A systematic literature search of four databases [PubMed^® (National Library of Medicine, Bethesda, MD, USA), the Cochrane Library, Embase^® (Elsevier, Amsterdam, the Netherlands), and Web of Science™ (Clarivate™, Philadelphia, PA, USA)] was conducted by combining medical subject headings (MeSH) words with free words. We selected articles in English published between 7 July 2012 and 7 July 2022 and focused on human studies.

The search strategy was built by combining the following MeSH words with free words: (((“Diet”[Mesh]) OR (“Life Style”[Mesh]) OR (“Exercise”[Mesh]) OR (“Motor Activity”[Mesh]) OR (“Probiotics”[Mesh]) AND ((“Cohort Studies”[Mesh]) OR (Cohort studies [Title/Abstract])) AND ((16S rRNA [Title/Abstract])) AND ((“Diabetes, Gestational”[Mesh])). Various free words were added to improve the search results and identify articles that might otherwise have been missed. The search strings are provided in Supplementary Table 1.

Titles and abstracts were screened by two investigators. Any disagreements between the two researchers were resolved by a third one. Before the formal literature selection, the three investigators were trained to ensure that they had consistent screening standards. Studies were included if they met the following criteria (1): they were cohort studies or case studies among pregnant women with GDM (2); metagenomics sequencing or 16S rRNA sequence analysis was carried out (3); they reported maternal outcomes such as HbA1c, fasting blood glucose level, and gestational weight gain (4); they were published from 7 July 2012 to 7 July 2022 (5); they were published in English; and (6) the population studied was aged >18 years.

Articles were excluded if they met any of the following criteria (1): they reported on trials that were not carried out in humans (2); they were non-randomized controlled studies (3); they analyzed probiotics in conjunction with other GDM therapies in the same intervention group (4); they were abstracts, case reports, expert opinions, reviews, letters, or editorials; or (5) they lacked sufficient data or did not meet the inclusion criteria.

Articles on pregnant women under 45 years old who had been diagnosed with GDM at a specific time point were chosen. The diagnosis of GDM was confirmed by a national or international standard. A further requirement was that the control groups should be healthy and the GDM group should not also have other metabolic diseases.

All articles were screened, and those deemed ineligible by the two researchers were removed; any disagreements were resolved by the third researcher. Detailed information was then recorded in an Excel spreadsheet. This included basic information such as authors, publication year, district, and study types, as well as maternal outcomes such as HbA_1c and fasting blood glucose levels and details of the gut microbiome, including diversities in richness, evenness, and similarity between communities.

We used the Newcastle–Ottawa Scale (NOS), which was developed for the assessment of cohort studies and case studies, to score the included studies (16). For the purpose of reaching accurate methodological quality, the NOS was evaluated through three dimensions (1): selection (2), comparability, and (3) outcome. A quality score ranging from 0 to 9 was obtained through a rating algorithm, with a score of 0–5 meaning poor quality, a score of 6–7 meaning moderate quality, and a score of 8–9 meaning high quality. The specific scores of each article according to the NOS are shown in Figure 1 and Supplementary Table 2.

Standardized mean differences (SDMs) were used to summarize and study differences among the studies and relevant measures, and 95% confidence intervals (CI) were used to estimate the differences in gut microbiota diversity between the GDM and the non-GDM groups. Operational taxonomic unit (OTU) data for each study were analyzed using RevMan 5.3 software, provided by the Cochrane Collaboration Network. Meta-analysis was performed using a random-effects model or a fixed-effects model. Sensitivity analyses were performed using the Egger test. Forest plots were created to visualize differences in microbial community structure between samples using a random-effects (RE) model and a fixed-effects (FE) model according to the reported I²-value. An I²-value > 50% and a p-value > 0.05 were considered statistically significant (17).

Qualitative data, such as the relative abundance of specific genera, were recorded for analysis. Subgroup analyses for confounding factors, such as diet, were performed. Because qualitative beta diversity was not provided, descriptive analyses could not be performed. We therefore chose to perform semiquantitative analysis. If two or more articles reported consistent study results, it was considered that the results were related to the disease and were worthy of further exploration and explanation.

