Label-Free Protein Profiling of Adipose-Derived Human Stem Cells

May 23, 2011 - Patel , V. J.; Thalassinos , K.; Slade , S. E.; Connolly , J. B.; Crombie , A.; Murrell , J. C.; Scrivens , J. H. A comparison of label...
1 downloads 0 Views 4MB Size
ARTICLE pubs.acs.org/jpr

Label-Free Protein Profiling of Adipose-Derived Human Stem Cells under Hyperosmotic Treatment Elizabeth S. Oswald,†,‡ Lewis M. Brown,*,‡,§ J. Chlo€e Bulinski,§ and Clark T. Hung*,† †

Department of Biomedical Engineering, §Department of Biological Sciences, Columbia University, New York, New York, United States

bS Supporting Information ABSTRACT: Our previous work suggested that treatment of cells with hyperosmotic media during 2D passaging primes cells for cartilage tissue engineering applications. Here, we used label-free proteomic profiling to evaluate the effects of control and hyperosmotic treatment environments on the phenotype of multipotent adipose-derived stem cells (ASCs) cultivated with a chondrogenic growth factor cocktail. Spectra were recorded in a data-independent fashion at alternate low (precursor) and high (product) fragmentation voltages (MSE). This method was supplemented with data mining of accurate mass and retention time matches in precursor ion spectra across the experiment. The results indicated a complex cellular response to osmotic treatment, with a number of proteins differentially expressed between control and treated cell groups. The roles of some of these proteins have been documented in the literature as characteristic of the physiological states studied, especially aldose reductase (osmotic stress). This protein acted as a positive control in this work, providing independent corroborative validation. Other proteins, including 50 -nucleotidase and transgelin, have been previously linked to cell differentiation state. This study demonstrates that label-free profiling can serve as a useful tool in characterizing cellular responses to chondrogenic treatment regimes, recommending its use in optimization of cell priming protocols for cartilage tissue engineering. KEYWORDS: label-free, proteomics, data mining, adipose, stem cells, chondrocytes, cartilage

’ INTRODUCTION Osteoarthritis (OA), the destruction of articular cartilage that covers the surfaces of our mobile joints (e.g., knee and hip), costs an estimated $128 billion in annual healthcare and job-related losses in the United States. The poor healing capacity of articular cartilage has led to intense research toward development of cellbased therapies for cartilage repair, including efforts to identify cell sources with potential clinical relevance for cartilage tissue engineering. Challenges associated with these sources include donor site availability and capability of cells to produce levels of extracellular matrix sufficient for the survival of engineered tissues upon implantation in the knee joint environment. Cell expansion and priming with chemical or physical factors are two interleaved approaches often attempted to address these challenges.1 In these strategies, culturing of cells on a twodimensional (2D) tissue culture dish provides a platform for increasing cell number (expansion) as well as an opportunity for applying chemical and physical stimuli (priming) that can induce differentiation of cells toward a desired lineage, for example, a chondrogenic lineage. Cells subjected to expansion and priming have previously yielded more robust tissue production when subsequently seeded in 3D scaffolds that are supportive of the chondrogenic phenotype.2,3 Methods for accurately and rapidly assessing the impact of 2D priming techniques on cell cartilage tissue generation in subsequent 3D culture are currently lacking. Comparison of RNA r 2011 American Chemical Society

levels expressed in cells during 2D culture in control and experimental conditions may capture only transient phenomena. Furthermore, RNA levels may not correlate well with production of functional tissue proteins, given protein synthetic capability and the extensive trafficking and post-translational modification pathways that cartilage tissue proteins must undergo before secretion from the cell. Additionally, one-by-one evaluation of the efficacy of each 2D treatment condition, by comparison of tissue component elaboration in subsequent 3D cultures, requires a large investment of time and resources in both the 2D expansion and 3D culture stages of the experiment. Therefore, an ability to identify during 2D expansion the experimental treatments that best improve cells’ capacity to form functional engineered cartilage tissues in subsequent 3D culture would enhance the rapid optimization of cartilage tissue engineering protocols. Proteomics techniques could allow the rapid assessment of the influence of 2D treatment conditions on cell protein, rather than RNA, production at early stages of cell culture.4 However, in spite of this potential, few studies have applied modern proteomic technologies to the biological assessment of chondrocytes and chondrogenic potential.511 Studies using human cell sources, as well as proteomics techniques that provide quantitative Received: January 12, 2011 Published: May 23, 2011 3050

