Exploring the Interspecific Interactions and the Metabolome of the Soil Isolate Hylemonella gracilis

ABSTRACT Microbial community analysis of aquatic environments showed that an important component of its microbial diversity consists of bacteria with cell sizes of ~0.1 μm. Such small bacteria can show genomic reductions and metabolic dependencies with other bacteria. However, so far, no study has investigated if such bacteria exist in terrestrial environments like soil. Here, we isolated soil bacteria that passed through a 0.1-μm filter. The complete genome of one of the isolates was sequenced and the bacterium was identified as Hylemonella gracilis. A set of coculture assays with phylogenetically distant soil bacteria with different cell and genome sizes was performed. The coculture assays revealed that H. gracilis grows better when interacting with other soil bacteria like Paenibacillus sp. AD87 and Serratia plymuthica. Transcriptomics and metabolomics showed that H. gracilis was able to change gene expression, behavior, and biochemistry of the interacting bacteria without direct cell-cell contact. Our study indicates that in soil there are bacteria that can pass through a 0.1-μm filter. These bacteria may have been overlooked in previous research on soil microbial communities. Such small bacteria, exemplified here by H. gracilis, can induce transcriptional and metabolomic changes in other bacteria upon their interactions in soil. In vitro, the studied interspecific interactions allowed utilization of growth substrates that could not be utilized by monocultures, suggesting that biochemical interactions between substantially different sized soil bacteria may contribute to the symbiosis of soil bacterial communities. IMPORTANCE Analysis of aquatic microbial communities revealed that parts of its diversity consist of bacteria with cell sizes of ~0.1 μm. Such bacteria can show genomic reductions and metabolic dependencies with other bacteria. So far, no study investigated if such bacteria exist in terrestrial environments such as soil. Here, we show that such bacteria also exist in soil. The isolated bacteria were identified as Hylemonella gracilis. Coculture assays with phylogenetically different soil bacteria revealed that H. gracilis grows better when cocultured with other soil bacteria. Transcriptomics and metabolomics showed that H. gracilis was able to change gene expression, behavior, and biochemistry of the interacting bacteria without direct contact. Our study revealed that bacteria are present in soil that can pass through 0.1-μm filters. Such bacteria may have been overlooked in previous research on soil microbial communities and may contribute to the symbiosis of soil bacterial communities.

IMPORTANCE Analysis of aquatic microbial communities revealed that parts of its diversity consist of bacteria with cell sizes of ;0.1 mm. Such bacteria can show genomic reductions and metabolic dependencies with other bacteria. So far, no study investigated if such bacteria exist in terrestrial environments such as soil. Here, we show that such bacteria also exist in soil. The isolated bacteria were identified as Hylemonella gracilis. Coculture assays with phylogenetically different soil bacteria revealed that H. gracilis grows better when cocultured with other soil bacteria. Transcriptomics and metabolomics showed that H. gracilis was able to change gene expression, behavior, and biochemistry of the interacting bacteria without direct contact. Our study revealed that bacteria are present in soil that can pass through 0.1-mm filters. Such bacteria may have been overlooked in previous research on soil microbial communities and may contribute to the symbiosis of soil bacterial communities. KEYWORDS Hylemonella sp., interspecific interactions, transcriptome analysis, metabolome analysis, Gram-negative bacteria, metabolomics, transcriptomics, volatile organic compounds B acteria are ubiquitous living organisms with various cell shapes and sizes surrounding us in all environments (1,2). Soil is the most complex habitat harboring the largest diversity and density of bacteria known to date (cell densities ranging from 10 7 to 10 10 cells/g of soil) (3)(4)(5). Soil bacteria are part of a community where they are in constant interaction with their own and other species (6)(7)(8). Bacteria produce and release a plethora of metabolites into their environment. In this way, they not only chemically modify their niche but also affect the behavior and the secondary metabolite production of nearby bacteria (9)(10)(11). Soil bacteria are known to produce a wide range of soluble and volatile secondary metabolites with different physicochemical and biological properties (7,(12)(13)(14). In contrast to soluble compounds, volatile organic compounds (VOCs) are rather small molecules (,300 Da) that can diffuse easily through air-and water-filled soil pores (15)(16)(17). These physicochemical properties make VOCs ideal metabolites for long-distance communication and interactions between soil microorganisms (18)(19)(20)(21).
In aquatic environments, bacteria are naturally found at lower cell densities compared to soil (10 3 to 10 6 cells/mL) (22)(23)(24). Recent studies have shown that a significant component of aquatic microbial diversity consists of bacteria with small cell sizes of about ;0.1 mm (25)(26)(27). However, little is known about bacteria with such cell sizes in soil environments, for instance in water-filled soil pores. There are indications that ultra-small bacteria might exist in rice paddy soils as described by Jansen (28) as "dwarf cells" and in other soil environments (29).
One can assume that a small cell size can be an advantage in challenging environments like soil. However, the distribution of microorganisms in soil is influenced by its water and moisture content, and a low soil moisture content leads to lower connectivity between soil pores, and thus to a lower number of accessible microhabitats.
Small bacterial cell size is often linked to a small genome size caused by genome streamlining (30). Recent metagenomic studies suggest that genome streamlining is ubiquitous in bacteria (31,32). In some cases, the primary metabolism of one organism can be directly built on the primary metabolism of another organism, known as syntrophic relationships (33,34). The Black Queen Hypothesis states that genome-streamlined organisms have an evolutionary advantage because of the loss of genes whose function can be replaced by bacteria in the surrounding environment, effectively conserving energy (35). Since bacteria with fewer genes have less adaptive capacity compared to bacteria with more genes, many of them are expected to depend on specific environmental conditions or on the presence of other specific organisms (36) to produce metabolites that support their persistence.
