Ultra-rare complement factor 8 coding variants in families with age-related macular degeneration

Summary Genome-wide association studies have uncovered 52 independent common and rare variants across 34 genetic loci, which influence susceptibility to age related macular degeneration (AMD). Of the 5 AMD-associated complement genes, complement factor H (CFH) and CFI exhibit a significant rare variant burden implicating a major contribution of the complement pathway to disease pathology. However, the efforts for developing AMD therapy have been challenging as of yet. Here, we report the identification of ultra-rare variants in complement factors 8A and 8B, two components of the terminal complement membrane attack complex (MAC), by whole exome sequencing of a cohort of AMD families. The identified C8 variants impact local interactions among proteins of C8 triplex in vitro, indicating their effect on MAC stability. Our results suggest that MAC, and not the early steps of the complement pathway, might be a more effective target for designing treatments for AMD.


INTRODUCTION
Age-related macular degeneration (AMD) is a late-onset neurodegenerative disease, which is a major cause of vision impairment in elderly population and constitutes 6-9% of legal blindness globally. 1 Prevalence of AMD is estimated to be over 280 million within the next 20 years 2 presenting a significant socioeconomic burden on individuals and society at large. Advanced age, environmental factors, lifestyle, and genetic predisposition contribute significantly to disease pathologies, which include disruption of retinal pigment epithelium (RPE) function, formation of large drusen and progressive degeneration of macular and perimacular photoreceptors. [3][4][5] Advanced AMD is classified as wet (because of subretinal neovascularization, termed neovascular AMD or nAMD) or dry (regional geographic atrophy (GA) with no neovascularization). 1 Early familial clustering and linkage studies suggested multiple genomic regions of potential AMD susceptibility. 5 Later, a large genome-wide association study (GWAS) uncovered 52 common and rare variants at 34 genomic loci that help explain over half of the AMD heritability, 6 with the complement system emerging as a major contributor, 5,7,8 together with chronic inflammation and disruption of the extracellular matrix. However, pathophysiological mechanisms of AMD progression remain unclear hindering effective diagnosis and development of therapies.
Population and GWAS-based strategies have been widely successful in identifying risk factors for multiple neurological conditions, but for the most part these risk alleles are present in non-coding or inter-genic parts of the genome making the interpretation of their function and association to specific genes or molecular mechanisms extremely difficult. In search of significant missing heritability and rare variants to target specific genes, we previously analyzed families with advanced AMD and suggested association in 13 additional genes. 9 In this study we leverage the advantages of familial analysis to identify ultra-rare variants in complement Factor 8A (C8A) and 8B (C8B) genes, which segregate with advanced AMD in 4 unrelated families. Complement factor 8 is part of the terminal step of the complement cascade, forming the membrane attack complex (MAC). Our results allow for a better understanding of the genetic factors contributing to disease development and propose a plausible and unifying model of MAC-associated AMD pathology.
Four families with multiple members diagnosed with advanced AMD (multigenerational manifestation of AMD in 3 families) were recruited for the study (Figure 1, See also Data S1- Table S1). A modified AREDS phenotype was assigned to 37 participants based on fundus photographs and relevant medical history (see STAR Methods for details) out of 39 enrolled in the study (age 52-89 years; Data S1 - Table S1). Diagnosis of advanced AMD was made when either GA, nAMD, or both were present in at least one eye. Inheritance patterns within the families suggest an autosomal dominant mode of inheritance, indicating a genetic variant with a large effect, which prompted us to perform whole exome sequencing (WES) on 13 affected and 4 unaffected individuals from these families to identify potential disease-associated variant/s. The initial screen of all previously reported AMD genes revealed no rare or pathogenic segregating variants. Further analysis uncovered two putative causal variants (See also Table S2); one in C8A (c.G1331A, p.R444H, ClinVar: rs143908758 in one family), and the other in C8B (c.G1144T, p.D382Y, ClinVar: rs139498867, in three families). We validated WES findings by performing Sanger sequencing of 21 additional family members ( Figure 2). The identified missense variants are observed at extremely low frequencies in public databases (Global MAFs of 0.00140 and 0.00280, respectively, https://www.ncbi. nlm.nih.gov/clinvar/), are predicted to be pathogenic by SIFT, PolyPhen, and CADD (See also Table S2, and Figure S1) and alter either conserved or semi-conserved residues in the protein ( Figure 2). Two common AMD risk variants are known to have a large genetic effect (ClinVar: rs1061170 in CFH and ClinVar: rs10490924 in ARMS2), but neither segregated well with the disease in the four families ( Figure 1). In Family 37, individual II-7 did not carry either of the common risk variants. Notably, affected individuals homozygous for the CFH rs1061170 risk allele in addition to a rare C8 variant were more likely to develop neovascularization. This co-occurrence was observed in five out of the six CFH homozygous patients, corresponding to a rate of 83% (4/6 nAMD versus 1/6 GA versus 1/6 both nAMD+GA; See also Table S1). Although the low number of individuals does not allow for statistical analysis, these data suggest that a cumulative burden in the complement cascade can drive nAMD development, consistent with the previously proposed multi-hit hypothesis. 5

