Identification of Paracrine Neuroprotective Candidate Proteins by a Functional Assay-driven Proteomics Approach*S

Glial cells support neuronal survival and function by secreting neurotrophic cytokines. Retinal Mueller glial cells (RMGs) support retinal neurons, especially photoreceptors. These highly light-sensitive sensory neurons receive vision, and their death results in blinding diseases. It has been proposed that RMGs release factors that support photoreceptor survival, but the nature of these factors remains to be elucidated. To discover such neurotrophic factors, we developed an integrated work flow toward systematic identification of neuroprotective proteins, which are, like most cytokines, expressed only in minute amounts. This strategy can be generally applied to identify secreted bioactive molecules from any body fluid once a recipient cell for this activity is known. Toward this goal we first isolated conditioned medium (CM) from primary porcine RMGs cultured in vitro and tested for survival-promoting activity using primary photoreceptors. We then developed a large scale, microplate-based cellular high content assay that allows rapid assessment of primary photoreceptor survival concomitant with biological activity in vitro. The enrichment strategy of bioactive proteins toward their identification consists of several fractionation steps combined with tests for biological function. Here we combined 1) size fractionation, 2) ion exchange chromatography, 3) reverse phase liquid chromatography, and 4) mass spectrometry (Q-TOF MS/MS or MALDI MS/MS) for protein identification. As a result of this integrated work flow, the insulin-like growth factor-binding proteins IGFBP5 and IGFBP7 and connective tissue growth factor (CTGF) were identified as likely candidates. Cloning and stable expression of these three candidate factors in HEK293 cells produced conditioned medium enriched for either one of the factors. IGFBP5 and CTGF, but not IGFBP7, significantly increased photoreceptor survival when secreted from HEK293 cells and when added to the original RMG-CM. This indicates that the survival-promoting activity in RMG-CM is multifactorial with IGFBP5 and CTGF as an integral part of this activity.

Cell to cell communication is crucial for any multicellular organism. A major portion of these communication events relies on secreted proteins that act on neighboring cells as well as on target cells far from the source cell. This has a particular bearing for nerve cells because neuronal survival depends on neurotrophic factors released from surrounding tissues, and in the absence of such factors the neurons undergo programmed cell death (1). Hereditary degeneration of retinal photoreceptor cells is a major cause of blindness and often the result of mutations in phototransduction-relevant genes (2). Intriguingly although the mutations are present from beginning, the degeneration often does not occur until late in life. One explanation for such a delayed photoreceptor apoptosis may be the existence of paracrine, neurotrophic activities produced by the highly reactive support cells of the retina, the retinal Mueller glial cells (RMGs). 1 Factors of this kind could be capable of upholding photoreceptor function for long periods even in diseased gene background, although finally the demand for support may exceed the supply upon which the balance shifts to cell death. As such, strategies to support degenerating photoreceptors via exogenously given recombinant neurotrophic factors are considered as a future treatment option to halt retinal degeneration and thereby preventing blindness.
There are several reports suggesting the existence of neurotrophic molecules of RMG origin. Medium conditioned by chick retinal explants or by chick RMGs has been shown to support neuronal survival (3). Moreover porcine RMG cultures have been shown to secrete factors that promote survival and neuritogenesis of adult porcine retinal ganglion cells in vitro (4) as well as cone photoreceptors (5). However, in none of these studies has the protein responsible for the neuronal survivalpromoting effects been identified.
To arrive at a situation where RMG-derived factors come to use in photoreceptor treatment, there is a need to develop and use strategies that advance from the description of a protective activity toward the identification of the proteins/ factors responsible for the activity. This could be achieved by combining high resolution proteomics approaches such as sensitive LC-MS/MS with high content functional assays that validate the samples throughout the process. Here we applied this strategy to investigate paracrine factors acting between RMGs and retinal photoreceptors (PRs). RMG-conditioned medium (RMG-CM) was tested for survival-promoting activity on primary PRs in a large scale survival assay specifically developed for this purpose. To further characterize the survival-promoting activity, the complexity of RMG-CM was reduced by subfractionation, and finally active subfractions were subjected to mass spectrometric identification of component proteins. We found that active subfractions of RMG-CM include at least 74 different proteins among them several potentially new candidates for neuroprotective activity in the context of PR survival. Expression cloning and retesting of three candidate proteins for their ability to promote PR survival revealed that two of them (connective tissue growth factor (CTGF) and insulin-like growth factor binding protein 5 (IGFBP5)) are able to increase PR survival when applied in combination with the original RMG-CM, indicating that the RMG-derived neurotrophic activity is multifactorial. Our results demonstrate the feasibility of the strategy that combines functional assay-driven subfractionation of complex secretomes with sensitive mass spectrometry to identify molecules responsible for cellular survival/communication. Importantly this strategy is also applicable to the identification of other bioactive proteins including growth factors and cytokines and opens up to the in-depth characterization of single molecules derived from complex protein mixtures.