Here, we aimed to explore if bacteria that can pass through 0.1-m filters are present in soil, and if such bacteria are cultivable. We further investigated their interaction with phylogenetically different bacteria commonly occurring in soil. The major research questions were if, and how interspecific interactions between bacteria that pass a 0.1mm filter and other common soil bacteria that cannot pass 0.1-mm filters affected their fitness, behavior, gene expression, and the production of secondary metabolites.

RESULTS
Isolation and identification of bacteria that pass through 0.1-lm filter. We isolated bacteria from a terrestrial soil sample that were able to pass through 0.22-mm and 0.1-mm pore size filters. After several days of incubation, only one type of bacterial colonies was observed on the inoculated plates. The grown colonies were identified as Hylemonella gracilis (Gram-negative; class, Betaproteobacteria; order, Burkholderiales) by 16S rRNA sequence analysis.
The colonies showed a round and colorless morphology when grown on 1/10th TSBA plates (Fig. 1A). Microscopically, the bacteria had a spiraled morphology with a length of approximately 6 to 12 mm, which is typical for Hylemonella species (Fig. 1B).
Hylemonella grows better in interaction with other bacteria. To test the hypothesis that small bacteria grow better in the presence of normal-sized bacteria, growth of H. gracilis was determined in coculture with two phylogenetically distantly related soil bacteria (Paenibacillus sp. AD87 and Serratia plymuthica PRI-2C) and compared to that of the monoculture. The bacterial CFU of H. gracilis (CFU/mL) obtained on 1/10th TSBA plates from monocultures and cocultures are summarized in Fig. 2. Cell counts of Paenibacillus sp. AD87 were 7.68 Â 10 7 CFU/mL in coculture with H. gracilis ( Fig. 2A). During the interaction with H. gracilis, the growth of S. plymuthica was significantly negatively affected (P = 0.037) after 5 days of incubation by reaching 1.47 Â 10 9 CFU/ mL compared to the monocultures ( Fig. 2A).
The bacterial CFU obtained from H. gracilis grown in the presence of cell free supernatants (CFS) of Paenibacillus sp. AD87 and of S. plymuthica are summarized in Fig. 2B. H. gracilis growth was significantly increased (P = 0.011) when growing in the presence of CFS of Paenibacillus sp. AD87, resulting in higher H. gracilis cell counts compared to the monoculture by reaching 1.10 Â 10 6 CFU/mL. In the presence of CFS from S. plymuthica PRI-2C, H. gracilis reached the highest cell counts at 1.72 Â 10 6 CFU/mL (P = 0.000) after 5 days of incubation (Fig. 2B).
Interspecific interaction between bacterial species allows use of additional substrates. During physiological or catabolic profiling, the metabolism of 31 carbon sources during bacterial growth are measured in 96-wells plates. The catabolic profiling assays revealed that Paenibacillus sp. AD87was able to utilize 11 out of the 31 carbon sources in monoculture, while S. plymuthica PRI-2C and H. gracilis were able to utilize 17 and 16 carbon sources, respectively. Interestingly, three compounds could be utilized only during the cocultivation of H. gracilis with one of the other species, but not by any of the monocultures. Specifically, a-cyclodextrin was utilized only during the cocultivation of H. gracilis with Paenibacillus sp. AD87, while L-threonine and glycyl-Lglutamic acid were utilized only during the cocultivation of S. plymuthica PRI-2C and H. gracilis (Fig. 3).
Genomic features of H. gracilis, S. plymuthica PRI-2C, and Paenibacillus sp. AD87. Sequencing of the complete genome of H. gracilis resulted in a genome size of 3.82 Mbp with 3,648 coding sequences (CDS). As expected, the genome analysis revealed that the genome of H. gracilis is smaller and contains fewer genes compared to S. plymuthica PRI-2C (5.4 Mbp) and Paenibacillus sp. AD87 (7.0 Mbp). The genome features of all three bacteria are summarized in Table 1.
In silico analysis of gene clusters encoding secondary metabolites. The in silico tool antiSMASH allows the rapid genome-wide identification, annotation, and analysis of secondary metabolite biosynthesis gene clusters in bacterial and fungal genomes (37). In silico analysis of H. gracilis revealed that H. gracilis possesses relatively few gene clusters related to secondary metabolism. A total of three gene clusters for H. gracilis were detected, of which one belonged to the class of bacteriocins, one to the class of Interspecific Bacterial Interactions mSystems terpenes, and one to aryl polyenes, the latter being a homolog to the aryl polyene gene cluster from Xenorhabdus doucetiae (GenBank accession no. NZ_FO704550.1) (Fig. 4A). For S. plymuthica PRI-2C, nine gene clusters were found, of which two gene clusters were annotated to encode the production of nonribosomal peptides (NRPs), one of homoserine lactones, one of aryl polyenes and/or non-NRP siderophores, one of hybrid polyketide-NRP metabolites, one of thiopeptides, one of butyrolactones, one of terpenes, and one of others (Fig. 4B).
