Metagenomic Sequencing Analysis Identifies Cross-Cohort Gut Microbial Signatures Associated With Age-Related Macular Degeneration

Abstract

Purpose: Alterations in the gut microbiota have been associated with age-related macular degeneration (AMD). However, the dysbiosis shared by different ethnicity and geographic groups, which may associate with the disease pathogenesis, remain underexplored. Here, we characterized dysbiosis of the gut microbiota in patients with AMD from Chinese and Swiss cohorts and identified cross-cohort signatures associated with AMD.

Methods: Shotgun metagenomic sequencing was performed on fecal samples from 30 patients with AMD and 30 healthy subjects. Published datasets with 138 samples from Swiss patients with AMD and healthy subjects were re-analyzed. Comprehensive taxonomic profiling was conducted by matching to the RefSeq genome database, metagenome-assembled genome (MAG) database, and Gut Virome Database (GVD). Functional profiling was performed by reconstruction of the MetaCyc pathways.

Results: The α-diversity of the gut microbiota was decreased in patients with AMD according to taxonomic profiles generated using MAG but not RefSeq database as reference. The Firmicutes/Bacteroidetes ratio was also decreased in patients with AMD. Among AMD-associated bacteria shared between Chinese and Swiss cohorts, Ruminococcus callidus, Lactobacillus gasseri, and Prevotellaceae (f) uSGB 2135 were enriched in patients with AMD, whereas Bacteroidaceae (f) uSGB 1825 was depleted in patients with AMD and was negatively associated with hemorrhage size. Bacteroidaceae was one of the major hosts of phages associated with AMD. Three degradation pathways were reduced in AMD.

Conclusions: These results demonstrated that dysbiosis of the gut microbiota was associated with AMD. We identified cross-cohort gut microbial signatures involving bacteria, viruses, and metabolic pathways, which potentially serve as promising targets for the prevention or treatment of AMD.

Age-related macular degeneration (AMD) is a major cause of irreversible visual impairment and blindness in the elderly, affecting 196 million people around the world.¹ The clinical manifestations of early AMD include discrete drusen deposits and pigmentary changes. Advanced AMD is characterized by either geographic atrophy (dry AMD) or neovascularization (wet AMD). Of note, neovascular AMD carries a greater risk of blindness than dry AMD.² Due to the fact that AMD pathogenesis remains unclear, preventive and therapeutic approaches for the disease are limited. Currently, there is no approved therapy for dry AMD, whereas treatment of neovascular AMD relies on repeated injections of antivascular endothelial growth factor (anti-VEGF) agents.³ 

Emerging evidence suggests that dysbiosis of the gut microbiota plays an important role in the onset or exacerbation of multiple ocular diseases, including autoimmune uveitis,⁴ glaucoma,^5,6 and AMD.^7–9 One possible explanation is that gut microbiota and metabolites induce systemic inflammation or irritate local (ocular) inflammation via the blood and lymphatic systems.¹⁰ An early study using mouse models of neovascular AMD provided the first evidence for the influence of the gut microbiota on pathological angiogenesis.⁷ Another study using wild-type aged mouse models revealed that specific members in the gut microbiota were associated with AMD features.⁹ These findings opened the possibility of AMD prevention and treatment by manipulating the gut microbiota. 

Characterizing the gut microbial dysbiosis in patients with AMD is a crucial step toward clinical translation. Two previous metagenomic studies demonstrated an association between aberrant gut microbiome profiles and AMD in the Swiss population.^11,12 However, among all factors influencing the composition of the gut microbiome, such as geographic differences, diet, lifestyle, and genetic background, ethnicity and geography are the two explaining more of the differences found in the fecal microbiome.¹³ Therefore, it is necessary to investigate the dysbiosis characteristics associated with AMD in populations from different ethnicity and geographic groups and the shared features of gut microbiome may represent a true association with AMD. On the other hand, common reference databases for taxonomic profiling are unable to capture full diversity of the gut microbiota, so that many gut microbial species may go undetected.¹⁴ Recently, thousands of new microbial species have been discovered by large-scale metagenomic assembly.^15–17 These new genome resources offer unique opportunities to elucidate the full spectrum of gut microbial diversity and explore the effects of the gut microbiome on host health and disease. 

