Gene profiling of scleroderma skin reveals robust signatures of disease that are imperfectly reflected in the transcript profiles of explanted fibroblasts.код для вставкиСкачать
ARTHRITIS & RHEUMATISM Vol. 54, No. 6, June 2006, pp 1961–1973 DOI 10.1002/art.21894 © 2006, American College of Rheumatology Gene Profiling of Scleroderma Skin Reveals Robust Signatures of Disease That Are Imperfectly Reflected in the Transcript Profiles of Explanted Fibroblasts Humphrey Gardner,1 Jeffrey R. Shearstone,1 Raj Bandaru,1 Tom Crowell,1 Matthew Lynes,1 Maria Trojanowska,2 Jaspreet Pannu,2 Edwin Smith,2 Stefania Jablonska,3 Maria Blaszczyk,3 Filemon K. Tan,4 and Maureen D. Mayes4 Objective. To determine whether biopsy specimens obtained from systemic sclerosis (SSc) lesions show a distinctive gene profile, whether that gene profile is maintained in fibroblasts cultured from SSc skin biopsy specimens, and whether results from tissue obtained from multiple clinical centers can be combined to yield useful observations in this rare disease. Methods. Biopsy samples and passaged fibroblasts were stored in RNAlater solution prior to processing for RNA. RNA from SSc and control skin biopsy specimens, as well as SSc and control explanted passage 4 fibroblasts, from 9 patients and 9 controls was hybridized to Affymetrix HG-U133A arrays. Data were analyzed using the BRB ArrayTools system. When appropriate, findings were followed up with immunohistochemical analysis or TaqMan studies. Results. Biopsy samples obtained from patients with SSc had a robust and distinctive gene profile, with ⬃1,800 qualifiers distinguishing normal skin from SSc skin at a significant level. The SSc phenotype was the major driver of sample clusters, independent of origin. Alterations in transforming growth factor ␤ and Wnt pathways, extracellular matrix proteins, and the CCN family were prominent. Explanted fibroblasts from SSc biopsy samples showed a far smaller subset of changes that were relatively variable between samples, suggesting that either nonfibroblast cell types or other aspects of the dermal milieu are required for full expression of the SSc phenotype. Conclusion. SSc has a distinct gene profile that is not confounded by geographic location, indicating that extended multicenter studies may be worthwhile to identify distinct subsets of disease by transcript profiling. Explanted SSc fibroblasts show an incomplete reflection of the SSc phenotype. Systemic sclerosis (SSc; scleroderma) is a devastating disease of connective tissue, the etiology and pathogenesis of which are poorly understood. Major pathogenetic components include endothelial injury and microvascular damage, excess production of extracellular matrix, and production of multiple autoantibodies, many of which are directed against nuclear and nucleolar components, and some of which are also associated with other connective tissue disorders such as rheumatoid arthritis and systemic lupus erythematosus. Disease outcome in SSc is not easily predicted, although general clinical patterns are discernable that, to an incomplete extent, overlap with autoantibody profiles (1,2). Most patients present with symptoms and signs of skin thickening and hardening in the distal extremities, which are frequently presaged (for months to years) by Raynaud’s phenomenon. Rapid progression of skin thickening is suggestive of a worse outcome, with a higher likelihood of internal organ involvement. However, different patients tend to have different patterns of end-organ damage in an unpredictable manner, and patients with limited disease have a distinctly increased 1 Humphrey Gardner, MD, Jeffrey R. Shearstone, MSc, Raj Bandaru, MSc, Tom Crowell, BSc, Matthew Lynes, BA: Biogen Idec, Cambridge, Massachusetts; 2Maria Trojanowska, PhD, Jaspreet Pannu, PhD, Edwin Smith, MD: Medical University of South Carolina, Charleston; 3Stefania Jablonska, MD, Maria Blaszczyk, MD: Warsaw Medical Academy, Warsaw, Poland; 4Filemon K. Tan, MD, PhD, Maureen D. Mayes, MD, MPH: University of Texas Medical School at Houston. Dr. Gardner has stock options in Novartis. Address correspondence and reprint requests to Humphrey Gardner, MD, Novartis Institutes for Biomedical Research, 250 Massachusetts Avenue, Cambridge, MA 02139. E-mail: Humphrey. email@example.com. Submitted for publication October 4, 2005; accepted in revised form February 21, 2006. 1961 1962 GARDNER ET AL risk of pulmonary hypertension. It is possible that the myriad outcomes in SSc represent different diseases that might ultimately be diagnosable by expression profiling. Fibroblast activation and the potential involvement of the transforming growth factor ␤ (TGF␤) axis have long been a focus of study in SSc, not least because explanted dermal fibroblasts are a highly accessible source of tissue, and cells from the same individual can be studied by multiple investigators. At the same time, there is concern that examination of SSc fibroblasts results in a focus on one potential aspect of the pathogenesis of the disease while missing the contribution of other tissues, e.g., endothelial cells and cells of the immune system. Furthermore, many studies and anecdotal reports have suggested that the phenotypic peculiarities of SSc fibroblasts are not stable and wear off with successive passages (3). Studies of fibroblasts in other contexts have suggested that fibroblasts retain some characteristics, such as transcript patterns enabling them to be clustered by site of origin (4), while losing others, such as surface integrin profiles (5). For these reasons, establishing whether the transcript profile of explanted SSc fibroblasts reflects the gene profiles associated with SSc skin biopsy specimens is of some importance. It is also advantageous to analyze in parallel both biopsy specimens and fibroblasts derived from them. In this way, at least to a first degree of approximation, fibroblast and nonfibroblast components of the SSc gene profile can be separated from one another. A current challenge with transcript profile studies of human tissue is variation in collection methodology or RNA preservation, leading to evanescent signatures that are more associated with artifacts of collection than with the disease of interest. In common diseases, rigorous collection protocols and studies limited to single sites have enabled identification of important prognostic signatures, using unsupervised clustering (6). However, to use similar techniques in rare diseases such as SSc, in which samples necessarily are collected