Identification and Validation of Novel Adipokines Released from Primary Human Adipocytes*

Adipose tissue is a major endocrine organ, releasing signaling and mediator proteins, termed adipokines, via which adipose tissue communicates with other organs. Expansion of adipose tissue in obesity alters adipokine secretion, which may contribute to the development of metabolic diseases. Although recent profiling studies have identified numerous adipokines, the amount of overlap from these studies indicates that the adipokinome is still incompletely characterized. Therefore, we conducted a complementary protein profiling on concentrated conditioned medium derived from primary human adipocytes. SDS-PAGE/liquid chromatography-electrospray ionization tandem MS and two-dimensional SDS-PAGE/matrix-assisted laser desorption ionization/time of flight MS identified 347 proteins, 263 of which were predicted to be secreted. Fourty-four proteins were identified as novel adipokines. Furthermore, we validated the regulation and release of selected adipokines in primary human adipocytes and in serum and adipose tissue biopsies from morbidly obese patients and normal-weight controls. Validation experiments conducted for complement factor H, αB-crystallin, cartilage intermediate-layer protein, and heme oxygenase-1 show that the release and expression of these factors in adipocytes is regulated by differentiation and stimuli, which affect insulin sensitivity, as well as by obesity. Heme oxygenase-1 especially reveals to be a novel adipokine of interest. In vivo, circulating levels and adipose tissue expression of heme oxygenase-1 are significantly increased in obese subjects compared with lean controls. Collectively, our profiling study of the human adipokinome expands the list of adipokines and further highlights the pivotal role of adipokines in the regulation of multiple biological processes within adipose tissue and their potential dysregulation in obesity.

Adipose tissue is a major endocrine organ, releasing signaling and mediator proteins, termed adipokines, via which adipose tissue communicates with other organs. Expansion of adipose tissue in obesity alters adipokine secretion, which may contribute to the development of metabolic diseases. Although recent profiling studies have identified numerous adipokines, the amount of overlap from these studies indicates that the adipokinome is still incompletely characterized. Therefore, we conducted a complementary protein profiling on concentrated conditioned medium derived from primary human adipocytes. Obesity has become a critical global health problem that frequently associates with the development of chronic diseases, including type 2 diabetes and cardiovascular dysfunction (1). It is now considered that adipose tissue is one of the major endocrine organs (2,3), besides acting as a lipid depot, providing an important function in storage and release of energy-rich substrates. By secreting a huge diversity of signaling and mediator molecules, termed adipokines, adipose tissue communicates with other tissues, such as liver, skeletal muscle, heart, brain, and vasculature (3)(4)(5)(6). Recent data indicate that these adipokines create a complex, interconnected network mediating the cross talk between these tissues (7,8). In obesity, enlargement of adipose tissue has been linked to a dysregulation of adipokine secretion and adipose tissue inflammation (9). This switch to a chronic state of low-grade inflammation represents a critical pathogenic link between obesity and the development of multifactorial diseases, such as type 2 diabetes and the metabolic syndrome.

SDS-PAGE/liquid chromatography-electrospray ionization tandem MS and two-dimensional SDS-PAGE/matrixassisted laser desorption ionization
Because of the relevance of adipose tissue in the progression of these common diseases, multiple unbiased, proteomic approaches have characterized the secretome from both rodent and human adipocytes and adipose tissue (10)(11)(12)(13)(14)(15)(16)(17). These studies have emphasized the complex nature of the adipokinome and have identified hundreds of adipokines. However, the amount of overlap between the identified adipokines thus far indicates that the adipokinome is still incompletely characterized.
Furthermore, it is largely unclear whether novel adipokines found in these approaches are dysregulated in obesity and type 2 diabetes. Because differentiated primary human adipocytes secrete factors, which induce insulin resistance in skeletal muscle cells and aberrant proliferation of smooth muscle cells (18,19), we have conducted an extensive proteomic profiling of conditioned media (CM) derived from differentiated, primary human adipocytes. This resulted in the identification of 347 proteins, of which to the best of our knowledge 44 have not been described as adipokines before. We subsequently assessed the regulation of (i) complement factor H (CFH) 1 , (ii) ␣B-crystallin (CRYAB), (iii) cartilage intermediate-layer protein (CILP), and (iv) heme oxygenase-1 (HO-1) in both primary human adipocytes and serum and adipose tissue samples from patients with obesity versus controls with normal body weight. The identified proteins have been implicated in multiple biological processes, thereby further highlighting the regulatory role of adipokines in intra-organ cross talk.

