Peripheral Blood Gene Expression in Alopecia Areata Reveals Molecular Pathways Distinguishing Heritability, Disease and Severity

ORIGINAL ARTICLE Peripheral blood gene expression in alopecia areata reveals molecular pathways distinguishing heritability, disease and severity

AB Coda, V Qafalijaj Hysa, K Seiffert-Sinha and AA Sinha Center for Investigative Dermatology, Division of Dermatology and Cutaneous Sciences, College of Human Medicine, East Lansing, MI, USA

Alopecia areata (AA) is an autoimmune hair loss disorder in which systemic disturbances have been described, but are poorly understood. To evaluate disease mechanisms, we examined gene expression in the blood of defined clinical subgroups (patchy AA persistent type, AAP, n ¼ 5; alopecia universalis, AU, n ¼ 4) and healthy controls (unaffected relatives, UaR, n ¼ 5; unaffected non-relatives, UaNR, n ¼ 4) using microarrays. Unsupervised hierarchical clustering separates all four patient and control groups, producing three distinct expression patterns reflective of ‘inheritance’, ‘disease’ and ‘severity’ signatures. Functional classification of differentially expressed genes (DEGs) comparing disease (AAP, AU) vs normal (UaR) groups reveals upregulation in immune response, cytokine signaling, signal transduction, cell cycle, proteolysis and cell adhesion- related genes. Pathway analysis further reveals the activation of several genes related to natural killer-cell cytotoxicity, apoptosis, mitogen activated protein kinase, Wnt signaling and B- and T-cell receptor signaling in AA patients. Finally, 35 genes differentially expressed in AA blood overlap with DEGs previously identified in AA skin lesions. Our results implicate innate and adaptive immune processes while also revealing novel pathways, such as Wnt signaling and apoptosis, relevant to AA pathogenesis. Our data suggest that peripheral blood expression profiles of AA patients likely carry new biomarkers associated with disease susceptibility and expression. Genes and Immunity (2010) 11, 531–541; doi:10.1038/gene.2010.32; published online 10 June 2010

Keywords: alopecia areata; alopecia; microarray; pathogenesis; genetics

Introduction highlighting the role of immune dysfunction as a crucial feature in AA pathogenesis. Alopecia areata (AA) is an autoimmune skin disorder The etiopathogenesis of AA is not completely under- typically characterized by the spontaneous onset of stood. The most widely accepted hypothesis describes patchy hair loss of the scalp. Although 80% of patients AA as a T-cell mediated autoimmune response targeting present with a minimal number of patches (AA; patchy an unknown antigen in anagen-stage hair follicles.8,9 persistent, AAP or transient, AAT), 7% of cases progress It has also been postulated that compromise of immune in clinical severity to total scalp-hair loss (alopecia totalis; privilege in normal hair follicles, mediated in part by AT) or total body hair loss (alopecia universalis, AU). interferon-gamma (INF-g),10 has a role in the suscep- The prevalence of AA in the United States is reported to tibility and pathogenesis of AA.11,12 Considerable evi- be 0.1–0.2% in the general population, with a lifetime risk dence indicates that genetic elements have a significant of 1.7%.1,2 The histopathological hallmark of AA is the role in the etiology of AA, as evidenced by the recent presence of a perifollicular and intrafollicular lympho- identification of a number of suggested susceptibility cytic infiltrate. T lymphocytes (predominantly CD4 þ loci,13 the high incidence of positive family history in cells), as well as Langerhans cells, macrophages and affected individuals14,15 and high twin concordance natural killer (NK) cells are found in the dermal papilla rates.16–20 Many epidemiologic studies have shown and matrix of the involved hair follicles.3,4 Lymphocytic a higher prevalence of autoimmune diseases, such as infiltration of the anagen hair bulb and dermal papilla is vitiligo, thyroid disease and collagen vascular disease, in accompanied by aberrant expression of human leukocyte AA patients and unaffected relatives (UaR)21,22 suggest- antigen class I and class II antigens and intercellular ing a genetic predisposition to autoimmunity associated adhesion molecule-1 on the follicular epithelium,5–7 with familial background of AA.23 These findings imply that systemic disturbances contribute to this T-cell Correspondence: Dr AA Sinha, Division of Dermatology and dependent, organ-specific disease. Overall, AA is a Cutaneous Sciences, Michigan State University, 4179 Biomedical complex, polygenic disease likely initiated by the and Physical Sciences Building, East Lansing, MI 48824-1320, USA. E-mail: [email protected] interactions between genetic and environmental factors Received 1 December 2009; revised and accepted 20 April 2010; triggering the immune dysfunction that results in published online 10 June 2010 autoimmune hair loss. Peripheral blood gene expression in alopecia areata AB Coda et al 532 Global gene-expression profiling by DNA microarray Results has become a valuable tool for the study of autoimmune and/or inflammatory diseases.24 Microarray studies on Unsupervised hierarchical clustering separates sample groups organ-specific autoimmune diseases have advanced our by familial inheritance, absence or presence of disease and understanding of autoimmune disease pathogenesis, and clinical subtype revealed targets for clinical biomarkers, prognostic We used Affymetrix U133 Plus 2.0 microarray chips factors and subtype distinction for individualized ther- (Affymetrix, Santa Clara, CA, USA) to generate gene apy.25–27 Our recent microarray analysis in AA patient expression profiles from buffy coat samples obtained skin samples,28 along with a previous microarray study in 5 alopecia areata, patchy persistent (AAP) patients, in murine and human AA skin,29 help to substantiate 4 alopecia universalis (AU) patients, 5 unaffected relatives AA as a Th1-cell-mediated autoimmune disorder with (UaR) and 4 unaffected non-relatives (UaNR) (Table 1). To a late secondary humoral response. No study, however, explore the similarities and differences in global gene has previously used gene microarray technology as a expression patterns among the samples without apriori large-scale screening tool to investigate gene regulatory knowledge of sample identity, we used an unsupervised alterations in the peripheral blood of AA patients. As hierarchical clustering method using those 551 probe sets with many organ-specific diseases, the systemic envi- with the highest variation across samples. This unbiased ronment in AA patients is likely to be critical in the approach was used to determine the extent to which global generation and regulation of the immune response gene expression analysis can be used to distinguish clinically and inflammatory cascade that is ultimately directed to defined subgroups (class prediction), as well as identify novel hair follicles. or unexpected gene expression patterns (class discovery). The aims of the present study were to (1) establish We report that unsupervised hierarchical clustering peripheral gene expression profiles associated with AA reveals distinct separation of AA patients from controls and AA subgroups and (2) examine AA-associated (Figure 1). Upon further evaluation, three unique transcriptional signatures to identify functional path- patterns emerge. First, our data reveal that AA patients ways and genes relevant to AA. On the basis of the and first-degree relatives of AA patients (AAP, AU and global microarray analysis in AA patients and healthy UaR) cluster distinctly from the healthy controls with no control blood, we report three distinct transcriptional familial connection to AA (UaNR), conceivably produ- signatures relevant to (i) the genetic predisposition to cing an ‘inheritance’ signature. Second, within this disease, (ii) the clinical expression of disease and (iii) the ‘inheritance’ cluster, AAP and AU samples further degree of phenotypic severity. Evaluation of each of these separate from UaR controls, with the exception of a signatures implicates several immune-related and single AU (‘AU 4’) sample, to produce a ‘disease’ signaling pathways in AA pathogenesis. signature. Third, within the ‘disease’ cluster, AAP

