Gut microbial interactions based on network construction and bacterial pairwise cultivation

doi:10.21203/rs.3.rs-3328085/v1

Download PDF

Research Article

Gut microbial interactions based on network construction and bacterial pairwise cultivation

https://doi.org/10.21203/rs.3.rs-3328085/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Association networks are widely applied to predict bacterial interactions in the human gut microbiome studies. However, experimental validation of the predicted interactions is challenging due to the complexity of microbial composition and the limited number of cultivated bacteria.

Results: In this study, we addressed this challenge by integrating in vitrotime series network association inference and co-culture of taxon pairs in network. Fecal samples were cultivated on YCFA agar plates for 13 days. Cells from agar were temporally harvested for DNA extraction and metagenomic sequencing. A total of 198 metagenome-assembled genomes (MAGs) were recovered and 360 bacterial isolates were cultivated belonging to 59 species. Temporal dynamics of bacteria growing on the YCFA agar were used to infer microbial association networks. To experimentally validate the interactions of taxon pairs in networks, we selected 43 bacterial strains that represented 43 MAGs. Among these, 19 strains were collected from the human Gut Microbial Biobank (hGMB) and 24 were isolated from this study. The co-culture experiments revealed that the majority of the interactions between taxa in networks were identified as neutralism (51.67%), followed by commensalism (21.67%), amensalism (18.33%), competition (5%) and exploitation (3.33%). Genome-centric analysis further revealed that the commensal members in human gut extensively involved the exchange of amino acids with greatest biosynthesis cost, short-chain fatty acids, and vitamins. We also validated the 12 beneficiaries by adding 16 additives into the basic YCFA medium. As a result, we found that the growth of 66.7% of the strains was significantly promoted.

Conclusions:

We have established a method that infers microbial interactions from association networks and validates these interactions using representative strains. This approach provides new insights into reducing the complexity of gut community and confirming microbial interactions in association networks through co-culture experimental. Our work highlights that the inferred gut microbial association networks tend to overestimate positive relationships in the real-world gut microbial communities. Moreover, the confirmed positive relationships between gut microbes are likely mediated by the exchange of amino acids, fatty acids, and vitamins.

Human gut microbiome

Association network

Microbial interaction

Co-culture experiments

Time series analysis

Community dynamics

The human gut harbors trillions of bacterial cells, comprising intricate association and interaction networks [1, 2]. Previous studies have primarily relied on in silico co-occurrence or correlation analyses to discuss microbial relationships in gut microbiome, resulting in a highly descriptive of the field [3–10]. It is still a challenge to get a comprehensive insight into bacterial interactions predicted from intertwined association network. With the growing available of cultivated gut bacterial resources, it is possible to address the question of whether and how gut microbes associated in networks are truly or not interacted with each other.

Inference of microbial association networks is the most fundamental and critical step to map microbial interaction [3, 11]. Based on the strategies of sampling campaign, the microbial association networks could be constructed from longitudinal cohort and cross-sectional cohort studies. Cross-sectional studies provide a “snapshot” of microbial community composition at a specific time point [12]. Considering that samples collected from different individuals at single time point are easier accessible than that from individuals over time, the sample sizes of cross-sectional studies are typically larger than longitudinal studies, thus cross-sectional studies provide numerous but not related information [12, 13]. Cross-sectional studies also have limitations, particularly in being unable to study the dynamic changes of the samples [14]. In contrast to cross-sectional cohort studies, longitudinal cohort studies enable the investigation of temporal dynamics in microbial communities and inference of local or potentially time-delayed associations [15]. However, the compositions of microbial communities were highly dynamic in human gut, resulting in the changed microbial interactions among community members [16]. The dynamic changes of the gut communities were greatly influenced not only by interactions among bacteria but also by the host [17]. In our study, the primary focus was on investigating the interactions among bacteria within the gut. Therefore, we considered that employing in-vitro time-series cultivation and sequencing to predict the association network was highly necessary.

In terms of sequencing methods, compared to 16S rRNA gene amplicon sequencing, metagenomic sequencing is a more suitable approach. In our previous work, we found that microbes from different species share identical 16S rRNA gene V4 or V3V4 region [18]. It could lead to an overestimation of node abundance, thereby increasing its significance within the network. Extended Local Similarity Analysis (eLSA) was a common method to infer the microbial association [19]. In terms of inferring association networks, the eLSA method stands out by capturing specific and potential delayed association patterns within time series data that conventional correlation analysis would not otherwise reveal.

Besides the computational methods to infer the microbial association, experimental methods are required to validate the relationships between inferred taxon pairs in network. In this study, temporal shotgun metagenomics in combination with co-culture experiment were performed to link the genotype and interaction relationships among microbes in human gut. We first performed in vitro cultivation with the diluted gut microbiota (10^− 2 and 10^− 4) on YCFA agar plates. Cultures on plates were then collected in batches of every 2 days (6 time points) for metagenomic shotgun sequencing. We retrieved 198 bacterial metagenome-assembled genomes (MAGs) representing the majority (63.4%) of microbial communities in fecal sample for microbial association network inference. We further isolated the representative strains of taxa in networks and conducted co-culture experiments. Comparative genomic analysis and supplementation culture experiments revealed that the commensalism among phylogenetically closed members of Eubacteriales and Bacteroidales in gut microbiome might be dependent on nutrient exchange (e.g., amino acids, vitamins, and short-chain fatty acid (SCFA)).

