HMDB0249244], [HMDB0251353], [HMDB0251427], [HMDB0257953], [HMDB0302133]	0.00141	2.65E‐09
397.40402	C26H52O2	Hexacosanoic acid; 6‐Hydroxy‐8‐hexacosanone; cerotic acid	[HMDB0002356], [HMDB0035628], [MMDBc0057084]	0.00137	9.77E‐11
398.23254	C24H31NO4	Drotaverine; 1‐(3,4‐Diethoxybenzyl)‐6,7‐diethoxy‐3,4‐dihydroisoquinoline	[HMDB0015669], [HMDB0243791]	0.00132	0.0069
400.34211	C23H45NO4	L‐Palmitoylcarnitine; O‐Palmitoylcarnitine; 12‐; 9‐; 5‐; 6‐; 11‐; 3‐; 8‐; 7‐; 13‐; 4‐; 14‐; 10‐Methylpentadecanoylcarnitine	[HMDB0240774], [HMDB0240783], [HMDB0240843], [HMDB0240844], [HMDB0240845], [HMDB0240846], [HMDB0240847], [HMDB0240848], [HMDB0240849], [HMDB0240850], [HMDB0240851], [HMDB0240852], [HMDB0240853], [HMDB0241914]	0.00124	1.65E‐10
383.25811	C25H34O3	12′‐Apo‐violaxanthal; Phorone A; Levonorgestrel butyrate; 12′‐Apo‐capsorubinal; (3S,5R,6R)‐3,5‐dihydroxy‐6,7‐didehydro‐5,6‐dihydro‐12′‐apo‐β‐caroten‐12‐al; (3S,5R,6S)‐5,6‐epoxy‐3‐hydroxy‐5,6‐dihydro‐12′‐apo‐β‐caroten‐12‐al; Anhydroophiobolin A	[HMDB0034953], [HMDB0256461], [HMDB0259613], [HMDB0302028], [HMDB0303973], [HMDB0303974] [MMDBc0015251]	0.00124	1.15E‐09
336.32611	C22H41NO	Pipericine; N‐Isobutyloctadeca‐trans‐2‐trans‐4‐dienamide	[HMDB0031678], [HMDB0302880]	0.00121	5.94E‐24
282.27912	C18H35NO	Oleamide; 9‐Octadecenamide; Laurocapram	[HMDB0002117], [HMDB0247607], [HMDB0253992]	0.00116	1.36E‐26
493.49784	C33H64O2	10,12‐Tritriacontanedione	[HMDB0035578]	0.00115	3.79E‐15
325.11291	C12H20O10	Difructose anhydride III	[HMDB0251271]	0.00115	2.44E‐17
369.2272	C20H32O6	19‐Hydroxy‐PGE2; 6,15‐Diketo,13,14‐dihydro‐PGF1a; Prostaglandin G2; 20‐Hydroxy‐PGE2; 6‐Ketoprostaglandin E1; 11‐Dehydro‐thromboxane B2; Thromboxane B3; 5(6)‐Epoxy Prostaglandin E1; 20‐COOH‐10,11‐dihydro‐LTB4; 20‐dihydroxyleukotriene B4; Cinncassiol D2; Cinncassiol D3; 19‐hydroxyprostaglandin H2(1‐); 20‐hydroxylipoxin A4; (E)‐7‐[4‐Hydroxy‐2‐[(E)‐3‐hydroxyoct‐1‐enyl]‐6‐oxooxan‐3‐yl]hept‐5‐enoic acid; 19‐Hydroxyprostaglandin E2	[HMDB0001908], [HMDB0001979], [HMDB0003235], [HMDB0003247], [HMDB0004241], [HMDB0004242], [HMDB0005099], [HMDB0012110], [HMDB0012633], [HMDB0012635], [HMDB0034678], [HMDB0036854], [HMDB0062683], [HMDB0062748], [HMDB0244347], [HMDB0246952]	0.00111	9.0E‐24
423.19895	C20H32O8	Cinncassiol B	[HMDB0036855]	0.00111	1.07E‐19

Note: After performing the RF, a total of 50 metabolites were identified as having the highest scoring values when ranked by Gini importance. Among them, 15 were annotated. Due to the presence of isomers, the same m/z has multiple identifications. Hence, for each case, the metabolites were manually annotated. Note that the Gini importance values are relative, and its magnitude depends entirely on the scale of the dataset.

