The bacterial strains used in this work were isolated from the endosphere of flowers, leaves, and stems of the officinal plant O. heracleoticum, as described in Semenzato et al. (2023) [32]. Strains are referred to as OH (O. heracleoticum), followed by the letters F, L, and S (for Flowers, Leaves, or Stems, respectively), and numbered (Table 1). The taxonomic affiliation of the bacteria was previously obtained via amplification and sequencing of the 16S rRNA coding gene; GenBank accession numbers are available in Semenzato et al. (2023) [32], except for strains OHF10 (OR880887), OHS9 (OR880888), and OHS20 (OR880889). The endophytes were stored in a 20% glycerol (453752, Carlo Erba, Milan, Italy) stock at –80 °C. They were grown on Tryptic Soy Agar (TSA, Oxoid LTD, Hampshire, UK) at 30 °C for 48 h.

Eleven strains of the Burkholderia cepacia complex (Bcc) belonging to four different species were selected on the basis of their origin, i.e., a clinical (Cystic Fibrosis patients, CF) or an environmental (ENV) one (Table 2). Each strain was grown on LB agar medium (NaCl 10 g/L, yeast extract 5 g/L, tryptone 10 g/L, agar 15 g/L, Oxoid LTD) at 37 °C for 48 h.

The other 36 pathogenic strains used in this work were isolated from different sources (hospital devices, foods, patients, healthy subjects, and the environment) and were previously characterized for their resistance to multiple antibiotics, using the disk-diffusion method [37]. Staphylococcus aureus, Coagulase-Negative Staphylococci (CoNS), Pseudomonas aeruginosa, and Klebsiella pneumoniae strains were provided by the Applied Microbiology laboratory (Health Sciences Department, University of Florence, Italy), while the standard bacteria S. aureus ATCC 25923, P. aeruginosa ATCC 27853, and K. pneumoniae ATCC 700603 were obtained from Thermo Fisher Diagnostics S.p.A. All the strains were grown on TSA plates at 37 °C for 24 h.

Endophytes antibacterial activity against Bcc strains was evaluated through the cross-streaking method (Supplementary Fig. 1). Petri dishes with or without a septum separating the two compartments (to permit the growth of the tester and the target strains without any physical contact) were used. Tester strains (i.e., the O. heracleoticum endophytes) were streaked across one half of a TSA plate and grown at 30 °C for 48 h, to allow the synthesis of antibacterial compounds. Single colonies of each target strain were then suspended in 50 µL of a 0.9% NaCl w/v solution. Target strains belonging to the Bcc were then streaked perpendicularly to the tester strain and plates were incubated at 30 °C for a further 48 h. Additionally, Bcc strains were streaked on half of a Petri plate in the absence of the tester and were allowed to grow at 30 °C for 48 h (positive growth control). The antagonistic effect was evaluated as the reduction or absence of the target strains growth compared to control plates. The different inhibition levels were indicated as follows: complete (3), strong (2), weak (1), and absence (0) of inhibition.

Bacterial suspensions of the other MDR pathogenic strains were prepared as mentioned before and then streaked perpendicularly to the tester strain using an inoculation needle; plates were incubated at 37 °C for a further 24 h. Additionally, target strains were grown at 37 °C for 24 h in the absence of the tester (positive growth control). The antagonistic effect was evaluated as described above. The different inhibition levels were indicated as follows: complete (4), strong (3), moderated (2), weak (1), and absence of inhibition (0).

The Inhibition Score (IS) and Sensitivity Score (SS) were calculated for each tester and target strain, respectively, as the sum of the values obtained in each antagonism experiment; the total IS for each plant compartment and the total SS for CF and ENV Bcc groups, and for each MDR target group were calculated as the sum of all IS/SS, normalized by the number of tester/target strains per group (NIS and NSS). The average value of inhibition towards each MDR target group was calculated as the ratio between the sum of all IS and the number of the target strains in each group (TIS $\overline{X}$); the total average value (TOT $\overline{X}$) was calculated for each tester by dividing the total IS by the total number of MDR target strains. ANOVA was performed using R’s aov() function. The degree of significance was set at p 0.05. The heatmaps were obtained using the R package pheatmap [38].

