Countless gut microbiota colonizes the human intestine. The number of bacteria in the adult gut increases to 10¹⁴, weighing approximately 1.13 kg, which is 10 times more than the total number of human cells (10¹³) (Shivaji, 2017). These microorganisms constitute a complex symbiotic ecosystem and are believed to interact with the host environment, thus influencing important physiological processes such as nutrient production, signaling pathway modulation, energy homeostasis, and anti-inflammatory responses (Pflughoeft and Versalovic, 2012). In recent years, disorder in gut bacterial composition and function has been associated with many human systemic diseases, including diabetes mellitus (Qin et al., 2012), obesity (Haro et al., 2016), depression (Zheng et al., 2020), Alzheimer’s disease (Wang et al., 2019), and cardiovascular disease (Battson et al., 2018). However, dysbiosis of the gut microbiome varies from person to person, showing a wide range of genomic variations that need to be elucidated.

Diabetes is one of the fastest growing health challenges of the 21^st century, with the number of adults living with the disease having more than tripled over the past two decades. According to the latest edition of the International Diabetes Federation (IDF) Diabetes Atlas, there are 463 million adults currently living with diabetes, and this number is estimated to reach 578 million by 2030, and 700 million by 2045 (IDF, 2019). Type 2 diabetes mellitus (T2DM) is the most common type of diabetes and is characterized by high blood glucose levels caused by the inability of the body’s cells to fully respond to insulin (IDF, 2019). T2DM is a multifactorial disease that is influenced by both environmental and genetic factors and is the leading cause of kidney failure, cardiovascular disease, and retinopathy (IDF, 2019). A 2010 study, which initially demonstrated the influence of gut microbiota on diabetes, reported the association between Firmucutes and Bacteroidetes was altered in Denmark diabetic patients compared with non-diabetic individuals (Larsen et al., 2010). Subsequent studies confirmed the function of microbiota in systemic metabolism and T2DM. Specifically, Qin et al. (2012) and Karlsson et al. (2013) undertook metagenomic sequencing in Chinese and Swedish diabetic individuals, respectively, and established that T2DM was characterized by a dysbiotic gut microbiota. Further studies indicated that microbial dysbiosis is associated with insulin resistance, abnormal lipid metabolism, and T2DM (Aron-Wisnewsky et al., 2021). Several underlying mechanisms have been demonstrated in mouse studies, including host signaling through lipopolysaccharides (LPS) derived from cell walls of gram-negative bacteria, short-chain fatty acids (SCFAs) produced by bacterial fermentation of dietary fiber, and bacterial regulation of bile acids (Allin et al., 2015).

Diabetic retinopathy is a major complication of diabetes and a leading cause of blindness and vision impairment. It can lead to vision loss due to macular edema and ischemia, as well as neovascularization of the retina and iris, such as vitreous hemorrhage, retinal detachment, and neovascular glaucoma (Cheung et al., 2010). In the USA, approximately 50% of individuals with T2DM may develop retinopathy or vision-threatening retinopathy (Kempen et al., 2004). The complications and increased prevalence of T2DM necessitate a new understanding of its biology and the development of effective approaches for its prevention and treatment. Increased understanding of the gut microbiota’s link to T2DM will provide new clinical strategies against diabetes. For example, several randomized controlled trials have indicated that standardized fecal bacterial transplantation can reduce insulin resistance and improve insulin sensitivity, halting the progression of diabetes (Udayappan et al., 2014; de Groot et al., 2021).

Since the idea of “gut–retina axis”, in which the gut microbiome modulated by diet, probiotics, or antibiotics, influences the development of retinal disease, was proposed, the significance of gut microbiome as a major modulator of eye disease has been increasingly recognized (Rowan et al., 2017; Rowan and Taylor, 2018). Gut microbiome dysbiosis is associated with many eye diseases, such as uveitis (Janowitz et al., 2019), glaucoma (Gong et al., 2020), and age-related macular degeneration (Zinkernagel et al., 2017). Furthermore, Beli et al. (Beli et al., 2018) reported that alterations in the gut microbiome via intermittent fasting could prevent retinopathy and prolong survival in mice, indicating that the gut flora can participate in the pathological process of diabetic retinopathy. However, it is still unclear whether differences exist in human gut microbiota between diabetic patients with and without retinopathy. Studies on gut microbial dysbiosis have identified the Bacteroidetes phylum using culture methods or have focused on mycobiome dysbiosis through Illumina sequencing of the ITS2 region (Moubayed et al., 2019; Jayasudha et al., 2020). Therefore, this study set out to systematically detect and identify differences in gut microbial compositions of diabetic individuals with (DR) or without (DM) retinopathy and to compare each group with healthy individuals (HC) based on 16S ribosomal RNA (rRNA) gene sequencing. The study aimed to provide a non-invasive method for diagnosis and identification of potential microbial targets that would enable further studies on the pathogenesis and therapeutic strategies for diabetic retinopathy.

