Associations of the intestinal microbiome with the complement system in age-related macular degeneration

Background: Age-related macular degeneration (AMD) is a leading cause of severe vision loss in the aged population. The etiology of AMD is multifactorial and includes nutritional factors, genetic variants mainly in the complement pathway, environmental risk factors and alterations in the intestinal microbiome. However, it remains largely unexplored whether there is an interdependency of these factors leading to the development of AMD. To investigate this issue, a comprehensive shotgun metagenomics analysis of 57 AMD and 58 healthy controls as well as of 16 complement C3 deficient mice and 16 wildtypes was performed. Single nucleotide polymorphisms (SNPs) in the complement factors were assessed with pre-designed TaqMan® SNP genotyping assays. Results: The composition of the intestinal microbiome differed significantly between AMD patients and controls. Whereas the class Negativicutes was more abundant in patients, the genus Oscillibacter and Bacteroides species had a significantly higher prevalence in persons without AMD. While SNPs within the complement factor B gene were more abundant in controls, SNPs within the high temperature requirement A serine peptidase 1 and complement factor H (CFH) genes were associated with AMD. Using a classification model, Negativicutes was identified as a potential biomarker for AMD and furthermore, it positively correlated with CFH. In addition, similar taxonomic features were identified that distinguished wildtype mice from C3 deficient mice. Conclusion: The composition of the intestinal microbiome differs between AMD patients and controls as well as between C3 deficient mice and wildtype mice. Moreover, since the phylum Firmicutes has been identified as potential biomarker for AMD and positively correlates with the genetic risk factor CFH, the study suggests an association between the intestinal microbiome and the complement system in AMD.


Introduction
Age-related macular degeneration (AMD) is the most frequent cause of blindness among older people in developed countries [1]. AMD has become a significant economic and social burden on public health [1] and its prevalence is projected to grow by 50 percent in the next decades [2]. The early stage of AMD results in accumulation of extracellular material underneath the retina, called drusen. Drusen are a sign of impaired retinal pigment epithelium (RPE) function and disruption of the metabolic transport between RPE and choroid [3]. Whereas the intermediate stage of AMD is characterized by larger drusen, the late stages are characterized by either choroidal neovascularization (CNV) in the exudative form or geographic atrophy in the atrophic form of AMD [4]. The pathogenesis of AMD is multifactorial and is thought to be a combination of oxidative stress, impaired RPE function, increased apoptosis and aberrant immune system activation [5].
Despite major research efforts, treatment options for early and intermediate AMD are only very limited. Smoking is the strongest modifiable risk factor for AMD, resulting in oxidative stress, ischemia, hypoxia and ultimately development of CNV. Other environmental risk factors are sunlight exposure [6,7] and obesity [8]. To date, the only factor that has been shown to be protective is a healthy diet, rich in omega − 3 fatty acids, lutein, zeaxanthin and antioxidants, consistent with the Age-Related Eye Disease Study 2 (AREDS2) formulation [9]. Therefore, preventative strategies targeting nutritional intake to avoid the development of AMD seem to be most promising [9][10][11][12][13][14]. Recently we have shown that the composition of the intestinal microbiome may impact AMD development and progression [15]. This link has recently been verified in an animal model, where modifications in the intestinal microbiome have resulted in exacerbation of CNV [16].
In addition to the effect of modifiable environmental risk factors, genetic variants may be associated with AMD such as polymorphisms in complement factor H (CFH) [17]. The discovery of genetic variants in components of the complement system indicated the potential role of local inflammation and complement regulation in the pathogenesis of AMD. Moreover, since toll-like receptors (TLRs) recognize microbe-specific molecules and are expressed on various cells along the gastrointestinal tract, they may be crucial in signaling between the immune system and the microbiota. Moreover, the complement system is a key part of the innate immune system operating via TLR signaling. When overactivated or de-regulated, the complement becomes a major link between infection and inflammation with profound involvement in inflammatory and degenerative diseases such as AMD [18]. In this study, we investigated gut microbiome alterations in patients with AMD and screened for associations between the intestinal microbiome and the complement system.

