Proteome and Secretome Dynamics of Human Retinal Pigment Epithelium in Response to Reactive Oxygen Species

Age-related macular degeneration (AMD) is the leading cause of blindness in developed countries, and is characterized by slow retinal degeneration linked to chronic reactive oxygen species (ROS) in the retinal pigmented epithelium (RPE). The molecular mechanisms leading to RPE dysfunction in response to ROS are unclear. Here, human stem cell-derived RPE samples were stressed with ROS for 1 or 3 weeks, and both intracellular and secreted proteomes were quantified by mass spectrometry. ROS increased glycolytic proteins but decreased mitochondrial complex I subunits, as well as membrane proteins required for endocytosis. RPE secreted over 1,000 proteins, many of which changed significantly due to ROS. Notably, secreted APOE is decreased 4-fold, and urotensin-II, the strongest known vasoconstrictor, doubled. Furthermore, secreted TGF-beta is increased, and its cognate signaler BMP1 decreased in the secretome. Together, our results paint a detailed molecular picture of the retinal stress response in space and time.


Results
Workflow and general overview. An overview of the experimental strategy is shown in Fig. 1A. RPE derived from hESC were treated with PQ for one or three weeks, and the conditioned media containing secreted proteins was collected to quantify changes in the composition of secreted proteins (secretome, SEC). Parallel experiments were carried out where RPE cells were harvested in order to quantify the intracellular proteome changes. We collected DIA data from each sample and identified and quantified proteins with Spectronaut 29 and the pan-human spectral library 21 . Using this strategy, we identified a total of 3,257 proteins from the intracellular proteome samples, and a total of 1,319 proteins from the secretome samples. Among those proteins, 1,054 proteins were identified in both groups (Fig. 1B). We assessed the quality of our proteome quantification by comparing the coefficients of variation for all identified protein groups within each treatment group (Fig. 1C,D). Based on the proportions of each group that had CV <20% or <10%, the quality of the data from the intracellular proteome quantification was better than the quality of the secretome. This difference in variability between the RPE and SEC experiments is reflected in the number of protein changes we were able to detect from each experiment (Fig. 1E, Supplementary Table 1). For both the RPE and SEC experiments, one-week or three-week PQ treatments were compared to controls. In addition, three-week treatment results were also compared to the one-week treatment data in order to determine which proteins might be involved in the transition from short term to long term ROS adaptation. intracellular proteome changes. Intracellular proteome changes were examined to understand how RPE responds to short term or long term ROS. After one week of PQ treatment, only 128 proteins were altered in abundance. KEGG pathway enrichment analysis of those proteins revealed only a handful of pathways: the citric acid cycle, glycolysis, and a cluster of pathways related to oxidative phosphorylation (OxPhos, Fig. 2A). All of these pathways are related to cellular energy production. Notably, the direction of protein changes in these pathways clearly suggests a decrease in mitochondrial metabolism and an increase in glycolysis. After 3 weeks of ROS stress in RPE, the changes were much more pronounced. Over 600 proteins changed and could be mapped to 20 KEGG pathways (Fig. 2B). Again, most of these pathways were associated with energy production, except for the protein synthesis-related pathway aminoacyl-tRNA biosynthesis, members of which mostly increased.
www.nature.com/scientificreports www.nature.com/scientificreports/ Many pathways related to amino acid metabolism, fatty acid metabolism, and glycolysis were found. After the three-week PQ treatment, changes to the cluster of OxPhos pathways were more pronounced potentially due to activation of cellular compensatory mechanisms to chronic oxidative stress. To better understand how short-term and long-term ROS from PQ treatment were influencing the mitochondrial electron transport chain (ETC), we generated a cartoon representation of ETC complex structures and colored their subunits according to whether they were increased or decreased (Fig. 2C,D). Remarkably, short term ROS caused a concerted decrease in subunits of ETC complex I, as well as a decrease of complex II subunits SDHA and SDHB (Fig. 2C). This result supports reports that complex I is the major site of ROS production by paraquat 30 . Chronic ROS also caused a nearly concerted decrease in ETC complex I subunits; however, concerted increases in subunits of complexes III, IV, and V were now also apparent (Fig. 2D). This data suggests that cells compensate for the reduction in complex I activity by increased translation of complex III-V subunits.