In this study, we performed shotgun metagenomic sequencing on fecal samples from patients with neovascular AMD and healthy subjects in the Chinese population. Comprehensive taxonomic profiling was conducted leveraging the RefSeq genomes and genome resources provided by large-scale metagenomic assembly studies. Then, functional profiling was performed to assess the metabolic potential of the gut microbiota. In addition, the possible association between gut microbiome and clinical features of wet AMD diseases, including visual acuity, hemorrhage size, lesion size, angiogram dye leakage, and fovea thickness, were demonstrated. Furthermore, we re-analyzed the AMD dataset with shotgun metagenomic samples from previous studies in the Swiss population to identify cross-cohort microbial signatures of AMD. Our analysis revealed that dysbiosis of gut bacteria, viruses, and metabolic functions were involved in AMD across cohorts. These findings lay the groundwork for potential microbiome-based interventions against AMD. 

Materials and Methods

Participant Recruitment

This study was conducted in compliance with the tenets of the Declaration of Helsinki for biomedical research involving human subjects and was approved by the Ethics Committee of China-Japan Friendship Hospital, Beijing, China (Protocol #ZRLW-2015-2). All participants received full explanation of study goals and thereafter provided written informed consent. Participants (n = 60) were recruited from the China-Japan Friendship Hospital. All participants were aged between 50 and 86 years. Patients (n = 30) with clinically confirmed active and treatment-naive neovascular AMD were included in the study. The demographic and clinical information are listed in Supplementary Table S1. The healthy controls (n = 30) free of ocular manifestations were selected to represent an age- and sex-matched group. All participants were screened by an ophthalmologist with subspecialty training in retina and received ophthalmic examinations, including fundus photography, optical coherence tomography (OCT), and fundus fluorescein angiography (FFA). None of the participants had a history of chronic inflammatory or gastrointestinal diseases (including surgery in the gastrointestinal tract) or received systemic antibiotics within the last 6 months before fecal sampling. 

Metagenomic Shotgun Sequencing

Fecal samples were stored in sterile tubes at −80°C at the collection point until DNA extraction. Metagenomic DNA was extracted from fecal samples using the PowerFecal DNA Isolation Kit (MoBio, Jefferson City, MO, USA) according to the manufacturer's instructions and sonicated into 250 to 350 bp fragments using a Bioruptor sonicator (Diagenode, Seraing, Belgium). Sequencing libraries were prepared using the VAHTS Universal DNA Library Prep Kit for Illumina (Vazyme, Nanjing, China) and quantified by qPCR using the KAPA SYBR FAST qPCR Kit (Kapa Biosystems, Wilmington, MA, USA). Paired-end 2 × 125-bp sequencing was performed on a Hiseq2500 instrument (Illumina, San Diego, CA, USA). 

Publicly Available Metagenomic Datasets From the Swiss Population

Gut metagenomic datasets of patients with AMD and control subjects from the Swiss population were obtained from the European Nucleotide Archive (accession numbers: PRJEB13835, PRJEB35615, and PRJEB24557).^11,12 

Taxonomic and Functional Profiling

Raw sequencing reads were quality-filtered using Trimmomatic¹⁸ version 0.36 and PRINSEQ¹⁹ version 0.20.4 and human reads were removed using KneadData version 0.6.1 (https://huttenhower.sph.harvard.edu/kneaddata). Filtered reads were mapped against the RefSeq genome database and metagenome-assembled genome (MAG) database using Kraken2²⁰ version 2.0.9. Taxonomic classification results were filtered using a confidence score of 0.2. The RefSeq database was created by adding fungal genomes to the standard Kraken database including genomes in RefSeq for the bacterial, archaeal, and viral domains. In total, it contains 29,943 microbial genomes, of which 19,362 were bacterial, 368 were archaeal, 9,346 were viral, and 867 were fungal. The custom MAG database contains the 4930 representative genomes of species-level genome bins (SGBs)¹⁷ downloaded from http://segatalab.cibio.unitn.it/data/Pasolli_et_al.html. Reads were mapped to the Gut Virome Database (GVD)¹⁵ viral populations using bowtie2.²¹ CoverM (https://github.com/wwood/CoverM) was used to remove reads that mapped at <95% nucleotide identity to the GVD contigs. Metabolic function profiles were generated using HUMAnN2²² version 0.11.1 with default settings. 