dx.doi.org/10.1021/pr200030v | J. Proteome Res. 2011, 10, 3050–3059

Journal of Proteome Research information with multiple replicates, are particularly scant. Of the large-scale protein profiling strategies currently available, labelfree protein profiling has attracted interest due to its simplicity of implementation and its potential to support complex experimental designs. Increased reproducibility of chromatography at nanoliter per minute flow rates, coupled with increased sensitivity and resolution of mass spectrometers, have reinforced the attractiveness and feasibility of the label-free protein profiling approach. In a variant of the method, spectra are recorded at alternately low (precursor) and high (product) fragmentation voltages, in a data-independent process called MSE acquisition,12 allowing high data density and predictable data acquisition. The conventional approach to protein identification and quantification employs data-dependent acquisition (DDA). Published findings from a number of groups indicate that MSE is effective for a wide range of biological systems including microbial cells,12 mammalian tissues,13 tumor samples,14 stem cell knockout cell lines,15 and many other applications.1621 Therefore, we describe here our use of label-free proteomics to detect reproducible alterations in chondrocyte precursor cell phenotype upon application of a potential chondrogenic stimulus in 2D cell culture. We prepared proteomic samples from multipotent ASCs that had been expanded in 2D in control or physiologically relevant hyperosmotic culture environments22 that included a chondrogenic growth factor cocktail. We chose the hyperosmotic stimulus based upon our laboratory’s previous success in applying physiologic stimuli in cartilage tissue engineering applications and upon preliminary data indicating that expansion of human chondrocytes in media of higher osmolarity increases cells’ expression of a gene (aggrecan) related to production of cartilage tissue (unpublished data). Additionally, we chose ASCs as a clinically relevant allogeneic and autologous cell type that is advantageous in donor tissue accessibility and abundance relative to other multipotent cell types, due to the prevalence of elective cosmetic liposuction procedures. The results of this study provide an initial assessment of the feasibility of using proteomics to evaluate the effects of 2D cell priming treatments on cell phenotype, working toward our ultimate goal of using this technology to rapidly optimize such treatments for 3D cartilage engineering applications.

’ MATERIALS AND METHODS Cell Culture

ASC medium was prepared from a 1:1 ratio of high-glucose (4.5 g/L) DMEM:DMEM/F12 containing 3.25 mM L-glutamine (Invitrogen) by the addition of 1% penicillin-streptomycin (Invitrogen), 0.1% gentamycin (Invitrogen), 0.5% Fungizone (Invitrogen), 100 nM dexamethasone (Sigma), and 10% FBS (Invitrogen). Control and treatment osmotic media were prepared from a base medium consisting of high-glucose (4.5 g/L) DMEM containing 4 mM L-glutamine (Invitrogen), 10 mM each HEPES, BES, and TES buffers (Mediatech), 1 each MEM and nonessential amino acids (Mediatech), 1% penicillin/streptomycin (Invitrogen), and 5% FBS (Atlantic Biologicals). This base medium was diluted with deionized water to 300 mOsM (control medium), from which 400 mOsM media (treatment medium) was prepared by the addition of NaCl (biotechnology grade, Sigma). Both osmotic media were further supplemented with growth factor cocktail (1 ng/mL TGF-β3, 10 ng/mL PDGF, 5 ng/mL FGF-2, and 5 ng/mL EGF), a chondrogenic cocktail similar to that used previously in our laboratory.23