Phylogenetic trees were constructed in MEGA XI [39] by aligning, using the Muscle algorithm, the 16S rRNA gene sequences of endophytic strains belonging to the Bacillus and Arthrobacter-Pseudarthrobacter genera with 35 type strains’ 16S rRNA gene sequences downloaded from the Ribosomal Database Project (RDP), selecting those exhibiting a higher percentage of identity with the endophytes sequences ($\ge$90%) [40]. The alignment was used to build the phylogenetic trees by applying the Neighbor-Joining algorithm, with a 1000-bootstrap resampling.

Biomass obtained from bacterial culture (30–100 mg) was collected in Head Space (HS) vials. Vials were sealed and immediately analyzed. HS vials were conditioned at 40 °C for 20 minutes before extraction. Bacterial volatile organic compounds (VOCs) produced by bacterial strains were extracted from the vial headspace and injected into the Gas chromatograph (GC). Headspace extraction was performed with a 2.5 mL Syringe-HS (0.64-57-R-H, PTFE, GERSTEL) conditioned and held at 40 °C from sample collection to injection. Splitless injection was used.

Gas chromatographic analysis was performed adapting a previously reported method [41]. In detail, an Agilent 7000C GC (Agilent Technologies, Inc., Santa Clara, CA, USA) system was used, equipped with a split/splitless injector, fitted with an Agilent HP5-MS UI capillary column (30 m $\times$ 250 µm; 0.25 µm film thickness), coupled to an Agilent triple quadrupole Mass Selective Detector MSD 5973 (Agilent Technologies, Inc., Santa Clara, CA, USA), with ionization voltage 70 eV; electron multiplier energy 2000 V; transfer line temperature, 270 °C; solvent Delay: 0 min. Helium was used as the carrier gas (1 mL/min). Concerning the oven program, the temperature was initially kept at 40 °C for 5 min, then gradually increased to 250 °C at a 2 °C/min rate; this temperature was held for 15 min and finally raised to 270 °C at a 10 °C/min rate. Samples were injected at 250 °C automatically. Interval scan: 35–450 m/z; Scan speed: 10,000 amu$\cdot$s${}^{-1}$ (25 Hz). The GC–MS mass spectrum data were analyzed using MassHunter Qualitative Analysis B.06.00 and the database of the National Institute Standard and Technology (NIST) was used to interpret the analyzed data. A comparison between the mass spectrum of the unidentified components released by the bacterial isolates and the mass spectrum of already known components available in the NIST 11 MS library was carried out (Match factor $>$800).

We first checked the ability of the entire panel of bacterial endophytes to inhibit the growth of eleven bacterial strains belonging to the Bcc, bacterial strains known for their resistance to numerous conventional antibiotics. Bcc species are responsible for infections in Cystic Fibrosis (CF) patients, leading to severe health complications such as necrotizing pneumonia and sepsis, reducing patients’ survival rates [42]. Treatment of Bcc infections is challenging because of their intrinsic resistance to various antibiotics and their tolerance to antibiotic exposure, especially in biofilm formations [43]. Current therapeutic approaches lack evidence-based guidelines and often follow various antibiotic protocols. However, the complete eradication of Bcc infection remains difficult, prompting the exploration of alternative strategies, such as the identification of compounds able to enhance antibiotic activity by targeting resistance mechanisms, or the employment of alternative therapies such as quorum-sensing inhibitors, natural antimicrobial peptides, and specifically designed vaccines [44].