Statistical Analysis

All statistical analysis was performed using R version 4.0.2. The Shannon diversity index was computed using the vegan R package. Principal coordinates analysis (PCoA) was performed using the ade4 R package. For all boxplots, the box edges denote the first and third quartiles and the horizontal line denotes the median, with the whiskers extending up to the 1.5-fold interquartile ranges. We used the default settings (P < 0.05 based on Kruskal-Wallis test and linear discriminant analysis [LDA] effect size > 2) of the linear discriminant analysis effect size (LEfSe) algorithm²³ to identify microbial features (taxa or pathways) with differential abundance between patients with AMD and healthy subjects in our Chinese dataset for basic analyses. We further used multivariable regression models in MaAsLin2²⁴ to identify associations between disease status and abundance of microbial features in the Chinese and Swiss datasets, so as to adjust for potential batch effects of metagenomic data from the two Swiss studies. This MaAsLin2 analysis was performed on the Chinese and Swiss datasets, respectively. We analyzed associations between clinical characteristics and microbial features using multivariable regression models in MaAsLin2. Only microbial features present in >80% of patients with AMD in our cohort and associated with AMD in at least one of the Chinese and Swiss datasets were included in the analysis. Abundance of each microbial feature was modeled with a function of a clinical parameter (one of visual acuity, hemorrhage size, lesion size, angiogram dye leakage, and fovea thickness) as a variable, while adjusting for age, sex, and body mass index (BMI) as covariates. We also used MaAsLin2 to screen the gut microbial differences between the Chinese and Swiss datasets. Abundance of each microbial feature was modeled with a function of country as a variable and disease status as a covariate. 

Results

Dysbiosis of Gut Bacteria in AMD

We performed shotgun metagenomic sequencing on fecal samples collected from 30 patients with neovascular AMD and 30 healthy subjects of the Chinese cohort (Fig. 1A). Taxonomic compositions of the gut microbiota were initially examined by alignment with microbial genomes in the RefSeq database. We observed that bacteria accounted for the majority of sequencing reads (99.49% on average). In addition, we found no significant differences in fungal and viral compositions between patients with AMD and healthy subjects (Supplementary Figs. S1, S2). Therefore, we first focused on the gut microbial dysbiosis involving bacteria in AMD. 

Figure 1.

 

Overview of the study design. (A) Fecal samples from the Chinese cohort composed of 30 patients with AMD and 30 healthy subjects were subjected to shotgun metagenomic sequencing. Taxonomic profiles of the gut microbiota were generated using the RefSeq genome database, the metagenome-assembled genomes database, and the Gut Virome Database (GVD). Functional profiles were reconstructed based on the MetaCyc pathway database. (B) Metagenomic samples of Swiss patients with AMD and healthy subjects from previously published studies were re-analyzed. Datasets from the Chinses and Swiss cohorts were integrated to identify cross-cohort gut microbial signatures associated with AMD. AMD, age-related macular degeneration.

In order to capture full diversity of the gut microbiota, we further conducted taxonomic profiling by mapping to MAGs recapitulated in 4930 bacterial and archaeal species populating the human microbiome.¹⁷ We found that MAGs increased mapping rates of metagenomic reads of samples from both patients with AMD and healthy subjects compared to the RefSeq genomes (P < 0.05; Fig. 2A). Unknown species-level genome bins (uSGBs) accounted for a higher proportion of reads for healthy individuals (average of 45.9%) than patients with AMD (average of 39.2%, P = 0.046; Fig. 2B). We also evaluated the α-diversity of the species profiles generated using the RefSeq and MAG databases, respectively. As expected, mapping to MAGs yielded higher α-diversity than RefSeq genomes (Fig. 2C). For species profiles generated using the RefSeq genome database, the α-diversity showed no significant difference between patients with AMD and healthy subjects (P = 0.208). Nevertheless, the α-diversity of patients with AMD was significantly lower than healthy subjects for species profiles generated using the MAG database (P = 0.024). 

Figure 2.