Fecal sample collection and cultivation/extraction of gut microbiomes

The whole project was approved by the Research Ethics Committee of the Institute of Microbiology, Chinese Academy of Science, and the assigned number authority of the ethical approval is APIMCAS2017049. Fecal samples were collected from a healthy female from China who had not taken antibiotics within the last 3 months (Additional file 1). The samples were kept fresh and transferred into an anaerobic workstation (Coy Laboratory Products Inc., MI, USA). The gas flow composition in the anaerobic workstation was 85% (N₂), 5% (CO₂), and 10% (H₂). Fecal samples were homogenized in sterilized PBS solution containing NaCl 136 mM, Na₂HPO₄ 8 mM, KH₂PO₄ 2 mM, KCl 2.6 mM, L-Cysteine 4 mM at pH 7.4, with a ratio of 1 g stool per 1 ml PBS. The homogenate was filtered using a 40 µm cell strainer (BD Falcon, USA) and equally divided into three 1.5 mL tubes, named fecal sample (FS)-1, -2, and − 3. The samples were then serially diluted, and the dilutions 10^− 2 and 10^− 4 were plated directly onto YCFA agar respectively, named 10^− 2-FS1, -FS2, -FS3 (10^− 2 group) and 10^− 4-FS1, -FS2, -FS3 (10^− 4 group), in sextuplicate. A total of 36 YCFA plates (2 (dilution level) × 3 (time series) × 6 (sextuplicate)) were inoculated at anaerobic and dark conditions and at 37℃. One plate from each fecal sample (FS1, FS2, and FS3) was collected at 72, 120, 168, 216, 264 and 312 hours, resulted in 3 plates for each group (10^− 2 and 10^− 4) at given sampling point. Biomass collected from same dilution rate at same time point was harvested. Half of the biomass was resuspended using 1 mL 15% glycerol and stored at -80℃ until use.

Metagenomic sequencing, assembly, and genome annotation

The fecal samples (1 sample) and harvest biomass from plates (36 samples) were subjected to DNA extraction and purification using DNeasy PowerSoil Kit (Qiagen, Germany). Paired-end libraries were constructed and sequenced on Illumina NovaSeq 6,000 platform to generate 150 bp paired-end reads with a 350 bp insert size by Novogene Technology Co., Ltd., China. The raw data were trimmed by SOAPnuke (v2.1.7) [20] and were subsequently aligned to the Homo sapiens (GCF_000001405.39) reference genome to remove the host contamination using bowtie2 (v2.3.5.1) [21]. The clean reads were de-novo assembled using SPAdes (v3.13.0) [22] with the default parameters. Contigs ≥ 1,500 bp were used for binning with MetaWRAP (v1.3.2) [23]. MAGs with a completeness ≥ 50% and contamination ≤ 10% were used for subsequent analysis. The set of quality filtered MAGs were dereplicated using dRep (v2.2, nc 0.6 ANI 99) [24]. A total of 198 non-redundant bacterial MAGs were retained. The MAGs were classified with GTDB-TK (v1.5.1) based on the Genome Taxonomy Database (release95) [25]. The phylogenetic tree was inferred from concatenated alignment of 120 bacterial marker genes from the non-redundant MAGs and the selected reference genomes under the WAG model [26]. Gene prediction and annotation of the non-redundant MAGs were performed using Prokka (v1.13) [27]. Metabolic reconstruction of these MAGs was performed using EnrichM (v0.6.4) (https://github.com/geronimp/enrich). KEGG module completeness was defined as the proportion of Steps_found in Steps_needed. To assess the presence of modules in bacterial genomes, we set a threshold of 70%. If the module completeness was > 70%, we considered the module as complete in the genome [28, 29].

Microbial association networks construction and time series representations analysis

The relative abundance of each MAG was estimated using coverM (v0.6.1) (github.com/wwood/CoverM) with “--min-read-percent-identity” 0.9 and “--min-read-aligned-percent” 0.7. MAGs which were detected in less than 50% of the samples in each group (i.e., dilution rate 10^− 2 and 10^− 4) were filtered before association network construction. Moreover, the relative abundance outlier among biological triplicates was eliminated using R (v3.6.2) (https://www.r-project.org/) package “outliers” (v0.15). Microbial association networks were inferred using eLSA [19] (-r 3, -s 6, -d 2) and visualized with Cytoscape (v3.9.0) [30]. Bacterial pairs (MAG pairs) exhibit robust (|local similarity score|>0.6, |ρ|>0.6, p < 0.05) local associations and time-delayed associations were filtered for the downstream analysis. For the microbial community of each biological replicate, we conducted time series representations (TSR) analysis using R package “Tsrepr” (v.1.1.0) to cluster bacteria with similar temporal dynamic features. Bacteria clustered into consistent dynamic pattern (e.g, sustained increase or sustained decrease) across biological replicates were deemed to exhibit strong dynamic reproducibility. If both members of a robust association were classified as highly reproducible bacteria, the taxon association would be selected for validation through interaction experiments.

Bacterial isolation and cultivation

Bacterial isolation was performed using two types of samples: fecal samples and cells scraped from the surface of YCFA agar mediums (from the same sample donor). For the fecal sample, insoluble particles were removed by filtration using a 40 µm cell strainer (BD Falcon, USA), followed by serial dilution ranging from 10^− 1 to 10^− 5 using PBS. Glycerol-stored samples, scraped from the surface of YCFA agar mediums, were diluted directly under the same conditions. Subsequently, 50 µL aliquots of the diluted samples were spread onto YCFA [31] or mGAM [32] agar plates and incubated for 5 to 20 days. Single colonies were then picked from the agar plates, and their purity was confirmed by sequencing their 16S rRNA gene, following the method described in our previous study [33]. Taxonomic classification of the isolates was performed by comparing their 16S rRNA gene sequences against NCBI database with a sequence identity cutoff of 98.7%. Taxonomic assignment of MAGs was analyzed using GTDB-Tk [25]. An isolate was assigned as a representative bacterium of a given MAG if they belonged to the same bacterial species.

