Advertisement
Research Article

Peripheral Blood Gene Expression as a Novel Genomic Biomarker in Complicated Sarcoidosis

  • Tong Zhou equal contributor,

    equal contributor Contributed equally to this work with: Tong Zhou, Wei Zhang

    Affiliations: Institute for Personalized Respiratory Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America, Section of Pulmonary, Critical Care, Sleep and Allergy, Department of Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America

    X
  • Wei Zhang equal contributor,

    equal contributor Contributed equally to this work with: Tong Zhou, Wei Zhang

    Affiliations: Institute of Human Genetics, The University of Illinois at Chicago, Chicago, Illinois, United States of America, Department of Pediatrics, The University of Illinois at Chicago, Chicago, Illinois, United States of America

    X
  • Nadera J. Sweiss,

    Affiliation: Section of Rheumatology, Department of Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America

    X
  • Edward S. Chen,

    Affiliation: Division of Pulmonary and Critical Care Medicine, Department of Medicine, The Johns Hopkins University, Baltimore, Maryland, United States of America

    X
  • David R. Moller,

    Affiliation: Division of Pulmonary and Critical Care Medicine, Department of Medicine, The Johns Hopkins University, Baltimore, Maryland, United States of America

    X
  • Kenneth S. Knox,

    Affiliation: Section of Pulmonary and Critical Care, Department of Medicine, The University of Arizona, Tuscon, Arizona, United States of America

    X
  • Shwu-Fan Ma,

    Affiliation: Section of Pulmonary/Critical Care, Department of Medicine, The University of Chicago, Chicago, Illinois, United States of America

    X
  • Michael S. Wade,

    Affiliations: Institute for Personalized Respiratory Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America, Section of Pulmonary, Critical Care, Sleep and Allergy, Department of Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America

    X
  • Imre Noth,

    Affiliation: Section of Pulmonary/Critical Care, Department of Medicine, The University of Chicago, Chicago, Illinois, United States of America

    X
  • Roberto F. Machado,

    Affiliations: Institute for Personalized Respiratory Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America, Section of Pulmonary, Critical Care, Sleep and Allergy, Department of Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America

    X
  • Joe G. N. Garcia mail

    jggarcia@uic.edu

    Affiliations: Institute for Personalized Respiratory Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America, Section of Pulmonary, Critical Care, Sleep and Allergy, Department of Medicine, The University of Illinois at Chicago, Chicago, Illinois, United States of America

    X
  • Published: September 12, 2012
  • DOI: 10.1371/journal.pone.0044818

Abstract

Sarcoidosis, a systemic granulomatous syndrome invariably affecting the lung, typically spontaneously remits but in ~20% of cases progresses with severe lung dysfunction or cardiac and neurologic involvement (complicated sarcoidosis). Unfortunately, current biomarkers fail to distinguish patients with remitting (uncomplicated) sarcoidosis from other fibrotic lung disorders, and fail to identify individuals at risk for complicated sarcoidosis. We utilized genome-wide peripheral blood gene expression analysis to identify a 20-gene sarcoidosis biomarker signature distinguishing sarcoidosis (n = 39) from healthy controls (n = 35, 86% classification accuracy) and which served as a molecular signature for complicated sarcoidosis (n = 17). As aberrancies in T cell receptor (TCR) signaling, JAK-STAT (JS) signaling, and cytokine-cytokine receptor (CCR) signaling are implicated in sarcoidosis pathogenesis, a 31-gene signature comprised of T cell signaling pathway genes associated with sarcoidosis (TCR/JS/CCR) was compared to the unbiased 20-gene biomarker signature but proved inferior in prediction accuracy in distinguishing complicated from uncomplicated sarcoidosis. Additional validation strategies included significant association of single nucleotide polymorphisms (SNPs) in signature genes with sarcoidosis susceptibility and severity (unbiased signature genes - CX3CR1, FKBP1A, NOG, RBM12B, SENS3, TSHZ2; T cell/JAK-STAT pathway genes such as AKT3, CBLB, DLG1, IFNG, IL2RA, IL7R, ITK, JUN, MALT1, NFATC2, PLCG1, SPRED1). In summary, this validated peripheral blood molecular gene signature appears to be a valuable biomarker in identifying cases with sarcoidoisis and predicting risk for complicated sarcoidosis.

Introduction

Individuals with sarcoidosis, a systemic inflammatory and non-caseating granulomatous disease of unknown origin affecting multiple organs and invariably the lung [1], [2], typically undergo spontaneous resolution. However, ~20% of affected individuals experience progressive disease with respiratory, cardiac or nervous system involvement. Complicated sarcoidosis is defined as exhibiting either cardiac manifestations (e.g., ventricular arrhythmias) [3], neurologic involvement (e.g., with evidence of hyperdense MRI lesions) [4] or deteriorating lung function (e.g., FVC <50%). Currently, FDA-approved therapies for complicated sarcoidosis do not exist and corticosteroids and corticosteroid-sparing immunosuppressive agents (TNFα inhibitors) have met with only limited success [5]. The accurate identification of individuals with or at risk for complicated sarcoidosis is a vexing clinical challenge with attempts to define clinically-useful biomarkers largely unsuccessful. Sarcoidosis biomarkers are desperately needed to deliver targeted therapies in individuals with complicated sarcoidosis and to identify patients at risk for increased morbidity and significant mortality as a consequence of complicated sarcoidosis.

Previous sarcoidosis candidate gene studies focused on granuloma formation and immune response pathways implicated several genes linked to sarcoidosis susceptibility [6] including HLA antigens such as class I HLA-B8 [7] and HLA-DRB1 [6], [8], [9]. Additional candidate genes involved in antigen processing, antigen presentation, macrophage and T-cell activation, and injury repair have also been associated with sarcoidosis susceptibility [10][23]. Whole genome scanning studies based upon unbiased, genome-based approaches identified genes implicated in sarcoidosis susceptibility via linkage analysis (i.e., D6S1666 in 63 German families with affected siblings) [24] with further scanning suggesting rs2076530 in BTNL2 (butyrophilin-like 2) gene to be associated with sarcoidosis development [25]. A significant challenge remains, however, in the assessment of sarcoidosis susceptibility in specific high-risk populations as well as in the identification of sarcoidosis patients at risk for complicated, progressive disease.