from multiple sites, we need to understand and control collection artifacts. We attempted to control these sources of variation with our methodology and explicitly tested for their presence during our analysis. PATIENTS AND METHODS Patients and controls. All patients fulfilled the American College of Rheumatology (formerly, the American Rheumatism Association) preliminary classification criteria for scleroderma (7), and all but 1 had diffuse cutaneous disease according to the classification criteria described by LeRoy et al (8). Consecutive outpatients with early disease (⬍3 years from the onset of the first non–Raynaud’s phenomenon scleroderma disease manifestation) and either diffuse disease or skin thickening involving the forearms were chosen, in an attempt to diminish heterogeneity of disease expression. To be eligible for the study, patients could not have been receiving doses of prednisone ⱖ20 mg during the 4 weeks prior to the biopsy, nor could they have been receiving other immunosuppressive treatment within this time period. Healthy controls were chosen on the basis of a lack of Raynaud’s phenomenon by history, lack of a diagnosis of a systemic autoimmune disease, and willingness to participate. Control subjects listed their medications, and none were receiving glucocorticoids or immunosuppressive agents. This study was approved by the institutional review boards of the University of Texas Health Science Center at Houston and the Medical University of South Carolina. Written informed consent was obtained from all participants. Biopsy method. Two 3-mm punch biopsy specimens were obtained from the same distal forearm (forearm with involved skin in patients with scleroderma) of each subject, in a side-by-side manner. One biopsy specimen was immediately placed in RNAlater (Ambion, Austin, TX), shipped on wet ice to Affymetrix (Santa Clara, CA), and stored at 4°C prior to RNA preparation. RNA was prepared within 15 days of tissue harvest. The other biopsy piece was placed in culture. Fibroblast culture conditions. Biopsy tissue was rinsed several times with antibiotic–antimycotic solution (catalog no. 15240-062; Life Technologies, Gaithersburg, MD). Tissue was then placed in 1 ml of collagenase solution and incubated for 24 hours at 37oC. Collagenase solution was prepared at the Medical University of South Carolina and was distributed to all investigators. The collagenase solution contained 0.25% collagenase 1 (Sigma, St. Louis, MO) and 0.05% DNase I (Sigma) in Dulbecco’s modified Eagle’s medium (DMEM) with 20% fetal bovine serum (HyClone, Logan, UT). The entire 1-ml solution was mixed together with 5 ml of media (DMEM plus 20% fetal calf serum), plated into a 25-cm2 flask, and left undisturbed for 48 hours at 37°C in a 5% CO2 atmosphere. The resulting confluent culture was then designated passage 1 (P1). Cells were then split (1:4 ratio) to generate 4 ⫻ 25–cm2 flasks (P2). Two flasks at passage 4 were trypsinized, the trypsin was then neutralized using soybean trypsin inhibitor, and cell pellets were washed twice with phosphate buffered saline prior to the addition of RNAlater (Ambion). Cell pellets were shipped frozen on dry ice by overnight delivery and stored frozen at ⫺80°C at the Affymetrix facility prior to RNA preparation. Total RNA purification. Skin biopsy specimens were removed from the RNAlater solution (Ambion), placed in a weighing dish containing 1 ml of TRIzol reagent (Invitrogen, Carlsbad, CA), minced using a razor blade, and poured into a 2-ml tube. RNAlater was removed from the fibroblast pellets, which were then resuspended in 1 ml of TRIzol reagent. Fibroblasts and biopsy specimens were homogenized using a PowerGen 125 homogenizer (Fisher Scientific, Hampton, NH) for 2–5 minutes at top speed. Total RNA was extracted from TRIzol according to the manufacturer’s protocol. The extraction process included an optional centrifugation step that was suggested for biopsy samples with a high content of fat and DISTINCT GENE PROFILE OF SCLERODERMA extracellular material. Total RNA was resuspended in 100 l of water and further purified using an RNeasy Mini column (Qiagen, Valencia, CA), according to the manufacturer’s protocol. Probe labeling, hybridization, and scanning. Sample labeling, hybridization, and staining were carried out according to the Eukaryotic Target Preparation protocol in the Affymetrix Technical Manual (701021 rev. 4) for Genechip Expression Analysis (Affymetrix). According to this protocol, 1–5 g of purified total RNA was used in a 20-l first-strand reaction with 200 units of SuperScript II (Invitrogen) and 0.5 g of dT-T7 primer in 1⫻ first-strand buffer (Invitrogen), with incubation at 42°C for 1 hour. Second-strand synthesis was carried out by adding 40 units of Escherichia coli DNA polymerase, 2 units of E coli RNase H, and 10 units of E coli DNA ligase in 1⫻ secondstrand buffer (Invitrogen), followed by incubation at 16°C for 2 hours. The second-strand synthesis reaction was purified using the Genechip Sample Cleanup Module according to the manufacturer’s protocol (Affymetrix). Purified complementary DNA (cDNA) was amplified and biotinylated using the BioArray HighYield RNA Transcript Labeling Kit (Enzo Life Sciences, Farmingdale, NY) according to the manufacturer’s protocol. Fifteen micrograms of labeled complementary RNA was fragmented and resuspended in 300 l of 1⫻ hybridization buffer containing 100 mM MES, 1M Na⫹, 20 mM EDTA, 0.01% Tween 20, 0.5 mg/ml aceylated bovine serum albumin, 0.1 mg/ml herring sperm DNA, Control Oligo B2 (Affymetrix), and eukaryotic control transcripts. Sample was applied to a Human Genome 133A GeneChip (Affymetrix). Hybridized arrays were washed and stained on a GeneChip Fluidics Station 450 and visualized using a GeneChip Scanner 3000. Quantitative reverse transcription–polymerase chain reaction (RT-PCR). First-strand cDNA was synthesized from RNA using the SuperScript III Platinum Two-Step qRT-PCR Kit (Invitrogen). For each reaction, 1 l of RNA containing 100 ng–1 g was used. RT-PCRs were set up in Optical 96-well reaction plates (Applied Biosystems, Foster City, CA) using TaqMan Gene Expression assays protocol for 50-l reactions (Applied Biosystems). The concentration of cDNA was determined by spectroscopy, using a BioPhotometer (Eppendorf, Madison, WI). For all reactions, 40 ng of cDNA was used. The RT-PCR thermal profile was as follows: 3 minutes at 95°, then 55 cycles of 15 seconds at 95°, and 1 minute at 56°. RT-PCR data were collected using an Mx3000P PCR system (Stratagene, La Jolla, CA) and analyzed using Mx3000P software (Stratagene). Standard TaqMan probes were used, as follows: for COL11A1, Hs00266273_m; for COL1A2, Hs00164099_m1; for COL1A1, Hs00164004_m1; for GAPDH, Hs99999905_m1. Array analysis. Array data in the form of CEL files were imported into BRB ArrayTools, developed by Dr. Richard Simon and Amy Peng Lam (http://linus.nci.nih.gov/BRBArrayTools.html). Biopsy data and fibroblast data were imported and normalized independently, using the robust multichip average (RMA) algorithm. Data sets normalized using the Affymetrix MicroArray Suite 5.0 probe reduction algorithm were used when comparisons between fibroblasts and biopsy specimens were necessary. Class comparisons and class predictions were carried out using the BRB software package. Specific methods for given analyses are described in the figure legends. 