EXPERIMENTAL PROCEDURES
Adipocyte Isolation and Culture-Subcutaneous adipose tissue was obtained from lean or moderately overweight women (for proteomic analysis n ϭ 5, body mass index 28.5 Ϯ 5.7, including one obese woman, and aged 24.9 Ϯ 1.6 years; for biochemical analysis n ϭ 13, body mass index 26.2 Ϯ 0.6, and aged 44.6 Ϯ 3.3 years) undergoing plastic surgery. The procedure was approved by the ethical committee of the Heinrich-Heine-University (Duesseldorf, Germany) and written consent was obtained from each donor. All subjects were healthy, free of medication, and had no evidence of diabetes, according to information from the respective family doctors, and were known not to be affected by human immunodeficiency virus, hepatitis C, or tuberculosis. Preadipocytes were isolated by collagenase digestion of adipose tissue, as we have previously described (18). Isolated cell pellets were resuspended in Dulbecco's modified Eagles/Hams F12 medium supplemented with 10% FCS, seeded in 75-cm 2 culture flasks or six-well culture dishes and maintained at 37°C with 5% CO 2 . After overnight incubation, cultures were washed and further incubated in an adipocyte differentiation medium (Dulbecco's modified Eagles/Hams F12 , 33 mol/l biotin, 17 mol/l d-panthothenic-acid, 66 nM insulin, 1 nM triiodo-L-thyronine, 100 nM cortisol, 10 g/ml apo-transferrin, 50 g/l gentamycin, 15 mmol/l HEPES, 14 nmol/l NaHCO 3 , pH 7.4) for 15 days, with medium change every 2-3 days and addition of 5 M troglitazone for the first three days. The degree of differentiation was determined by oil red staining and induction of adiponectin expression. (Supplementary Fig. 1). Differentiated adipocytes were used for the generation of adipocyte-CM, as described previously (20). Macrophages were isolated from human adipose tissue, using a method described by Curat et al. (21). For hypoxia treatment, differentiated adipocytes were incubated with a gas mixture containing 1% O 2 , 5% CO 2 , and 94% N 2 in MIC-101 modular incubator chambers (Billups-Rothenburg, Del Mar, CA) at 37°C for indicated times.
Sample Preparation for Secretome Analysis-For sample preparation, CM from five different donors (200 ml derived from 7 ϫ 10 7 adipocytes) were collected. CM was first analyzed for the capacity to induce insulin resistance of the level of insulin-stimulated Akt phosphorylation in skeletal muscle cells, as described previously (18,19), and only active CM-inducing insulin resistance was further used. Prior to protein profiling studies, these CM were pooled and centrifuged for 20 min at 40,000 ϫ g at 4°C. Subsequently, supernatants were concentrated to a final volume of 200 l using Amicon TM Ultra 15 centrifugal filter devices (Millipore, Billerica, MA) with a cut-off mass of 3,000 Dalton. Protein concentrations were measured using Advanced Protein Assay (Cytoskeleton, Denver, CO). Concentrated CM (4.9 mg/ml) was diluted 1:3 in a buffer containing 25 mM Tris, 4% CHAPS (w/v), 7 M urea, and 2 M thio-urea, and stored as aliquots at Ϫ80°C until use.

SDS-PAGE and Protein Identification by Liquid Chromatography (LC)-MS-For
LC-MS analysis, aliquots of concentrated CM were initially separated by one-dimensional SDS-PAGE. 20 g of protein was mixed with 10 l of 2ϫ SDS-PAGE sample buffer and 3 l of 60 mM dithiothreitol. The samples were boiled for 10 min at 96°C and cooled to room temperature. After addition of 3 l of 150-mM iodoacetamide, the samples were incubated for 30 min in the dark before being subjected to SDS-PAGE on a 5% to 15% gradient gel. Proteins were visualized with Imperial Protein Stain (Thermo-Fisher, Bonn, Germany). The stacking gels were removed and the entire lanes (60 ϫ 5 ϫ 0.75 mm) were divided into 24 equally sized portions, based on different molecular weights. The bands were chopped into small cubes and washed three times with acetonitrile-water (1:1). Acetonitrile was added and removed to dehydrate the gel pieces before they were dried for 5 min in a speed vac. The dry gel pieces were rehydrated in an ice-cold solution of 12.5 ng/l Trypsin (sequencing grade, Promega) in 10 mM NH 4 HCO 3 . After 45 min on ice, excessive trypsin solution was replaced by 20 l of buffer without enzyme, and proteins were digested at 37°C for 4 h. The digest was stopped by the addition of 20 l 10% formic acid. Peptides were extracted for 30 min and the volume was reduced to 20 l before the extracts were stored at Ϫ80°C.