 

Taxonomic composition of the gut microbiota in patients with AMD and healthy subjects in our dataset. (A) Metagenomic read mapping rates of fecal samples from patients with AMD and healthy subjects using the RefSeq and MAG databases. P values were computed using Wilcoxon signed rank test. (B) The proportions of uSGBs in AMD and healthy samples based on alignment with the MAG database. P value was computed using Wilcoxon rank sum test. (C) The α-diversity measured by the Shannon index of the gut microbiota in patients with AMD and healthy subjects based on taxonomic profiles generated using the RefSeq and MAG databases. P values were computed using Wilcoxon rank sum test. (D) Phylum-level compositions of the gut microbiota in patients with AMD and healthy subjects based on alignment with the RefSeq and MAG databases. (E) Genus enriched or reduced in patients with AMD based on the RefSeq and MAG databases. MAG, metagenome-assembled genome; uSGB, unknown species-level genome bin.

Examination of taxonomic compositions at the phylum-level revealed that Bacteroidetes and Firmicutes were dominant in both patients with AMD and healthy groups (Fig. 2D). Firmicutes/Bacteroidetes ratio was decreased and Synergistetes was depleted in patients with AMD regardless of the databases (P < 0.05). Based on taxonomic profiles generated using MAGs, we further detected the enrichment of Verrucomicrobia as well as the reduction of Actinobacteria and Elusimicrobia in patients with AMD (P < 0.05). Among genera with average abundance >0.1% in the gut microbiota of either patients with AMD or healthy subjects, Veillonella and Lactobacillus were enriched in patients with AMD, whereas Faecalibacterium, Anaerostipes, Blautia, and Eggerthella were reduced in patients with AMD, regardless of the databases (P < 0.05; Fig. 2E). By analyzing taxonomic profiles of MAGs, we further found that Akkermansia was enriched in patients with AMD, whereas Clostridioides, Clostridium, Collinsella, Dorea, Bilophila, and Butyricicoccus were depleted in patients with AMD (P < 0.05). 

We next performed principal coordinates analysis based on the species profiles of the gut microbiota generated using the RefSeq and MAG databases, respectively (Figs. 3A, 3B). These analyses yielded clear delineation between samples from AMD and healthy groups (P < 0.001). We then performed LEfSe analysis to identify species associated with AMD. The analysis using RefSeq species profiles identified the enrichment of 9 species and the depletion of 29 species in patients with AMD (LDA effect size >2). The top five species sorted by LDA effect size for each group are shown in Figure 3C. We found that members of Veillonella and Lactobacillus were overrepresented among species enriched in AMD. Lachnospiraceae including Anaerostipes hadrus, Ruminococcus gnavus, and Anaerobutyricum hallii were depleted in AMD. The analysis using MAG species profiles yielded 31 overabundant species and 68 reduced species (LDA effect size >2), the majority of which belonged to uSGBs. The top five uSGBs sorted by LDA effect size for each group are shown in Figure 3D. We observed that some uSGBs classified at the same family as RefSeq species associated with AMD showed a consistent trend, such as Veillonellaceae (f) uSGB 6925 and Blautia (g) uSGB 4834. However, some uSGBs showed an opposite trend of associations with AMD. For instance, Lachnospiraceae (f) uSGB 4752 was more abundant in patients with AMD than healthy subjects. 

Figure 3.

 

Differences in the gut bacterial profiles between patients with AMD and healthy subjects in our dataset. (A, B) Principal coordinates analysis (PCoA) of the gut microbiota based on the species-level Bray-Curtis distance. Species profiles were generated by matching to the RefSeq database (A) and the MAG database (B). (C) The top five species (sorted by LDA effect sizes) enriched in the patients with AMD and the healthy groups, respectively, according to the RefSeq database. (D) The top five uSGBs (sorted by LDA effect sizes) enriched in the patients with AMD and the healthy groups, respectively, according to the MAG database.