Data obtained are shown in Fig. 1, whose analysis revealed that almost all the endophytes (with few exceptions, see for instance Peribacillus sp. OHF10 and Arthrobacter sp. OHS14) were able to inhibit the growth of at least two Bcc strains, even though at a different extent: some of them were able to completely interfere with the growth of the entire panel of Bcc strains (Bacillus sp. OHF9, Cytobacillus sp. OHF13, and Pseudarthrobacter sp. OHS10), while others exhibited a moderate/low degree of inhibition. In general, O. heracleoticum-associated strains were much more able to reduce or inhibit the growth of Bcc strains isolated from CF patients (NSS = 115) than those with an environmental origin (NSS = 72). This was previously observed for medicinal plants-associated bacteria isolated from different plant compartments and rhizospheric soil of Origanum vulgare L., Lavandula angustifolia Mill., and Echinacea purpurea L. [45, 46, 47]. The most sensitive strain was FCF3 (SS = 153), while LMG1222 (SS = 56) was the most resistant. Although the two strains belong to the same species (B. cepacia), their behavior in response to the antagonistic activity of the endophytes was completely different, suggesting the relevant role of the strains origin in determining their resistance profiles [48]. This finding is quite interesting, and it is worth further investigation. In our opinion, CF Bcc isolates obtained from healthcare settings may have been exposed to specific antibiotics or antimicrobial agents, potentially influencing their resistance profiles. In contrast, ENV strains might face a different selective pressure, including exposure to various stressors and competition, which could influence their differential response to antibacterial compounds [49]. Over time, these strains may have developed mechanisms to survive in the presence of antimicrobial agents found in their natural habitat, leading to higher resistance levels compared to clinical isolates.

The analysis of the NIS revealed that the flowers compartment hosted the community with the highest inhibitory potential against the Bcc strains, with a NIS of 21, compared to the leaves and stems compartments (NIS = 17). Concerning the single strains, the highest IS (from 33 to 26) were registered for strains belonging to the genera Arthrobacter (OHF5, OHF22, OHL24), Pseudarthrobacter (OHF15, OHF21, OHF24, OHS3, OHS10), Bacillus (OHL23, OHS8), Cytobacillus (OHF13), Peribacillus (OHF2B), Priestia (OHF7, OHF9), Pantoea (OHS9), and Curtobacterium (OHS12). It is worth noting that the majority of the most active strains are Gram-positive, belonging to the Phyla Firmicutes and Actinobacteria, except for Pantoea sp. OHS9. It is widely recognized that the phylum Actinobacteria is responsible for over half of the natural bioactive compounds documented in literature surveys, encompassing antibiotics, immunosuppressive agents, antitumor agents, and enzymes [50]. They can be isolated from various environmental sources, such as soil samples, plants, and the marine environment, and in many cases, their potential to attenuate and/or inhibit the growth of Gram-negative and Gram-positive human pathogenic strains was documented [51, 52]. Antibacterial activities were also reported for the Phylum Firmicutes, especially for isolates belonging to the genus Bacillus [53, 54]. A previous work on bacterial endophytes isolated from O. vulgare L. reported similar results; indeed, the most active endophytic strains were all Gram-positive (with few exceptions), many of which belonged to the genera Arthrobacter and Bacillus [45]. Consistently, the most active bacterial endophytes associated with E. purpurea rhizospheric soil were affiliated to the genus Arthrobacter [47]. From these first observations, we can hypothesize that the different antibacterial patterns observed for O. heracleoticum-associated bacterial strains seem not related to the ecological niche from which they were isolated but might be linked to their taxonomical affiliation at the genus level.

The same results are also reported in Fig. 2. In this case, the tester and target strains were sorted based on hierarchical clustering (dendrograms on the top and the left of the heatmap), highlighting well-defined groups of strains.