Pairwise co-culture with representative bacterial isolates

To validate the interactions between MAG pairs in network, representative isolates were cultivated on YCFA agar plates. Monocultures of the representative bacterial isolates were prepared in 5 mL mGAM medium with 1% (v/v = 1:1) inoculation and cultured at 37℃ for 3–4 days. The growing cells were harvested at 3,000 rpm at 4℃ for 30 min and resuspended in 0.2 mL YCFA medium. For experimental validation of microbial interactions inferred by network, the bacterial isolates representing the correlated MAGs were co-cultured on YCFA agar plate. Specifically, 3 µl of the representative bacterial isolate cultures were dripped onto the YCFA agar plate surface, followed by the addition of 3 µl of another representative bacterial isolate at external tangency to the first representative isolate. The plates were then cultivated at 37℃ for 5 days, and photographs were recorded. According to the pairwise interaction result, we could identify the interaction relationship of pairwise cultivated bacterial strains, including neutralism (0/0), commensalism (0/+), exploitation (-/+), amensalism (0/-) and competition (-/-). Then, we could visualize the validated interaction relationship by using Cytoscape (v3.9.0) [30].

Validation of metabolic interactions of beneficiaries

We defined the microbes within profitable relationship (commensalism, exploitation) that could facilitate the growth of other microbe (“beneficiaries) as “helpers”, and we defined the microbes in competitive relationship (competition and amensalism) as “competitor_A” and “competitor_B”. Module prevalence was determined as the proportion of complete modules within helpers, beneficiaries, competitor_A or competitor_B MAGs. If the prevalence > 50%, we considered the module to have high prevalence within given group. To validate whether the predicted substrate could effectively promote the growth of beneficiaries, we measured the growth curves of beneficiaries within profitable group when potential cross-feeding substrates were added or absent from the YCFA medium. We established a control group and experimental groups for this study. The control group was cultured in YCFA medium without any additives, while the experimental groups were added with 16 additives, including 9 amino acids (valine, leucine, isoleucine, lysine, histidine, methionine, tryptophan, ornithine and arginine) (100 mg/L respectively, pH 7.3), 4 vitamins (riboflavin, thiamine, tetrahydrofolate, and cobalamin) (0.5, 0.5, 0.5, and 0.01 mg/L respectively, pH 7.3) and 3 SCFAs (acetate, propionate, and butyrate) (100 mg/L respectively, pH 7.3). Subsequently, the mixed YCFA medium was added into a 96-well plate at a volume of 150 uL per well. Monocultures of the beneficiaries were prepared in 5 mL mGAM medium with 1% (v/v = 1:1) inoculation and cultured at 37℃ for 3–4 days. The growing cells were harvested at 6,000 rpm at room temperature for 5 min and remove the supernatant. The pellet was resuspended in YCFA medium, and the optical density (OD600) was adjusted to 0.5. Inoculated the adjusted bacterial culture with 1% (v/v = 1:1) into a 96-well plate, with three replicates for each sample and measured the OD₆₀₀ absorbance every 30 minutes with a microplate reader (SPECTROstar Omega, Germany). The generation time of each beneficiary was calculated as

$$Gt \left(\text{h}\text{o}\text{u}\text{r}/\text{g}\text{e}\text{n}\text{e}\text{r}\text{a}\text{t}\text{i}\text{o}\text{n}\right)= (\text{t}2-\text{t}1) / \text{l}\text{o}\text{g}2(\text{N}2/\text{N}1)$$

where Gt represented the generation time, t1 represented the starting time of the logarithmic growth phase, t2 represented the ending time of the logarithmic growth phase, N1 represented the OD₆₀₀ at the starting time of the logarithmic growth phase, N2 represented the OD₆₀₀ at the ending time of the logarithmic growth phase.

Profiling fecal microbial communities with metagenomic sequences

The experimental flow chart is shown in Fig. 1A. To characterize the microbial communities in the fecal samples, we sequenced fecal metagenomic DNAs. Meanwhile, in attempt to detect low abundant gut microbes and to extract their interactions with other bacteria, we grew fecal microbial communities with YCFA plates, and 36 cultivated fecal microbial communities (two dilutions 10^− 2 and 10^− 4, and 6 time-points for each dilution) were further sequenced. A total of 419 Gb paired-end reads of high-quality sequences were generated, with approximately 10.6 Gb per cultivated microbial communities and 35.9 Gb of fecal sample (Table S1). We recovered 198 non-redundant MAGs (Table S2), of which 84 MAGs were shared by all samples and 60 MAGs were solely assembled from fecal metagenomic data (Fig. 1B). Still, those cultivated microbial communities provided 7 (from 10^− 2) and 27 (from 10^− 4) MAGs that were not assembled from the fecal sample metagenome, suggesting the contribution of enrichment for recovering low abundant bacteria. As expected, the MAGs recovered from microbial communities cultivated at lower dilution (10^− 2 group) covered most of the MAGs from the higher dilution (10^− 4 group). Additionally, the microbial diversity of the 10^− 2 group was higher than that observed within the 10^− 4 group (Fig. 1C). The principal coordinate analysis (PCoA) based on Bray-Curtis distance showed that the microbial community compositions of cultivated microbial communities were the major factor that differentiated the 10^− 2 from 10^− 4 samples (explained 59.78% of the total variation) (Fig. 1D).