Results
Taxonomic characterization of the intestinal microbiome in AMD patients and healthy controls In total, 115 stool samples were sequenced from 57 AMD patients and 58 age-and sexmatched healthy controls. The AMD patients and controls were similar for gender, age at collection, smoking habit and BMI (p > 0.05, Table 1). A total of 2.8 billion 100 bp pairedend reads with an average insert size of 350 bp were generated, with an average of 24.3 ± 9.3 (s.d.) million reads per sample. After trimming and filtering, we kept about 2.5 billion non-human high-quality reads, with an average of 21.5 ± 8.5 (s.d.) million reads per sample. The majority of mapped reads were bacterial with 98.8 ± 8.1% (s.d.). The phyla Firmicutes and Bacteroidetes dominated the microbiome composition (Fig. 1).
Bacteroidia and Clostridia were the most abundant classes, and Bacteroides, Alistipes and Subdoligranulum the most abundant genera in the cohort, consistent with previous observations [19,20].  [19]. To group the AMD patients and control samples into enterotype clusters, we applied the PAM algorithm using Jensen-Shannon distance for relative genus abundances (Fig. 2). The optimal number of enterotypes was three as suggested by the CH index (Fig. 2b). Between-class-analysis was applied to cluster the samples into three enterotypes. (Fig. 2a). The enterotypes were characterized by the following contributors at genus level: Enterotype 1 had a relatively high level of Bacteroides, enterotype 2 of Prevotella and Escherichia was indicative for enterotype 3 (Fig. 2c). However, no association between the enterotypes and the disease status was found (p > 0.05, Fisher's exact test), showing an uniform distribution of the samples across the three enterotypes.
A principal component analysis with the health status as grouping variable showed that differences in microbial species abundances separated the patient group from the control group (p = 0.001, PERMANOVA analysis with nrepet = 10'000, Fig. 1c). To further examine features of the intestinal microbiome in AMD patients, we compared the relative abundances of taxa between patients and controls, showing that the class Negativicutes was more abundant in patients (p = 0.0016) and the genus Oscillibacter (p = 0.035) and Bacteroides species (p = 0.0035, Wilcoxon rank sum test) were more abundant in controls ( Fig. 1D).

Identification of AMD patients based on the intestinal microbiome
To illustrate the potential diagnostic value of the intestinal microbiome for AMD, a species-based classifier was used in an attempt to predict AMD samples among a mixture of samples from patients and healthy controls. Based on the relative species abundance profile from MetaPhlAn, a classification model was constructed using linear shrinkage discriminant analysis and features were ranked according to their correlation adjusted t Association between single nucleotide polymorphisms and AMD Out of the genetic polymorphisms previously shown to be associated with risk for AMD [21][22][23][24][25][26][27][28][29][30][31][32][33][34], single nucleotide polymorphisms rs800292 (p = 0.0012), rs1410996 (p = 0.00019) and rs1061170 (p = 0.000064) within the complement factor H (CFH) gene and rs11200638 (p = 0.00012) within the high temperature requirement A serine peptidase 1 (HTRA1) gene were associated with a higher risk for having AMD in our cohort. However, rs12614 (p = 0.0057) within the complement factor B (CFB) gene was inversely correlated with AMD in the cohort (Fisher's exact test, Table 2).  A principal component analysis showed that differences in microbial species abundances separated the C3 − /− mice from the wildtypes (p = 0.001, PERMANOVA analysis with nrepet = 10000, Fig. 4C). To further examine features of the intestinal microbiome in C3 − /− mice, we compared the relative abundances of taxa between C3 deficient mice and wildtypes, showing that the phylum Firmicutes was more abundant in C3 − /− mice (p < 0.0001) and the phylum Bacteroidetes was more abundant in wildtypes (p < 0.0001; Fig. 4D). There was no association of sex or week at collection with the intestinal microbiome in the cohort (q > 0.02). Bacteroides species were more abundant in healthy controls relative to AMD patients. In contrast, the class Negativicutes was more abundant in AMD patients compared to controls. These results are consistent with previous studies, reporting associations between several of these species and AMD, [15,36].