Target strains (columns) were divided into three clusters: the first one, on the left, groups the most sensitive strains, all of clinical origin (B. cepacia FCF3, B. cenocepacia LMG21465, and LMG506); the middle one includes three CF strains and one ENV strain (B. ambifaria LMG19182), which were strongly or completely inhibited by around 60% of tester endophytic strains; finally, the last group (on the right of the heatmap) is mostly composed of ENV strains, except for B. multivorans LMG13010, and include the most resistant Bcc strains. Hence, the hierarchical clustering confirmed that environmental strains are generally more resistant than clinical isolates (p = 0.02, ANOVA test).

Concerning the endophytes’ clustering (rows), the first group is formed by the most active tester strains, already listed in the previous paragraph; the second group consists of strains that were able to strongly or completely inhibit Bcc strains included in the first two target clusters; the third one groups five endophytes with very low or no antagonistic effect on Bcc strains; lastly, the fourth group includes strains with low IS.

For each endophyte, the taxonomic affiliation at the genus level and the compartment of origin were color-coded on the left of the heatmap. 56.3% of endophytic strains belonging to the first group were isolated from the flowers compartment (9 out of 16 strains), while only two strains (12.5%) belonged to the leaves-associated endophytic community. The nine OHF strains belong to the genera Arthrobacter, Pseudarthrobacter, Bacillus, Cytobacillus, Peribacillus, and Priestia, the latter four all belonging to the family Bacillaceae. On the other hand, 4 out of 5 endophytes (80%) of the third group were associated with the stems compartment and belonged to the genera Pseudomonas, Bacillus, and Arthrobacter. From these first observations, we can hypothesize that the different antibacterial patterns observed for O. heracleoticum-associated bacterial strains seem not related to the ecological niche from which they were isolated but might be linked to their phylogenetic relatedness. For example, Pseudarthrobacter sp. OHF24 and OHS10, isolated from flowers and stems, respectively, have similar activity against Bcc target strains and belong to the same species [32]. Indeed, the ANOVA test (Supplementary Table 1) confirmed that the IS obtained for each endophytic strain was significantly related to the taxonomic affiliation of the strain at the genus level (p 0.05), but not to the compartment from which they were isolated. A non-significant interaction between the two factors was also observed. Thus, taxonomic affiliation is apparently involved in determining endophytes antagonistic activity against Bcc strains, but it cannot be excluded an interplay between taxonomy and ecological niche; indeed, it has been demonstrated that the composition of the bacterial communities living inside plants’ different compartments is not casual, and this might determine different antagonistic patterns in different anatomical parts [47, 55].

To ascertain which of these two parameters, i.e., the “ecological niche” or “taxonomic affiliation”, is strongly associated with the ability of endophytes to inhibit Bcc strains, our attention focused on the two main phylogenetic subgroups: the first is composed by 13 strains belonging to the genera Arthrobacter and Pseudarthrobacter, while the second one includes 16 strains belonging to the genus Bacillus. For each of the two subgroups, a phylogenetic tree based on 16S rRNA gene sequences and a heatmap portraying cross-streaking results against Bcc strains were obtained (Figs. 3 and 4).

Both subgroups include strains with (very) dissimilar inhibition patterns isolated from different plant organs. Concerning the Arthrobacter subgroup heatmap (Fig. 3B), tester strains, sorted based on hierarchical clustering, were divided into 5 clusters of strains, formed by (i) OHS10, OHF24, and OHL24, which exhibited the strongest antibacterial activity towards both CF and ENV Bcc strains; (ii) OHS3, OHF22, OHF21, OHF5, and OHF15, which were slightly less efficient against ENV Bcc strains; (iii) OHF17, OHL4, and OHL10, with reduced anti-Bcc activity compared to the previous two groups; (iv) OHL14 strain, which was able to interfere with the growth of Burkholderia strains FCF3, LMG21462, LMG24506, and LMG16656 only; (v) OHS14, with no antibacterial activity. Concerning the Bacillus subgroup (Fig. 4B), the 4 clusters are composed of (i) OHS8 and OHL23, which strongly or completely inhibited the growth of all target strains; (ii) OHF16, OHF1, OHL12, OHF2C, OHL2, OHS5, OHF2A, and OHF3, which were active only towards clinical target strains; (iii) OHF11 and OHS18, whose antagonistic effect was limited to Burkholderia strains FCF3, LMG21462, and LMG24506; (iv) OHF20, OHS4, OHF6, and OHS2, which have no or low antagonistic effect.

