Sézary Syndrome Is a Unique Cutaneous T-Cell Lymphoma As

Se´zary syndrome is a unique cutaneous T-cell lymphoma as identiﬁed by an expanded gene signature including diagnostic marker molecules CDO1 and DNM3

N Booken1,11, A Gratchev1,11, J Utikal1, C Wei2,XYu3, M Qadoumi1, M Schmuth4, N Sepp4, D Nashan5, K Rass6,TTu¨ting7, C Assaf8, E Dippel1,9, R Stadler10, C-D Klemke1 and S Goerdt1

1Department of Dermatology, Venereology and Allergology, University Medical Centre Mannheim, Ruprecht Karl University of Heidelberg, Mannheim, Germany; 2Institute of Medical Statistics, University Medical Centre Mannheim, Ruprecht Karl University of Heidelberg, Mannheim, Germany; 3Medical Research Center (ZMF), University Medical Centre Mannheim, Ruprecht Karl University of Heidelberg, Mannheim, Germany; 4Department of Dermatology, Innsbruck Medical University, Innsbruck, Austria; 5Department of Dermatology, University of Freiburg, Freiburg, Germany; 6Department of Dermatology, The Saarland University Hospital, Homburg/Saar, Germany; 7Department of Dermatology, University of Bonn, Bonn, Germany; 8Department of Dermatology, Charite-University Medicine Berlin, Berlin, Germany; 9Department of Dermatology, Academic Medical Centre, Lemgo, Germany and 10Department of Dermatology, Academic Medical Centre, Minden, Germany

Sezary syndrome (SS) is a rare, aggressive CD4 þ cutaneous While MF usually is a slowly progressive disease, SS runs a more T-cell lymphoma (CTCL); molecular traits differentiating SS aggressive course with a high mortality rate and a median survival from nonleukemic mycosis fungoides (MF) and from inflamma- time of 2–4 years. The probability of survival in CTCL can be tory skin diseases (ID) are not sufficiently characterized. accurately predicted by a formula based on the clinical CTCL- Peripheral blood mononuclear cells (PBMC) of 10 SS patients 1 and 10 healthy donors (HD) were screened by Affymetrix severity index (CTCL-SI) that evaluates the involvement of the skin, 2 U133Plus2.0 chips for differential gene expression. Ten candi- lymph nodes, blood and visceral organs. date genes were confirmed by qRT-PCR to be significantly Sezary cells are malignant circulating CD4 þ T cells that þ overexpressed in CD4 T cells of SS versus HD/ID. For easier derive from the same T-cell clone as the malignant T cells in skin clinical use, these genes were re-analyzed in PBMC; qRT-PCR and other organs as demonstrated by T-cell receptor (TCR) gene confirmed five novel (DNM3, IGFL2, CDO1, NEDD4L, KLHDC5) 3,4 and two known genes (PLS3, TNFSF11) to be significantly rearrangement analysis studies. The detection of Sezary cells overexpressed in SS. Multiple logistic regression analysis in the peripheral blood is primarily based on morphological revealed that CDO1 and DNM3 had the highest discriminative features such as cerebriform nuclei; therefore, ‘Sezary cells’ may power in combination. Upon comparison of PBMC and skin be detected in healthy individuals as well as in inflammatory samples of SS versus MF, CDO1 and DNM3 were found dermatoses. As a consequence, an arbitrary threshold has been upregulated only in SS. Using anti-CDO1 antisera, differential set at 5% (20%) Sezary cells among peripheral blood mono- expression of CDO1 protein was confirmed in SS CD4 þ T cells. Interestingly, DNM3 and CDO1 are known to be regulated by nuclear cells (peripheral blood lymphocytes) for the diagnosis of SS-associated transcription factors TWIST1 and c-myb, respec- SS in the blood. Loss of CD26 or CD7, expression of CD158k/ tively. Furthermore, CDO1 catalyzes taurine synthesis and KIR3DL2, analysis of Her2 neu gene copy number, and lack of taurine inhibits apoptosis and promotes chemoprotection. In T regulatory cells may be informative,5,6 but only apply to a summary, CDO1 and DNM3 may improve the diagnosis of SS subpopulation of SS patients. Recently, multi-gene qRT-PCR was and open novel clues to its pathogenesis. proposed to improve the molecular diagnosis of SS cells in the Leukemia (2008) 22, 393–399; doi:10.1038/sj.leu.2405044; peripheral blood, but the method still awaits confirmation by published online 22 November 2007 7 Keywords: Se´zary syndrome; DNA arrays; cysteine dioxygenase; other groups and validation using independent patient samples. dynamin 3; diagnostic markers In SS and MF, chemotherapy does not improve survival, but impairs quality of life; due to frequent recurrences, allogeneic bone marrow transplantation is no successful treatment option either. Stage-adapted therapy as pursued currently by most groups Introduction aims at mitigating the course of the disease, but a potentially curative treatment regimen for SS or MF does not exist, indicating the urgent necessity to identify novel therapeutic targets. Cutaneous T-cell lymphomas (CTCL) are a group of lymphoproli- We here performed Affymetrix microarray-based gene ex- ferative disorders that are characterized by localization of malignant pression profiling of SS peripheral blood mononuclear cells T lymphocytes in the skin, mainly in the epidermis. The classical (PBMC) versus healthy controls followed by qRT-PCR and forms of CTCL are plaque-type mycosis fungoides (MF) and Sezary statistical analysis to identify and validate candidate genes as syndrome (SS). SS is a rare, leukemic variant of CTCL that typically novel diagnostic markers and putative therapeutic targets for the presents with erythroderma, peripheral lymphadenopathy, severe disease. We also assessed the predictive capacity of SS- pruritus and malignant, circulating T lymphocytes, the Sezary cells. associated genes identified by others.