RMG Preparation
Adult porcine eyes were provided by a local slaughterhouse. They were removed from the animals within 5 min after death and kept on ice in CO 2 -independent medium (Invitrogen) until further use. Retinas were dissected from the eye, and RMGs were prepared as described previously (6,7). Briefly major blood vessels were removed, and the retina was cut into small pieces that were washed twice in Ringer's solution. Dissociation of retinal tissue was obtained by treating each retina with 2.2 units of activated papain (Worthington) for 40 min at 37°C; papain enzyme activity was stopped by addition of DMEM-HEPES (Invitrogen) with 10% FCS. Then 160 units of DNase (Sigma) were added, and the tissue was further dissociated by gentle trituration using a fire-polished Pasteur pipette. Dissociated cells were collected by centrifugation (800 ϫ g for 5 min), resuspended in DMEM containing 10% FCS, and plated directly onto cell culture plates (Nunc). The plated cells were allowed to attach for 16 h at 37°C in an incubator. Non-attached cells were then removed by gentle agitation (panning), and RMGs were cultured for up to 21 days to near confluence.

Photoreceptor Preparation
The retina was dissected from porcine eyes as described above, and photoreceptors were isolated from retina as described previously (8,9). Briefly retinal pieces were cut, washed, and subsequently papain-treated for 15 min, and then the reaction was stopped by addition of DMEM/Nutrient Mix F-12 (Invitrogen) with 2% FCS. After gentle trituration, PRs were collected by centrifugation (800 ϫ g for 5 min), resuspended in DMEM/Nutrient Mix F-12 containing 2% FCS, and plated on 96-well plates, which were precoated with poly-D-lysine (Sigma; 2 g/cm 2 for 3 h) and laminin (BD Biosciences; 1 g/cm 2 overnight). PR quality was controlled before plating by propidium iodide staining, and only preparations with above 80% living cells were used. After incubation for 20 h, the medium was removed and replaced by different conditioned media or subfractions.

Conditioning of Culture Medium
Plates with either RMGs, rat embryonic fibroblast cell line R6, or HEK293 fibroblast cells were washed twice with serum-free medium and incubated for 3 h in serum-free medium, and then medium was replaced with DMEM/Nutrient Mix F-12 containing 1 mM sodium pyruvate (Invitrogen) and penicillin/streptomycin (Invitrogen). Medium was conditioned for 16 h, then filtered (Ͻ0.2 m) to remove nonadherent cells and debris, and either directly applied onto photoreceptors in 96-wells (100 l/well) or subfractionated as described below.

Photoreceptor Survival Assay
Conditioned media from either RMGs (day 7, 5.6 Ϯ 1.3 g/ml; day 14, 12.6 Ϯ 1.1 g/ml; day 21, 21.3 Ϯ 1.9 g/ml) or R6 (11.9 Ϯ 1.5 g/ml) were applied onto attached photoreceptors (100 l/well) 20 h after preparation. Photoreceptor survival was monitored by performing an esterase-calcein fluorophore (Molecular Probes) assay as described before (9). Briefly living cells fluoresce bright green, and total fluorescence of respective wells was measured daily in a microplate fluorescence reader (MWG Biotech FL600), and values were compared with the initial fluorescence at the beginning of the survival assays. Relative fluorescence is given in percent of initial fluorescence, and every experiment was performed at least three times.
In some experiments, inhibitors U0126 and LY294002 (both from Cell Signaling Technology) were added to the photoreceptor survival assay to a final concentration of 1 and 5 M, respectively. For proteinase treatment, proteinase K (Sigma) was added to RMG-conditioned medium at a final concentration of 100 g/ml and incubated for 30 min at 37°C, and the reaction was stopped by addition of 12.5 g/ml ␣ 2 -macroglobulin (Serva).

Subfractionation of Conditioned Media
Size Fractionation-14 ml of conditioned media were centrifuged at 5000 ϫ g for 40 min at 4°C through 50-or 10-kDa Macrosep devices (Millipore); the remaining size-enriched concentrate was applied on the photoreceptor survival assay.
Mono Q FPLC-Prior to FPLC separation, conditioned media (RMG-CM: 57 ml with a total protein content of 718 g; R6-CM: 60 ml with a total protein content of 714 g) were concentrated and equilibrated to 30 mM sodium phosphate buffer via Amicon 10-kDa filters. 10-fold concentrated conditioned media were loaded onto Mono Q columns (HR_5/5, GE Healthcare) in 30 mM sodium phosphate buffer (pH 7.5), flow-through was collected, and proteins binding to the Mono Q matrix were eluted by a gradient of 0 -60% 1 M NaCl within 25 min into 96-well plates. Collected fractions were combined as described under "Results," concentrated 10-fold via Microsep 10-kDa devices, and rediluted 10-fold in DMEM/Nutrient Mix F-12 before application on the photoreceptor survival assay. Protein contents for each pooled fraction where calculated after Bradford assay (performed on an acetone-precipitated aliquot of each pooled fraction) and were sufficiently similar between RMGs and R6 to allow direct comparisons of survival effects (RMGs: A, 3.35 g/100 l; B, 4.16 g/100 l; C, 3.45 g/100 l; D, 3.62 g/100 l; E, 2.87 g/100 l; and flow-through, 3.53 g/100 l; R6: A, 3.47 g/100 l; B, 3.74 g/100 l; C, 3.61 g/100 l; D, 2.93 g/100 l; E, 3.78 g/100 l; and flow-through, 3.43 g/100 l).