For Paenibacillus sp. AD87 the in silico analysis revealed a total of 10 gene clusters coding for secondary metabolites. From which two gene clusters encode pathways for producing terpenes, one for bacteriocins, one for lasso peptides, two for lanthipeptides, one for NRPs, one for others, one for polyketides (type III enzyme mechanism) and one gene cluster for non-NRP Siderophores (Fig. 4C). Pathway analysis in H. gracilis compared to S. plymuthica PRI-2C and Paenibacillus sp. AD87. For annotation and pathway analysis Rapid Annotation using Subsystem Technology (RAST) and OrthoFinder were used. The RAST comparison of Paenibacillus sp. AD87 and H. gracilis revealed 504 unique enzymes (according to their Enzyme Commission (EC) numbers.) exclusive for Paenibacillus sp. AD87, while 434 were present only in H. gracilis and 532 EC numbers were shared by both genomes (Fig. 5A). The RAST comparison of S. plymuthica PRI-2C and H. gracilis revealed that 751 enzymes were present only in S. plymuthica PRI-2C, and 260 were present only in H. gracilis. A total of 727 EC numbers participating in diverse metabolic pathways were found in both genomes (Fig. 5B).  The missing genes and pathways found by OrthoFinder and EggNOG were annotated with Gene Ontology (GO) terms. The analysis revealed that five genes related to metabolic pathways were absent in H. gracilis. Those missing genes were annotated with the following molecular function ontology terms: GO:0008473 (ornithine cyclodeaminase activity), GO:0008696 (4-amino-4-deoxychorismatelyase activity), GO:0003920 (GMP reductase activity), GO:0004035 (alkaline phosphatase activity), and GO:0008442 (3-hydroxyisobutyrate dehydrogenase). We verified if the absence of these molecular functions would render specific pathways obsolete or unavailable in H. gracilis. However, alternative pathway routes are present for these genes encoding certain molecular functions according to KEGG database annotations. The pathway analysis by RAST did not reveal the absence of essential genes in H. gracilis. Still, the comparison of the number (n) of genes present in each bacteria revealed major differences in several pathways, specifically in the categories  Effect of interspecific interactions on gene expression. The transcriptome analysis of monocultures and cocultures revealed a total of 277 significant differentially expressed genes; where 100 genes were downregulated and 177 genes were upregulated between the different treatments ( Table 2).
(i) Effect of interspecific interactions on gene expression in Paenibacillus sp. AD87 and H. gracilis. Genes related signal transduction (T) was the category with the most differentially expressed genes during the cocultivation of H. gracilis with Paenibacillus sp. AD87 compared to the monoculture of H. gracilis ( Fig. 6A and B, Table  S6 and S7).
In Paenibacillus sp. AD87 histidine biosynthesis and dephosphorylation, genes were upregulated, while cellular-growth-related genes were downregulated (see Table S1 in the supplemental material) at day 10 of the interaction with H. gracilis (Fig. 6B). For the interaction of H. gracilis with Paenibacillus sp. AD87 15, significant differentially expressed genes were found (0 at day five and 15 at day 10). At day five, genes related to sulfur assimilation, chemotaxis, and response to (chemical/external) stimuli were upregulated in H. gracilis in the presence of Paenibacillus sp. AD87.
(ii) Effect of interspecific interactions on gene expression S. plymuthica PRI-2C and H. gracilis. During the interaction of S. plymuthica PRI-2C with H. gracilis, 61 genes were significantly differentially expressed at day five and 10 at day 10. At day five, iron-sulfur cluster-assembly-related genes, a sulfur transferase, and a transaminase were upregulated, while genes related to inorganic diphosphatase activity, exonuclease activity, and DNA repair were downregulated. At day 10, genes related to sulfur transmembrane transport, sulfur compound catabolism, and cysteine biosynthesis were upregulated, and genes related to sulfur compound metabolism and translation were downregulated. (Table S2 and S3). For S. plymuthica PRI-2C, genes related to signal transduction and translation, ribosome structure, and biogenesis were the most differentially expressed gene categories (Fig. 6C). For H. gracilis in interaction with S. plymuthica PRI-2C, 182 differentially expressed genes were identified at day 10 and only one at day five. At day five, genes related to the ribosome/ribonucleoproteins, organelle organization/assembly, and iron-sulfur cluster assembly were upregulated, and genes related to the innate immune response (toll-like receptor signaling) were downregulated (Table S4 and S5). At day 10, genes related to signal transduction and chemotaxis were upregulated in H. gracilis. For H. gracilis, the most upregulated genes were linked to chemotaxis pathway and iron scavenging, suggesting activity in competition (Fig. 6A). Metabolomic analysis of volatile compounds. The volatile blend composition of the monocultures differed from that of the cocultures. Clear separations between the controls, monocultures and cocultures were obtained in Partial least squares discriminant analysis (PLS-DA) score plots (Fig. 7A). The analysis revealed a total of 25 volatile organic compounds produced by mono-and cocultured bacteria that were not detected in the noninoculated controls (Table 3). Of these, 17 were identified and categorized in six chemical classes (alkenes, benzenoids, sulfides, thiocyanates, terpenes, furans). The remaining eight compounds could not be assigned with certainty to a known compound. The most abundant volatile organic compounds were sulfur-containing compounds such as dimethyl disulfide (C 2 H 6 S 2 ) and dimethyl trisulfide (C 2 H 6 S 3 ). These two sulfur compounds were produced by all three bacteria. Interestingly an unknown compound with a retention time (RT) of 26.4 min produced by the monocultures of H. gracilis was not detected in the interactions with S. plymuthica PRI-2C (Table 3). Two other unknown compounds with an RT of 4.15 min and 24.34 min produced by the monocultures of Paenibacillus sp. AD87 were not detected in the cocultivation with H. gracilis (Table 3).
