Patterns of Gene Expression, Splicing, and Allele-Specific Expression Vary among Macular Tissues and Clinical Stages of Age-Related Macular Degeneration

Age-related macular degeneration (AMD) is a leading cause of blindness, and elucidating its underlying disease mechanisms is vital to the development of appropriate therapeutics. We identified differentially expressed genes (DEGs) and differentially spliced genes (DSGs) across the clinical stages of AMD in disease-affected tissue, the macular retina pigment epithelium (RPE)/choroid and the macular neural retina within the same eye. We utilized 27 deeply phenotyped donor eyes (recovered within a 6 h postmortem interval time) from Caucasian donors (60–94 years) using a standardized published protocol. Significant findings were then validated in an independent set of well-characterized donor eyes (n = 85). There was limited overlap between DEGs and DSGs, suggesting distinct mechanisms at play in AMD pathophysiology. A greater number of previously reported AMD loci overlapped with DSGs compared to DEGs between disease states, and no DEG overlap with previously reported loci was found in the macular retina between disease states. Additionally, we explored allele-specific expression (ASE) in coding regions of previously reported AMD risk loci, uncovering a significant imbalance in C3 rs2230199 and CFH rs1061170 in the macular RPE/choroid for normal eyes and intermediate AMD (iAMD), and for CFH rs1061147 in the macular RPE/choroid for normal eyes and iAMD, and separately neovascular AMD (NEO). Only significant DEGs/DSGs from the macular RPE/choroid were found to overlap between disease states. STAT1, validated between the iAMD vs. normal comparison, and AGTPBP1, BBS5, CERKL, FGFBP2, KIFC3, RORα, and ZNF292, validated between the NEO vs. normal comparison, revealed an intricate regulatory network with transcription factors and miRNAs identifying potential upstream and downstream regulators. Findings regarding the complement genes C3 and CFH suggest that coding variants at these loci may influence AMD development via an imbalance of gene expression in a tissue-specific manner. Our study provides crucial insights into the multifaceted genomic underpinnings of AMD (i.e., tissue-specific gene expression changes, potential splice variation, and allelic imbalance), which may open new avenues for AMD diagnostics and therapies specific to iAMD and NEO.


Introduction
Age-related macular degeneration (AMD) is a complex neurodegenerative disease with both intermediate and advanced forms and is the leading cause of visual disability in the aging population.The intermediate form may be a clinical biomarker indicating an increased risk of progression to either of the two advanced forms: neovascular AMD (also referred to as wet AMD), which correlates to rapid vision loss, and geographic atrophy (also referred to as dry AMD), in which it can take longer for vision loss to occur [1,2].In either form, this condition involves the progressive degradation of the macula, leading to central vision loss, which impairs reading, facial recognition, and driving abilities [1].
While there is no cure for AMD, and treatments are limited in their efficacy, the development of anti-vascular endothelial growth factor (VEGF) has helped to mitigate visual loss associated with neovascular AMD, though it cannot fully restore anatomic or visual integrity [3][4][5].While there are no FDA-approved therapies for intermediate AMD (iAMD), an over-the-counter supplementation of antioxidant AREDS2 formula has been demonstrated to modestly reduce the rate of progression to advanced AMD [6].Therefore, our focus in the current study was intermediate AMD (iAMD) and neovascular AMD (NEO) to uncover pathways and mechanisms specific to these AMD subtypes.It is only through understanding disease mechanisms that appropriate therapies can be developed.
A large AMD genome-wide association study (GWAS) identified significant association at 34 loci [7] and directed us to potential pathways underlying disease mechanism(s).Studies that utilize single-cell, single-nuclei and/or bulk RNAseq are agnostic and unbiased approaches to examine gene expression, further adding value by demonstrating whether disease-associated loci are expressed in disease-affected tissue [8][9][10][11][12][13][14][15][16][17][18].Additionally, RNA-Seq can uncover differentially spliced genes and non-coding RNAs [19][20][21][22][23][24][25].The relationship between differential splicing and gene expression is crucial in shaping the proteome diversity observed in cells, with up to 95% of human multi-exon genes estimated to undergo alternative splicing [26].Splicing fine-tunes gene expression via the generation of multiple protein isoforms from a single gene, yet our understanding of how splicing contributes to transcriptome variation is limited.Recent studies have explored how splicing diversity and gene expression vary across human traits, implicating aberrant splicing patterns in disease and illustrating the therapeutic potential of spliceosome-targeted therapies [27,28].Thus, these findings demonstrate the contribution of differential splicing to differential gene expression and emphasize the need to investigate the interplay between these distinct yet interconnected mechanisms.
To date, RNAseq studies in AMD have not evaluated differential gene and splice expression simultaneously in the macular RPE/choroid and macular neural retina within the same well-characterized donor eye across disease states.We focused the present study on tissues specifically affected by AMD, the macula of the retinal pigment epithelium/choroid (RPE)/choroid and the macula of the neural retina, between iAMD, NEO and, separately, normal condition.To address the complexity of a multi-faceted disease like AMD, we utilized a systems biology approach, as previously employed [13,[29][30][31].In addition, motivated by previous studies showing evidence of allele-specific expression (ASE) [32] in genes associated with a risk of autism, stroke progression, Alzheimer disease and cancer [33][34][35][36][37][38], we interrogated the DNA of each donor for previously reported AMD GWAS coding variants [7].This was undertaken to determine whether an imbalance of expression between alleles may underlie phenotypic variation, and hence the pathophysiology of AMD.To our knowledge, this is the first study to assess ASE across the clinical spectrum of AMD at a genome-wide level.Methods for human donor eye collection have been previously described in detail according to a standardized protocol [39], and moreover have been successfully used in several downstream genomics studies [13,[40][41][42].In brief, donor eyes were procured within a 6 h post-mortem interval time.Both eyes from each donor underwent postmortem phenotyping with ocular imaging, including spectral domain optical coherence tomography (SD-OCT) and color fundus photography, as published.Retinal pigment epithelium/choroid was immediately dissected from the overlying retina, and the macula separated from the periphery using an 8 mm macular punch.For both peripheral and macular tissues, RPE/choroid was separated from the overlying retinal tissue using microdissection; tissue planes were optimized to minimize retinal contamination of RPE/choroid samples, and a subsequent 6 mm RPE/choroid tissue punch was taken from the 8 mm punch.For this experiment, retina and/or RPE/choroid tissues were placed in RNAlater (Ambion), an RNA stabilizing reagent stored as previously described [13] In brief, AMD phenotyping employed the modified Age-Related Eye Disease Study severity grading scale, where AREDS category 0/1 was considered normal, AREDS category 3 was intermediate AMD (iAMD), and AREDS category 4b was neovascular AMD [43].Phenotype analysis was performed as described [39], by a team of four retinal specialists and ophthalmologists at the University of Utah School of Medicine, Moran Eye Center, and the Massachusetts Eye and Ear Infirmary Retina Service.The agreement of all four specialists upon independent review of the color fundus and SD-OCT imaging was deemed diagnostic; discrepancies were resolved by a collaboration between a minimum of three specialists to ensure a robust and rigorous phenotypic analysis.Only one eye per donor was analyzed for further study.In the case of discordant phenotypes within the same donor, the more severe diseased eye was used for inclusion in the study.For example, if a patient had a diagnosis of AREDS 3 (iAMD) in one eye and AREDS 0/1 in the contralateral eye, only the AREDS 3 (iAMD) eye was used in the study.Similarly, if a patient had a diagnosis of AREDS 3 in one eye and neovascular AMD in the contralateral eye, only the neovascular eye was used in the study.In the case where donor eyes were concordant for disease status, one randomly selected eye per donor was chosen.Although AREDS category 2 (early AMD), category 4a (geographic atrophy) and AREDS category 4c (both geographic atrophy and neovascular AMD) were collected, they were not included in this study.Resultant transcriptomic and epigenomic data from these well-characterized donor eye phenotypes have been previously published [13].