Table 1 Demographic data for study subjects

Sample code Race Age (years) Gender Disease type Length of disease (years)

Unaffected non-relatives (UaNR) UaNR 1 Caucasian 40 F NA NA UaNR 2 Caucasian 32 F NA NA UaNR 3 Hispanic 33 F NA NA UaNR 4 Asian 37 M NA NA

Unaffected relatives (UaR) UaR 1 Caucasian 27 F UaR NA UaR 2 Caucasian 57 M UaR NA UaR 3 Caucasian 22 M UaR NA UaR 4 African–American 42 F UaR NA UaR 5 Caucasian 48 F UaR NA

Alopecia areata, patchy persistent (AAP) AAP 1 Caucasian 34 F AAP 9 AAP 2 Caucasian 25 F AAP 21 AAP 3 Hispanic 37 F AAP 13 AAP 4 African–American 42 F AAP 1 AAP 5 Caucasian 54 M AAP 29

Alopecia universalis (AU) AU 1 Caucasian 66 F AU 47 AU 2 Caucasian 62 M AU 47 AU 3 Hispanic 31 F AU 26 AU 4 Caucasian 23 M AU 11

Abbreviations: AAP, alopecia areata, patchy persistent; AU, alopecia universalis; F, female; M, male; UaNR, unaffected non-relatives; UaR, unaffected relatives.

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 533 samples split from the AU samples to define a ‘severity’ signature. It should be noted, however, that one AAP patient (AAP 5) did share gene expression characteristics with control samples. It is noteworthy that this patient was the only one in this study who was undergoing treatment with systemic corticosteroids at the time of blood sampling, suggesting treatment-related reversal of some disease-associated gene expression. Finally, we find that AU samples group more closely to UaR samples than did AAP samples, which will be discussed further below. Overall, these results suggest that whole-blood gene expression analysis may distinguish familial association to AA (‘inheritance’ signature), AA from controls (‘disease’ signature), as well as AAP from AU (‘severity’ signature).

The AA ‘inheritance’ signature reveals differential expression of functional pathways relevant to adaptive immunity and signaling pathways, such as mitogen activated protein kinase (MAPK), Wnt and Hedgehog (Hh) signaling The ‘inheritance’ cluster (AAP þAU þ UaR) is composed of 3750 differentially expressed genes (DEGs), including 2470 upregulated transcripts and 1280 downregulated transcripts, which distinguish it from the UaNR group (on the basis of a 1% false-discovery rate and greater than þ /À2.5-fold change cutoff (Table 2)). Functional annotation and pathway analysis generated by the database for annotation, visualization and integrated discovery (DA- VID) database and PubMed literature searches of ‘inheritance’ signature DEGs reveals upregulation of MAPK signaling, Wnt signaling, ubiquitin-mediated proteolysis, chemokine/cytokine signaling and immune receptor signaling pathways (Figure 2). The ‘inheritance’ signature also exhibits the upregulation of several apoptosis-related genes, including TNF alpha, CASP1

Table 2 Numbers of genes differentially expressed in inheritance, disease and severity signatures

Differentially expressed genesa Total Up Down

AAP+AU+UaR vs UaNR (Inheritance) 3750 2470 1280 AAP+AU vs UaR (Disease) 882 764 118 AAP vs AU (Severity) 464 144 320

Abbreviations: AAP, alopecia areata, patchy persistent; AU, alopecia universalis; UaNR, unaffected non-relatives; UaR, unaffected relatives. aThe following filtering criteria were used to generate the differentially expressed genes: (i) Benjamini–Hochberg procedure with 1% false-discovery rate; (ii) greater than a +/À2.5 fold-change in the mean expression values between experimental and control groups.

Figure 1 Unsupervised hierarchical clustering reveals three distinct gene expression patterns. Unsupervised hierarchical clustering of those 551 genes of highest variance across samples from 5 AAP patients, 4 AU patients, 5 UaR controls and 4 UaNR controls is shown. Each row signifies one of the 551 genes, whereas the columns represent samples processed by a Euclidean distance algorithm in dChip. Expression value intensities are illustrated by the color scale with a range of À3–3 on a log2 scale. Sample columns are colored to indicate subset: green ¼ UaNR, blue ¼ UaR, orange ¼ AU and red ¼ AAP.

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 534

Figure 2 ‘Inheritance’ signature functional annotation and pathways. DEGs in AAP, AU and UaR as compared with UaNR, functionally annotated by DAVID. Bars indicate fold change levels for AAP and AU) and UaR vs UaNR.