As can be seen from the annotation on the left of the heatmaps, there is no evident clustering of strains based on the compartment of origin (p $>$0.05, ANOVA test), except for the Arthrobacter cluster II, where 4 out of 5 strains were isolated from the flowers compartment. Observing the Arthrobacter and Pseudarthrobacter phylogenetic tree (Fig. 3A), only a few strain clusters are evident. One of them is formed by the endophytic strains OHF5 and OHS3: they belong to the same Operational Taxonomic Unit (OTU, percentage of identity = 99%) and both obtained an IS of 28, indeed they are grouped in the same section of the heatmap. However, on the other hand, strains OHS14 and OHF22 belong to the same OTU (percentage of identity = 99%) but have completely different inhibitory activities against Bcc strains (IS${}_{\text{OHF22}}$ = 27; IS${}_{\text{OHS14}}$ = 0). A similar pattern can be found also in the Bacillus subgroup. Strains OHL23 and OHS2 belong to the same OTU (percentage of identity = 99%), but the obtained IS for the two strains are divergent (IS${}_{\text{OHL23}}$ = 30; IS${}_{\text{OHS2}}$ = 4). Nevertheless, Bacillus strains OHL18 and OHL11, which are the only strains included in the heatmap cluster III, are in the same OTU (percentage of identity = 99%), and the same can be observed for strains OHF6, OHF20 and OHS4 (IS ranging from 4 to 7 and all belonging to cluster IV of the heatmap), which, however, share the same OTU with OHL12, OHF3, OHS5, and OHF1, strains with higher IS (ranging from 19 to 22 and all belonging to cluster II of the heatmap). In general, the inhibition pattern varied between strains probably belonging to the same species, suggesting that antibacterial activity might be due to different inhibitory mechanisms within the same species and so be related to intra-specific genomic differences.

The observed antagonistic activity of endophytes might be attributed to the production of diffusible but also volatile molecules with antibacterial activity. Since volatile organic compounds (VOCs) are involved not only in inter/intra-species communication but also in antagonism between microbes [56], the production of VOCs was analyzed on a smaller subset of 12 endophytes, all clustered in the first group of the heatmap (Fig. 2), isolated from all three compartments, and belonging to the genera Arthrobacter (OHF5, OHF24, and OHL24), Pseudarthrobacter (OHF15, OHF21, OHS3, and OHS10), Bacillus (OHL23 and OHS8), Priestia (OHF7 and OHF9) and Curtobacterium (OHS12). The cross-streaking tests were performed on Petri dishes divided into two halves by a physical septum, not allowing the diffusion of antibacterial molecules from the tester to the target strains. The obtained results are reported in Fig. 5.

All the selected strains, apart from Curtobacterium sp. OHS12, were able to strongly or completely antagonize the growth of CF and ENV Bcc strains. Due to the presence of the physical septum, the antibacterial activity of these endophytes could be attributed solely to the production of VOCs. Also in this case, the antibacterial effect was stronger towards CF targets but still substantial against ENV Bcc strains. Overall, the inhibition patterns and the IS obtained in this experiment were the same or slightly lower than the ones obtained using plates without septum. However, three strains exhibit interesting discrepancies with respect to the IS shown in the cross-streaking experiments performed with Petri dishes without a septum. The two Priestia strains (OHF7 and OHF9), exhibiting an IS of 29 and 33, respectively, in the previous experiments, showed a decreased IS, which is particularly marked for Bcc strains of environmental origin. The third strain, Curtobacterium sp. OHS12, did not affect the growth of Bcc strains through the production of VOCs.