We further investigated whether the severity of AMD was associated with bacterial species among the gut microbiota. To this end, we performed association analysis between the abundance of bacterial species and clinical parameters, including visual acuity (best corrected visual acuity [BCVA], tested using the Early Treatment Diabetic Retinopathy Study [ETDRS] chart), hemorrhage size (area of hemorrhage on fundus image), lesion size (area of lesion on fundus image), angiogram dye leakage (area of lesion on FFA), and fovea thickness (measured by OCT). Among the five parameters, we did not find any bacterial species whose abundance correlated with the variation of AMD visual acuity, whereas lesion size was associated with the greatest number of bacterial species (Supplementary Table S2). Interestingly, approximately 83% (14 out of 17) bacterial species that were found positively associated with fovea thickness were also positively associated with lesion size (Fig. 4). Specifically, the top three bacterial species positively associated with lesion size were Eubacterium (g) uSGB 4348, Dialister invisus, and Citrobacter freundii (see Fig. 4, false discovery rate [FDR] < 0.05). Two species consistently depleted in patients with AMD were negatively associated with the severity of AMD (Supplementary Fig. S3). Phascolarctobacterium (g) uSGB 5790 was negatively associated with lesion size (FDR < 0.1; see Supplementary Fig. S3). Bacteroidaceae (f) uSGB 1825 was negatively associated with hemorrhage size (FDR < 0.2; see Supplementary Fig. S3). 

Figure 4.

 

Associations between clinical characteristics and bacterial species that were enriched in the gut microbiota of patients with AMD. The associations with coefficients > 0.5 and FDR < 0.25 were shown.

Taken together, these observations suggest that AMD is associated with dysbiosis of gut bacteria. Further, the MAG database enabled the identification of numerous uSGBs as important components of bacterial dysbiosis in AMD. 

Signatures of AMD-Associated Gut Bacterial Dysbiosis Across Cohorts

We hypothesized that gut bacteria associated with AMD across populations are more likely to be relevant to disease susceptibility or pathogenesis. To prioritize the microbial features involved in AMD, we re-analyzed the previously published dataset, including metagenomic samples from 69 patients with AMD and 69 healthy subjects in Switzerland and examined common characteristics of gut microbial dysbiosis between the Chinese and Swiss cohorts (see Fig. 1B). The bacterial species associated with AMD according to at least one of the two datasets are shown in Supplementary Table S3 (P < 0.05). As expected, compositions of the gut microbiome were significantly different between the Chinese and Swiss cohorts (Supplementary Fig. S4). Specifically, approximately 75% bacterial species showed differences in abundance between the two cohorts (P < 0.05). Despite the marked difference, we detected three (accounted for 3/58 in Chinese and 3/123 in Swiss cohort) bacterial species concordantly enriched in AMD across cohorts (P < 0.05; Fig. 5A), namely Ruminococcus callidus, Prevotellaceae (f) uSGB 2135, and Lactobacillus gasseri (Figs. 5B–D). Meanwhile, we found six bacterial species depleted in patients with AMD in two cohorts (P < 0.05; Fig. 5E), including Cloacibacillus evryensis, Slackia piriformis, and four uSGBs (Figs. 5F–K). Consistent with findings in the Swiss study,¹²Bacteroides cellulosilyticus showed a decreasing trend in AMD according to our dataset (P = 0.076; Fig. 5L). These results demonstrated that common characteristics of bacterial dysbiosis can be found in the gut microbiota of patients with AMD across cohorts. 

Figure 5.

 

Bacterial species consistently enriched or reduced in patients with AMD across Chinese (CHN) and Swiss (SUI) datasets. (A) Venn diagram showing the number of species enriched in AMD. (B-D) Relative abundance distributions of species enriched in AMD. (E) Venn diagram showing the number of species reduced in AMD. (F-L) Relative abundance distributions of species reduced in AMD. ↑, enriched species in AMD; ↓, reduced species in AMD.

Dysbiosis of Gut Phage in AMD

We next investigated viral compositions of the gut microbiota in patients with AMD and healthy subjects using the GVD. According to the GVD, the vast majority of gut viruses were bacteriophages (98.4% on average), whereas only 0.2% were eukaryotic viruses. Thus, we hypothesized that the α-diversity of viral component of the gut microbiota was correlated with that of bacterial component. To test this hypothesis, we computed the correlation between the α-diversity of taxonomic profiles generated using GVD and MAGs. The results showed that the α-diversity of viral and bacterial components were positively correlated (Fig. 6A). 

Figure 6.