Bacterial growth patterns and reconstruction of association networks of fecal microbiota

To determine the gut microbial associations, we performed an extended Local Similarity Analysis (eLSA) using genome-based relative abundance data of the10^− 2 and 10^− 4 groups at 6 time points (see Method, Table S3, Fig. 2A and 2D). The eLSA analysis provides statistically local similarity and potentially time-delayed association patterns in replicated time series data. As expected, it was found that serial dilution reduced the number of community members and the complexity of microbial association networks. The reconstructed network for the 10^− 2 group included 110 nodes and 496 edges (containing 396 directly associations and 100 delayed associations) (Fig. 2A and Table S3), while the network for the 10^− 4 group included 56 nodes and 143 edges (containing 102 directly associations and 41 time-delayed associations) (Fig. 2D and Table S3). In 10^− 2 and 10^− 4 group, 12 identical association combinations were predicted, including 7 combinations with consistently predicted associations (positive or negative correlations in both groups) (e.g., bin_26-bin_27, bin_81-bin_37, bin_89-bin_102, bin_37-bin_69, bin_79-bin_8, bin_91-bin_99, bin_47-bin_27), and 5 combinations with predicted opposite associations (e.g., bin_101-bin_67, bin_70-bin_60, bin_89-bin_35, bin_23-bin_54, bin_82-bin_69). This suggests that serial dilution could not only simplify community composition, but also influence interactions among microbes.

We analyzed the bacterial growth patterns in each dilution groups using K-medoids clustering method. In the 10^− 2 dilution group (Fig. 2B), three patterns were recognized: Pattern 1 presented microbes (as represented by MAGs) with constantly increasing abundances; Pattern 2 represented microbes (as represented by MAG) with constantly decreasing abundances; and Pattern 3 represented microbes (as represented by MAG) with initial increases followed by decreases. Similar growth patterns were identified for the 10^− 4 dilution cultivation group (Fig. 2E). For example, bin_36 (Christensenella hongkongensis) and bin_94 (Parabacteroides distasonis) displayed constantly increased dynamic patterns in biological repeats of 10^− 2 and 10^− 4. Then, we considered that bin_36 and bin_94 showed a stable growth pattern in both groups (Pattern 1), and we considered them as one of the candidate nodes for validation.

To retrieve the robust associations from the microbial networks, only those nodes of which corresponding bacteria (as represented by MAG) exhibited consistent growth patterns were selected (Table S4). After this filtration, 100 MAGs were selected, and they showed 239 associated pairs (either local or time-delayed, Fig. 2C and 2F, Table S5) were identified from the association networks. Those MAGs and pairs were applied for guidance of microbial strain isolation and for experimental validation of the predicted microbial interactions from association network analysis.

Bacterial isolation and cultivation

In order to validate the predicted interactions, we made efforts on cultivation and isolation of microbial strains corresponding to the above selected 100 MAGs. A total of 360 isolates representing 35 taxa at genus level were obtained (Fig. 3A, Table S6). At species level, 24 isolates were matched MAGs of the 100 selected ones (Fig. 3B). To obtained further strain resources for testing the predicted interactions from association network, we searched the human Gut Microbial Biobank (hGMB) [32], and found 19 bacterial strains matching the 100 selected MAGs. Thus, we obtained totally 43 microbial isolates/strains representing 43 MAGs out of the 100 selected ones (Fig. 3B). With those 43 microbial strains, we would be able to test experimentally 60 association pairs, including 32 predicted positive and 28 predicted negative correlations (Table S7).

Co-culture of taxon pairs for validation of predicted interactions

Using the cultivated bacterial isolates, we conducted co-culture experiments to validate the microbial interactions of the robust taxon pairs predicted from in silico association networking analysis. In total, we identified 60 cultivated bacterial isolate combinations representing 60 taxon pairs in the association network. Subsequently, all 60 cultivated bacterial isolate combinations were co-cultured on YCFA agar plates, and the results (Fig. 4) showed the relationships between positive correlated taxa could be commensalism, exploitation, amensalism, and neutralism, and relationships between negative correlated taxa could be amensalism, commensalism, competition, exploitation, and neutralism (Table S7, Fig. 5). Despite the complexity of observed interactions in co-culture experiments, we found that the phenotype of neutralism was identified as the most prevalent relationship (51.67%) between the robust taxon pairs with positive or negative correlations inferred by network analysis (Table S7). Additionally, the interactions between taxa with positive and negative correlations in networks were primarily identified as neutralism (65.63% and 35.71%, respectively).

Based on the observed phenotypes, we classified the relationships between the bacterial isolates as profitable (commensalism and exploitation) or competitive (amensalism, competition, and exploitation). The phenotype of profitable interaction showed that at least one isolate could be enhanced. One example for the commensalism phenotype is that Eggerthella lenta (bin_77) is beneficial to Bacteroides ovatus (bin_86) as evidenced by the higher density of Bacteroides ovatus at the junction of these two bacterial colonies, while the Eggerthella lenta does not be affected by Bacteroides ovatus (Fig. 4B). On the other hand, for the competition phenotype, the growth of Intestinimonas butyriciproducens (bin_67) and Shigella flexneri (bin_85) were inhibited at the area closed of each other.