Correspondence: Dr N Booken, Department of Dermatology, Venerolo gy and Allergology, University Medical Center Mannheim, Ruprecht-Karl University Heidelberg, Theodor-Kutzer-Ufer 1-3, Patients, materials and methods D-68167 Mannheim, Germany. E-mail: [email protected] 11These authors contributed equally to this work. Patients Received 18 April 2007; revised 13 October 2007; accepted 24 A total of 41 CTCL patients were included in the study, 27 October 2007; published online 22 November 2007 patients with SS (14 male and 13 female), median age 68 years, Se´zary syndrome gene expression N Booken et al 394 ranging from 48 to 83 years, stage III/IVA and 14 patients with U133 Plus 2.0 Arrays from Affymetrix (ten control samples and MF (10 male and 4 female), median age 59 years, ranging from ten samples from SS patients) were utilized. Raw data from 16 to 95 years, stage IB-IVA. The patients were diagnosed Affymetrix CEL files were analyzed using SAS software package according to the WHO-EORTC classification of cutaneous Microarray Solution version 1.3 (SAS Institute, Cary NC). Gene lymphomas and the criteria of the International Society of annotation was obtained through the Affymetrix NetAffx website Cutaneous Lymphomas. The extent of CTCL involvement was (http://www.affymetrix.com/analysis/index.affx). The quality quantified by the CTCL-SI.2 Analysis of the TCR g or b chain control, normalisation and statistical modeling were performed genes was performed using the well-established polymerase by array group correlation, mixed model normalisation and chain reaction-based GeneScan technique.3 All patients with SS mixed model analysis, respectively. Analysis of differential gene had erythroderma, peripheral lymphadenopathies, a highly expression was based on a loglinear mixed model of perfect elevated CD4/CD8 ratio and atypical circulating Sezary cells matches,8 where group (SS or control) were considered to be in the blood (Supplementary Table 1). For comparative analysis, with constant effects and the chip ID with random effect. blood samples were obtained from 10 healthy donors (4 males and 6 females, median age 57 years, ranging from 34 to 84 years). In addition for qRT-PCR experiments, peripheral blood Real-time RT–PCR analysis samples and lesional skin biopsy specimens of 24 patients with RT–PCR analysis was used to determine lack of gene expression different inflammatory skin diseases (ID) such as psoriasis (16 of candidate genes in CD4 þ T cells of healthy donors and ID patients, 13 male and 3 female, median age 53 years, ranging versus SS patients. Primers used were from Metabion (Martins- from 28 to 76 years), atopic dermatitis (6 patients, 5 male and 1 ried, Germany); primer sequences and the standard RT–PCR female, median age 46 years, ranging from 24 to 78 years) and procedures applied will be supplied by the authors upon erysipelas (2 patients, 1 male and 1 female of 28 and 78 years of request. Real-time PCR analysis was performed using TaqMan age) were obtained. Furthermore 15 additional samples of PCR master mix (Applied Biosystems, Darmstadt, Germany) healthy donors (5 male and 10 female, median age 48 years, together with TaqMan probes and primers (MWG-Biotech, ranging from 25 to 89 years) and peripheral blood T cells Martinsried, Germany) using standard conditions. Each primer activated for 3 days with phytohemagglutinin were also and probe concentration was optimized before use. The included in the study (Supplementary Text 1). Sezary cell lines sequences and concentrations used for quantification of the (SeAx, HH and Hut 78), MF cell line (Myla), leukemia cell lines selected genes are listed in the Supplementary Table 2. Human (Molt, Peer) were used as a calibrators for real-time RT–PCR GAPD was used as an internal control. The experiments were analysis. The study was approved by the Medical Ethics performed on ABI PRISM 7000 sequence detection system Committee II of the University of Heidelberg, Medical Faculty (Applied Biosystems). The expression levels of analyzed genes Mannheim. Informed consent was provided according to the were normalized to GAPD mRNA expression. Each assay Declaration of Helsinki. included a negative control (no template DNA) and all were carried out in quadruplicate (96-well maximum). Relative quantification of the gene expression was performed using Preparation of PBMCs, RNA isolation and cDNA relative quantification module of 7000 system software version synthesis 1.2.3. As a calibrator (expression value taken as 1), one of the Peripheral blood mononuclear cells (PBMCs) were isolated from cell lines or SS2 sample was taken depending on the level of Sezary patients and control individuals using the Ficoll gradient expression (Supplementary Table 2). centrifugation. CD4 þ T cells were selected from PBMC using CD4 þ -Isolation Kit I and CD4 þ -Isolation Kit II for cell culture according to the manufacturer’s instructions (Miltenyi Biotec, Immunuhistochemistry Bergisch Gladbach, Germany). For additional purging of CD4 þ An anti-human CDO1 antiserum was raised in rabbits against T cells from residual monocytes and dendritic cells, monocytes synthetic peptide representing the amino acids 52–69 (gb and dendritic cells were depleted using magnetic labeling with AAB58352) using a standard immunization protocol (PSL reagents provided in Blood Dendritic Cell Isolation Kit II GmbH, Heidelberg, Germany). Pre-immune serum was used (Miltenyi Biotec). Using this procedure, 495% of residual as a negative control. cryosections of liver tissue (5 mm) and monocytes and dendritic cells were removed as confirmed by cytospins of CD4 þ T cells were fixed with acetone and air- FACS analysis. RNA was isolated from whole PBMCs and CD4 þ dried. For immunohistochemical analysis, the CDO1 antiserum T cells using RNeasy Mini or Midi Kit and from tissue sections was used at a dilution of 1:250; HRP-conjugated-anti-rabbit IgG using RNeasy Mini fibrous tissue Kit according to the (Santa Cruz Biotechnology, Heidelberg, Germany) was used as a manufacturer’s instructions (QIAGEN, Hilden, Germany). For secondary antibody at a dilution of 1:200. Detection was cDNA synthesis, total RNA (2 mg) was treated with 2 U Turbo performed using AEC substrate (Dako, Hamburg, Germany). DNase (Ambion, Austin, TX, USA) and used for Reverse Transcription (RT) with Superscript II reverse transcriptase (Invitrogen, Karlsruhe, Germany) using oligo dT primers Statistical analysis of real-time RT–PCR data according to the recommendations of the manufacturer. The The comparison of gene expression between the groups was obtained cDNA was diluted 1:10 with ddH2O and 1 ml was used performed by using the Wilcoxon–Mann–Whitney U-test (for for each PCR reaction. two samples) or the Kruskal–Wallis test (for more than two samples). The test result was considered significant when P-value was less than 0.05. In order to determine overexpression Microarray processing of which genes is specifically related to SS, we further analyzed Microarray hybridization, scanning and analysis was performed the quantitative real-time PCR data for SS patients, patients with by Zentrum fu¨r Medizinische Forschung (Medical Research ID and control samples using the receiver operating character- Center (ZMF), University Medical Centre Mannheim, University istic (ROC) methodology based on logistic regression. Logistic of Heidelberg, Mannheim, Germany). Twenty Human Genome regression analysis is a statistical method for analyzing