The most-and least-changed intracellular proteins after three weeks of ROS stress do not appear in the pathway enrichment analysis. The most increased proteins were GDF15, PLIN2, and SQSTM1 (Fig. S1A). Known NRF2-regulated proteins NQO1, SRXN1, and HMOX1 are also found in the six most-increased proteins 9,31 . The most-decreased proteins were COL6A1, LRP2, PCBP4, CD74, GLUL, and the complement cascade inhibitor, SERPING1 (Fig. S1B). Human embryonic stem cells (hESC) were differentiated into retinal pigment epithelium (RPE) cells, allowed to mature, and then treated with paraquat (PQ) or vehicle control for one week or three weeks. After treatment, serum-free media was collected to measure the protein secretome (SEC), and cells were harvested to measure the intracellular proteome (RPE). (B) Venn diagram showing the number and overlap of proteins identified in the two datasets. (C) Number of protein groups that meet certain criteria from each condition. Total identified protein groups are in red, the subset of protein groups that were quantified with <20% coefficient of variation (CV) in dark grey, and the subset of protein groups that were quantified with <10% CV are in light grey. (D) The same as (C) for SEC data. (E) Heatmap showing overview of statistically significant changes (q-value <0.01, fold-change >50%) found in each of the comparisons. For the intracellular proteomics experiments, three biological replicates from each treatment timepoint were compared with four control biological replicates (two controls from each timepoint). For the secreted proteome experiments, four biological replicates from each timepoint were compared with eight control biological replicates (four from each timepoint).
www.nature.com/scientificreports www.nature.com/scientificreports/ intracellular adaptation from short-term and long-term RoS. Because proteomic data was collected from both one-week and three-week ROS treatments, the evolution of proteome response to ROS exposure from short-term to long-term can be computed by directly comparing these two time points (3w/1w). In order to understand only the adaptation of the ROS response from short term to long term, any proteins that changed in the 1w/controls or 3w/controls comparisons were excluded, leaving 92 protein changes. KEGG pathway enrichment analysis using this subset of altered proteins found Glyoxylate and dicarboxylate metabolism and Propanoate metabolism pathways (enrichment p-values < 5e-4, Fig. 3A). Only 5 proteins (CAT, GLDC, MUT, ABAT ACSS3) were found in these two pathways, all of which were increased. Wikipathway term enrichment analysis (Fig. 3B) revealed changes to the ETC pathway (enrichment p-value = 4e-7), including SCO1 and SURF1 which assemble cytochrome C oxidase. Finally, the set of altered proteins were checked for enrichment of gene ontology (GO) molecular functions and biological processes (Fig. 3C). Notably, several proteins involved in mitochondrial membrane organization and protein import, as well as several mitochondrial ribosomal proteins translation were increased. Among these proteins, alpha-synuclein and NDUFA7 were the only downregulated members. Proteins with peroxidase activity (Catalase, MGST2 and GPX8) were also upregulated.
In the list of altered proteins there were many interesting changes that were not part of these concerted pathway changes. Of these, two categories jump out: calcium signaling and structural protein changes. Among the www.nature.com/scientificreports www.nature.com/scientificreports/ significantly decreased proteins were structural proteins Zyxin (Zyx, UniProt: Q15942, Fig. 3E) and fibulin 5 (FBLN5, UniProt: Q9UBX5, Fig. 3F). The latter was the most downregulated protein. FBLN5 is especially interesting in the context of AMD because it localizes to Bruch's membrane in the eye 32 and mutations in FBLN5 are known to be associated with AMD 33,34 . Three of the most decreased proteins are involved in calcium signaling, Multiple coagulation factor deficiency protein 2 (binds to calcium ions) 35 , alpha-taxilin (may play a role in calcium-dependent exocytosis) 36 , and calcium/calmodulin-dependent protein kinase kinase 2. Related to calcium signaling, a member of the calcium-activated chloride channels, Anoctamin-10 37 (UniProt: Q9NW15), was increased almost 2-fold (Supplementary Table 1). Finally, in the lists of the most increased and decreased proteins (at least 2-fold change from three-weeks /one-week), several proteins have yet unknown function in the retina, such as CLPTM1, BPIFA3, MAOA, and FAM129B (Fig. 3D).