1963 Immunohistochemical analysis. Specimens from the individuals involved in the Affymetrix analysis were not available for immunohistochemical analysis. Recuts were obtained from paraffin-embedded biopsy specimens (5 SSc and 2 control) from subjects whose age distribution and length of history of SSc were similar to those of the Affymetrix group. Pretreatment of tissue sections for antigen retrieval was required and was performed as follows: for type I collagen, pretreatment with pepsin (P7000; Sigma) at 1 mg/ml in 0.5N acetic acid for 2 hours at 37°C; for CD3, pretreatment with 1 mM EDTA (pH 8.2) in a pressure cooker (Biocare Medical, Walnut Creek, CA). Endogenous peroxidase was quenched using 0.3% hydrogen peroxide in methanol for 30 minutes. Tissue sections were rinsed in several changes of 50 mM Tris HCl, 150 mM NaCl, 0.05% Tween 20 (pH 7.6) prior to immunostaining. Unconjugated primary antibodies included monoclonal mouse antihuman CD20, clone L26 (Dako, Carpinteria CA); monoclonal mouse anti-human fibrillin 1, clone 12A5.18 (LabVision, Fremont, CA); monoclonal rat anti-human CD3, clone CD3-12 (Serotec, Raleigh, NC); and polyclonal rabbit anti-human type I collagen (Biogenesis, Poole, UK). Binding of primary antibodies to tissue sections was detected using biotinylated secondary antibodies (Vector, Burlingame, CA) followed by an avidin–biotin–horseradish peroxidase reagent (Elite ABC; Vector), using diaminobenzidine as the substrate. RESULTS Biopsy and cell pellet RNA recovery and hybridization. We noted in pilot studies that yields from fresh-frozen human skin samples were ⬃2-fold higher than those from skin samples stored in RNAlater (data not shown). However, the attraction of the RNAlater method in terms of ease of use was thought to outweigh the reduced yield. In general, a 3-mm biopsy sample stored in RNAlater according to the manufacturer’s instructions provided ample material for a labeling and hybridization reaction without additional amplification. Twenty-eight biopsy specimens were processed. Of the specimens submitted for processing, 16 of 28 gave yields in excess of 2 g RNA and were hybridized. One group of 10 biopsy samples processed as a single batch during the study yielded ⬍0.1 g RNA, but there were no clear parameters distinguishing these samples from those preceding or following them, and these failures were assumed to be attributable to an undiagnosed error in processing. Of the fibroblast culture pellets, all yielded at least 5 g of RNA, but 3 of 26 failed hybridization. As an indicator of RNA quality, the total mean ⫾ SD percentage of present calls was noted to be 54.9 ⫾ 4.6% (range 61.3–46.4%) for the biopsy samples and 50.7 ⫾ 2.9 (range 57.0–45.8%) for the fibroblast samples. Overall, samples from 9 patients and 9 controls provided data. Twelve subjects (5 patients and 7 con- 1964 Figure 1. Dendrogram showing results of unsupervised clustering of the biopsy specimens used in this study, with relevant clinical information. Intersection of the lists created from the 50th intensity percentile for each sample resulted in 8,736 probe set qualifiers with a consistently strong signal in all samples. These probe sets were used as the basis for hierarchical clustering using the BRB ArrayTools system with a centered correlation metric and average linkage. Note that disease status was more important than any other parameter in classifying the biopsy specimens. SSc ⫽ systemic sclerosis; NL ⫽ normal; SC ⫽ site 1; TX ⫽ site 2; C ⫽ caucasian; AA ⫽ African American. trols) had both biopsy samples and fibroblast cultures analyzed. The mean ⫾ SD age of patients was 45.7 ⫾ 8.5 years (range 29–56 years), and that of controls was 46.9 ⫾ 8.8 years (range 34–56 years). The mean ⫾ SD disease duration among patients was 10.6 ⫾ 6.1 months (range 3–24 months). Comparison of normal and SSc biopsy specimens. Unsupervised (agglomerative) clustering of all samples (Figure 1) demonstrated that the scleroderma phenotype was the dominant influence on the expression profile and was not confounded by patient sex, age, or race, or by the origin of the biopsy sample. Class comparisons on all 22,283 qualifiers between normal and SSc biopsy specimens showed 1,839 qualifiers distinguishing normal skin from SSc skin at a univariate P value of ⬍0.01, and 506 qualifiers distinguishing normal skin from SSc skin at a univariate P value of ⬍0.001 (by unpaired t-test, with random variance model). Of these, the 50 most significant (according to P value) are shown in Figure 2. GARDNER ET AL Comparison of normal and SSc fibroblasts and relationship to the biopsy specimens. Class comparisons between normal and SSc fibroblasts showed 223 qualifiers distinguishing normal from SSc at a P value of ⬍0.01, of which 21 were significant at a P value of ⬍0.001. These results reflect the far higher variance seen in the fibroblast culture profiles than in the biopsy specimens. Of the 223 qualifiers, 105 were up-regulated in SSc samples, and 118 were down-regulated. The intersection between qualifiers dysregulated in biopsy specimens and those dysregulated in fibroblasts (n ⫽ 26 [9 discordant, 17 concordant]) (Figure 3A) was far greater than that which would be seen by chance (P ⱕ 0.02 by Fisher’s exact test of observing an intersection of that many or more by random sampling of gene lists from a pool of 23,000 qualifiers), and included a large proportion of extracellular matrix genes, including type VIII collagen, fibulin 1, fibrillin 2, and decorin. Interestingly, a subset of these showed inverse regulation in the context of the biopsy samples and the fibroblasts, notably ephrin B2. Determining fibroblast and