Leukemia Se´zary syndrome gene expression N Booken et al 395 dichotomous response while accommodating adjustments for SS patients (Supplementary Table 1, No. 1–10) and 10 healthy one or more explanatory variables.9 Wald’s test has been used blood donors (HBD), labeled and hybridized onto Affymetrix in order to assess if the explanatory variables have a significant HG-U133 Plus 2.0 Genchip. Significant alterations in gene influence on the dichotomous response variable.10 Using this expression were found at a log 10 P-value 46.04 (Bonferroni approach and depending on a chosen cut-off point, it is possible correction) in 294 probe sets. Among these 294 probe sets, the to assess the specific values for sensitivity and for specificity of expression of only 16 genes was increased at least twofold gene expression, related to a special cut-off value. Usually, excluding several genes reported previously to be associated when sensitivity and specificity are regarded as equally with SS such as PLS3 and KIR3DL2.6,11,12 To include a broader important, the optimal cut off is the one in which the sum of range of genes for further investigation, Po0.02 was then set as sensitivity and specificity is maximal. Therefore, in Table 1b the the threshold. Using this setting, we found the expression of 53 values for sensitivity and specificity are given for the optimal cut candidate genes to be increased at least twofold (Supplementary off. Plotting both quality measures for all possible cut-off points Table 3). simultaneously into a graph (sensitivity on the x axis versus 1Fspecificity on the y axis) creates the ROC curve (receiver operator statistics). The area under the ROC curve (AUCFarea Real-time RT–PCR confirmation of candidate genes in þ under the curve) is a measure of the quality of the model, that is, CD4 T cells þ a measure of association of predicted probabilities and observed As the malignant lymphocytes in SS are CD4 T cells, purified þ responses. The nearer the AUC value is close to 1, the better is CD4 T cells were used for confirmation of candidate genes. the diagnostic procedure. Values of about 0.5 symbolize that RT–PCR analysis was performed to determine lack of gene þ this procedure is only as good as random assignments. Hence, expression of candidate genes in CD4 T cells of healthy AUC is suitable to compare several diagnostic tests. In addition, donors and ID versus SS patients. Ten genes were selected for a multivariable analysis including several genes was also validation by qRT-PCR: PLS3 (Plastin 3), DNM3 (Dynamin 3), performed. All statistical calculations were performed using RDH10 (Retinol dehydrogenase 10), IGFL2 (Insulin growth SAS (release 8.02; SAS institute Inc., Cary, NC, USA).9 factor-like family member 2), DUSP4 (dual specificity phospha- tase 4), CDO1 (cysteine dioxygenase, type I), IGFBP4 (Insulin- like growth factor binding protein 4), TNFSF11 (tumor necrosis Results factor superfamily member 11), NEDD4L (neural precursor cell expressed) and KLHDC5 (Kelch domain containing 5). Differential gene expression in leukemic CTCL cells in Samples of purified CD4 þ T cells were obtained from 11 SS SS patients (Supplementary Table 1), 12 healthy blood donors and In order to identify genes differentially expressed in leukemic 10 patients with inflammatory diseases (8 psoriasis and 2 cells of SS patients, total RNA was isolated from the PBMC of 10 atopic dermatitis). Real-time PCR analysis revealed statistically