The occurrence of diabetes and related retinopathy is closely related to gut microbial dysbiosis. Here, we applied 16S rRNA gene sequencing to obtain a comprehensive composition of gut microbiota in DM and DR patients and in healthy individuals. In addition to the unique microbial genera identified in each group, we also compared distinct genera and families of bacteria in DM and DR groups with the HC group and with each other. Furthermore, we selected and confirmed 25 biomarkers at the family level using the random forest prediction model, which could distinguish DR from DM, and both from HC, with a reliable diagnostic accuracy. Therefore, our results should be explored further in clinical trials. Moreover, the selected biomarkers can be used not only for clinical diagnosis but also as targets of novel therapeutic interventions for diabetic retinopathy.

In this study, we found that patients in the DM and DR groups had distinct gut microbiota compared with controls, based on reduced diversity and altered microbial composition. Overall, α-diversity indices (Invsimpson, Shannoneven, and Heip) were lower in DM and DR than in HC. Other indices (Ace, Chao, and Shannon) were lower in the DM group, but non-significantly lower in the DR group. When DR was compared with DM, an increased trend was observed in DR, although it did not reach statistical difference. DR was characterized by alterations in specific OTUs mainly assigned to the families Acidaminococcaceae, Muribaculacea, Atopobiaceae, and norank_o_Coriobacteriales. Herein, we found twenty-two genera in DR only compared with eight in DM and thirteen in HC ( Figure 3C ). We speculate that the possible explanation was the microbiota in DR might tend to exhibit a more complex and higher pathology diversity. Moreover, differences in β-diversity (PLS-DA) were observed in the three groups. Our results were consistent with those of previous studies, which reported that lower α-diversity indices (Shannon and Chao) of gut microbiota and different β-diversity were observed in T2DM patients compared with non-diabetic individuals (Nuli et al., 2019; Zhao et al., 2019; Reitmeier et al., 2020). Therefore, we postulate that the onset and progression of diabetes and its retinal complications are related to alterations in the gut microbiota.

We also found that DM and DR were associated with disturbance of Firmicutes, Bacteroidetes, and Desulfobacterota phyla. An earlier study showed that the proportions of Firmicutes and Clostridia (belonging to phylum Firmicutes) were significantly reduced, while that of Bacteroidetes was increased, in the gut of T2DM patients (Larsen et al., 2010). Similarly, we observed that abundances of Firmicutes and related Clostridiaceae in the DM and DR groups were lower than in healthy participants. Additionally, we found that Bacteroidetes was more abundant in the DR than in the DM and HC groups, which is in line with other study findings (Moubayed et al., 2019). Bacteroidetes are gram-negative bacteria whose cell walls are mainly composed of LPS. LPS derived from members of Bacteroidetes not only inhibit innate immune signaling and endotoxin tolerance, but are also associated with diabetic pathogenesis (Davis-Richardson et al., 2014; Vatanen et al., 2016). Vagaja et al. (Vagaja et al., 2013) reported that systemic LPS exposure in hyperglycemic mice could accelerate the injury of retinal capillary endothelium and thinning of the retina. Interestingly, we noted that, compared with the HC group, Desulfobacterota was enriched in the DR group but reduced in the DM group. Desulfobacterota is composed of many organisms that can reduce sulfur compounds via the DsrAB-dissimilatory sulfite reduction pathway (Waite et al., 2020), and these organisms participate in butyrate degradation by carrying out a butyrate beta-oxidation pathway, indicating that Desulfobacterota is involved in the equilibrium of the catabolic reaction (Hao et al., 2020). Therefore, it may be hypothesized that bacteria in the Bacteroidetes and Desulfobacterota phyla can either release LPS to trigger inflammatory injuries or aggravate energy metabolism abnormalities, both of which are pathological features of diabetes.