Direct analysis in real-time mass spectrometry (DART-MS) based metabolomics. Metabolomics analysis based on direct analysis in real-time mass spectrometry (DART-MS) revealed separations between the controls, monocultures, and cocultures as presented in PLS-DA score plots (Fig. 7B). The metabolomic composition of the monocultures differed from that of the cocultures (Fig. 7B). Statistical analysis (one-way ANOVA and post hoc Tukey's HSD test) revealed 617 significant mass features present on day five and day 10 of which 48 could be tentatively assigned to specific compounds. Most of the significant peaks were found in the cocultures of H. gracilis with Paenibacillus sp. AD87. The significant mass features and the corresponding tentative metabolites can be found in Table 4.
Mass spectrometry imaging metabolomics. Laser ablation electrospray ionizationmass spectrometry imaging (LAESI-MSI) was performed to visualize the localization of In the coculture of H. gracilis with Paenibacillus sp. AD87, genes related to signal transduction (T) were the category with the most differentially expressed genes. In the coculture of H. gracilis with S. plymuthica PRI-2C, genes related to signal transduction (T), translation, ribosome structure and biogenesis (J) were the most prevalent differentially expressed gene categories. COG C, energy production and conversion; D, cell cycle control, cell division, and chromosome partitioning; E, amino acid transport and metabolism; F, nucleotide transport and metabolism; G, carbohydrate transport and metabolism; H, coenzyme transport and metabolism; I, lipid transport and metabolism; J, translation, ribosomal structure, and biogenesis; K, transcription; L, replication, recombination, and repair; M, cell wall/membrane/envelope biogenesis; N, cell motility; NA, not assigned; O, posttranslational modification, protein turnover, and chaperones; P, inorganic ion transport and metabolism; Q, secondary metabolites biosynthesis, transport, and catabolism; R, general function prediction only; S, function unknown; T, signal transduction mechanisms; U, intracellular trafficking, secretion, and vesicular transport; V, defense mechanisms.
Interspecific Bacterial Interactions mSystems metabolites in their native environments in monoculture as well as during interaction without performing any extraction. Across all treatments, clear separation was observed among the samples for controls, monocultures, and interactions (Fig. 8A).  9B). This is also evident in the box-and-whisker plots and the ion intensity maps that are presented for four statistically significant metabolites belonging to this set (Fig. S2B).
For the pairwise analysis for the H. gracilis monoculture (H) and the coculture of S. plymuthica PRI-2C and H. gracilis (SH), 57 mass features (in green) displayed significantly high concentration in H and 42 mass features had significantly low concentration in HM compared to SH (Fig. 9C). The box-and-whisker plots along with the ion Interspecific Bacterial Interactions mSystems intensity maps for four statistically significant metabolites belonging to this set are shown in Fig. S2C. For the pairwise analysis for S. plymuthica PRI-2C monoculture (S) and the interaction of S. plymuthica PRI-2C and H. gracilis (SH), 135 mass features (in green) displayed significantly high concentration in S and 65 mass features had significantly low concentration in S compared to SH (Fig. 9D). The box-and-whisker plots along with the ion intensity maps for four statistically significant metabolites belonging to this set are shown in Fig. S2D.
To visualize the number of shared and unique metabolites among the monoculture and interaction samples, Venn diagrams were plotted. The Venn diagram (Fig. 9E) for monocultures H. gracilis and Paenibacillus sp. AD87 and their interaction shows 80 metabolites unique to H. gracilis monoculture, 75 metabolites unique to Paenibacillus sp. AD87 monoculture, and 100 metabolites that are unique during their interaction. A total of 1,062 metabolites were shared within these three treatments. Similarly, the Venn diagram (Fig. 9F) for monocultures H. gracilis and S. plymuthica PRI-2C and their interaction shows 196 metabolites unique to H. gracilis monoculture, 48 metabolites unique to S. plymuthica PRI-2C monoculture, and 120 metabolites that are unique during their interaction.

DISCUSSION
Here, we report the first-time isolation of H. gracilis from a terrestrial soil sample. This bacterium passed a 0.1-mm filter, which suggests a very small cell size, theoretically justifying referring to these bacteria as ultrasmall bacteria (26). However, against our expectation, the microscopic analysis revealed that this bacterium is not ultrasmall in cell size but possesses a very thin diameter and showed the typical spiraled morphology known for these species (38)(39)(40)(41). These observations are in line with previous research by Wang et al. (42)  showing that H. gracilis can pass through filters of various pore sizes ranging from 0.45 mm to 0.1 mm, most probably thanks to their cell shape and cell morphology. In silico analysis of 16 terrestrial metagenome data available on MG-RAST (https://www.mg-rast.org/) showed that H. gracilis was not present in terrestrial metagenome data (not shown), suggesting that H. gracilis is not commonly present in terrestrial soils. The bacterial interaction assays revealed that H. gracilis grows better when interacting with Paenibacillus sp. AD87 or S. plymuthica PRI-2C. The cell numbers of H. gracilis were higher when exposed to cellfree supernatants of Paenibacillus sp. AD87 and S. plymuthica PRI-2C, suggesting that the metabolites released by the latter bacteria in cocultures with H. gracilis are associated with improved growth of H. gracilis. We hypothesized that H. gracilis grows better in coculture, either because growth is stimulated by signals produced by the other organism, or   because the environment that is created by the other organism allows H. gracilis to make more efficient use of certain metabolic pathways. Indeed, the metabolic experiments with Biolog plates showed that during interspecific interactions of H. gracilis with Paenibacillus sp. AD87 or with S. plymuthica PRI-2C, more carbohydrates could be utilized compared to the monocultures. This is an interesting observation, and it may indicate that interaction of bacteria can trigger the production of exoenzymes, enabling the degradation of carbohydrates, which the bacteria were not able to degrade in monoculture. We speculated that since H. gracilis grows better in interaction with other bacteria and is of relatively small cell size, H. gracilis might have evolved according to a genome streamlining strategy, i.e., the adaptive loss of genes for which functions it relies on interaction with other bacteria in the immediate environment. The whole-genome sequencing of H. gracilis revealed a genome size of 3.82 Mbp. This is a relatively small genome size for free-living soil bacteria that typically have estimated average genome sizes of ;4.7 Mbp (36,(43)(44)(45)(46). The in silico antiSMASH (37) comparison of genes that are part of secondary metabolite gene clusters showed that the H. gracilis genome contained only three gene clusters encoding the production of secondary metabolites (bacteriocins, terpenes, and aryl polyenes). Terpenes and aryl polyenes are known as protective compounds against abiotic stressors, while bacteriocins have antimicrobial activities against closely related bacteria (17,(47)(48)(49)(50)(51). We hypothesize that H. gracilis genome streamlining has allowed it to be more competitive, by retaining only the most essential metabolic functions while having roughly about one quarter less DNA to replicate during each cell division. Gene loss and reduced genome size may cause dependency on other microbes in their surroundings, and this may explain a considerable part of the phenomenon that most of the detectable bacteria in the environment are not cultivable under laboratory conditions.