Two bacterial species may have similar environmental preferences or may be responding to the same external factors, will leading to a positive correlation in their dynamic profile. However, this statistical correlation does not necessarily mean that they are interacting directly or that their relationship is biologically relevant. We, therefore, assigned the neutral relationship of co-culture experiment as correct confirmation of taxon pairs in networks. By doing so, the agreement rate of the experimental results with the prediction for the 10^− 2 group was 64.29% and that of the prediction for the 10^− 4 group was 72.22%. The agreement rate of the prediction for positive correlation was 81.82% and 100% in the 10^− 2 and 10^− 4 groups, respectively, while that for negative correlation was 55% and 37.5% in 10^− 2 and 10^− 4 groups, respectively. Additionally, the final phenotype agreement rate of potentially time-delayed association patterns was 66.67% and 60% in the 10^− 2 and 10^− 4 groups, respectively. In summary, the majority of experimental results fit with the prediction results, and the positive correlations in the community were predicted more correctly.

Genome-scale metabolic analysis revealing the nutrient flow within positive taxon pairs

To link the observed phenotypic results to the genotypic relationships, we reconstructed metabolic pathway of the corresponding bacterial MAGs involved in the experimentally studied taxon pairs and summarized their KEGG module completeness. Firstly, the results of co-culture experiment were divided into two groups of interactions based on their interaction phenotype: profitable group, including commensalism (0/+), exploitation (-/+); and competitive group, including amensalism (0/-) and competition (-/-) (Table S8). We noticed that certain microbes in the profitable group acted as “helpers” that promote the growth of correlated microbe (“beneficiaries”). Interestingly, the helpers in the profitable group are highly versatile as evidenced by their significantly greater number of metabolic modules than that of beneficiaries (Fig. S1). However, this difference was not observed in the competitive group (Competitor_A vs Competitor_B). Besides, the module number in Competitor_A or Competitor_B did not significantly differ from that of helpers in the profitable group (Fig. S1).

Furthermore, the incomplete or missing metabolic modules typically found in beneficiaries (> 50% of group members) are amino acid biosynthesis, fatty acid biosynthesis, and vitamin biosynthesis. Interestingly, we found that beneficiaries were incapable of synthesizing the amino acids with the high metabolic costs, such as tryptophan, methionine, and lysine (Fig. 6, Table S8). These results suggested that the amino acid, fatty acids, and vitamins were the potential cross-feeding components between helpers and beneficiaries. In summary, the microbes with high metabolic diversity could promote the growth of auxotrophic microbes by providing them amino acids, SCFAs, and vitamins.

Moreover, within the profitable group, it was found that helpers and beneficiaries tended to have a close phylogenetic relationship, and this relationship could even be observed at a relatively higher taxonomic level (e.g., class level) (e.g., Clostridia and Bacteroidia) (Table S7, Fig S2). Specifically, 46.67% of taxon pairs of the validated profitable group were found to be affiliated with the same order, suggesting that kin selection might determine the gut microbial community assembly [34]. However, this trend was not observed in the competitive group. Members of the taxon pairs in competitive groups showed highly taxonomic divergence as none of them belonging to the same bacterial orders.

We introduced 16 additives into YCFA medium (see Methods), and monitored the growth curves of these beneficiaries (Fig. S3) for 48 hours (normalized with the first time point). Our findings revealed that additives could help most microbials (66.67%) to increase their maximum biomass, such as Bacteroides ovatus, Bacteroides clarus, Bifidobacterium adolescentis, Enterocloster lavalensis, Neobittarella massiliensis, Odoribacter splanchnicus, Parabacteroides distasonis, and Parabacteroides merdae. However, some microbes reduced of their maximum biomass, including Enterocloster citroniae. The additives had no impact on the growth status of the three microbes, including Anaerococcus vaginalis, Christensenella minuta, and Finegoldia magna (Fig. 7). Additives not only enhance the maximum biomass of microbes but also significantly reduce their generation time, including Christensenella minuta, Enterocloster citroniae, Enterocloster lavalensis, and Parabacteroides merdae. However, additives could increase the maximum biomass of Bacteroides ovatus but increasing its generation time (Fig. S4).

The aim of this study was to establish a workflow based on cultivation that could serve as a platform for predicting and investigating bacterial interactions in the gut (Fig. 1). The YCFA medium, which is commonly used for isolating and cultivating intestinal microbes [35–38].

Association network analyses based on time series metagenomics has been frequently used to infer microbial relations in environmental or host-associated microbiomes, including the gut microbiome [39]. Although microbial association networks are readily available, how to distinguish true interactions from the false ones has become an important step towards understanding the gut microbial community assembly and further controlling the gut microbiome function. This necessitates a high throughput experimental method to validate the microbial interactions inferred from in-silico analyses. However, the high microbial diversity and unstable conditions are limiting factors for the understanding of organism interactions in human gut [40]. To address this challenge, we present a strategy to simplify the gut microbial community and in vitro time course cultivation on YCFA agar medium, which is of great importance in the identification of reliable correlation in gut microbiome. In vitro cultivated time series samples could effectively avoid interference from the host gut immune system, and the community behavior can be predictable.

One hurdle in conducting co-culture experiments is cultivation of “uncultivable” microbes [41]. We found that 63.4% of indigenous microbes could be identified on YCFA plates, confirmed the effectiveness of the YCFA medium in cultivating most human gut microbes [35–38]. This also provides the basis for inferring representative microbial association networks of the human gut using in-vitro culturing experiments. Since both the microbial association networks and bacterial isolates were obtained from the in-vitro cultivation experiment, bacterial strains isolated from the YCFA medium could be directly linked to the taxon nodes in the microbial association network. The bacterial isolates obtained from the present study in combination with the exiting isolates in hGMB covered approximately 25.1% of robust taxon pairs in microbial association networks. Co-culture experiment conducted in a controlled environment has been widely applied to verify microbial interactions [42–44]. It is worth noting that the majority of (65.63%) positively associated taxon pairs in the microbial association networks have been experimentally confirmed as neutral, which was not consistent with the previous inferred that the main interaction in the gut community was negative correlation [45, 46]. On the other hand, our finding suggested that the in-silico analysis is inadequate to explore the real interactions between microbes in human gut.

