HTRA1, an age‐related macular degeneration protease, processes extracellular matrix proteins EFEMP1 and TSP1

Summary High‐temperature requirement protein A1 (HTRA1) is a serine protease secreted by a number of tissues including retinal pigment epithelium (RPE). A promoter variant of the gene encoding HTRA1 is part of a mutant allele that causes increased HTRA1 expression and contributed to age‐related macular degeneration (AMD) in genomewide association studies. AMD is characterized by pathological development of drusen, extracellular deposits of proteins and lipids on the basal side of RPE. The molecular pathogenesis of AMD is not well understood, and understanding dysregulation of the extracellular matrix may be key. We assess the high‐risk genotype at 10q26 by proteomic comparison of protein levels of RPE cells with and without the mutation. We show HTRA1 protein level is increased in high‐risk RPE cells along with several extracellular matrix proteins, including known HTRA1 cleavage targets LTBP‐1 and clusterin. In addition, two novel targets of HTRA1 have been identified: EFEMP1, an extracellular matrix protein mutated in Doyne honeycomb retinal dystrophy, a genetic eye disease similar to AMD, and thrombospondin 1 (TSP1), an inhibitor of angiogenesis. Our data support the role of RPE extracellular deposition with potential effects in compromised barrier to neovascularization in exudative AMD.

with potential effects in compromised barrier to neovascularization in exudative AMD.

K E Y W O R D S
age related macular degeneration, genetics, mass spectrometry, Neurodegenerative diseases

| INTRODUCTION
Age-related diseases, including age-related macular degeneration (AMD), are an enormous burden and a growing public health concern, given the aging population. Molecular pathways underlying AMD have yet to be defined, mainly for two reasons: (i) studies use inappropriate cells. Most use either immature cell types (e.g., immortal cell lines) or cells derived postmortem from patients with late-stage disease. Neither kind of cell can model the important early stages of an age-related disease, and the latter are in limited supply and thus not amenable to high-throughput applications.
Furthermore, neither can be used to study the function of a single gene in isolation. (ii) Most age-related diseases are not amenable to study because they have multiple genetic risk factors, the underlying cellular pathophysiology is poorly understood, and the affected cells are physically inaccessible and not easily studied in vitro.
Retinal pigment epithelial (RPE) dysfunction in age-related macular degeneration (AMD) leads to central vision loss. The early stages of AMD are characterized by yellow deposits of drusen outside of dysfunctional RPE. Increasing size and number of drusen and the associated inflammation eventually lead to retinal degeneration as patient's age. Advanced disease is characterized by either death of macular tissue in general (termed geographic atrophy or dry AMD) or choroidal neovascularization that causes fluid to leak into the macula (termed wet AMD).
Recently, key inroads into finding the cause of AMD have been furnished by genomewide association studies (GWAS) that identified mutations at two genetic loci as risk factors for AMD.