nonfibroblast components of the transcript profile of SSc and normal biopsy specimens. To make an approximation of the cellular source of differentially expressed genes in SSc versus normal biopsy specimens, we reasoned that genes 5-fold up-regulated in cultured fibroblasts compared with biopsy samples might be considered as likely to be of fibroblast origin in the biopsy specimen, and, similarly, that genes down-regulated 5-fold in fibroblasts relative to the biopsy sample were unlikely to be of fibroblast origin in the biopsy specimen, with genes between these 2 extremes of differential expression being considered indeterminate in source. To ensure that genes influenced by the disease process would not be averaged out, class comparisons of SSc biopsy specimens versus SSc fibroblasts were made independently of control biopsy specimens versus control fibroblasts, and a union was made of the 5-fold up-regulated or downregulated qualifiers. These lists were then intersected with the class comparison of SSc versus control biopsy samples. Thus, for example, qualifiers q dysregulated in the fibroblast component were found, such that ([(qfibro(ssc) ⬎ 5 ⫻ qbiopsy(ssc); P ⬍ 0.01) or (qfibro(control) ⬎ 5 ⫻ qbiopsy(control); P ⬍ 0.01)] and [qbiopsy(control) ⫽ qbiopsy(ssc); P ⬍ 0.01]). Using these criteria, 121 probable fibroblast and 358 probable nonfibroblast genes were identified; expression of the 30 most significant of each is shown in Figures 4A and B, respectively. In order to provide an independent form of confirmation of the probable cellular source of transcripts, we created an independent RMA data set com- DISTINCT GENE PROFILE OF SCLERODERMA Figure 2. Top 50 (by P value) class-distinguishing genes for systemic sclerosis (SSc) versus normal biopsy specimens. Class comparison between SSc and control biopsy specimens was performed on the robust multichip average– normalized data set in BRB ArrayTools using a nominal significance level of 0.01 for each univariate test. Log2 intensity values for each biopsy sample are presented as a color map for each qualifier (left group of columns of color maps). Nine colors, from bright red to bright green, represent equal-size bins, from the highest to the lowest values in the row. In order to give a rough guide as to the likely source of the transcript, the pair of columns in the middle of the figure represent the ratio of the mean expression in biopsy specimens versus that in fibroblasts, as determined in a MicroArray Suite–normalized data set containing all fibroblast and biopsy samples, with bright red indicating ⬎16-fold higher expression in biopsy specimens, and bright green representing ⬎16-fold lower expression. The left column of the pair represents the comparison of control biopsy specimens versus control fibroblasts, and the right column represents scleroderma biopsy specimens versus scleroderma fibroblasts. The group of columns on the right represents data for SSc and control fibroblasts for the same qualifiers, in the same format and representation as for the biopsy specimens. These are heat-mapped relative to one another, independent of the biopsy specimens, and illustrate that very few of the genes that distinguish SSc from control biopsy specimens also distinguish SSc from control fibroblasts. 1965 1966 GARDNER ET AL Figure 3. A, Intersection of qualifiers in class comparisons of biopsy specimens and fibroblasts. Class comparisons identified 1,839 qualifiers distinguishing normal skin from systemic sclerosis (SSc) skin and 223 qualifiers distinguishing normal fibroblasts from SSc-derived fibroblasts (P ⱕ 0.01). Intersection of these lists yielded 26 qualifiers, 17 of which demonstrated concordant disease state expression between fibroblasts and biopsy specimens (probe sets highlighted in green), while 9 showed discordant values (probe sets highlighted in red). The heat map of the expression intensity data is false colored, as described in Figure 2. B, A subset of fibroblast classifier genes was successful in predicting the disease class of the biopsy specimens (see Patients and Methods). The upper part of the figure shows heat maps of expression of these genes in SSc and normal biopsy samples, and the group of columns on the right shows expression of the cognate HU95A qualifiers in SSc and normal biopsy specimens based on the data set described by Whitfield et al (9). The lower part of the figure illustrates whether each biopsy specimen was correctly classified as SSc or normal, using this 26-gene set or the subset of cognate qualifiers present in the data set of Whitfield et al. Note that 1 of the SSc biopsy specimens from this study was misclassified by all but 1 algorithm, and 1 of the normal biopsy specimens in the study by Whitfield et al was misclassified by 1 algorithm. All others were correctly classified by all algorithms. Ctl ⫽ control. DISTINCT GENE PROFILE OF SCLERODERMA Figure 4. A and B, Top 30 genes (by P value) derived from the class comparison shown in Figure 2, separated into those showing 5-fold increased expression in fibroblasts relative to biopsy specimens (A) and those showing 5-fold increased expression in biopsy specimens relative to fibroblasts (B). The 3 columns on the right (F ⫽ fibroblast, K ⫽ keratinocyte, E ⫽ endothelial) represent the mean of 3 samples each of cultured fibroblasts, basal keratinocytes, and endothelial cells for the same qualifiers in an independent robust multichip average–normalized data set. These illustrate that, on the whole, the distinction between fibroblast and nonfibroblast attained by comparing fibroblasts with biopsy specimens is consistent with an estimate of the cellular source by comparing cultured cell lines. Notable exceptions are IGF1, COMP, and SFRP4, which are likely to originate from fibroblasts but appear to be dramatically down-regulated in culture. 