(caspase-1), DIP (death-inducing protein), interleukin (IL)-1 cluster (AAP and AU) from unaffected relative controls receptor signaling genes (IL1R1, IRAK1, IL1RN) and (UaR), again on the basis of a 1% false-discovery rate modulators of NF Kappa B pathways (IKBKG, NFKBIA). and greater than þ /À2.5-fold change cutoff (Table 2). It has been shown that overexpression of Bcl-2 and Bcl-xL, Functional annotation and pathway analysis of the both upregulated in the ‘inheritance’ signature, leads to ‘disease’ signature DEGs reveals an enrichment of genes premature termination of anagen and catagen acceleration, related to immune system, signal transduction, cell cycle resulting in hair follicle regression and alopecia.30 and apoptosis-related pathways. A number of the genes that are upregulated in the A total of 167 DEGs (152 up, 15 down) are involved in ‘inheritance’ signature have previously been implicated immune system processes, including T- and B-cell in AA pathogenesis (TNF, TRAF1, CTLA4, IL1RN, response, NK-cell cytotoxicity and cytokine/chemokine ICAM1)9,10 or other autoimmune diseases (NKTR, IL-6 signaling (Figure 3). Several of these genes and pathways signal transducer).31,32 These findings reflect an activated have a previously described role in AA pathogenesis, immune system in AA-prone individuals with enhanced validating that our experimental approach is sufficiently inflammatory cytokine expression and the enhanced sensitive to capture disease-relevant processes.10,33 expression of genes promoting trans-endothelial cell Although it has been shown that NK cells aggregate migration. perifollicularly in AA lesional skin3,4,34 and peripheral Conversely, we observe the downregulation of antigen blood,35,36 there is limited data regarding the role of NK processing and presentation related genes, integrin cells in AA pathogenesis. Our data show an upregulation signaling, keratinocyte differentiation and of multiple of NK-cell cytotoxicity in AA peripheral blood that, as metabolic pathways. Hedgehog-signaling components, suggested by others,11,12,37 may promote the failure of including PTCH1, IHH, PKA and FBXW11 (b-TrCP), hair follicle immune privilege. are also underexpressed in the ‘inheritance’ cluster We also documented increased levels of Th17 path- compared with UaNR. way-related genes, including IL-23 alpha subunit p19 See Supplemental Table 1 for full list of DEGs within (FC ¼ 2.6), IL-17 receptor A (FC ¼ 2.6), interleukin 12 the ‘inheritance’ gene signature with chromosomal receptor beta 1 (FC ¼ 6.7), and CCR2 (FC ¼ 3.5) in the locations. disease signature. Th17, a newly described IL-17-producing, CCR2 þ /CCR5À helper T-cell subset, is being The ‘disease’ signature reveals dysregulation of immune increasingly recognized as important in infection clear- system, signal transduction, cell cycle and apoptosis-related ance and autoimmune tissue inflammation. Although we genes did not find the IL-17A/F gene itself to be upregulated, To evaluate alterations in gene expression that are our data nonetheless provide support for enhanced Th17 specific to AA patients we eliminated UaNR subjects pathway activity in AA blood. from our next analysis, thus enriching our gene list for We observe the dysregulation of 28 apoptosis-related transcriptional changes directly associated with clinical genes that may (1) contribute directly to hair follicle disease. We find 882 DEGs distinguish the ‘disease’ degeneration and/or catagen induction and (2) amplify

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 535

Figure 3 ‘Disease’ signature functional annotation and pathways. DEGs in AAP and AU, functionally annotated by DAVID. Bars indicate fold change levels for AA patients (AAP, AU) vs UaR. the apoptotic debris that generates antigens capable of Wnt/b-catenin signaling. Wnt/b-catenin signaling is stimulating the immune system. Apoptotic processes essential for hair follicle morphogenesis, the onset of also have a role in modeling the architecture of the hair cycling and regeneration.41,42 developing follicle epithelium during morphogenesis, See Supplemental Table 2 for full list of DEGs that and have a major role in the control of the hair growth comprise the ‘disease’ gene signature. cycle.38,39 These findings correlate with the pervasive shift of hair follicle cycling from anagen to catagen The AA ‘severity’ signature reveals a more robust expression characteristic of involved follicles and the presence of of immune response-related genes in AAP when compared pro-apoptotic programs we observed in AA lesional with AU skin.28 To uncover the factors mediating progression of clinical Several unanticipated ontologies are significantly upregu- severity and disease extent, we compared the genes and lated in the ‘disease’ signature, including genes involved pathways differentially expressed between AAP and in lipid metabolism, innate immunity and G-protein AU. AAP samples expressed 144 transcripts that are signal transduction, with prominence of Rho-GTPase upregulated and 320 transcripts that are downregulated signaling (Figure 3). Rho signaling has been identi- relative to AU samples (Table 2). Functional annotations fied as essential to many cellular functions, including reveal that AAP patients express higher levels of genes immune system development and cell cycle control,40 involved in signal transduction (34 genes), general although its potential role in AA pathogenesis requires immune function (42 genes), cytokine/chemokine sig- further study. Pathway analysis by DAVID database also naling (19 genes) and cell adhesion/motility (18 genes) identifies 28 upregulated genes within the ‘disease’ (Figure 4). We found relative upregulation of multiple signature related to MAPK signaling and 17 genes in cell adhesion molecules, including CXCR1, CXCL2,

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 536

Figure 4 ‘Severity’ signature functional annotation and pathways. DEGs between AAP and AU, functionally annotated by DAVID. Bars indicate fold change levels for AU patients vs AAP patients. A ‘negative’ fold change indicates higher relative expression in AAP patients.

L-selectin, P-selectin and cutaneous lymphocyte antigen, overlapping genes are involved in T-cell immunity, in AAP compared with AU. Cutaneous lymphocyte cell adhesion/leukocyte migration, stress response and antigen, the major E-selectin ligand on human skin- protein metabolism and modification. These data may homing T cells, shows the highest upregulation, implying reflect that peripheral blood contains skin-homing a more active T-cell mediated immune reaction is lymphocytes, and their precursors, which exhibit operative in clinically ‘less severe’ AAP patients com- changes in gene expression relevant to AA pathology pared with individuals with clinically ‘more severe’ AU (Figure 5). As a result, AA blood DEGs within the that is measurable in peripheral blood. ‘disease’ and ‘inheritance’ signatures overlapping with Pathway analysis reveals dysregulation in transform- skin DEGs may represent good candidate markers ing growth factor-beta signaling (7 genes), apoptosis (15 for AA diagnosis and susceptibility, pending larger genes) and Wnt signaling (7 genes) in AAP relative to confirmatory studies and advances in proteomic AA AU. Significant upregulation of TGF-beta signaling in studies. AAP is shown by increased expression of TGF-b2, TSP-1, LTBP-1 and ACVR1 compared with AU. Previous studies have shown TGF-b to be localized in the outermost layer of outer root-sheath cells and bulge Discussion region in normal hair follicles and, as noted above, this The study of complex diseases must confront two major molecule has been implicated in the maintenance of confounding factors: (1) the interplay of genetic immune privilege.12,43 Thus, decreased levels of TGF-b2 and environmental factors at work and (2) broad clinical in AU relative to AAP may indicate a more extensive loss heterogeneity between patients. In this study, we used of hair follicle immune privilege in AU. global transcriptional analysis of peripheral blood See Supplemental Table 3 for full list of DEGs that by microarray to (a) determine molecular classification constitute the ‘severity-specific’ gene signature. of disease and disease subtypes and (b) offer a comprehensive view of system-wide functional pathway Comparison of differential gene expression in AA blood vs that disturbances relevant to disease susceptibility, clinical in skin reveals a set of shared genes and pathways expression and severity in the largest known microarray A total of 35 dysregulated genes in the AA blood study on AA to date. As with many organ-specific ‘disease’ signature correspond to those genes found diseases, the overlying systemic environment is likely to be dysregulated in AA lesional skin in our own to be critical in the generation and regulation of the studies28 and in others29 (Table 3). Several of these immune response that is directed to hair follicles,

