The diverse roles of the microbiome with respect to health-related physiological processes and the development of different diseases have yet to be sufficiently elucidated. Some scholars have proposed that the physiological adaptation of the microbial pattern, present in pregnancy, is altered in women with metabolic diseases, such as GDM, as a consequence increased inflammation, IR, and weight gain in this population [27, 28]. However pregnant women with GDM typically maintain a stable gut microbiome for a period of time, as GDM status may interfere with the flexibility of the maternal gut microbiome, thus limiting the ability of GDM patients to respond to dietary interventions [29]. In the present study, we found that compared with healthy controls, the gut microbiota in GDM group subjects was characterized by reductions in species OTUs and beta diversity. Moreover, the distribution of species OTUs in GDM group subjects was observed to be more aggregated than that in control group individuals, with this difference in distribution found to be significantly different. Such aggregated distribution is presumed to be a characteristic manifestation of the disease state, although the specific discriminative criteria need to be further defined. In addition, we also detected significant changes in the α diversity of the GDM group microbiota compared with that in the healthy controls. Specifically, we recorded increases in Shannon and Simpson index values, and reduction in those of the Chao1 and ACE indices. These findings accordingly indicate that disease status can lead to large variations and a reduced abundance in gut microbiota among pregnant women, which can modify the interactions between different bacterial species, thereby complicating the diagnosis and treatment of GDM. We also detected significant differences with respect to the species composition ratios in GDM group and healthy control group subjects. At the phylum level, the proportion of Firmicutes in the GDM group was significantly higher than that detected in the control group, whereas in contrast, we detected significant reductions in the proportion of Actinobacteria and Proteobacteria in the GDM group. Similar differential patterns were detected at the genus level, with the proportions of Lactobacillus, Bacteroides, and Megamonas being significantly higher in the GDM group than in the control group, and the proportions of Bifidobacterium and Dialister being significantly lower Furthermore, at the species level, the proportions of Bacteroides stercoris and Bacteroides coprocola were found be significantly higher in the GDM group than in the controls. whereas the proportions of Bifidobacterium pseudocatenulatum and Escherichia coli were significantly lower (Attached Table 1).
However, there tends to be a lack of consensus regarding changes in gut microecology during pregnancy. Whereas the findings of some studies have indicated that the gut microbiome undergoes distinct changes during pregnancy, other studies have found little evidence in this respect [27, 30]. For example, in a study of gut microbiota in non-pregnant women, Fugmann et al. [31] found that the proportion of Firmicutes in women with a history of GDM was smaller than that in women without GDM. Wang et al. [32] similarly detected a lower Firmicutes composition in the oral microbiome of GDM women, although did not identify similar difference in the gut microbiome. In contrast, Jost et al. [33], who followed up seven healthy pregnant women from the third trimester to the postpartum period, found that the maternal microbiota was dominated by Firmicutes. In addition, some scholars believe that an imbalance in the ratio between Firmicutes and Bacteroidetes may represent a manifestation of biological disorder [34–36], which is consistent with the findings of the present study, in which we found that the Firmicutes/Bacteroidetes ratio in GDM group subjects (0.5898/0.3169 = 1.86) was higher than that in the controls (0.4867/0.3040 = 1.60). However, further clinical trials are needed to determine appropriate threshold values for difference in specific proportions.
By displaying our data in the form of box plot, we have more intuitively demonstrated the common bacteria characterized by significant differences in composition ratios between the GDM and healthy control group subjects, such as Bifidobacterium (Fig. 3c) and Dialister (Fig. 3d) at the genus level and Bifidobacterium pseudocatenulatum (Fig. 3e) and Bacteroides coprocola (Fig. 3f) at the species level. Given that the proportions of gut microorganisms can vary between groups, we considered this as a means whereby disease states could be distinguished. In this regard, we used linear discriminant analysis (LEfSe, LDA > 2.0) to identify 14 differential OTUs that could be applied in discriminating between the two groups (Fig. 4a), with those OTUs enriched in the GDM group mainly belonging to Lachnospiraceae and Bacteroidetes, whereas depleted OTUs were mainly from the Bifidobacterium and Actinobacteria.
At present, machine learning, such as random forest analysis, is increasingly being applied in the field of medical diagnosis [37, 38], and using this approach, researchers have established that specific combinations of different gut microbes can be used to effectively distinguish GDM individuals from healthy controls. For example, in a study published in 2017, which sought to improve the predictive power of the model, researchers attempted to provide taxonomic and functional information for unknown or unanalyzed species to enhance discrimination, and accordingly succeeded in increasing the area under the receiver operating characteristic (ROC) curve (AUC) from 0.80 (95%CI = 0.73 0.86) to 0.91 (95%CI = 0.87 0.96) [39]. However, the best combination of the model still required the inclusion 20 gut microbes. In the present study, we sought to combine different numbers of bacteria with inter-group differences at the species level and generate respective ROC curves (Fig. 4b). Contrary to our expectations, we found that a peak AUC value could be obtained using combination of considerably fewer bacteria (Eubacterium hallii, Butyrate-producing bacterium GM2.1, and Clostridium disporicum) (Fig. 4c). This combined marker panel could distinguish GDM individuals from healthy controls, with an AUC of 93.64% (95%CI: 89.83–97.45%, P < 0.001, Fig. 4d). This finding will facilitate the identification of pregnant women with GDM, contribute to disease management and treatment, and even provide a detection basis and therapeutic assistance for assessment and improvement of the intestinal status of the postpartum GDM population and newborns. Moreover, we anticipate that it will provide a new model for the diagnosis and treatment of gestational diabetes mellitus, which warrants further evaluation.
The composition ratio of Eubacterium hallii, Clostridium disporicum, and Butyrate-producing bacterium GM2.1 in the GDM group was 0.9%, 0.1%, and 0.03%, respectively. Although the composition of these bacteria is relatively small, they were nevertheless enriched in the GDM group. As a reference, these bacteria have been linked to possible disease mechanisms identified in previous studies. For example, the lactic acid or butyrate produced by bacteria can regulate intestinal permeability and induce intestinal inflammation, thereby leading to the occurrence of diabetes [40, 41]. Clostridium disporicum and other bacteria are characterized as producers of the cell wall component lipopolysaccharide (LPS), and by increasing intestinal permeability, dysregulation of the intestinal flora has been shown promote increases in levels of LPS entering the systemic circulation, thereby leading to inflammation and metabolic dysfunction [42]. In addition, the findings of studies on intestinal microbial function have indicated that LPS biosynthesis and export are involved in the regulation of blood glucose levels [39]. This will accordingly be a focus of future research to further investigate the mechanisms whereby intestinal microbes influence the occurrence and development of disease, thereby enabling the development of effective interventions.
Among the limitations of this study was the fact that stool samples were not collected from the assessed populations during the early stages of pregnancy, and consequently, although AUC values could be used to explain the difference between different populations, we were unable to identify candidate predictor bacteria during early pregnancy. In addition, all the subjects assessed in this study were recruited from a single institute, and thus prior to commencing further clinical studies, additional large-scale multicenter studies should be conducted to verify the findings reported herein.