Secretome remodeling.
Remodeling of the protein secretome in response to short-term and long-term ROS was examined. Protein changes representative of several KEGG pathways changed after one week of PQ treatment, including "complement and coagulation cascades", "protein digestion and absorption", and a cluster related to 'ECM-receptor interaction" and "Focal adhesion" (Fig. 4A). After three weeks of PQ treatment, proteins in the glycolysis and lysosome pathways were also found to increase in the extracellular medium (Fig. 4B).
In common between the one-week and three-week changes were the cell and ECM structure-related pathways of "ECM-receptor interaction", "focal adhesion", "amoebiasis", as well as the "complement and coagulation cascades". Figure 4C shows these common pathways plotted with the protein changes that make up and connect these pathways. Related to the structural protein pathways, we see decreases in several intracellular and extracellular organization proteins, including actin, actinin, thrombospondins 1 and 2, fibronectin, and collagens. The Laminin changes were mixed, with subunits alpha-5 and beta-2 increased, but alpha-4 and beta-1 were decreased. Within the complement cascade several factors were decreased, including: CFI, CFB, C3, C4A, and C4B. Overall, the data indicates significant deregulation of the protein machinery for the complement, coagulation, and organizational systems results from ROS stress.
There are several notable protein changes among the secretome quantification that did not appear in the pathway analysis. Since we noticed an increase in TGF-β1 (Figs 4C, S2Ai) we looked for related extracellular signaling proteins, and found that BMP1 decreased three-to four-fold with short-term and long-term treatment (Fig. S2Aii). TGF-β2 was decreased more subtly around 30% after one or three weeks of stress, but did not reach our arbitrary cutoff for significance (Fig. S2Aiii). Also altered was apolipoprotein E (APOE), which is linked to www.nature.com/scientificreports www.nature.com/scientificreports/ AMD wherein the APOE2 allele increases risk, while APOE4 is protective, in contrast to Alzheimer's disease 38,39 . We observed that APOE was down over 2-fold after 1 week of stress, and nearly 4-fold after 3 weeks of stress (Fig. S2Bi). Another related protein strongly associated with high myopia, amyloid β-like protein 2 (APLP2), was significantly increased after three weeks of stress (Fig. S2Bii). Finally, given that development of AMD is associated with both loss and gain of vasculature, it is notable that urotensin-II (UTS2), the strongest known vasoconstrictor, was more than tripled after one or three weeks of PQ induced stress (Fig. S2Biii).
Given the apparent ECM remodeling we also checked for any changes in secreted proteases. We found a large number of proteases and protease-related proteins altered in the secretome data. We did identify two matrix metalloproteases (MMPs), MMP2 and MMP15, but only MMP2, a type IV collagenase, was significantly decreased (Fig. S2Ci). One of the most upregulated proteases was ADAM9, a known 'sheddase' protease that cleaves extracellular regions of membrane proteins 40 and is known to regulate pathologic angiogenesis in the retina 41 (Fig. S2Cii). Finally, a serine carboxypeptidase, Retinoid-inducible serine carboxypeptidase (SCPEP1), was increased over 50% after one or three weeks of PQ stress (Fig. S2Ciii).
Short term to long term adaptation of the Secretome. As described for the intracellular proteome data, we can understand the adaptive response in the secretome using the proteins that only change between one www.nature.com/scientificreports www.nature.com/scientificreports/ week and three weeks of stress. This filtering leaves only 29 increased and 34 decreased proteins. Reactome pathway analysis of these altered proteins returned many altered candidate pathways centered around the decreased levels of tubulins (Fig. 5A). Overall, there are many less concerted changes in this data subset than the others. However, there are more interesting individual changes. Amyloid precursor protein (APP) was increased between one and three weeks of stress (Fig. 5B). The most increased protein in the adaptation of the Secretome was tetranectin (almost 14-fold increase), which is a potential marker of several diseases and may play a role in exocytosis [42][43][44] . Retinol-binding protein 4 (RBP4) was also increased over 2-fold (Fig. 5C). RBP4 plays an important role in facilitating transportation and cellular uptake of retinol 45 .