ARTICLE

Twice-passaged human ASCs isolated from the abdominal fat of a 42-year-old female patient were obtained in a cryopreserved state from Dr. Kacey Marra, at the University of Pittsburgh (IRBexempt). Cells were thawed and expanded to ∼90% confluence for two additional passages in tissue culture-treated flasks containing ASC medium. At passage five, cells were split into flasks containing either control or treatment media (plating density 1.4  102 cells/cm2). Cells were grown until the control group reached ∼90% confluence (3 days). Due to slight differences in the cell doubling rate between osmotic groups at each passage, treatment group cells were at ∼80% confluence when control group cells were at ∼90% confluence. Cells were then subjected to an additional passage in media of the same osmolarity at the same density (8 days). Upon confluence at passage six, cells were plated for a final passage (three dishes per osmotic group) at 6  103 cells/cm2 and maintained in the same osmotic media used in previous passages. Once the control group had reached 90% confluence (10 days), cells from each dish were harvested as follows for proteomics analysis, yielding three unique biological replicate samples per osmotic culture condition (control and treatment). Sample Preparation

Cells were washed three times with 10 mL of ice cold PBS. Careful attention was paid to the cell washing procedure, to rid the cell samples of excess BSA and other proteins from the culture media. Cells were lysed in 0.3% SDS, Tris-buffered saline with 1% protease inhibitor cocktail (Sigma), precipitated using methanol/chloroform,24 then dissolved in 0.1% Rapigest (Waters Corp.). Dithiothreitol (6 mM) was added; the solution was sonicated in a bath sonicator for 5 min and then boiled for five minutes. Approximate protein content in each sample was estimated with the Bradford Protein Assay (Biorad). Cysteines were alkylated with iodoacetamide. Proteins were digested with trypsin (6 ng/μL (#V511A, Promega Corp.) in 50 mM NH4HCO3) and 50 fmol of a digest of yeast alcohol dehydrogenase was added as an internal detection control. Chromatography and Mass Spectrometry

Three chromatograms were recorded for each of six biological replicates (three isotonic, three hypertonic), yielding 18 chromatograms. Prior to analytical separation on a NanoAcquity UPLC (Waters Corp.), peptides were trapped on a Symmetry C18 Trap column, 5 μm particles, 180 μm  20 mm (Waters Corp.), for 1 min at 15 μL/minute in 1% solvent B (0.3% formic acid in acetonitrile)/99% solvent A (0.3% formic acid, aqueous). Peptides were analyzed in a 120-min chromatogram on a 75 μm ID x 10 cm reverse phase 1.7 μm particle diameter bridged ethyl hybrid (BEH) C18 column at a flow rate of 300 nL/minute. For the analytical separation, Solvent B was increased in a 90-min linear gradient between 3 and 40%, and postgradient cycled to 95% B for 7 min, followed by postrun equilibration at 3% B. Spectra were recorded in V-positive mode with a Synapt quadrupole-time-of-flight mass spectrometer (Waters Corp). Source settings included extraction cone at 3.5 V, sampling cone at 24 V, and source temperature 80 °C. Collision energy was held at 6 V for low energy scan and ramped from 15 to 35 V for the high energy scan with a collision gas flow (Ar) of 1.5 mL/minute. Alternate 0.6 s scans at low and high energy were recorded for the range between 100 and 1990 m/z. A reference sprayer was operated at 300 nL/minute to produce a lockmass spectrum with Glu-1-Fibrinopeptide B (m/z 785.8426) every 30 s. 3051