To investigate the mechanisms of interaction, we performed transcriptome analysis on the interaction pairs of H. gracilis with S. plymuthica PRI-2C and Paenibacillus sp. AD87. Interestingly, a larger amount of significantly differentially expressed genes was induced by H. gracilis in the other two competing bacteria compared to the transcriptomic changes in H. gracilis. Several processes, enriched according to GO term enrichment analysis, could be part of a mechanism(s) mediating interactions between H. gracilis and S. plymuthica PRI-2C and Paenibacillus sp. AD87, for example genes related to chemotaxis. Moreover, the GO terms for signal transduction, secondary metabolite production, and cell motility were enriched in the transcriptome of H. gracilis during the cocultivation with Paenibacillus sp. AD87, suggesting that chemotaxis and cell movement is an important feature during interspecific interactions between these two bacterial taxa (52,53). In addition, GO terms referring to iron-sulfur (Fe-S) complex assembly were enriched in the transcriptomes of H. gracilis during the cocultivation with S. plymuthica PRI-2C and Paenibacillus sp. AD87. Fe-S clusters are important for sustaining fundamental life processes: they participate in electron transfer, substrate binding/activation, iron or sulfur storage, regulation of gene expression, and enzyme activity (54,55). This upregulation could indicate that, potentially, in coculture, normal-sized bacteria released metabolites that H. gracilis used for synthesizing Fe-S complexes. It is also possible that iron-sulfur complex assembly is activated during competition with the interacting bacteria for sulfur or iron collection (scavenging) (56)(57)(58)(59).
The metabolic pathway analysis showed that the loss of genes in H. gracilis does not appear to have resulted in functional loss of metabolic pathways. Loss of nonessential and possibly redundant genes in several metabolic pathways could explain why and how the genome of H. gracilis has become so small. The missing genes are not essential to complete metabolic pathways and only appear to result in limited options in certain metabolic pathways. RAST analysis showed that all basal metabolic pathways   (41,60,61) and in the citric acid cycle (41,62). Based on the available data, it cannot be unambiguously determined which alternative pathway may preferably be used by H. gracilis to synthesize fumarate. The metabolomics analysis revealed the production of specific antimicrobial compounds such as pyrollnitrin (S. plymuthica PRI-2C) and 2,5-bis(1-methylethyl)-pyrazine (Paenibacillus sp. AD87) which are well known for their broad-spectrum antimicrobial activity (63)(64)(65)(66)(67). However, the produced antimicrobial compounds didn't show activity against H. gracilis: in both interactions, H. gracilis showed increased growth when growing in coculture with either Paenibacillus sp. AD87 or S. plymuthica PRI-2C.
The understanding of natural metabolites that mediate interactions between organisms in natural environments is the key to elucidate ecosystem functioning. The detection and identification of the compounds that mediate such interactions is still challenging. Techniques such as mass spectrometry imaging (MSI) provide new opportunities to study environmentally relevant metabolites in their spatial context (68)(69)(70). In this study, the metabolomics was performed using three independent approaches namely, DART-MS analysis, GC/MS-Q-TOF analysis and LAESI-MSI from living bacterial colonies. LAESI-MSI analysis revealed that several mass features were detected in higher abundance during the cocultivation of H. gracilis with Paenibacillus sp. AD87, these mass features were m/z 425.2886 and m/z 558.2832. LAESI-MSI is not suitable for unambiguous compound annotation, but can still be used for putative compound annotation. To annotate the detected mass features to compounds with high certainty, LAESI-MSI should be coupled with ion mobility separation as previously suggested (71)(72)(73). Yet, LAESI-MSI can help to spatially distinguish the produced secondary metabolites of living bacterial colonies with limited sample preparation and can give insight into the spatial distribution of metabolites.
Several studies indicate that the volatile blend composition of the volatiles greatly depends on biotic interactions and on growth conditions (15,19,(74)(75)(76). Here, a higher number of volatile compounds were detected in the bacterial cocultures, most likely due to the combination of emitted volatiles of the interacting bacteria. The high number of sulfur-containing compounds indicates that these compounds are commonly produced by bacteria and might play an important role in signaling during interspecific interactions (77,78). No novel volatile compounds were detected during the coculture of the three bacteria.