Nucleic Acid Extraction and RNA-Sequencing
DNA and RNA were extracted from macular retina and macular RPE/choroid tissues using the Qiagen All-prep DNA/RNA mini kit (cat #80204) according to the manufacturer's protocol from a total of 27 donors; 12 AREDS 0/1, 10 AREDS 3; 5 4b (neovascular) (a total of 54 samples).The quality of RNA samples was assessed with an Agilent Bioanalyzer.Total RNA samples were poly-A selected and cDNA libraries were constructed using the Illumina TruSeq Stranded mRNA Sample Preparation Kit (cat# RS-122-2101, RS-122-2102) according to the manufacturer's protocol.Sequencing libraries (18 pM) were chemically denatured and applied to an Illumina TruSeq v3 single read flow cell using an Illumina cBot.Hybridized molecules were clonally amplified and annealed to sequencing primers with reagents from an Illumina pTruSeq SR Cluster Kit v3-cBot-HS (GD-401-3001).Following transfer of the flowcell to an Illumina HiSeq instrument (HCS v2.0.12 and RTA v1.17.21.3), a multiplexed, 50 cycle single read sequence run was performed using TruSeq SBS v3 sequencing reagents (FC-401-3002).

Primary Processing of RNA Sequencing Data
Each of the 54 samples (50 bp, poly-A selected, non-stranded, Illumina HiSeq) from the fastq datasets were processed as follows: reads were aligned using NovoCraft's novoalign 2.08.03 software (http://www.novocraft.com/),accessed 15 July 2017, with default settings plus the -o SAM -r All 50 options to output multiple repeat matches.The genome index used contained human hg19 chromosomes, phiX (an internal control), and all known and theoretical splice junctions based on Ensembl transcript annotations.Additional details for this aspect of the protocol are described elsewhere (http://useq.sourceforge.net/usageRNA-Seq.html),accessed 15 July 2017.
Estimates of sample quality were determined by running the Picard CollectRNA-SeqMetrics application (http://broadinstitute.github.io/picard/),accessed 15 July 2017, on each sample.These QC metrics were then merged into one spreadsheet to identify potential outliers.Agilent Bioanalyzer RNA integrity number (RIN) and library input concentration columns were similarly added for QC purposes (http://www.genomics.agilent.com),accessed 15 July 2017.

Differential Gene Expression of Poly A Tail Sequencing and Splicing Analysis of Poly A Tail
Sample sets were analyzed using the DefinedRegionDifferentialSeq (DRDS) utility of USeq to detect differentially expressed and differentially spliced genes.This application accepts as input a conditions directory containing folders with biological replicates for each macular sample type (Normal Retina, Intermediate AMD Retina, Neovascular AMD Retina, Normal RPE/choroid, Intermediate AMD RPE/choroid, and Neovascular AMD RPE/choroid) and an Ensemble gene table in UCSC refFlat format.Gene models were created by merging gene transcripts into a single composite "gene" with the USeq MergeUCSCGeneTable utility.A table containing alignment counts from each sample for each gene was created with DRDS.Data in this table provided the basis for estimating count-based differential abundance using the DESeq2 Bioconductor package (http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html),accessed 15 July 2017 and 23 August 2023 [44].This program estimates the over-dispersion in the count data and calculates adjusted p-values using a negative binomial test.Benjamini-Hochberg p-value correction was applied to our adjusted p-values to control for multiple testing.Fold changes for differences in gene expression indicated the degree of change between conditions.DESeq2 also generates a log2 ratio estimate of difference in gene abundance using variance corrected counts as well as rLog values for clustering and principal component analysis (PCA).Library size and within-replica variance were estimated for each sample.Pairwise comparisons were made between the normal and disease subgroups.Differences in splicing were assessed for merged replica counts for each exon with ≥10 counts in each gene in each subgroup using a chi-square test.A Bonferroni multiple testing correction was applied and the exon with the biggest absolute log2 normalized gene count ratio was noted.Similarly, an adjusted p-value was calculated and fold changes for differences in splicing indicated the degree of potential splicing difference between conditions.A per-base normalized gene count read coverage log2 ratio graph was created, enabling the visualization of the relative exon coverage difference for each pairwise comparison.To identify potential outlier samples, unsupervised hierarchical clustering (HC) and PCA were performed with the aid of the Partek Genomic Suite (http://www.partek.com/pgs),accessed 15 July 2017 and 23 August, 2023, using the default settings.DESeq2 rLog values from genes with ≥20 counts were included in this pipeline.For HC visualization, row values were mean-centered at zero and scaled to a standard deviation of one.In addition to further demonstrating the quality of our data, violin plots of log 10 -transformed FPKM val-ues were generated with the ggplot2 package (https://ggplot2.tidyverse.org/),accessed 11 September 2023 [45].To display differential gene and splice expression, volcano plots were produced using freely accessible software (https://huygens.science.uva.nl/VolcaNoseR),accessed 19 May 2023 [46].
We used our previously published bulk RNAseq dataset from 85 donor eyes as our validation dataset [13].Differential gene expression was performed as described above for DESeq2 and using the limma/voom package in R, as previously described [13,47].Briefly, for the limma/voom package in R, normalization was carried out using TMM, controlling for age and sex in the analysis, and p-values were adjusted with Benjamini-Hochberg correction [47][48][49].Genes were considered significant if they had an adjusted p-value less than 0.05 and a fold change ≥ 1.5 in either direction.