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 537 Table 3 Differentially regulated genes found in AA skin and blood microarray analyses

Gene name Entrez gene Blood Skin

Corresponding expressiona ARID1A: AT rich interactive domain 1A 8289 mm CD1C: CD1c molecule 911 mm FAM65B: family with sequence similarity 65, member B 9750 mm FCGR2B: Fc fragment of IgG, low affinity IIb, receptor (CD32) 2213 mm GPR171: G protein-coupled receptor 171 29909 mm HCLS1: hematopoietic cell-specific Lyn substrate 1 3059 mm IL1R1: interleukin 1 receptor, type I 3554 mm MDS1: ecotropic viral integration site 1 2122 mm PTPRJ: protein tyrosine phosphatase, receptor type, j 5795 mm QDPR: quinoid dihydropteridine reductase 5860 mm RAC2: ras-related C3 botulinum toxin substrate 2 5880 mm TBXAS1: thromboxane A synthase 1 6916 mm VAV1: VAV 1 oncogene 7409 mm BAGE: B melanoma antigen 574 kk CDH12: cadherin 12, type 2 (N-Cadherin 2) 1010 kk EZR: villin 2 (EZRIN) 7430 kk FRMPD1: ferm and PDZ domain containing 1 22844 kk HNF1A: transcription factor 1, hepatic 6927 kk LY6G6D: lymphocyte antigen 6 complex, locus G6D 58530 kk REEP2: receptor accessory protein 2 51308 kk RPS8: ribosomal protein S8 6202 kk TACSTD2: tumor-associated calcium signal transducer 2 4070 kk VDR: vitamin D (1,25- dihydroxyvitamin d3) receptor 7421 kk

Opposite expressionb C18orf10: chromosome 18 open reading frame 10 25941 mk CRIP2: cysteine-rich protein 2 1397 mk DNAL4: dynein, axonemal, light polypeptide 4 10126 mk EPOR: erythropoeitin receptor 2057 mk FIG4: FIG4 homolog, SAC1 lipid phosphatase domain containing 9896 mk MYCL1: v-raf murine sarcoma viral oncogene homolog B1 4610 mk NCAM1: neural cell adhesion molecule 1 (CD56 Ag) 4684 mk NCKIPSD: NCK interacting protein with SH3 domain 51517 mk SIGLEC7: sialic acid binding Ig-like lectin 7 27036 mk TCRA@: T cell receptor alpha locus 6955 mk FKBP5: FK506 binding protein 5 2289 km SEPP1: selenoprotein P, plasma, 1 6414 km aThese genes include those which show corresponding differential expression in AA skin and AA blood as compared with normal skin/ blood. bThese genes show opposite differential expression between AA skin and AA blood.

Unsupervised hierarchical clustering of microarray data generated from peripheral blood reveals three distinct gene signatures in our analysis, providing a means for molecular classification through class prediction (assigning samples to known categories of disease) and class discovery (identifying new classes of samples). Clustering data exhibit class prediction by (1) clearly distinguishing AA patients from controls (‘disease’ signature)—a comparison with relevance for diagnosis Figure 5 Overview of major AA dysregulated pathways. Pathways and identifying possible biomarkers for disease diag- are grouped by dysregulated genes found within a signature. nosis and (2) distinguishing clinically severe (AU) from ‘Inheritance’ signature is defined as upregulated pathways in AAP, milder (AAP) forms of disease (‘severity’ signature)—a AU and UaR as compared with UaNR. ‘Disease’ signature is comparison relevant for understanding clinical hetero- defined as upregulated pathways in AAP and AU vs UaR. ‘Severity’ geneity and disease prognosis. signature is defined as upregulated pathways in AAP as compared with AU. *Denotes dysregulated pathway in AA skin. Furthermore, unsupervised clustering reveals an unexpected ‘inheritance’ signature (class discovery) that seems to separate AA patients and individuals who are related to AA patients (and may be at risk of developing and therefore relevant to pathogenic events. Ultimately, AA) from healthy unrelated individuals. The illumina- these studies intend to facilitate the search for bio- tion of an ‘inheritance’ signature supports the premise of markers relevant to the diagnosis, prognosis and a genetic component underlying the etiopathogenesis of treatment of AA. AA and suggests that peripheral blood gene expression