At the genus level, decreased Faecalibacterium, Eubacterium_hallii_group, and Clostridium (Clostridium_sensu_stricto_1 and norank_f_norank_o_Clostridia_UCG-014) genera and increased Blautia, Bifidobacterium, and Lactobacillus were observed in both the DM and DR groups. For genera with decreased abundance, Faecalibacterium, Eubacterium_hallii_group, and Clostridium are known human gut colonizers and butyrate producers and are highly discriminant for T2DM (Karlsson et al., 2013; Hiippala et al., 2018). These bacteria ferment dietary fibers to produce butyrate through gut microbial metabolism, thus playing a regulatory role in improving insulin sensitivity, alleviating inflammation, and ameliorating diabetes in humans and mice (Vrieze et al., 2012; Udayappan et al., 2016; Hiippala et al., 2018). Similarly, reduced Clostridium was also found in both Chinese and European T2DM patients (Qin et al., 2012; Karlsson et al., 2013). Karlsson et al. (2013) further reported that Clostridium correlated negatively with fasting glucose, insulin, and plasma triglycerides, and positively with adiponectin and high-density lipoprotein, all of which are closely related to T2DM. Hence, we speculate that the occurrence of diabetes and retinopathy may be associated with reduction in these beneficial bacteria.

Blautia and lactobacillus genera were enriched in participants with diabetes and were more abundant in the DM group than in the DR group. These increased Blautia and lactobacillus genera in diabetic patients were also observed in other studies (Larsen et al., 2010; Karlsson et al., 2013; Wang et al., 2020). Wang et al. (2020) reported that Blautia was positively correlated with tauroursodeoxycholic acid (TUDCA) levels. TUDCA, a farnesoid X receptor (FXR) antagonist, regulates the glycolipid metabolism via its receptor for FXR and bile acid G-protein-coupled membrane receptor (TGR5). TGR5 can be found in retinal primary ganglion cells, and the activation of TGR5 could prevent retinopathy and prolong survival in db/db mice (Beli et al., 2018). This may explain why diabetic patients with higher Blautia are less likely to develop retinopathy. Increased Lactobacilli levels were positively correlated with fasting glucose, glycosylated hemoglobin, and a long-term measure of blood glucose control (Karlsson et al., 2013). Thus, increased Lactobacillus in the intestines could be a consequence of increased glucose levels in diabetic patients. However, lactobacillus genus represents a common probiotic with well immunomodulatory and antioxidant properties, which involved with the mechanism of diabetic retinopathy. Zeuthen et al. (Zeuthen et al., 2006) reported that lactic acid bacteria (mainly belong to lactobacillus) differentially up-regulated dendritic cells’ surface maturation markers and potentially contribute to the regulation of chronic inflammation. Another study revealed that oral administration of lactobacillus strains improved antioxidant capacity in diabetic mice (Dang et al., 2018). These beneficial effects of lactobacillus could partially explain the result that it was more abundant in the DM group than in the DR group. The current study found that some drugs for diabetes treatment had important effects on gut microbiome alterations. For example, Wu et al. (2017) reported that metformin significantly influenced the gut microbiome (including increased Bifidobacterium) to improve glucose tolerance and enhance the antidiabetic effects of the drug. Most subjects in the DM and DR groups of our study used metformin, and we noticed that Bifidobacterium abundance was higher in these groups than in the HC group. Therefore, the increased abundance of Bifidobacterium may be attributed to the effect of metformin.

Furthermore, apart from identifying the gut microbiota, microbial biomarkers should be screened that can be used to distinguish diseases. In this study, we identified 25 important families that could distinguish individuals with DR from those with DM or HC, with AUC values ranging from 0.69 to 0.85. To further develop the potential reliability of these biomarkers as diagnostic tools, we took the duration and fasting plasma glucose levels and determined their correlation with selected biomarkers. We found 14 different families, apart from Eubacteriaceae, which showed a positive correlation with diabetes-related clinical parameters; all the other families showed negative correlations. Most families, such as Clostridiaceae, Peptostreptococcaceae, norank_o_Clostridia_UCG-014, and Butyricicoccaceae were highly expressed in the HC group. Generally, Clostridiaceae and Peptostreptococcaceae are negatively correlated with metabolic diseases and participate in the production of SCFAs, such as acetate, propionate, and butyrate (Fei et al., 2019; Oliphant and Allen-Vercoe, 2019). Previous studies reported that SCFAs could improve glucose homeostasis and were associated with reduced risk of T2DM, obesity, and cardiovascular disease (Chambers et al., 2018; Zhao et al., 2018). This may explain the observation why their abundances were decreased in the DM and DR groups compared to controls. However, we also noted that some bacteria in DR were closer to HC in comparison with the DM group; it was partially contradictory with expectation. We presume that one possible explanation is bacterial compensatory growth. Sometimes, the abundances of SCFA-producing bacteria and the fecal SCFA concentrations are not consistent. For example, Zhao et al. (2019) reported that in most T2DM patients with diabetic complications, the abundances of SCFA-producing bacteria were increased, while the fecal SCFA concentrations, which reflect a balance status between gut production and absorption of SCFAs (Fernandes et al., 2014), were decreased. On the other hand, hyperglycemia drives intestinal barrier disruption and dysfunction of intestinal epithelial cells, which absorb and utilize SCFAs as their key energy source (Thaiss et al., 2018; Yao et al., 2020). Most of these contradictory microbiota are SCFA-producing bacteria; we presume that they might present compensatory growth in order to promote energy harvest by producing more beneficial SCFAs. Another possible reason is that the role of these bacteria was not simply to produce SCFAs. However, further research is needed, as there remains no mechanistic explanation of these microbiota toward diabetic development. In addition, we identified some microbiota, including norank_o_Clostridia_UCG-014 and Butyricicoccaceae, which are worth further analysis, even though they have not been previously reported.