Overall, our study showed that H. gracilis can pass through a 0.1-mm filter and is present in terrestrial environments. The growth performance and physiological behavior of H. gracilis were dependent on the cocultivated bacterial partner and they might be metabolically depending on the cocultivated bacteria. At the same time, H. gracilis was able to change the physiology, release of volatile organic compounds, and secreted enzymes of the cocultivated bacteria without direct cell-cell contact.
Microbial interspecific interactions play an important role in the functioning of the terrestrial ecosystem. Soil microbial communities are very diverse and dynamic and involve frequent and sporadic interspecific interactions. Our study indicates that sulfur and Fe-S clusters could play an important role in microbial interspecific interactions in terrestrial environments and more studies are required to understand their role. The study of sporadic interspecific interactions and the inclusion of rare taxa in future analysis could help to better understand microbial communities and functions of those.

MATERIALS AND METHODS
Isolation and identification of bacteria that pass through 0.1-lm filters. (i) Isolation of H. gracilis from soil. After removing the grassland vegetation, a topsoil core was collected and mixed; a sample of 10 g was suspended in 90 mL of 10 mM phosphate-buffer (pH 6.5) and shaken with gravel (2 to 4 mm) at 250 rpm for 45 min. The extract was filtered through sterile gauze pads and subsequently through sterile Whatmann 1-mm paper filter using Buchner funnel. Purified extract was filtered again through a syringe filter (0.2 mm) and afterwards through a 0.1-mm filter (GE-Healthcare). The filtered extract was plated on 1/10th tryptic soy broth agar (TSBA) plates immediately after isolation and Interspecific Bacterial Interactions mSystems incubated at 25°C (Fig. S1). The plates were inspected daily using a stereomicroscope (Leica M205C) screening for bacteria colonies.
(iii) Identification of H. gracilis. For the identification of H. gracilis, 16S rRNA PCR was performed from grown colonies in a 50-mL PCR-GoTaq green master mix (Promega Corp. Madison, USA; cat. no. M712). For 16S rRNA gene amplification, the following primers were used: forward primer 27f (59-AGA GTTT GAT CMT GGC TCAG-39), reverse primer 1492r (59-GRT ACC TTG TTA CGA CTT-39), amplifying ;1,465 bp from the 16S rRNA gene (83, 84) (modified). All PCRs were performed on a Bio-Rad C1000 Touch PCR machine (BIO-RAD, Veenendaal, the Netherlands) with these settings: initial cycle of 95°C for 3 min; 30 cycles of 94°C for 30 sec, 55°C for 45 sec, and 72°C for 1 min; and final extension at 72°C for 5 min. The PCR products were purified using the Qiagen PCR purification kit and sent to MACROGEN (MACROGEN Europe, Amsterdam, the Netherlands) for 16S rRNA sequencing.
(iv) Microscopy. Microscopy pictures of H. gracilis cells were taken at 400-fold magnification with an Axio Imager M1 microscope (Carl Zeiss, Germany) under phase-contrast illumination with an AxioCam MRm camera. Macroscopic colony pictures of H. gracilis were taken with an Olympus Binocular at Â20 magnification. Images were analyzed with AxioVision v4.7 (Carl Zeiss Imaging Solutions GmbH, Germany).
(v) Bacterial interactions assays. After four days of preculture, a single colony of Paenibacillus sp. AD87 and S. plymuthica PRI-2C and H. gracilis was picked and inoculated in 20 mL 1/10th TSB (Paenibacillus sp. AD87 and S. plymuthica PRI-2C) and grown overnight at 24°C, 220 rpm. For the inoculation of H. gracilis, a single colony was picked from a TSBA plate and inoculated in 20 mL LB-medium and grown for 3 days at 24°C, 200 rpm. For the interaction assay, an inoculation mix of each bacterial strain (Paenibacillus sp. AD87, S. plymuthica PRI-2C, and H. gracilis) was prepared by diluting the bacterial isolates in 20 mL of 10 mM phosphate-buffer (pH 6.5) to an optical density at 600 nm (OD 600 ) of 0.005 (Paenibacillus sp. AD87 and S. plymuthica PRI-2C) or to an OD 600 of 0.05 (H. gracilis), which corresponds to 10 5 CFU/mL. A droplet of 10 mL inoculation mixture was added in the middle of a 6-cm diameter petri dish (monocultures) or next to each other in a distance of ;0.5 cm (pairwise interactions). All treatments were performed in triplicates on 1/10th TSBA plates incubated at 24°C. After the growth time of 3 days, the bacteria were scratched and washed from the plates by using sterile cell scratchers and Phosphate buffer. For the enumeration of the cell counts (CFU/mL), dilution series of the scratched bacteria were prepared and plated in triplicates on 1/10th TSBA plates and grown for 48 h. Enumeration was carried out on an aCOlyte colony counter (Don Whitley Scientific, Meintrup DWS Laborgeräte GmbH, Germany) based on their different morphology.