Liquid chromatography (LC)-MS data were acquired on a HCT ETD II ion trap mass spectrometer (Bruker Daltonics, Bremen, Germany), equipped with a nano ESI source (Bruker Daltonics). Samples were introduced by an easy nano LC system (Proxeon, Odense, Dennmark) using a 0.1-by-200-mm analytical column, self-packed with ReproSil-Pur C18-AQ, 5 m (Dr. Maisch, Ammerbuch, Germany). Eighteen microliters of the sample were aspirated into the sample loop and a total of 25 l was loaded onto the column using a flow rate of 2 l/min. Loading pump buffer was 0.1% formic acid. Peptides were eluted with a gradient of 0% to 35% acetonitrile (ACN) in 0.1% formic acid over 170 min at a column flow rate of 500 nl/min. Subsequently, the acetonitrile content was raised to 100% over 2 min and the column was regenerated with 100% acetonitrile for additional 8 min.
Data-dependent acquisition of MS and tandem MS (MS/MS) spectra was controlled by the Compass 1.3 software. MS1 scans were acquired in standard enhanced mode. Five single scans in the mass range from m/z 400 to m/z 1400 were combined for one survey scan. Up to three doubly and triply charged ions rising above a given threshold were selected for MS/MS experiments. Ultra scan mode was used for the acquisition of MS2 scans in the mass range from m/z 100 m/z 1600 and three single scans were added up. The ion charge control value was set to 250,000 for all scan types. Raw data were processed with Data Analysis 4.0 (Bruker Daltonics) and xml-formatted peak lists were transferred to Proteinscape 2.1 (Bruker Daltonics).
MASCOT 2.2 (Matrix Science, London, UK) was used to search a composite decoy database, which was built from SwissProt_57.4 (468851 sequences; 166,149,756 residues). The composite database was generated with the Perl script makeDecoyDB (Bruker Daltonics), which added a randomized sequence and a tagged accession number for each entry. The tagged accessions were used for the calculation of false positive rate in Proteinscape 2.1. Searches were submitted via Proteinscape (Bruker Daltonics) and the following parameter settings: enzyme "trypsin," species "human," fixed modifications "carbamidomethyl," optional modifications "Methionine oxidation," and missed cleavages "2." The mass tolerance was set to 0.4 Da for peptide and fragment spectra. Protein lists were compiled in Proteinscape. Peptide hits were accepted when the ion score exceeded a value of 20. Protein hits required at least one peptide hit exceeding a peptide score of 40. In addition, the hits to decoy entries were used to calculate a minimal protein score, which is required to keep the false positive rate below 2% on the protein level (22).
Two-dimensional PAGE and Protein Identification by Matrix-Assisted Laser Desorption Ionization-MS-For two-dimensional PAGE analysis, aliquots (150 g) of concentrated CM were separated in the first dimension by isoelectric focusing using pH 4 -7 and pH 6 -9 linear IPG strips performed on a MultiPhor II electrophoresis unit (GE Healthcare, Freiburg, Germany), and in the second dimension by large format SDS-PAGE (12%), as previously described (23). Subsequent to electrophoretic separation, gels were stained with a ruthenium fluorescent stain (24) and protein patterns were visualized by laser scanning using blue laser source (457 nm) on a Typhoon 9400 (GE Healthcare) and a resolution of 100 m. Detection of protein spots and calculation of relative spot abundances was carried out automatically using Proteomweaver 4.0 image analysis software (Bio-Rad).
Protein spots exceeding intensity levels of 0.2 and matching gel sets of six individual two-dimensional gels were selected for protein identification by MALDI-MS. Proteins matching these criteria were subsequently excised from four replicate gels (4 -7 and 6 -9 each) using a Gelpix spot picker (Genetix, Dornach, Germany). For in-gel digestion, gel pieces were washed for 10 min in digestion buffer (10 mM NH 4 HCO 3 ) and digestion buffer containing 50% acetonitrile (1:1, v/v). Acetonitrile was added and removed to dehydrate the gel pieces. The dry gel pieces were rehydrated in an ice-cold solution of 3.5 ng/l Trypsin (sequencing grade, Promega) in 10 mM NH 4 HCO 3 . Proteins were digested at 37°C for 4 h. Peptides were extracted for 30 min with 10 l of 0.1% trifluoroacetic acid and directly applied to a MALDI Pre-spotted AnchorChip target (Bruker Daltonics) according to the manufacturer's instructions.
Subsequently, samples were analyzed in a time-of-flight Ultraflex-Tof/Tof mass spectrometer (Bruker Daltonics). Acquired mass spectra were automatically calibrated and annotated using Compass 1.3 software (Bruker Daltonics) and xml-formatted peak lists were transferred to Proteinscape 2.1 (Bruker Daltonics,). Because all experiments were performed on primary human adipocytes, MS spectra from each individual spot were used to search a human subset of Swiss-Prot (Sprot_57.4, 20401 protein entries) nonredundant database using Mascot search engine (Version 2.2; Matrix Science) in consideration of the following settings: enzyme "trypsin," species "human," fixed modifications "carbamidomethyl," optional modifications "Methionine oxidation," and missed cleavages "1." Mass tolerance was set to 50 parts per million (ppm). Using these settings, a mascot score of greater than 70 was taken as significant (p Ͻ 0.01). Calculated pI and molecular mass data were obtained by Mascot. For peptides matching to different isoforms or multiple members of a protein family, we used the following reporting criteria: The experimental pI and molecular mass taken from the two-dimensional gels were compared with the theoretical data of the different isoforms/ protein members. If no conflicts in molecular mass or pI were found, the isoform/protein member with the highest mascot score was reported. For verifying the results, each protein spot was picked and identified from at least three physically different two-dimensional gels.