Statistical Tests
Statistical significance was calculated using Student's t test.

1D PAGE
Proteins from conditioned media or from Mono Q FPLC fractions were precipitated with acetone and dissolved in buffer for 1D PAGE (50 mM Tris, pH 7.4, 250 mM NaCl, 25 mM EDTA, 1% Nonidet P-40, 10% glycerol) with protease inhibitors (Complete, Roche Applied Science), protein content was measured by Bradford assay (Bio-Rad), and samples were resolved by 9 -15% PAGE. Protein bands were visualized by silver staining.

2D PAGE
To discriminate between residual serum factors and de novo synthesized proteins from RMGs, medium was conditioned in the presence of 0.8 mCi of 35 S (trans-35 S label, ICN)/10-cm culture dish and filtered after conditioning, and proteins from conditioned medium were precipitated with acetone. Precipitates were dissolved in 2D lysis buffer (9 M urea, 2 M thiourea, 1% dithioerythritol, 4% CHAPS) and protease inhibitors (Complete mini, Roche Applied Science), and protein content was determined by Bradford assay (Bio-Rad). Electrophoretic separation was carried out as described previously (7,10) with IPG strips pH 3-10 (GE Healthcare) in the first dimension and gradient SDS-PAGE gels (9 -15%) in the second dimension. Gels were silver-stained (11) and dried between cellophane sheets. Hyperfilms (GE Healthcare) were exposed to dried gels for 1 week to produce autoradiographs.

Mass Spectrometry
Electrospray Analysis-Peptide sequence information was obtained as described before (10) by LC-coupled MS/MS analysis on a Q-TOF2 system (Waters) coupled with a CapLC system (Waters). Proteins from aliquots (400 l) of active FPLC flow-through fractions were precipitated with acetone, dissolved in 50 mM ammonium bicarbonate supplemented with 0.2% Rapigest (Waters), and reduced in 2.5 mM dithiothreitol (Merck) for 10 min at 60°C followed by alkylation in 7.5 mM iodoacetamide (Merck) for 30 min at room temperature. Proteins were then subjected to tryptic digest (5 ng/l of sample) at 37°C overnight. Peptide samples were acidified with HCl (37%) to a final pH below 2 to hydrolyze the Rapigest surfactant. Samples were centrifuged for 30 min (13,000 rpm at 4°C), and the aqueous phase was transferred to new tubes. 15-l aliquots were loaded onto the CapLC system, trapped on a C 18 precolumn (5 m, Symetry300 C 18 NanoEase trap column, Waters), and separated on a 75-m C 18 column (75 m ϫ 150 mm, Atlantis C 18 NanoEase column, Waters) by elution with a gradient (0 -60% solution B (95% acetonitrile, 0.1% formic acid)). The peptides were subjected to nanoelectrospray ionization followed by MS/MS in a Q-TOF2 system (Waters) coupled on line to the CapLC system.
LC MALDI MS/MS Analysis-Sample preparation was similar as for electrospray except acidification was done with TFA to a final concentration of 0.5%. Peptide separation by HPLC was performed on an UltiMate nano-LC system (LC Packings) equipped with a 75-m C 18 column. Mobile phases were 5% acetonitrile, 0.1% TFA (solution A) and 80% acetonitrile, 0.1% TFA (solution B). Peptides were separated by a gradient of 5-50% solution B in 80 min followed by a step of 50 -100% solution B in 5 min at a flow rate of 200 nl/min. The eluate was collected in 20-s fractions that were diluted 4-fold in 2 mg/ml ␣-cyano-4-hydroxycinnamic acid, 70% acetonitrile, 0.1% TFA and spotted on MALDI targets by a Probot liquid handling system (LC Packings). Protein identification was performed on a 4700 Proteomics Analyzer (Applied Biosystems, Framingham, MA).

Database Searching
MS/MS spectra from MALDI were extracted by the 4000 Series Explorer Software (Applied Biosystems). Charge state deconvolution and deisotoping were not performed, and a limit for peak detection of a signal to noise ratio of 5 was used. MS/MS spectra from electrospray were extracted and processed by Protein Lynx Global Server version 2.1 RC5 (Waters). Background was subtracted (threshold ϭ 35%, polynomial order ϭ 5), and spectra were smoothed (Savitzky-Golay, two iterations, smoothing window ϭ 3 channels). Deisotoping was performed using the deisotoping type "medium" (MaxEnt Lite algorithm) with a threshold of 1%.
All MS/MS samples were analyzed using Mascot (Matrix Science, London, UK; version 1.9.05) and X! Tandem (The Global Proteome Machine Organization; version 2007.01.01.1). Mascot and X! Tandem were set up to search the Swiss-Prot database (selected for Mammalia, version July 25, 2006, 44,506 sequences) assuming the digestion enzyme trypsin with one missed cleavage allowed. Mascot and X! Tandem were searched with a fragment ion mass tolerance of 0.50 Da and a parent ion tolerance of 100 ppm. Iodoacetamide derivative of cysteine was specified in Mascot as a fixed and in X! Tandem as a variable modification. Oxidation of methionine was specified in Mascot and in X! Tandem as a variable modification.