Bioinformatic Analysis
Gene Set Enrichment Analysis (GSEA v.4.3.2) software was utilized to profile our expression dataset (UC San Diego and Broad Institute, https://www.gsea-msigdb.org/gsea/index.jsp), accessed 6 September 2023 [50,51].A gct file was created with our normalized FPKM count data, along with a phenotype cls file for input.The gmt file "h.all.v2023.1.Hs.symbols" and the chip file "Human_Ensembl_Gene_ID_MSigDB.v2023.1.Hs" were used in parallel with our input.The resulting output represented hallmarks or gene sets found to be enriched in our dataset.Based on the direction of our comparison (i.e., iAMD vs. Normal, NEO vs. Normal, etc.), genes that were higher-ranking or more associated with the phenotype on the left contributed positively to the enrichment score (ES) and the lower-ranking genes or genes associated to a lesser extent with the phenotype contributed negatively to the ES.This enrichment score was then normalized based on the variation in our gene set, as it was by others [51], giving us our normalized enrichment score (NES).
Next, we examined genes previously demonstrated to be associated with AMD in candidate gene studies and/or GWAS conducted by the International AMD Genetics Consortium (IAMDGC) and Gorman et al. (2022) [7,29,30,[52][53][54][55][56][57][58][59] for overlap with our DEGs or DSGs.Genes that overlapped (e.g., were both differentially spliced and differentially expressed) were then validated for expression between the same tissue type and disease comparison in a different macular bulk RNAseq dataset from our lab [13].We further explored how the validated genes we identified may be regulated via the UCSC genome browser [60].
QIAGEN Ingenuity Pathway Analysis (IPA) (QIAGEN Inc., Redwood City, CA 94063, United States, (https://digitalinsights.qiagen.com/IPA),software accessed on 8 October 2023, was utilized, as previously described, for the functional analysis of our validated genes (significant differential expression, significant differential splice expression, and significant in a second RNAseq dataset) [30,61].An IPA-generated network was created to visualize interconnectedness between our validated genes.This network was overlaid with our differential expression data and differential splice data separately.

Allele-Specific Expression (ASE)
SNPs previously identified by GWAS as being associated with AMD (determined using the GWAS Catalog, accessed 15 July 2017, https://www.ebi.ac.uk/gwas/) [7] were investigated for allele-specific expression (ASE) in our dataset.Specifically, we genotyped the exonic AMD SNPs using either the genotypes from the HumanOmni2.5-8BeadChip Kit or TaqMan assays.Bam files of individuals showing a heterozygous genotype were examined to determine the number of reads for each of the two alleles.Genotypes of heterozygotes determined from the SNP Chip showing monoallelic expression were confirmed using proxies (r 2 ≥ 0.8), as determined by the 1000 Genomes phase 3 CEU reference panel.
Only individuals with ≥10 reads were used.A binomial test, corrected using Benjamini-Hochberg, was used to determine statistically significant allelic imbalance within each individual [63].

Differential Expression Validation with Real-Time PCR
RNA was reverse-transcribed using oligo-dT primers (Invitrogen, 5781 Van Allen Way Carlsbad, CA 92008, United States) and SuperScript III reverse transcriptase (Invitrogen, 5781 Van Allen Way Carlsbad, CA 92008, United States), according to the manufacturer's protocol.Then, cDNA was used as a template for real-time PCR reactions, and run in triplicate using pre-designed Taqman Gene Expression Assays (Life Technologies, 5781 Van Allen Way Carlsbad, CA 92008, United States) for UCHL1, PFKP, LPCAT1, PDPN, GAS1, and CST3, and for UBC as an endogenous control.Assays were run on the Taqman 7500 Real Time PCR system (Life technologies).Mean Ct values were normalized to UBC and analyzed using the 2 -∆∆CT method, as previously described [64].

Results
Numbers and characteristics of subjects available for the analysis of each tissue type after QC are shown in Table 1.For clustering analysis, using both hierarchal clustering (HCA) and principal component analysis (PCA) based on the samples' whole transcriptome expression, samples were split into two primary groups, comprising retina and RPE/choroid samples.A clear separation of macular RPE/choroid and macular retina tissue types was observed.Hierarchal clustering demonstrated greater variability among the macular RPE/choroid samples than for the macular retina samples.Log 10 -transformed FPKM values were plotted and demonstrated no significant difference in our overall count data between groups (Figure 1).Out of the 54 samples, 47 samples passed QC analysis (Table 1).To evaluate the quality of our tissue dissection, we calculated the number of reads mapped to genes known to be expressed exclusively in the neural retina and RPE/choroid, respectively, using an approach as previously described for the retina [65].Retina genes involved in phototransduction (GNGT1, GUCA1A, PDE6A, GNB1, CNGB1, GNAT1, CNGA1, PDE6B, PDE6G, PRPH2, RHO, ROM1, SAG, and SLC24A1) accounted for an average of 2.3% of reads in the total only in the normal retina library, and accounted for only 0.06% of our normal RPE/choroid tissue reads, proportions which are similar to those reported in a previous study [66].In our study, RPE/choroidal genes (BEST1, RDH5, and RPE6) accounted for an average of 0.65% of reads in the total RPE/choroid library and To evaluate the quality of our tissue dissection, we calculated the number of reads mapped to genes known to be expressed exclusively in the neural retina and RPE/choroid, respectively, using an approach as previously described for the retina [65].Retina genes involved in phototransduction (GNGT1, GUCA1A, PDE6A, GNB1, CNGB1, GNAT1, CNGA1, PDE6B, PDE6G, PRPH2, RHO, ROM1, SAG, and SLC24A1) accounted for an average of 2.3% of reads in the total only in the normal retina library, and accounted for only 0.06% of our normal RPE/choroid tissue reads, proportions which are similar to those reported in a previous study [66].In our study, RPE/choroidal genes (BEST1, RDH5, and RPE65) accounted for an average of 0.65% of reads in the total RPE/choroid library and only 0.02% of total reads in the retina library.These findings demonstrate that neither the macular retina nor the macular RPE/choroid was relevantly contaminated (e.g., if there was contamination of the retina genes in the RPE/choroid library, reads would be greater than 1% compared to the observed proportion of 0.06%).In addition, we plotted our log 10 -transformed FPKM values and showed a similar distribution across our sample conditions, illustrating that our expression results were not due to sample variability.

Gene Expression Differences
A total of 26,650 genes were expressed in the macula RPE/choroid and/or macula retina.Within phenotyped normal eyes, 16,638 genes showed significant (FDR ≤ 0.05) differential expression between macular RPE/choroid and macular retina tissues with a minimum fold change ≥ |1.5|.As illustrated in Figure 2, within macular RPE/choroid tissues, significant differential expression was observed for 40 genes between iAMD and normal eyes, 1204 genes between NEO and normal eyes, and 1194 genes between iAMD and NEO eyes (Figure 2A-C, Table S1).Within macular retina tissues, 30 genes were differentially expressed between iAMD and normal eyes, 41 genes were differentially expressed between NEO and normal eyes, and 50 genes were differentially expressed between iAMD and NEO donor eyes (Figure 2D-F).Of these differentially expressed genes in the macula RPE/choroid, 29 were unique to iAMD vs. normal, 285 were unique to NEO vs. normal, and 276 were unique to iAMD vs. NEO (Table S1).Of the 40 significant DEGs in the iAMD vs. normal comparison of the macular RPE/choroid and the 1204 significant DEGs in the NEO vs. normal comparison of the macular RPE/choroid, only six genes (MTRN2L1, CLEC2L, CCM2L, CYP4X1, GLDN, and SMAD7) were found to overlap (Figure 3A).However, none of the above genes were found to be statistically significant in the iAMD vs. NEO comparison of the macular RPE/choroid.Genes that were unique to the macular retina in iAMD vs. normal (n = 27), NEO vs. normal (n = 38), and iAMD vs. NEO had not been previously associated with AMD (Table S1).Of the 30 significant DEGs in iAMD vs. normal of the macular retina and the 41 significant DEGs in neovascular AMD vs. normal of the macular retina, only two genes (FRG1 and CERKL) were found to overlap (Figure 3B).However, none of the above genes were found to be statistically significant in iAMD vs. NEO in the macular retina.
Of note, only one gene, mitochondrial-derived peptide humanin MTRNR2L1 [67], overlapped between any RPE/choroid and retina disease comparisons (Table S1).MTRNR2L1 was found to be differentially expressed in iAMD vs. normal for both the macular neural retina and macular RPE/choroid.
A total of nine unique microRNAs (miRNAs) were identified (MIR146A, MIR3918, MIR4657, MIR17HG, MIR3620, MIR3064, MIR197, MIR4680, and MIR4647) across all disease comparisons.Of these miRNAs, six (MIR4657, MIR17HG, MIR3620, MIR197, MIR3064, and MIR3918) were found to be differentially expressed in iAMD vs. NEO within the macular RPE/choroid, while three (MIR146A, MIR197, and MIR3918) were identified in NEO vs. normal in the macular RPE/choroid (Table S1).In the macular retina, two miRNAs were observed: MIR4680, in neovascular AMD vs. normal, and MIR4647 in intermediate AMD vs. NEO (Table S1).Also noteworthily, a unique lncRNA (AC000124.1)was downregulated in iAMD compared to NEO RPE/choroid, while PIWL1 was upregulated in NEO compared to normal macular RPE/choroid (Table S1).Genes that were unique to the macular retina in iAMD vs. normal (n = 27), NEO vs. normal (n = 38), and iAMD vs. NEO had not been previously associated with AMD (Table S1).Of the 30 significant DEGs in iAMD vs. normal of the macular retina and the 41 significant DEGs in neovascular AMD vs. normal of the macular retina, only two genes (FRG1 and CERKL) were found to overlap (Figure 3B).However, none of the above genes were found to be statistically significant in iAMD vs. NEO in the macular retina.
Of note, only one gene, mitochondrial-derived peptide humanin MTRNR2L1 [67], overlapped between any RPE/choroid and retina disease comparisons (Table S1).MTRNR2L1 was found to be differentially expressed in iAMD vs. normal for both the macular neural retina and macular RPE/choroid.

Gene Splicing Differences
Similar to the DEG results, the highest number of DSGs in our RNAseq data was observed in NEO vs. normal, with 1154 significant DSGs in the macular retina and 629 in the macular RPE/choroid (Figure 4B,E).Similar to the DEG analysis, fewer DSGs were observed for comparisons of iAMD vs. NEO (810 in the macular retina, 608 in macular RPE/choroid; Figure 4C,F).The lowest number of DSGs was identified in the iAMD vs. normal comparison (210 in the macular retina, 177 in the macular RPE/choroid; Figure 4A,D).When comparing DSGs between the macular retina and the macular RPE/choroid for each disease comparison, there were 13 DSGs that overlapped between tissue types in iAMD vs. normal, 152 DSGs overlapping between NEO vs. normal, and 102 DSGs overlapping between iAMD vs. NEO (Table S2).

Gene Splicing Differences
Resembling the DEG results, the highest number of DSGs in our RNAseq data was observed in NEO vs. normal, with 1154 significant DSGs in the macular retina and 629 in the macular RPE/choroid (Figure 4B,E).Similar to the DEG analysis, fewer DSGs were observed for comparisons of iAMD vs. NEO (810 in the macular retina, 608 in macular RPE/choroid; Figure 4C,F).The lowest number of DSGs was identified in the iAMD vs. normal comparison (210 in the macular retina, 177 in the macular RPE/choroid; Figure 4A,D).When comparing DSGs between the macular retina and the macular RPE/choroid for each disease comparison, there were 13 DSGs that overlapped between tissue types in iAMD vs. normal, 152 DSGs overlapping between NEO vs. normal, and 102 DSGs overlapping between iAMD vs. NEO (Table S2).For the macula RPE/choroid, there were 26 significant DSGs overlapping all three disease state comparisons in the RPE/choroid and 56 DSGs overlapping between iAMD vs. normal and NEO vs. normal.Of these 56 genes, 8 genes, CCPG1, GALNT15, PLEKHA, RGS20, TMEM14B, ULK4, VPS13C, and VPS37A, were not regulated in the same direction, leaving us with 48 (identically regulated) overlapping genes (Figure 3C).When examining DSGs unique to each disease state comparison, there were 94 in the intermediate AMD vs. normal macula RPE/choroid, 247 in iAMD vs. NEO in RPE/choroid, and 263 in NEO vs. normal in macula RPE/choroid (Table S2).When evaluating DSGs in the macular retina, 68 genes overlapped between iAMD vs. normal and NEO vs. normal.Of these 68 genes, 9 genes, CCT2, CNOT2, EXOC3, GPATCH2, HDAC9, KIAA1841, RPF2, and SNHG14, were For the macula RPE/choroid, there were 26 significant DSGs overlapping all three disease state comparisons in the RPE/choroid and 56 DSGs overlapping between iAMD vs. normal and NEO vs. normal.Of these 56 genes, 8 genes, CCPG1, GALNT15, PLEKHA, RGS20, TMEM14B, ULK4, VPS13C, and VPS37A, were not regulated in the same direction, this resulted in 48 (identically regulated) overlapping genes (Figure 3C).When examining DSGs unique to each disease state comparison, there were 94 in intermediate AMD vs. normal macula RPE/choroid, 247 in iAMD vs. NEO in RPE/choroid, and 263 in NEO vs. normal in macula RPE/choroid (Table S2).When evaluating DSGs in the macular retina, 68 genes overlapped between iAMD vs. normal and NEO vs. normal.Of these 68 genes, 9 genes, CCT2, CNOT2, EXOC3, GPATCH2, HDAC9, KIAA1841, RPF2, and SNHG14, were not regulated in the same direction, this resulted in 59 (identically regulated) overlapping genes (Figure 3D).There were 115 genes unique to iAMD vs. normal, 645 to NEO vs. normal comparison, and 329 unique to iAMD vs. NEO (Table S2).

Analysis of DEGs and DSGs for Overlap with Genes Previously Associated with AMD
DEGs and DSGs: Normal Macular RPE/Choroid vs. Normal Macular Retina Genes demonstrated to be associated with AMD risk in in prior candidate or GWAS studies were examined [7,29,30,52-59] in our normal tissue, comparing the macular RPE/choroid to the macular retina, to characterize the transcriptomic landscape at a baseline state in the tissue affected by disease.A total of 94 DEGs out of 115 of the previously identified AMD loci were found to have statistically significant differential  A-C) show the significant hallmarks (p-value < 0.05, FDR q-value ≤ 0.05) identified in the macular RPE/choroid across disease state.(D,E) show the significant hallmarks (p-value < 0.05, FDR q-value ≤ 0.05) identified in the macular retina across disease state.Please note: no hallmark was identified to be significantly upregulated in neovascular AMD for the macular retina.Abbreviations: GSEA, gene set enrichment analysis, AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Overlap of Differentially Expressed Genes and Differentially Spliced Genes
Utilizing our systems biology approach to drill further down to the disease mechanism in AMD pathophysiology, we looked for overlap in our DEGs and DSGs in each disease comparison and tissue type.Only a small proportion of the overall DSGs in the macular RPE/choroid were also found to be DEGs in each disease state comparison: 6 significant genes in iAMD vs. normal macular RPE/choroid; 162 NEO vs. normal macular RPE/choroid; and 137 iAMD vs. NEO macular RPE/choroid.Of the DSGs in the macular RPE/choroid, 97 overlapped between NEO vs. normal and iAMD vs. NEO RPE/choroid comparisons (Table S3).When comparing DSGs to the DEGs in the retina, there were no overlapping DSGs and DEGs.

Validation of Overlapping DEGs and DSGs through Bulk RNAseq
To validate our findings from the overlapping DEG/DSGs, we utilized our previously published bulk macula RNAseq dataset of well-characterized donor eye tissue [13].As noted in the methods, we reanalyzed this data set so that the same comparisons could be made.We found no statistically significant differentially expressed genes when comparing iAMD vs. NEO in the macula RPE/choroid in our bulk RNAseq dataset [13].Only one DEG/DSG, STAT1, was validated in our iAMD vs. normal macular RPE/choroid comparison (Table 7).Seven DEG/DSGs (AGTPBP1, FGFBP2, CERKL, BBS5, RORα, ZNF292, and KIFC3) were validated in the bulk RNAseq data set in NEO vs. normal macula RPE/choroid.Of these, only gene variants in RORα have been previously associated with AMD risk [7,30,55,56] (Table 7).Additionally, of these validated genes, the proteins of AGTPBP1, CERKL, BBS5, and KIFC3 were expressed in the cytoplasm as opposed to the nucleus.Using the UCSC Genome Browser, we found that all of our validated DEG/DSGs contained numerous transcription-factor binding sites at the splice site coordinates identified.In addition, DEG/DSG ZNF292 splice coordinates overlapped with a transcription start site (TSS).
Table 7. Genes identified as DEGs/DSG and validated in an independent bulk RNAseq dataset.A "+" sign indicates the gene was upregulated, while a "−" indicates the gene was downregulated.An asterisk (*) indicates an opposing log fold change between DEG and DSG expression.When interrogating the seven validated DEG/DSGs (AGTPBP1, BBS5, CERKL, FGFBP2, KIFC3, RORα, and ZNF292) from the NEO vs. normal comparison in Ingenuity Pathway Analysis (IPA), a network was generated of upstream regulators and downstream targets forming possible interconnections between these DEG/DSGs (Figure 7A,B).Eleven miR-NAs were found to form relationships with these genes.Next, STAT1, from our iAMD vs. normal comparison, was added to the network to visualize how it may interact with our seven validated genes (Figure 7C,D).All networks were overlaid with our expression data corresponding to the respective disease state (iAMD or NEO) for DEGs and DSGs (Figure 7A-D).(A,B) show the 7 identified genes from NEO vs. normal; (C,D) show our 7 identified genes from NEO vs. normal combined with STAT1, our validated gene from iAMD vs. normal.Each network is overlaid with expression values/fold changes from either our DEG or DSG dataset for that disease state comparison.Red or green indicates the gene was found in our dataset and associated with increased or decreased measurement, respectively.Orange or blue indicates the gene was not in our dataset but is predicted to be associated with activation or inhibition.Further clarification is provided in the legend.Abbreviations: DEG, differentially expressed gene, DSG, differentially spliced gene.

The Allele-Specific Expression (ASE) of Known AMD-Associated SNPs
According to annotation information for published AMD genome-wide association studies included in the NHGRI-EBI Catalog (https://www.ebi.ac.uk/gwas/home; accessed 28 March 2017), 12 AMD-associated SNPs are located in coding regions of APOE, ARMS2 (1), C2 (1), C3 (2), CFB (1), CFH (4), CFI (1), and PLA2G12A (1), and therefore were investigated for allele-specific expression (Table 8).No heterozygotes were found in our sample for CFH rs121913059, CFI rs141853578, C3 rs147859257, or APOE rs429358.We found no expression of ARMS2 in either the RPE/choroid or neural retina, and therefore we could not investigate the coding SNP rs10490924.For those heterozygotes showing mono-allelic expression (n = 6), CFH rs10754199 was used to confirm heterozygotes for CFH coding SNPs rs10661170 and rs1061147 (r 2 = 1 for rs10754199 and both coding SNPs), CFB rs2242572 was used to confirm heterozygous genotypes for CFB rs641153 (r 2 = 1), and C3 rs1047286 was used to confirm the heterozygous genotype of C3 rs2230199 (r 2 = 0.843).Significant ASE was detected within individuals in four SNPs: CFH rs1061170 (Y402H), CFH rs1061147, CFB rs641153, and C3 rs2230199.Specifically, for CFH rs1061170 we found Each network is overlaid with expression values/fold changes from either our DEG or DSG dataset for that disease state comparison.Red or green indicates the gene was found in our dataset and associated with increased or decreased measurement, respectively.Orange or blue indicates the gene was not in our dataset but is predicted to be associated with activation or inhibition.Further clarification is provided in the legend.Abbreviations: DEG, differentially expressed gene, DSG, differentially spliced gene.