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 538 may contain the underpinnings of susceptibility to AA. catenin signaling through UDP-mediated b-catenin Previous studies have suggested a polygenic ‘inheri- degradation.47 Absence of b-catenin has been shown to tance’ model is operative in AA9,23,44 but this is the first impair anagen initiation and the ability of stem cells to study revealing that transcriptional profiling of periph- drive differentiation of hair keratinocytes, instead lead- eral blood may reflect distinct patterns of heritability in ing to epidermal differentiation.41 These findings may be AA. Larger scale studies involving immunohistopatho- associated with the decreased expression in AA lesional logy and protein levels will be needed to validate the skin samples of WNT10B, which itself stimulates existence of an ‘inheritance’ pattern with the potential to canonical Wnt signaling,48 and BMP2, important in identify at-risk individuals. coordination of follicular cycling.28,49 Furthermore, Functional classification and pathway analysis of downregulation of Hh components, such as PTCH1 DEGs within the three AA-associated transcriptional and b-TrCP, as seen in our ‘inheritance’ signature, may signatures highlight genes and pathways disturbances confer an impaired ability to regenerate hair and with relevance to genetic susceptibility, disease induction promote anagen hair growth. Inhibition of Hh signaling and maintenance and disease severity and progression. components in adult skin also has been previously Multiple dysregulated pathways (MAPK, Wnt, Polyubi- shown to prevent anagen hair growth and cycling.50–52 quitination and G-protein signaling) overlap all three Thus, individuals with a personal or family history of signatures, whereas disturbances in NK-cell immunity AA appear to harbor an inherent dysfunction in hair and apoptosis-related pathways exclusively overlap the regeneration that may contribute to disease susceptibil- ‘inheritance’ and ‘disease’ signatures. Some pathways ity. Coupled with previous work showing that Wnt are uniquely altered in the ‘inheritance’ (Hh signaling), signaling can promote wound healing and hair follicle disease (innate immunity) and severity (cell adhesion regeneration,53 our data suggest that an arrested folli- and TGF-beta signaling) signatures. Finally, the dys- cular cycling and impaired follicular regeneration regulation of T-cell immunity, Wnt signaling, apoptosis through impaired Hh and Wnt signaling may have a and cell adhesion-related pathways found in transcrip- role in the pathogenesis and severity of AA, and should tional analyses of both AA patient blood and lesional be a focus of future studies. skin suggest direct pathogenic links between systemic Unique to the ‘disease’ signature is the upregulation of disturbances and peripheral pathology. genes involved in the innate immune response. The Four major pathway disturbances (MAPK, G-protein upregulation of genes involved in innate immunity may signaling, Wnt signaling and polyubiquitination) inter- suggest that AA patients have an inappropriate immune sect all three comparisons, suggesting that these and/or inflammatory response to normally innocuous pathways may be relevant to many aspects of AA, stimuli leading to the initiation of disease. As previously including susceptibility, induction and severity progres- suggested, AA is likely the result of polygenic alterations sion. Upregulation of p38-MAP signaling components coupled with undetermined environmental stimuli that (MAP2K3 and MAPK11) and Rac2 in the ‘inheritance’ trigger AA in susceptible individuals. The prominent and ‘disease’ signatures indicate active Th1-cell differ- role of innate immunity has been previously implicated entiation and enhanced production of IFN-g,45 support- in the pathogenesis of autoimmune diseases,54,55 how- ing the importance of an active Th1-mediated immune ever its role in AA remains poorly understood. response in AA. Rac2, essential in T-cell and integrin Our data reveal several genes and pathways in activation and development of Th1 cells,46 also has a role peripheral blood that may be particularly relevant to in amplifying the production of IFN-g at the effector disease severity and response to treatment. For example, phase of the T-cell response.44 This amplification of IFN-g a specific type 1 interferon (IFNA13) is expressed highly may contribute to the underlying mechanisms related to in AU samples. This correlates with upregulation of loss of hair follicle immune privilege. IFN-a inducible genes, IFI-15 and IRF7, in AA lesional NK cells have recently gained momentum as contri- skin compared with non-lesional skin.28 Several case butors to the collapse of immune privilege, and thus reports have described severe AA/AU as a side AA pathogenesis. Although the ‘inheritance’ signature effect of interferon-a therapy for various diseases.56–58 shows upregulation of intracellular components of Furthermore, IFN-a has been shown to promote NK-cell cytotoxicity (ZAP70, LAT, LFA-ICAM1), the NK-cell cytotoxicity and promote autoimmune disease,59 ‘disease’ signature exhibits increased expression of thus it is plausible that IFN-a may function as a systemic perforin and NK-cell activating receptors (NKG2E, amplifier in terms of disease extent and severity of AA. CD244/2B4 and FCGR3A). Alternatively, several NK- We also report that expression of FK506-binding proteins cell activating receptors (CD94, NKG2D and NKG2E) (FKBP), such as FKBP4 and FKBP15, are downregulated were found to be downregulated in UaR samples when in AU compared with AAP. Tacrolimus (FK506) tropical compared with UaNR (data not shown). These findings treatment has been shown to be efficacious in some suggest the presence of hyperresponsive NK cells in AA patients with AA,60 and is known to form a FK506–FKBP patient blood, whereas the suppression of NK-cell complex that binds to and blocks calcineurin. Several activity may have a protective role in susceptible, but case reports have shown treatment failure of tacrolimus disease-free, relatives of AA patients (UaR). Future work in severe AA/AU patients.61–63 It may be that inade- assessing gene expression levels of multiplex AA quate expression of FKBPs in a subset of AU patients families might reveal diagnostic or prognostic markers contributes to therapeutic resistance to tacrolimus, from this pathway for clinical use. revealing a potentially useful marker for therapeutic In the ‘disease’ signature, we found upregulation of efficacy. Nevertheless, the limited number of patients GSK3b, p53, NLK, CtBP2 and components of the non- within this study limits its direct translation to the canonical Wnt/calcium pathway (CaMK2 and PPP3CA), bedside. Although several of these pathways and all of which are known to inhibit canonical Wnt/b- findings may be relevant to disease susceptibility,