Enumeration of growth-inhibitory or growth-promoting effects of cell-free supernatants of Paenibacillus sp. AD87and S. plymuthica PRI-2C on the growth of H. gracilis. A bacterial growth assay in liquid media supplemented with cell-free supernatants (CFS) of Paenibacillus sp. AD87 and S. plymuthica PRI-2C was conducted. For the assay, single colonies of Paenibacillus sp. AD87, S. plymuthica PRI-2C, and H. gracilis were inoculated in 20 mL 1/10th TSB (Paenibacillus sp. AD87 and S. plymuthica PRI-2C) or in 20 mL LB medium (H. gracilis) and grown overnight (Paenibacillus sp. AD87 and S. plymuthica PRI-2C) or for 3 days (H. gracilis) at 24°C, 220 rpm. For the preparation of Paenibacillus sp. AD87 and S. plymuthica PRI-2C CFS, the grown cultures were centrifuged at 5,000 rpm for 20 min (at room temperature) and filtered through 0.2-mm filters (GE Healthcare). For the assay, H. gracilis was inoculated into 20 mL liquid LB media at an OD 600 of 0.05. The growth medium was then supplemented either with 20% (vol/vol) CFS of Paenibacillus sp. AD87 or S. plymuthica PRI-2C or with 20% (vol/vol) of filter-sterilized liquid 1/10th TSB media (control). The cultures were incubated at 24°C at 220 rpm for 7 days and the bacterial growth was monitored by optical density (absorbance at 600 nm) measurements and by plate counting. After 5 days of growth, the CFU/mL of H. gracilis grown in the presence of CFS of Paenibacillus sp. AD87 or S. plymuthica PRI-2C were enumerated by plate counting. For this, the cultures were sampled, and dilution series were prepared in triplicates and a volume of 100 mL of each serial dilution was plated in three replicates with a disposable Drigalski spatula on 1/10th TSBA plates. CFU Enumeration was carried out on an aCOlyte colony counter.
DNA isolation and genome sequencing of H. gracilis. Genomic DNA of H. gracilis was extracted using a Qiagen genomic-tip 500/G DNA kit (Qiagen, cat. no. 10262). Genome sequencing was performed on the PacBio RS II platform (Pacific Biosciences, Menlo Park, CA, USA) using P6-C4 chemistry at the Institute for Genome Sciences (IGS), Baltimore, MD, USA. The sequencing resulted in a total of 70,101 reads with N 50 of 17,309 nucleotides. The PacBio raw sequences were analyzed using SMRT portal v2.3.0.140936 p.4150482. Sequences were assembled de novo with the RS_HGAP_assembly 3 software (Pacific Biosciences) with default settings on an estimated genome size of 3.8 Mbp. The resulting assemblies were subjected to scaffolding using the RS_AHA_scaffolding 1 software. The genome assembly properties are shown in Table 1. Final contigs were annotated using PROKKA v1.11 (85) and InterproScan v5. 16 55.0 (86). The whole-genome sequence was submitted as Hylemonella gracilis strain NS1 to NCBI GenBank (https://www.ncbi.nlm.nih.gov/genbank/) under accession no. CP031395.
RNA isolation and sequencing. Sampling for total RNA extractions was performed in triplicates after five and 10 days of incubation on bacteria grown on 1/10th TSBA plates either in coculture or monoculture as described previously (bacterial interactions assays on 1/10th TSBA plates). For the isolation of bacterial cell material, a volume of 1 mL of 10-mM phosphate buffer (pH 6.5) was added to the surface of the 1/10th TSBA plates and grown bacterial cells were suspended from the plate surface with a disposable cell scratcher (VWR International B.V., the Netherlands). For total RNA extraction the obtained cell suspension was transferred to a tube containing RNAprotect bacteria reagent (Qiagen, cat. no. 76506) and centrifuged for 20 min at 20,000 g, 4°C. The supernatant was discarded, and the resulting cell pellets were stored at 280°C. Total RNA was extracted using the Aurum Total RNA minikit (Bio-Rad) according to the manufacturer's protocol. Samples were treated with the Ambion TURBO DNA free kit according to the manufacturer's protocol. The RNA concentration and quality was checked on a NanoDrop Spectrophotometer (ND 2000, Thermo Fisher Scientific, the Netherlands) and on a 1.0% Trisborate-EDTA (TBE) agarose gel. Samples were subjected to RNA sequencing at the Erasmus Center for Biomics (www.biomics.nl), Erasmus MC, Rotterdam, the Netherlands using the Illumina HiSeq 2500 sequencing platform. The obtained reads were checked for quality using Fastq. For the estimation of the transcripts, the filtered sequences were aligned against the cDNA sequences of H. gracilis, Paenibacillus sp. AD87 and S. plymuthica PRI-2C using Bowtie 2 (v2.2.5) (87) with the following settings:no-mixedno-discordantgbar 1000end-to-end. Transcript abundance was calculated using RSEM v1.1.26 (88) and differential expression between the treatments was calculated using the edgeR v3.2 package in the R environment (89)(90)(91).
Pathway annotations. Please see Supplemental Methods (Text S1). Exploration of missing genes and genome streamlining in Hylemonella. RAST annotations of S. plymuthica PRI-2C, Paenibacillus sp. AD87, and H. gracilis were used to compare their genomes and to explore the genomes for missing genes in metabolic pathways (http://rast.nmpdr.org) (92)(93)(94). The missing gene sequences were extracted and assigned with KEGG orthology (95,96). Presence/absence of genes belonging to metabolic pathways was compared across the three genomes to identify shared genes and pathways and to determine incomplete metabolic pathways in H. gracilis.