An AMD-associated genetic locus was found at 10q26, where the risk from three mutations is inseparable due to linkage disequilibrium ( Figure 1). The mutations are composed of a coding SNP (rs10490924) that produces the A69S mutation in the putative Age-Related Maculopathy Susceptibility 2 (ARMS2) gene, an insertiondeletion del443ins54 that deletes the polyadenylation signal sequences of the RNA transcript, and the SNP, rs11200638, in the promoter of High Temperature Requirement A Serine Peptidase 1 (HTRA1; Yang et al., 2006). The other major genetic locus linked to AMD is at chromosome 1q31, where a single nucleotide polymorphism (SNP) rs1061170 causes a missense mutation Y402H in complement factor H (CFH; Edwards et al., 2005;Klein et al., 2005;Hageman et al., 2005;Haines et al., 2005). These two genetic loci, 1q31 and 10q26, were the first to be identified in human GWAS, and they confer the most significant genetic risk of AMD alleles. The initial cohort studies showed the risk of developing AMD was 4.6 times that of wild type among people heterozygous and 7.4 times increased for those homozygous for the CFH risk allele (Klein et al., 2005). Those heterozygous for the ARMS2/HTRA1 risk allele were at~2.7 times greater risk of AMD, whereas homozygotes had 8.2 times increased risk (Rivera et al., 2005). Interestingly, a retinal degenerative disease similar to AMD called Doyne honeycomb macular dystrophy was recently discovered to be caused by EFEMP1 coding mutations, and families carrying disease mutations develop drusen and retinal degeneration decades earlier than patients with AMD (Stone et al., 1999). Although there is phenotypic overlap, whether the two diseases are mechanistically linked is not known.
Further molecular studies of these genetic risk factors might provide great insight into the mechanisms of these diseases.
How the HTRA1/ARMS2 risk allele at 10q26 causes AMD remains unclear, but increasing evidence suggests the gene products may also be involved in drusen formation and inflammation in the extracellular space (Iejima, Nakayama & Iwata, 2015;Jones et al., 2011). Whereas the biological relevance of the putative ARMS2 protein is not well understood, more is known about HTRA1. HTRA1 mutations also cause the genetic vascular disease, cerebral autosomal recessive arteriopathy with subcortical infarcts and leukoencephalopathy (CARASIL; Hara et al., 2009). The protein is implicated as a tumor suppressor and shown to play an important role in extracellular matrix remodeling during placental development and arthritis (Chien et al., 2004;Grau et al., 2006;Nie et al., 2006). It cleaves and inhibits TGF beta, and several extracellular matrix proteins, including fibronectin and amyloid precursor protein (Beaufort et al., 2014;Grau et al., 2005;Oka et al., 2004). Also, when HTRA1 is overexpressed in mice, the elastic network of extracellular matrix becomes fragmented between the RPE and the choroid (Vierkotten, Muether & Fauser, 2011).
We have overcome some of the hurdles to studying AMD pathogenesis by developing a method for creating patient-derived RPE. To determine how the HTRA1/ARMS2 risk allele at 10q26 causes AMD, we make use of new technologies in stem cells and protein biology. RPE is generated from patients' own skin cells through induced pluripotent stem (iPS) cells, yielding matched RPE that expresses key molecular markers (Liao et al., 2010). This allows us to study patientspecific genomes. Our previous study used iPS-RPE derived from an AMD patient with the high-risk HTRA1/ARMS2 allele and demonstrated decreased SOD2 activity in high-risk RPE, implying oxidative stress might drive AMD pathogenesis (Yang et al., 2014). To provide further mechanistic insight at the functional level, we compared wild-type RPE to that derived from an AMD patient with the highrisk HTRA1/ARMS2 allele, in a high-throughput, mass spectrometry experiment. As expected, mass spectrometry detected increased HTRA1 expression in the high-risk RPE cells. Surprisingly, we also found increased expression of several components of extracellular matrix proteins that HTRA1 is known to cleave and decreased expression of RNA processing proteins. Our study also identifies EFEMP1 as a cleavage target of HTRA1 implicating a potential link between the two diseases AMD and DHRD.