A total of 632 articles identified from four databases (PubMed, 155; Embase, 34; the Cochrane Library, 48; and Web of Science, 395) were retrieved. Of these, 106 articles were duplicates and were removed, 354 articles were excluded because they reported on different study purposes, and 20 were excluded because they described other study designs. After excluding publications that did not meet the inclusion criteria, 22 (8, 18–27) studies remained and were subjected to full-text scanning for this systematic review. The flow chart illustrating the selection process is provided in Figure 1.

Among all the selected studies, 18 studies were conducted in Asia (16 from China, one from Malaysia, and one from Japan). Nine were cohort studies, while 13 were case studies. The age of the participants ranged from 28 to 45 years. The information on participants that was recorded included findings related to blood biochemistry, chronic inflammation, and biomarkers. Most of these studies used the International Association of Diabetes Pregnancy Group’s criteria; only one study, from Thailand, chose its own criteria (22). Owing to different methodologies of analysis, the storage temperature of the stool sample varied (–20°C in one study and –80°C in 19 studies). We focused on the differences emerging when comparing GDM patients with healthy people. One article divided participants into four groups according to their blood pressure and lipid measurement (20). Therefore, we took only the GDM group and the healthy group into consideration. One article did not exclude those who had probiotics or antibiotic treatment during pregnancy (23). Only two articles included the sample size calculation, which makes these studies more reliable than the others (18, 28) (Supplementary Table 4).

Quality was assessed using the NOS. Each study was evaluated based on three criteria, namely selection of the participants, comparability of groups (i.e., absence of confounder bias), and exposure (measurement of outcome influencing). Ten articles were assessed as being of fair quality and 12 as being of good quality, no studies were of poor quality The detailed scores for every study are shown in Figure 2.

In general, GDM is diagnosed at 24–28 gestational weeks, using established criteria from the International Association of Diabetes and Pregnancy Study Groups (IADPSG) based on the results of a standard 2-h, 75-g oral glucose tolerance test (OGTT) (29). Pregnant women are diagnosed with GDM if one or more glucose levels is elevated, as follows: fasting ≥ 5.1 mmol/L, 1 h ≥ 10.0 mmol/L, and 2 h ≥ 8.5 mmol/L.

The basic demographics and clinical characteristics of the participants are shown in Supplementary Table 3. The included studies involved 965 participants with GDM and 1,508 healthy people. Only four studies recorded and accounted for pre-study body mass index (BMI) in order to obtain more objective results (18, 19, 28, 30). Participants with other metabolic diseases or who had been treated with antibiotics or probiotics were excluded.

Stool samples were generally stored at –80°C, except in one study, in which the samples were stored at –20°C (18). Studies used a wide range of DNA extraction kits. Only the QIAamp Fast DNA StoolMini Kit (Portsmouth, NH, USA) was used more than twice. Target DNA sequencing in the V3–V4 region was the most common technique, used in 15 studies, whereas four studies (20, 21, 31, 32) sequenced in the V4 region, one (23) sequenced in the V1–V2 region, and one (26) sequenced in the V6–V8 region. There were four criteria for defining clustering of OTUs, namely 95% OTUs (1/22), 97% OTUs (13/22), 99% OTUs (5/22), and 100% OTUs (1/22). Taxonomy annotation was conducted mostly with an RDP (Ribosomal Database Project) classifier trained on the SILVA (6/20), RDP (3/22), and Greengenes (5/22) databases. Specific information is shown in Supplementary Table 4.

Diversities are commonly known as alpha diversity and beta diversity. They are used to describe the species composition of the gut microbiota, which may be considered a significant factor influencing outcomes.

Alpha diversity can predict both the number of the species and individual distribution, known as richness and evenness, and can be measured by the ACE (abundance-based coverage estimator), Chao1, Simpson, and Shannon indexes. Among the included studies, four (18, 19, 33, 34) compared the ACE in GDM and non-GDM patients [SMD –0.28 (95% CI –1.98 to 1.42), p = 0.75, I² = 98%]. Six studies (18, 19, 24, 33–35) reported the Chao1 index [SMD 0.48 (95% CI 0.23 to 0.73), p < 0.05, I² = 67%] for quality assessment. The Shannon index [SMD –0.06 (95% CI –0.67, 0.94), p = 0.84, I² = 88%] was provided in six studies (18, 19, 28, 33–35) and the Simpson index in seven studies (18, 19, 23, 24, 33–35) [SMD –0.44 (95% CI –1.30 to 0.41), p = 0.31, I² = 95%]. The indexes were the same in each group (Figure 3).