dx.doi.org/10.1021/pr200030v |J. Proteome Res. 2011, 10, 3050–3059

Journal of Proteome Research

ARTICLE

Data Analysis

Spectra were analyzed with ProteinLynx Global Server (Vers. 2.4, RC7) (Waters Corp.) and searched against a database of human protein sequences (reviewed canonical sequences with isoforms) from UniProt release 15.5 (July 7, 2009). The database also contained sequences for yeast alcohol dehydrogenase, porcine trypsin, bovine serum proteins (BSA, serotransferrin, fibrinogen alpha chain, fibrinogen beta chain and fibrinogen gamma-B). This database was comprised of 34 602 protein sequences (20,005,831 amino acid residues). Data mining and statistical analysis was performed with the Elucidator Protein Expression Data Analysis System, Version 3.3 (3.3.0.1.SP3_CRE52.21) (Rosetta Biosoftware). Proteins were identified with a PeptideTeller predicted error (false positive rate) of 1% and a calculated decoy error rate of 0.2% (see Keller et al.25). Ratio P-values for differential expression were calculated by the Elucidator program using the xdev parameter.26,27 The P-values calculated are not from an analysis of variance, but instead derive from an application of an error model developed for large scale microarray data as adapted for proteomics within the Elucidator program. Analysis of results was also enhanced by use of the DAVID28 and KEGG29 resources for pathway analysis. The data associated with this manuscript may be downloaded from ProteomeCommons.org Tranche using the following hashes. (Users of this string are urged to verify the accuracy of copy paste operations, especially when copying string containing the (þ) symbol.) A spreadsheet with details on detected peptides can be accessed with this hash: keRkXr0LAKZduC5IMYhwlEywWoY grkctqxInmcbWUDnop6ZJEtkzJU5n2Fo6Oo/UK70kdtvzaGge V8N0t2mW4Bþ1qV4AAAAAAAACUw== All 18 raw data files may be accessed with this hash: zJqh LDUH1FRNpLaHO6WqoþE34i9xmuR7K1q8RBKaþzcaQF Zv3L2PfHygfiWY2fRjCo4I7o0k3uhz43xRtq2fknKwLOsAAAA AAACRZQ== These hashes may be used to prove exactly what files were published as part of this manuscript’s data set, and the hash may also be used to check that the data has not changed since publication.

’ RESULTS MSE technology identified and quantified proteins in whole cell lysates of six separate cultures (biological replicates of three control and three experimental treatments). Principal component analysis revealed good reproducibility for the technical replicates for each biological replicate (Figure 1). It also revealed clear separation in the protein abundance patterns between control and experimental treatments. Hierarchical clustering revealed that treatment of cells with media of higher osmolarity resulted in increased abundance of some proteins and decreased abundance of others (Figure 2). This figure represents intensity data transformed via Z-score, which may magnify small differences in protein abundance. No statistical or protein identification filters were applied to improve reliability at this level of the analysis or in the expanded graphic covering the entire data set (Supporting Information Figure 1). Thus, important overall trends in the data are apparent, but individual proteins are evaluated based on other factors (see below). Figures 1 and 2 show overall good reproducibility of replicate LCMS runs. Figure 2 also shows less variability in the mass spectrometric measurement than in the biological variability, thus suggesting that the effects of treatment were an important factor in this experiment. A total of 294 proteins were detected with a peptide error (false positive rate) of 1% and a calculated decoy error rate of 0.2%

Figure 1. Principal component analysis of Z-score transformed intensity data processed by the Elucidator program for all 18 LCMS Chromatograms in this experiment. Each data point represents a chromatogram. Data from control cells grown in 300 mOsM media are all found on the low end of the first principal component axis. Data from the 400 mOsM grown cells (treated) are grouped on the higher end of that axis, suggesting an overall global effect of the osmolarity treatment. Replicate LCMS runs (of like color) are relatively close together in most cases indicating excellent reproducibility of chromatography and mass spectrometry.

Figure 2. Agglomerative hierarchical cluster of Z-score transformed intensity data processed by the Elucidator program for all 18 LCMS chromatograms in this experiment. Proteins identified in each chromatogram are labeled as control (300 mOsM) or treated (400 mOsM). Groups of three identically colored boxes above the results of each LCMS run indicate the results for three replicate LCMS runs for each biological replicate (three biological replicates of control (300 mOsM) or treated (400 mOsM) cells). Z-score coloration indicates protein abundance in the sample (red indicates higher abundance, green indicates lower abundance and black equal abundance). The cluster suggests many proteins were downregulated at 400 mOsM (upper part of cluster) and many proteins upregulated at 400 mOsM (lower part of cluster). It should be emphasized that there was no filtering applied for statistical significance, fold change or the relative reliability of the protein identification. Z-score transformation as presented emphasizes small differences.