Our study was designed to identify novel genomic biomarkers by comparing genome-wide gene expression data in African American (AA) and European descent ancestry (EA) sarcoidosis cases. We identified a universal gene signature that differentiates sarcoidosis patients from healthy controls and distinguishes complicated sarcoidosis (pulmonary- FVC<50%, cardiac, or neurologic sarcoidosis) from uncomplicated sarcoidosis. This gene signature was superior in prediction accuracy in each of the AA and EA populations when compared to a second signature comprised of genes within the T cell receptor–innate immunity pathway that includes genes previously associated with sarcoidosis. These signatures distinguished sarcoidosis patients from idiopathic pulmonary fibrosis (IPF) cases with signature validation provided by significant association of genetic variants within signature genes with sarcoidosis susceptibility. These results highlight the utility of peripheral blood molecular gene signatures as valuable biomarkers for predicting individuals at risk for complicated sarcoidosis and for potentially facilitating individualized therapies in this enigmatic disorder.

Results

Patient Characteristics

PBMC samples were collected from subjects with sarcoidosis (n = 39) and healthy controls (n = 35) (Table 1). The clinical characteristics of study patients are displayed in Table 2. Significant differences in age, gender, race and pulmonary function studies did not exist between uncomplicated and complicated sarcoidosis cases (P>0.05 by χ2 test for gender and p>0.05 by t-test for the other characteristics). Uncomplicated sarcoidosis cases trended toward higher corticosteroid usage whereas complicated sarcoidosis cases trended toward higher methotrexate usage and were more likely to be receiving anti-TNFα therapy. However, these differences were not statistically significant (P>0.05 for all drugs) (Table 2). Predictably, complicated pulmonary sarcoidosis cases exhibited significantly reduced pulmonary function compared to the other study groups (data not shown).

thumbnail

Table 1. Study subjects with racial and complication status.

doi:10.1371/journal.pone.0044818.t001
thumbnail

Table 2. Patient characteristics and concomitant medications.

doi:10.1371/journal.pone.0044818.t002

Identification of Differentially-expressed Genes in Sarcoidosis

All cases with diagnoses of cardiac, neurologic, or severe pulmonary sarcoidosis (FVC<50%) comprised the cohort labeled as ‘complicated sarcoidosis’. At the specified significance level (fold-change >1.4, q-value <0.05), 316 genes were differentially expressed between all sarcoidosis cases and healthy controls in the combined samples (pooled AAs and EAs). For individual populations, 118 genes were differentially-expressed between all AA cases and controls, whereas 861 genes were differentially expressed between all EA cases and controls. In contrast, 1124 genes were differentially expressed between complicated sarcoidosis cases and healthy controls in the combined samples. For individual population, 730 and 980 genes were differentially expressed between AA and EA cases with complicated sarcoidosis and healthy controls, respectively with the TCR signaling pathway significantly enriched among complicated sarcoidosis-associated genes in both populations (adjusted P<0.05) (Figure 1A).

thumbnail

Figure 1. Identifying gene signatures in sarcoidosis. Panel A. Enriched pathways among complicated sarcoidosis-associated genes.

The top ranking KEGG pathways are listed for each population. The red line indicates the cutoff of significance (adjusted p-value<0.05). The number of genes in each pathway is shown beside the pathway name. Panel B. Heatmap of patients with complicated sarcoidosis and healthy controls. Red represents increased gene expression; Blue represents down-regulation. “++”: patients with complicated sarcoidosis; “−”: healthy controls. Panel C. Principal component analysis on expression values of the 20-gene signature. X-axis: principal component 1 with eigenvalue; Y-axis: principal component 2 with eigenvalue. Left panel: patients with complicated sarcoidosis and healthy controls; middle panel: patients with complicated sarcoidosis, uncomplicated sarcoidosis and healthy controls; and right panel: patients with complicated sarcoidosis and uncomplicated sarcoidosis. HC: healthy controls; US: patients with uncomplicated sarcoidosis; and CS: patients with complicated sarcoidosis. Panel D. Comparison between the 20-gene signature and the TCR/JS/CCR signaling pathway gene signature. The distribution of prediction accuracy is based on 1,000 times of five-fold cross-validation. The dashed lines indicate the average classification accuracy for the 20-gene signature or the TCR/JS/CCR signaling pathway gene signature. Left panel: all sarcoidosis patients versus healthy controls; and right panel: patients with complicated sarcoidosis versus patients with uncomplicated sarcoidosis.

doi:10.1371/journal.pone.0044818.g001

Identifying a Gene Signature for Complicated Sarcoidosis

To identify a universal gene signature for complicated sarcoidosis in both AA and EA populations, an initial analysis set comprised of 1233 genes differentially expressed between AA or EA complicated sarcoidosis cases vs. healthy controls was utilized for the SVM algorithm. Figure S1 depicts the distribution of the prediction accuracy for gene signatures with the number of genes during recursive feature selection (see Supplementary Text S1 for details). A 20-gene signature (Table 3) was chosen as the most parsimonious signature with the peak prediction accuracy (Figure S1) and accurately distinguished patients with complicated sarcoidosis from healthy controls (Figures 1B and 1C), or from uncomplicated sarcoidosis (Figure 1C). Two genes within the unbiased 20-gene signature, HBEGF (heparin-binding EGF-like growth factor) and SAP30 (Sin3A-associated protein, 30kDa), were strongly up-regulated in complicated sarcoidosis whereas the remaining 18 signature genes were down-regulated in complicated sarcoidosis (Figure S1). The non-targeted 20-gene signature distinguished all sarcoidosis patients from healthy controls with an accuracy of 86.0% (sensitivity = 88.2% and specificity = 83.3%) in the combined samples (pooled AAs and EAs) (Figure 1D). The discriminative accuracy became 88.2% and 94.2% in separating sarcoidosis cases from healthy controls in AA and EA, respectively (Figure S2). When distinguishing complicated sarcoidosis cases from uncomplicated sarcoidosis cases, the accuracy was 81.4% (sensitivity = 87.0% and specificity = 74.2%) in the combined samples (Figure 1D) but was reduced to 83.7% and 64.5% in separating complicated sarcoidosis cases from uncomplicated sarcoidosis cases in AA and EA, respectively (Figure S2).

thumbnail

Table 3. The unbiased 20-gene signature for complicated sarcoidosis.