Criteria for Protein Identification-Scaffold (version 01_06_19, Proteome Software Inc., Portland, OR) was used to validate MS/MSbased peptide and protein identifications. Peptide identifications were accepted if they could be established at greater than 95.0% probability as specified by the Peptide Prophet algorithm (12). Protein identifications were accepted if they could be established at greater than 95.0% probability and contained at least two identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm (13). Proteins that contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to satisfy the principles of parsimony. All details for protein and peptide identifications are given as supplemental tables.

RESULTS
Proteomics Strategy to Identify Secreted Survival-promoting Factors-To identify candidate proteins correlated with cellular activity, it is favorable to develop a strategy that integrates biological activity on the basis of high content cellular assays within a proteomics work flow. Inspired by the principle of multidimensional chromatography (multidimensional protein identification technology (MudPIT) (14, 15)), we chose an off-line approach that decreases throughput yet enabled us to (i) increase input of sample material for preparative demands and (ii) validate generated subproteomes for cellular function within the work flow ( Fig. 1). Starting with the total secretome from retinal Mueller glial cells (RMG-CM) that carries a survival-promoting activity for retinal PRs, we first subfractionated large quantities of the secretome by size exclusion applying a 10-kDa cutoff filter. Then the concentrated protein pools Ͼ10 kDa carrying the PR survival activity were further fractionated by anion exchange chromatography (FPLC). Pooled subfractions from FPLC were retested for activity, only the active subfractions were subjected to tryptic cleavage, and resulting peptides were identified by mass spectrometry. This led to the identification of several candidates for survival activity that were then PCRcloned and expressed in a mammalian system and retested for activity.
RMG-conditioned Medium Promotes Porcine Photoreceptor Survival in Vitro-RMGs have been suggested as important players in rescuing retinal cells including PRs during retinal degeneration or after toxic insults. For instance, it has been proposed that RMGs release substances into the culture medium (conditioned medium) that are able to induce neurite sprouting in cultures of retinal ganglion cells (4) and enhance survival of cones in vitro (5). However, the identity of these factor(s) is unknown. To define potentially neuroprotective factors released by RMGs, we designed an in vitro survival assay that enables monitoring of the protective effects of RMG-CM or subfractions on PRs (see "Experimental Procedures"). PR survival was measured on a daily basis after replacing regular medium with respective conditioned media and compared with the starting population of living cells. Survival of cells was measured as total fluorescence per well after calceinesterase reaction. As shown in Fig. 2A, cell counts correlated linearly with total fluorescence. RMG-CM resulted in consistently higher photoreceptor survival rates than application of either conditioned medium from control cells (R6-CM) or basic medium (DMEM) alone (see Fig.  2, B and C). The most significant time point of photoreceptor survival promoted by RMGs compared with controls proved to be 6 days after application of conditioned media.
The RMG-derived Survival-promoting Activity Is Lost during Prolonged Culturing-We reported previously that RMGs dedifferentiate upon prolonged culturing (7). It could thus be hypothesized that RMGs may lose, along with other specific marker proteins such as glutamine synthetase, the PR survival-promoting activity over time in culture. Starting with the total secretome from retinal Mueller glial cells that carries a survival-promoting activity for retinal photoreceptors, large quantities of the secretome were subfractionated by anion exchange chromatography (FPLC). Pooled subfractions were retested for activity, and active subfractions were then subjected to tryptic cleavage. The resulting peptides were analyzed by mass spectrometry (LC-MS/MS). This led to identification of several candidates for survival activity that were then expressed in a mammalian system and retested for activity.
We tested CM from RMGs that were cultured for 7 (data not shown), 14, and 21 days before conditioning. The highest survival-promoting activity was found in CM from 14-day-old RMGs. After 21 days in culture the activity was lost compared with the controls (see Fig. 3A). Medium conditioned by 7-dayold RMGs also showed lower survival effects on PRs compared with day 14 CM; this was accompanied by a lower overall protein content in the CM. This probably resulted from lower cell densities in the culture plates of 7-day-old RMGs compared with 14-and 21-day-old RMGs.
Survival-promoting Activity from RMGs Is a Protein Separating into Size Fraction above 10 kDa-To confirm that the survival-promoting activity is a protein, RMG-CM was treated with proteinase K. Proteinase K treatment was stopped by addition of ␣ 2 -macroglobulin to not directly reduce survival of PRs. Proteinase K treatment completely abolished the survival-promoting activity of RMG-CM (Fig. 3A), whereas the addition of ␣ 2 -macroglobulin alone reduced the survival of PRs to a lesser extent.
To further characterize the survival-promoting factor, the CM was enriched into molecular mass fractions Ͻ10 kDa, between 10 and 50 kDa, and Ͼ50 kDa. RMG-derived survivalpromoting activity was significantly enriched in the fraction above 50 kDa (Fig. 3B) and to a lesser extent also in the 10 -50-kDa fraction when compared with the control (R6-CM).