1967 1968 GARDNER ET AL Figure 5. A, Summary of dysregulation of extracellular matrix genes in systemic sclerosis (SSc). B, Relative expression of qualifiers for different collagens in biopsy specimens and fibroblasts. Data are expressed as the ratio of the geometric means of SSc versus normal fluorescence intensity. All significantly dysregulated collagens are ⬃2-fold up-regulated in SSc biopsy specimens relative to control biopsy specimens, with the notable exception of type XI collagen, which is increased to at least 4-fold. Explanted SSc fibroblasts show relatively minor and attenuated differences in collagen expression relative to control. C, Quantitative reverse transcription–polymerase chain reaction for COL1A1, COL1A2, and COL11A1 in 4 SSc biopsy specimens and 7 normal skin samples, normalized to GAPDH expression. Up-regulation of type XI collagen expression is far more dramatic than that of type I collagen expression in the SSc samples. Values are the mean and SD. bining published expression profiles on HU133A chips for cultured basal keratinocytes (3 lines), endothelial cells (3 lines), and fibroblasts (3 lines). The mean values for the same set of qualifiers for each of these cell types were heat-mapped relative to one another, independently of the biopsy specimens, and placed to the right of the heat-mapped data for the biopsy specimens. This comparison showed that although, in general, tissue assignments based on the fibroblast-to-biopsy comparison were consistent with evidence from the cell lines, there were exceptions. Examples include secreted Frizzled-related protein 4 (SFRP4), insulin-like growth factor 1 (IGF-1), and cartilage oligomeric matrix protein (COMP), which may well be derived predominantly from fibroblasts but which are so down-regulated in tissue culture relative to the biopsy samples that they are interpreted, using this approach, as being of nonfibroblast origin. Comparison of our data set with previously published data. The generous provision of Affymetrix data by Whitfield et al (9) enabled us to compare the genes that in their study were determined to be significantly modulated in SSc with the genes in our study. We generated a list of qualifiers in common between HU95A and HU133A chips and generated classifier lists at a significance level of P ⬍ 0.05 for our data set and that of Whitfield et al. Overall, 10,508 qualifiers from the HU95A chip were matched to 9,530 unique qualifiers on HU133A chips. Among these 9,530 qualifiers, 2,250 were significantly different between control and SSc biopsy specimens in our data set. Of those, 572 were also present in the data set described by Whitfield and DISTINCT GENE PROFILE OF SCLERODERMA colleagues (P ⬍ 10⫺44). To give a better idea of concordance of genes of high significance in each data set, 52 of the 100 most significant qualifiers in the data set described by Whitfield et al were also present in our data set, at a significance level of P ⬍ 0.05, and 60 of the most significant qualifiers in our data set were present at the same level of significance in the Whitfield data set. In our experience, these levels of concordance compare favorably with those observed in other repeated studies using different platforms. Generation of a predictive qualifier set from fibroblasts. We reasoned that, should there be an expression profile predictive of SSc in fibroblasts, we should be able to identify a group of qualifiers that would distinguish normal from SSc fibroblasts and also effectively distinguish SSc biopsy specimens from controls. To this end, we performed a class comparison between SSc and normal fibroblasts, at a P value of ⬍0.01. The 223-member qualifier list that resulted from this comparison (described above) was used to generate a classifier for the biopsy specimens. Leave-one-out cross-validation analysis selected a subset of 26 genes that were successful, by multiple models, in distinguishing SSc from normal biopsy specimens. Fifteen of these could be matched to qualifiers in the HU95A chip. Those 15 were successful at distinguishing SSc from normal biopsy specimens in the data set described by Whitfield et al (9), although close examination of the expression levels suggests that some qualifiers, especially those down-regulated in SSc in our data set, were not similarly dysregulated in the data presented by Whitfield and colleagues (Figure 3B). Among those found in common were 3 interferon targets: OAS1, MX2, and IFI16. To reach a more general understanding of the relationship between cultured fibroblasts and disease, we examined clustering of normal and SSc fibroblast and biopsy gene sets by principal component analysis, using various qualifier lists (data not shown). The list of 223 genes from the class comparison of fibroblasts enabled distinction of SSc and control fibroblasts as loose but resolvable clusters, and could also divide SSc and control biopsy specimens into tighter independent clusters. However, no qualifier list derived from a class comparison of SSc versus control biopsy samples, even the probable fibroblast subset, could resolve the fibroblast cultures into SSc and control groups, whereas with each gene list used the biopsy samples were resolved into 2 tight clusters. As well as suggesting that nonfibroblast drivers of disease may be important in the biopsy specimens, this finding is consistent with the possibility 1969 that cultured fibroblasts may derive from quite small subpopulations within the biopsy samples. From biopsy specimen to biopsy specimen, these may be more or less representative of the scleroderma phenotype. Collagens in SSc. Collagens are well known to be up-regulated in SSc. We therefore surveyed qualifiers for collagens. Qualifiers showing significant upregulation in SSc are shown in Figure 5B. Two qualifiers for type XI collagen showed a remarkably high level of up-regulation in SSc compared with other qualifiers. We therefore reconfirmed this observation using quantitative RT-PCR for 3 collagens: COL1A1, COL1A2, and COL11A1 (Figure 5C). Immunohistochemical analysis of SSc biopsy specimens. Five SSc biopsy specimens and 3 control specimens were examined by immunohistochemistry for fibrillin 1, type I collagen, CD3, and CD20, of which a representative subset is shown in Figure 6. Infiltration of T cells and B cells in SSc biopsy specimens was sparse and not significantly different from that in controls. Type I collagen and fibrillin 1 showed significantly increased staining in SSc specimens. Immmunostaining for smooth-muscle actin showed staining in dermal adnexae and arterioles in control and SSc samples, but no dermal fibroblasts were positive for ␣-smooth muscle actin in any sample (data not shown). DISCUSSION The abundance of autoantibodies in SSc indicates involvement of the immune system, but the precise role of the immune system in SSc remains unclear. Immune cell signals in the biopsy specimens include upregulation of the T cell markers CD4 and CD28, the monocyte markers CD14, CD163, and CD11b, the antigen-presenting cell costimulatory B7-2 protein (CD86), and B cell antigens CD83 and CD84 as well as Ig. There were no obvious differences in CD3 (T cell) counts between control and SSc