doi:10.1371/journal.pone.0044818.t003

Evaluation of a Sarcoidosis-related TCR/JS/CCR Signaling Pathway Gene Signature

As the T cell receptor pathway (TCR), the JAK STAT signaling pathway (JS) and the cytokine-cytokine receptor signaling pathway (CCR) have all been implicated in sarcoidosis [6], [26], a 31 gene signature comprised of TCR/JS/CCR signaling pathway genes implicated associated with sarcoidosis was assessed as a potential molecular biomarker in identifying cases or risk for complicated sarcoidosis (Table 4) (see Supplementary Text S1 for details). Overall, this TCR/JS/CCR signaling pathway signature differentiated sarcoidosis from healthy controls with a prediction accuracy of 82.2% (Figure 1D), but exhibited a substantially reduced prediction accuracy of <60% in distinguishing complicated sarcoidosis from uncomplicated sarcoidosis (Figure 1D). The discriminative accuracy of this TCR/JS/CCR signature was 83.2% in separating all AA sarcoidosis patients from healthy controls but only 69.7% for distinguishing AA complicated sarcoidosis cases from uncomplicated sarcoidosis. Similarly, in EA cases, the accuracy of the TCR/JS/CCR signature was 75.1% for distinguishing sarcoidosis patients from healthy controls, but only 37.5% in distinguishing EA patients with complicated sarcoidosis from uncomplicated EA sarcoidosis cases. Comparison of the prediction accuracy in both the TCR/JS/CCR and unbiased 20-gene signatures in combined EA and AA cases (Figure 1D, Figure S3) revealed the superior performance of the unbiased 20-gene sarcoidosis signature (P<10−15 by t-test). Finally, as sarcoidosis and IPF represent the most common interstitial lung diseases (ILDs) of unknown etiology, the capacity for the unbiased 20-gene and TCR/JS/CCR sarcoidosis gene signatures to distinguish sarcoidosis cases from IPF cases (GEO - GSE38958) was assessed. Each signature performed with comparable prediction accuracy in IPF and sarcoid with the 20-gene signature (77.2%) slightly superior to the TCR/JS/CCR signaling pathway signature (76.5%) in distinguishing sarcoidosis from IPF cases (Figure S4, P<10−5 by t-test).

thumbnail

Table 4. The 31 differentially-expressed TCR/JS/CRR signaling pathway genes in sarcoidosis.

doi:10.1371/journal.pone.0044818.t004

Validation on Independent Datasets

We evaluated the performance of our gene signatures in two different independent sarcoidosis blood gene expression datasets. One dataset (GEO - GSE19314) from University of California, San Francisco (UCSF) [27] and another one (GEO - GSE18781) is from Oregon Health Sciences University (Oregon) [28]. The discriminative power is very similar between the unbiased 20-gene and the TCR/JS/CCR signatures in the both datasets. The 20-gene signature classified sarcoidosis cases from healthy controls with accuracy of 75.9% and 78.3% for the USCF and Oregon datasets, respectively, while the discriminative accuracy became 75.4% and 80.0% when the TCR/JS/CCR signature was applied for the USCF and Oregon datasets, respectively (Figure 2). Again, principal component analysis indicates that patients with sarcoidosis can be well distinguished from healthy controls in the two independent datasets, just based on the expression of our unbiased 20-gene signature (Figure 2).

thumbnail

Figure 2. Validation in independent datasets.

The upper panels show the comparison between the 20-gene signature and the TCR/JS/CCR signaling pathway gene signature. The distribution of prediction accuracy is based on 1,000 times of five-fold cross-validation. The dashed lines indicate the average classification accuracy for the 20-gene signature or the TCR/JS/CCR signaling pathway gene signature. The lower panels show the results of principal component analysis on expression values of the 20-gene signature. X-axis: principal component 1 with eigenvalue; Y-axis: principal component 2 with eigenvalue.

doi:10.1371/journal.pone.0044818.g002

Use of Genetic Variants to Validate Sarcoidosis Gene Signatures

A genome-wide association study (GWAS) (Affymetrix 6.0 SNP array) involving 407 sarcoidosis cases including 212 AAs (including 68 complicated cases) and 195 EAs (including 46 complicated cases) was performed and allele frequencies of ~1,300 common SNPs residing in unbiased sarcoidosis signature genes analyzed in sarcoidosis cases and healthy controls (see Supplementary Text S1 for details). At the nominal P-value <0.01, 30 SNPs from 6 unbiased 20-gene signature genes were found to be significantly associated with sarcoidosis (Table 5), including 4 genes which overlapped between the AA and EA samples (NOG [noggin], RMB12B [RNA binding motif protein 12B], SESN3 [sestrin 3], TSHZ2 [teashirt zinc finger homeobox 2]). The most highly significant signature gene SNP in AAs was rs629508 (P = 1.7×10−3) in SESN3, whereas in EA cases, the most significant SNP was rs2618134 (P = 4.7×10−5) in RBM12B. Interestingly, several SNPs were also significantly associated with complicated sarcoidosis, including rs629508 (P = 5.4×10−5) and rs1294689 (P = 3.6×10−5) in the AA samples and rs10485815 (P = 2.8×10−5) in the EA samples (Table 5). In comparison, from ~3,800 common SNPs residing in TCR/JS/CCR signature genes, 37 SNPs were associated with sarcoidosis in AA samples, whereas 34 SNPs were significant in EA samples, respectively (Table S1). The most highly significant TCR-JS-CCR signature gene SNP in AAs was rs2131817 (P = 1.4×10−5) in AKT3, whereas in EA cases, the most significant SNP was rs7614488 (P = 7.8×10−7) in CBLB. Several TCR/JS/CCR signature gene SNPs, rs2953040 and rs6791765 in CBLB (Cas-Br-M, murine, ecotropic retroviral transforming sequence b) and rs2131817 in AKT3 were significantly associated with sarcoidosis in both EA and AA sarcoidosis cases (P<0.01) (Table S1).

thumbnail

Table 5. SNPs significantly associated with sarcoidosis within the unbiased 20 signature genes (P<0.01).

