Introduction

Diabetes mellitus (DM), a metabolic disorder characterized by disturbance of glucose metabolism and chronic hyperglycemia, is a significant threat to global health. The prevalence of DM has increased rapidly over the past few decades and is estimated to rise from 463 million in 2019 to 700 million in 2045 worldwide1. In Thailand, the estimated number of DM may reach 4.3 million in 20352. The most common eye complication in DM patients is diabetic retinopathy (DR), which is the leading cause of blindness worldwide2. Other ophthalmic complications include cataract, glaucoma, and ocular surface changes3,4.

The ocular surface changes in DM patients include tear film dysfunction, increased conjunctival metaplasia, decreased conjunctival goblet cell density, decreased sub-basal nerve density, decreased corneal sensitivity, and delayed epithelial and stromal wound healing5,6. Previous studies revealed that such changes were more commonly associated with type 2 DM and proportional to the severity of DR and HbA1c level3,7,8,9. Moreover, DM increases the risk of infection such as conjunctivitis, corneal ulcer, and endophthalmitis10,11,12,13,14 by altering ocular protective immune response, including decreased cytokine production and impaired cellular immune response functions15,16,17. Alterations of immune function, together with ocular surface changes in DM, may disturb the normal ocular surface microenvironment and microbiome. Moreover, the common treatment modalities in the advanced stage of DR are intravitreal drug injection or intraocular surgery. These procedures may introduce the ocular surface microbes into the eye, resulting in intraocular infection13,18. The previous reports showed more bacterial colonization and the presence of gram-negative bacteria on the ocular surface of the DM than in the non-DM groups19,20,21,22,23. Consistently, the most common organisms causing postoperative endophthalmitis reported in DM patients were coagulase-negative Staphylococcus and gram-negative bacteria, which correlates with conjunctival flora of the DM13,18,24.

The culture method is the gold standard for detecting microorganisms. This conventional technique has significant limitations when studying the ocular surface microbiome due to the detection of fastidious microorganisms and small amounts of specimens25. Therefore, the development of advanced technology using next-generation sequencing (NGS) has tremendously contributed to the current research. This high-throughput method can generate a million reads per run, thus reducing the cost and analytic time26,27,28. Many studies have reported the changes in the ocular surface microbiome in DM using the culture technique19,20,21,22,23, but only a few applied the NGS method29,30,31,32.

To our knowledge, there is no previous study on the ocular surface microbiome in different stages of DR and the level of glycemic control using the NGS method. Therefore, we conducted a study to identify the difference in the microbial community between DM and non-DM groups using the NGS method and determine the possible effects of the DR severity and glycemic control on the ocular surface microbiome. Results from this study may help improve the understanding of the pathogenesis underlying the increased risks of ocular infections in DM.

Results

A total of 80 eyes (80 subjects) included 60 eyes of DM and 20 eyes of non-DM subjects. The DM group was further classified into 3 subgroups according to diabetic retinopathy (DR) staging as no diabetic retinopathy (DM-no DR), non-proliferative diabetic retinopathy (DM-NPDR), and proliferative diabetic retinopathy (DM-PDR) (20 eyes/subgroup). The demographics and clinical details are shown in Supplementary Table 1. The mean age of all subjects was 55.6 years old (range 37–79 years old), and 50% were male. Most of the subjects were office workers who resided in the central area of Thailand. The DM subjects were also classified by glycemic control into 2 subgroups, well-controlled DM (HbA1c < 7%) and poorly controlled DM (HbA1c ≥ 7%), which was the majority (60%). The duration of DM ranged from 4 months to 36 years. Among subjects with DR, 13 eyes had diabetic macular edema, and 12 eyes had vitreous hemorrhage.

Culture and antibiotic susceptibility

A higher culture-positive rate was shown in the DM (15%), compared to the non-DM group (5%), without significant difference (p value = 0.437). There were no significant differences in the culture-positive rate among non-DM and the three DM subgroups (p value = 0.748). Seven microorganisms were identified, mostly gram-positive cocci, which Staphylococcus epidermidis being the most common. In addition, Providencia rettgeri was isolated from the DM-NPDR subgroup, while Kocuria palustris and Micrococcus luteus were isolated from the DM-PDR subgroup. There was no significant difference in the culture-positive rate between well-controlled and poorly controlled DM subgroups (16.7% and 13.9%, respectively, p value = 0.571).