C8 variants are prevalent in AREDS1 and AREDS2 cohorts
To investigate the role of C8 variants on a population level, we interrogated the available sequencing data from the two large AMD cohorts that participated in the AREDS 1 and 2 studies. 10,11 Our analysis of the genotyping data from AREDS1/2 AMD cohorts revealed that while the overall frequencies of these variants remained low, these were significantly higher compared to those reported in public databases ( Figure 2). Our data indicate that these ultra-rare C8 variants could be contributing to genetic risk even in non-familial AMD cases.
Complement factor 8 is part of the membrane attack complex and is expressed by the RPE C8 is a protein triplex comprising two main subunits (a and b), and a smaller g subunit which is not essential for MAC formation. The variants we identified are located in the main subunits and could impact both interactions with C8g as well as other components of the MAC (Figure 3). There is an ongoing debate as to the primary location/cells that initiate the disease process; nonetheless, it is well established that RPE function is significantly affected in AMD. Therefore, we first confirmed the presence of C8 subunits, with C7 and C9 (known interacting partners in the formation of the membrane attack complex) in human donor RPE, a key target cell type in AMD (Figure 3).

In-silico analysis and modeling predict that C8 variants affect protein interaction
We then constructed a homology model of the C8 heterotrimeric complex (STAR Methods, Data S2) and introduced the variants in the a and b subunits to evaluate their potential impact on local interactions (Figure 3; See also Figure S2). R444H variant in C8a is predicted to result in the addition of a hydrogen bond that would interact with a residue on C8b ( Figure 3) and perturb another residue reported to interact with C8g (PDB: 3OJY, https://www.ebi.ac.uk/pdbe/entry/pdb/3ojy/protein/1). D382Y variant in C8b is anticipated to disrupt local structure with the loss and addition of several H-bonds ( Figure 3). Protein stability calculations predicted that both variants have non-native stable conformations (See also Table S3), with C8b D382Y indicating a more pronounced stability effect. Thus, the two C8a and b variants not only show familial segregation but are also predicted to impact the formation, stability, or function of the heterotrimeric C8 complex. In-vitro analysis of the C8 triplex confirms the potential impact of variants on protein interaction To investigate the effect of identified variants on the assembly of the C8 triplex, we co-transfected all three C8 subunits, with and without variants, in HEK293 cells ( Figure S2). All constructs could form a C8 complex, Figure 1. Rare C8 variants segregate in familial AMD cases Pedigree structures of four AMD families, with symbols indicating sex and disease phenotype, and genotypes of rare C8A or C8B variants as well as two common large effect AMD risk alleles noted below each symbol (see keys at bottom of the figure). To the right of each pedigree, fundoscopy images of select individuals with advanced AMD are shown. aAMDadvanced age-related macular degeneration; GA -geographic atrophy; nAMD-neovascular age-related macular degeneration.

OPEN ACCESS
iScience 26, 106417, April 21, 2023 3 iScience Article as indicated by co-immunoprecipitation (Co-IP) experiments ( Figure 3C). As predicted by stability modeling, the introduction of the C8 variants increased the immunoprecipitation of C8 subunits with C8a or C8b antibody ( Figure 3D). Notably, the C8a variant displayed a higher affinity to C8G as shown by the higher amount in co-immunoprecipitation experiment ( Figure 3E), confirming previous reports of C8A R444 facilitating the interaction with C8G.