doi:10.1371/journal.pone.0044818.t005

PubMatrix Evaluation

The medical informatic tool PubMatrix (http://pubmatrix.grc.nia.nih.gov) tool was next used to evaluate the relevance of sarcoidosis signature genes in the published biomedical literatures (PubMed). Each signature gene was searched against a series of terms related to lung fibrosis or sarcoidosis including: “sarcoidosis”, “tuberculosis”, “granulomatous disease”, “hypersensitivity pneumonitis”, and “pulmonary fibrosis”. The majority of 20-gene signature genes were highly novel to these terms (Table 6) with only 2/20 genes having any PubMed citations linked to these terms (HBEGF, LOC100132356). Of the 31 TCR/JS/CCR-gene signature genes, 8/31 genes were cited in the sarcoidosis literature with CD28, IFNG, IL7R, AKT3, IL2RA, IL2RB, and STAT4 demonstrating a robust relationship with these terms (Table S2).

thumbnail

Table 6. PubMatrix search results for the 20-gene signature against sarcoidosis-related search terms.

doi:10.1371/journal.pone.0044818.t006

Discussion

The major aim of this work was to identify potential universal and racially-specific gene signatures to serve as novel biomarkers for the presence of sarcoidosis as well as for the presence and/or susceptibility of the development of complicated sarcoidosis. Leveraging whole genome expression profiles in a cohort of sarcoidosis patients, an unbiased gene signature comprised of 20 autosomal genes was identified which distinguished sarcoidosis cases from healthy individuals and, importantly, differentiated patients with complicated sarcoidosis from patients with uncomplicated sarcoidosis. The 20-gene signature exhibited equivalent prediction accuracy to other sarcoidosis signatures containing a greater number of genes (such as 39-gene and 78-gene sarcoidosis signatures) with each signature superior in accuracy to signatures with fewer genes (e.g., the 10 gene signature) (Figure S1). The expression levels of the majority of these 20 signature genes showed a pattern of an additive model between uncomplicated and complicated sarcoidosis (Figure 3), i.e., when the signature gene is up-regulated, patients with complicated sarcoidosis exhibited higher expression levels than patients with uncomplicated sarcoidosis. In the sarcoidosis signature, 19 of 20 genes performed unidirectionally (up-regulation or down-regulation) in both complicated and uncomplicated sarcoidosis. Therefore, the 20-gene signature appears to not only capture differences between complicated sarcoidosis and healthy controls, but potentially conveys information regarding differences between sarcoidosis cases (both complicated and uncomplicated) and healthy controls.

thumbnail

Figure 3. Boxplot of expression of the 20 signature genes.

The dark grey points and lines indicate the geometric mean of expression in each category. HC: healthy controls; US: patients with uncomplicated sarcoidosis; and CS: patients with complicated sarcoidosis. Y-axis: log2-transformed expression values.

doi:10.1371/journal.pone.0044818.g003

Gene products encoded by TCR/JS/CCR signaling pathway genes have been implicated in sarcoidosis pathogenesis [6], [26] and these signature genes were enriched among the differential genes between EA and AA cases with complicated sarcoidosis cases and healthy controls. The utility of a TCR/JS/CCR signaling pathway gene signature in classifying sarcoidosis cases was compared to the unbiased 20-gene signature. Both signatures performed with high level prediction accuracy (>80%) in distinguishing cases with sarcoidosis from healthy controls. In contrast, the prediction accuracy of the 20-gene signature was much superior to the TCR/JS/CCR signaling pathway gene signature in classifying combined AA and EA patients with complicated and uncomplicated sarcoidosis (81.4% vs. 58.8%, P<10−15, t-test). We speculate that the unbiased nature of the 20-gene signature allows better capture of the characteristics of complicated sarcoidosis compared to the more restrictive TCR/JS/CCR signaling pathway signature genes. The potential role of TCR/JS/CCR signaling pathways genes in the development of sarcoidosis was confirmed by the capacity of this signature to successfully differentiate the majority of sarcoidosis and healthy controls. However, we speculate that either sarcoidosis disease progression or the development of complicated sarcoidosis likely requires the participation of genes and pathways extending beyond the TCR/JS/CCR pathway. These findings underscore the complex pathobiology of this disorder and implicate the necessity of global and unbiased approaches.

We further evaluated the classification accuracy of the 20-gene sarcoid signature separately in EA and AA samples and found the 20-gene signature to demonstrate >85% accuracy for classifying either EA or AA sarcoidosis cases (complicated and uncomplicated) from healthy controls. In contrast, the 20-gene sarcoidosis signature differentiated complicated sarcoidosis and uncomplicated sarcoidosis cases with an accuracy >80% in AA cases, but only ~60% in EA cases, potentially the relative smaller complicated EA sample size or a bias for AA expression dysregulation driven by greater genetic variation, an issue which requires further examination. Both the 20-gene signature and TCR/JS/CCR-gene signature successfully discriminated sarcoidosis cases from IPF patients with similar prediction accuracies reflecting the differences in immunopathogenesis, clinical course, prognosis, and response to steroid treatment [29] in these two fibrotic lung disorders. This finding may infer additional clinical utility of the signature as a diagnostic biomarker for sarcoidosis.

As evidenced by the paucity of PubMed citations (PubMatrix results), the 20-gene signature is comprised of highly novel candidate genes in sarcoidosis susceptibility and severity of disease. As a complementary method to validate our findings [30][35], we examined the allele frequencies of both unbiased 20-gene sarcoidosis signature single nucleotide polymorphisms (SNPs) as well as TCR/JS/CCR signaling pathway signature gene SNPs in sarcoidosis cases and healthy controls embedded within a GWAS dataset constructed by genome-wide assessment of genetic variants in over 400 EA and AAs with sarcoidosis. As genetic variants, such as SNPs and copy number variants (CNVs), contribute significantly to variations in gene expression, SNPs were annotated to the genomic regions of these signature genes (based on the Affymetrix annotation) and, therefore, potentially contribute to gene expression variation by acting as cis-eQTLs. From ~1,300 SNPs in our 20 signature genes, we identified 30 SNPs (corresponding to 6 signature genes) which were significantly associated with sarcoidosis in either EA or AA samples, suggesting a potential role of these cis-acting SNPs in regulating the expression of sarcoidosis signature genes. Similarly, from ~3,800 SNPs in TCR/JS/CCR signature genes, relationships between SNPs and sarcoidosis were observed. While these findings serve to validate the potential importance and relevance of signature genes, a direct association between these SNPs and expression is necessary to validate these relationships. Our results suggest that genetic variants via cis-acting eQTLs may contribute to the variation in expression of sarcoidosis signature genes. We further recognize that additional factors, such as trans-acting eQTLs, environmental factors, or epigenetic pathways, may contribute substantially to signature gene expression variation. Further investigations involving genome-wide genotypic data (e.g., for mapping trans-acting eQTLs) and expression data on the same samples could potentially provide greater insights into the contribution of genetics to the identified gene signature.