Antibiotic-resistant microorganisms were only found in the DM-NPDR and DM-PDR subgroups. Resistance to benzylpenicillin, erythromycin, clindamycin, and fusidic acid in Staphylococcus hominis and resistance to ampicillin, cefazolin, and tetracycline in P. rettgeri were found in the DM-NPDR subgroup. Methicillin-resistant S. epidermidis (MRSE) was isolated from the DM-PDR subgroup. Antibiotic-resistant microorganisms were found in both well-controlled and poorly controlled DM subgroups. The details of the culture and antibiotic susceptibility are shown in Table 1.

Table 1 Results of bacterial culture in the non-DM and DM groups (with three subgroups of DR staging).
Full size table

Next-generation sequencing analysis

Illumina sequencing of 16S rRNA genes generated 3,698,312 total reads. After quality data processing, 1,786,948 high-quality reads were obtained. There was an average of 22,337 reads per sample (ranging from 4534 to 54,924). The number of amplicon sequence variants (ASVs) was 1359. The approximate saturation of microbial richness of all samples was 4534 sequencing depths, as estimated by the rarefaction curves. The plateau curve in rarefaction was observed when approximately 2000 sequencing depths were reached.

Core ocular surface microbiome

The 100% core ASVs sample matching was used to identify the core ocular surface microbiome. At the phylum level, Proteobacteria, Firmicutes, and Actinobacteria were the core microbiome of the non-DM group, and Proteobacteria and Firmicutes of the DM group. At the class level, Alphaproteobacteria, Gammaproteobacteria, Actinobacteria, and Bacilli were the core microbiome of the non-DM group, and Gammaproteobacteria and Bacilli of the DM group.

Taxonomic composition of ocular surface microbial community

Twenty-three bacteria phyla, 40 classes, 190 families, and 318 genera were identified. The bacteria of phylum Proteobacteria were highly abundant in both non-DM and DM groups (40.56% and 42.18%), followed by Firmicutes (32.43% and 34.83%), Actinobacteria (13.43% and 10.71%), and Bacteroidetes (9.35% and 6.71%) (Fig. 1a). The abundance of Actinobacteria was significantly decreased in the DM compared to the non-DM group (p value = 0.021) (Fig. 1b). Bacteria in the class of Alphaproteobacteria were significantly abundant in the non-DM group (p value = 0.007), while Gammaproteobacteria was significantly abundant in the DM group (p value = 0.036). At the family level, Enterobacteriaceae was significantly more abundant in the DM than non-DM groups (p value = 0.008). Neisseriaceae was significantly abundant in the DM-NPDR compared to the DM-no DR and non-DM groups (p value = 0.016 and 0.029, respectively). At the genus level, Escherichia-Shigella was significantly abundant in the DM compared to the non-DM group (p value = 0.016) (Fig. 1c). Pseudomonas was more abundant in the DM-PDR (6.05%) compared to the DM-NPDR (2.35%), DM-no DR (2.42%), and non-DM group (3.05%), with no statistical significance (p value = 0.78) (Fig. 1d). In subgroups classified by glycemic control, Escherichia-Shigella was also significantly abundant in both well-controlled (10.38%) and poorly controlled DM (10.02%), compared to the non-DM group (5.39%) (p value = 0.028 and 0.035, respectively) (Fig. 1e).

Figure 1
figure 1

Taxonomic composition (a) Top 15 bacteria in phylum and genus levels of the ocular surface microbiome in the non-DM and DM groups. (b) Comparison of top 5 phyla between the non-DM and DM groups. (c) Comparison of top 10 genera between the non-DM and DM groups, (d) between the non-DM and DM subgroups classified by DR staging, and (e) between the non-DM and DM subgroups classified by glycemic control. *p value < 0.05; **p value < 0.001. (Abbreviations: DM diabetes mellitus, Non-DM non-diabetes mellitus, No DR  no diabetic retinopathy, NPDR non-proliferative diabetic retinopathy, PDR  proliferative diabetic retinopathy, HbA1c hemoglobin A1c).

Full size image

To identify the possible biomarkers for differentiating the DM from non-DM subjects, the linear discriminant analysis effect size (LEfSe) was performed to determine the significant difference in the bacterial distribution between groups. The bar plot represents the effect size LDA (LDA; the linear discriminant analysis) for a significant taxon in a particular group. Bacterial taxa with LDA scores greater than 2 were considered significant. The results are shown in Fig. 2.