| Differential expression of proteins in RPE cells carrying the AMD risk allele at 10q26
To assess the functional significance of the high-risk allele composed of mutations in HTRA1/ARMS2 at 10q26, RPE from an AMD patient heterozygous (T-in/del-A; G-wt-G) at the 10q26 locus and a low-risk homozygote (G-wt-G; G-wt-G) control at 10q26 locus was com-  (Table S1). Differentially expressed proteins were analyzed bioinformatically using the PANTHER GO-Slim Overrepresentation Test.

| RPE with the heterozygous (T-in/del-A; G-wt-G) risk 10q26 allele shows increased expression of HTRA1 and extracellular matrix proteins
The 61 proteins with increased expression in high-risk cells were found to have overrepresentation of the gene ontology terms "extracellular matrix structural constituent" (41-fold, p = 5.60 9 10 À4 ), "actin binding" (8.98-fold; p = .0437), "structural constituent of cytoskeleton" (5.35-fold; p = 2.87 9 10 À3 ), and "structural molecule activity" (4.8-fold; 1.67 9 10 À4 ) in the Molecular Function classification; "protein folding" (12.22; p = .0144) in the Biological Function classification; and "extracellular matrix" (8.59-fold; p = .0183) in the Cellular Component classification. Actin-binding proteins were predominantly increased in cells from high-risk AMD at the 10q26 locus, along with many extracellular matrix proteins, especially proteins related to fibrillin microfibrils (Table S2). The largest gene ontology group increased in RPE cells with the AMD risk allele at 10q26 was extracellular matrix proteins, which are listed in Table 2.
High-risk RPE cells contained more HTRA1, providing support that the high-risk allele at 10q26 causes HTRA1 to be overexpressed. Also upregulated were several extracellular matrix proteins, as the positive control for HTRA1 cleavage (Figure 3b). Our negative control was WT CFH, which previous reports showed is not cleaved by HTRA1 (Figure 3c). In these assays, recombinant HTRA1 cleaved recombinant EFEMP1, partially after a 90-min incubation and completely after an overnight incubation (Figure 3d). TSP1 was also cleaved after an overnight incubation ( Figure 3e).  traits of in vivo disease pathology. Induced pluripotent stem (iPS) cells offer an additional advantage to study specific genetic risk alleles from afflicted individuals. In this study, we used untargeted proteomics to explore the impact of the mutant allele of the 10q26 locus. We used label-free mass spectrometry to compare the protein expression between iPS-RPE derived from a human subject with a heterozygous mutant allele and a human subject wild type at the 10q26 locus.
Our results showed increased expression of HTRA1 in the highrisk RPE cells, but we did not detect the ARMS2 protein, perhaps due to low expression levels. Interestingly there was a greater than For example, GWAS data also revealed that patients homozygous for both the CFH and ARMS2/HTRA1 risk alleles had over 50 times the risk of developing AMD (Schaumberg, Hankinson, Guo, Rimm & Hunter, 2007). Despite the well-established disease risk conferred by variants at the genetic loci of CFH and HTRA1/ARMS2, the mechanisms linking these gene products to disease are not well understood.
Nevertheless, this association between CFH and AMD implicated the alternative complement pathway in AMD pathogenesis. Complement activation is normally inhibited when CFH binds C3b and Complement Reactive Protein (CRP). Thus, if the CFH Y402H risk mutation lowers its affinity for CRP, then the lack of inhibition might be expected to trigger overactive inflammation in the extracellular space.
The CFH risk allele is associated with increased soft drusen in the eye, and immunohistochemical and proteomic studies of drusen composition revealed several complement proteins in addition to CFH (Crabb et al., 2002;Magnusson et al., 2005;Wang et al., 2010). Several clinical trials have tried using anti-inflammatory agents to target the complement system and systemic inflammation, but these had limited success (Ambati, Atkinson & Gelfand, 2013).
In addition to the extracellular matrix proteins found to be increased in AMD high-risk RPE cells, there was decreased expression of RNA splicing proteins. These proteins have not been observed before in relation to retinal degenerative disease, and while it is possible that other, unidentified, genetic differences between the two cell lines could explain the abundance difference in RNA processing proteins, their importance remains to be identified.
We identified two novel substrates of the HTRA1 protease, EFEMP1 and TSP1. EFEMP1 is an extracellular matrix protein that is a component of fibrillin microfibrils, which are important both for maintaining the structure of the extracellular matrix and for regulating TGF beta activation of angiogenesis. The R345W mutation of EFEMP1 is responsible for Doyne honeycomb retinal dystrophy (DHRD), a retinal degenerative disease clinically similar to AMD that afflicts patients decades earlier than AMD ( Figure 5). Mutant EFEMP1 was shown to be misfolded and retained in the cell rather than secreted into the extracellular matrix in human cell culture (Marmorstein et al., 2002). Mice with EFEMP1 knocked-in exhibited extracellular drusen-like deposits between the RPE and Bruch's membrane characteristic of DHRD and AMD (Marmorstein, McLaughlin, Peachey, Sasaki & Marmorstein, 2007). The sub-RPE deposits were found to contain EFEMP1 and to activate the complement pathway (Fu et al., 2007). Our study showed increased EFEMP1 expression due to HTRA1/ARMS2 mutation and cleavage of EFEMP1 by HTRA1, which potentially links AMD with DHRD given that in both disease models EFEMP1 levels were increased in ways possibly

| Protease cleavage assay
The enzymatic activity of recombinant HTRA1 (Abcam, 134441) was confirmed with a protease activity assay kit (Abcam, 112152) con- For the protease cleavage assays on native protein from primary cell culture, cells were lysed and protein concentrations measured using the BCA protein assay. Proteins were separated by SDS-PAGE (4%-15%; Bio-Rad) and transferred to nitrocellulose (Bio-Rad). After blocking, membranes were incubated in polyclonal rabbit antifibulin 3 Ab ( wrote the manuscript.

CONFLI CT OF INTEREST
The authors declare no competing financial interests.