 

Differences in the gut viral compositions between patients with AMD and healthy individuals. (A) Correlation between viral and bacterial α-diversity (as measured by the Shannon index) in the patients with AMD and the healthy groups in our dataset. Viral and bacterial profiles were based on GVD and MAG database matching, respectively. (B) The α-diversity of gut viral profiles based on GVD matching for patients with AMD and healthy subjects in our dataset. P value was computed using Wilcoxon rank sum test. (C) PCoA of viral profiles for the patients with AMD and healthy samples based on Bray-Curtis distance. (D) The GTDB host range of phages showing differences in abundance between patients with AMD and healthy subjects in our dataset. (E) Venn diagram showing the number of phages enriched in AMD for Chinese (CHN) and Swiss (SUI) datasets. (F) Venn diagram showing the number of phages reduced in AMD for CHN and SUI datasets. (G) The GTDB host range of phages enriched in CHN and SUI patients with AMD. (H) The GTDB host range of phages enriched in CHN and SUI healthy subjects. GVD, Gut Virome Database; GTDB, Genome Taxonomy Database. ↑, enriched phages in AMD; ↓, reduced phages in AMD.

We compared viral compositions of gut microbiota between patients with AMD and healthy subjects. Consistent with our previous finding that the bacterial α-diversity was decreased in AMD, the viral α-diversity was slightly lower in patients with AMD than healthy subjects (P = 0.061; Fig. 6B). We performed principal coordinates analysis based on viral profiles generated using GVD (Fig. 6C). The result showed that the AMD group was separated from the healthy group, suggesting that dysbiosis of gut viruses is associated with AMD. 

Subsequently, we performed LEfSe analysis to identified gut viruses that showed differences in abundance between patients with AMD and healthy subjects. We found no significant difference in eukaryotic viruses. Thus, we focused on the characteristics of phage dysbiosis. In total, we detected 50 phages enriched and 62 phages depleted in AMD (LDA effect size >2). We examined hosts of these phages based on the Genome Taxonomy Database (GTDB)²⁵ assignment provided by the GVD (Fig. 6D). Among phages whose hosts were assigned, those targeting Bacteroidaceae ranked the first in numbers, followed by phages targeting Selenomonadaceae, Ruminococcaceae, and Lachnospiraceae. 

We further integrated the Swiss dataset to identify cross-cohort gut phage signatures of AMD. The phages associated with AMD according to at least one of the two datasets are shown in Supplementary Table S4. A total of 12 phages were enriched in AMD (P < 0.05; Fig. 6E) and 41 phages were reduced (P < 0.05; Fig. 6F) in both datasets. Of these, Bacteroidaceae was the most frequently identified host both for phages with increased and decreased abundance in AMD across cohorts (Figs. 6G, 6H). Interestingly, gut phages that were negatively associated with lesion area were dominated by those targeting Bacteroidaceae (Supplementary Table S5). Other assigned hosts of phages consistently enriched in AMD included Veillonellaceae, Akkermansiaceae, Lactobacillaceae, Ruminococcaceae, and Acutalibacteraceae (see Fig. 6G). Phages targeting Rikenellaceae, Lachnospiraceae, and Coriobacteriaceae were consistently reduced in patients with AMD according to the two datasets (see Fig. 6H). Collectively, these results demonstrated that phage dysbiosis in the gut microbiota may be associated with AMD. However, it is unclear which aspects of phage dysbiosis are causes or consequences of bacterial dysbiosis in AMD. 

Functional Alterations of the Gut Microbiome in AMD

To describe differences in metabolic potential of the gut microbial community between patients with AMD and healthy subjects, we performed functional profiling using HUMAnN2. Principal coordinates analysis based on abundances of the reconstructed MetaCyc pathway showed that samples of the patients with AMD and healthy groups can be separated by differences in functional profiles (Fig. 7A). We identified 15 metabolic pathways enriched in AMD and 45 metabolic pathways reduced (LDA effect size >2). The top five pathways sorted by LDA effect size for each group are shown in Figure 7B. We found that pathways in nucleic acid processing were enhanced in patients with AMD, including queuosine biosynthesis and preQ0 biosynthesis. Pathways in enzyme cofactor and carrier biosynthesis including N10-formyl-tetrahydrofolate biosynthesis and pyridoxal 5'-phosphate biosynthesis were also enriched in AMD. Conversely, pathways in amino acid biosynthesis were reduced, including the superpathway of branched amino acid biosynthesis and L-isoleucine biosynthesis III. A total of 24 pathways were positively associated with lesion size in patients with AMD (Supplementary Table S6; FDR < 0.2). Finally, we examined common metabolic characteristics associated with AMD in the gut microbiome across the Chinese and Swiss cohorts. The pathways associated with AMD according to at least one of the two datasets are listed in Supplementary Table S7. We did not identify metabolic pathways that were consistently enriched or reduced in patients with AMD across cohorts (P > 0.05). Nevertheless, three degradation pathways reduced in AMD patients of Chinese cohort (P < 0.05) also showed a decreasing trend in Swiss cohort, including the superpathway of N-acetylneuraminate degradation, glycerol degradation to butanol, and glycogen degradation I (Fig. 7C). 