Discussion
After smoking, obesity is the second most important environmental risk factor for AMD [37]. Numerous studies have shown that a higher relative abundance of Firmicutes and a lower Bacteroidetes abundance are typically associated with obesity [38]. In a previous study, we observed a shift of relative abundance in Firmicutes at the expense of Bacteroidetes in AMD patients [15]. In this study, the class Negativicutes belonging to the phylum Firmicutes was enriched in AMD patients. Moreover, Bacteroides species belonging to the phylum Bacteroidetes was more abundant in controls. Looking at the prediction model in Fig. 3A, Negativicutes as well as the order Selenomonadales, both belonging to the phylum Firmicutes, may be identified as potential biomarkers for AMD. On the other hand, Bacteroides species, especially including B. cellulosilyticus and B. caccae, may have a protective effect on the development of AMD as these species were more abundant in the control group. Taken together, these results suggest an elevated Firmicutes to Bacteroidetes ratio, also at lower taxonomic levels in AMD patients compared to healthy controls. These data are also consistent with observations in mice, demonstrating that high-fat diets exacerbate choroidal neovascularization, which is a hallmark of exudative AMD, by increasing the relative abundance of Firmicutes [ 36].
In addition to environmental risk factors, genetic risk factors mostly associated with the complement system are involved in the development of AMD. The first genome-wide association study for AMD performed in 2005 identified a highly associated genetic variant in complement factor H (CFH) [39]. This genetic link between AMD and the complement system was further expanded by several studies showing an association between other complement factors such as complement component 3 (C3), complement component C2 (C2) and complement factor B (CFB) [40,41]. In our cohort, while single nucleotide polymorphisms (SNPs) within the CFH gene are associated with a higher risk for developing AMD, SNP within the CFB gene has a protective effect (Table 2). Furthermore, to assess associations between the complement system and the intestinal microbiome, both involved into the development of AMD, microbiota of C3 deficient mice and wildtypes was analyzed in terms of taxonomic profile and correlated to clinical metadata. While the phylum Firmicutes was more abundant, the phyla Bacteroidetes was decreased in C3 deficient mice compared to wildtypes (Fig. 4). Since Firmicutes was proposed as potential biomarker for AMD, it may also contribute to the development of AMD in C3 deficient mice.
These results show that the complement alters the intestinal microbiota composition in mice and as such microbiota-derived nutrients, metabolites and antigens within the host.
This in turn may promote degenerative diseases such as AMD. This association between the complement system and the intestinal microbiome was further confirmed in humans by the correlation of SNPs within the CFH gene with the taxa Negativicutes, Clostridiales, Bacteroides species, and Ruminococcus torques (Fig. 3C). Complement factor H is the soluble inhibitor of the alternative pathway of complement, mediating anti-inflammatory housekeeping functions by protecting cells from complement activation. Thus, mutations in the CFH gene result in uncontrolled complement activation, which is involved in the pathogenesis of AMD [42]. Since Negativicutes, which was proposed as potential biomarker for AMD (Fig. 3A), positively correlated with CFH, it may influence the development and/or progression of AMD via the regulation of the alternative pathway of complement. Moreover, while the relative abundance of Bacteroides species were proposed as potential biomarker for controls and negatively correlated with CFH, they may be protective for AMD.
Limitations of this study include the lack of longitudinal data and the limited number of patients.