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 539 pathogenesis and severity, further validation with larger included 18 subjects in this study (Table 1): 9 subjects cohorts will be necessary to confirm mechanistic path- with AA (5 AAP and 4 AU), 5 UaR and 4 UaNR. Apart ways and clinically relevant biomarkers. from one patient who was treated with systemic We also show that despite representing a clinically corticosteroids (‘AAP 5’), all other AA subjects were ‘less severe’ phenotype, AAP shows evidence of a more untreated at the time of sample collection. Written active immune response than AU, based on both informed consent was obtained from all study subjects unsupervised clustering data and functional pathway before venous blood draws. Buffy coat samples were analyses. AU samples group more closely to UaR isolated from fresh whole blood at the time of collection, samples than AAP samples, with one AU sample even stored at À80 1C and subsequently used for RNA clustering within the UaR group. One explanation for extraction. Information regarding demographic data, this potentially paradoxical finding may be that an AA disease manifestation, current medication and co- absence of hair follicles in AU patients essentially morbid autoimmune diseases was obtained at the time of removes the immune response stimulus that may be blood collection. present in the clinically ‘less severe’ AAP. Alternatively, chronic inflammation and subsequent exhaustion of the RNA extraction immune response in AA may also help to explain our Total RNA was isolated from buffy coat samples using microarray results. TRIzol reagent (Invitrogen, San Diego, CA, USA) per Overall, the data presented here are a key initial step manufacturer’s protocol. The aqueous phase was iso- for identifying diagnostic and prognostic markers of lated and treated with DNase (Qiagen, Valencia, CA, disease, establishing a molecular classification of defined USA). RNA was then purified using the RNeasy Mini kit AA clinical subtypes, and providing insights into disease (Qiagen). After purifying steps, RNA was eluted with mechanisms. The substantial number of DEGs between RNase-free water. The purity and concentration of the patients and controls underscores the clinical hetero- final RNA product was analyzed by Nanodrop 100 geneity and multifactorial nature that characterizes AA, spectrophotometry (Thermo Scientific, Waltham, MA, whereas the distinct identification of various subgroups USA). RNA integrity was assessed by using an RNA Pico through unsupervised hierarchical clustering highlights Chip with an Agilent 2100 Bioanalyzer (Agilent Tech- the utility of microarray analysis for disease classification nologies Inc., Palo Alto, CA, USA). Purified RNA was and illuminating disease mechanisms. The clinical and stored at À80 1C for up to 30 days until proceeding with pathogenetic significance of any of the newly identified the amplification step. markers remains unclear, and larger scale studies will be required to confirm key genes and mechanistic pathways cDNA production, amplification, fragmentation and labeling identified in our study. In future work, it will be of cDNA important to evaluate AA patients stratified by a Purified RNA was used to produce first- and second- comprehensive list of constant (for example, age of strand cDNA using the WT-Ovation FFPE RNA Ampli- onset, human leukocyte antigen types or positive family fication system V2 (NuGEN, San Carlos, CA, USA) by history of AA) and variable (for example, disease activity PCR. Double-stranded cDNA was amplified according to or response to therapy) clinical parameters that are likely manufacturer’s instructions and purified by DNA Clean to affect gene expression profiles and provide clues to & Concentrator-25 spin columns (Zymo Research, explain disease heterogeneity. The elucidation of increas- Orange, CA, USA). Only cDNA of high purity (260/280 ingly specific transcriptional profiles, or ‘fingerprints’, ratios: 1.75 –2.1) was used for this study. Fragmenting assigned to clearly defined disease subgroups can be and labeling of cDNA was performed using FL-Ovation expected to facilitate the discovery of biomarkers cDNA Biotin Module V2 (NuGEN). To assure the success relevant to disease classification, prognosis and response of fragmentation and size distribution before proceeding to therapy for individualization of treatment. with microarray hybridization, cDNA samples were bioanalyzed before and after fragmentation and labeling using an RNA 6000 NanoChip on the Agilent Bioanalyzer Materials and methods (Agilent Technologies Inc., Palo Alto, CA, USA). Patient recruitment Microarray analysis The study was approved by the Institutional Review Labeled cDNA samples were hybridized to Affymetrix Boards of Weil Medical College of Cornell University/ HG-U133 Plus 2.0 chips (Affymetrix), and then stained New York Hospital (Institutional Review Board No. on the Affymetrix Fluidics Station 750 and scanned by 1098–429) and Michigan State University (Institutional the Affymetrix GeneChip Scanner 3000. The U133 Plus Review Board No. 05–1026). Study subjects were 2.0 chip contains more than 54000 probe sets represent- recruited through the National AA Foundation (NAAF) ing more than 33 000 well-characterized genes and and the dermatology outpatient clinics at New York Unigenes. The Affymetrix microarray suite 5.0 was used Hospital and Michigan State University. Patients were to obtain gene expression signal values for each gene. classified according to National Alopecia Areata Foun- Each chip was scaled to an average intensity of 500 to dation clinical assessment guidelines.64,65 Patients with adjust for minor variations in the overall intensity of less than 100% scalp hair loss for more than or equal to 1 hybridization. dChip (www.dchip.org) was then used for year were classified as AA, patchy persistent (AAP). probe-(normalization and modeling) and high-level Patients with 100% scalp and body hair loss were (hierarchical clustering and sample comparison) analy- classified as AU. AAP patients in this study had not sis. Expression values were log2 transformed and then progressed to AT (100% scalp hair loss and varying used to perform an unsupervised hierarchical clustering degrees of body hair loss) or AU before study. We method using those 551 probe sets with the highest