Survival-promoting Activity Acts through PI 3-Kinase-dependent Pathway in Photoreceptors-Cell survival can be promoted by several distinct or interacting pathways. To define the pathways responsible for RMG-mediated prolonged survival of photoreceptors in vitro, inhibitors were applied along with CM in the survival assay. LY294002 inhibits selectively PI 3-kinase, and U0126 is an inhibitor of MEK1/2, the kinases upstream of ERK1/2. Application of LY294002 completely blocked the survival-promoting activity of RMG-CM, whereas U0126 only partially reduced survival of PRs (Fig.  3C). To ensure that the application of the inhibitors itself would not interfere with PR survival both inhibitors were also applied in combination with 2% FCS on the PR survival assay. FCS at 2% carries strong survival-promoting activities, and neither LY294002 nor U0126 were able to prevent PR survival under those conditions (data not shown). PR survival promoted by RMG-CM thus depends significantly on the PI 3-kinase pathway and only to a lesser extent on the MEK kinase pathway.
RMG-CM Is a Complex Mixture of Proteins-To define the factor responsible for RMG-derived PR survival, we first studied the complexity of RMG-CM. For this purpose, medium was conditioned in the presence of [ 35 S]methionine, and subsequently proteins were resolved by high resolution 2D PAGE. Radioactive labeling allows detection of proteins transcribed during the conditioning process on a very sensitive level and also enables discrimination of these from residual, serumderived proteins.
RMG-CM shows highly complex protein patterns (supplemental figure). About 2200 different protein spots were consistently detected on autoradiographs of 2D gels. Thus, a prerequisite for identifying proteins responsible for the neuro- protective effect of RMG-conditioned medium was to reduce complexity.
RMG-CM Complexity Can Be Diminished by Separation by Mono Q FPLC-To reduce complexity, large amounts of RMG-CM were first volume-reduced by size exclusion through a 10-kDa cutoff filter. The concentrated protein fraction above 10 kDa was then subjected to further fractionation by anion exchange (Mono Q) FPLC. RMG-CM gave reproducible elution profiles from Mono Q columns (Fig. 4A), and the complexity of the eluting fractions was determined by 1D PAGE (see Fig. 4B). A total of 33 fractions from each run were collected into 96-well plates, and according to the UV 280 nm detection the majority of proteins were found in fractions 20 -33 (Fig. 4A). 10 g of total protein from every second fraction were resolved by 1D PAGE, which shows that prominent bands, probably representing one protein species, appear in ϳ6 -8 consecutive fractions.
Survival-promoting Factor Does Not Bind Strongly to Anion Exchange Column-All pooled fractions from Mono Q FPLC along with the flow-through were tested on the PR survival assay for survival-promoting activity. Activity of every tested fraction was compared with an analogously treated conditioned medium from the R6 cell line (R6-CM) to evaluate selective RMG-derived activity.
Compared with R6-CM, most of the survival-promoting activity from RMGs did not bind to Mono Q-Sepharose. The major part of the activity reappeared in the flow-through from FPLC, and some appeared in pooled fraction A, which covers very low eluting salt concentrations (below 50 mM NaCl; Fig.  4C). We concluded that RMG-derived survival-promoting activity does not bind strongly to anion exchange matrices and could therefore be of acidic nature.
Identification of Proteins in Mono Q Flow-through Fractions-The enrichment of the survival-promoting activity in the Mono Q flow-through, which is not very complex, made it feasible to identify component proteins directly and generate a list of probable candidates for this activity. The flow-through fraction was treated with trypsin and applied to Q-TOF MS/MS as well as LC-MS/MS MALDI mass spectrometry to identify its component proteins.
Seventy-four different proteins were identified above the significance threshold from RMG FPLC flow-through (Table I).