Conclusions
In conclusion, we found alterations of the intestinal microbiome in AMD patients. While the phylum Firmicutes was more abundant in patients with AMD, Bacteroides species may be protective for AMD, both correlating with the genetic risk factor CFH. This data points towards a possible interconnection between the intestinal microbiome, the complement system and the development of AMD. However, the data do not allow to determine whether the altered intestinal microbiome is the consequence of the disease or is involved in its pathogenesis via uncontrolled activation of the complement system. Switzerland). To exclude low-quality reads and reads mapping to human DNA, the resulting 150 bp paired-end reads were quality filtered with Trimmomatic v.0.32 [43] and mapped to the human reference genome hg19 using Bowtie2 v.2.2.4 [44]. For taxonomical analysis, the high-quality non-human reads were mapped against a set of clade-specific marker sequences using the Metagenomic Phylogenetic Analysis tool v.2.6.0 (MetaPhlAn2) and the marker database v.20 [45] using default settings. In order to provide the relative abundance of each taxonomic unit, Bowtie2 v.2.2.4 was applied for alignment followed by normalization of the total number of reads in each clade by the nucleotide length of its marker.

Methods
Genetic analysis of germline DNA Genomic DNA was extracted from EDTA blood samples using a QIAamp DNA Blood Midi Kit (Qiagen AG, Basel, Switzerland). To identify genetic risk factors for AMD, 16 previously described SNPs were analyzed (supplementary Table 1). Real-time quantitative PCR (see below) was used for allelic discrimination of all SNPs analysed except for CFB (rs4151667,   (Fig. 3B).
The R package sda provided the functionality for high-dimensional linear discriminant analysis with feature selection. A species-based classifier using the relative species abundance profile from MetaPhlAn was trained using Stein-type shrinkage estimators and features are ranked according to correlation-adjusted t (cat) scores to provide a set of candidate biomarkers (Fig. 3A). The minimum error was calculated by 100-fold crossvalidation by the "rfcv" function of the randomForest package with sequentially reduced numbers of features. The optimal number of features was selected by cross-validation.
The model was tested on the testing set using this set of features and a receiver operating (ROC) curve within the pROC package in R.

Mice
Adult (6-8 weeks of age) C3 − /− mice on a C57BL/6 background (B6;129S4-C3 tm 1Crr /J, n = 16) and C57BL/6 mice as controls (n = 16), both from Jackson Laboratory (Bar Harbor, ME, USA), were used for this study. Mice were housed in groups of 4 animals of the same gender under temperature and humidity-controlled conditions in individually ventilated cages giving sterile food and water ad libitum and exposed to 12:12 h light:dark cycles.
One and 4 weeks after arrival of the mice, faeces were collected and immediately frozen at -20 °C. At the end of the study, mice were euthanized by carbon dioxide (CO 2 ) inhalation. Metagenomic DNA was isolated and sequenced as described above. To exclude low-quality reads and reads mapping to murine DNA, the resulting 150 bp paired-end reads were quality filtered with Trimmomatic v.0.32 [43] and mapped to the mouse reference genome GRCm38 using Bowtie2 v.2.3.0 [44]. The high-quality non-murine reads were taxonomically profiled using MetaPhlAn2, and associations of microbial abundances with genetics (C3 − /− versus wildtype mice) and demographic parameters (sex and week at collection) were identified by MaAsLin (for details see above). All animal studies were conducted at the University of Bern and approved by the local Animal Ethics Committee (Veterinärdienst des Kantons Bern / BE99-16).

Ethics approval and consent to participate
This study was approved by the Ethics Committee of the Canton of Bern (ClinicalTrials.gov: NCT02438111) and all participants gave written informed consent to participate in the study.
All animal studies were approved by the local Animal Ethics Committee (Veterinärdienst des Kantons Bern / BE99-16).

Consent for publication
Not applicable.

Availability of data and material
The Diversity of the intestinal microbiome in AMD patients and healthy controls.
Relative abundances of microbiota at phyla level in all study subjects (a) and averaged for study groups (b). c: Principal component analysis of microbial species abundance grouped patients and controls separately, with PERMANOVA confirming a significant difference between the groups (p = 0.0018). d: Relative abundances of taxa associated with AMD (Wilcoxon rank sum test, p < 0.05).

Figure 2
Enterotypes of the intestinal microbiome in AMD patients and healthy controls. a: