In this study, we analyzed transcriptome profiles of the tissues most affected in AMD, the RPE-choroid and neural retina. Using pairwise statistical analysis in combination with significance score-based clustering, we identified 32 novel gene expression signatures, termed disease modules, associated with major clinical AMD phenotypes as well as two potential pre-AMD phenotypes (Figure 2). Among the 32 disease modules, we discovered 4 disease modules common to multiple disease phenotypes (Figure 2). Next, we showed by SVM analysis that expression levels of the top 20 genes in the RPE-choroid Global Up module (Iowa cohort) can significantly discriminate between clinically documented cases of AMD and normal donors in an independent RPE-choroid dataset (Oregon cohort). Finally, using disease module genes with statistically enriched functional concordance, we assembled detailed interactomes that reveal both global and phenotype-specific processes associated with AMD (Figures 4 and 5). Collectively composed of nearly 200 differentially expressed genes, these interactomes are statistically enriched in genes previously associated with AMD (23 total genes; Figures 4 and 5; Table S2 in Additional file 3), thus validating our analytical strategy and further implicating the previously unexplored network components in AMD pathology.
Our findings reveal that cell-based inflammatory responses within the RPE-choroid are a core feature of AMD. Remarkably, this global process is even detectable in one of the earliest potential stages of the disease prior to perceptible vision loss (MD1; Figures 4b and 5b). Thus, Oregon donors classified by SVM models as AMD (Figure S10a in Additional file 1), but initially labeled 'normal' (Materials and methods), may have been in the beginning stages of AMD. Recently, elevated levels of CXCL10 were reported in the sera and choroid of individuals with AMD , and elevated intraocular CCL2 levels were observed in neovascular AMD . Here we show that all AMD phenotypes in the RPE-choroid are associated with elevated expression of all, or a subset, of the following chemokines: CXCL1, CXCL2, CXCL9, CXCL10, CXCL11, CCL2, and CCL8 (Figure 2a, right). These chemokines are known to recruit macrophages, dendritic cells, granulocytes, CD4+ Th1 cells, CD8+ T cells, and natural killer cells to damaged tissue . Although activated macrophages and other leukocytes are known expressers, most of these chemokines are also expressed in cultures of human fetal RPE following exposure to IL1B . (Notably, IL1B is found in the RPE-choroid Global Up module.) Determining the cellular source(s), context-specific cellular target(s), and regulatory mechanism(s) of AMD-associated chemokines remains an important goal of future work.
A number of other factors in the RPE-choroid Global Up module further implicate a cellular immune response in AMD. For example, CD86 is expressed in drusen-associated dendritic cells , CD69 is expressed in activated leukocytes , and ILI41 and the elastolytic protease CTSL2 are both expressed by macrophages upon activation . In addition, the up-regulation of immunoglobulin genes supports an adaptive, autoimmune response in AMD that is consistent with previous reports of immunoglobulins in drusen and drusen-associated RPE  as well as anti-carboxyethylpyrrole adduct antibodies  and anti-retinal antigen auto-antibodies [131–133] in AMD sera.
We also identified a Retina Global Up module composed of more than 50 genes common to sub-clinical AMD (MD2), Dry AMD (non-GA), neovascular AMD, and GA, but not to donors at risk for AMD with macular hard drusen alone (MD1) with little to no vision loss (Figure 2b, right, and Figure 5b). In addition to wound response genes (for example, TGFB2, CYR61), this module is highly enriched in complement genes previously associated with AMD (C3, C4, C1S, CFI, SERPING1; Table S4 in Additional file 6). Given the prevailing view that the role of complement in AMD is linked to deposition of the terminal complement complex in drusen and the capillary pillars of the choroid [6–8, 10], this finding is unanticipated, and indicates a possible direct role for complement in retinal degeneration. In addition, the up-regulation of both complement and major histocompatibility complex I genes (HLA-A/B/C genes, and B2M) in the Global Up module, along with APOE elevation in the Retina GA Up and CNV Up modules (Figure 5b) may reflect activation of resident microglia in the retina [134, 135]. This is further supported by reports of a similar microglial immunological response in a CX3CR1
-/- mouse model of AMD , as well as in animal models of photo-induced retinal damage [136, 137].
Finally, by integrating functionally enriched gene sets with interactome data, this work reveals many attractive candidates for AMD therapeutics and diagnostics (Figures 4 and 5). Examples of wet AMD candidate targets within the RPE-choroid interactome (Figure 4) include IGF2, an imprinted gene adjacent to H19 (also in Figure 4) whose product up-regulates VEGF expression , CYR61, a matricellular protein that modulates angiogenesis and apoptosis , and SPP1, a matricellular CFH binding protein that is both a promoter of VEGF-induced endothelial migration and an immunomodulator [140, 141]. GA Up genes with potential relevance for pharmaceutical intervention include EPO, which inhibits oxidative damage-induced apoptosis in cultured RPE  and IL6, a mediator of RPE degeneration . Furthermore, the interactome genes constitute only a small subset of all identified disease module genes with potential utility for AMD therapy. For example, 62 angiogenesis-related genes in the RPE-choroid CNV Up module encode secreted proteins (Table S3 in Additional file 5), any one of which may represent an effective drug target.