Discussion
The results presented here indicate that oxidative stress induces proteome remodeling inside RPE indicative of a metabolic shift from respiration toward glycolysis. A shift to glycolysis has been observed in other systems to reduce oxidative damage under high energy demand 46 . We also found intracellular proteome changes suggestive of ECM and cellular organization changes. Some intracellular protein changes have less-clear implications and require follow-up experiments. For example, CD74, a major histocompatibility class II antigen that acts as a transcription regulator linked to cell survival 47 was decreased. GDF15, a distant member of the BMP subfamily and target of p53 48 , was also highly upregulated after chronic ROS, although its function in RPE has not been studied.
Short-term and long-term oxidative stress also induce significant remodeling of the secreted proteome. Notably, there were many altered proteins in the complement system which is known to be affected in AMD 14,49,50 . We also observe many changes in proteins related to ECM, focal adhesion, and TGFβ signaling 51 , which together may regulate neovascularization. Most interestingly, we find a number of Alzheimer's-related proteins change in the secretome of oxidatively stressed RPE, including APP, APLP2, and APOE. The different isoforms of APOE have been associated with altered risk of AMD 39,52,53 , and APOE was previously demonstrated to be secreted from RPE and regulate lipid uptake 15 .
It is also interesting to consider the proteins we identified that did not change, such as HTRA1. Single-nucleotide polymorphisms (SNPs) near the ARMS2/HTRA1 genes are associated with altered risk of developing AMD [54][55][56] . We detected HTRA1 protein both inside RPE and in the secretome. Intracellularly, HTRA1 was increased between one and three weeks of stress but did not meet our initial statistical criteria for defining a protein change (39% increase, q-value = 0.016). Extracellularly, HTRA1 was increased 65% but the q-value was only 0.047. We also identified HTRA2 from the intracellular proteome data, and HTRA2 was significantly increased between the chronic stress and control groups (54% increase, q-value < 0.001).
Our study uniquely addresses both the temporal and spatial reorganization of the RPE proteome. We expect this data will serve as an important resource for future mechanistic and therapeutic studies. For example, the data suggest new proteins that may mediate the observed retinal disfunction in AMD in tissues adjacent to RPE, the photoreceptors and vasculature (Fig. 6). Increased secretion of TGFβ1 protein and decreased secretion of BMP1 protein may promote harm to adjacent photoreceptors and retina tissue; TGFβ1 protein was reported to play a role in proliferative vitreoretinal diseases 57 and neovascular AMD 58 , and BMP-Smad1/5/8 signaling was reported to promote survival of retinal ganglion cells 59 . Signaling by TGFβ1 and BMP proteins are also known to modulate www.nature.com/scientificreports www.nature.com/scientificreports/ macrophage transitions to pro-or anti-inflammatory state [60][61][62][63] , and macrophage activity in the retina has been connected to AMD 64 . The decreased APOE secretion may decrease the RPE's ability to uptake lipids and possibly the shed photoreceptor outer segments. Increased secretion of proteases ADAM9 and SCPEP1 might breakdown the basement membrane, and increased urotensin-II might cause choroidal vasculature to constrict and limit the oxygen delivery to the region. All of these ideas are opportunities for follow-up research uniquely enabled by this study. An important limitation to note about our study is that in order to quantify extracellular proteins without interference from proteins in serum used to prepare media, secretome experiments were carried out by switching cells serum-free media for 24 hours. This was not done for the intracellular proteomics experiments, and somewhat limits direct comparison of the two datasets. Regardless of this caveat, this intracellular and secretome remodeling of RPE cells shown by proteomics provides a highly granular view of the short-term and long-term oxidative stress responses. Methods chemicals and reagents. LC-MS grade acetonitrile and water were manufactured by Burdick and Jackson and purchased from Thermo-Fisher. Urea and the BCA assay kit (Pierce) were also purchased from Thermo-Fisher. LC-MS grade formic acid and 1 M triethanolamine bicarbonate (TEAB), pH 8.5, paraquat (PQ) and tris base were purchased from Sigma-Aldrich. Sequencing-grade modified trypsin was purchased from Promega.