Figure 2
figure 2

Linear discriminant analysis effect size (LEfSe) represents the significant difference in bacterial distribution between groups. Bacterial taxa with LDA scores greater than 2 were considered significant. (a) Comparison of taxa lists between the non-DM and DM groups. (b) Comparison of taxa lists between the non-DM and DM subgroups classified by DR staging. (c) Comparison of taxa lists between the non-DM and DM subgroups classified by glycemic control. (Abbreviations: p phylum, c class, o order, f family, and g genus, DM diabetes mellitus, Non-DM non-diabetes mellitus, No DR no diabetic retinopathy, NPDR non-proliferative diabetic retinopathy, PDR proliferative diabetic retinopathy, HbA1c hemoglobin A1c).

Full size image

Alpha-diversity

Alpha-diversity was performed using observed ASVs, Chao1, Shannon, and phylogenetic diversity (PD) whole tree. All analyses showed no significant difference between the non-DM and DM groups (p value > 0.05). For the subgroups of DM classified by DR staging and glycemic control, there was no statistically significant difference in the alpha-diversity indices between each subgroup. The box plots of the Shannon diversity index are demonstrated in Supplementary Figure 1.

Beta-diversity

The principal coordinate analysis (PCoA) using unweighted UniFrac distance showed that the human ocular surface microbial community in the DM group was significantly different from the non-DM group (p value = 0.0038) (Fig. 3a). In addition, subgroup analysis showed that the microbial community in the DR subgroup (DM-NPDR and DM-PDR) significantly differed from DM-no DR and the non-DM group (p value < 0.05). However, there was no significant difference in the microbial community within the DM-NPDR and DM-PDR subgroups, as well as in the DM-no DR and non-DM groups (both p value > 0.05) (Fig. 3b). Interestingly, when classified by glycemic control, the poorly controlled DM had a significantly different microbial community compared to the well-controlled DM and non-DM groups (p value < 0.001) (Fig. 3c).

Figure 3
figure 3

This figure represents beta-diversity between groups by principal coordinate analysis (PCoA) plot based on unweighted UniFrac distance and comparison of ocular surface microbiome between groups (box plot): (a) between the non-DM (green-dotted circle) and DM groups (red-dotted circle); (b) between the non-DM (green-dotted circle) and DM subgroups classified by the DR staging (no DR: red-dotted circle, NPDR: blue-dotted circle, and PDR: purple-dotted circle); and (c) between the non-DM (green-dotted circle) and DM subgroups classified by the glycemic control (HbA1c < 7%: red-dotted circle, HbA1c ≥ 7%: blue-dotted circle) (*p value < 0.05; **p value < 0.001). (Abbreviations: DM diabetes mellitus, Non-DM non-diabetes mellitus, No DR no diabetic retinopathy, NPDR non-proliferative diabetic retinopathy, PDR proliferative diabetic retinopathy, HbA1c hemoglobin A1c).

Full size image

Discussion

This study is the first to report the ocular surface microbiome in DM according to DR staging and glycemic control using the NGS method. Our results signify the effect of these factors on the ocular surface microbial community. Previous studies described the ocular surface microbiome in DM, mainly using conventional culture technique. The culture-positive rate was shown to be higher in DM than non-DM group (21.7–94.13% and 4.3–73.3%, respectively)19,20,21,22,23. The result was reinforced by our study, although not statistically significant (15% and 5%, respectively; p = 0.437). The lower culture-positive rate in our study may be due to different specimen collection and culture techniques19,20,21,22,23.

Most microorganisms identified were gram-positive cocci with mainly coagulase-negative Staphylococcus. Gram-negative bacteria were more commonly found in the DM group, consistent with prior reports19,20. Furthermore, the colonization of pathogenic bacteria was more detected in the DM than in the non-DM group20,21. P. rettgeri, which can be found in normal healthy gut but can lead to opportunistic infection, was found in the DM-NPDR group in this study. Koreishi et al.33 reported that P. rettgeri was a potential microorganism causing ocular infections such as conjunctivitis, keratitis, dacryocystitis, and endophthalmitis in immunocompromised patients, including DM. Kocuria spp., gram-positive bacteria found in the DM-PDR group, was reported to cause canaliculitis, dacryocystitis, recurrent conjunctivitis, and severe marginal keratitis with corneal melt34,35.

Martins et al.20 reported that DM with DR had more culture-positive rate than DM-no DR subjects. However, this study found no significant difference in culture-positive rate when subgroup analysis by either the DR severity or glycemic control was done. Remarkably, this study detected the antibiotic-resistant organisms only in DM with DR group (including P. rettgeri and S. hominis in the DM-NPDR group and MRSE in the DM-PDR group). One possible explanation for this finding is that DM with DR subjects is likely to require intraocular intervention and, as a consequence, exposed to antibiotics36. For instance, several studies reported the emergence of antibiotic-resistant organisms after multiple sessions of intravitreal drug injection using prophylactic antibiotics36,37,38.

