Functional genomics in age-related macular degeneration: From genetic associations to understanding disease mechanisms

Abstract

Genome-wide association studies have been remarkably successful in identifying genetic variants associated with age-related macular degeneration (AMD), demonstrating a strong genetic component largely driven by common variants. However, progress in translating these genetic findings into a deeper understanding of disease mechanisms and new therapies has been slow. Slow progress in this area can be attributed to limited knowledge of the functional impact of AMD-associated non-coding variants on gene function, the molecular mechanisms and cell types underlying disease. This review offers a comprehensive overview of functional genomics approaches to uncover the genetic, epigenetic, cellular and molecular mechanisms underlying AMD and outlines future directions for research.

1. Introduction

The human genome sequence, genome-wide association studies (GWAS) and more recently the application of next-generation sequencing have revolutionized the study of genetics of multifactorial or complex diseases. Several GWAS for Age-related Macular Degeneration (AMD) have established a strong role for genetic variants in the disease etiology (Fritsche et al., 2016; Gorman et al., 2022). A sentinel GWAS for common and rare variants by the International AMD Genomics Consortium identified 52 independent genetic variants at 34 loci associated with AMD (Fritsche et al., 2016). The number of loci was recently been expanded to 46 in a larger GWAS meta-analysis (Han et al., 2020) and to 63 in a cross-ancestry GWAS (Gorman et al., 2022).

Collectively, these studies have led to several key insights. First, these risk variants explain a substantial proportion of the estimated heritability of late AMD in European ancestries, and position AMD at the extreme end of complex disease genetics with a substantial proportion of genetic heritability explained by a small number of strong susceptibility variants. Second, the functional interpretation of causal variants, genes and the underlying disease mechanisms has not been straightforward since the majority of AMD-associated variants lie within non-coding regions of the genome, and for most AMD loci, no functional follow-up studies have been performed (Strunz et al., 2020). Finally, the full potential of these studies remains unmet, as translational and clinical advances have been slower than anticipated. While GWAS findings have contributed to development of anti-complement therapies for treating atrophic AMD (Wilke and Apte, 2024), functional improvements for patients on these therapies are modest (Csaky et al., 2024). Additionally, many other drugs have failed in clinical trials despite seemingly promising GWAS and in vitro studies (Holz et al., 2014; Qin et al., 2021), in part due to limited understanding of disease biology. Thus, accelerating the progress from genetic association to functional interpretation is critical for realizing the true clinical potential of genetic findings.

The Multi-Omics task group of the Ryan Initiative for Macular Research (RIMR) aimed to provide direction on how genomics--combined with other omics approaches--can be used to better understand the pathogenesis of AMD. Discussions focused on various omics approaches that have been used to decipher the effect of AMD-associated genetic variants on the disease mechanisms. This review captures parts of the discussions held during the meeting and provides a comprehensive overview of functional genomic approaches that can help turn genetic associations into understanding of AMD disease mechanisms.

2. Gene prioritization and functional annotation of genetic variants

2.1. Bioinformatic gene prioritization and fine mapping

After a GWAS, markers that show significant statistical associations, typically with odds ratios >1 for the risk allele or < 1 for the protective allele, and p-values < 5 x 10⁻⁸ are selected using standard genome-wide significance thresholds (Roeder and Wasserman, 2009; Xu et al., 2014). Larger GWASs, such as those described above for AMD, include multiple datasets with meta-analyses displaying replication of the top results (Kraft et al., 2009). The first step in selecting loci for fine mapping includes prioritizing high confidence loci that display consistent odds ratios and replicate across multiple cohorts. Fine mapping can be done solely by relying on the local genomic (DNA) architecture, or by integrating information from other functional omics assays such as gene expression, DNA methylation and other epigenetic marks, and protein structure/function assays. The initial size of the chromosomal region that shows association (haplotype blocks together with the 3D chromatin architecture), and the concordance of the results across various ethnic groups, may drive specific strategies for narrowing down to causal variants.

Initially in many association studies, variant and gene prioritization (i.e., identification of the most likely causal variant and thus affected gene) is usually done by ranking each variant by its impact on genes near the association peak. This can be done automatically by assigning each variant a pathogenicity score such as a Combined Annotation Dependent Depletion (CADD) Score (Rentzsch et al., 2019) or by employing prediction algorithms (Wang et al., 2022a). Those methods, however, generally assume that a single variant is causal for the observed association and that the variant is in linkage disequilibrium (linked or correlated) with the associated variants. As knowledge of cell-type-specific gene expression patterns and other regulatory variant types has improved, a role for multiple variants at a GWAS locus has emerged (Abell et al., 2022). At some loci the primary GWAS signal may be explained by two (or more) independent causal variants, as described below for the CFH and C2/CFB loci.

2.2. Examination of the local DNA architecture

At the ARMS2/HTRA1 locus on chromosome 10 a single haplotype (i.e., variants in a chromosome section that are inherited together) confers the risk from this region, with no evidence for additional haplotypes (Grassmann et al., 2017). In essence, this means that all risk-increasing alleles are inherited at the same time and thus carry nearly the same probability of association. Here, prioritizing the most impactful variant could be useful to identify the underlying mechanisms behind the observed association.

In contrast, there are several regions where the assumption of a single variant is not true (12). Figure 1A shows the local Manhattan plot of associated signals in the complement factor H (CFH) region on chromosome 1 (Fritsche et al., 2016). The most strongly associated variant is (rs10922109) is intronic. However, this variant (and correlated variants) might not primarily drive the association signal. Instead, there are two protective haplotypes in this region: one carrying the CFH I62V variant (H2), the other a large deletion of CFH-related genes 1 and 3 (CFHR1/CFHR3, haplotype H3, Figure 1B). In addition, there is a risk-increasing haplotype H1 that carries the well-known CFH rs1061170 (Y402H) variant. Importantly, the risk-reducing allele of the main association signal (rs10922109) is present on both protective haplotypes H2 and H3, but absent from the risk-increasing H1 haplotype. Therefore, the association signal itself is potentially not the driving factor in this region, but rather different haplotypes capture the underlying risk.

Figure 1.

Overview of approaches in identifying and interpreting GWAS associations results. A: DNA based genotyping and sequencing can be used to identify novel loci. B: Functional genomic approaches leverage on transcriptome and epigenome data integration with genetics data to identify the causal mechanisms. C: Available model systems that could be employed to model disease.

A similar example is seen in the complement component 2/complement factor B (C2/CFB, Figure 2C) region on chromosome 6. Both protective haplotypes in this region, H1 and H2, carrying the L9H and R32Q variants respectively, are “tagged” by the lead variant in this region (rs429608, incorrectly labeled in its first reporting as rs116503776). Again, the risk conveyed by rs429608 is likely explained by the combined risk from two different haplotypes carrying two distinct functional variants, which are only weakly linked to rs429608 (R² of 0.52 to R32Q and 0.28 to L9H, respectively). These findings thus highlight that, for variant and consequently gene prioritization, a deep understanding of the underlying genomic architecture and biological function is essential.

Figure 2. Haplotype structure at the CFH and C2/CFB loci associated with AMD.

For variant and gene prioritization, understanding the underlying genomic architecture and thus haplotype structure is essential. A: Regional Manhattan plot of the CFH locus on chromosome 1. The negative decadic logarithm of the P-value of association is plotted on the y axis and the genomic position of each variant on chromosome 1 is plotted on the x axis. The lead variant, rs10922109 is highlighted in purple. B: Four common haplotypes in this region are shown with their respective defining and likely causal variants. The risk increasing haplotype H1 is characterized by the Y402H mutation, while H2 carries the I62V variant. H3 contains a common deletion encompassing CFHR1 and CFHR3. The lead variant perfectly tags both protective haplotypes, thus represents the combined signal of both haplotypes without having a functional effect itself. C: At the C2/CFB locus on chromosome 6, both protective haplotypes H1 and H2 are tagged by rs429608, which is the lead variant with the smallest P-value in this region. Neither rs429608 nor strongly correlated variants have an obvious impact on genes in the region.

