Microbial imbalance in inflammatory bowel disease patients at different taxonomic levels

We collected faecal samples from 30 individuals (9 CD patients, 11 UC patients and 10 healthy volunteers) and performed 16S rRNA taxonomic profiling to understand changes in community structure during disease with resulting data analysed using the DADA2 pipeline. Amplicon sequence variants (ASVs) were used as a measure of diversity. As previously reported, species richness in the gut of IBD patients (both CD and UC) was lower than that of healthy volunteers [10, 13]. Moreover, within the IBD patients groups, the gut of CD patients exhibit substantially lower species richness than that of UC patients (Fig. 1a). As a measure of diversity, we identified a total of 2261 ASVs, of which 81% belong to the Firmicutes, 9.25% Bacteroidetes, 5.13% Proteobacteria, and 3.14% ASVs are from Actinobacteria. Combined, these four phyla constitute more than 98% of the total identified ASVs [22, 23]. To understand how this compared to other studies, we investigated the global microbial species-richness in the human gut across > 22,000 samples from 113 different studies from a variety of conditions (EBI metagenomics accession numbers [1] in Additional file 1: Table S1). We have considered only the known species in each study and made a unique list of gut bacterial species across studies. Similar to our experimental observations, the four phyla including Firmicutes, Bacteroidetes, Proteobacteria and Actinobacteria comprised > 94% of the total diversity in the gut, with the Firmicutes being the most species-rich phylum across conditions (Fig. 1b). However, the number of species identified as belonging to the Firmicute Phyla in our dataset (81% of all identified species from 20 IBD and 10 healthy condition samples) is substantially higher than what is usually reported at a global level (38.18%). As the observation comes from 2/3 of IBD patient and 1/3 of healthy control samples, this increased level of Firmicutes species-richness was attributed to the high number of IBD samples. Analysis of global studies for species-richness of Proteobacteria demonstrated this phyla usually accounts for 31.18% of all species, however this was substantially lower in our dataset (5.13%). Finally, the richness of Bacteroidetes was also reduced in our dataset compared to the global datasets (9.24% in our study compared to 14.35% globally). We find a similar observation when considering all OTUs from individual EBI gut microbiome studies (Additional file 2: Figure S1).

After identifying the differences at different taxonomic levels for each disease condition, for future work it is crucial to understand the reasons for such dysbiosis and whether they are causative or consequential of disease. Studies suggest that metabolic dependency [24] and nutritional preferences [25] between microorganisms are driving forces in microbial community formation. For example, metabolic cooperation between bacteria is crucial to microbial assemblages and changes to this could cause shifts across the whole community. For future work, it would be interesting to investigate the microbial metabolic interactions during disease and how this compares to a healthy gut.

Compared to the healthy controls, both IBD patient groups (CD and UC patients) demonstrated strong microbial imbalance at different taxonomic levels (Fig. 2). At the phylum level, both IBD conditions exhibit an increased abundance of Firmicutes and Actinobacteria, relative to the controls. In particular, the abundance of this phyla during UC, was far greater than CD or healthy controls. For the two other dominant bacterial phyla, the Proteobacteria and Bacteroidetes, the abundance profiles varied across disease conditions (Fig. 2a). In CD, the abundance of Bacteroidetes, which is often associated with a healthy gut, was deceased 2.4-fold, whereas the abundance of Proteobacteria, a phyla associated including wide variety of pathogens, was increased 3.8 fold. Interestingly, for patients with UC the abundance of Proteobacteria was decreased (3.4 fold) relative to controls and there was no significant differences in levels Bacteroidetes [26]. Several studies have reported the gut microbial imbalances for IBD, however, the imbalance at the level of different phylum is variable across studies [10, 13, 18, 21, 27, 28]. This could likely be a reflection on the lack of standardisation across microbiome techniques, or perhaps due to a heterogeneity in the microbiome associated with disease.

We further investigated how different taxonomic levels belonging to each of the main four phyla, Firmicutes, Proteobacteria, Bacteroidetes, and Actinobacteria, were changed during IBD. For Firmicutes, the most abundant phylum in the gut in all conditions, we observed four classes and nine different families which changed in abundance relative to healthy controls. For CD patients, the abundance of two classes including Clostridia and Erysipelotrichia, was reduced, and three families including Ruminococcaceae, Christensenellaceae and Erysipelotrichaceae were reduced relative to healthy controls. The level of two other two classes such as Negativicutes and Bacilli (obligately aerobic) and five families including Veillonellaceae, Lactobacillaceae, Acidaminococcaceae, Streptococcaceae and Peptostreptococcaceae was increased, similar to the imbalance in their parent phylum Firmicutes. Interestingly, Lachnospiraceae, the most abundant Firmicutes family, was at a similar level to the control. For UC patients, the abundance of the Firmicute classes Clostridia, Negativicutes and Bacilli, and Firmicute families Ruminococcaceae, Lachnospiraceae, Veillonellaceae, Streptococcaceae and Peptostreptococcaceae were increased. The Erysipelotrichia class and Erysipelotrichaceae family were the same as the controls, whereas, three families including Acidaminococcaceae, Christensenellaceae and Lactobacillaceae, were reduced in abundance. For Bacteroidetes, which is the only reduced phylum in CD, we observed a reduced abundance in the Bactersoidia class and Bacteroidia families Bacteroidaceae and Rikenellaceae. The Prevotellaceae family, in particular, was increased in CD patients. For UC patients, the only families to change of the Bacteroidetes was the Rikenellaceae and Tannerellaceae families which were decreased in abundance, and the Prevotellaceae which increased in abundance as with CD. Finally, for the Proteobacteria phylum, we observed an imbalance in Enterobacteriaceae and Burkholderiaceae families, with the abundance level of Enterobacteriaceae increased in CD patients and decreased in UC patients compared to the controls. Burkholderiaceae abundance was increased for both CD and UC patients. Finally, for the Actinobacteria phylum, the abundance level of both the class Coriobacteriia and family Coriobacteriaceae was increased in both IBD conditions relative to the controls, whereas the class Actinobacteria and family Bifidobacteriaceae was reduced [13, 29]. In summary, we demonstrate that multiple families of a class, classes of a phyla differ between both the IBD conditions, and between each IBD condition and healthy control. This suggests that changes in one bacterial family has consequences for others. To investigate this further, we used co-occurrence network analysis to establish patterns of how bacterial groups increase and decrease across global studies.