The result for the heterogeneity assessment was not as good as we had anticipated. The I²-value was >50%, indicating strong heterogeneity. The source of this heterogeneity should be further explored. Therefore, we further performed sensitivity analyses, omitting each study in turn, to ensure accuracy and stability of the results. If, after removing an article from the indexes, both the I²-value and the p-value were stable, indicating that the article has stable sensitivity analysis results.

In addition, to test our hypothesis that the results were reliable, we assessed the risk of publication bias using Egger’s tests, based on the symmetry of a funnel chart. The results indicated no evidence of publication bias, since the p-value of each index was above 0.05, showing that the conclusions of the meta-analysis were relatively robust. These results can be seen in Supplementary Figure 1.

As for beta diversity,17 articles reported beta diversity, of which 11 used principal component analysis (PCA) and 10 reported results using principal coordinate analysis (PCoA). We were unable to conduct a robust analysis of beta diversity because results were mostly provided in graphical form, rather than as specific data.

Owing to the significant statistical heterogeneity encountered in the analysis, several subgroup analyses were conducted separately. Analyses of classifications of gut microbiome and dietary intakes were carried out provided these were reported in at least two articles.

Subgroup analysis of the gut microbiome at genus level showed that the abundances of Blautia [SMD 0.36 (95% CI 0.02 to 0.71), p = 0.04, I² = 0%] and Collinsella [SMD –4.18 (95% CI –8.73 to 0.38), p = 0.38, I² = 78.9%] were significantly higher in the GDM group than in the control group. There were no differences in-between groups in the abundances of Clostridium [SMD –0.47 (95% CI –0.92 to –0.01), p = 0.187, I² = 42.4%] and Faecalibacterium [SMD –0.19 (95% CI –0.42 to 0.04), p = 0.971, I² = 0%].

Similarly, subgroup analysis showed that high fiber intake, compared with lower fiber intake, protects against GDM [SMD –0.96 (95% CI –0.95 to –0.43), p < 0.05, I² = 92%], whereas there were no differences between patients and control participants in energy intake [SMD –0.46 (95% CI –1.84 to 0.93), p = 0.52, I² = 97%], cereal intake [SMD –0.31 (95% CI –0.92 to 0.29), p = 0.31, I² = 86%], meat intake [SMD –0.08 (95% CI –0.31 to 0.15), p = 0.15, I² = 93%], or milk intake [SMD –0.66 (95% CI –1.55 to 0.24), p = 0.15, I² = 93%], although there remained considerable heterogeneity in all analyses. Specific results and data can be viewed in Table 1.

The current study compared the gut microbiota in patients with GDM and healthy participants. Nine (18, 19, 21, 23, 24, 27, 30, 34, 36) articles analyzed the taxa at the phylum level. Three (18, 19, 24) articles mentioned Firmicutes and concluded that they were less abundant in GDM patients, whereas two (30, 36) articles reported the opposite results. Bacteroidetes are more abundant in GDM patients than in participants without DGM. Two articles (23, 34) reported that the abundance of Actinobacteria was lower in GDM patients than in those without GDM. Four studies (24, 34, 36, 37) reported that the abundance of Proteobacteria was higher in GDM patients than in healthy controls. Only one study (23) reported the opposite. Three studies (24, 27, 30) compared Firmicutes/Bacteroides, two of which (24, 30) found that their abundance was higher in the GDM group.

At the class level, one (21) study mentioned that Rothia, Actinomyces, Bifidobacterium, Adlercreutzia, and Coriobacteriaceae from Actinobacteria were reduced in the GDM population.