Prediction and Annotation of Secretory Proteins-Secretory protein prediction and functional annotation was completed by using different, independent methods. First, protein information of all identified proteins was extracted from the Swiss-Prot/TrEMBL database (http://www.expasy.ch/sprot/). Gene names were used to screen the Bio-GPS gene annotation portal (http://biogps.gnf.org) for expression data in adipose tissue (supplemental Table S1).
Clinical Study for the Analysis of Adipokine Concentration in Serum and Adipokine Expression in Subcutaneous and Visceral Adipose Tissue-23 male obese patients scheduled for bariatric surgery and 17 lean age-matched controls undergoing an elective abdominal surgery were recruited at Ghent University Hospital. For all patients, anthropometric and routine blood parameters were assessed (supplemental Table S2). Fasting blood samples were collected and frozen at Ϫ80°C, and adipose tissue biopsies were fixed for microscopic evaluation of adipocyte surface area analysis or frozen at Ϫ80°C for protein expression analysis. The study protocol was approved by the local ethics committees and all participants gave written informed consent (registration no B67020084018).
Determination of Adipose Tissue Cell Size-Visceral and subcutaneous adipose tissue specimens were obtained from the patients at the end of the surgical intervention. Adipose tissue was fixed in buffered 4% formol solution (Klinipath, Olen, Belgium) at room temperature. Further fixation, dehydration, cleaning, and paraffin impregnation of tissues was performed (Tissue Tek Vip; Sakura, Heppenheim, Germany) and tissues were embedded with TBS 88 Paraffin Embedding System (Medite, Burgdorf, Germany). Hematoxylin-eosin staining and film coverslipping of 3-m slides was completed by a Tissue Tek Prisma (Sakura). Adipocyte size was assessed by using the approach of Tchoukalova et al. (25) Digital photographs of the paraffin slides were taken with a Mirax Midi camera (Zeiss, Jena, Germany) and the average surface area of 10 adipocytes of each slide was calculated using the Mirax Viewer software (Zeiss).
Immunoblotting-Adipocytes and macrophages were treated as indicated and lysed in a buffer containing 50 mM HEPES, pH 7.4, 1% TritonX100, Complete protease inhibitor and PhosStop phosphatase inhibitor mixture (both Roche, Mannheim, Germany). After incubation for 2 h at 4°C, the suspension was centrifuged at 10,000 ϫ g for 15 min. Thereafter, 5-10 g of adipocyte lysates were separated by SDS-PAGE using 10% horizontal gels and transferred to polyvinylidene fluorid membranes in a semidry blotting apparatus. Membranes were blocked with Tris-buffered saline containing 0.1% Tween and 5% nonfat dry milk, and were subsequently incubated overnight with a 1:1000 dilution of the appropriate antibodies. After washing, filters were incubated with secondary horseradish peroxidase coupled antibody and processed for enhanced chemiluminescence detection using Immobilon horseradish peroxidase substrate (Millipore). Signals were visualized and evaluated on a LUMI Imager (Boehringer, Mannheim, Germany) or VersaDoc 4000 MP (BioRad, Munich, Germany) workstation. Detection of actin was used to normalize for protein loading. Analysis of serum probes by Western blot was done by application of defined serum volumes not requiring normalization for protein loading. Antibodies were purchased from Abcam (CFH, CRYAB, HO-1, actin) and R&D Systems (CILP).
ELISA-HO-1 release by cells and serum concentration was determined by ELISA (R&D Systems, Stressgene). The assay was performed in duplicate according to the manufacturer's instructions.
Gene Expression-RNA was isolated from 100-200-mg adipose tissue biopsy with a RNeasy mini kit for lipid-rich tissue using the Qiacube robot (Qiagen, Hilden, Germany). DNA remove was done by a DNAse I digestion, followed by a clean up with RNeasy mini column using the Qiacube robot. For complementary DNA (cDNA) synthesis 500 ng total RNA was applied to Superscript first-strand cDNA Kit (Invitrogen, Karlsruhe, Germany) according to the manufacturer's instructions. Measurements of relative CILP1 expression were performed on a StepOne Plus Real-Time PCR system (Applied Biosystems, Carlsbad, CA) using SybrGreen and Qiagen Quantitect primer assay for human CILP1. The cycling conditions comprised a polymerase activation at 95°C for 10 min followed by 40 cycles at 95°C for 15 s, 60°C for 30 s, and 72°C for 30 s. Messenger RNA (mRNA) expression was normalized to RPS18 mRNA content and expressed as arbitrary units.