Table 1 Statistical analysis of quantitative real-time PCR results for gene expression in Sezary syndrome CD4+ T cells in comparison to CD4+ T cells of healthy blood donors (HBD) and inﬂammatory diseases (ID) samples

(a) P-values calculated using the Wilcoxon–Mann–Whitney -U-test Gene SS vs HBD SS vs ID SS vs ID+HBD

TNFSF11 0.0001 0.0001 o0.0001 DNM3 0.0001 0.0002 o0.0001 IGFL2 0.0001 0.0005 o0.0001 CDO1 0.001 0.001 0.0001 NEDD4L 0.0015 0.0011 0.0002 PLS3 0.0012 0.0022 0.0002 IGFBP 4 0.0023 0.0035 0.0005 RDH10 0.0028 0.0043 0.0006 DUSP 4 0.0178 0.0004 0.0006 KLHDC5 0.0074 0.0221 0.0035

(b) AUC values for the discrimination between Sezary syndrome patients and controls calculated using ROC statistic Gene AUCa P-valueb Sensitivity Speciﬁcity Sumc

TNFSF11 0.988 0.028 1.0 0.95 1.95 DNM3 0.983 0.093 0.91 1.0 1.91 IGFL2 0.967 0.035 1.0 0.86 1.86 CDO1 0.911 0.045 0.82 1.0 1.82 NEDD4L 0.909 0.016 0.91 0.95 1.86 PLS3 0.89 0.141 0.82 0.91 1.73 IGFBP 4 0.882 0.092 0.82 0.95 1.77 RDH10 0.876 0.017 0.73 0.91 1.64 DUSP 4 0.872 0.015 0.91 0.82 1.73 KLHDC5 0.818 0.083 0.82 0.95 1.77 AUC, area under curve; ROC, receiver operating characteristic. All genes included were upregulated in Sezary syndrome. aArea under curve, sensitivity and speciﬁcity are related to the optimal cut-off point. bP-value of the Wald’s test. cSum of sensitivity and speciﬁcity.