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 540 variation across samples. To compare samples and 10 Gilhar A, Kalish RS. Alopecia areata: a tissue specific generate a list of DEGs, the following filtering criteria autoimmune disease of the hair follicle. Autoimmun Rev were used: (i) Benjamini–Hochberg procedure with a 1% 2006; 5: 64–69. false-discovery rate; (ii) greater than a þ /À2.5-fold 11 Paus R, Ito N, Takigawa M, Ito T. The hair follicle and immune change in the mean expression values between the privilege. J Investig Dermatol Symp Proc 2003; 8: 188–194. experimental and control groups. Redundant probe sets 12 Albers GW, Amarenco P, Easton JD, Sacco RL, Teal P. for a single gene were then eliminated. Next, we used Antithrombotic and thrombolytic therapy for ischemic gene ontology terms to functionally annotate DEGs, and stroke: the Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl): subsequently perform pathway discovery analysis using 483S–512S. the DAVID (http://david.abcc.ncif-crf.gov/) website. 13 Martinez-Mir A, Zlotogorski A, Gordon D, Petukhova L, Mo J, Chromosomal mapping of 100 genes assessed for Gilliam TC et al. Genomewide scan for linkage reveals significance by generating P-values for all stretches evidence of several susceptibility loci for alopecia areata. Am containing p20 DEGs to assess the significance of ‘gene J Hum Genet 2007; 80: 316–328. proximity’. Significant stretches (of p20 genes) were 14 Blaumeiser B, van der Goot I, Fimmers R, Hanneken S, filtered at Po0.01. Ritzmann S, Seymons K et al. Familial aggregation of alopecia areata. J Am Acad Dermatol 2006; 54: 627–632. 15 van der Steen P, Traupe H, Happle R, Boezeman J, Strater R, Hamm H. The genetic risk for alopecia areata in first degree Conflict of interest relatives of severely affected patients. An estimate. Acta Derm Venereol 1992; 72: 373–375. The authors declare no conflict of interest. 16 Scerri L, Pace JL. Identical twins with identical alopecia areata. J Am Acad Dermatol 1992; 27(5 Part 1): 766–767. 17 Jackow C, Puffer N, Hordinsky M, Nelson J, Tarrand J, Duvic M. Alopecia areata and cytomegalovirus infection in Acknowledgements twins: genes versus environment? J Am Acad Dermatol 1998; 38: 418–425. The authors are grateful for resources provided by The 18 Hendren OS. Identical alopecia areata in identical twins. Arch Research Technology Support Facility (RTSF) at Michi- Derm Syphilol 1949; 60(5, Part. 1): 793–795. gan State University and the Genomics Core. We also 19 Lenane P, Pope E, Krafchik B. Congenital alopecia areata. JAm thank Jason R Smith for his technical assistance. We are Acad Dermatol 2005; 52(2 Suppl 1): 8–11. grateful to the National Alopecia Areata Foundation for 20 de Viragh PA, Gianadda B, Levy ML. Congenital alopecia their help in patient recruitment. Finally, we thank the areata. Dermatology 1997; 195: 96–98. study participants without whom this study would not 21 Puavilai S, Puavilai G, Charuwichitratana S, Sakuntabhai A, be possible. Sriprachya-Anunt S. Prevalence of thyroid diseases in patients with alopecia areata. Int J Dermatol 1994; 33: 632–633. 22 Shellow WV, Edwards JE, Koo JY. Profile of alopecia areata: a questionnaire analysis of patient and family. Int J Dermatol References 1992; 31: 186–189. 23 Muller SA, Winkelmann RK. Alopecia areata. An evaluation of 1 Safavi K. Prevalence of alopecia areata in the first national 736 patients. Arch Dermatol 1963; 88: 290–297. health and nutrition examination survey. Arch Dermatol 1992; 24 Kunz M. DNA microarray technology in dermatology. Semin 128: 702. Cutan Med Surg 2008; 27: 16–24. 2 Madani S, Shapiro J. Alopecia areata update. J Am Acad 25 Oertelt S, Selmi C, Invernizzi P, Podda M, Gershwin ME. Dermatol 2000; 42: 549–566; quiz 567-70. Genes and goals: an approach to microarray analysis in 3 Perret C, Wiesner-Menzel L, Happle R. Immunohistochemical autoimmunity. Autoimmun Rev 2005; 4: 414–422. analysis of T-cell subsets in the peribulbar and intrabulbar 26 Orban T, Kis J, Szereday L, Engelmann P, Farkas K, Jalahej H infiltrates of alopecia areata. Acta Derm Venereol 1984; 64: et al. Reduced CD4+ T-cell-specific gene expression in human 26–30. type 1 diabetes mellitus. J Autoimmun 2007; 28: 177–187. 4 Ranki A, Kianto U, Kanerva L, Tolvanen E, Johansson E. 27 Fossey SC, Vnencak-Jones CL, Olsen NJ, Sriram S, Garrison G, Immunohistochemical and electron microscopic characteriza- Deng X et al. Identification of molecular biomarkers for tion of the cellular infiltrate in alopecia (areata, totalis, and multiple sclerosis. J Mol Diagn 2007; 9: 197–204. universalis). J Invest Dermatol 1984; 83: 7–11. 28 Dudda-Subramanya R, Coda AB, Sinha AA. Transcriptional 5 Brocker EB, Echternacht-Happle K, Hamm H, Happle R. profiling in alopecia areata defines immune and cell cycle Abnormal expression of class I and class II major histocom- control related genes within disease specific signatures. patibility antigens in alopecia areata: modulation by topical Genomics (in press). immunotherapy. J Invest Dermatol 1987; 88: 564–568. 29 Carroll JM, McElwee KJ, L EK, Byrne MC, Sundberg JP. Gene 6 Gupta AK, Ellis CN, Cooper KD, Nickoloff BJ, Ho VC, Chan array profiling and immunomodulation studies define a cell- LS et al. Oral cyclosporine for the treatment of alopecia areata. mediated immune response underlying the pathogenesis of A clinical and immunohistochemical analysis. J Am Acad alopecia areata in a mouse model and humans. J Invest Dermatol 1990; 22(2 Part 1): 242–250. Dermatol 2002; 119: 392–402. 7 McDonagh AJ, Snowden JA, Stierle C, Elliott K, Messenger 30 Muller-Rover S, Rossiter H, Lindner G, Peters EM, Kupper TS, AG. HLA and ICAM-1 expression in alopecia areata in vivo Paus R. Hair follicle apoptosis and Bcl-2. J Investig Dermatol and in vitro: the role of cytokines. Br J Dermatol 1993; 129: Symp Proc 1999; 4: 272–277. 250–256. 31 Deng YJ, Huang ZX, Zhou CJ, Wang JW, You Y, Song ZQ et al. 8 McElwee KJ, Tobin DJ, Bystryn JC, King Jr LE, Sundberg JP. Gene profiling involved in immature CD4+ T lymphocyte Alopecia areata: an autoimmune disease? Exp Dermatol 1999; responsible for systemic lupus erythematosus. Mol Immunol 8: 371–379. 2006; 43: 1497–1507. 9 Alexis AF, Dudda-Subramanya R, Sinha AA. Alopecia areata: 32 Ishihara K, Hirano T. IL-6 in autoimmune disease and chronic autoimmune basis of hair loss. Eur J Dermatol 2004; 14: inflammatory proliferative disease. Cytokine Growth Factor Rev 364–370. 2002; 13: 357–368.