samples, and staining of CD20 cells (B cells) showed very few cells in all specimens (Figure 6). This suggests that very small numbers of lymphocytes can be detected by gene profiling. The extent of immune cell infiltration is known to decline with time from the onset of disease, which was 3 months to 2 years for both the biopsy specimens submitted for RNA analysis and those used for immunohistochemical analysis. However, our histology findings are typical and are rather dissimilar from the findings described by Whitfield et al (9), in which some SSc biopsy specimens appeared to include secondary lymphoid tissue, interpreted as B cell aggregates. In summary, although there 1970 GARDNER ET AL Figure 6. CD20, CD3, type I collagen, and fibrillin immunostaining in biopsy specimens from 3 patients with systemic sclerosis (SSc) and 2 normal controls. A few CD20⫹ cells can be seen adjacent to an adnexal structure in 1 biopsy specimen (top left panel), but otherwise CD20⫹ cells were not seen. CD3⫹ cells were present in a predominantly perivascular distribution in all biopsy specimens and were not increased in SSc. Both type I collagen and fibrillin 1 show increased staining in the SSc biopsy specimens. is undeniably an immune cell signature from B cells, T cells, and macrophages, consistent with the abundant autoantibodies seen in the disease, these results do not clarify our understanding of the relationship between autoimmunity and disease pathogenesis. Changes in mRNA expression for extracellular matrix proteins found in SSc biopsy specimens are summarized in Figure 5A. It has long been known that collagens are up-regulated in SSc, but the major focus has been on the fibrillar collagens, types I, III, and V. However, chondrogenesis-associated collagens, such as type XI collagen and type X collagen, as well as the chondrogenesis-associated protein COMP, are also upregulated, as well as the basement membrane type IV collagen, the network-forming collagens types X and VIII, and the endothelial basement membrane type XV collagen, and the collagen- and biglycan-processing en- zyme bone morphogenetic protein 1 (10). Most of these up-regulations have also been seen in TGF␤-treated fibroblasts (11). Significantly, in this study, type XI collagen was up-regulated disproportionately relative to other collagens (Figures 5B and C) (⬃5-fold versus ⬃2-fold for most others). Type V collagen forms fibrils with type I collagen to control fibril diameter, and type XI collagen forms fibrils with type II collagen in an analogous manner. Settings in which type XI collagen is found in association with type I collagen in normal tissue include fibrocartilage of the intervertebral disc, as well as embryonic tendon. Thus, it is possible that the dermal fibrotic response is a specific developmental program directed toward fibrocartilage (12) rather than a nonspecific response. In contrast, in a study of lung fibroblast responses to TGF␤, type IV collagen was the only collagen that was significantly dysregulated (13). DISTINCT GENE PROFILE OF SCLERODERMA The small leucine-rich family of proteoglycans (SLRPs) also show some characteristic alterations in SSc. Decorin is well known to be down-regulated in SSc and is believed to have some TGF␤ antagonistic properties of its own. In contrast, biglycan, versican, and lumican are up-regulated, as is asporin, a relative of decorin. The potential regulatory roles of the SLRPs in matrix assembly are beginning to be scrutinized. Although all of them are associated with collagen fibrils, each may be synthesized by different subsets of cell types (14). All of these proteins seem to have generally inhibitory effects on collagen fibril diameter and on fibroblast proliferation; thus, the reciprocal downregulation of decorin and up-regulation of biglycan and lumican in SSc have a pathophysiologic significance that is not fully understood. However, these abnormalities in SLRPs may be connected to ultrastructural and biochemical changes in skin, namely increased collagen solubility (15) and fibril abnormalities, including both abnormally thin fibrils (16) and abnormally tightly packed and thick fibrils (17). The strength of activation of the TGF␤ pathway in cultured SSc fibroblasts has been discussed extensively (18). Our study and that by Whitfield et al (9) show extensive evidence of dysregulation of the TGF␤ and Wnt pathways in vivo. Many TGF␤ targets are up-regulated in SSc biopsy specimens and are extremely similar to those seen in normal fibroblasts treated with TGF␤ (11). Many of these targets are no longer differentially expressed in explanted fibroblasts, notably the bulk of the collagens, which suggests that the profibrotic drivers are from cell types that do not persist in fibroblast culture. Transcript for TGF␤ itself is not increased in SSc biopsy specimens, suggesting that increased TGF␤ signal is attributable to increased activation of latent TGF␤ protein. One of the best known activators of latent TGF␤ is thrombospondin 1, the transcript of which was reported to be significantly up-regulated in SSc biopsy specimens and in cultured SSc fibroblasts in some studies (19). Studies have suggested that the subset of pulmonary fibroblasts that are Thy-1 positive are in fact TGF␤ insensitive due to failure of TGF␤ activation (20). Thy-1, along with thrombospondin 1, is strongly up-regulated in SSc skin, suggesting that this relationship does not hold in dermal fibrosis. Changes in the Wnt pathway that are seen in SSc biopsy samples include down-regulation of Wnt inhibitory factor 1, Frizzled-related protein, and Frizzled homolog 7, as well as up-regulation of SFRP4. Taken together, these changes are consistent with a general 1971 up-regulation of Wnt signaling. It is notable that a Wnt regulatory system has been shown to be active in mesenchymal stem cells in culture: down-regulations in Dkk and reciprocal up-regulations in Wnt-5A cause a deceleration in cell growth (21). Synergies between the Wnt and TGF␤ pathways have been proposed, which may be mediated by interaction of Smad4 and lymphoid enhancer factor 1 (22). The best known member of the CCN family is connective tissue growth factor (CTGF; CCN2), a putative downstream target of TGF␤ and a profibrotic cytokine in its own right, which is well known to be up-regulated in SSc skin and SSc fibroblasts (23–25). CTGF requires insulin or IGF-1 (the level of IGF-1 is also elevated in SSc biopsy samples) for induction of collagen in SSc fibroblasts (26). Cyr61 (CCN1), which provides