To further preselect likely candidates for paracrine activity, all identified proteins were screened for the presence of a secretion signal by the program SignalP and for their likelihood to occur extracellularly by SecretomeP (Center for Biological Sequence Analysis, Technical University of Denmark) (16). We found that a total of 38 of 74 proteins were predicted to be FIG. 3. Active compound from RMG-CM is a protein rather than a peptide. A, porcine photoreceptor survival 6 days after application of medium alone (medium), R6-conditioned medium (R6-CM), and RMG-CM generated from RMGs 14 days in vitro (day14) or RMGs 21 days in vitro (day21). Highest survival-promoting activity was found in conditioned medium from 14-day-old RMGs; after 21 days in culture the activity was lost and was as low as in the controls. Furthermore RMG-CM day 14 treated with proteinase K (proteinase K ϩ ␣ 2macroglobulin) was applied to the photoreceptor survival assay. Proteinase K treatment was stopped by addition of ␣ 2 -macroglobulin, which itself quenched PR survival but to a lesser extent compared with proteinase K treatment, which completely destroyed the survivalpromoting activity of RMG-CM (**, p Ͻ 0.01; ***, p Ͻ 0.001). B, size fractionation of RMG-CM. RMG-CM (day 14) was enriched into three molecular mass fractions Ͻ10 kDa, between 10 and 50 kDa, and above 50 kDa. Similar fractionations were performed with R6-conditioned medium (R6-CM). All fractions were monitored for their survival-promoting activity on photoreceptors. Survival is given as relative fluorescence compared with initial fluorescence 6 days after application of samples. RMG-derived survival-promoting activity was significantly enriched (*, p Ͻ 0.05) in the fraction above 50 kDa as compared with R6-CM and slightly enriched in the fraction between 10 and 50 kDa. RMG-CM fractions containing proteins with a molecular mass below 10 kDa did not carry PR survival-promoting activity; however, the Ͻ10-kDa fraction of R6-CM did promote PR survival (*, p Ͻ 0.05). C, the RMG-derived survival effect is abolished by PI 3-kinase inhibitor LY294002. RMG-CM (day 14) was applied to the photoreceptor survival assay either alone or together with PI-3 kinase inhibitor LY294002 (5 M) or MEK inhibitor U0126 (1 M). Survival is given as relative fluorescence signal 6 days after application compared with initial fluorescence. Application of LY294002 completely blocked survival-promoting activity of RMG-conditioned medium, whereas U0126 only partially reduced survival of PRs (**, p Ͻ 0.01; ***, p Ͻ 0.001). The error bars represent the standard deviation. located extracellularly with either a classical endoplasmic reticulum/Golgi secretion signal or a score above 0.5 from SecretomeP (see Table I). Among those 38 proteins are several growth factor-related proteins that are good candidates for the PR survival activity. For validation of activity, we selected IGFBP5, IGFBP7, and CTGF. Furthermore Cu,Zn-su-peroxide dismutase (SODC) and thioredoxin (THIO) were identified in this fraction; both proteins might protect photoreceptors by preventing oxidative damage. To confirm the mass spectrometric identifications, we additionally monitored the expression of the selected five candidate proteins by RT-PCR from RMGs in vitro. RMGs were confirmed to express CTGF, IGFBP5, and IGFBP7 (Fig. 5A) as well as SODC and THIO (data not shown).
PR Survival-promoting Activity of Identified Candidate Proteins-To identify which protein from the active fraction is responsible for the PR survival-promoting effect of RMG-CM, we cloned CTGF, IGFBP5, and IGFBP7; expressed them in HEK293 cells; and generated stable HEK293 producer cell lines secreting the respective proteins into the culture medium (proper secretion was monitored by Western blots with antibodies against V5 epitope; Fig. 5B). Additionally we tested commercially available SODC and THIO for their activity on PR survival.
We found that supplementation of medium with SODC or THIO had no survival-promoting effect on PRs in vitro when compared with negative control (medium alone) (data not shown). Possible effects of the expression-cloned candidates as secreted by HEK293 cells were directly evaluated by comparison with empty vector-transfected HEK293 cell medium for several days in the PR survival assay. Both CTGF and IGFBP5 but not IGFBP7 increased survival of PRs (Fig. 5C). Additionally we wanted to test whether those factors would be sufficient to increase the survival effect of RMGs when added to the original RMG-CM. Again CTGF and IGFBP5 significantly increased (p Ͻ 0.05) the survival-promoting effect of RMG-CM (Fig. 5D), whereas SODC, THIO, and IGFBP7 failed to do so (Fig. 5D). This indicates a survival-promoting effect for CTGF and IGFBP5, which may act together with other factors present in RMG-CM. DISCUSSION Many efforts have been made to study extracellular fluid proteomes to identify bioactive molecules that bear therapeutic potential or may qualify as biomarkers. For instance, interstitial fluids, which contain secreted proteins from adjacent cells, are a valuable source to learn more about the microenvironment of a given tissue and are especially interesting in cancer development (17). Because of their relatively low abundance, these molecules are often difficult to identify. Given the complexity of an organism, tissue, or body fluid, reduction of this complexity is a key objective. The isolation of primary cells from organs or tissues and their subsequent cultivation in serum-free conditions can allow the recognition of their cell-specific contribution to a complex body fluid (18 -20). Even then, the secretome derived from this homogeneous pool of cells remains too complex to allow unequivocal identification of distinct bioactive molecules. Decreasing the level of complexity again is of key importance for successful identification both with respect to an increase in the depth of analysis and a reduction in the number of candidates

MW (kDa) c) Taxonomy LC-MALDI d)
Q-TOF e) extracellular localization f) 1  that are linked to a specific biological activity. MudPIT (21) has been developed as a powerful approach to increase analytical depth. The strategy presented here derives some of its steps from a MudPIT approach. We dissected the cell-specific secretome first based on a specific function (photoreceptor protection of cell survival), and then candidates for this function (from a subpool of proteins included in bioactive fractions) were identified by mass spectrometry. In the present study, we therefore used this strategy with the intention of providing proof of principle as well as shedding light on the important biological question of endogenous neurotrophic activities in the retina. To this end, we chose pure primary retinal glial cell populations (RMGs) (7) and monitored the secreted protein pool for its photoreceptor survival-promoting activity. The complex protein pool was then subfractionated by size enrichment followed by anion exchange chromatography. Resulting subfractions were again monitored for activity, and only the components of active subfractions of the total protein pool were then identified by mass spectrometry. This approach has two major advantages. 1) There is a possibility to scale up the approach to large input amounts, which allows an increase in sensitivity to a level where one can expect to also identify low abundance molecules, such as cytokines.