Genes and Immunity Peripheral blood gene expression in alopecia areata AB Coda et al 541 33 Benson M, Carlsson L, Guillot G, Jernas M, Langston MA, target of Sonic hedgehog in hair follicle morphogenesis. Mech Rudemo M et al. A network-based analysis of allergen- Dev 2001; 107: 69–82. challenged CD4+ T cells from patients with allergic rhinitis. 49 Plikus MV, Mayer JA, de la Cruz D, Baker RE, Maini PK, Genes Immun 2006; 7: 514–521. Maxson R et al. Cyclic dermal BMP signalling regulates stem cell 34 Imai R, Takamori K, Ogawa H. Changes in populations of activation during hair regeneration. Nature 2008; 451: 340–344. HLA-DR+CD3+ cells and CD57ÀCD16+ cells in alopecia 50 Wang LC, Liu ZY, Gambardella L, Delacour A, Shapiro R, areata after corticosteroid therapy. Dermatology 1994; 188: Yang J et al. Regular articles: conditional disruption of 103–107. hedgehog signaling pathway defines its critical role in hair 35 Imai R, Miura J, Numata K, Takamori K. Increase in natural development and regeneration. J Invest Dermatol 2000; 114: killer cells in the peripheral blood lymphocytes in patients 901–908. with alopecia areata]. Nippon Hifuka Gakkai Zasshi 1988; 98: 51 Yamago G, Takata Y, Furuta I, Urase K, Momoi T, Huh N. 1505–1507. Suppression of hair follicle development inhibits induction of 36 Ito T, Ito N, Saatoff M, Hashizume H, Fukamizu H, Nickoloff sonic hedgehog, patched, and patched-2 in hair germs in mice. BJ et al. Maintenance of hair follicle immune privilege is linked Arch Dermatol Res 2001; 293: 435–441. to prevention of NK cell attack. J Invest Dermatol 2008; 128: 52 Chiang C, Swan RZ, Grachtchouk M, Bolinger M, Litingtung 1196–1206. Y, Robertson EK et al. Essential role for Sonic hedgehog during 37 Albers GW, Amarenco P, Easton JD, Sacco RL, Teal P. hair follicle morphogenesis. Dev Biol 1999; 205: 1–9. Antithrombotic and thrombolytic therapy for ischemic stroke: 53 Fathke C, Wilson L, Shah K, Kim B, Hocking A, Moon R et al. american college of chest physicians evidence-based clinical Wnt signaling induces epithelial differentiation during cuta- practice guidelines (8th Edition). Chest 2008; 133(6 Suppl): neous wound healing. BMC Cell Biol 2006; 7:4. 630S–669S. 54 Holers VM, Boackle SA. Complement receptor 2 and 38 Polakowska RR, Haake AR. Apoptosis: the skin from a new autoimmunity. Curr Dir Autoimmun 2004; 7: 33–48. perspective. Cell Death Differ 1994; 1: 19–31. 55 Boackle SA, Holers VM. Role of complement in the develop- 39 Stenn KS, Paus R. Controls of hair follicle cycling. Physiol Rev ment of autoimmunity. Curr Dir Autoimmun 2003; 6: 154–168. 2001; 81: 449–494. 56 Radny P, Bauer J, Caroli UM, Eigentler TK, Kamin A, Metzler 40 Cantrell DA. GTPases and T cell activation. Immunol Rev 2003; G et al. Alopecia areata induced by adjuvant treatment with 192: 122–130. alpha-interferon in malignant melanoma? Dermatology 2004; 41 Huelsken J, Vogel R, Erdmann B, Cotsarelis G, Birchmeier W. 209: 249–250. beta-Catenin controls hair follicle morphogenesis and stem 57 Kernland KH, Hunziker T. Alopecia areata induced by cell differentiation in the skin. Cell 2001; 105: 533–545. interferon alpha? Dermatology 1999; 198: 418–419. 42 Van Mater D, Kolligs FT, Dlugosz AA, Fearon ER. Transient 58 Kartal ED, Alpat SN, Ozgunes I, Usluer G. Reversible alopecia activation of beta -catenin signaling in cutaneous keratino- universalis secondary to PEG-interferon alpha-2b and ribavirin cytes is sufficient to trigger the active growth phase of the hair combination therapy in a patient with chronic hepatitis C cycle in mice. Genes Dev 2003; 17: 1219–1224. virus infection. Eur J Gastroenterol Hepatol 2007; 19: 817–820. 43 Meyer KC, Klatte JE, Dinh HV, Harries MJ, Reithmayer K, 59 Chakrabarti D, Hultgren B, Stewart TA. IFN-alpha induces Meyer W et al. Evidence that the bulge region is a site of autoimmune T cells through the induction of intracellular relative immune privilege in human hair follicles. Br J adhesion molecule-1 and B7.2. J Immunol 1996; 157: 522–528. Dermatol 2008; 159: 1077–1085. 60 Price VH, Willey A, Chen BK. Topical tacrolimus in alopecia 44 de Andrade M, Jackow CM, Dahm N, Hordinsky M, areata. J Am Acad Dermatol 2005; 52: 138–139. Reveille JD, Duvic M. Alopecia areata in families: association 61 Feldmann KA, Kunte C, Wollenberg A, Wolfe H. Is topical with the HLA locus. J Investig Dermatol Symp Proc 1999; 4: tacrolimus effective in alopecia areata universalis? Br 220–223. J Dermatol 2002; 147: 1031–1032. 45 Rincon M, Flavell RA, Davis RA. The JNK and P38 MAP 62 Thiers BH. Topical tacrolimus: treatment failure in a patient kinase signaling pathways in T cell-mediated immune with alopecia areata. Arch Dermatol 2000; 136: 124. responses. Free Radic Biol Med 2000; 28: 1328–1337. 63 Park SW, Kim JW, Wang HY. Topical tacrolimus (FK506): 46 Yu H, Leitenberg D, Li B, Flavell RA. Deficiency of small treatment failure in four cases of alopecia universalis. Acta GTPase Rac2 affects T cell activation. J Exp Med 2001; 194: Derm Venereol 2002; 82: 387–388. 915–926. 64 Duvic M, Norris D, Christiano A, Hordinsky M, Price V. 47 Aberle H, Bauer A, Stappert J, Kispert A, Kemler R. beta- Alopecia areata registry: an overview. J Investig Dermatol Symp catenin is a target for the ubiquitin-proteasome pathway. Proc 2003; 8: 219–221. EMBO J 1997; 16: 3797–3804. 65 Olsen E, Hordinsky M, McDonald-Hull S, Price V, Roberts J, 48 Reddy S, Andl T, Bagasra A, Lu MM, Epstein DJ, Morrisey EE Shapiro J et al. Alopecia areata investigational assessment et al. Characterization of Wnt gene expression in developing guidelines. National alopecia areata foundation. J Am Acad and postnatal hair follicles and identification of Wnt5a as a Dermatol 1999; 40(2 Part 1): 242–246.

Supplementary Information accompanies the paper on Genes and Immunity website (http://www.nature.com/gene)

Genes and Immunity