integrin-dependent promigratory stimuli to fibroblasts and vascular smooth muscle cells (27,28), is also up-regulated in SSc, while Wnt-1–inducible signaling pathway protein 2 (WISP2; CCN5) is downregulated. Loss of Cyr61 is associated with differentiation of mesenchymal stem cells into any daughter lineage (29), and thus its up-regulation may reflect expansion of an uncommitted fibroblast phenotype. Down-regulation of WISP2 and reciprocal up-regulation of CTGF have also been noted in the response of normal fibroblasts to TGF␤ (11), but the significance of this remains unclear. CCN3 (NOV) was not dysregulated at the transcript level in SSc. Using the observations described above, profiling the data from both biopsy samples and fibroblasts enabled us to make some informed guesses about the tissue-type contribution of different genes in SSc. For example, up-regulation of the monocyte/macrophage genes CD14, Toll-like receptor 1 (TLR-1) and TLR-2, and integrins ␤2, ␣X, and ␣M appears to be associated with up-regulation of interleukin-6 (IL-6), IL-16, and CXCL3, all of which might be synthesized by fibroblasts. Similarly, examination of IGFs, CCN family members, and proteases showed down-regulation of IGF binding protein 5 (IGFBP-5), IGFBP-6, and WISP2 and reciprocal up-regulation of IGFBP-3 and IGFBP-4 as well as CTGF, plasminogen activator inhibitor 1, and C1q (which was likely from a nonfibroblast source). It was also noteworthy that the vast majority of fibroblastassociated transcripts showing significant change in SSc biopsy specimens were up-regulated (105 of 121 [87%]), whereas approximately half of the fibroblast-associated transcripts in the nonfibroblast compartment (161 of 358 [45%]) were down-regulated. A theme may be emerging involving loss of nonfibroblast-derived inhibitory signals 1972 and concomitant up-regulation of fibroblast profibrotic behavior in SSc lesions. It is possible that some of the down-regulation of epithelium-derived genes is attributable to loss of epithelial adnexae in sclerodermatous skin. However, although there was a 2-fold downregulation of keratin 15, associated with early hair follicle “bulge” cells (30), no other markers of follicles such as keratin 19 and the hair keratins were significantly or consistently altered in the SSc biopsy specimens, nor were adnexae lost in the biopsy samples examined by immunohistochemistry. Furthermore, a systematic review of endothelial, keratinocyte, and fibroblast-specific transcripts of the highest abundance did not show a change in the relative size of these 3 compartments in the SSc biopsy specimens (data not shown). This study includes the first extensive comparison of fibroblast cultures and primary biopsy specimens in SSc and provides an opportunity to examine fibroblastdriven versus non–fibroblast-driven behavior. We reasoned that if fibroblasts truly reflect the disease state, then a subset of genes that distinguishes between SSc fibroblasts and normal fibroblasts should be able to distinguish SSc skin from normal skin. Leave-one-out cross-validation analysis selected a subset of genes that were successful, by multiple models, in distinguishing SSc specimens from normal biopsy specimens, in both our data set and that of Whitfield et al (Figure 3B). From this we conclude that there is a set of gene expression profiles that distinguishes SSc from normal fibroblasts; however, the number of genes that distinguish SSc fibroblasts from control fibroblasts is far reduced compared with the number of genes that distinguish SSc biopsy specimens from control specimens at the same level of significance (223 versus 1,839; P ⫽ 0.01). Indeed, Figure 2 demonstrates that the majority of the most significant differentially expressed genes in SSc biopsy specimens are not consistently differentially expressed in explanted fibroblasts. Differences between the differential gene profiles of fibroblasts and biopsy specimens suggest further conclusions. For example, CTGF, which is well known as a fibrosis driver downstream of TGF␤, as well as Thy-1 are up-regulated in scleroderma biopsy specimens relative to normal controls, but expression in cultured fibroblasts does not clearly correlate with disease. Aspects of the SSc phenotype that are retained in fibroblasts include primarily a small residual subset of changes in matrix gene expression. One interpretation of this is that there is a key disease-driver population of a GARDNER ET AL different cell type in the dermis, such as the first known target of SSc-mediated injury, the endothelial cell, or its partner the pericyte. Another possibility is that a small driver population of aberrant fibroblasts or pericytes, which creates an altered cytokine milieu and matrix, is progressively lost in culture. A third, related possibility is that fibroblast-synthesized matricellular proteins assembled in the dermis contribute to an autocrine loop causing altered fibroblast behavior, and that the process of explantation and growth in monolayer causes progressive loss of those signals. Our results demonstrate that multicenter studies of the gene profiles in SSc skin can generate meaningful results that are not confounded by the source of material, and show that the SSc transcript profile is very robust. We also observed previously unreported changes in matrix expression, such as a disproportionate upregulation of type XI collagen, which suggest a move toward fibrocartilage in the fibrotic process. Without comparison with other fibrotic conditions, it is impossible to comment on the specificity of the gene profile changes for scleroderma, but the consistency does suggest that this method of analysis is sufficiently stable that, with larger sample numbers, different subtypes of SSc may become discernable. Indeed, a sample obtained from 1 patient with SSc, who was notably younger than the other patients, was something of an outlier (Figure 1). We have validated the use of a room-temperature/ 4°C RNA-stabilizing agent (RNAlater) to ease sample collection, and have since determined that this form of preservation is effective for immunohistochemical analyses (data not shown). Thus, prospects are good for obtaining meaningful results from remote biopsy specimens and centralized Affymetrix analysis for a large number of patients with this and other relatively rare diseases. Conversely, the utility of explanted scleroderma fibroblasts in the study of SSc is clearly circumscribed. Phenotypic characteristics of SSc fibroblasts appear to be degraded independent of the dermal milieu, since their differential expression versus that of control fibroblasts is reduced relative to the biopsy samples and shows higher variance, for reasons that are yet to be elucidated. Nonetheless, fibroblast culture remains a necessary technique for assessing therapeutic strategies and the function of humoral factors ex vivo. Finally, in this study, we did not witness a strong signature from B cells or other cells of the immune system. Dedicated to the memory of Joseph Korn, professor, advocate, and student of scleroderma. DISTINCT GENE PROFILE OF SCLERODERMA 1973 ACKNOWLEDGMENTS Dr. Gardner thanks Anthony Guidi and Jim Kiernan for consultation in sample handling, Margaret McCrann for pilot hybridization experiments, John McCoy for resource support, and Uwe Hansen for valuable discussions. Dr. Mayes thanks Julio Charles for assistance in biopsy processing. 