The NGS analysis showed that the top four most common phyla in both non-DM and DM groups were Proteobacteria, Firmicutes, Actinobacteria, and Bacteroidetes, which were consistent with previous studies30,31,32. At the class level, we found that Alphaproteobacteria predominated in the non-DM group, and Gammaproteobacteria predominated in the DM group, similar to the reported from Ham et al.29 Prior studies have demonstrated the abundance of potentially pathogenic bacteria in the DM group, using NGS technique29,32. Our study likewise found a higher proportion of Enterobacteriaceae, Neisseriaceae, Escherichia-Shigella, and Pseudomonas in the DM group, especially in DM with the DR group.

The ocular surface microbiome was previously reported to be more diverse in alpha-diversity in the DM compared to the non-DM group29,30,31,32. Nevertheless, our study using 4 indices did not confirm such findings. The dissimilarity may be from the different sample collection techniques, analysis, and populations. However, from the beta-diversity analysis, we can first demonstrate that the presence of DR and poorly controlled DM status can significantly change the microbial community on the ocular surface. Many groups have reported more severity of ocular surface changes in DR and poorly controlled DM3,7,8,9. As the level of advanced glycation end product (AGE) is significantly increased in patients with DM-PDR39. This level affects the ocular surface by reducing proliferation, enhancing apoptosis of epithelial cells, and inducing inflammation in the sub-basal nerve plexus5,6. Together with the alteration of protective immune response in DM15,16,17, these can lead to ocular surface microbiome dysbiosis and increase the abundance of potential pathogens in DM, especially in the DR subgroup.

The limitations of this study were small sample size and 16S rRNA incapability to detect viruses and fungi, which may be a part of the ocular surface microbiome. In addition, the data collection in a single time point may not represent the whole picture of ocular surface microbiome change with time. Further studies are undoubtedly warranted to understand the cause-and-effect relationship. Additional assessments in the metagenomics profiling, protein expression, and metabolic activity may lead us to a better understanding of ocular surface microbiome change in this significant and common disease.

Conclusion

This study demonstrated definite differences in the ocular surface microbiome between the DM and non-DM groups, the DM with DR and no DR groups, and finally, poorly controlled and well-controlled DM groups. In addition, a higher prevalence of pathogenic bacteria and antibiotic-resistant strains was more commonly found in DR compared to the other groups. These results may provide further knowledge on ocular surface diseases and the pathogenesis of many infectious diseases related to the ocular surface microbiome in DM patients.

Methods

The study was conducted in accordance with the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board, Faculty of Medicine, Chulalongkorn University (IRB No. 377/64 and COA No.704/2021). In addition, the study was registered to the Thai Clinical Trial Registration (TCTR No. TCTR20210427010).

Subjects and study design

This cross-sectional study was conducted at King Chulalongkorn Memorial Hospital (KCMH), Bangkok, Thailand. The inclusion criteria were subjects aged over 18 years old with or without diabetes mellitus. Exclusion criteria were any subjects with the following conditions: having ocular surface diseases; history of ocular and periocular infection/inflammation; having allergic conjunctivitis; previous contact lens used; history of ocular medication, systemic antibiotic, steroid, or immunosuppressive drug use; history of intravitreal drug injection or ocular surgery within the past three months. This study also excluded the subject, who was a healthcare worker.

A sample size of 20 eyes per group was calculated based on a formula comparing two independent means and standard deviation from the previous study30. From February to March 2022, 20 eyes of 20 non-diabetic subjects (non-DM group) with normal blood sugar or HbA1c tests documented within 1 year and 60 eyes of 60 subjects with a known diagnosis of type 2 diabetes mellitus by ADA criteria40 (DM group) (total 80 eyes) were recruited. The DM group was further classified by DR staging41 into three subgroups: DM-no DR, DM-NPDR, and DM-PDR groups (20 eyes per group). All groups were age- and gender-matched. This same group was also divided into well-controlled DM (HbA1c < 7%) and poorly controlled DM (HbA1c ≥ 7%). Written informed consent was obtained from all subjects before enrollment.

The subjects were interviewed for the demographic data, the Ocular Surface Disease Index (OSDI) questionnaires, and received complete ophthalmic examinations. The conjunctival swab was done before any examinations that touched the ocular surface and the instillation of any drugs, except for topical anesthesia. The processes of conjunctival swab, culture, DNA extraction, and 16S rRNA sequencing processes were done by the masked investigators.