DISCUSSION
GWAS using common variants have identified genetic loci of interest for many complex traits, but rare variants in familial cases can help ascertain causal genes and provide clues to molecular mechanisms. Clustering of AMD in familial structure, mimicking an autosomal dominant mode of inheritance, is strongly indicative of genetic variants having a large effect size. In concordance, we uncovered ultra-rare missense variants in C8A and C8B genes segregating in 4 families, and no other reported AMD-associated common or rare variant 6,9 showed segregation with AMD phenotype. Furthermore, the rare variants in both a or b C8 subunits could impact C8 heterotrimer formation and/or stability, likely altering MAC-mediated immune response. In addition, a recent study identified C8G as an inhibitor of neuroinflammation, 12 our results show that C8 variants could affect the C8G binding, and thus potentially prevent its role as an inhibitor and exacerbate the inflammatory process in AMD.
Three C8 subunits (a, b, and g encoded by C8A, C8B, and C8G genes, respectively) form the terminal MAC together with C5b, C6, C7, and C9. 13 A rare variant in C9, previously implicated in AMD, is shown to iScience Article enhance MAC polymerization. 14 Nonsense mutations in C8 can cause C8 deficiency, 15 but their impact on vision has not been assessed.
The three complement pathways (classical, lectin and alternative) converge on C5b-guided sequential assembly of C6-C9 resulting in the formation of MAC, and deposition or activation of MAC can initiate divergent signaling pathways triggering an immune response and cellular defense mechanisms. 7,16 Together with the reported rare variant in C9, 14 our study further strengthens the hypothesis of dysregulated terminal MAC as a key contributor to AMD pathology and suggests that rare variants in C8 genes could also affect the complex formation or its stability.
Molecular diagnosis is not a part of routine AMD screening, and aside from the two strongly associated common genetic variants (ClinVar: rs1061170 in CFH and ClinVar: rs10490924 in ARMS2/HTRA1), patients are generally not tested for any of the other common or rare variants that are reported in the literature.
Most clinical studies have also focused on the correlation of only these two variants with divergent disease phenotypes. Notably, common variants in distinct complement genes have been linked to both protective Quantification of C8 a, b, and g subunits is indicated in green, teal and magenta, respectively. Wilcoxon test was done to assess statistical significance, and comparison approaching significance are indicated.

OPEN ACCESS
iScience 26, 106417, April 21, 2023 5 iScience Article and deleterious effects on disease progression, in addition to stage-specific (and at times conflicting) effects. 17 Except for C9, all reported variants in the complement cascade are clustered in the early steps of the pathway, leading to a concentrated effort to modulate or augment these steps for the development of therapies. Of the 21 targets specifically directed to the complement cascade, only one trial is focusing on the MAC, using an AAV to produce CD59, which prevents the binding of C9 and disrupts the assembly of the MAC (ClinicalTrials.gov: NCT04358471). 18 However, their success in clinical trials has been limited at best. 19 The concept of ''inflammaging'' was introduced to provide a framework for many age-related conditions, including AMD, 20,21 with chronic low levels of inflammation being a key contributor. Inflammasome activation has been reported in RPE from AMD patients as well as in ARPE-19 cells. 22,23 Furthermore, multiple studies have shown that MAC dysregulation can lead to inflammasome activation. 24-28 Notably, inflammasome activation can drive inflammation in C9 À/À mice lacking the ability to form MAC. 24 These studies, together with our data, indicate a complex interplay between AMD, MAC and the inflammasome. In Figure 4, we propose a model to explain the role of MAC in AMD. Terminal MAC can therefore be mentioned as a target of therapies for age-related diseases such as AMD, where ''inflammaging'' might be a contributor to pathology.

Limitation of the study
Our genetic analysis is based on exome capture, which is limited in the coverage of the genome to the (mostly) coding regions that are covered by the probes provided. This might lead to incomplete coverage of some genes or regions, particularly in the regulatory or deep-intronic regions. WES is also prone to uneven coverage because of differences in binding of the probes, leading to improper detection of variants.
The effect of variants is deduced by computational analysis or in HEK293 cells. Further investigations are needed with relevant target tissues or cells and using in vivo model systems to examine the contribution of C8 and MAC in AMD pathology.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

AUTHOR CONTRIBUTIONS
Overall

Lead contact
Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Anand Swaroop (swaroopa@nei.nih.gov) and Michael L. Klein (kleinm@ohsu.edu).

Materials availability
This study did not generate new unique reagents.
Data and code availability d All data reported in this paper will be shared by the lead contact upon request and is available at NCBI SRA bioproject PRJNA805222.
d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
d Whole exome and Sanger sequencingdata were deposited to the NCBI SRA under bioproject PRJNA805222.