Generation of RPE from human embryonic stem cells.
Much of the methods for RPE generation from stem cells and the treatment with PQ was identical to a previous publication 9 , and therefore some of the following text matches those methods verbatim. All stem cell work was approved by the Buck Institute SCRO committee, and all methods were performed in accordance with federal and local stem cell research guidelines and regulations. Human ESCs (WA-01; National Institutes of Health registry #0043) were maintained in Essential 8 medium (Gibco, Grand Island, NY, USA) and 1% penicillin-streptomycin-amphotericin B solution (Lonza, Walkersville, MD, USA). Cells were grown on Matrigel (BD Biosciences, St. Paul, MN, USA)-coated plates and serially passaged using 0.5 mM EDTA solution. Genotyping revealed that cells did not have any key AMD single nucleotide polymorphism (SNP) 65,66 . Human ESCs were differentiated to retinal lineage using our previously published protocol 26,[67][68][69] . The RPE regions were manually picked and expanded.
The RPE cells generated from hESCs were cultured in α-MEM medium (Life Technologies, Grand Island, NY, USA) containing 1% fetal bovine serum (Atlanta Biologicals, Flowery Branch, GA, USA), L-glutamine (VWR, Radnor, PA, USA), taurine (Sigma-Aldrich Corp., St. Louis, MO, USA), hydrocortisone (Sigma-Aldrich Corp.), and tri-iodo-thyronine (Sigma-Aldrich Corp.) on Matrigel-coated plates or filter membranes (VWR) 70 . Cells were subcultured using Accutase (Gibco) in the presence of thiazovivin (1µM), a Rho-associated protein kinase pathway inhibitor that allows passaging RPE cells for over eight passages 71 . The RPE cells were in culture for up to seven passages without any appreciable loss in their ability to mature into polarized RPE cells. The cells were allowed to grow to 100% confluence and used for experiments following maturation.
cell treatments and collection. RPE cells were treated as described previously 9 . Briefly, RPE cells were seeded in 12-well dishes and allowed to mature for up to three weeks following final passage. Fully confluent plates of cells showing typical cobblestone morphology and presence of pigmentation were treated every other day with 160 μM Paraquat (PQ) diluted in RPE media described above for one week or three weeks. In parallel on the same multi-well dish, control RPE were grown with the same media lacking PQ. After treatment, the cells were washed once with PBS and then trypsinized. Media containing KSR was then added to quench the trypsin activity after 5 minutes, and cells were concentrated by centrifugation. The media was aspirated and the cells were washed once with PBS before being frozen at −80 °C until further processing. www.nature.com/scientificreports www.nature.com/scientificreports/ Sample preparation for intracellular proteome measurement. RPE cell pellets were lysed and proteins were simultaneously precipitated according to the MPLEx protocol 72 . To each frozen cell pellet, 100 μL of PBS was added, and then 5 volumes of cold (−20 °C) chloroform-methanol (2:1 [vol/vol]) was added to each pellet and cells were vortexed until they thawed and dissolved in the solvent. Samples were incubated on ice for 5 minutes, vortexed in the cold room for 5 minutes, and then centrifuged at 15,700 relative centrifugal force (RCF) for 10 minutes at 4 °C, which helped clarify a biphasic solution with proteins at the interphase. The top aqueous layer was removed taking care to avoid the protein disc at the interphase, and then 500 μL of cold (−20 °C) methanol was added resulting in a monophasic solution. Samples were then vortexed briefly and centrifuged at 15,700 RCF for 10 minutes at 4 °C to pellet proteins. The supernatant was removed and the protein pellet was dissolved again with 1 mL of cold (−20 °C) methanol by vortexing to wash away any remaining lipids, and proteins were pelleted again by centrifugation at 15,700 RCF for 10 minutes in the cold room. The methanol wash was removed, and the protein pellet was dried briefly in a vacuum centrifuge. The dried protein pellet was then dissolved in 8 M Urea and 100 mM TEAB containing protease inhibitor cocktail, and protein concentration was quantified using the BCA assay. Protein disulfides were reduced by adding 4.5 mM DTT (final concentration) and heating to 37 degrees for 30 minutes, and then samples were cooled to room temperature before adding 10 mM iodoacetamide (final concentration) to alkylate free thiols. Samples were then diluted 4-fold with 100 mM TEAB to reduce the Urea concentration, and enzymatic protein hydrolysis was initiated by the addition of trypsin (Sequencing Grade Modified Trypsin, Frozen, Promega) at a ratio of 1:50, trypsin: substrate weight.