Figure 7.

 

Functional differences in gut microbiota between patients with AMD and healthy subjects. (A) PCoA of samples from patients with AMD and healthy subjects in our dataset based on the pathway-level Bray-Curtis distance. (B) The top five pathways (sorted by LDA effect sizes) enriched and depleted, respectively, in patients with AMD of our dataset. (C) Relative abundance distributions of pathways showing differences in abundance between patients with AMD and healthy subjects for both Chinese (CHN) and Swiss (SUI) datasets.

Discussion

To have a better understanding of the gut microbiome alterations in AMD, we characterized compositions of the gut microbiome in patients with AMD and healthy subjects using shotgun metagenomic sequencing. Importantly, considering the influence of ethnicity, geography, and diet difference is crucial when reaching to the conclusion of association between disease phenotype and gut microbiome composition. We therefore integrated the metagenomic datasets of from both Chinese and Swiss cohorts to identify shared gut microbial signatures of AMD.^11,12 Our analysis revealed the cohort-independent, therefore ethnicity-, geography-, and diet-independent associations between AMD and multiple bacterial species, viral populations (mainly bacteriophages) as well as metabolic pathways (see Figs. 5–7). 

We comprehensively surveyed taxonomic signatures of the gut microbiota in AMD by mapping metagenomic reads to multiple databases, ranging from traditional genomic references to new genomic resources provided by recent studies in large-scale metagenomic assembly.^15–17 Mapping to newly discovered genomic sequences enables us to capture the full spectrum of the gut microbial diversity. Notably, we observed a decrease in the α-diversity of the gut bacteria and phages in patients with AMD using MAGs and GVD as references, respectively. In contrast, the α-diversity measured on the species profiles generated using the RefSeq database did not exhibit a significant difference between the AMD and healthy groups. Although we found no significant alterations in the gut mycobiome of patients with AMD, we believe that future efforts to reconstruct new fungal genomes via large-scale metagenomic assembly will facilitate the examination of possible involvement of gut fungal dysbiosis in AMD. 

We reported for the first time the associations between AMD and gut phage dysbiosis. Previous studies suggest that gut phages contribute to the transition from health to disease by causing an imbalance between symbiotic bacteria and pathobionts.²⁶ This outcome can be achieved, for example, by selectively killing the dominant commensal gut bacteria or by reducing the number of commensal bacteria and allowing deleterious competitors to dominant the gut environment.^27,28 Whether dysbiosis of gut phages may trigger inflammatory reactions and promote AMD directly or by shifting the balance of gut bacteria warrants further investigation. 

In the gut microbiota of patients with AMD, we observed a reduced Firmicutes to Bacteroidetes ratio, which was reported in some immune-mediated diseases, including dry eye syndrome and systemic lupus erythematosus.^29,30 This observation implies that an immune imbalance caused by gut dysbiosis contributes to AMD onset or progression. Specifically, we detected the enrichment of Veillonella and Lactobacillus in patients with AMD. Veillonella spp. have been implicated in several severe inflammatory conditions, including recurrent Crohn's disease, osteomyelitis, and endocarditis,^31–33 and a recent study reported an increased abundance of gut Veillonella that was positively correlated with the status of autoimmune hepatitis.³⁴ Lactobacillus is regarded as an opportunistic pathogen emanating from the gut³⁵ and has been reported to enhance nonspecific cellular immune responses.³⁶ More importantly, we identified the enrichment of several bacterial species in patients with AMD across the Chinese and Swiss datasets. These species may represent prioritized targets for future investigations in AMD pathogenesis. 