Leukemia Se´zary syndrome gene expression N Booken et al 396 significant overexpression of all 10 genes in SS CD4 þ T cells in alternative to the analysis of single gene expression in CD4 þ comparison to HBD or ID samples (Table 1a). Further statistical T-cell samples. evaluation of the quantitative data was done using ROC methodology based on logistic regression. Good discrimination capacity was seen for all 10 genes analyzed (Table 1b). Out of Comparison of CDO1/DNM3 expression with the 5 these 10 genes, 5 showed an AUC value (maximum 1) above 0.9 gene qRT-PCR assay proposed by Nebozhyn et al.7 and among these genes, 2 (DNM3 and CDO1) showed a Recently, Nebozhyn et al.7 proposed a qRT-PCR assay analyzing specificity of 100% (Table 1b). To exclude any contribution of STAT4 (kk), CD1D (m), PLS3 (mm), TRAIL (m) and GATA-3 (m)in CD4 þ monocytes and dendritic cells to the results of the PBMC as a highly reliable test for the diagnosis of SS. Real-time analyses of CD4 þ T cells, additional purging procedures using PCR analysis of these five genes in our patient samples as magnetic cell sorting for definitive depletion of monocytes and compared to HBD or ID confirmed the strong reduction in dendritic cells were performed. Real-time PCR analysis revealed expression of STAT4 and the strong overexpression of PLS3 in SS similar or even higher gene expression of the candidate genes in while expression of TRAIL and GATA-3 was not significantly highly purified SS CD4 þ T cells (data not shown). We also altered in contrast to the results of Nebozhyn et al.7 Also in analyzed the expression of factors involved in the biochemical strong contrast to Nebozhyn et al., CD1D was even significantly pathway of taurine and glutathione biosynthesis including downregulated in our SS patient samples (Table 2a). Statistical CSAD (cysteine sulfinic acid decarboxylase), SLC6A6 (taurine evaluation using ROC methodology based on logistic regression transporter) and GCLM (gamma Glutamylcysteine Synthetase) showed an excellent discriminative capacity for STAT4 (Supplementary Table 2); these genes were expressed, but not (AUC ¼ 0.906) and a medium value for CD1D (AUC ¼ 0.796) upregulated (data not shown). (Table 2b). When analyzing expression of our 10 genes and the genes of Nebozhyn et al.7 using the multiple logistic regression model on data generated in our patient and control samples, the Real-time RT–PCR confirmation of candidate genes in combination of DNM3 and STAT4 (AUC ¼ 0.957) and the PBMC combination of DNM3 and CDO1 (AUC ¼ 0.956) had the best Since PBMC samples are easier to obtain from patients and predictive values for the discrimination between SS patients and easier to process for routine analysis than CD4 þ T cells, we control samples. analyzed the expression of the 10 candidate genes also in PBMC samples from patients and controls, although PBMC contain clearly more cellular contaminants beyond the tumor cells Analysis of the expression of CDO1 and DNM3 in tissue (Sezary cells) than CD4 þ T-cell isolates. PBMC samples were samples of SS and MF patients and in activated T cells obtained form 17 additional cases of SS patients (Supplementary As a CTCL, MF is related to, but different from SS. A major Table 1, No. 11–27), 10 healthy blood donors (different from difference between the two diseases is both the negligible tumor those used for microarray analysis), and 10 patients with load as assessed by TCR gene rearrangement analyses and the inflammatory diseases (4 psoriasis, 4 atopic dermatitis and 2 lack of Sezary cells identified morphologically in the peripheral erysipelas). Relative quantification of gene expression revealed blood of MF patients. We therefore predicted that the marker that the observed increase of gene expression in SS PBMCs in molecules of the Sezary cells identified here would