Study cohort
This research was approved and conducted in accordance with the Oregon Health & Science University Institutional Review Board. This cohort is comprised of 152 families recruited in the Pacific Northwest United States, with 1287 ascertained family members phenotyped for AMD. Patients seen at the Casey Eye Institute (Portland, OR, USA) who were identified as having a family history of AMD were invited to participate in genetic research and to assist us in reaching out to other members of their family. All subjects enrolled in the study agreed to participate after informed consent. Participants were asked to provide their medical history of AMD including fundus photographs, provide a blood sample, answer a questionnaire, describe known family history of AMD, and agree to follow up communications. The questionnaire included questions regarding demographics, ocular disease history, cardiovascular disease history, diabetes history, smoking habits, medications, and supplements. Over the years, many participants were re-contacted for follow up information and to obtain additional fundus photographs. All fundus images were carefully graded using a modification of the AREDS scale. 32 Each eye was graded as follows based on drusen size and/or evidence of GA or CNV, and the phenotype category for the worse eye was used for each individual.
We then applied a segregation filter. Variants segregating in all affected individuals or all affected individuals but one within a family were retained. Finally, the variant level data were collapsed into gene-level data by combining all variants observed in each gene across different families.

Cell transfections
Before transfection 2 3 10 5 HEK293 cells were seeded into individual wells of 6 well plates. After a 48 h incubation in growth medium, transfections were performed using lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. The ratio Lipofectamine 2000/DNA was 3:1. After transfection, the cells were incubated at 37 C in a humidified incubator with 5% CO 2 for 24 h. Each transfection was carried out in triplicates and repeated 3 times. Cells were co-transfected with the following combination of plasmids (500 ng each): (1)   The following files from the RCSB PDB database (https:// www.rcsb.org) 2rd7, 3ojy, and 2gos were used as structural templates to build the homology model of the C8 trimer. PyMol (https://pymol.org/2/) was used for visualization (variants introduced using the mutagenesis option in the wizard) and hydrogen bond interactions. Mutant variants R444H (CO8A subunit) and D382Y (CO8B subunit) were generated by the 'edit: swap' procedure and equilibrated using 500 ns molecular dynamics in water in Yasara. Protein stability was evaluated using the FoldX plugin in Yasara. 39 Before protein stability calculations, trimeric molecules C8, isolated alpha and beta subunits, and corresponding mutant variants were repaired by FoldX (https://foldxsuite.crg). Changes in protein stability were calculated as the following: DDG = DG(mutant) -DG(wild type).
Here DG(mutant) and DG(wild type) are protein stabilities for mutant variant and wild type protein respectively.
Structural model files (PDB format) are included as separate files (Data S2).

QUANTIFICATION AND STATISTICAL ANALYSIS AREDS1 and AREDS2 analysis
Variants called from whole-genome sequencing of the AREDS and AREDS2 cohorts were filtered to only C8A rs143908758 and C8B rs139498867. Only AMD cases were kept, of any subtype (N = 2451). A custom R script (R version 4.0.3) processed VCF data and counted the number of heterozygous carriers, homozygous carriers, and non-carriers. The sum of all carriers with AMD was compared to the number of carriers in the ExAC exome sequencing cohort (N = 60706). ExAC carrier rates were obtained from gnomAD browser. 40 Chi-squared tests were run to approximate the expected carrier and noncarrier counts for AMD and ExAC samples. Since the number of carriers was small, Fisher's Exact tests were run. P-values from Fisher's Exact and carrier counts were reported. R packages used included data. Table 1.14.2, ggplot2 3.3.5, and base stats 4.0.3. Samtools and bcftools from htslib 1.13 were used to manipulate VCFs. 41 This work used the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov).

IP signal quantification
Raw images of IP membranes were analyzed using Image Lab software (BioRad). Signal intensity was determined by selecting the appropriate bands on each membrane using the lanes and bands function in the software. Background level subtraction was done as recommended by the manufacturer tutorial. Adjusted signal intensity (found in supplementary Excel file 1) was exported to an Excel file for further analysis. To enable comparison the intensity of the protein bands in each transfection and IP conditions was quantified as previously described in the STAR protocol. 42 Briefly, the intensity was first normalized to the amount of protein in the input control (indicating the total amount of protein generated in the cells) and then divided by the quantity in the WT conditions, for ease of visualization. WT levels are always indicated by 1, increase in the amount of protein will give a ratio >1, decrease in the amount of protein will give a ratio <1. Wilcoxon test was performed to assess statistical significance.