We examined the phenotypes of validated positive and negative interactions among microbes. The most prevalent interaction was neutralism. Owing to the time series shotgun metagenomic sequencing, we could investigate the genome-scale metabolic models among the validated interactions. Our reconstructed metabolic models highlighted that the exchange of amino acids, SCFAs, and vitamins played a significant role in the development of positive associated microbe relationships in the human gut. One commensalism relationship between Bacteroides ovatus and Eggerthella lenta was experimentally confirmed by a previously reported study, which found that the primary cross-feeding components were amino acids (e.g., phenylalanine, valine, gamma-glutamylisoleucine) [42]. We also inferred that the competition relationship might improve the colonization resistance in the gut. For instance, Shigella flexneri, an opportunistic pathogen, could cause a gut infection known as shigellosis, with more than 160 million cases of shigellosis occurring worldwide each year [47]. We found that Intestinimonas butyriciproducens (bin_67) could inhibit the growth of Shigella flexneri (bin_85) by competition. We attempted to evaluate the effects of amino acids, SCFAs, and vitamins on the growth of beneficiaries by adding them to YCFA medium. We found that 66.7% of the experimental groups showed an enhancement in growth. Microbes that did not exhibit noticeable changes in growth may be due to the absence of certain unidentified growth factors in the medium, possibly because the incompleteness of the genome for metabolic pathway prediction.

In addition to the direct microbial interactions, it is worth noting that the high-order interaction might play an important role in community assembly and stability. While this high-order interaction can be hardly tested by co-culture experiment, we could attempt to infer the indirect relationships between microbes if one microbe involved in multiple taxon pairs. Prediction of this kind of high-order interactions may provide new insights into the gut microbial community assembly and play a critical role in regulating the community composition and function. For instance, the inferred relationship between bin_101-bin_23 was strong negative interaction, but the co-culture experiment of the representative isolates (Bacteroides clarus- Phocaeicola vulgatus) was commensalism. Remarkably, we identified isolates from two strains, Intestinimonas butyriciproducens (bin_67) and Phocaeicola vulgatus (bin_23), that showed a competition relationship when the isolates were closed. We speculated that the Intestinimonas butyriciproducens (bin_67) could inhibit the growth of Phocaeicola vulgatus (bin_23), and the Phocaeicola vulgatus (bin_23) could enhance the growth of Bacteroides clarus (bin_101) leading to the strong negative interaction between Bacteroides clarus (bin_101) and Phocaeicola vulgatus (bin_23).

It should be noted that the in vitro cultivation provides an incomplete view of the gut microbial community interaction network as microbes from given lineage could not be cultivated by YCFA medium. Despite having a high relative abundance in the fecal sample (> 2%), some gut microbes (e.g., bin_113, bin_124, bin_131, bin_153, bin_154, bin_164, bin_166, bin_176, bin_183, bin_193) from Firmicutes could not be cultivated by YCFA agar medium. Additionally, the nutritional composition can also affect the cultivable microbes. For example, some rare microbes (e.g., bin_47 and bin_117), which were not detect in the original fecal sample, were able to grow onto YCFA agar medium. To increase the diversity of the cultivable microbes, it may be necessary to combine YCFA with a multi-culture condition [32]. For the future study, we need to construct a voluminous interaction network using comprehensiveness isolates collection and we will use the enumeration method to validate the higher-order interactions of gut microbe communities.

In this study, a research workflow was proposed that integrated time-series metagenomes and co-culture of taxon pairs for validation of association network predicted bacterial interactions in gut microbiota. The high recovery rate of gut microbes on YCFA agar medium suggested that in-vitro cultivation can be a viable method to investigate the gut microbial community dynamics and interactions under relatively stable conditions. Time-series metagenomes provided a higher prediction accuracy. It provided a guidance for using representative strains to validate association nodes. Although the microbial association networks have been simplified through serial dilution, co-culture experiment demonstrated that the majority of (65.63%) the observed positive associations from it were neutralism. Therefore, the microbial association networks should be examined carefully and interpreted with caution. In addition to the neutral outcomes, confirmed positive and negative phenotypes between gut microbes suggested that the microbial association networks inferred by temporal dynamics showed relatively higher accuracy for positive interactions than negative relationships. Furthermore, genome-scale analysis and experimental analyses revealed that the microbes with versatile metabolic abilities acted as helpers in the commensalism relationships and provided the amino acid biosynthesis, fatty acid, and vitamin to the beneficiaries. Although high-order (indirect) interactions could be speculated from the experimentally validated microbial association network, interactions between multiple species experiment should be conducted in the future studies.

MAGs

Metagenome-assembled genomes

hGMB

Human Gut Microbial Biobank

SCFAs

Short-chain fatty acids

eLSA

Extended local similarity analysis

TSR

Time series representations

PCoA

Principal coordinates analysis

Optical density.

Ethics approval and consent to participate

This study was approved by the Research Ethics Committee of the Institute of Microbiology, Chinese Academy of Science. All subjects provided informed consent to be included in the study.

Consent for publication

Not applicable.

Availability of data and materials

The reconstructed metagenome-assembled genomes in the present study have been deposited in the National Genomics Data Center (NGDC) database with BioProject accession number PRJCA018951.

Competing interests

The authors declare that they have no competing interests.

Funding

This work was supported financially by the National Key Research and Development Program of China (No. 2021YFA0717002) and Taishan Young Scholars (tsqn202306029).

Author contributions

M.J. performed sampling and the experiments. C.L., C.X., and H.J. provided strain support from hGMB. M.J., Y.W. and S.L. analyzed the data and wrote the manuscript.

Acknowledgements

Not applicable

Ghoul M, Mitri S: The Ecology and Evolution of Microbial Competition. Trends Microbiol. 2016; 24: 833–845.
Mitri S, Foster KR: The genotypic view of social interactions in microbial communities. Annu Rev Genet. 2013; 47: 247–273.
Matchado MS, Lauber M, Reitmeier S, Kacprowski T, Baumbach J, Haller D, et al: Network analysis methods for studying microbial communities: A mini review. Comput Struct Biotechnol J. 2021; 19: 2687–2698.
Nash AK, Auchtung TA, Wong MC, Smith DP, Gesell JR, Ross MC, et al: The gut mycobiome of the Human Microbiome Project healthy cohort. Microbiome. 2017; 5: 153.
Coker OO, Dai Z, Nie Y, Zhao G, Cao L, Nakatsu G, et al: Mucosal microbiome dysbiosis in gastric carcinogenesis. Gut. 2018; 67: 1024–1032.
Qian Y, Yang X, Xu S, Huang P, Li B, Du J, et al: Gut metagenomics-derived genes as potential biomarkers of Parkinson's disease. Brain. 2020; 143: 2474–2489.
Peng W, Yi P, Yang J, Xu P, Wang Y, Zhang Z, et al: Association of gut microbiota composition and function with a senescence-accelerated mouse model of Alzheimer's Disease using 16S rRNA gene and metagenomic sequencing analysis. Aging (Albany NY). 2018; 10: 4054–4065.
Backhed F, Roswall J, Peng Y, Feng Q, Jia H, Kovatcheva-Datchary P, et al: Dynamics and Stabilization of the Human Gut Microbiome during the First Year of Life. Cell Host Microbe. 2015; 17: 852.
Baxter NT, Schmidt AW, Venkataraman A, Kim KS, Waldron C, Schmidt TM: Dynamics of Human Gut Microbiota and Short-Chain Fatty Acids in Response to Dietary Interventions with Three Fermentable Fibers. mBio. 2019; 10: e02566-02518.
Halfvarson J, Brislawn CJ, Lamendella R, Vazquez-Baeza Y, Walters WA, Bramer LM, et al: Dynamics of the human gut microbiome in inflammatory bowel disease. Nat Microbiol. 2017; 2: 17004.
Faust K, Raes J: Microbial interactions: from networks to models. Nat Rev Microbiol. 2012; 10: 538–550.
Odamaki T, Kato K, Sugahara H, Hashikura N, Takahashi S, Xiao JZ, et al: Age-related changes in gut microbiota composition from newborn to centenarian: a cross-sectional study. BMC Microbiol. 2016; 16: 90.
Bian GR, Gloor GB, Gong AH, Jia CS, Zhang W, Hu J, et al: The Gut Microbiota of Healthy Aged Chinese Is Similar to That of the Healthy Young. Msphere. 2017; 2: e00327-00317.
Wang X, Cheng Z: Cross-Sectional Studies: Strengths, Weaknesses, and Recommendations. Chest. 2020; 158: S65-S71.
Mars RAT, Yang Y, Ward T, Houtti M, Priya S, Lekatz HR, et al: Longitudinal Multi-omics Reveals Subset-Specific Mechanisms Underlying Irritable Bowel Syndrome. Cell. 2020; 183: 1137–1140.
Palmer JD, Foster KR: Bacterial species rarely work together. Science. 2022; 376: 581–582.
Hijova E, Bertkova I, Stofilova J: Dietary fibre as prebiotics in nutrition. Cent Eur J Public Health. 2019; 27: 251–255.
Jiang MZ, Zhu HZ, Zhou N, Liu C, Jiang CY, Wang Y, et al: Droplet microfluidics-based high-throughput bacterial cultivation for validation of taxon pairs in microbial co-occurrence networks. Sci Rep. 2022; 12: 18145.
Xia LC, Steele JA, Cram JA, Cardon ZG, Simmons SL, Vallino JJ, et al: Extended local similarity analysis (eLSA) of microbial community and other time series data with replicates. BMC Syst Biol. 2011; 5 Suppl 2: S15.
Chen Y, Chen Y, Shi C, Huang Z, Zhang Y, Li S, et al: SOAPnuke: a MapReduce acceleration-supported software for integrated quality control and preprocessing of high-throughput sequencing data. Gigascience. 2018; 7: 1–6.
Langmead B, Salzberg SL: Fast gapped-read alignment with Bowtie 2. Nat Methods. 2012; 9: 357–359.
Bankevich A, Nurk S, Antipov D, Gurevich AA, Dvorkin M, Kulikov AS, et al: SPAdes: a new genome assembly algorithm and its applications to single-cell sequencing. J Comput Biol. 2012; 19: 455–477.
Uritskiy GV, DiRuggiero J, Taylor J: MetaWRAP-a flexible pipeline for genome-resolved metagenomic data analysis. Microbiome. 2018; 6: 158.
Olm MR, Brown CT, Brooks B, Banfield JF: dRep: a tool for fast and accurate genomic comparisons that enables improved genome recovery from metagenomes through de-replication. ISME J. 2017; 11: 2864–2868.
Chaumeil PA, Mussig AJ, Hugenholtz P, Parks DH: GTDB-Tk: a toolkit to classify genomes with the Genome Taxonomy Database. Bioinformatics. 2020; 36: 1925–1927.
Parks DH, Chuvochina M, Waite DW, Rinke C, Skarshewski A, Chaumeil PA, et al: A standardized bacterial taxonomy based on genome phylogeny substantially revises the tree of life. Nat Biotechnol. 2018; 36: 996–1004.
Seemann T: Prokka: rapid prokaryotic genome annotation. Bioinformatics. 2014; 30: 2068–2069.
Zimmermann J, Kaleta C, Waschina S: gapseq: informed prediction of bacterial metabolic pathways and reconstruction of accurate metabolic models. Genome Biol. 2021; 22: 81.
Johnson MD, Scott JJ, Leray M, Lucey N, Bravo LMR, Wied WL, et al: Rapid ecosystem-scale consequences of acute deoxygenation on a Caribbean coral reef. Nat Commun. 2021; 12: 4522.
Cline MS, Smoot M, Cerami E, Kuchinsky A, Landys N, Workman C, et al: Integration of biological networks and gene expression data using Cytoscape. Nat Protoc. 2007; 2: 2366–2382.
Poyet M, Groussin M, Gibbons SM, Avila-Pacheco J, Jiang X, Kearney SM, et al: A library of human gut bacterial isolates paired with longitudinal multiomics data enables mechanistic microbiome research. Nat Med. 2019; 25: 1442–1452.
Liu C, Du MX, Abuduaini R, Yu HY, Li DH, Wang YJ, et al: Enlightening the taxonomy darkness of human gut microbiomes with a cultured biobank. Microbiome. 2021; 9: 119.
Liu C, Zhou N, Du MX, Sun YT, Wang K, Wang YJ, et al: The Mouse Gut Microbial Biobank expands the coverage of cultured bacteria. Nat Commun. 2020; 11: 79.
Harcombe W: Novel Cooperation Experimentally Evolved between Species. Evolution. 2010; 64: 2166–2172.
Forster SC, Kumar N, Anonye BO, Almeida A, Viciani E, Stares MD, et al: A human gut bacterial genome and culture collection for improved metagenomic analyses. Nat Biotechnol. 2019; 37: 186–192.
Lagkouvardos I, Overmann J, Clavel T: Cultured microbes represent a substantial fraction of the human and mouse gut microbiota. Gut Microbes. 2017; 8: 493–503.
Browne HP, Forster SC, Anonye BO, Kumar N, Neville BA, Stares MD, et al: Culturing of 'unculturable' human microbiota reveals novel taxa and extensive sporulation. Nature. 2016; 533: 543–546.
Duncan SH, Hold GL, Harmsen HJM, Stewart CS, Flint HJ: Growth requirements and fermentation products of Fusobacterium prausnitzii, and a proposal to reclassify it as Faecalibacterium prausnitzii gen. nov., comb. nov. Int J Syst Evol Microbiol. 2002; 52: 2141–2146.
La Fata G, Weber P, Mohajeri MH: Probiotics and the Gut Immune System: Indirect Regulation. Probiotics Antimicrob Proteins. 2018; 10: 11–21.
Fan Y, Pedersen O: Gut microbiota in human metabolic health and disease. Nat Rev Microbiol. 2021; 19: 55–71.
Faust K, Raes J: Microbial interactions: from networks to models. Nat Rev Microbiol. 2012; 10: 538–550.
Venturelli OS, Carr AC, Fisher G, Hsu RH, Lau R, Bowen BP, et al: Deciphering microbial interactions in synthetic human gut microbiome communities. Mol Syst Biol. 2018; 14: e8157.
Liu P, Zhang T, Zheng Y, Li Q, Su T, Qi Q: Potential one-step strategy for PET degradation and PHB biosynthesis through co-cultivation of two engineered microorganisms. Engineering Microbiology. 2021; 1: 100003.
Weiss AS, Burrichter AG, Durai Raj AC, von Strempel A, Meng C, Kleigrewe K, et al: In vitro interaction network of a synthetic gut bacterial community. ISME J. 2022; 16: 1095–1109.
Coyte KZ, Schluter J, Foster KR: The ecology of the microbiome: Networks, competition, and stability. Science. 2015; 350: 663–666.
Chang CY, Bajić D, Vila JCC, Estrela S, Sanchez A: Emergent coexistence in multispecies microbial communities. Science. 2023; 381: 343–348.
Macmicking JD: Bacteria disarm host-defence proteins. Nature. 2017; 551: 303–305.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Gut microbial interactions based on network construction and bacterial pairwise cultivation

Status:

Version 1

Abstract

Figures

Background

Methods

Fecal sample collection and cultivation/extraction of gut microbiomes

Metagenomic sequencing, assembly, and genome annotation

Microbial association networks construction and time series representations analysis

Bacterial isolation and cultivation

Pairwise co-culture with representative bacterial isolates

Validation of metabolic interactions of beneficiaries

Results

Profiling fecal microbial communities with metagenomic sequences

Bacterial growth patterns and reconstruction of association networks of fecal microbiota

Bacterial isolation and cultivation

Co-culture of taxon pairs for validation of predicted interactions

Genome-scale metabolic analysis revealing the nutrient flow within positive taxon pairs

Discussion

Conclusions

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1