rather not be comparison to HBD was statistically significant for all genes expressed in PBMC and skin samples of MF patients. Out of the with the exception of IGFBP4 (Table 2a). In the case of 10 selected genes, the best predictive markers, that is, CDO1 comparison of SS PBMC with ID samples, a significant increase and DNM3, were also tested in PBMC and skin biopsy samples in expression was observed for 7 genes excluding RDH10, from MF patients. qRT-PCR was used to analyze PBMC samples DUSP4 and IGFBP4 (Table 2a). Further statistical evaluation of from MF patients (n ¼ 10) and the results were compared with the quantitative data was again done using ROC methodology the results from the 27 SS patients and the 10 inflammatory skin based on logistic regression. In the PBMC samples, CDO1 and disease patients (see above). For PBMC samples, threshold DNM3 showed the highest AUC values of 0.896 and 0.884, values were set using the results from the ROC analyses. The respectively (Table 2b). The lowest discrimination rates were expression of CDO1 (Figure 1a) and DNM3 (Figure 1b) was only seen in PBMC for the genes RDH10, IGFBP4 and DUSP4, which increased in 1 out of 10 MF PBMC samples similar to control was in accordance with the results of the Wilcoxon–Mann–Whitney samples from HBD and ID while the percentage of positive SS U-test (Table 2a). samples was 74 and 78%, respectively. qRT-PCR was also used Since none of the genes analyzed provided 100% sensitivity to analyze skin samples from SS patients (n ¼ 10; Supplementary in combination with 100% specificity, we next tested whether Table 1), MF patients (n ¼ 10) and ID patients (psoriasis, n ¼ 5). multiple analyses of expression values of several genes had a Since the number of skin samples was not large enough to predictive capacity better than that of single genes. A multiple calculate threshold values, expression values higher than the logistic regression model including several genes was performed highest expression value in the control group (ID) were using FORWARD-selection in SAS Procedure PROC LOGISTIC considered positive. The expression of CDO1 (DNM3) was in order to find the ideal combination of genes and to optimize detected to be increased in 6 (5) out of 10 SS patients (Figures 1c the discrimination between SS patients and controls. The and d), while expression of the two genes was not increased combined analysis of CDO1 and DNM3 gene expression levels among the 10 MF skin samples. These data indicate that CDO1 in PBMC samples had the best predictive value (AUC ¼ 0.956) and DNM3 are specifically upregulated in SS, as there was no for the discrimination between SS patients and control samples. difference between MF and ID. A higher number of skin In contrast, analysis of CD4 þ T-cell samples revealed that samples, however, are needed to confirm this finding. In combined analysis of two or more genes did not result in an addition, CDO1 expression was analyzed in activated T cells improved predictive capacity compared to analysis of any single of healthy donors. However, T-cell activation did not exert any gene alone. These results indicate that the combined analysis of effects on CDO1 expression indicating that expression of CDO1 the expression of CDO1 and DNM3 in PBMC samples may is an event associated with the molecular pathogenesis of SS suffice for the differential diagnosis of SS versus ID and is an (Supplementary Text 1).

Leukemia Se´zary syndrome gene expression N Booken et al 397 Table 2 Statistical analysis of quantitative real-time PCR results for gene expression in Sezary syndrome peripheral blood mononuclear cells (PBMC) in comparison to PBMC of healthy blood donors (HBD) and inﬂammatory diseases (ID) samples

(a) P-values were calculated using the Wilcoxon-Mann–Whitney U-test Gene SS vs HBD SS vs ID SS vs ID+HBD