Quantitative abnormalities in T cells have been described in the peripheral blood of patients with sarcoidosis [36] with significant lymphopenia, involving CD4, CD8, and CD19 positive cells, common in sarcoidosis patients and correlating with disease severity [37]. Individual signatures genes may not only have a role in the pathophysiology of sarcoidosis but could be potentially approached as novel therapeutic targets for the disease. For example, HBEGF, a member of the EGF family of growth factors, is a potent mitogen and chemoattractant for many cell types including fibroblasts, smooth muscle cells and epithelial cells [38][41]. A substantial body of evidence suggests that HBGEF plays a role in wound healing and response to injury [42][45] leading to speculation that HBEGF may represent a target involved in the pathobiology of chronic lung sarcoidosis and a novel therapeutic target, an observation supported by the PubMatrix search results.

Among our 20-gene signature, LOC100132356 was most cited in PubMed literatures, though it only codes a hypothetical protein. This gene was linked to the terms such as sarcoidosis, tuberculosis, granulomatous disease, hypersensitivity pneumonitis, and pulmonary fibrosis. However, the detailed function of this gene is still unclear.

Recently, lung gene expression profiles were compared between patients with self-limiting sarcoidosis and those with progressive restrictive fibrotic disease [46] with a greater number of down-regulated genes versus up-regulated genes identified in patients with progressive pulmonary sarcoidosis. These findings are highly consistent with the expression profile of our signature genes in patients with complicated sarcoidosis. Interestingly, we failed to identify any overlap between sarcoidosis signature genes and the differentially expressed genes produced by comparison of self-limited and progressive lung sarcoidosis. The lack of overlap may reflect greater severity of disease in our cohort with cardiac and neurologic sarcoidosis in addition to cases with severe lung disease. In addition, our studies did not involve lung tissue expression but rather analysis of PBMCs and therefore tissue-specific expression may also contribute to this lack of overlap.

Furthermore, our sarcoidosis gene signatures performed well in two independent validation cohorts (UCSF and Oregon) [27], [28]. We should point out two challenges in our validation. Firstly, our microarray platform (Affymetrix Human Exon 1.0 ST Array) was different from that used for the validation cohorts (Affymetrix Human Genome U133 Plus 2.0 Array). Secondly, our study focused on gene expression in PBMCs while whole blood expression profiles were analyzed for the USCF cohort [27].

In summary, despite significant limitations including a relatively small size of the EA complicated cases in the analysis set, an unbiased 20-gene molecular gene signature was identified as a potential novel molecular biomarker in the diagnosis of sarcoidosis as well for the presence of complicated sarcoidosis with substantial accuracy in both EA and AA sarcoidosis cases. With validation in a replicate sarcoidosis cohort and testing against other granulomatous disorders like Wegener’s disease, hypersensitivity pneumonitis, and tuberculosis, this sarcoidosis gene signature may represent a novel universal gene signature for complicated sarcoidosis and serve as a springboard to individualized therapies in this enigmatic disorder.

Materials and Methods

Subjects and PBMC Samples

The study was approved by the Institutional Review Board (IRB) of the University of Illinois at Chicago (UIC) with written informed consent obtained from all subjects. The UIC’s IRB committee members (Chairs) include: Indru Punwani, D.D.S., Susan Labott, Ph.D., Paul Heckerling, M.D., and Kathryn Rugen, Ph.D. The DNA samples provided by the Johns Hopkins University investigators, and their use in this study, were approved by the IRB of the Johns Hopkins University. PBMC samples were collected from subjects with sarcoidosis (n = 39) and healthy controls (n = 35) (Table 1). The diagnosis of sarcoidosis was based on established joint international criteria [47]. Subjects with other concurrent systemic inflammatory diseases were excluded. A total of 29 African descent American (AA) and 10 European descent American (EA) patients with sarcoidosis were included in the overall sarcoidosis cohort with 18 AA and 4 EA patients diagnosed with complicated sarcoidosis defined as cardiac sarcoidosis (e.g., ventricular arrhythmias) [3], neurologic sarcoid (e.g., evidence of hyperdense MRI lesions) [4] or severe pulmonary sarcoidosis (FVC<50%). The detailed description of the therapy status of each patient has been listed in Table S3.

RNA Microarray Hybridization

Total RNA was isolated from PBMCs using standard molecular biology protocols (n = 74) without DNA contamination or RNA degradation. Sample processing (e.g., cDNA generation, fragmentation, end labeling, hybridization to Affymetrix GeneChip Human Exon 1.0 ST arrays) was performed by the University of Chicago Functional Genomics Facility per manufacturer’s instructions.

Identification of Genes Differentially Expressed in Sarcoidosis and Complicated Sarcoidosis

Human Exon 1.0 ST arrays were summarized using the Affymetrix Power Tools v.1.12.0 (http://www.affymetrix.com/) (see Supplementary Text S1 for details). The microarray data has been uploaded into NCBI GEO database (GEO accession number: GSE37912). Genes on chromosomes X and Y were removed to avoid the potential confounding factor of gender. SAM (Significance Analysis of Microarrays) [48], implemented in the samr library of the R Statistical Package [49], was used to compare log2-transformed gene expression levels between patients with complicated sarcoidosis and normal controls in the combined (AA and EA), EA, and AA samples, respectively. False discovery rate (FDR) was controlled using the q-value method [50]. Transcripts with a fold-change greater than 1.4 and q-value less than 0.05 were deemed differentially expressed. We searched for any enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) [51] physiological pathways among the differential genes relative to the final analysis set using the NIH/DAVID [52], [53]. An adjusted P-value<0.05 after the Benjamini-Horchberg procedure [54] was used as the cutoff.