2.3. Coding variants in AMD

Identifying rare coding variants in AMD patients can help pinpoint causal genes and identify therapeutic targets, ideally increasing likelihood of success through the drug development pipeline (Plenge et al., 2013). The population attributable risk of each rare variant is likely low, but discovery of a cluster of deleterious/protective rare variants in a single gene provide knowledge of the role of a specific protein in disease, and can be useful in development of diagnostics and/or therapeutics. For instance, nonsense mutations in the PCSK9 gene have been shown to reduce mean LDL cholesterol levels and the risk of coronary heart disease (Cohen, 2006), which led to the successful development of a drug targeting PCSK9 (Sabatine et al., 2017).

Traditional GWAS, which primarily focuses on common variants (minor allele frequency > 5%), has now been adapted to include rare population variants (Fritsche et al., 2016; Gorman et al., 2022). However, this approach only analyzes rare variants represented on the arrays used in the study and does not uncover novel rare variants. The plummeting cost of genome sequencing and improvements in sequencing efficiency have enabled large-scale efforts, both population-based and familial, to expand the catalog of potential causal variants associated with AMD within and beyond known GWAS loci (de Jong et al., 2021b; Geerlings et al., 2017; Kwong et al., 2024; Ratnapriya et al., 2020; Saksens et al., 2016).

In several AMD-associated genes, such as CFH, CFI, and C3, rare or ultra-rare coding variants have been identified, underscoring the critical role of the complement system in AMD. The deleterious effect of several of these coding variants on complement activity has been demonstrated by molecular and cellular studies, but the majority remain to be functionally validated (de Jong et al., 2021a),. Additionally, rare coding variants can affect disease characteristics and disease severity. For example, rare variants in complement genes are associated with earlier disease onset (de Breuk et al., 2021) and with faster disease progression to advanced AMD (Seddon et al., 2023; Seddon et al., 2015). Achieving statistical significance remains a primary challenge in evaluating the contribution of rare variants in AMD, requiring very large sample sizes. As an alternate strategy, genetic studies may focus on isolated populations to aid the identification of rare and low-frequency genetic variants which have arisen recently, and are still concentrated in limited geographical areas.

3. Linking genetic variants to RNA, protein, and metabolite levels through quantitative trait loci (QTLs)

The vast majority of AMD-associated variants are located in non-coding regions and often overlap enhancers, promoters and open-chromatin regions, suggesting a role in regulating gene expression. As regulation of gene expression is highly cell-type and tissue-specific, it is critical to perform functional genomics studies in disease-relevant cells and tissues. However, for many diseases, these pathologically relevant cell types and cell states are unknown. QTL (quantitative trait loci) analysis has emerged as a popular method to connect at the molecular level with the risk variants to identify the causal genes and mechanisms affected in the disease-relevant cell/tissues. Various types of QTLs have been studied in AMD, including gene expression (eQTLs), exon splicing (sQTLs), single cell (scQTLs), DNA methylation (mQTLs), chromatin accessibility (caQTLs), protein (pQTLs) and metabolic (metQTLs) QTLs, which have contributed to gaining insight into causal genes at AMD GWAS loci (Table 1).

Table 1.

AMD-GWAS loci with lead SNPs, their putative target genes and location within open chromatin in retina and macula.

GWAS					eQTL						Open chromatin
Locus	Chr	Pos	rs ID	Known/Novel	RPE (P)	RPE (M)	Retina (P)	Retina (M)	GTEx	GTEx Target Genes	RPE (P)	RPE(M)	Retina(P)	Retina(M)
CFH	1	196664793	rs485632	Known					25	3
COL4A3	2	228177181	rs56123646	Known					2	3
ADAMTS9	3	64730121	rs17676309	Known					1	1
COL8A1	3	99201587	rs17826006	Known					0	0
CFI	4	110656730	rs7439493	Known			PLA2G12A, CFI		35	3
C9	5	39327888	rs62358361	Known					0	0
CFB/C2/SKIV2L	6	29605935	rs116306799	Known			HLAFAS1, HLAV, HCG4, POLR1HASP		0	0
VEGFA	6	43826627	rs943080	Known					2	2	Y	Y
NT5DC1/COL10A1	6	116433265	rs476330	Known			TSPYL4		20	7	Y		Y	Y
PILRA	7	99991548	rs7803454	Known	PILRA, PILRB	PILRB	PILRB, STAG3L5P	PILRA, PILRB	49	17
KMT2E	7	104756355	rs1144	Known					7	5			Y
TNFRSF10A	8	23082971	rs79037040	Known					28	3	Y	Y	Y
RORB	9	76600134	rs11143745	Known		RORB			1	1	Y
TGFBR1	9	101926594	rs334362	Known					1	1
ABCA1	9	107665739	rs2575876	Known					0	0			Y	Y
ARHGAP21	10	25054210	rs4570483	Known					0	0
ARMS2/HTRA1	10	124215211	rs36212733	Known					11	3	Y	Y	Y	Y
BLOC1S1-RDH5	12	56115778	rs3138141	Known	RDH5		BLOC1S1	AC009779.3, BLOC1S1	46	6
ACAD10	12	112080669	rs73203613	Known					0	0
B3GALTL	13	31821240	rs9564692	Known			B3GLCT	B3GLCT	22	2
RAD51B	14	68799787	rs1956526	Known					0	0
LIPC	15	58680638	rs2414577	Known					3	4
CETP	16	56994528	rs17231506	Known					11	2
CTRB2	16	75234872	rs72802342	Known					10	4	Y
NPLOC4	17	79619408	rs7406891	Known	TSPAN10		NPLOC4, NPLOC4, ARL16		36	7
ABCA7/CFD	19	1059243	rs1137776	Known					0	0
NRTN/FUT6	19	5832773	rs78060698	Known					0	0
C3	19	6716279	rs11569415	Known					7	2
APOE	19	45411941	rs429358	Known					4	3
PLTP/MMP9	20	44549476	rs6073957	Known			SNX21, ACOT8, CTSA, PLTP		22	8
C20orf85	20	56653122	rs145616409	Known					0	0
SYN3/TIMP3	22	33091812	rs7289865	Known					0	0
SLC16A8	22	38478666	rs77968014	Known					0	0	Y
CD46/CD55	1	207960708	rs4844619	Novel					32	2
LRP2	2	170067947	rs11689553	Novel			LRP2		0	0
ZNF385B	2	180725310	rs2170240	Novel					5	1	Y	Y	Y	Y
CAND2/TMEM40	3	12819789	rs28731356	Novel			RAF1		9	6
ZBTB38	3	141096185	rs9821337	Novel			ZBTB38		14	2
IGFBP7	4	57927876	rs1277297	Novel			REST, NOA1	REST	0	0
AFF1	4	87988590	rs2280953	Novel					14	5			Y
RP1L1	8	10573698	rs75731123	Novel					9	6	Y	Y	Y
HSDL2	9	115150612	rs10981380	Novel			HSDL2, KIAA1958		19	4
C5	9	123759044	rs7040033	Novel			CUTALP		42	12
ME3	11	86326044	rs4439535	Novel			ME3		14	3
TYR	11	89011046	rs1393350	Novel					0	0
LINC02343	13	34912828	rs621736	Novel					0	0
SERPINA1	14	94838142	rs112635299	Novel					6	1
HERC2	15	28365618	rs12913832	Novel					1	1	Y	Y
TRPM1	15	31394868	rs7182946	Novel	TRPM1	TRPM1	MTMR10, MTMR10		12	1	Y	Y
MYO1E	15	59588995	rs4238336	Novel					0	0
SMAD3	15	67338914	rs6494627	Novel					4	3
CSK	15	75097780	rs35206230	Novel	RPP25		MPI		36	15
RLBP1	15	89760997	rs2070780	Novel	RLBP1		RLBP1	RLBP1	20	2
CHD9	16	53370021	rs62049817	Novel				AC007342.9	6	2
CLUL1	18	597950	rs9973159	Novel			CLUL1		0	0	Y	Y	Y	Y
ACAA2	18	47285763	rs880448	Novel			ACAA2		0	0	Y	Y	Y	Y
TGFB1/CCDC97	19	41825191	rs73045269	Novel					17	6
EXOC3L2/MARK4	19	45748362	rs73036519	Novel					4	3
RASIP1	19	49241015	rs427248	Novel	NTN5	FUT2	NTN5	FUT2, NTN5	35	8
RRAS	19	50193163	rs201107314	Novel					0	0	Y		Y	Y
LBP	20	36997655	rs2232613	Novel					0	0
PDGFB	22	39647168	rs77516118	Novel					3	2