Catabolic profiling. To determine the carbon source usage abilities of H. gracilis and Paenibacillus sp. AD87 and S. plymuthica PRI-2C strains, Biolog EcoPlate (Labconsult S.A.-N.V., Brussels, Belgium) assays were performed (97,98). Bacteria were cultured in monoculture or in coculture in single wells of the Biolog EcoPlate. One single colony of each bacterial strain was picked and inoculated in 15 mL 1/10th TSB or 15 mL LB liquid media. Bacteria were grown overnight (Paenibacillus sp. AD87 and S. plymuthica PRI-2C) or for 2 days (H. gracilis) at 24°C at 250 rpm. Grown bacterial cultures were washed twice by centrifugation at 4.500 rpm for 15 min at room temperature, the supernatant was discarded, and the pellet was washed and resuspended in 5 mL of 10 mM phosphate buffer. The reuspended cultures were diluted to an OD 600 of 0.005 in 20 mL of 10-mM phosphate buffer either in monoculture or in coculture. For the experiment, Biolog EcoPlates were inoculated with 100 mL of each bacterial inoculation suspension (monocultures or cocultures) in each well. For each bacterial monoculture one Biolog EcoPlate was inoculated, as well for each cocultivation pair (S. plymuthica PRI-2C with H. gracilis and Paenibacillus sp. AD87 with H. gracilis). Plates were incubated for 1 week and absorbance was measured at 590 nm every 24 h on a BIOTEK plate reader to determine the ability of the bacterial cultures to use the carbon sources present in the wells.
Trapping of volatile organic compounds and GC-Q-TOF analysis. Please see Supplemental Methods (Text S1).
DART-MS data analysis. Please see Supplemental Methods (Text S1). Ambient mass-spectrometry imaging LAESI-MS data analysis. For LAESI-MS analysis, a single colony of each bacterial isolate was picked and inoculated in 20 mL 1/10th TSB (Paenibacillus sp. AD87 and S. plymuthica PRI-2C) and grown overnight at 24°C at 220 rpm. For the inoculation of H. gracilis, a single colony was picked from plate and inoculated in 20 mL LB medium and grown for 3 days at 24°C at 200 rpm. The inoculation mixture was prepared by diluting the bacterial isolates in 20 mL of 10 mM phosphate buffer (pH 6.5) to an OD 600 of 0.005. The inoculation mixture was pulse-vortexed for 30 sec and a droplet of 10 mL was added in the middle of a 6-cm diameter petri dish (monocultures) or next to each other in a distance of approximately 0.5 cm (pairwise interactions). All treatments were inoculated in triplicates on 1/10th TSBA and incubated at 24°C for five and 10 days. After five and 10 days of incubation, bacterial colonies were cut out of the agar (size approximately 1 to 3 cm 2 ) and subjected to LAESI-MS measurement. The LAESI-MS analysis was carried out on a Protea Biosciences DP-1000 LAESI system (Protea Bioscience Inc., Morgantown, WV, USA) coupled to a Waters model Synapt G2S (Waters Corporation, Milford, MA, USA) mass spectrometer. The LAESI system was equipped with a 2940-nm midinfrared laser yielding a spot size of 100 mm. The laser was set to fire 10 times per x-y location (spot) at a frequency of 10 Hz and 100% output energy. A syringe pump was delivering the solvent mixture of methanol/water/formic-acid (50:50:0.1% vol/vol) at 2 mL/min to a PicoTip (5 cm Â 100 mm diameter) stainless steel nanospray emitter operating in positive ion mode at 4,000 V. The LAESI was operated using LAESI desktop software v2.0.1.3 (Protea Biosciences Inc., Morgantown, WV, USA). The time of flight (TOF) mass analyzer of the Synapt G2S was operated in the V-reflectron mode at a mass resolution of 18.000 to 20.000. The source temperature was 150°C, and the sampling cone voltage was 30 V. The positive ions were acquired in a mass range of 50 to 1,200 m/z. The MS data were lock mass corrected post data acquisition using leucine encephalin (C 25 H 37 N 5 O 7 m/z = 556.2771), which was used as an internal standard. All the acquired Waters *.RAW data files were converted to open file format *.imzML using an in-house script written in R. Later, these data were preprocessed in multiple steps to remove noise and to make the data comparable. First, square root transformation was applied to the data to stabilize the variance. Then, baseline correction was performed to enhance the contrast of peaks to the baseline. For better comparison of intensity values and to remove small batch effects, Total Ion Current (TIC)-based normalization was applied. This was followed by spectral alignment and peak detection to extract a list of significant mass features for each sample replicate per treatment. In the end, a mass feature matrix was generated with sample replicates for each treatment in columns and mass features in rows. This feature matrix was used to perform further statistical analysis. The preprocessing and peak-detection steps were applied using R scripts developed in-house and using functions available within the MALDIquant R package (99). To perform multivariate analysis, the feature matrix was imported into the online version of Metaboanalyst 4.0 (100). Ion intensity maps displaying the spatial distribution for statistically significant mass features were created using R. Before generating the ion maps, the intensity values for the selected mass features were normalized to the maximum intensity within the image, measured for each mass value individually. Venn diagrams displaying unique and common masses among different treatments were drawn using the jvenn tool (101).
Data availability. The raw data of this article will be made available by the authors to any qualified researcher upon request. The whole-genome sequence of Hylemonella gracilis strain NS1 is available at the NCBI GenBank (https://www.ncbi.nlm.nih.gov/genbank/) under accession no. CP031395, and the raw reads of the transcriptomics data are available at the Sequence Read Archive (SRA) (https://www.ncbi.nlm.nih.gov/sra) under accession no. PRJNA483535. The metabolomics raw data (DART-MS and LAESI-MSI) are available at metabolights (www.ebi.ac.uk/metabolights/MTBLS5841).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.02 MB.