At the family level, there were slight differences between studies. Both Veillonellaceae (21, 37) and Prevotella group 9 (30, 37) were increased in GDM patients, whereas Lachnospiraceae and Ruminococcaceae (24) were decreased. Some studies drew controversial conclusions as regard Streptococcaceae, Enterobacteriaceae, Lachnospira, Clostridiales, Clostridia, and Firmicutes. Two (24, 36) studies found that, at the family level, members of the family Clostridiales were increased in GDM patients.

Studies reporting findings at the genus level were the most common. We found enrichment of Ruminococcaceae, Lactococcus, Escherichia, Lachnospiraceae, Clostridia, Alistpes, Firmicutes, and Phascolarctobacterium. Results regarding Streptococcus and Bacteroidetes varied. We also found enrichment of Coprococcus, Staphylococcus, Oscillospira, Burkholderiales, Akkermansia, Prevotella group 9, and Faecalibacterium in participants without GDM. Abundances of Blautia, Staphylococcus, Sutterella, Oscillospira, Enterococcus, and Lactobacillus also varied, and this needs further study.

We performed further random forest analysis of four dominant bacteria. Proteobacteria are relatively abundant in GDM [SMD –0.44 (95% CI –1.3 to 0.41), p = 0.31, I² = 95%], whereas the abundances of Actinobacteria [SMD –1.64 (95% CI –2.32 to –0.95), p < 0.05, I² = 94%], Bacteroides [SMD 7.96 (95% CI –7.77 to 23.69), p = 0.32, I² = 100%], Firmicutes [SMD 0.25 (95% CI –0.01 to 0.51), p = 0.06, I² = 49%], and Proteobacteria [SMD 1.10 (95% CI –1.59 to 3.80), p = 0.42, I² = 99%] did not differ significantly between the two groups.

To further explore the potential correlations of key clinical indexes with altered gut microbiome in GDM, correlation analyses were performed using Spearman analysis. The results showed that phylum Bacteroidetes was positively associated with 1hPG, whereas Proteobacteria, Verrucomicrobiota, and Actinobacteria were all negatively associated with 1-hour plasma glucose level (1hPG) levels. Analysis at the genus level revealed a negative association between Ruminococcaceae UCG014 and 1hPG, but a positive association between Ruminococcaceae UCG014 and high-density lipoprotein (HDL) levels (36). The genus Akkermansia was negatively correlated with 1hPG and positive correlated with HDL levels. According to Chen et al., genera of Clostridiales, Ruminococcaceae and Lachnospiraceae, within the phylum Firmicutes, were significantly negatively correlated with at least one OGTT value (19). An unassigned genus of Enterococcaceae within the phylum Firmicutes, the genus Atopobium within the phylum Actinobacteria and the genus Sutterella within the phylum Proteobacteria were significantly positively associated with 1-h or 2-h OGTT values (Supplementary Table 5; Figure 4).

Indexes of inflammation, fecal calprotectin (FCALP), lipopolysaccharide (LPS), lipopolysaccharide-binding protein (LBP), and fecal LPS (FLPS), were reported in two (20, 26) studies. Levels of zonulin, FCALP, LPS, LBP, and FLPS were higher in GDM patients than in those without GDM, and the results were statistically significant. Cui et al. reported that Enterococcus and Vagococcus are in direct proportion to FCALP and LPS, Streptococcus is inversely proportional to LBP, and Staphylococcus is directly proportional to FLPS (20). Women’s diet, including total energy and fiber intake, remained unchanged between sampling times.

Four articles reported Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis. The conclusions drawn in each article vary widely. Chen et al. identified several microbial gene pathways including the glycan biosynthesis and vitamin metabolic pathways (19). Li et al. showed that the species of gut microbes found in normoglycemic pregnant women (NOR) and GDM are involved in cell wall/membrane/envelope biogenesis, organic ion transport and metabolism, post-translational modifications, protein turnover, chaperones, transcription, unknown function, intracellular transport, secretion, and vesicular transport (24). Su et al. found a positive relationship between Bacteroides species enriched in patients with GDM, and amino sugar and nucleotide sugar metabolism (34). Wang et al. found that the predicted metagenome of women who developed GDM was enriched in organisms involved in starch and sucrose metabolism, whereas those implicated in lysine biosynthesis and nitrogen metabolism were reduced (37).