Validation and Replication of RNAseq Findings
We validated our RNA-Seq methodology by choosing genes that varied in fold expression from a range of +20 to −20 (FDR of p < 0.05) between the normal RPE/choroid and retina-UCHL1, PFKP, and LPCAT1 (down-regulated in RPE/choroid vs. retina) and PDPN, GAS1, and CST3 (up-regulated in RPE/choroid vs. retina)-using real-time qPCR reactions run in triplicate on a subset of samples that were used for the RNAseq experiments.We confirmed the direction of effect for five of the six genes examined (Supplementary Table S4).We were unable to detect PFKP expression in all of the RPE/choroid tissue, and therefore this gene could not be validated.Additionally, we were able to replicate all our top 20 genes from the normal RPE/choroid vs. normal retina with the Human Eye Integration data (https://eyeintegration.nei.nih.gov/),accessed 15 July 2017.This database is a collection of healthy human RNAseq datasets generated from various studies of human eye tissue.To the best of our knowledge, no public database is yet available that contains gene expression data of macular retina and macular RPE/choroid tissues from the same donor eyes across the different clinical stages of AMD.

Discussion
In this study, we utilized a global RNAseq approach to investigate gene, splice, and allele-specific expression profiles in the macular retinal pigment epithelium/choroid (RPE/choroid) and the macular retina of post-mortem eyes from individuals with intermediate AMD or neovascular AMD, comparing them to normal age-matched controls.Additionally, this is the first study of its kind to compare macular RPE/choroid and macular retina in this manner within the same deeply phenotyped donor eye (obtained in a post mortem interval time < 6 h).
While it is clear that both the macular RPE/choroid and macular retina are important to AMD pathophysiology, studies have hypothesized that macular RPE/choroid cell function is more significantly related to AMD pathophysiology than retinal cell function [7,13,65,[69][70][71][72].However, there are only a few studies that compare the pathological changes occurring in the macula area (RPE/choroid) to the inner retina [69,73], even though AMD predominantly impacts the macula; thus, the lack of information leaves the relationship between the tissue types at given stages of disease unclear.In this study, at a global RNAseq level, we show the importance of gene expression, splicing, and allele-specific expression in the macular RPE/choroid compared to the macular retina in iAMD and NEO pathophysiology.The bulk RNASeq approach allowed us to identify genes that otherwise would not have been identified via a single-nuclei RNASeq approach, as their expressed transcripts are located in various cellular compartments, including the cytoplasm for validated genes: AGTPBP1, CERKL, BBS5, and KIFC3 (https://www.genecards.org/),accessed 28 August 2023.However, this was also a limitation of our study, as a single-cell approach may have been more appropriate due to the diverse cell types that comprise the macular retina.Therefore, we may have failed to identify significant gene expression changes critical to AMD pathophysiology in the macular retina, as described as by others [15,16,74,75].
As previously demonstrated by others, across human tissues [28] and in neurodegenerative conditions [76,77], there is very little overlap between DEGs and DSGs; thus, these two sets of biological processes appear to operate through distinct mechanisms.This could be due to the fact that the majority of splice isoforms undergo nonsense-mediated decay [78] and do not become functional proteins, which may be the case for some of the differentially spliced genes identified herein.Additionally, techniques used to detect differential splicing compared to differential gene expression have different sets of biases and therefore are inherently noisier compared to overall gene expression changes [79].Our data suggest that AMD's genomic underpinnings are multifaceted and may involve various regulatory mechanisms that require further exploration.
In this study, we only found overlap between DEGs and DSGs in the macular RPE/choroid between any disease state comparison (Table S3).Furthermore, we validated a handful of our DEG/DSGs in an independent bulk RNAseq data set.STAT1 was significantly increased in iAMD compared to normal in our DEGs and an independent bulk RNAseq data set [13], but was significantly decreased in our DSGs (Table 7).Notably, STAT1 was the only validated gene in our iAMD vs. normal comparison, and we hypothesize the difference between its DEG and DSG state may be a compensatory mechanism during the intermediate stage of disease development.Studies have also tied interferon-γ to STAT1 signaling, where interferon-γ has been connected to RPE cell death [80] and shown to negatively regulate HTRA1 expression by activating the p38 MAPK/STAT1 pathway; further it has been shown that STAT1 can bind the HTRA1 promoter [81].Additionally, interferon-γ signaling was also an enriched hallmark in our GSEA analysis.In NEO vs. normal, we validated seven genes (AGTPBP1, BBS5, CERKL, FGFBP2, KIFC3, RORα, and ZNF292) from our DEGs, DSGs, and in an independent bulk-RNAseq dataset (Table 7).Independently, these genes have been linked to AMD (RORα [30,55]), eye disease (BBS5 [82], CERKL [83][84][85], KIFC3 [86,87]) and other immune/neurodegenerative condi-tions (AGTPBP1 [88], FGFBP2 [89,90], KIFC3 [91,92], ZNF292 [93]).As previously reported, RORα has also been shown to interact with the AMRS2/HTRA1 locus [30,55].In addition, BBS genes [94], RORα [95], and the circular RNA of ZNF292 [96] have been demonstrated to interact with the Wnt/β-Catenin signaling pathway, which we found as an enriched hallmark (Figure 5) in our AMD disease states (iAMD and NEO).We previously found RORα to be downregulated in peripheral blood samples from patients with NEO [30], but when investigating the affected tissue herein, we found it to be upregulated in our DEGs and further validated in our independent bulk RNAseq dataset [14] (Table 7).This further highlights the need for additional studies exploring the relationship between biomarkers identified in patient samples of serum/blood and the actions of those markers in diseaseaffected tissue, as we still do not know whether AMD is a local or systemic disease [97,98].As illustrated in our network, an interesting picture emerged when we investigated our seven validated genes in the NEO vs. normal comparison (Figure 7A,B).The bulk of the interconnections between these seven genes were miRNAs, predicted to have an inhibitory effect.Once we added STAT1 (from iAMD vs. normal), this continued to remain the case (Figure 7C,D).Thus, we hypothesize that miRNA degradation may be underlying disease development of AMD and remains an avenue to be explored.
Various factors in addition to gene and splice expression, such as microRNAs (miR-NAs), RNA-binding proteins, and long non-coding RNAs (lncRNAs), may influence transcript stability and modulate translation at a tissue-specific level [98].As discussed above, once we generated a network(s) for our cross validated genes (DEGs, DSGs, and an independent bulk RNAseq dataset), relationships were primarily seen with miRNAs.Thus, miRNAs may represent a potential therapeutic target for diagnosis, prognosis and/or treatment [98,99].MiRNA-146a was downregulated in macular RPE/choroid donor tissues from neovascular AMD subjects compared to controls (Table S1).Previously, miRNA-146a was shown to be upregulated in the serum of patients with neovascular AMD [100][101][102], again underscoring the importance of tissue specificity with gene expression.Targets of miRNA-146a have been implicated in the modulation of the immune response in endothelial tissue, including in the negative regulation of complement factor H [103][104][105].While the involvement of the non-coding genome is under active investigation, miRNAs and other non-coding RNAs (lncRNA-AC000124.1 and PIWL1 in our results-Table S1) have been found to have key roles in cellular homeostasis, with disruption leading to human diseases such as cancer [106].Further studies need to be conducted to fully characterize the role of lncRNAs and miRNAs as biomarkers and determine their potential as therapeutic targets.In this study, we demonstrated that genes previously associated with AMD risk, relevant signaling pathways, and miRNAs and other ncRNAs are expressed differently between tissue types and disease states.
MTRNR2L1 was the only gene found to overlap between the macular retina and macular RPE/choroid in any disease state comparison (found in iAMD vs. normal).While it is known that MTRNR2L1 is a nuclear-encoded humanin isoform gene, its biological function is currently unknown.Recent studies have demonstrated that humanin, a small peptide derived from the mitochondria, can protect the RPE cells against mitochondrial damage induced by oxidative stress and endoplasmic reticulum stress [107,108].The upregulation of MTRNR2L1 in the retina and the RPE/choroid tissues with AMD indicates a potential role of MTRNR2L1 in protecting against retinal and RPE damage during disease development and the progression of AMD.However, the exact role of this gene warrants future investigation.
While we highlight above the need to explore the relationship between genome-wide association studies (GWAS)'s loci and gene expression, determining the causative genes responsible is another matter altogether [13].The challenge arises from a few factors.First, a significant proportion of these SNP associations are found in regions of the genome that are non-coding, known as intra-or intergenic regions [109].Second, each identified association may involve more than one candidate gene [110].Thirdly, gene expression is highly tissueand cell-specific [111], so what is found in these GWAS studies is not necessarily found in the macular RPE/choroid or macular retina.Thus, when considering previously identified AMD risk loci [1,7,29,68,112], it was interesting that 67 of our DEGs (Table 2) showed significantly higher expression in the normal macular RPE/choroid compared to the normal macular retina, underscoring the importance of tissue and geographic location [10,13,15,71].For the DSGs, only a small portion overlapped with previously reported AMD loci, and these were fairly evenly distributed between the normal macular RPE/choroid and normal macular retina in expression differences (Table 3).DEGs that were found to have a higher expression in the normal macular RPE/choroid were found to have a higher expression as DSGs in the normal macular retina (Figure 6).This observed bidirectional change in genes that overlapped in the macula suggests an expression-dependent, homeostatic mechanism in unaffected tissue.We did not find a large portion of the previously identified AMD risk loci differentially expressed or spliced in our disease states of iAMD and NEO when compared to each other, or separately to normal (Tables 4-6).No previously associated AMD genes were found to overlap with our DEGs in the macular retina when considering disease state, but previously associated AMD genes were found to overlap with our DSGs in the macular retina at a higher number compared to the macular RPE/choroid (Table 6).This could be due to a lack of tissue specificity [28] for differential splicing in the macula, which is further supported by our normal macular RPE/choroid vs. normal macular retina DSG findings (where an approximately equivalent expression between macular RPE/choroid and macular retina was observed).Of note, while a prior report demonstrated that a splice variant in TRPM1 was expressed more highly in the retina of late-stage AMD donor eyes and a second splice variant in TRPM1 was expressed in the RPE/choroid of AMD donor eyes [14], we did not find splicing for TRPM1 in the RPE/choroid.We found the DSG for TRPM1 to be significantly down regulated in the macular retina in NEO vs. normal and significantly upregulated in iAMD vs. NEO (Table 6).However, when examining our DEGs, TRPM1 was significantly upregulated in iAMD vs. NEO in macula RPE/choroid (Table 4).
The allele-specific expression (ASE) of CFH demonstrated significant allelic imbalance in both iAMD and NEO depending on the SNP being interrogated (Table 8), although CFH did not demonstrate differential gene expression between disease states.It may be this unequal expression of alleles at a given variant within the CFH gene that contributes to the disease pathophysiology of AMD.The mechanisms that underlie ASE are under active investigation and include epigenetics [15].The evaluation of known coding regions in previously reported GWAS loci demonstrated that significant ASE for C3, rs2230199, and CFH, rs1061170, occurred in the macular RPE/choroid for normal and iAMD, while ASE for CFH, rs1061147, occurred in the macular RPE/choroid for normal and intermediate and neovascular AMD (Table 8).The protective variant for CFB, rs641153, only demonstrated ASE in the normal macular RPE/choroid (Table 8).Findings regarding the complement genes C3 and CFH suggest that coding variants at these loci may influence AMD development via an imbalance in gene expression in a tissue specific manner.A similar circumstance has been noted for the inverse pattern of association of the APOE alleles; the ε4 allele increases the risk of AD and the ε2 allele is protective, whereas the effects of these alleles on AMD risk are the opposite [7,[113][114][115][116].Interestingly, the FDA has approved two inhibitors of complement, pegcetacoplan (C3) and avacincaptad pegol (C5), as the first medications to treat geographic atrophy (dry AMD).
In summary, this RNA-Seq experiment identified novel DEGs/DSGs that may be acting in concert, along other with factors such as ASE and miRNAs, contributing to the development of intermediate and neovascular AMD.It also expanded upon previous gene expression studies that demonstrated differential gene expression in affected tissues.Our results may provide insight into why some, but not all individuals with intermediate AMD develop advanced forms of the disease.Gene expression, along with splicing, may help to refine the pool of candidates for further investigation for therapeutic targets.