The gut microbiota abundance profiles from > 22,000 samples across a variety of conditions from 113 different studies was analysed to explore how different bacterial groups change across global studies. Using the Pearson’s correlation test, we built a network of significantly co-abundant (Pearson’s correlation coefficient > 0.3 and P value < 1e−10) bacterial families across a range of conditions collected from global studies (Fig. 3ai). We observed that the majority of bacterial families in the network belonged to the phyla Proteobacteria, Actinobacteria, Firmicutes and Bacteroidetes (Fig. 3aii). In the global gut bacterial family–family co-abundance network (Fig. 3ai), family nodes of 45% of the total connections are from the same phylum, compared to a random network of the same size where percentage connections were much lower (~ 30%) (Fig. 3aiii). This suggests that groups from the same phylum, which likely have similar metabolic requirements, are likely to change abundance as a collective. To understand the family level microbial imbalance during IBD, we further considered a subnetwork of the global family co-abundance network, where at least one family node was from the most abundant bacterial families in either CD, UC or healthy subjects. In this subnetwork, we highlighted bacterial families which were changed during CD or UC, compared to the healthy (Fig. 3b, c).

In CD, seven bacterial families including Coriobacteriaceae, Prevotellaceae, Burkholderiaceae, Veillonellaceae, Streptococcaceae, Pseudomonadaceae and Acidaminococcaceae have increased abundance compared to the healthy controls (Fig. 2c), however, only two families including Prevotellaceae and Veillonellaceae are connected in the global network (Fig. 3b). In contrast, the level of five other families including Erysipelotrichaceae, Christensenellaceae, Ruminococcaceae, Bacteroidaceae and Rikenellaceae, were well connected in the global network (Fig. 3b) and had reduced abundance in CD (Fig. 2c). Similarly, for UC, families with an increased abundance in UC were less well connected on the global scale (Fig. 3c). This suggests that bacterial groups which increase in abundance during IBD, are not typically associated with the healthy gut microbiome, nor do they commonly co-exist with commensals observed in the healthy gut. Furthermore, families which had increased abundance levels in healthy conditions compared to CD are very well connected, suggesting microbes in the gut of healthy individuals exist as a co-operative microbial assemblage. In particular, the connection between families such as Bacteroidaceae and Ruminococcaceae in the co-abundance network indicates that they may coexist together in across conditions, potentially due to similarities in physiology or the presence of metabolically cooperating species. For future work, it would be highly interesting to examine species of these families and investigate the relationships between these organisms.

Samples were collected from patients in standard 300 ml sterilin tubes and frozen immediately in −80 °C. Patients were asked to produce the first-morning sample for consistency and to avoid alcohol the previous 24 h. Samples were thawed and DNA was extracted using FastDNA Spin Kit for Soil (MPBiomedicals) [30] as per the manufacturer’s instructions.

16S rRNA high-throughput sequencing data was analysed by following the workflow from Callahan et al. [33]. Quality checking, filtering and trimming of fastq files were performed by functions from the dada2 package in R [34]. After filtering the reads, high-resolution Amplicon Sequence Variants (ASVs) were inferred using dada function [34]. ASVs are a higher-resolution analogue of the traditional OTUs. Chimeric sequences were removed and taxonomy assigned to ASVs based on the naive Bayesian classifier method with silva_nr_v132_train_set.fa as the training set [34]. Species-richness and alpha diversity (Shannon) were analysed by plot_richness function from the phyloseq package in R [35]. To make the rarefaction species richness curve, ‘rarecurve’ function from the vegan package [36] in R was used.

Taxonomic assignments, containing a detailed taxonomy and abundance data of OTUs or ASVs in samples, of 113 gut microbiome studies, covering more than 22,000 samples, were downloaded from the EBI metagenomics database [1]. The data were then parsed and tables containing bacterial abundance from different phyla, classes and families were generated. Abundance at the level of phylum was then used for the global gut microbial abundance. The family-level bacterial abundance was used to construct the bacterial family–family coexistence network. For each pair of bacterial families, Pearson’s correlation test was performed. Family nodes were connected when P-value < 1e−10 and Pearson’s correlation coefficient > 0.3.