^a p‐values were calculated as described in the method section.

Contrary to existing reports, the recurrent presence of carnitine fatty acyl esters in spore samples indicates a universal role in their functional biology. Acylcarnitines derived from tetradecanoic acid (methyltetradecanoylcarnitine, C14) were detected in every strain, with particularly high levels in the wild isolate J3 (Figure 3e). In contrast, pentadecanoic (methylpentadecanoylcarnitine, C15) and heptadecanoic (methylheptadecanoylcarnitine, C17) acylcarnitines were absent in pathogenic B. cereus reference strains and B. cytotoxicus, yet highly abundant in the nonlethal dairy‐associated reference strain J630. Notably, C17 acylcarnitine was also abundant in the wild food isolate J271, which has not been linked to outbreaks (Figure 5a,b). Together, these findings suggest that longer‐chain acylcarnitines (C15–C17) may be associated with nonpathogenic or environmentally adapted strains, whereas shorter‐chain species (C14) are broadly distributed. This is the case of C15 acylcarnitine, that is elevated only in the nonpathogenic strains. This pattern supports the hypothesis that acylcarnitine fatty acyl chain length reflects strains' metabolic traits, potentially related to spores' pathogenic potential.

FIGURE 5.

Random forest identifies key metabolites driving spore discrimination. (a) Heatmap of the top 20 annotated m/z ratios ranked by Gini importance from the random forest model. m/z values 428.37344, 400.34211, and 383.25811 (bold) correspond to protonated molecular species ([M+H]⁺) correspondent to metabolites consistent with the bacterial/spore biological framework. The first two were annotated as acylcarnitines, whereas m/z 383.25811 corresponds to two C25 β‐apo‐12′‐carotenal isomers. Heatmap colors indicate relative metabolite abundance (blue, lower; red, higher). (b) Chemical formulas and generalized structures of the annotated metabolites, including positional methyl isomers of C17 acylcarnitine (methylheptadecanoylcarnitine, C3–C16), positional methyl isomers of C15 acylcarnitine (methylpentadecanoylcarnitine, C3–C14), L‐ and O‐palmitoylcarnitine (C16), and stereochemical/geometrical isomers of C25 β‐apo‐12′‐carotenal ((3S,5R,6R)‐3,5‐dihydroxy‐6,7‐didehydro‐5,6‐dihydro‐12′‐apo‐β‐caroten‐12′‐al and (3S,5R,6S)‐5,6‐epoxy‐3‐hydroxy‐5,6‐dihydro‐12′‐apo‐β‐caroten‐12′‐al). Wavy bonds indicate positions of methyl branches along the fatty acyl chains.

Lastly, the presence of diterpenoids, particularly carotenoids, has been widely reported as a key mechanism contributing to spore resistance against ultraviolet (UV) radiation. As membrane‐associated antioxidants, carotenoids can scavenge reactive oxygen species (ROS) generated upon irradiation, thereby protecting essential cellular components from oxidative damage [56]. In Bacillus species, C30 glycosylated apocarotenoids are among the most frequently reported carotenoid derivatives and are commonly associated with yellow to orange pigmentation [57]. Remarkably, this study identifies two novel oxygenated C25 β‐apo‐12′‐carotenal isomers (C₂₅H₃₄O₃). These compounds were uniquely enriched in the dairy‐associated reference strain J630 and in the diarrheal reference strain J649 (Figure 5a,b).