Sample preparation for secreted proteome measurement. RPE cells were stressed as described above for one week or three weeks, and then cells were washed three times with 1x PBS before adding 2 mL of serum-free RPE media. Although many cell lines are sensitive to serum-free conditions, RPE can be cultured serum-free after maturation 73 and can even be passaged in serum-free media without affecting their viability 74 . Cells were cultured for 24 hours and then the media was collected and frozen at −80 °C. Media containing secreted proteome was then concentrated using a centrifugal ultrafiltration membrane (Amicon Ultra-15, MWCO 30 kDa) by centrifugation at 7,500 X gravity for 10 minutes at 4 °C. To denature proteins the buffer was exchanged three times by adding 1 mL of 2 M Urea in 50 mM Tris, pH 8.0 and centrifuging. The hold-up volume of the buffer-exchanged protein was then removed from the ultrafilter and protein was quantified using the BCA assay. Protein disulfides were then reduced by adding 4.5 mM DTT (final concentration) and heating to 37 degrees for 30 minutes, and then samples were cooled to room temperature before adding 10 mM iodoacetamide (final concentration) to alkylate free thiols. Enzymatic protein hydrolysis was initiated by the addition of trypsin (Sequencing Grade Modified Trypsin, Frozen, Promega) at a ratio of 1:50, trypsin: substrate weight.
Mass spectrometry data collection and analysis. Peptides were analyzed by nanoflow liquid chromatography -tandem mass spectrometry analysis (nanoLC-MS/MS) on a Sciex 5600 TripleTOF mass spectrometer using data-independent acquisition (DIA) using variable-width precursor isolation windows as described previously 75 . Peptide separation for DIA data collection was performed using a linear gradient of 5% mobile phase B (98% Acetonitrile, 0.05% formic acid, 1.95% water) to 40% B over 200 minutes with a flow rate of 300 nL/min. Data analysis. Proteins were identified and quantified from DIA-MS data using Spectronaut 29 and the pan-human spectral library 21 . Spectronaut settings were all defaults except that quantification filtering was set to Q-value percentile = 25% for intracellular proteome, and Q-value percentile = 20% for secreted proteome. Spectronaut performs local retention time-based signal normalization. Statistically-significant protein changes (defined as q < 0.01 and at least 50% change) were assessed for enriched pathways and visualized using the ClueGO plugin of Cytoscape 76 . Additional analysis was carried out using custom scripts in R 77 .
Sample groups and statistical comparisons. Data from the intracellular proteome was collected from three wells of treated cells and two wells of control cells per time point, resulting in a total of 10 samples for intracellular proteome. To ensure that we only detect changes that result from short term and long term ROS, we compared the treated samples in triplicate with the four control samples, two from each treatment timepoint. Data from the protein secretome was collected from four replicates of treated cells and four replicates of control cells at each timepoint. As described for the intracellular proteomics comparison, samples from each treatment timepoint were compared with a pool of all eight control samples.

Data availability
All raw mass spectrometry data files are available from massive.ucsd.edu under dataset identifier MSV000083398 (password: intrpesec).