Identification of Gene Signature for Classifying Sarcoidosis and Complicated Sarcoidosis

To identify gene signatures useful in the diagnosis and classification of sarcoidosis, a machine learning algorithm based on support vector machine (SVM) using a linear kernel, was applied in combination with recursive feature elimination (RFE) for generating a predictive model (see Supplementary Text S1 for details) [55][58]. The e1071 library of the R Statistical Package [49] was used to conduct SVM and RFE. In each round of RFE, the SVM linear classifier was trained by the pooled samples from both AA and EA, including all the healthy controls and sarcoidosis patients. The gene signature that was comprised of the smallest number of genes with significant peak prediction accuracy was used in subsequent analyses. To test the performance of our gene signature, 1,000 times of five-fold cross-validation was conducted using SVM. In addition, the gene signature was also tested for classification accuracy in AA and EA samples, separately. We also used two independent sarcoidosis datasets using different microarray platforms [27], [28] to validate our gene signature.

Supporting Information

Figure S1.

Distribution of the classification accuracy in each RFE step. X-axis: the number of genes in each step; Y-axis: the classification accuracy from a five-fold cross-validation (repeated 1,000 times). The red line shows the average accuracy for each RFE step.

doi:10.1371/journal.pone.0044818.s001

(PDF)

Figure S2.

Distribution of classification accuracies of the 20-gene signature. X-axis: the classification accuracy from a five-fold cross-validation (repeated 1,000 times). The dashed lines indicate the average classification accuracy. (A) All sarcoidosis patients versus healthy controls in the AA samples; (B) Patients with complicated sarcoidosis versus patients with uncomplicated sarcoidosis in the AA samples; (C) All sarcoidosis patients versus healthy controls in the EA samples; and (D) Patients with complicated sarcoidosis versus patients with uncomplicated sarcoidosis in the EA samples.

doi:10.1371/journal.pone.0044818.s002

(PDF)

Figure S3.

Comparison between the 20-gene signature and the TCR/JS/CCR signaling pathway gene signature in individual populations. The distribution of accuracy is based on 1,000 times of five-fold cross-validation. The dashed lines indicate the average classification accuracy for the 20-gene signature or the TCR/JS/CCR signaling pathway gene signature. HC: healthy controls; US: patients with uncomplicated sarcoidosis; and CS: patients with complicated sarcoidosis.

doi:10.1371/journal.pone.0044818.s003

(PDF)

Figure S4.

Capability of the the 20-gene signature and the TCR/JS/CCR signaling pathway gene signature in separating sarcoidosis patients from IPF patients. The distribution of accuracy is based on 1,000 times of five-fold cross-validation. The dashed lines indicate the average classification accuracy for the 20-gene signature or the TCR/JS/CCR signaling pathway gene signature.

doi:10.1371/journal.pone.0044818.s004

(PDF)

Table S1.

SNPs significantly associated with sarcoidosis within the 31 TCR/JS/CRR signature genes (P<0.01).

doi:10.1371/journal.pone.0044818.s005

(PDF)

Table S2.

PubMatrix search results for the TCR/JS/CCR signature genes against sarcoidosis-related search terms.

doi:10.1371/journal.pone.0044818.s006

(PDF)

Table S3.

Patient therapy description.

doi:10.1371/journal.pone.0044818.s007

(PDF)

Text S1.

Supplementary methods.

doi:10.1371/journal.pone.0044818.s008

(PDF)

Acknowledgments

The authors would like to thank Dr. Steven M. Dudek and Dr. Jeffrey R. Jacobson for valuable discussions. The authors also would like to thank Dr. Eun A Ko for figure editing. This work was also benefited from the Johns Hopkins Sarcoidosis gene bank/database and the Hospital for the Consumptives of Maryland (Eudowood).

Author Contributions

Conceived and designed the experiments: TZ WZ NJS DRM KSK IN RFM JGNG. Performed the experiments: TZ WZ SFM MW. Analyzed the data: TZ WZ. Contributed reagents/materials/analysis tools: TZ WZ NJS ESC DRM KSK SFM MW IN. Wrote the paper: TZ WZ RFM JGNG.