Presentation of Data and Statistics-Data are expressed as mean Ϯ SE. The Shapiro-Wilcoxon test was used to test the Gaussian distribution of biological parameters. The Student's t test was used for all experiments where only two experimental groups were compared; analysis of variance (ANOVA) followed by p for linear trend posttests were used for experiments where more than two experimental groups were compared. Correlations were performed by Pearson. All statistical analyses were completed using JMP statistics software (SAS Institute , Cary, NC) or Prism (GraphPad, La Jolla, CA), considering a p value of less than 0.05 as statistically significant. Corresponding significance levels are indicated in the figures.

RESULTS
Proteomic Profiling of the Adipocyte Secretome-As summarized in Fig. 1, two complementary orthogonal approaches-1-dimensional SDS-PAGE/LC-electrospray ionization (ESI)tandem MS (MS/MS) and two-dimensional SDS-PAGE/ MALDI-MS-were used to profile the CM derived from primary human adipocytes. To account for biological variability, CM was collected from five different donors. Prior to protein profiling, CMs derived from the different donors were pooled to make profiling more feasible. Subsequently, CM was concentrated by a factor of 1,000 in order to achieve a protein concentration in the mg/ml range. Using LC-ESI-MS/MS analysis, automated data processing by Proteinscape TM 2.1 led to the consistent identification of 341 unique protein species in protein slices (24 slices per lane), derived from two replicate 5-15% gradient SDS-PAGE gels with a false positive rate below 2% (see supplemental Table S1). Complementary to this approach, the CM was analyzed using two-dimensional SDS-PAGE/MS. Image analysis of the merged large format two-dimensional gels covering a pH range of 4 -9 (overlapping gel couple pH 4 -7 and pH 6 -9, exemplarily shown in supplemental Fig. S2) reproducibly detected more than 1200, nonredundant protein spots within the CM. Consistently detected protein spots were excised from four replicate gelcouples, subjected to in-gel digestion and MALDI-MS. Proteinscape TM 2.1 data processing identified a total of 351 protein spots, which were consistently found in at least three technical replicate gels. Subsequent analysis assigned these 351 spots to 89 different protein species (supplemental Table S1).
Overall, the combined MS data resulted in the identification of 347 distinct protein species. Of these, 83 proteins were detected via both approaches (supplemental Table S1).
Secretory Protein Selection and Comparison with Reported Secretomes-We used different web-based bioinformatic tools (Swiss-Prot/TrEMBL, BioGPS, Ingenuity IPA8.5, Sig-nalP3.0, SecretomeP2.0, NCBI PubMed) to assess whether the 347 proteins could be secretory proteins. Applying these consecutive filter methods, 263 proteins (263/347, 76%) were predicted to be or to have been annotated as secretory proteins (supplemental Table S1). Of these, 219 proteins have been described in previous profiling studies on adipocytes or adipose tissue (Table I, supplemental Table S1). However, to the best of our knowledge, the 44 proteins listed in Table II have not been reported in previous approaches, and can therefore be considered as potential novel adipokines.
Validation of Novel Adipokines Using Human Adipocytes and Adipocyte-conditioned Medium-The available webbased bioinformatical and Gene Onthology Biological Process and Molecular Function annotation tools,ascribe a wide array of biological functions to the adipokines identified in this study. Depending on the immediate availability of suitable biochemical tools, we therefore selected four adipokines eliciting completely distinct biological functions for further validation. These are (i) CFH, which has been implicated in inflammation; (ii) CRYAB, which has been implicated in apoptosis; (iii) CILP, which participates in extracellular matrix structure and remodeling; and (iv) HO-1, a component of the oxidative stress response.
Although CFH and CRYAB were recently described as novel adipokines (15)(16)(17), these proteins have not been validated so far. Western blot analysis of concentrated CM confirmed the presence of CRYAB, CFH, HO-1, and CILP, thereby validating the secretion of these factors from primary human adipocytes ( Fig. 2A). Like adiponectin, which was used as a control marker for adipocyte differentiation, protein expression of CRYAB and CILP gradually increased during the transition of preadipocytes in mature primary human adipocytes (Fig. 2B). Protein expression of HO-1 peaked at day 7 of differentiation and was slightly decreased in mature cells (day 14; Fig. 2B). The abundance of CFH gradually decreased during adipocyte differentiation (Fig. 2B, Fig. 3A). Finally, all of these factors are more highly expressed in adipocytes when compared with macrophages isolated from human adipose tissue (Fig. 2C).