Meanwhile, we identified the reduction of several bacterial species in the gut microbiota of patients with AMD in both the Chinese and Swiss datasets. These species may serve as promising candidates of probiotics that apply to patients with AMD or individuals at a high risk of AMD across populations. Alterations of the gut microbiota after probiotic treatment have been shown to alleviate ocular diseases, including autoimmune dry eye and uveitis.^37,38 On the other hand, these results imply that narrow spectrum antibiotics may be safer alternatives of broad spectrum antibiotics by reducing collateral damage to the host gut microbiome. 

We found that composition of the gut microbiota is associated with AMD severity. The previous gut microbiota study based on mouse models revealed that Clostridiales and Firmicutes were positively associated with retinal damage, whereas protection from AMD features was associated with Bacteroidales.⁹ Our analysis reproduced the associations between these taxa and AMD severity in human. Specifically, we identified a list of Clostridiales and Firmicutes uSGBs that were positively associated with lesion size, fovea thickness, hemorrhage size, or angiogram dye leakage (see Fig. 4). In addition, a few Bacteroidales species were negatively associated with lesion or hemorrhage size, including B. salyersiae, B. finegoldii, B. dorei, B. caccae, and four Bacteroidales uSGBs (see Supplementary Fig. S3). The gut microbiome may affect AMD severity through multiple mechanisms. Gut dysbiosis can lead to an increase in intestinal permeability, thereby facilitating the translocation of microbes. Toll-like receptor 7 dependent translocation of Lactobacillus outside the gut has been implicated in the exacerbation of systemic lupus erythematosus by increasing plasmacytoid dendritic cell and interferon signaling.³⁹ Likewise, gut pathobionts can breach the gut mucosal barrier, reach the eye through the systemic circulation, and induce local inflammatory response in which angiogenic factor VEGF-A may be involved.⁷ Interestingly, we previously reported the presence of intraocular microbiota in human patients and animal models with a disease-specific profile.⁴⁰ Other mechanisms may also link gut microbial dysbiosis to ocular inflammation and AMD. For example, gut microbial peptides may act as antigens to induce host immune responses leading to ocular inflammation. In support of this notion, a microbiota-dependent activation signal in the gut was shown to activate retina-specific T cells and intraocular inflammation in a mouse model of spontaneous uveitis.⁴¹ Additionally, metabolites from the abnormal gut microbiota may dysregulate host energy metabolism and induce chronic, low-grade inflammation that increases disease susceptibility.⁴² As for protective role of gut commensals in AMD, it has been reported that gut abundance of Bacteroides was negatively correlated with complement factor H.¹² Thus, it is worth pursuing whether Bacteroides species have a protective effect on the development of AMD via suppressing the complement system. 

Our study only analyzed the association between wet AMD and gut microbiome composition in three patient cohorts from China and Switzerland, with a limited sample size. Therefore, a larger patient cohort including multiple additional ethnic and geographic groups may help to reveal the true microbial factors contributing to AMD etiology. Differences in the gut microbiota between the Chinese and Swiss cohorts can be partially driven by vastly different diets in the two populations. A limitation of this study is that diet was not controlled. Future studies incorporating dietary data are warranted to interpretate the gut microbial variation across patients with AMD and to disentangle interactions between diet, gut microbiome, and AMD. Furthermore, the complex analysis of clinical features and therapeutic interventions of AMD will help to elucidate the role of the gut microbial dysbiosis in the susceptibility, onset, and progression of AMD. 

In conclusion, our study characterized the dysbiosis of the gut microbiota in patients with AMD and identified cross-cohort gut microbial signatures of AMD. This work potentially provides microbiome-based targets, such as prebiotics, probiotics, and limited spectrum antibiotics, for AMD prevention and treatment. 

Acknowledgments

Supported by the National Key R&D Program of China (2021YFA1101204) to F.Z., X. Wen, L.W.; the National Key R&D Program of China (2019YFA0111200) and the National Natural Science Foundation of China 81700803 to X. Wen; and the National Natural Science Foundation of China 82171041 to L.W. 

Disclosure: W. Xue, None; P. Peng, None; X. Wen, None; H. Meng, None; Y. Qin, None; T. Deng, None; S. Guo, None; T. Chen, None, X. Li, None; J. Liang, None; F. Zhang, None; Z. Xie, None; M. Jin, Smilebiotek Ltd. (C); Q. Liang, None; L. Wei, None