References

  1. 1. Iannuzzi MC, Rybicki BA, Teirstein AS (2007) Sarcoidosis. N Engl J Med 357: 2153–2165. doi: 10.1056/nejmra071714
  2. 2. Newman LS, Rose CS, Maier LA (1997) Sarcoidosis. N Engl J Med 336: 1224–1234. doi: 10.1056/nejm199704243361706
  3. 3. Nunes H, Freynet O, Naggara N, Soussan M, Weinman P, et al. (2010) Cardiac sarcoidosis. Semin Respir Crit Care Med 31: 428–441. doi: 10.1055/s-0030-1262211
  4. 4. Zajicek JP, Scolding NJ, Foster O, Rovaris M, Evanson J, et al. (1999) Central nervous system sarcoidosis–diagnosis and management. Qjm 92: 103–117. doi: 10.1093/qjmed/92.2.103
  5. 5. Morgenthau AS, Iannuzzi MC (2011) Recent advances in sarcoidosis. Chest 139: 174–182. doi: 10.1378/chest.10-0188
  6. 6. Iannuzzi MC, Rybicki BA (2007) Genetics of sarcoidosis: candidate genes and genome scans. Proc Am Thorac Soc 4: 108–116. doi: 10.1513/pats.200607-141jg
  7. 7. Brewerton DA, Cockburn C, James DC, James DG, Neville E (1977) HLA antigens in sarcoidosis. Clin Exp Immunol 27: 227–229.
  8. 8. Grunewald J, Eklund A, Olerup O (2004) Human leukocyte antigen class I alleles and the disease course in sarcoidosis patients. Am J Respir Crit Care Med 169: 696–702. doi: 10.1164/rccm.200303-459oc
  9. 9. Rossman MD, Thompson B, Frederick M, Maliarik M, Iannuzzi MC, et al. (2003) HLA-DRB1*1101: a significant risk factor for sarcoidosis in blacks and whites. Am J Hum Genet 73: 720–735. doi: 10.1086/378097
  10. 10. Maliarik MJ, Chen KM, Sheffer RG, Rybicki BA, Major ML, et al. (2000) The natural resistance-associated macrophage protein gene in African Americans with sarcoidosis. Am J Respir Cell Mol Biol 22: 672–675. doi: 10.1165/ajrcmb.22.6.3745
  11. 11. Hutyrova B, Pantelidis P, Drabek J, Zurkova M, Kolek V, et al. (2002) Interleukin-1 gene cluster polymorphisms in sarcoidosis and idiopathic pulmonary fibrosis. Am J Respir Crit Care Med 165: 148–151. doi: 10.1164/ajrccm.165.2.2106004
  12. 12. Akahoshi M, Ishihara M, Remus N, Uno K, Miyake K, et al. (2004) Association between IFNA genotype and the risk of sarcoidosis. Hum Genet 114: 503–509. doi: 10.1007/s00439-004-1099-5
  13. 13. Kelly DM, Greene CM, Meachery G, O’Mahony M, Gallagher PM, et al. (2005) Endotoxin up-regulates interleukin-18: potential role for gram-negative colonization in sarcoidosis. Am J Respir Crit Care Med 172: 1299–1307. doi: 10.1164/rccm.200411-1594oc
  14. 14. Bombieri C, Luisetti M, Belpinati F, Zuliani E, Beretta A, et al. (2000) Increased frequency of CFTR gene mutations in sarcoidosis: a case/control association study. Eur J Hum Genet 8: 717–720. doi: 10.1038/sj.ejhg.5200524
  15. 15. Schurmann M, Albrecht M, Schwinger E, Stuhrmann M (2002) CFTR gene mutations in sarcoidosis. Eur J Hum Genet 10: 729–732. doi: 10.1038/sj.ejhg.5200868
  16. 16. Kruit A, Grutters JC, Ruven HJ, van Moorsel CH, Weiskirchen R, et al. (2006) Transforming growth factor-beta gene polymorphisms in sarcoidosis patients with and without fibrosis. Chest 129: 1584–1591. doi: 10.1378/chest.129.6.1584
  17. 17. Pabst S, Baumgarten G, Stremmel A, Lennarz M, Knufermann P, et al. (2006) Toll-like receptor (TLR) 4 polymorphisms are associated with a chronic course of sarcoidosis. Clin Exp Immunol 143: 420–426. doi: 10.1111/j.1365-2249.2006.03008.x
  18. 18. McGrath DS, Foley PJ, Petrek M, Izakovicova-Holla L, Kolek V, et al. (2001) Ace gene I/D polymorphism and sarcoidosis pulmonary disease severity. Am J Respir Crit Care Med 164: 197–201. doi: 10.1164/ajrccm.164.2.2011009
  19. 19. Furuya K, Yamaguchi E, Itoh A, Hizawa N, Ohnuma N, et al. (1996) Deletion polymorphism in the angiotensin I converting enzyme (ACE) gene as a genetic risk factor for sarcoidosis. Thorax 51: 777–780. doi: 10.1136/thx.51.8.777
  20. 20. Arbustini E, Grasso M, Leo G, Tinelli C, Fasani R, et al. (1996) Polymorphism of angiotensin-converting enzyme gene in sarcoidosis. Am J Respir Crit Care Med 153: 851–854. doi: 10.1164/ajrccm.153.2.8564143
  21. 21. Valentonyte R, Hampe J, Croucher PJ, Muller-Quernheim J, Schwinger E, et al. (2005) Study of C-C chemokine receptor 2 alleles in sarcoidosis, with emphasis on family-based analysis. Am J Respir Crit Care Med 171: 1136–1141. doi: 10.1164/rccm.200405-658oc
  22. 22. Spagnolo P, Renzoni EA, Wells AU, Sato H, Grutters JC, et al. (2003) C-C chemokine receptor 2 and sarcoidosis: association with Lofgren’s syndrome. Am J Respir Crit Care Med 168: 1162–1166. doi: 10.1164/rccm.200303-456oc
  23. 23. Morohashi K, Takada T, Omori K, Suzuki E, Gejyo F (2003) Vascular endothelial growth factor gene polymorphisms in Japanese patients with sarcoidosis. Chest 123: 1520–1526. doi: 10.1378/chest.123.5.1520
  24. 24. Schurmann M, Reichel P, Muller-Myhsok B, Schlaak M, Muller-Quernheim J, et al. (2001) Results from a genome-wide search for predisposing genes in sarcoidosis. Am J Respir Crit Care Med 164: 840–846. doi: 10.1164/ajrccm.164.5.2007056
  25. 25. Valentonyte R, Hampe J, Huse K, Rosenstiel P, Albrecht M, et al. (2005) Sarcoidosis is associated with a truncating splice site mutation in BTNL2. Nat Genet 37: 357–364. doi: 10.1038/ng1519
  26. 26. Grunewald J, Eklund A (2007) Role of CD4+ T cells in sarcoidosis. Proc Am Thorac Soc 4: 461–464. doi: 10.1513/pats.200606-130ms
  27. 27. Koth LL, Solberg OD, Peng JC, Bhakta NR, Nguyen CP, et al. (2011) Sarcoidosis blood transcriptome reflects lung inflammation and overlaps with tuberculosis. Am J Respir Crit Care Med 184: 1153–1163. doi: 10.1164/rccm.201106-1143oc
  28. 28. Sharma SM, Choi D, Planck SR, Harrington CA, Austin CR, et al. (2009) Insights in to the pathogenesis of axial spondyloarthropathy based on gene expression profiles. Arthritis Res Ther 11: R168. doi: 10.1186/ar2855
  29. 29. Antoniou KM, Tzouvelekis A, Alexandrakis MG, Sfiridaki K, Tsiligianni I, et al. (2006) Different angiogenic activity in pulmonary sarcoidosis and idiopathic pulmonary fibrosis. Chest 130: 982–988. doi: 10.1378/chest.130.4.982