Specimen collection

The conjunctival swab was collected from one eligible eye of each subject. If both eyes were eligible, only the right eye was included. One drop of 0.5% Tetracaine Hydrochloride solution (Alcon, USA) was applied to the inferior conjunctival fornix of the selected eye. Three minutes later, the subject’s lower eyelid was pulled down to avoid touching the lid margin or eyelashes, and a sterile cotton swab was swept gently on the inferior conjunctival surface from nasal to temporal 5 times without touching eyelids or eyelashes (rotating the sterile cotton swab each time for 360° of swab collection). Then, the swab was placed in 1 ml phosphate buffer saline (PBS) as a sterile transport medium and sent to the microbiology laboratory. All conjunctival swabs were done by the same investigator. A sterile cotton swab was placed in transport media without swabbing (blank swab) and used as a negative control.

Culture and antibiotic susceptibility test

The sterile pipette was used to transfer the sample onto a chocolate agar plate (approximately 0.1 ml). The plate was incubated in a CO2 incubator at 37 °C, 5% CO2, and checked daily for bacterial growth for a week. Bacterial identification was determined by an automated identification system (VITEK 2 XL) and mass spectrometry microbial identification (VITEK MS, France).

The antibiotic susceptibility test was determined using the Kirby-Bauer disk diffusion method and the VITEK 2 XL system. The antibiotics included benzylpenicillin, ampicillin, oxacillin, piperacillin/tazobactam, cefazolin, cefuroxime, cefuroxime axetil, ceftriaxone, cefepime, ertapenem, imipenem, meropenem, amikacin, gentamycin, ofloxacin, ciprofloxacin, levofloxacin, moxifloxacin, erythromycin, clindamycin, chloramphenicol, linezolid, teicoplanin, vancomycin, tetracycline, tigecycline, fusidic acid, rifampicin, and trimethoprim/sulfamethoxazole. The results were interpreted according to the Clinical Laboratory Standards Institute (CLSI) guidelines.

Next-generation sequencing analysis

Bacterial DNA extraction was performed using the QIAamp DNA Microbiome Kit (QIAGEN, Germany), following the manufacturing protocol. DNA was subjected to 16S metagenomic sequencing library preparation. All samples were processed in a single sequencing run.16S rRNA gene was amplified using 341F and 805R primers, targeting V3-V4 variable regions and 2X sparQ HiFi PCR Master Mix (Quantabio, USA). The PCR amplification included an initial denaturation step of 2 min at 98 °C, followed by 30 cycles of 98 °C for 20 s, 55 °C for 30 s, and 72 °C for 1 min, followed by a final extension step at 72 °C for 1 min for one time. Subsequently, 16S amplicons were purified using sparQ Puremag Beads (Quantabio, USA) and indexed using 2.5 µL of each Nextera XT index primer in a 25 µl PCR reaction, followed by 10 cycles of PCR amplification condition above. The final PCR products were cleaned, pooled, and diluted to the final loading concentration at 4 pM. Cluster generation and 250-bp paired-end read sequencing were performed on an Illumina MiSeq (Illumina, USA) at the Omics Sciences and Bioinformatics Center (Chulalongkorn University, Bangkok, Thailand).

The raw sequences were categorized into groups based on the 5’ barcode sequences. The sequences were processed following the DADA2 v1.16.0 pipeline (URL: https://benjjneb.github.io/dada2/). The microbial diversity and community structures using unique amplicon sequence variants (ASVs) were described by the DADA2 pipeline42. Microbial taxa were classified from Silva version 138 as a reference database43. Alpha-diversity index (Observed ASVs, Chao1 richness, Shannon, and PD whole tree) was computed using DADA2 software. For Beta-diversity, principal coordinate analysis (PCoA) of unweighted UniFrac distances were plotted from Phyloseq data. Linear discriminant analysis effect size (LEfSe) was performed to identify the bacterial biomarkers.

Statistical analysis

Demographic data were analyzed using descriptive statistics. The Chi-square test was used for comparing categorical data variables between groups. Independent t test or Mann–Whitney U test was performed for comparing continuous variables between two groups, while analysis of variance (ANOVA) or Kruskal–Wallis test was used for comparing more than two groups. Pairwise comparison of alpha-diversity (Observed ASVs, Chao1, Shannon, and PD whole tree) was calculated using the Kruskal–Wallis test. Permutational multivariate analysis of variance (PERMANOVA) was performed to evaluate the differences in beta-diversity among groups. Moreover, the Kruskal–Wallis sum-rank test was also used in LEfSe analysis to identify bacterial biomarkers that differed significantly in abundant taxon between sample groups. p value less than 0.05 was considered statistically significant.