3.1. Expression quantitative trait loci (eQTLs)

In 2019, the first eQTL map of the human retina (EyeGEx) was established, constructed from >500 post-mortem donor eyes. This study revealed that approximately 60% of expressed genes in the retina are under genetic regulation, involving 14,565 variants (eVariants) influencing the expression of 10,474 genes (eGenes) at a false discovery rate (FDR) of 5% (Ratnapriya et al., 2019). Integration of retina eQTL results with AMD GWAS loci revealed nine lead variants at AMD GWAS loci that are significant eQTLs in the retina (Table 1). These were further evaluated for causality by co-localization studies (described in section 3.2).

In a second eQTL study involving 129 donor eyes, separate analyses of the retina and retinal pigment epithelium (RPE) identified eQTLs in 18,935 genes across four eye regions (retina periphery/macula and RPE periphery/macula). Among these genes, approximately 32% showed eQTLs in both the RPE and retina, with 20% and 47% exclusive to the RPE/choroid and retina, respectively (Orozco et al., 2020). The identified QTLs were used to prioritize candidate genes underlying AMD GWAS loci (Table 1), and co-localization was used to determine causality of these genes in AMD disease (described in section 3.2). Finally, eQTL studies on 23 primary human fetal RPE (fRPE) cells from 23 primary human identified eQTLs for 1173 gene expressed in RPE (Liu et al., 2019).

These studies established the role of common genetic variants, including those associated with AMD risk, in regulating gene expression in the retina and RPE. These QTL studies have proven to be a valuable resource to identify causal variants and genes at AMD GWAS loci, using the statistical methods described in sections 3.2–3.4.

3.2. Colocalization

Co-localization methods can be used to determine whether the same variant responsible for a GWAS locus also affects gene expression. Two co-localization methods, eCAVIAR (Hormozdiari et al., 2016) and COLOC (Giambartolomei et al., 2014), have been implemented in AMD GWAS to identify the effector genes and causal variants. Implementation of the eCAVIAR using retina eQTL and AMD GWAS data identified likely causal variants and underlying effector genes at six AMD loci (Ratnapriya et al., 2019). At two of these loci (B3GALTL and RDH5/CD63), the lead GWAS variant was identified as the most likely causal variant, whereas the likely causal variant was distinct from the lead variant at four other loci (Table 2).

Table 2.

Co-localization of AMD GWAS SNPs with retina/RPE eQTLs

Ratnapriya et al., 2019
GWAS locus	Lead SNP	Causal SNP	Target Gene	Target tissue
TMEM97/VTN	rs11080055	rs241777	TMEM199	Retina (P)
RDH5/CD63	rs3138141	rs3138141	BLOC1S1	Retina (P)
SLC16A8	rs8135665	rs5756908	CTA-228A9.3	Retina (P)
ACAD10	rs61941274	rs7398705	SH2B3	Retina (P)
B3GALTL	rs9564692	rs9564692	B3GLCT	Retina (P)
CFI	rs10033900	rs9995011	PLA2G12A	Retina (P)
Orozco et al., 2020
GWAS locus	Lead SNP	Causal SNP	Target Gene	Target tissue
ARMS2-HTRA1	rs3750846	rs11528744	HTRA1	Retina (M)
PILRB-PILRA	rs7803454	rs28495773	PILRA	Retina (M)
MMP9	rs142450006	rs3761159	SLC12A5-AS1	Retina (P)
B3GALTL	rs9564692	rs4381465	B3GLCT	Retina (M)
RDH5-CD63	rs3138141	rs56108400	BLOC1S1	Retina (P)
PILRB-PILRA	rs7803454	rs6965458	PILRB	Retina (M)
TMEM97-VTN	rs11080055	rs7212510	TMEM199	Retina (M)
COL4A3	rs11884770	rs7559693	COL4A3	Retina (P)
CTRB2-CTRB1	rs72802342	rs9928736	BCAR1	Retina (M)
Orozco et al., 2020
GWAS locus	Lead SNP	Causal SNP	Target Gene	Target tissue
TNFRSF10A	rs79037040	rs11777697	TNFRSF10A	RPE (P)
TRPM3	rs71507014	rs12440880	TRPM1	RPE (P)
RDH5-CD63	rs3138141	rs3138142	RDH5	RPE (P)
NPLOC4-TSPAN10	rs6565597	rs7405790	TSPAN10	RPE (P)
SLC16A8	rs8135665	rs760975	BAIAP2L2	RPE (P)
CFI	rs10033900	rs141098594	CFI	RPE (M)
Gorman et al., 2024
GWAS locus	Lead SNP	Causal SNP	Target Gene	Target tissue
CD46/CD55	rs2724360	rs2724360	CD46	Multi-tissue
RDH5-CD63	rs3138141	rs3138142	RDH5	Multi-tissue
RDH5-CD63	rs3138141	rs3138141	BLOC1S1	Multi-tissue
PDGFB	rs77516118	rs3827357	FAM227A	Multi-tissue
CFI	rs10033900	rs7439493	CFI	Multi-tissue
TNFRSF10A	rs79037040	rs13278062	TNFRSF10A	Multi-tissue
ABCA1	rs2740488	rs1800978	ABCA1	Multi-tissue

Colocalization of retina and RPE eQTL data with AMD GWAS data identified 6 causal variants regulating the expression of 6 effector genes (BAIAP2L2, TSPAN10, TRPM1, RDH5, TNFRSF10A, CFI) in RPE, and 9 causal variants regulating expression of 9 effector genes in retina (PILRA, PILRB, B3GLCT, SLC12A5-AS1, BLOC1S, TMEM199, BCAR1, COL4A3 and HTRA1) (Orozco et al., 2020) (Table 2). The findings of RDH5 was also replicated in fRPE(Liu et al., 2019). Co-localization using eCAVIAR with a recent multi-ancestry AMD GWAS identified seven candidate causal variants modifying the expression of 7 genes (CD46, RDH5, BLOC1S1, FAM227A, CFI, TNFRSF10A, and ABCA1) in eQTL data from multiple tissues(Gorman et al., 2022) (Table 2). Caution should be exercised when interpreting co-localization results due to uncertainty induced by linkage disequilibrium, as well as the potential presence of multiple causal variants at the same locus.