STAT 4 0.0002 0.0002 o 0.0001 DNM3 0.0002 0.0004 o 0.0001 CDO1 0.0003 0.0004 o 0.0001 PLS3 0.0028 0.0005 o 0.0001 TNFSF11 0.0015 0.0042 0.0001 NEDD4L 0.0125 0.0025 0.0005 KLHDC5 0.0028 0.0132 0.0006 CD1D 0.0098 0.0043 0.0006 IGFL2 0.0228 0.0208 0.0038 RDH10 0.0175a 0.0533a 0.0065 DUSP 4 0.0076 0.1878a 0.0118 IGFBP 4 0.2519a 0.1042a 0.0795a GATA-3 0.4021a 0.7195a 0.445a TRAIL 0.4216a 0.0623a 0.5117a

(b) AUC values for the discrimination between Sezary syndrome patients and controls calculated using ROC statistic Gene AUCa,b P-valuec Sensitivity Speciﬁcity Sumd

STAT4 0.906 0.0005 0.74 1.0 1.74 DNM3 0.896 0.033 0.85 0.80 1.65 CDO1 0.884 0.032 0.74 0.95 1.69 PLS3 0.854 0.066 0.74 1.0 1.74 TNFSF11 0.828 0.077 0.55 1.0 1.55 NEDD4L 0.801 0.013 0.63 0.95 1.58 KLHDC5 0.799 0.045 0.59 0.95 1.54 CD1D 0.796 0.009 0.7 0.9 1.6 IGFL2 0.749 0.061 0.67 0.95 1.62 RDH10 0.735 0.068 0.63 0.8 1.43 DUSP 4 0.717 0.043 0.55 0.90 1.45 IGFBP 4 0.652 0.039 0.48 1.0 1.48 GATA-3 0.563 0.278 0.48 0.8 1.28 TRAIL 0.555 0.285 0.59 0.65 1.24 AUC, area under curve; ROC, receiver operating characteristic. All genes included except for Stat4 and CD1D were upregulated in Sezary syndrome. aNot significant. bAUC, sensitivity and specificity are related to the optimal cut-off point. cP-value of the Wald’s test. dSum of sensitivity and specificity.

Immunohistochemical analyses of CDO1 in PBMC cells.11–13 While single gene analysis with any of these Immunohistochemical staining was performed with rabbit anti- seven genes may suffice in CD4 þ T-cell samples, high human CDO1 antiserum on samples of CD4 þ cells isolated sensitivity and high specificity in PBMC samples is achieved from SS patients and healthy donors. Staining of hepatocytes in preferentially by combined analysis of mRNA expression liver samples served as a positive control. Strong staining of levels of CDO1 and DNM3. Analysis of CDO1 and DNM3 about 30% of CD4 þ T cells was observed in the case of SS in PBMC and skin samples of SS versus MF patients also patients while only few CD4 þ T cells of healthy donors showed indicated that these markers identify SS as a molecularly unique positive staining which was, however, much weaker than that of CTCL. SS CD4 þ cells (Supplementary Figure 1). The three microarray studies previously performed in SS have not yielded uniform and reproducible molecular signatures for the disease.11,13,14 This is probably due to the Discussion differences in microarray platforms used by the different groups,11,13 to strategic differences in identifying common In the present study, we used global gene transcription analysis versus differential traits of SS and MF,11,13,14 and to selection with Affymetrix Gene Chip technology to identify genes of control populations including Th2 T cells.13 Despite the characteristic for leukemic CTCL cells in SS patients. Special differences between previous gene profiling studies in SS, our interest was directed to genes that are highly informative as microarray analysis partially confirmed the results of those two single genes or in combination for diagnostic purposes as studies comparing SS PBMCs/CD4 þ T cells with normal determined by qRT-PCR. Seven genes (PLS3, DNM3, IGFL2, donors.13,14 CDO1, TNFSF11, NEDD4L, KLHDC5) were found to be Since the analysis of whole gene expression profiles is an significantly overexpressed in SS. While DNM3, IGFL2, CDO1, expensive and complicated method, the markers identified NEDD4L and KLHDC5 are described to be overexpressed in by microarray experiments are usually used for the design of Sezary cells for the first time in this study, PLS3 and TNFSF11 real-time PCR-based test systems. In this respect, the selection have previously been noted to be highly expressed in Sezary of the statistical model used for evaluating the data obtained by