  30. 30. Duan S, Huang RS, Zhang W, Bleibel WK, Roe CA, et al. (2008) Genetic architecture of transcript-level variation in humans. Am J Hum Genet 82: 1101–1113. doi: 10.1016/j.ajhg.2008.03.006
  31. 31. Morley M, Molony CM, Weber TM, Devlin JL, Ewens KG, et al. (2004) Genetic analysis of genome-wide variation in human gene expression. Nature 430: 743–747. doi: 10.1038/nature02797
  32. 32. Stranger BE, Forrest MS, Dunning M, Ingle CE, Beazley C, et al. (2007) Relative impact of nucleotide and copy number variation on gene expression phenotypes. Science 315: 848–853. doi: 10.1126/science.1136678
  33. 33. Stranger BE, Nica AC, Forrest MS, Dimas A, Bird CP, et al. (2007) Population genomics of human gene expression. Nat Genet 39: 1217–1224. doi: 10.1038/ng2142
  34. 34. Spielman RS, Bastone LA, Burdick JT, Morley M, Ewens WJ, et al. (2007) Common genetic variants account for differences in gene expression among ethnic groups. Nat Genet 39: 226–231. doi: 10.1038/ng1955
  35. 35. Zhang W, Duan S, Kistner EO, Bleibel WK, Huang RS, et al. (2008) Evaluation of genetic variation contributing to differences in gene expression between populations. Am J Hum Genet 82: 631–640. doi: 10.1016/j.ajhg.2007.12.015
  36. 36. Kataria YP, LoBuglio AF, Bromberg PA, Hurtubise PE (1976) Sarcoid lymphocytes: B- and T-cell quantitation. Ann N Y Acad Sci 278: 69–79. doi: 10.1111/j.1749-6632.1976.tb47017.x
  37. 37. Sweiss NJ, Salloum R, Gandhi S, Alegre ML, Sawaqed R, et al. (2010) Significant CD4, CD8, and CD19 lymphopenia in peripheral blood of sarcoidosis patients correlates with severe disease manifestations. PLoS One 5: e9088. doi: 10.1371/journal.pone.0009088
  38. 38. Besner G, Higashiyama S, Klagsbrun M (1990) Isolation and characterization of a macrophage-derived heparin-binding growth factor. Cell Regul 1: 811–819. doi: 10.1091/mbc.1.11.811
  39. 39. Higashiyama S, Abraham JA, Klagsbrun M (1993) Heparin-binding EGF-like growth factor stimulation of smooth muscle cell migration: dependence on interactions with cell surface heparan sulfate. J Cell Biol 122: 933–940. doi: 10.1083/jcb.122.4.933
  40. 40. Wilson SE, He YG, Weng J, Zieske JD, Jester JV, et al. (1994) Effect of epidermal growth factor, hepatocyte growth factor, and keratinocyte growth factor, on proliferation, motility and differentiation of human corneal epithelial cells. Exp Eye Res 59: 665–678. doi: 10.1006/exer.1994.1152
  41. 41. Kiso S, Kawata S, Tamura S, Higashiyama S, Ito N, et al. (1995) Role of heparin-binding epidermal growth factor-like growth factor as a hepatotrophic factor in rat liver regeneration after partial hepatectomy. Hepatology 22: 1584–1590. doi: 10.1002/hep.1840220535
  42. 42. Marikovsky M, Breuing K, Liu PY, Eriksson E, Higashiyama S, et al. (1993) Appearance of heparin-binding EGF-like growth factor in wound fluid as a response to injury. Proc Natl Acad Sci U S A 90: 3889–3893. doi: 10.1073/pnas.90.9.3889
  43. 43. McCarthy DW, Downing MT, Brigstock DR, Luquette MH, Brown KD, et al. (1996) Production of heparin-binding epidermal growth factor-like growth factor (HB-EGF) at sites of thermal injury in pediatric patients. J Invest Dermatol 106: 49–56. doi: 10.1111/1523-1747.ep12327214
  44. 44. Ito N, Kawata S, Tamura S, Kiso S, Tsushima H, et al. (1994) Heparin-binding EGF-like growth factor is a potent mitogen for rat hepatocytes. Biochem Biophys Res Commun 198: 25–31. doi: 10.1006/bbrc.1994.1004
  45. 45. Homma T, Sakai M, Cheng HF, Yasuda T, Coffey RJ Jr, et al. (1995) Induction of heparin-binding epidermal growth factor-like growth factor mRNA in rat kidney after acute injury. J Clin Invest 96: 1018–1025. doi: 10.1172/jci118087
  46. 46. Lockstone HE, Sanderson S, Kulakova N, Baban D, Leonard A, et al. (2010) Gene set analysis of lung samples provides insight into pathogenesis of progressive, fibrotic pulmonary sarcoidosis. Am J Respir Crit Care Med 181: 1367–1375. doi: 10.1164/rccm.200912-1855oc
  47. 47. No authors listed (1999) Statement on sarcoidosis. Joint Statement of the American Thoracic Society (ATS), the European Respiratory Society (ERS) and the World Association of Sarcoidosis and Other Granulomatous Disorders (WASOG) adopted by the ATS Board of Directors and by the ERS Executive Committee, February 1999. Am J Respir Crit Care Med 160: 736–755. doi: 10.1164/ajrccm.165.2.ats01
  48. 48. Tusher VG, Tibshirani R, Chu G (2001) Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 98: 5116–5121. doi: 10.1073/pnas.091062498
  49. 49. R_Development_Core_Team (2005) R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.
  50. 50. Tibshirani R, Hastie T, Narasimhan B, Chu G (2002) Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 99: 6567–6572. doi: 10.1073/pnas.082099299
  51. 51. Kanehisa M, Goto S, Kawashima S, Okuno Y, Hattori M (2004) The KEGG resource for deciphering the genome. Nucleic Acids Res 32: D277–280. doi: 10.1093/nar/gkh063
  52. 52. Huang da W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4: 44–57. doi: 10.1038/nprot.2008.211
  53. 53. Dennis G Jr, Sherman BT, Hosack DA, Yang J, Gao W, et al. (2003) DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol 4: P3. doi: 10.1186/gb-2003-4-5-p3
  54. 54. Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Statist Soc B 57: 289–300.
  55. 55. Vapnik V (1998) Statistical Learning Theory. New York: John Wiley & Sons.
  56. 56. Guyon I, Weston J, Barnhill S (2002) Gene selection for cancer classification using support vector machines. Machine Learning 46: 389–422. doi: 10.1023/a:1012487302797
  57. 57. Thuerigen O, Schneeweiss A, Toedt G, Warnat P, Hahn M, et al. (2006) Gene expression signature predicting pathologic complete response with gemcitabine, epirubicin, and docetaxel in primary breast cancer. J Clin Oncol 24: 1839–1845. doi: 10.1200/jco.2005.04.7019
  58. 58. Zhang X, Lu X, Shi Q, Xu XQ, Leung HC, et al. (2006) Recursive SVM feature selection and sample classification for mass-spectrometry and microarray data. BMC Bioinformatics 7: 197.