3.3. Transcriptome-wide association studies (TWAS)

Transcriptome-wide association studies (TWAS) can aggregate the genomic information into functionally relevant units to identify genes whose expression is significantly associated with AMD without necessity for direct measurement of expression levels (Gusev et al., 2016). TWAS integrates the information from eQTL, GWAS summary statistics and linkage disequilibrium reference panels to associate the genetically regulated component of gene expression. A recent study leveraged EyeGEx for which both gene expression and genetic variation (eQTL) were measured to impute the cis genetic component of expression into summary-level data from the most recent International AMD Genomics Consortium study of 16,144 AMD patients and 17,832 controls (Fritsche et al., 2016). This analysis identified 61 transcriptome-wide significant gene-AMD associations (FDR < 0.05); of these, 38 genes were within 1 Mb of 14 AMD-GWAS loci, and three genes (HIC1, RLBP1 and PARP12) had not been previously associated with AMD. At eight of these loci, only a single target gene was identified (Ratnapriya et al., 2019).

3.4. Mendelian randomization

Mendelian Randomization (MR) uses genetic variation to address causal questions about whether modifiable exposures influence health, developmental or social outcomes (Sanderson et al., 2022). Summary Data-based Mendelian Randomization (SMR) uses the MR framework to estimate the effect of a molecular phenotype on a disease/trait (Zhu et al., 2016). Recently, SMR has been applied to assess the causal or pleiotropic association between mQTL data and AMD, which identified 5 causal associations at 3 loci (KMT2E/SRPK2, PILRB/PILRA and ARMS2/HTRA1 (Advani et al., 2024). Similarly, SMR analysis using retina eQTLs identified 8 causal associations at 5 loci (CFI, C2/CFB/SKIV2L, PILRB/PILRA, RDH5/CD63 and TMEM97/VTN) with notable overlap of target genes identified through co-localization studies (Ratnapriya et al., 2019). Since each method has unique assumptions, integrating these analyses can offer a comprehensive view of causal relationships between variants and molecular phenotypes.

3.5. eQTLs at the single cell level

Gene regulation exhibits high context specificity, which necessitates the study of gene regulation at the single cell level. Yet, most eQTL studies have been performed on bulk samples, where the estimated effect sizes reflect an average effect across various cell types. As most tissues are composed of multiple cell types in varying proportions, the cell types driving the inferred eQTLs remain elusive. The significant expense associated with single-cell studies has been a major hindrance to conducting large-scale single-cell eQTL (sc-eQTL) analyses.

A recent study examined 20 human retinas to map eQTLs, caQTLs, allelic-specific expression, and allelic-specific chromatin accessibility in major retinal cell types (Wang et al., 2023). The study identified approximately 4,000 genes under regulation. In contrast to bulk eQTL studies (Ratnapriya et al., 2019), where significant eQTLs are shared across tissues, this study reported that the majority of sc-eQTLs (~90%) were specific to particular cell types. However, caution is warranted in interpreting these results. The observed specificity of sc-eQTLs could also be partially attributed to the potential incompleteness in identifying all possible eQTLs across different cell types, as evidenced by the discovery of 4,069 sc-eGenes, with varying numbers of sc-eQTL sets per cell type, ranging from 704 to 1,175. Integrating these data with AMD GWAS and generating larger ocular eQTL datasets at single-cell resolution would be a valuable resource for understanding the impact of AMD-associated variants on specific disease-relevant cell types.

Single cell RNA sequencing studies of the retina, RPE and choroid have been helpful to determine which cell types are enriched for expression of genes in or near AMD GWAS loci. AMD-associated genes were enriched in neural cells types such as cone photoreceptors, bipolar cells and horizontal cells, in addition to non-neural cell types including RPE, glial cells, monocytes/myeloid cells, vascular cells, choroidal cells, and fibroblasts (Cowan et al., 2020; de Breuk et al., 2021; den Hollander et al., 2022). The role of glia in AMD and use of scRNAseq datasets to understand glial disease biology in AMD is discussed in section 3.6.

3.6. Retinal microglia expression of AMD-GWAS Loci by scRNA-seq

As an innate immune cell, the role of macrophages in AMD has been a significant area of investigation (Essner and Gorrin, 1979; Gupta et al., 2003; Lad et al., 2015; Penfold and Provis, 1986; Sennlaub et al., 2013; Thanos, 1992). Microglia, the resident macrophages of the central nervous system (CNS) are yolk-sac derived, long-lived and self-renewing, and the only macrophage type present in normal neuroretina (Ginhoux et al., 2010; O’Koren et al., 2016; O’Koren et al., 2019; Reyes et al., 2017) . Perivascular macrophages are also present in the steady state retina but reside in the perivascular spaces and are poorly understood (Ginhoux et al., 2010; O’Koren et al., 2016; O’Koren et al., 2019; Reyes et al., 2017). In retinal degeneration, monocyte-derived macrophages are recruited from the blood into the retina (Beguier et al., 2020; Chen et al., 2011; Guo et al., 2012; Karlen et al., 2018; Levy et al., 2015; O’Koren et al., 2016; Sennlaub et al., 2013). Thus, three macrophage types exist in diseased retinas: microglia, perivascular, and monocyte-derived macrophages (Karlen et al., 2018; Reyes et al., 2017). The choroid also contains macrophages (and other immune cells), but these are even less well characterized (Droho et al., 2023; McMenamin et al., 2019; O’Koren et al., 2019; Voigt et al., 2022; Yang et al., 2020).

Contrary to the assumption that macrophages are pathologic in AMD, work in the CNS suggested otherwise (Park et al., 2022). Further, O’Koren et al. showed that microglia are in fact protective in two preclinical models of retinal degeneration, exerting protective effects by clearing photoreceptor debris (O’Koren et al., 2019), a finding supported by others (Funatsu et al., 2022; He et al., 2022; Zhou et al., 2023). Yu et al. demonstrated that protective microglia share a common transcriptional signature across four distinct preclinical models (Yu et al., 2024). This signature protective program involves microglia migrating to the subretinal space to clear otherwise cytotoxic debris (Yu et al., 2024). Yu et al. also demonstrated that this protective signature is highly conserved in human AMD. Using scRNA-seq, they found that 87% of differentially expressed genes (DEG) were conserved between mouse models and human AMD (Yu et al., 2024) (GSE208434). Immunohistochemistry and flow cytometry confirmed the presence of these microglia in the subretinal space of AMD donors, with enrichment in the macula. Integration of additional scRNA-seq datasets GSE183320, GSE203499) further validated these findings.

A comprehensive map of eQTLs for retina cell types, including immune cells is not available. However, using the atlas of retinal macrophages by Yu et al., genes in AMD-GWAS loci were examined. Three human microglial clusters were identified: hMG1, hMG2, and hMG3. Cluster hMG2, validated as subretinal microglia enriched in the macular region in AMD donors, shared 87% DEGs with protective microglia in mice. Other clusters included monocyte-derived and perivascular macrophages (mo-MF and pvMF, respectively) (Figure 3A, B). Interestingly, microglial clusters showed the highest expression of GWAS loci genes (Figure 3C, D) with hMG3 expressing genes like C3, APOE, VEGFA, PDGFB, KMT2E, SERPINA1, PILRA, and TGFBR1 (Figure 3D) (Yu et al., 2024). This pattern suggests that microglia in AMD significantly express genes of AMD-GWAS loci, particularly subretinal microglia of the macula. Generation of larger, comprehensive single-cell eQTL maps of retinal cell types are needed to further understand the effect of AMD-associated variants on glial cell subsets.

Figure 3.