(see list in Supporting Information Table 1). Of those proteins, 185 proteins were identified based on two or more peptides. There were 113 595 peptide hits (matched spectra) to the forward database 3052

dx.doi.org/10.1021/pr200030v |J. Proteome Res. 2011, 10, 3050–3059

Journal of Proteome Research

ARTICLE

Figure 3. Example relative quantification data and supporting identification data for a protein of equal unchanged abundance (upper) panel, a protein with increased abundance (middle panel) and one with decreased abundance (lower panel) as a result of treatment (400 mOsM). Relative abundance (%) plots show the results for three replicate LCMS runs for each biological replicate (three biological replicates of control (300 mOsM) or treated (400 mOsM). For each panel, example graphics for two peptides are shown, including a single component of an isotopic cluster (feature plotted at relative % abundance) generated by the Elucidator program (charge state indicated), and a derived ms/ms spectrum generated by the PLGS program from high collision energy scan of MSE data. Red peaks are y-ions, blue peaks are b-ions, and green are modified ions (e.g., loss of NH3) in the MSE scans. The feature peak in each case is an overlay of aligned MS spectra (the most intense from an isotopic cluster representing a single charge state) of matching m/z and retention time for the 18 LCMS runs in the experiment. Below the graphics is a table representing relative expression of these proteins in (millions of) relative units. Nine replicates for control and treated are shown. They are ordered sequentially in groups of three representing replicate LCMS runs of the same biological replicate sample. Glyceraldehyde-3-phosphate dehydrogenase had a ratio of treated/control 1.0 (P = n.s.), ALDR had an abundance ratio of 3.5 (P < 1045) and 5NTD had an abundance ratio of 0.62 (P < 1041). Abundance ratio P-values were calculated within Elucidator as described previously.26.

including multiple detections of the same peptide at the same or different charge states. A total of 2251 unique peptide sequences were identified. Overall, the average mass error for these peptides was 7.0 ppm. Average protein sequence coverage was 19.9%. Three examples of protein expression patterns obtained for the two treatment conditions are shown in Figure 3. Glyceraldehyde3-phosphate dehydrogenase exhibited largely equal expression between control and treatment samples. This was seen in the MS signal for a single feature from an isotopic cluster for each of two peptides (LISWYDNEFGYSNR and GALQNIIPASTGAAK), illustrating the lack of demonstrable difference in relative abundance between samples from control and treated cultures. Also,

an MS/MS spectrum derived from each of these peptides extracted from the high collision energy scan is shown illustrating y- and b-series of fragment ions. In contrast, aldose reductase exhibited a significant increase (3.5-fold) in its abundance in the treatment group, as reflected in MS features from two peptides and in increased relative expression values calculated for the individual LCMS chromatograms. Corresponding prominent y-ion series in MSE-derived spectra for two peptides support the identification of this protein. Finally, an example of a protein, 50 -nucleotidase (5NTD), with decreased abundance (0.6-fold) in medium of higher osmolarity is also illustrated in Figure 3. As for two other proteins, example MS features from two peptides, and 3053

dx.doi.org/10.1021/pr200030v |J. Proteome Res. 2011, 10, 3050–3059

Journal of Proteome Research

ARTICLE

Figure 4. Expression data for selected proteins in agglomerative hierarchical cluster of Z-score transformed intensity data processed by the Elucidator program for all 18 LCMS chromatograms in this experiment. These proteins have at least 1.5-fold response to the treatment, and are represented by at least two peptides for both identification and quantification at ratio P-values as calculated by Elucidator.26 See text for other acceptance criteria. Mean relative abundance of each protein in control and treated as well as abundance ratio (as calculated by Elucidator) are listed (see Supporting Information Table 1 for individual abundance intensities for each replicate and each protein). Cluster coloration indicates protein abundance in the sample (red indicates higher abundance, green indicates lower abundance and black, unchanged abundance). Some proteins were upregulated at 400 mOsM (upper part of cluster) and some were downregulated at 400 mOsM (lower part of cluster). See Supporting Information Figure 1 for graphic that includes all proteins in an agglomerative hierarchical cluster of Z-score transformed intensity data processed by the Elucidator program. Elucidator identified fibronectin as the (shorter) isoform 3, based on the peptides actually detected (the algorithm seeks to provide the simplest interpretation of the data). However, no spectra were annotated to any unique peptides ascribed to isoform 3, thus the protein is listed here as the canonical isoform (P02751).

decreased relative expression values and y- and b-ion series in MSE-derived spectra, are illustrated. Analogous differences were recorded for additional proteins from this data set that were selected based on the magnitude of their fold-change in abundance ((1.5-fold ratio), significant P-value (