The most active strains, scoring an IS of 30, were Arthrobacter sp. OHF24 and OHL24, isolated respectively from the flowers and the leaves compartment and not close phylogenetically (Fig. 3A). In a previous work, the antibacterial potential of VOCs produced by a panel of 8 endophytic strains isolated from O. vulgare was evaluated against 10 Bcc strains [41]. Also in this case, the most active strain, OVS8, belonged to the genus Arthrobacter, which was also able to interfere with the growth of human MDR Klebsiella pneumoniae strains through the production of VOCs [57]. Within the Actinomycetes phylum, research studies mainly focused on the genus Streptomyces, which is widely known for its remarkable ability to produce a wide array of antibiotics and other bioactive compounds and thus employed in various medical, agricultural, and industrial applications [58]. Indeed, there is still limited evidence regarding the antibacterial compounds synthesized by Arthrobacter species, with most studies focusing, to date, on strains isolated from Antarctic or marine environments [51, 59, 60]. The results here obtained suggest the potentiality of Arthrobacter strains isolated from medicinal plants as a promising source of antibacterial molecules of diffusible and volatile nature.

The endophytic strains showing the highest inhibitory activity against Bcc strains were also tested against 36 MDR human pathogens, belonging to S. aureus, P. aeruginosa, K. pneumoniae, and CoNS groups. The strains were selected because of their resistance to multiple antibiotics (Table 2). Moreover, the interest in these species is validated by their inclusion in the AR-ISS program in Italy, which actively participates in the European Antimicrobial Resistance Surveillance Network (EARS-Net), overseen by the European Centre for Disease Prevention and Control (ECDC). The primary goal of AR-ISS surveillance is to describe the antibiotic resistance within a specific group of pathogens isolated from invasive infections belonging to eight species (Staphylococcus aureus, Streptococcus pneumoniae, Enterococcus faecalis, Enterococcus faecium, Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, and Acinetobacter species), to provide insights into the current status of antibiotic resistance and contribute in global efforts to address this growing concern in healthcare [61]. The same pathogenic species were previously employed in cross-streaking experiments aimed at demonstrating the antibacterial potential of O. vulgare bacterial endophytes; some of them, belonging to the genera Bacillus and Arthrobacter, were mostly active towards CoNS and S. aureus strains [45].

The antibacterial activity of the target strains was evaluated qualitatively, as described in the previous paragraphs. Data shown in Fig. 6 revealed that the selected endophytes reported the highest antimicrobial activity against the CoNS group (except for strain 5419, isolated from a food sample), with a NSS = 27, followed by K. pneumoniae (NSS = 26) and S. aureus (NSS = 16) groups. On the contrary, P. aeruginosa strains were the most resistant (NSS = 8). Interestingly, the most inhibited strains belonging to the CoNS group are the ones of human origin, followed by strains isolated from hospital devices. This is quite relevant considering the potential applications of the 12 endophytic strains as antibiotics producers able to face the spread of MDR human pathogens. As verified by the ANOVA analysis, MDR target strains’ sensitivity was significantly related to both taxonomic affiliation and isolation source (p 0.001).

Contrary to what was observed for the Bcc group, the flowers compartment hosted the community with the lower inhibitory potential, with a NIS of 45, compared to the leaves (NIS = 58) and stems compartments (NIS = 50). The highest IS was reported for the tester strain Bacillus sp. OHS8 (IS = 80), which was able to completely inhibit the growth of almost all the CoNS strains and induced a strong-moderate inhibition against K. pneumoniae and S. aureus groups. Apart from OHS8, the highest IS (from 56 to 66) were registered for strains belonging to the genera Pseudarthrobacter (OHF21, OHF24, OHS3), Arthrobacter (OHF5, OHL24), and Bacillus (OHL23). In particular, Bacillus sp. OHL23 was the only endophyte able to strongly inhibit the growth of all the S. aureus strains. On the other hand, the lowest antibacterial activity was reported for the two Priestia strains (OHF7 and OHF9), which had almost no antagonistic effect on P. aeruginosa and S. aureus groups.