Human retinal macrophage expression of genes in AMD-GWAS loci by scRNA-seq. A, B: Workflow by Yu et al(Yu et al., 2024) for identification of protective subretinal microglia signature in human AMD eyes (human microglia cluster [hMG1, 2, 3], monocyte-derived and perivascular macrophage [mo-MF, pv MF, respectively]). Cluster hMG2, validated as subretinal microglia enriched in the macular region in AMD donors, shared 87% DEGs with protective microglia in mice. C: Module analysis of n=64 genes containing AMD-GWAS loci. D: Percent expressed and average expressions of gene by macrophage cell type (GSE208434 GSE183320, GSE203499).

3.7. Protein quantitative trait loci (pQTLs)

Proteomics involves a comprehensive analysis of protein function and structure, providing valuable insights into health vs. disease. Integrating proteomics with genetics, transcriptomics, and epigenetic data enhances our ability to identify disease genes, pathological processes and discover robust biomarkers. Despite its potential, proteomics faces challenges such as low throughput and technical limitations, including issues with the aggregation and precipitation of membrane proteins, leaving a key knowledge gap in biomedical research in the post-genomic era. Recent advancements in proteomic methodologies, particularly high-throughput approaches such as mass spectrometry, protein pathway arrays, next-generation tissue microarrays, single-cell proteomics, and single-molecule proteomics (Cui et al., 2022), have significantly improved protein identification and quantitation. These developments offer new insights into molecular mechanisms of pathogenesis and the signaling networks associated with specific diseases.

In the context of AMD, proteomic studies have been conducted on donor eyes, as well as various ocular and systemic fluids, such as aqueous humor, vitreous humor, tears, plasma, and urine (Garcia-Quintanilla et al., 2022; Kersten et al., 2018). The application of robust proteomics studies in the eye has been constrained primarily by small sample sizes that complicate replication across studies. As tissue availability is scarce, a recent study traced the cellular origin of 5,900 proteins in locally enriched fluid compartments (Wolf et al., 2023). It identified hundreds of cell-specific protein markers, including those for individual retinal cell types, suggesting that systemic fluids can provide crucial information about AMD (Wolf et al., 2023). AMD-associated genetic variants have been described to alter complement proteins in the systemic circulation, and it has been suggested that pQTL studies – extending to proteins outside of the complement system – in plasma and serum samples may aid in understanding the effect of genetic variants and pinpointing causal genes in AMD (den Hollander et al., 2022).

As an alternative source for AMD-relevant tissues, pQTL studies can be performed in induced pluripotent stem cell (iPSC)-derived cells. For genomics studies, population-scale iPSC studies using “cell villages” has become feasible. section 5.3). A recent study explored genetic regulation of proteins in geographic atrophy by integrating the genotype profiles with transcriptomics and proteomics data from iPSC-derived RPE of 36 control individuals and 43 GA patients. This study identified co-localization of pQTLs for three proteins: PYROXD2 (rs942813), SPATA20 (rs989128), and CRYZ (rs1424671) (Senabouth et al., 2022).

These findings underscore the potential of proteomics and pQTL studies to advance our understanding of AMD and point to its future promise in identifying and validating effective biomarkers. Such pQTL studies must still be performed on eye samples, using analysis of systemic proteins for further insight into AMD pathophysiology:

3.8. Metabolic quantitative trait loci (metQTL)

Metabolomics are affected by genomics and by environmental factors, and hence have been shown to be closely related to disease phenotypes of multifactorial diseases such as AMD (Lains et al., 2019; Lains et al., 2018). MetQTLs, defined as genetic variants that correlate with changes in metabolites, can provide insights into the biological relevance of AMD-associated variants. A metQTL analysis was performed on plasma samples of 388 AMD patients and 98 controls from Boston and Coimbra (Lains et al., 2021). This study identified 28 significant metQTLs, which corresponded to 5 metabolites and 2 genes: ASPM and LIPC. The most significant and the largest number of associations were observed for variants in the LIPC gene and phosphatidylethanolamine metabolites, which are glycerophospholipids. These results suggest that this metabolic pathway likely plays a major role in AMD and was supported by work of the EYE-RISK Consortium (Acar et al., 2020).

The study was expanded by conducting metabolite GWAS (mGWAS) using multi-ethnic genetic and metabolomics data from up to 28,000 participants from publicly available cohorts (Nurses Health Study, Nurses Health Study II, Health Professionals Follow Up Study, Canadian Longitudinal Study on Aging, Finnish Men’s Health Study, Hispanic Community Health Study/Study of Latinos) and the Boston/Coimbra cohort (Han et al., 2023). Using bidirectional Mendelian Randomization analysis, this study identified 108 putative causal relationships between plasma metabolites and advanced AMD. The identified metabolites were enriched in the pathways of glycerophospholipid metabolism, lysophospholipid, triradylcglycerol, and long chain polyunsaturated fatty acids. Additionally, a Bayesian genetic colocalization analysis was performed with a customized metabolome-wide association pipeline (Han & Liang, 2022). to prioritize putative causal AMD-associated metabolites, and identified 249 shared common causal variants for metabolites and AMD. The top five colocalized genomic regions were located in or near APOE, C3, CFI, ABCA1, and LOC101929452 - LOC100506274. In the same work, limited evidence was found linking urine metabolites to AMD risk. This study emphasized the contribution of plasma metabolites, particularly in lipid-related pathways, and genes, to AMD risk (Han et al., 2023).

In addition to genetic risk factors, lifestyle can also have significant impact on the metabolite level that could be potentially leveraged to monitor disease progression in AMD (Thee et al., 2023). Future research should integrate factors such as lifestyle and rare genetic variants to achieve a more complete understanding of metabolite roles in disease progression as well as to develop reliable biomarkers for AMD. In addition, metabolomics and metQTL studies should preferably be extended to eye samples, since the relationship between systemic and ocular metabolites is not well-understood. As an alternative approach to local eye tissue samples, metabolic alterations can be studied in iPSC-derived AMD-relevant cell types (described in section 5.2), though these cannot recapitulate the disease environment.

4. Assessing the effect of non-coding variants on regulation of effector genes

Epigenetic changes, encompassing DNA methylation, histone modification, non-coding RNA, and posttranslational modifications, in addition to chromatin accessibility, 3D genome contacts, and protein-DNA interactions are fundamental regulators of gene expression in both health and disease. This section critically examines studies that elucidate these epigenetic and gene regulatory mechanisms, particularly in relation to AMD-associated loci and variants. By exploring how epigenetic and gene regulatory alterations influence gene expression dynamics, we aim to provide a comprehensive overview of their role in the context of AMD susceptibility.

4.1. Methylation

DNA methylation is recognized as the most stable epigenetic modification, with patterns that are specific to each cell type, underscoring the importance of studying disease-relevant tissues to elucidate their role in pathophysiology. An integrative analysis of DNA methylation profiles of 160 human donor retina samples with transcriptome and genetic variants was performed to uncover epigenomic regulation of disease pathogenesis. This study identified 37,453 genetic variants regulating methylation (cis-mQTLs) and 13,747 DNA methylation positions regulating gene expression (eQMPs), which were enriched for processes related to synapse, mitochondria, and catabolism (Advani et al., 2024). Further integration of the AMD GWAS data identified 87 target genes where methylation and gene-expression changes likely mediate the genotype effect on AMD.