15. 16. 17. REFERENCES 1. Gussin HA, Ignat GP, Varga J, Teodorescu M. Anti– topoisomerase I (anti–Scl-70) antibodies in patients with systemic lupus erythematosus. Arthritis Rheum 2001;44:376–83. 2. Kane GC, Varga J, Conant EF, Spirn PW, Jimenez S, Fish JE. Lung involvement in systemic sclerosis (scleroderma): relation to classification based on extent of skin involvement or autoantibody status. Respir Med 1996;90:223–30. 3. Krieg T, Perlish JS, Fleischmajer R, Braun-Falco O. Collagen synthesis in scleroderma: selection of fibroblast populations during subcultures. Arch Dermatol Res 1985;277:373–6. 4. Chang HY, Chi JT, Dudoit S, Bondre C, van de Rijn M, Botstein D, et al. Diversity, topographic differentiation, and positional memory in human fibroblasts. Proc Natl Acad Sci U S A 2002;99: 12877–82. 5. Szulgit G, Rudolph R, Wandel A, Tenenhaus M, Panos R, Gardner H. Alterations in fibroblast ␣1␤1 integrin collagen receptor expression in keloids and hypertrophic scars. J Invest Dermatol 2002;118:409–15. 6. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 2000;403:503–11. 7. Subcommittee for Scleroderma Criteria of the American Rheumatism Association Diagnostic and Therapeutic Criteria Committee. Preliminary criteria for the classification of systemic sclerosis (scleroderma). Arthritis Rheum 1980;23:581–90. 8. LeRoy EC, Black C, Fleischmajer R, Jablonska S, Krieg T, Medsger TA, Jr, et al. Scleroderma (systemic sclerosis): classification, subsets and pathogenesis. J Rheumatol 1988;15:202–5. 9. Whitfield ML, Finlay DR, Murray JI, Troyanskaya OG, Chi JT, Pergamenschikov A, et al. Systemic and cell type-specific gene expression patterns in scleroderma skin. Proc Natl Acad Sci U S A 2003;100:12319–24. 10. Scott IC, Imamura Y, Pappano WN, Troedel JM, Recklies AD, Roughley PJ, et al. Bone morphogenetic protein-1 processes probiglycan. J Biol Chem 2000;275:30504–11. 11. Gardner H, Strehlow D, Bradley L, Widom R, Farina A, de Fougerolles A, et al. Global expression analysis of the fibroblast transcriptional response to TGF␤. Clin Exp Rheumatol 2004;22(3 Suppl 33):S47–57. 12. Benjamin M, Ralphs JR. Biology of fibrocartilage cells. Int Rev Cytol 2004;233:1–45. 13. Renzoni EA, Abraham DJ, Howat S, Shi-Wen X, Sestini P, Bou-Gharios G, et al. Gene expression profiling reveals novel TGF␤ targets in adult lung fibroblasts. Respir Res 2004;5:24. 14. Matheson S, Larjava H, Hakkinen L. Distinctive localization and function for lumican, fibromodulin and decorin to regulate colla- 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. gen fibril organization in periodontal tissues. J Periodontal Res 2005;40:312–24. Uitto J, Ohlenschlager K, Lorenzen I. Solubility of skin collagen in normal human subjects and in patients with generalised scleroderma. Clin Chim Acta 1971;31:13–8. Davis EC, Blattel SA, Mecham RP. Remodeling of elastic fiber components in scleroderma skin. Connect Tissue Res 1999;40: 113–21. Sapadin AN, Esser AC, Fleischmajer R. Immunopathogenesis of scleroderma: evolving concepts. Mt Sinai J Med 2001;68:233–42. Pannu J, Trojanowska M. Recent advances in fibroblast signaling and biology in scleroderma. Curr Opin Rheumatol 2004;16: 739–45. Mimura Y, Ihn H, Jinnin M, Asano Y, Yamane K, Tamaki K. Constitutive thrombospondin-1 overexpression contributes to autocrine transforming growth factor-␤ signaling in cultured scleroderma fibroblasts. Am J Pathol 2005;166:1451–63. Zhou Y, Hagood JS, Murphy-Ullrich JE. Thy-1 expression regulates the ability of rat lung fibroblasts to activate transforming growth factor-␤ in response to fibrogenic stimuli. Am J Pathol 2004;165:659–69. Gregory CA, Singh H, Perry AS, Prockop DJ. The Wnt signaling inhibitor dickkopf-1 is required for reentry into the cell cycle of human adult stem cells from bone marrow. J Biol Chem 2003;278: 28067–78. Attisano L, Labbe E. TGF␤ and Wnt pathway cross-talk. Cancer and Metastasis Rev 2004;23:53–61. Igarashi A, Nashiro K, Kikuchi K, Sato S, Ihn H, Grotendorst GR, et al. Significant correlation between connective tissue growth factor gene expression and skin sclerosis in tissue sections from patients with systemic sclerosis. J Invest Dermatol 1995;105:280–4. Shi-wen X, Pennington D, Holmes A, Leask A, Bradham D, Beauchamp JR, et al. Autocrine overexpression of CTGF maintains fibrosis: RDA analysis of fibrosis genes in systemic sclerosis. Exp Cell Res 2000;259:213–24. Trojanowska M. Molecular aspects of scleroderma. Front Biosci 2002;7:608–18. Gore-Hyer E, Pannu J, Smith EA, Grotendorst G, Trojanowska M. Selective stimulation of collagen synthesis in the presence of costimulatory insulin signaling by connective tissue growth factor in scleroderma fibroblasts. Arthritis Rheum 2003;48:798–806. Grzeszkiewicz TM, Kirschling DJ, Chen N, Lau LF. CYR61 stimulates human skin fibroblast migration through integrin ␣ v␤ 5 and enhances mitogenesis through integrin ␣v␤3, independent of its carboxyl-terminal domain. J Biol Chem 2001;276:21943–50. Grzeszkiewicz TM, Lindner V, Chen N, Lam SC, Lau LF. The angiogenic factor cysteine-rich 61 (CYR61, CCN1) supports vascular smooth muscle cell adhesion and stimulates chemotaxis through integrin ␣(6)␤(1) and cell surface heparan sulfate proteoglycans. Endocrinology 2002;143:1441–50. Schutze N, Noth U, Schneidereit J, Hendrich C, Jakob F. Differential expression of CCN-family members in primary human bone marrow-derived mesenchymal stem cells during osteogenic, chondrogenic and adipogenic differentiation. Cell Commun Signal 2005;3:5. Liu Y, Lyle S, Yang Z, Cotsarelis G. Keratin 15 promoter targets putative epithelial stem cells in the hair follicle bulge. J Invest Dermatol 2003;121:963–8.