The enrichment of these C25 apocarotenoids may reflect the incorporation and subsequent oxidative cleavage of dietary β‐carotene present in food matrices. This hypothesis is particularly plausible for strain J630, which was isolated from dairy products known to contain bioavailable all‐trans‐β‐carotene. Sporulating cells in carotenoid‐rich environments may cleave β‐carotene via carotenoid oxygenases, retaining oxidized apo‐carotenoid derivatives within the spore membrane. Such retention could enhance oxidative and UV resistance during environmental persistence. Although the original isolation source of J649 is unknown, the presence of similar apo‐carotenoid derivatives suggests adaptation to foodborne niches rich in carotenoids. In the case of J649, this enrichment may not only contribute to increased UV resistance but could also enhance resistance to oxidative stress encountered during host colonization, including ROS generated by phagocytic immune cells in the intestinal environment. Similar antioxidant‐mediated protective mechanisms have been reported in other bacterial species [56].

Overall, metabolites such as acylcarnitines and diterpenoids contribute to the metabolic discrimination of spores. Pathogenic strains (B. cytotoxicus, B. cereus J648 and J649) lack medium‐ to long‐chain acylcarnitines, whereas nonpathogenic strains, such as B. cereus J630, contain higher levels of these metabolites, highlighting their potential relevance during outbreak investigations. In addition, the sporulation environment strongly shapes the metabolic signature of spores, which may help distinguish naturally occurring strains from those resulting from intentional dissemination. Beyond established biomarkers, the identification of metabolic outliers may further support the discrimination of laboratory‐produced ( B. cereus J630, J648, J649 and B. cytotoxicus) spores from natural ones ( B. cereus J3 and J271), emphasizing the value of metabolomic profiling for spore characterization.

4. Conclusion

The inherent similarity between bacterial species can lead to major biotyping challenges, particularly in implementing effective prevention and containment measures. The B. cereus group, a spore‐forming bacterial genus, exemplifies this challenge due to its wide range of pathogenic and nonpathogenic species, which are genetically indistinguishable using conventional methods such as genome sequencing. In this work, using FT‐ICR‐MS–based untargeted metabolomics, we show that B. cereus s.l. spores possess a core metabolome shaped by the sporulation environment while retaining sufficient metabolic specificity to enable species and strain discrimination. Lipids, amino acids/peptides, carbohydrates, terpenoids, and aromatic oxygenated compounds were consistently detected across species, suggesting the existence of a shared metabolic framework supporting spore resistance. Clear strain‐level segregation was observed. Wild isolates B. cereus J3 and J271 exhibited the highest metabolic richness and the most complex chemical profiles, whereas reference strain J648 showed the lowest diversity and lacked strain‐specific metabolites. B. cytotoxicus J650 displayed an intermediate metabolic profile, clustering more closely with the food‐derived strain J271 than with other B. cereus reference strains. Strain J649 was distinguished by its lipid profile, clustering with B. cytotoxicus based on similar HMDB weight distributions, while J630 was characterized by elevated levels of medium‐ to long‐chain acylcarnitines and unique diterpenoid signatures. Collectively, these features enabled discrimination of B. cereus strains and B. cytotoxicus based on reproducible metabolic signatures.

Overrepresentation analysis highlighted lipids as key contributors to spore physiology, not only as structural membrane components but also as mediators of environmental sensing and stress defense. The incorporation of exogenous lipids into spore structures may reflect a metabolic imprint of the original isolation environment, offering a potential indicator for distinguishing naturally occurring strains from laboratory‐produced ones. Additionally, the detection of medium‐ to long‐chain acylcarnitines and their association with pathogenicity suggests a possible strategy for virulence assessment.

Altogether, these findings demonstrate the potential of untargeted metabolomics as a promising strategy for bacterial spore biotyping. The identification of this specific cohort of metabolites provides a strong basis for spore classification and paves the way for methodologic transfer to more widely available MS instruments.

Conflicts of Interest

The authors declare no conflicts of interest.

Romão M. I., Gomes I., Antunes W., Novak P., Sousa Silva M., and Cordeiro C., “Decoding Biothreats With FT‐ICR‐MS: Metabotyping of Bacillus cereus Spores Through Untargeted Metabolomics,” Journal of Mass Spectrometry 61, no. 7 (2026): e70073, 10.1002/jms.70073.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.