Genome-wide DNA methylation profiling on RPE cells from 44 human donor eyes (25 AMD and 19 normal controls) identified differential methylation of multiple loci (including the SKI, GTF2H4, and TNXB genes) in AMD, which were further validated in the RPE of independent AMD donors (Porter et al., 2019). The novel differentially methylated genes SKI and GTF2H4 regulate disease pathways implicated in AMD, including TGF-beta signaling (SKI) and transcription-dependent DNA repair mechanisms (GTF2H4). Another study using DNA methylation microarrays on macular RPE/choroid samples of 39 controls, 29 early and intermediate AMD, and 15 geographic atrophy patients did not identify any significant differentially methylated regions (DMRs) but identified 23 differentially methylated positions (DMPs). Annotation of the DMPs based on their physical proximity to nearby genes identified 35 genes, primarily associated with non-AMD GWAS loci (Orozco et al., 2023). These studies support the use of complementary functional annotations for making biological sense of GWAS findings. At the same time, small sample sizes across these studies as well as lack of independent validation warrants functional confirmation of epigenetic association studies.

The difficulty in collecting disease-relevant tissues in a large number of individuals is a significant challenge, prompting studies of peripheral tissues such as blood. Although generalization from peripheral tissue to the retina is not appropriate, disease-associated changes in peripheral DNA methylation might uncover useful biomarkers. Genome-wide DNA methylation profiling of blood from 199 AMD patients (99 ‘wet’, 100 ‘dry’ AMD) and 99 controls did not yield any genome-wide significant findings (Oliver et al., 2015). However, investigations around the known AMD GWAS loci identified a significant correlation of methylation level of one of the CpG sites with the AMD risk variant rs10490924 at the ARMS2/HTRA1 locus.

4.2. Open chromatin

Non-coding variants residing within cis-regulatory elements (CREs) are likely to regulate transcription of neighboring genes. The assay for transposase-accessible chromatin with sequencing (ATAC-seq) has become a widely used proxy for CRE activity due to its straightforward implementation and its ability to represent a universal hallmark across various types of CREs.

A comprehensive analysis of chromatin accessibility using ATAC-seq in retina and RPE of 8 controls and 8 AMD (3 early dry, and 5 late dry) patients revealed overall higher accessibility in the retina than in the RPE, potentially reflecting the much greater diversity of cell types in the retina. Additionally, the analysis showed a widespread decrease in the chromatin accessibility in RPE of patients with early AMD, and in the retina of patients with advanced disease, suggesting the dysfunction in the RPE as the primary driver of the disease onset (Wang et al., 2018). These changes did not coincide more frequently with known AMD GWAS loci than would be expected by chance, suggesting that the observed differences in chromatin accessibility are unlikely to be solely attributed to AMD-associated genetic variants.

Wang et al, generated a single-cell atlas of gene expression and chromatin accessibility of eight adult human retinas, which they used to analyze risk variants associated with AMD and other eye diseases (Wang et al., 2022b). By integrating this atlas with eQTLs (Ratnapriya et al., 2019), they enhanced the functional annotation of rs66475830 on chromosome 6 in the FRK-NT5DC1-COL10A1 risk locus, identifying TSPYL1 and TSPYL4 as target genes (Wang et al., 2022b). In another study, integration of ATAC-seq data at single nucleus level (snATAC-seq) with AMD GWAS loci identified RPE and myeloid cells as the top disease-relevant cell types (Orozco et al., 2023). Correlation between cell type-specific chromatin accessibility and the expression of nearby genes can help prioritize putative regulatory elements and target genes at GWAS loci. However, interpreting these data is not always straightforward. For example, integration of single cell transcriptome and epigenome data across all retinal cell types identified 22 genes at 7 AMD-associated loci, with multiple candidate genes within a single locus (Orozco et al., 2023). These findings further emphasize the need for functional validation of these results with additional approaches such as CRISPR-based approaches and systematic functional characterization of individual ‘putative causal genes’ to further clarify their role in disease processes (Bock et al., 2022).

4.3. 3D genome

Though considerable work has focused on correlating AMD GWAS variants with nearby (cis) genomic annotations such as gene expression (eQTLs), methylation, and open chromatin regions, study of trans regulatory effects is scarce, due mainly to the limited statistical power and the large burden of multiple testing related to detection of trans-eQTLs. Alternatively, distal regulatory mechanisms can be captured by generating high-resolution contact maps for enhancer-promoter interactions. Chromosome conformation capture (3C) techniques can determine the 3D organization of chromosomes and the physical interactions that occur within and between them. Variations of 3C techniques including Hi-C, 4-seq (3C coupled to next-generation sequencing) and microcapture C are powerful techniques to determine 3D chromatin organization.

Hi-C analysis of four human retina samples identified 67,841 significant chromatin contacts and 2948 topologically associated domains (TADs), with genes expressed being separated from nonexpressed genes in two distinct domains (Marchal et al., 2022). Among AMD GWAS variants, 4 lead variants (C20orf85, CTRB2/CTRB1 and KMT2E/SRPK2 loci) and 136 filtered LD variants distributed among 10 loci (COL8A1, PILRB/PILRA, TGFBR1, ARHGAP21, RDH5/CD63, ACAD10, RAD51B, C3 (NRTN/FUT6), APOE (EXOC3L2/MARK4) and SLC1) were in contact with gene bodies or TSS through retina chromatin loops (Marchal et al., 2022).

An added virtue of this approach is the ability to identify trans regulation ie., variant regulating the gene expression of a distant genes or even located on different chromosome. For example, this approach combined with other annotations (e.g. eQTLs and CREs) can help identify high-confidence causal genes in AMD GWAS loci for further functional validation. However, these approaches do not directly establish their functional relationship in gene regulation and further studies to validate their impact on retina-specific gene expression are needed.

4.4. Protein-DNA interactions

Assessing which genetic variants affect protein-DNA interactions can further help understand the functional effects of AMD-associated genetic variants. SNP evaluation by Systematic Evolution of Ligands by EXponential enrichment (SNP-SELEX) can systematically assess the relative binding affinity of human transcription factors to non-coding variants in the human genome (Yan et al., 2021). Compared to the typical SELEX method that often uses randomized DNA sequences as input, SNP-SELEX uses a library of 40-bp DNA sequences from the human genome, with the center position corresponding to the genetic variant with four possible bases. By including the lead variants identified by GWAS in addition to other genetic variants that are in high linkage disequilibrium, the approach can assess the binding affinity of the library of DNA sequences to individual transcription factors (TF). Another approach to systematically examine TF-DNA interactions uses protein microarrays (Hu et al., 2009), which contain thousands of purified full-length human TFs. For each pair of high-risk and low-risk alleles of a GWAS variant, this approach can identify which TFs show differential binding to the two alleles.

To date, no systematic examination of differential protein-DNA interactions for AMD GWAS hits has been performed. However, some individual genetic variants have been analyzed in terms of protein-DNA interactions. One example involves the HTRA1 gene, which harbors a genetic variant (rs11200638) in its promoter region. Dewan et al. used chromatin immunoprecipitation (ChIP) followed by quantitative real-time polymerase chain reaction (qPCR) analysis to identify the transcription factors adaptor-related protein complex 2α (AP2α) and serum response factor (SRF) interacting with the variant (Dewan et al., 2006).

4.5. Massively parallel reporter assay (MPRA)

The massively parallel reporter assay (MPRA) has emerged to meet the need for high-throughput functional testing of putative CREs (Oh and Chen, 2019; White, 2015). MPRA builds on basic reporter assays by pairing each putative CRE with short DNA sequence barcodes that are transcribed alongside the reporter gene (Chang et al., 2023; Kwasnieski et al., 2012; Melnikov et al., 2012). These assays help determine the functionality of DNA elements in regulating reporter gene expression.