2) The functional assay-based subfractionation reduces the amount of potential candidate proteins and therefore results in a manageable number of validations.
With the aim of identifying secreted molecules that are specific for the primary retinal glial cells (RMGs), we always compared the activity of subfractionated RMG-CM to a similarly subfractionated conditioned medium derived from a fibroblast cell line (R6-CM). Interestingly R6-CM also displayed survival-promoting activity, but in contrast to RMG-CM, this activity was enriched in fractions below 10 kDa from sizefractionated conditioned media, and the activity strongly bound to the Mono Q column as was demonstrated by elution from FPLC in high salt-containing fractions (Pool E; Fig. 4C). This underlines the efficacy of the functional assay-driven approach we chose here in contrast to conventional "bottomup" approaches (22), which produce complex protein inventories but lack direct connection to specific functions.
It is at this point interesting to speculate that R6 cells secrete FGF-2 as this is a small protein (17 kDa) with basic properties, and thus predicted to bind strongly to anion exchange matrix, that has already been demonstrated to increase survival rates of photoreceptors in vitro (9). In contrast, the activity purified from RMG-CM does not bind to anion exchange resin and is enriched in fractions above 10 kDa. Moreover the production of this activity is diminished with prolonged culturing underlining the specific expression by primary RMGs and the loss of expression along with dedifferentiation in vitro (7). We identified a total of 74 different proteins from FPLC flow-through, which was demonstrated to be the most active fraction after FPLC separation of RMG-CM. To further preselect good candidates for the paracrine survival effect, we analyzed all identified proteins for the presence of an endoplasmic reticulum/Golgi-directed secretion signal with the SignalP program (23). A subset of 17 proteins were predicted to carry a secretion leader sequence, namely clusterin; collagens ␣-1(I), ␣-1(III), ␣-2(I), ␣-2(IV), and ␣-2(V); connective tissue growth factor; epididymal secretory protein E1; insulin-like growth factor-binding proteins 5 and 7; metalloproteinase inhibitor 1; peptidyl-prolyl cis-trans isomerase B; plasminogen activator inhibitor 1; polypeptide N-acetylgalactosaminyltransferase 2; proactivator polypeptide precursor; serotransferrin; and serpin H1. As a considerable proportion of extracellular proteins are secreted by a non-classical pathway (24), we additionally predicted the extracellular localization with the SecretomeP program (16) and found an additional 21 proteins with an extracellular prediction (score Ͼ0.5), among them proteins that have already been described to be secreted by alternative pathways, namely annexin A2 (25), ␣-enolase (26), cathepsin D (27), and thioredoxin (28).
FIG. 5. Survival-promoting activity of IGFBP5 and CTGF. A, RT-PCR on RMG-derived mRNA confirms expression of mass spectrometrically identified candidate proteins CTGF, IGFBP5, and IGFBP7 by RMGs. B, candidate factors (CTGF, IGFBP5, and IGFBP7) were PCR-cloned into mammalian expression vectors that contain C-terminal His tag and V5 epitope. After transfection into the HEK293 cell line, stably expressing clones were established, and secretion of respective factors into the culture medium was monitored by Western blotting with an antibody recognizing the C-terminal V5 epitope (lanes 1, conditioned medium; lanes 2, cell lysate). C, different factors were tested for their survival-promoting activity on photoreceptors: conditioned medium from HEK293 cells stably expressing CTGF, IGFBP5, or IGFBP7 compared with conditioned medium from HEK293 cells transfected with empty vector (vector). Survival of photoreceptors was measured on days 1, 2, 5, 6, and 7 after initial application of the respective conditioned medium. Survival is given as relative fluorescence compared with initial fluorescence (day 0). CTGF and IGFBP5 consistently result in higher survival rates of photoreceptors than IGFBP7 and empty vector control (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). D, different factors were tested for their survival-promoting activity in combination with RMG-CM: medium alone (negative control), RMG-CM (day 14, supplemented 1:5 with untransfected HEK293 cell-conditioned medium), and RMG-CM supplemented (1:5) with conditioned medium from HEK293 cells stably expressing IGFBP5 (RMG-CM ϩ IGFBP5), CTGF (RMG-CM ϩ CTGF), or IGFBP7 (RMG-CM ϩ IGFBP7) as well as SODC (500 ng/ml) and THIO (500 ng/ml). Survival is given as relative fluorescence compared with initial fluorescence 6 and 7 days after application of samples. IGFBP5 and CTGF increase the survivalpromoting activity of RMG-CM (*, p Ͻ 0.05). The error bars represent the standard deviation.
Altogether this sums up to a proportion of about 51% extracellular proteins within the subfraction of CM we analyzed. This finding is in agreement with recently reported proportions of extracellular proteins in secretomes, e.g. around 35% in the total CM of breast cancer cell lines (22).