Based on these results, 10 endophytic strains were tested for their ability to produce VOCs capable of antagonizing the growth of the 36 MDR human pathogenic strains, using Petri dishes with a septum physically separating the plate into two compartments, as described in the Materials and Methods section. Data obtained are shown in Fig. 7.

In general, O. heracleoticum endophytes antibacterial activity was lower compared to the previous cross-streaking test results. Tester strains VOCs exhibited a moderate antagonistic effect towards the CoNS strains (NSS = 15), especially against S. epidermidis 5318 and 5403, both of human origin. The highest IS, ranging from 42 to 39, were registered for strains belonging to the genera Arthrobacter (OHF5, OHL24) and Pseudarthrobacter (OHF24).

Concerning the cross streaking performed against MDR pathogenic strains, the total inhibition score (TIS) and the average value of TIS for each target group ($\overline{X}$) and the total average value for each tester (TOT $\overline{X}$) were calculated (Tables 3 and 4). The tester strains reported the highest antimicrobial activity against the CoNS group in both experiments (the $\overline{X}$ value ranged from 0.9 to 3.8 without septum; from 0.2 to 2.8 with septum), while S. aureus strains in cross-streaking experiments without septum ($\overline{X}$value from 0.0 to 1.4), and the K. pneumoniae group when using Petri dishes with a septum ($\overline{X}$value from 0.0 to 0.8) were only slightly inhibited.

The VOCs produced by the most interesting strains were analyzed using HS-GC/MS. The identified VOCs are listed in Table 5. Results are expressed as the area of each peak in the chromatograms normalized by the bacterial biomass. Headspace-GC (HS-GC/MS) was chosen since it reduces sample manipulation, does not require the use of solvents, and because of its easy preparative steps [41]. A total of 16 distinct metabolites were detected, mainly belonging to the following structurally distinct classes: alcohols (1-Butanol, 1-Butanol-3-methyl, 1-Hexanol, 2-ethyl-, 2-Propanol), ketones (2-Butanone, 2-Butanone, 3-methyl-, Acetone), hemiterpenes (isoprene), and sulphurated compounds (Bis(methylthio) methane, Carbon disulfide, Dimethyl sulfide, Dimethyl disulfide, Dimethyl trisulfide, Ethanethioic acid S-methyl ester, Metanthiol, Thiophene) (see Table 5).

Among all strains, Arthrobacter sp. OHL24 exhibited the highest VOCs content based on the normalized area of all peaks, followed by Priestia strains OHF9 and OHF7 (Fig. 8).

Based on the relative peak area, the compound dimethyl disulfide was the most abundant volatile for Arthrobacter sp. OHL24. Among sulfides, dimethyl sulfide was produced by almost all strains. It was demonstrated that dimethyl disulfide induced significant growth inhibition of different microbial pathogens, such as Rhizoctonia solani and Pythium ultimum, or many Gram-negative and Gram-positive bacteria [62].

For Pseudarthrobacter strains OHF15, OHF21, OHF24, and OHS10, 2-propanol was the most abundant compound, while it was acetone for Priestia sp. OHF7 and sp. OHF9, and Bacillus sp. OHS8. Notably, Priestia strains OHF7 and OHF9 produced acetone as the main compound ($>$93% total VOCs). 1-Hexanol-2-ethyl was the main component for Arthrobacter sp. OHF5, Bacillus sp. OHL23 and Pseudarthrobacter sp. OHS3. The distribution of the different classes of metabolites among all strains is shown in Supplementary Figs. 2–5.

The general antibacterial activity of the endophytic strains can be correlated to the production of thiophene, while the higher activity of Arthrobacter sp. OHL24 against S. aureus strains could be ascribed to the production of sulphurated compounds.