In the retina, MPRA has been used to identify enhancer and silencer elements driving gene expression in photoreceptors (Shepherdson et al., 2024). Recently, MPRA has been employed to assess the functionality of CREs linked to inherited retinal disease (IRD) genes identified through single-cell ATAC-seq (scATAC-seq) data (Li et al., 2023). Beyond assessing general enhancer activity of CREs, MPRA can be used to identify critical sequences within an element by comparing the activity of elements with and without genetic variants (Kwasnieski et al., 2012; Sahu et al., 2022; Xiao et al., 2024). Thus, candidate causal non-coding variants associated with AMD identified by GWAS can be tested for their functional activity using MPRA.

One important consideration is that gene regulation is cell-context dependent. MPRA designs should incorporate cell-specific promoters, such as BEST1 for RPE cells (Esumi et al., 2009; Johari et al., 2021) and CRX for photoreceptors (Chen et al., 1997; Hennig et al., 2008). A significant challenge in AMD research is selecting an appropriate model system for MPRA. The macula in the central retina is the primary site affected by AMD (Fleckenstein et al., 2021), but current models such as mouse retina explants, retinal organoids, and iPSC-RPE cells do not model the macula, potentially limiting the accuracy of MPRA in identifying the functional genomics of AMD. Finally, as gene regulation is affected by chromatin states, and given that MPRA is an episomal system, one limitation is that results might not reflect endogenous activity. Other complementary approaches, such as CRISPRi, could be considered. Despite this limitation, MPRA has significant advantages, including scalability and cost-effectiveness.

5. Functionalization of genetic variants using cellular disease models

Understanding the mechanisms through which genetic variants mediate disease is critical for developing preventative measures and effective treatments. Mouse models have been valuable to understand the mechanistic basis of human disease (Perlman, 2016). However, the evolutionary divergence in transcriptional regulation, along with differences in physiology, genetics, and lifespan, complicates the study of the functional impact of genetic variants in human disease genes when relying on mouse as the disease model (Becker et al., 2024; Powell et al., 2022). Additionally, gene expression in humans is influenced by aging, epigenetic changes, environmental factors, or a combination of these factors (Consortium, 2015), which can be challenging to model in mice. Some of the AMD-specific challenges are worth noting, including that vision loss is caused by the degeneration of photoreceptors in the macula, a structure unique to primates. Additionally, knockout of specific genes is one of the most widely used approaches in mouse models to study gene function (Groza et al., 2023). However, the majority of risk variants associated with complex diseases including AMD are not loss-of-function variants. As a result, while these models can offer valuable insights into the role of the gene in disease-relevant pathways, they do not directly provide variant-centric mechanistic understanding. Landowski et al. addressed this challenge by generating a transgenic mice expressing equal amounts of the fulllength normal human CFH Y402 (CFH-Y/0) or the AMD-risk associated CFH H402 (CFH-H/H) variant on a Cfh^−/− background to demonstrate the effect of Y402H polymorphism in promoting AMD-like pathology development and lipoprotein dysregulation (Landowski et al., 2019). However, such examples are rare and lack the regulatory context necessary to study the impact of variants, particularly those with smaller effects on disease risk. In recent years, cellular models have gained popularity as a scalable alternative for disease modeling (Bharti et al., 2022). They offer the advantage of enabling rapid investigations into cellular, transcriptional, and genetic mechanisms, and are of human origin. In this section, we explore the current status and challenges associated with the use of these models with a specific focus on hiPSC-RPE.

5.1. Effect of AMD-associated genetic variants in hiPSC-derived RPE

Human induced pluripotent stem cells (hiPSCs) have emerged as valuable cellular models for disease modeling. HiPSCs provide the unique ability to differentiate into various cell types, allowing the generation of multiple disease-relevant cell types that can be instrumental for studying the underlying mechanisms of various disorders. AMD disease mechanisms have in particular been studied in hiPSC-derived retinal pigment epithelial cells (hiPSC-RPE), which have been shown to display key physical, gene expression and functional properties similar to human RPE cells in vivo (Bharti et al., 2022; Dalvi et al., 2019). The effect of genetic variants on hiPSC-RPE function has been studied by using patient-derived hiPSCs carrying high- or low-risk genotypes underlying AMD, or by gene editing (e.g. CRISPR) of AMD-associated variants.

Several studies have investigated the effect of AMD-associated genetic variants at the CFH locus in iPSC-RPE. Hallam et al compared hiPSC-RPE derived from two AMD patients carrying the CFH rs1061170 (Y402H) high-risk genotype to two subjects without AMD carrying the low-risk genotype (Hallam et al., 2017). The hiPSC-RPE cells derived from high-risk patients mimic several key features of AMD including increased inflammation and cellular stress, accumulation of lipid droplets, impaired autophagy, and deposition of drusen-like deposits (Hallam et al., 2017). Keir et al compared hiPSC-RPE from one subject carrying the CFH rs1061170 (Y402H) high-risk genotype with one subject carrying the low-risk genotype (Keir et al., 2017). hiPSC-RPE of the high-risk carrier showed more cell-surface complement deposition, which was further exacerbated by anti-VEGF therapy (Keir et al., 2017). Sharma et al generated iPSC-RPE for four homozygous carriers of the CFH rs1061170 (Y402H) high-risk genotype, three homozygous carriers of the low-risk genotype, and three heterozygous carriers (Sharma et al., 2021). The authors demonstrated that drusen deposition in vitro is dependent on the high-risk CFH genotype and anaphylatoxin-triggered alternative complement signaling (Sharma et al., 2021). Ebeling et al generated hiPSC-RPE from 16 AMD patients and 10 individuals without AMD, which were genotyped for the CFH rs1061170 (Y402H) variant (Ebeling et al., 2021). HiPSC-RPE derived from AMD patients displayed decreased mitochondrial function under conditions of stress and elevated expression of inflammatory markers compared to hiPSC-RPE from individuals without AMD, and these effects were more pronounced in CFH high-risk carriers (Ebeling et al., 2021).

While the ARMS2/HTRA1 locus shows one of the strongest associations with AMD, the causal gene and variant remain enigmatic (Merle et al., 2023). Three variants at this locus are in tight linkage disequilibrium: the rs10490924 (A69S) variant in ARMS2, an insertion-deletion located in the 3’- UTR of ARMS2 and upstream of HTRA1, and the rs112000638 variant in the HTRA1 promoter. Chang et al performed CRISPR-editing to isolate the individual effects of these three variants in ARMS2 and HTRA1 in an isogenic background (Chang et al., 2023). HiPSC-RPE carrying the ARMS2 rs10490924 (A69S) variant raised oxidative stress in iPSC-RPE, while the other two variants did not affect oxidative stress (Chang et al., 2023). This study pinpoints the ARMS2 A69S variant as the potential causal variant at this locus, and serves as a proof of concept for the use of patient-derived hiPSCs and CRISPR-editing for functional annotation of tightly linked GWAS variants.

Biasella et al investigated the effect of the rs704 (T400M) missense variant in the VTN gene in iPSC-RPE of four individuals: one homozygous high-risk carrier, one homozygous low-risk carrier, and two heterozygous carriers (Biasella et al., 2020). The rs704 affected expression, secretion, and processing of vitronectin and altered functionality in cellular processes related to sub-RPE pathology in AMD (Biasella et al., 2020). Klipfel et al investigated the effect of a rare splice-site variant (c.214+1G>C, rs77968014) in the SLC16A8 gene in iPSC-RPE of one carrier and one-non carrier AMD patient. The authors demonstrated that the SLC16A8 splice-site variant leads to retention of intron 2, subsequent absence of MCT3 protein, and a deficit in transepithelial lactate transport (Biasella et al., 2020). Peters et al investigated the effect of the AMD-associated variants rs1883025 and rs22740488 in the ABCA1 gene in three AMD patients carrying the high-risk ABCA1 genotype and three AMD patients carrying the low-risk genotype (Peters et al., 2022). The risk-conferring alleles were associated with increased ABCA1 mRNA and protein levels and reduced cholesterol efflux from the RPE (Peters et al., 2022).