Figure 1 .
Figure 1.Violin plot of Log10-transformed FPKM counts from 27 donor eye samples with both the macular RPE/choroid and macular retina shown.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium, FPKM, fragments per kilobase of transcript per million mapped reads.

Figure 1 .
Figure 1.Violin plot of Log 10 -transformed FPKM counts from 27 donor eye samples with both the macular RPE/choroid and macular retina shown.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium, FPKM, fragments per kilobase of transcript per million mapped reads.

Cells 2023 , 30 Figure 2 .
Figure 2. Volcano plots of differentially expressed genes across disease states.(A-F) Each dot represents one of the 26,650 genes expressed.Blue and red represent significant genes, with red indicating upregulation and blue indicating downregulation in each disease comparison.Grey dots represent genes that did not meet the significance threshold of padj < 0.05 and a fold change ≥ |1.5|.The ten most significant genes in each disease comparison are labeled.Abbreviations: AMD, agerelated macular degeneration, RPE, retinal pigment epitheliumOf these differentially expressed genes in the macula RPE/choroid, 29 were unique to iAMD vs. normal, 285 were unique to NEO vs. normal, and 276 were unique to iAMD vs. NEO (TableS1).Of the 40 significant DEGs in the iAMD vs. normal comparison of the macular RPE/choroid and the 1204 significant DEGs in the NEO vs. normal comparison of the macular RPE/choroid, only six genes (MTRN2L1, CLEC2L, CCM2L, CYP4 × 1, GLDN,