Eubacteriaceae was the only family that showed a positive correlation with fasting glucose level and was significantly enriched in diabetic patients, especially in the DR group. Generally, Eubacteriaceae is one of the predominant components of human gut microbiota and is abundant in healthy people (Arrieta & Finlay, 2012). However, Eubacteriaceae is also abundant in some immune and metabolic diseases. For example, Eubacteriaceae level was elevated at the time of acute mucosal simian immunodeficiency virus replication, and it was more abundant in pre-surgery, among those who went into remission from T2DM in post-surgery (Allers et al., 2020; Davies et al., 2020). The above findings indicate that Eubacteriaceae may play an important role in immune and metabolic diseases. Although we cannot explain the reason for the observed increase in Eubacteriaceae in the DM and DR groups, this observation is consistent with the relationship between Eubacteriaceae and metabolic diseases. Moreover, a previous study showed that Pasteurellaceae, which mainly colonizes the mucosal surface of the respiratory and genital tracts, is an important human and animal pathogen, and can initiate infection by employing multiple molecular mechanisms of adherence (Jacques and Paradis, 1998). Paradoxically, we found that Pasteurellaceae was significantly enriched in the DM group but significantly reduced in the DR group. Furthermore, using only Pasteurellaceae as a biomarker to distinguish between the DM and DR groups reached the highest accuracy (AUC: 0.74). We propose that Pasteurellaceae might be the major microbiota, and its effects on DM and DR requires further studies involving larger sample sizes.

The main strength of this study is that the recruited subjects were strictly control-matched, with well-characterized clinical information, thus reducing bias arising from potential confounding effects such as age, sex, and BMI. In addition, previous studies have shown that the gut microbial composition may be influenced by antibiotics (Mikkelsen et al., 2015; Ianiro et al., 2016). We recruited subjects with no recent history of antibiotic use. However, there were some limitations to our study. For example, there is a certain admission-rate bias because a large proportion of subjects were selected in the hospital. In addition, owing to lack of detailed dietary information for the subjects, our study could not control for or assess whether and how dietary habits influenced gut microbial composition in the DM, DR, or HC groups. As Hunan province is located in southern China, and rice is the predominantly food here, whether the dietary habits of different regions have any effects on the microbial changes will require further multicenter controlled studies. Furthermore, our findings do not show a causal relationship between the identified differential gut microbial composition in each group, which is an inherent limitation of cross-sectional studies. To extend and advance the current work, a larger sample cohort and other emerging analytical strategies need to be considered in future studies.

We have characterized and identified distinct gut microbial compositions in subjects in the DR and DM groups and compared them with healthy controls. Moreover, we have developed and validated a gut microbial classifier model that can effectively discriminate the DR from the DM and HC groups. The current findings lay the foundations for future studies, based on distinct gut microbiota, for developing better clinical diagnostic methods and identifying potential therapeutic targets of diabetic retinopathy.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: NCBI BioProject, accession no: PRJNA688998.

The studies involving human participants were reviewed and approved by the Ethics Committee of the AIER Eye Hospital Group (Ethics No. AIER2018IRB21). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

YH was involved in the conception and design of the study. YH, ZW, HM, SJ, ZCh, and ZCu were involved in the collection and assembly of data. YH, JC, and ST were involved in interpreting the data and wrote the manuscript. All authors contributed to the article and approved the submitted version.

This work was supported by the National Natural Science Foundation of China (81871495) and the Science and technology project of Changsha, Hunan (kh1901251), Key Research and Development Program of Jiangxi Province (20203BBGL73193), and Science Research Foundation of Aier Eye Hospital Group (AM2001D1).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.