Although CFH protein expression is higher in preadipocytes when compared with mature adipocytes (Fig. 3A), CFH expression can be up-regulated in adipocytes by troglitazone, TNF␣, insulin, and hypoxia (Fig. 3B). Furthermore, CFH levels in serum from obese subjects, as determined by Western blotting, were elevated versus subjects with normal body weight (Fig. 3C).
CRYAB expression is strongly induced during adipogenesis (Fig. 3D), reaching levels that are 10 times higher in mature adipocytes than in preadipocytes. CRYAB expression is not regulated by troglitazone, TNF␣, insulin, adiponectin, or hypoxia (data not shown), and the amount of CRYAB detected in serum by Western blotting is not significantly different between lean and obese subjects (Fig. 3E).
CILP is a completely novel adipokine, first identified by our research team. The antibody used to validate CILP recognizes both the 140-kDa precursor form as well as the 90-kDa secreted N-terminal form. The abundance of the CILP precursor significantly increased during differentiation, reaching about 10 times higher concentrations in mature adipocytes compared with preadipocytes (Fig. 4A). Although the same trend could be observed for the abundance of the 90-kDa secreted product, this increase failed to reach significance. It seems plausible to ascribe this to release of the 90-kDa form during differentiation, based on prevalent abundance of this form in CM (Fig. 2B). Western blot analysis revealed that the 140-kDa precursor form is significantly decreased by troglitazone, while the 90-kDa secreted product is significantly inhibited by TNF␣ in cultured adipocytes (Fig. 4B). Immunoblotting failed to produce reproducible results in adipose tissue biopsies (data not shown). However, using real-time PCR, we observed that expression of CILP1 mRNA was up to 30-fold increased in subcutaneous versus visceral adipose tissue (Fig. 4C). Finally, plasma levels of CILP were significantly decreased in serum from obese males versus normal weight controls (Fig. 4D). Notably, only the secreted 90-kDa form was observed in serum. Protein expression of HO-1 significantly increases with adipocyte differentiation (Fig. 5A and 5B). Although HO-1 expression is not affected by TNF␣ (data not shown), its release is significantly decreased by this proinflammatory factor (Fig.  5C). ELISA analysis showed that circulating levels of HO-1 were significantly increased in obese subjects compared with lean controls (Fig. 5D). Interestingly, HO-1 serum concentrations correlate with subcutaneous adipocyte size (Fig. 5E).
Measurement of CRYAB and HO-1 expression in adipose tissue biopsies from subcutaneous and visceral adipose tissue of lean and obese men revealed that both proteins are expressed significantly higher in adipose tissue of obese subjects ( Fig. 6A and 6B). Although the increase in HO-1 expression in visceral fat is similar in lean and obese subjects (Fig.  6B), CRYAB expression in visceral adipose tissue of obese subjects is significantly higher compared with lean controls (Fig. 6A). DISCUSSION Proteomic Profiling of the Adipocyte Secretome-Detailed characterization of the human adipokinome is necessary to increase our understanding of the role of adipocytes in disease pathophysiology. Therefore, we profiled the adipokinome from primary human adipocytes. To enable a more general analysis, and considering the substantial biological variability in humans, we collected and pooled CM derived from five different donors. For proteomic profiling,  , gel-based examination provides the opportunity to display unrivaled additional information concerning protein isoforms or putative protein modifications, and 2DE maps also can serve as a reference for further quantitative profiling studies (e.g. DIGE). This is illustrated by our profiling results, where each identified protein species is represented, on average, by more than three distinct spots (351 protein spots, 89 distinct protein species) varying in molecular mass or pI. Based on this information, we already described pigment epithelium-derived factor as a major biological active secretion product of the human adipocyte (present in 15 spots and accounting for 6% of total spot intensities) (26). Furthermore, we also discovered dipeptidyl peptidase 4 (DPP4) as novel adipokine (27), further confirming the usefulness of our approach. A major challenge in secretome analysis is to discriminate between actually secreted proteins and those that may be contaminants introduced due to the operational procedure. To address this issue, we filtered the identified proteins for properties, disclosing them as secretory proteins. The testing of all proteins by hierarchical analysis, including screening of expression data and literature, as well as using signal sequence prediction programs, led to the identification of 263 putative secretory proteins. Applying this strategy utilizing mainly theoretical analysis tools assigned 84 proteins as "nonsecretory." A major limitation of this analysis strategy is the risk of losing interesting target proteins without further validation. In order to uncover those proteins and achieve additional information, usage of Brefeldin A treatment at 20°C has been described to monitor active secretion (28,29). Brefeldin A blocks major ER/Golgi-dependent, as well as independent, secretion pathways and therefore facilitates identification and discrimination of genuine secreted proteins. On the other hand, proteins released by cleavage processes like DPP4 (27), or proteins secreted by mechanisms not affected by Brefeldin A, will not be covered. Nevertheless, the high number of proteins identified in our study demonstrates the efficiency of our integrated profiling approach, which is in line with very recently published studies of the human adipocyte secretome (15,17). Comparison analysis reveals that 219 (83%) of our putative secretory proteins were already reported in previous studies (Table I), based on secretome analysis from other origins such as 3T3L1 adipocytes, tissue explants, or isolated adipocytes. A very recent study conducted by Rosenow et al. (16), applying an analogous approach to the human SGBS adipocyte cell line, identified only 80 secreted proteins, potentially indicating limitations of the cell model used.