In summary, patient-derived iPSC-RPE cells have been used study the functional consequences of AMD-associated variants at several AMD loci, and CRISPR-editing can aid in pinpointing causal variants. With the number of iPSC differentiation protocols rapidly expanding, these studies can be expanded to other cell types relevant to AMD expressing the gene of interest.

5.2. Quantifying mitochondrial metabolism in patient-derived iPSC RPE

RPE relies heavily on mitochondrial metabolism to support critical functions including nutrient transport, recycling, phagocytosis, and secretion of cytokines (Hurley, 2021; Zhang et al., 2021). Mitochondrial dysfunction of RPE is an important driver of AMD pathogenesis (Ferrington et al., 2020; Kaarniranta et al., 2020). Many AMD-associated genes are implicated in metabolic pathways, including lipid metabolism (LIPC, CETP, APOE, ABCA1, ACAD10), amino acid metabolism involved in extracellular matrix (ECM) modeling (COL8A1, COL4A3, HTRA1, TIMP3, MMP9), and sugar metabolism (B3GALTL, SLC16A8, ME3). Even common variants in genes such as CFH and ARMS2 influence mitochondrial metabolism (Ferrington et al., 2016; Kanda et al., 2007).

Metabolite analysis and mitochondrial oxygen consumption assays are common methods to assess mitochondrial metabolism. Metabolites, as intermediates in biochemical reactions, reflect the activity of nutrient transporters and metabolic pathways. Mass spectrometry or biochemical assays of medium metabolites offer non-invasive methods to evaluate mitochondrial function in nutrient utilization and transport (Chao et al., 2017; Zhang et al., 2021). Labeling specific nutrients such as glucose and amino acids with stable isotope tracers quantify the flux of the metabolic pathways (Chao et al., 2017; Zhu et al., 2023). Mitochondria consume oxygen to oxidize nutrients and measurement of oxygen consumption rate can assess RPE mitochondrial function and substrate utilization (Chichagova et al., 2017; Hass et al., 2023; Rountree et al., 2016).

Both methods have been used to investigate the functional impact of certain genetic variants. For example, patient-derived iPSC-RPE with ABCA1 risk alleles (rs1883025, rs2740488) have reduced efficiency of cholesterol efflux despite increased ABCA1 mRNA and protein levels (Peters et al., 2022). A splice variant in SLC16A8 results in lactate transport deficits in iPSC-RPE (Klipfel et al., 2021). Additionally, iPSC-RPE derived from AMD donors carrying the CFH high-risk allele (rs1061170, Y402H) display decreased mitochondrial function (Ebeling et al., 2021). Integrating metabolite analysis and mitochondrial oxygen consumption assays will provide robust functional readouts to investigate the effects of genetic variants on RPE metabolism and deeper insights into plasma metabolomics data from AMD patients.

5.3. Scaling of hiPSC models for population studies: village in a dish

Just as large datasets have been required to dissect the genetic architecture of AMD, similarly powered studies will be required to robustly determine the molecular pathogenesis of this complex disease. Linking genetic variation to disease and cellular phenotypes is dramatically enhanced with large sample sizes (Senabouth et al., 2022). However, both cost and logistics of maintaining multiple cell lines are major barriers impeding the large-scale modeling of AMD using hiPSCs. A promising approach to overcome these barriers is the “village in a dish” (Neavin et al., 2023; Wells et al., 2023). Here, patient-specific hiPSCs from multiple donors are pooled, before being differentiated and profiled together (Figure 4), reducing cost and labor requirements for hiPSC profiling, and dramatically reducing the technical variation inherent to cell culture. hiPSC co-culturing reduces stochastic variation between cell cultures (Neavin et al., 2023).

Figure 4. ‘Village in a dish’ - applying large-scale iPSC modeling to study AMD.

Here, human induced pluripotent stem cell (hiPSC)-derived RPE cells from people with or without AMD can be molecularly profiled and characterized. Pooling hiPSC lines from multiple individuals allows for rapid and economical profiling. Individual genetic signatures can be determined using single cell RNA-seq.

An important advance in the “village in a dish” model has been the incorporation of single cell RNA-seq profiling to dissect each individual hiPSC’s contribution to a cellular phenotype (Neavin et al., 2023). The ability to simultaneously map both an individual’s contribution to the village, as well as their cellular profile at scale, opens the exciting prospect for a variety of informative experimental designs. Specifically, this experimental model enables understanding of the precise contribution of genetic variation on cell phenotypes. The village in a dish model can be used for developing robust differentiation protocols (i.e. whereby the propensity for differentiation is confirmed across multiple individuals); large-scale disease modeling (e.g. eQTL and pQTL mapping in disease-specific cell types, as introduced in Section 3); as well as ‘clinical trials in a dish’ (i.e. where cells from multiple donors can be screened for therapeutic or toxic response to novel therapies) (Farbehi et al., 2024).

6. Conclusion and future directions

Significant progress has been made through GWAS in identifying genetic variants associated with AMD, but critical gaps remain in understanding the role of genetic variants in promoting disease onset, progression and phenotype. Most GWAS have predominantly involved individuals from European ancestries, highlighting the urgent need for more diverse and larger cohorts, particularly from African ethnicities, to better understand the global genetic landscape of AMD. Additionally, functional understanding of major AMD loci, including the ARMS2/HTRA1 locus which exhibits the strongest association with AMD, remains limited, requiring a detailed investigation of associated variants and target genes in disease-relevant model systems.

eQTL studies suggest that several AMD-associated variants influence gene expression, making it crucial to understand transcriptome regulation in disease-relevant cells and tissues to interpret these genetic findings. Thus, transitioning from bulk eQTL to single-cell RNA sequencing (scRNA-seq) is essential for identifying cell-type-specific QTLs and gaining a better understanding of cell-type-specific and context-dependent gene regulation in biological processes and disease mechanisms. Of note, a recent study demonstrated that QTLs and GWAS loci often do not overlap because GWAS and eQTL studies tend to focus on different types of variants (Mostafavi et al., 2023), and many disease-associated variants are neither coding nor located in open chromatin regions. These findings highlight the need for complementary functional approaches, such as exploring 3D genome architecture in AMD-relevant cell types with advanced techniques like 4C-seq and microcapture C, protein-DNA interactions, and MPRA.

Finally, hiPSC-derived cellular models offer a promising avenue for modeling AMD and evaluating personalized treatments, but most studies so far have been limited in scale (Chang et al., 2023; Saini et al., 2017; Sharma et al., 2021; Smith et al., 2019). Expanding these models through robotics or co-culturing patient lines in “cell villages” will facilitate population-scale studies and help reduce technical and biological variation. These approaches will enable a comprehensive population-scale assessment of genetic variants on cell function, and may facilitate clinical trials in a dish to assess the effect of drug candidates in a patient population (Zushin et al., 2023).

In conclusion, there is an urgent need for a shift towards functional genomics approaches to capitalize on the success of GWAS studies in AMD. Novel and highly effective functional genomics approaches are available and can enable the identification of causal variants, effector genes, and disease mechanisms in AMD. Functional genomics of AMD GWAS loci is a crucial next step to enable the development of effective treatments for AMD.

Highlights.

• Genome-wide association studies have identified >60 loci associated with AMD

• The vast majority of AMD-associated loci have not yet been assessed in functional studies

• Functional genomic approaches can help unravel causal and protective variants, genes and disease mechanisms

• A shift to functional genomics is needed to capitalize on the success of GWAS studies in AMD