Figure 2 .
Figure 2. Volcano plots of differentially expressed genes across disease states.(A-F) Each dot represents one of the 26,650 genes expressed.Blue and red represent significant genes, with red indicating upregulation and blue indicating downregulation in each disease comparison.Grey dots represent genes that did not meet the significance threshold of padj < 0.05 and a fold change ≥ |1.5|.The ten most significant genes in each disease comparison are labeled.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Cells 2023 , 30 Figure 3 .
Figure 3. Overlap of differentially expressed genes (DEGs) and differentially spliced genes (DSGs) between intermediate AMD (iAMD) vs. normal and neovascular AMD (NEO) vs. normal.(A-D) Each circle represents the number of significant DEGs or DSGs in macular RPE (retinal pigment epithelium)/choroid and macular retina.The overlap between these two circles shows the number of overlapping genes that were regulated in the same direction between each comparison.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Figure 3 .
Figure 3. Overlap of differentially expressed genes (DEGs) and differentially spliced genes (DSGs) between intermediate AMD (iAMD) vs. normal and neovascular AMD (NEO) vs. normal.(A-D) Each circle represents the number of significant DEGs or DSGs in macular RPE (retinal pigment epithelium)/choroid and macular retina.The overlap between these two circles shows the number of overlapping genes that were regulated in the same direction between each comparison.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Figure 4 .
Figure 4. Volcano plots of differentially spliced genes across disease states.(A-F) Each dot represents one of the 26,650 genes expressed.Blue and red represent significant genes, with red representing upregulation and blue representing downregulation in each disease comparison.Grey dots represent genes that did not meet the significance threshold of padj < 0.05 and a fold change ≥ |1.5|.The ten most significant genes in each disease comparison are labelled.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Figure 4 .
Figure 4. Volcano plots of differentially spliced genes across disease states.(A-F) Each dot represents one of the 26,650 genes expressed.Blue and red represent significant genes, with red representing upregulation and blue representing downregulation in each disease comparison.Grey dots represent genes that did not meet the significance threshold of padj < 0.05 and a fold change ≥ |1.5|.The ten most significant genes in each disease comparison are labelled.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

3. 4 .
Analysis of DEGs and DSGs for Overlap with Genes Previously Associated with AMD DEGs and DSGs: Normal Macular RPE/Choroid vs. Normal Macular Retina

Figure 5 .
Figure 5. Gene set enrichment analysis (GSEA) using our normalized expression dataset.Thresholds were set based on the nominal p-value < 0.05 and FDR q-value ≤ 0.05 generated by the GSEA software, v.4.3.2.(A-C) show the significant hallmarks (p-value < 0.05, FDR q-value ≤ 0.05) identified in the macular RPE/choroid across disease state.(D,E) show the significant hallmarks (p-value < 0.05, FDR q-value ≤ 0.05) identified in the macular retina across disease state.Please note: no hallmark was identified to be significantly upregulated in neovascular AMD for the macular retina.Abbreviations: GSEA, gene set enrichment analysis, AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Figure 5 .
Figure 5. Gene set enrichment analysis (GSEA) using our normalized expression dataset.Thresholds were set based on the nominal p-value < 0.05 and FDR q-value ≤ 0.05 generated by the GSEA software, v.4.3.2.(A-C) show the significant hallmarks (p-value < 0.05, FDR q-value ≤ 0.05) identified in the macular RPE/choroid across disease state.(D,E) show the significant hallmarks (p-value < 0.05, FDR q-value ≤ 0.05) identified in the macular retina across disease state.Please note: no hallmark was identified to be significantly upregulated in neovascular AMD for the macular retina.Abbreviations: GSEA, gene set enrichment analysis, AMD, age-related macular degeneration, RPE, retinal pigment epithelium.

Figure 7 .
Figure 7. Ingenuity Pathway Analysis (IPA)-generated network of our validated genes (DEGs, DSGs, and confirmed in an independent bulk RNAseq dataset): (A,B) show the 7 identified genes from NEO vs. normal; (C,D) show our 7 identified genes from NEO vs. normal combined with STAT1, our validated gene from iAMD vs. normal.Each network is overlaid with expression values/fold changes from either our DEG or DSG dataset for that disease state comparison.Red or green indicates the gene was found in our dataset and associated with increased or decreased measurement, respectively.Orange or blue indicates the gene was not in our dataset but is predicted to be associated with activation or inhibition.Further clarification is provided in the legend.Abbreviations: DEG, differentially expressed gene, DSG, differentially spliced gene.

Figure 7 .
Figure 7. Ingenuity Pathway Analysis (IPA)-generated network of our validated genes (DEGs, DSGs, and confirmed in an independent bulk RNAseq dataset): (A,B) show the 7 identified genes from NEO vs. normal; (C,D) show our 7 identified genes from NEO vs. normal combined with STAT1, our validated gene from iAMD vs. normal.Each network is overlaid with expression values/fold changes from either our DEG or DSG dataset for that disease state comparison.Red or green indicates the gene was found in our dataset and associated with increased or decreased measurement, respectively.Orange or blue indicates the gene was not in our dataset but is predicted to be associated with activation or inhibition.Further clarification is provided in the legend.Abbreviations: DEG, differentially expressed gene, DSG, differentially spliced gene.

Table 1 .
Subject characteristics of the bulk RNAseq discovery dataset.

Table 2 .
Comparison of previously identified AMD genes to differentially expressed genes (DEGs) in macular RPE/Choroid vs. macular retina of normal tissue.A "+" sign indicates the gene was upregulated, while a "−" indicates the gene was downregulated.

Table 3 .
Comparison of previously identified AMD genes: differentially spliced genes (DSGs) in macular vs. macular retina of normal tissue.A "+" sign indicates the gene was upregulated, while a "−" indicates the gene was downregulated.

Table 3 .
Comparison of previously identified AMD genes: differentially spliced genes (DSGs) in macular RPE/Choroid vs. macular retina of normal tissue.A "+" sign indicates the gene was upregulated, while a "−" indicates the gene was downregulated.Normal Macular RPE/Choroid vs. Normal Macular RetinaFigure 6. Visualization of significant DEGs and DSGs (previously associated with AMD) found in normal macular RPE/choroid vs. normal macular retina to illustrate directionality in normal tissues.Abbreviations: AMD, age-related macular degeneration, RPE, retinal pigment epithelium, DEG, differentially expressed gene, DSG, differentially spliced gene.

Table 5 .
Differentially spliced genes (DSGs) across disease states in the macular RPE/choroid found to be previously associated with AMD.A "+" sign indicates the gene was upregulated, while a "−" indicates the gene was downregulated.An asterisk (*) indicates that the gene is upregulated in the more severe disease state (i.e., intermediate AMD or neovascular AMD).

Table 8 .
Allele-specific expression (ASE) of known AMD-associated SNPs.Only individuals with more than 10 reads were counted, with significant ASE displayed (p < 0.05).