Although about 80% of the proteins are already published, our approach has identified 44 additional proteins (Table II) that we consider as novel adipokines secreted from the human adipocyte. Interestingly, to date, more than 700 different proteins are described as being potentially secreted from the

FIG. 2. Representative Western blots of CILP, HO-1, CRYAB, and CFH protein level and release by human primary adipocytes.
Adipocytes were differentiated and concentrated CM was generated as described in the text. A, Release of novel adipokines at day 14 of differentiation. 1-5 l of concentrated CM were analyzed by SDS-PAGE and Western blotting. B, Protein level of novel adipokines during adipocyte differentiation. 10 g of total lysates were analyzed by SDS-PAGE and Western blotting. Signals were detected by ECL. C, 10 g of total lysates derived from adipocytes (Ad) and macrophages (MØ) were analyzed by SDS-PAGE and Western blotting, with subsequent signal detection by ECL.
adipose tissue, irrespective of species differences. Though detection and identification of putative novel adipokines is the irreplaceable basis, this alone will not be sufficient to enhance our current knowledge of the endocrine function of adipose tissue. Therefore, the candidate proteins have to be further validated regarding their expression, secretion, and function.
Validation of Novel Adipokines-To address their putative biological relevance, we have chosen four candidate proteins for further validation experiments. CFH, CRYAB, CILP, and HO-1 are representatives of different critical functional pathways (inflammation, apoptosis, extracellular matrix structure, and oxidative stress) that have been demonstrated to be dysregulated in obesity (15,30).
CFH is an example of a large group of adipocyte-secreted proteins involved in inflammation. Showing a significant reduction of CFH protein expression during adipogenesis, we could corroborate findings from two other studies demon-strating that, at the mRNA level, this factor is predominantly expressed in preadipocytes but is also present in adipocytes (31,32). In addition, CFH in serum is elevated in insulinresistant subjects, and CFH expression was found to be elevated in subcutaneous adipose tissue (31). Consistently, we found higher CFH content in blood of obese men and an upregulation of CFH in adipocytes treated with TNF␣, insulin, and hypoxia. Collectively, these data point to a possible higher contribution of adipose tissue to circulating CFH levels in obesity due to adipose tissue inflammation. Whether adipocytes or preadipocytes contribute to increased adipose CFH levels in obesity remains an open question that requires further analysis.
The protection against apoptosis and oxidative stress is a function of CRYAB. It is known that CRYAB is expressed in adipose tissue (33). Here, in agreement with a recent report by Kim et al. (15), we confirm CRYAB as an adipokine. In addi- tion, the present study also demonstrates that CRYAB is increased during adipogenesis. Although CRYAB was not affected by stimuli that affect insulin sensitivity in adipocytes, CRYAB expression was found to be significantly increased in subcutaneous and visceral adipose tissue of obese patients compared with lean controls. Therefore, further characterization of CRYAB regulation in adipose tissue of obese patients and patients with the metabolic syndrome could be of potential interest. The increased relative abundance of CRYAB in visceral adipose tissue of obese patients could be especially important, since visceral adipose tissue is a critical player in the development of obesity-related complications, whereas subcutaneous adipose tissue may act merely as an innocent "bystander" (34). It should further be noted that CRYAB con-centrations, measured by Western blot in serum, are not different between lean and obese patients, suggesting that other tissues such as heart and skeletal muscle regulate circulating levels of CRYAB (35,36).