Both lines of evidence, the functional assay-driven subfractionation as well as the prediction of secretion for proteins identified in active subfractions, enabled us to select proteins for validation from a comparatively manageable amount of candidates. We selected five candidate proteins and retested them for their activity on PR survival. THIO and SODC were chosen because both proteins are known to be protective against reactive oxygen species, and in addition a member of the thioredoxin family was recently identified to promote survival of cone photoreceptors (29). CTGF, IGFBP5, and IGFBP7 were selected because they are described as secreted proteins with cell growth or maintenance activity (30, 31) (ExPASy) and might therefore be involved in paracrine signaling. As the complexity of RMG-CM indicated that the survival-promoting activity might be multifactorial rather than conferred by one single factor, we also tested the identified candidates in combination with the original RMG-CM. Whereas THIO, SODC, and IGFBP7 did not increase survival of PRs in vitro, CTGF and IGFBP5 were found to be active when applied as secreted from HEK293 cells as well as in combination with the original CM. This is in line with our hypothesis that the RMG-CM-derived activity on PR survival consists of multiple factors. Most interestingly, CTGF and IGFBP5, but not IGFBP7, promoted PR survival, and although all three molecules have an IGFBP domain in their N-terminal region, there is a clear functional difference in the specific context, which underlines the value of the functional approach used here. In a recent report about the expression of six IGFBPs (IGFBP1-6) by porcine RMGs in vitro, the authors found that only the expression of IGFBP5 decreased with prolonged culturing, whereas the expression of all other IGFBPs (IGFBP1, -2, -3, -4, and -6) increased (32). This leads to the assumption that IGFBP5 is likely to be expressed by RMGs in vivo and it underlines our finding that the activity (which may be related partly to IGFBP5) was lost with prolonged culturing of RMGs. The importance of IGFBP5 in retina in vivo is further underlined by its expression pattern in developing chick retina, which points to a significant role during retinal development (33). IGFBP5 binds with high affinity to IGF-1, is cleaved by several proteases (for a review, see Ref. 34), and is able to associate with several extracellular matrix proteins (for a review, see Ref. 30). When associated with the extracellular matrix it appears to be protected from proteolysis; however, once released from the extracellular matrix it is degraded (35) into a 22-kDa fragment that has only low affinity for IGF-1 and -2 as well as into a smaller 14-kDa fragment. In our study, expression of IGFBP5 in HEK293 cells resulted in conditioned media that primarily carried the uncleaved form of IGFBP5, and only a minor part of the protein was degraded (into 20-and 17-kDa fragments as estimated from the marker size; Fig. 5B). Nevertheless both the uncleaved and the cleaved forms of IGFBP5 may be responsible for the PR survival-promoting activity described here. CTGF, the other molecule found to be active, also contains an IGFBP domain (along with a C-terminal cysteine knot and a von Willebrand factor type C (VWFC) domain structure). Yet the binding affinity between CTGF and IGFs is 2-3 orders of magnitude lower than between IGFBP5 and IGFs (36). In contrast to IGFBP5, IGFBP7 has low affinity to the known IGFBP ligands, IGF-1, and IGF-2 (37), indicating that its individual character differs from that of other members of the IGFBP family.
Given the high affinity of IGFBP5 to IGFs, IGFBP5 may function as a deposit of IGFs, and thus the survival activity may be derived from IGF secreted from the cells and bound to IGFBP5. Interestingly IGF-induced survival is well known to be dependent on PI 3-kinase (38), and this would explain why inhibition of this kinase with LY294002 abolished the PR survival-promoting activity of RMG-CM.
However, IGFBP5 has also been reported to control cell survival, differentiation, and apoptosis directly and as such independently of IGF-1 or IGF-2 (for a review, see Ref. 30). These IGF-independent actions of IGFBPs involve the binding of other molecules (for a review, see Ref. 39). Most interestingly, two known binding partners of IGFBP5, namely fibronectin (40) and plasminogen activator inhibitor 1 (41), were also identified in the active fraction of RMG-CM (see Table I).
As the fractionation of the CM was performed in a native state, it is possible that IGFBP5 is bound to either or both of these molecules. This may also explain why the most active fraction after size exclusion was above 50 kDa, whereas IGFBP5 alone has an apparent molecular mass of ϳ35 kDa. Further studies with recombinant protein from mammalian expression systems are needed to decide whether IGFBP5 (and CTGF) directly induces survival of PRs or acts in combination with other molecules.
In conclusion, our functional assay-driven proteomics screening approach has led to the identification of previously unknown glial cell-secreted proteins. In addition, the validation of candidate proteins from active subfractions has produced two promising factors that increase the rate of photoreceptor survival in vitro. Our data therefore demonstrate the value of such a streamlined identification approach where screening for function drives the direction of in-depth biomarker identification. Bearing this in mind, we anticipate that this approach is applicable in several other contexts, such as cancer initiation and progression driven by tissue microenvironment, provided that this can be coupled with relevant functional assays. Finally it should be emphasized that our functional approach allows a high degree of automation because nearly all steps of subfractionation and activity testing can be adapted to high throughput requirements, which is a prerequisite for systematic biomarker identification and validation.