CILP is a secreted glycoprotein that resides in the extracellular matrix (37,38), maintaining cartilage homeostasis (39). Although CILP expression was originally suggested to be restricted to cartilage (37,(40)(41)(42)(43), we confirm Bio-GPS data showing that CILP is highly expressed in adipose tissue. Furthermore, we demonstrate, for the first time, that CILP is also secreted by primary human adipocytes. The gene product of CILP is a 140-kDa precursor protein for two secreted, proteolytically generated products, a 90-kDa N-terminal CILP, and a 62 kDa C-terminal domain. The N-terminal CILP domain FIG. 4. Protein level of CILP in adipocytes and relative CILP serum concentration in lean and obese subjects. A, Human primary adipocytes were differentiated as described in the text, and CILP protein level during differentiation was analyzed by SDS-PAGE and Western blot. Data were normalized to the protein level of actin and are expressed relative to day 0. Data are mean values Ϯ S.E., n Ն 3, *p Ͻ 0.05 versus preadipocytes. B, Differentiated adipocytes were treated as described in Fig. 3 legend. CILP protein level was analyzed by SDS-PAGE and Western blotting. Data were normalized to the protein level of actin and are expressed relative to unstimulated control (Con). Data are mean values Ϯ S.E., n ϭ 5, *p Ͻ 0.05 versus control. C, CILP-1 mRNA expression (Ra) in human adipose tissue. Expression of CILP-1 was determined by real-time PCR in subcutaneous (sc) and visceral (visc) adipose tissue biopsies collected during abdominal surgery of normal weight males (n ϭ 12). Data were normalized for RPS18 expression and expressed as mean Ϯ S.E. *p Ͻ 0.05 versus subcutaneous adipose tissue. D, Relative serum concentration of CILP was determined in samples obtained from lean (n ϭ 12) and obese (n ϭ 9) subjects participating in study 1. Sera samples were diluted and analyzed by SDS-PAGE and Western blotting. Data are mean values Ϯ S.E., *p Ͻ 0.05 versus lean controls.
polypeptide was shown to function as an insulin-like growth factor (IGF-1) antagonist and binding partner for TGF-␤1 (44,45). Several studies might show that CILP dysfunction contributes to various diseases affecting the cartilage. In this regard, the expression of CILP rises substantially in association with aging and also in the early stages of osteoarthritis and rheumatoid arthritis (37). CILP acts as negative regulator of TGF-␤1 by binding it directly in vitro, thereby disrupting normal TGF-␤ activity (45). Furthermore, TGF-␤ strongly inhibits adipogenesis and the amount of fat in adipocytes (46). Because CILP plasma levels are decreased in obese men, one could speculate that this decrease could result in a higher TGF-␤1 activity, and thus could result in an abrogation of adipogenesis. Accordingly, Fain et al. demonstrated that TGF-␤1 release by human adipose tissue is enhanced in obesity (47). Yet whether CILP levels are associated with TGF-␤1 remains to be characterized. HO-1 is a ubiquitously expressed enzyme that is involved in the reduction of oxidative stress and inflammation (48). It can be found in the circulation. Recent studies have shown elevated plasma levels in several chronic disorders, like Parkinson's disease and hemophagocytic syndrome, and in newly diagnosed type 2 diabetics (49)(50)(51). Until today, the sources of circulating HO-1 are unknown and it is postulated that plasma HO-1 is due to "leakage" of the enzyme from tissues to the plasma compartment (49,52). Here, we demonstrate that HO-1 is an adipokine expressed and released by human primary adipocytes in a differentiation-dependent manner and that circulating levels of HO-1 are increased in obese men, correlating with the size of subcutaneous adipocytes. Our data further suggest that adipocytes actively secrete HO-1 and possibly release it into the circulation, and that this secretory activity is dependent on adipocyte size.
We also demonstrate that HO-1 secretion is down-regulated by TNF␣, with no effect on its expression level, which could be a time-course-dependent effect. As other studies of human peripheral monocytes and human chondrocytes could demonstrate a TNF␣-induced reduction of HO-1 expression (53,54), as well as that HO-1 induction in an animal model leads to a reduction in the circulating TNF␣ level (55), it is assumable that there is a direct link between HO-1 and TNF␣, which has to be further investigated.
It is described that HO-1 reduces adipogenesis, and its induction in animal models is associated with decreased body weight, improved adipokine profile, and insulin sensitivity (56). These observations imply that HO-1 is both a marker of cellular stress (57) and acts as an important factor in maintaining cellular and tissue homeostasis. As our results show that HO-1 increases during adipocyte differentiation and is elevated in subcutaneous and visceral fat of obese patients, we suggest that HO-1 might play an important role in adipose tissue inflammation. Moreover, the elevated plasma levels in obese patients described in this study, and elevated levels of circulating HO-1 in newly diagnosed type 2 diabetics (49), suggest that HO-1 may be a potent biomarker for obesity and obesity-associated disorders like the metabolic syndrome and type 2 diabetes. CONCLUSIONS Taken together, the presented study contributes to our understanding of the complex endocrine function of human adipocytes. Extensive profiling led to the identification of 263 proteins to be released from adipocytes, including 44 putative novel adipokines. (see also http://www.diabesityprot.org/) Functional studies of four selected adipokines-CFH, CRYAB, HO-1, and CILP-provide valuable clues regarding their regulation during adipogenesis and the relationship of their circulation levels to obesity. Especially, HO-1 exhibiting significantly elevated plasma levels in obesity and increased expression in visceral fat of obese patients may be an attractive candidate adipokine for further investigations in the field of obesity and obesity-associated disorders like the metabolic syndrome and type 2 diabetes. For the novel adipokines presented in this study, additional basic research and clinical studies could provide insight into their involvement in obesityrelated diseases and their potential use as biomarkers.