Leukemia Se´zary syndrome gene expression N Booken et al 398 PBMC – samples Skin – samples 16 18 14 16 12 14 12 10 10 8 8 6 CDO1 6 CDO1 4 4 2 2 0 0 SS MF ID SS MF ID

b d 6000 600

5000 500

4000 400

3000 300 DNM3 2000 200 DNM3 1000 100

0 0 SS MF ID SS MF ID

Figure 1 Gene expression of CDO1 (a and c) and DNM3 (b and d) in PBMC/lesional skin biopsies of Sezary syndrome patients (SS; n ¼ 27/ n ¼ 10), Mycosis fungoides patients (MF; n ¼ 10/n ¼ 10) and inﬂammatory dermatoses/psoriasis patients (ID; n ¼ 10/n ¼ 5) was determined by quantitative RT–PCR. Sample values within the 95% conﬁdence interval are shown. (a and b) The expression of CDO1 (a) and DNM3 (b) was only increased in 1 out of 10 MF PBMC samples similar to control samples from HBD and ID while the percentage of positive SS samples was 74 and 78%, respectively. (c and d) The expression of CDO1 (DNM3) was detected to be increased in 6 (5) out of 10 SS patients while expression of the two genes was not increased among the 10 MF skin samples.

real-time PCR may considerably influence the outcome of the in which taurine chloramine is generated. It is important to note analysis. To evaluate our data and to identify useful diagnostic that impaired apoptosis is a hallmark of CTCL. CDO1 is not marker genes, we used ROC methodology based on logistic normally expressed in peripheral blood leukocytes, but it is regression, a statistical approach that is more robust than linear highly expressed in liver and brain. We have shown here that discriminant analysis used by others.7,11 As a result, we show CDO1 is not induced in CD4 þ T cells by either activation with here that combined analysis of the expression of CDO1 and PHA, IL-2, anti-CD2, anti-CD3 or a combination of anti-CD3 DNM3 in PBMCs achieves high specificity and sensitivity in the and anti-CD28. The CDO1 promoter, however, has been shown diagnosis of SS. When we compared our results with the to harbor a c-myb-responsive element (gttg) at À71.18 Interest- multigene qRT-PCR assay analyzing expression of STAT4, PLS3, ingly, we have previously reported that c-myb is consistently TRAIL, GATA-3 and CD1D as proposed by Nebozhyn et al.,7 it overexpressed in PBMC from SS patients, but not PBMC from turned out that only STAT4 (and PLS3) had a similarly high MF patients.19 Upregulated c-myb may well contribute to discriminative power for the diagnosis of SS as DNM3 and upregulation of CDO1 in SS PBMC indicating hierarchical CDO1 and that STAT4 analysis may complement DNM3 pathways in the molecular pathogenesis of SS. analysis as effectively as analysis of CDO1 in a combined The importance of transcription factor-mediated regulation of approach. We therefore suggest that a multi-center trial should gene expression in SS is further substantiated by the following evaluate a 4 gene qRT-PCR assay consisting of three upregulated findings: (a) TWIST1, a bHLH transcription factor known to be genes (PLS3, DNM3, CDO1) and one downregulated gene overexpressed in SS and also found upregulated by gene (STAT4) to minimize misclassifications in the diagnosis of SS. profiling in SS in this study (log 10 P46.2), regulates DNM3 The SS gene signature identified here does not only provide expression;20 (b) PEG-10 found upregulated by gene profiling in new SS marker molecules that can be used for diagnostic SS in this study (log 10 P43.1) is a target gene of c-myc, another purposes, but it also opens novel clues to unravel the molecular oncogene and transcription factor involved in the pathogenesis pathogenesis of SS. One of the most consistent SS marker genes of SS21 and (c) the PLS3/T-plastin promoter is not methylated in newly identified here, that is, cysteine dioxygenase (CDO1), is lymphocytes despite strong methylation in other leukocytes22 the rate limiting enzyme in the synthesis of taurine, an important rendering it accessible to SS-associated transcriptional regulators semi-essential amino acid. The key role of CDO1 in this still to be identified. One such putative transcriptional regulator biosynthetic pathway is underlined by the finding that the other could be TOX, another bHLH transcription factor found components of this biosynthetic pathway such as CSAD, upregulated by gene profiling in SS in this study (log 10 P46.3) SLC6A6 and GCLM (Supplementary Table 2) were expressed, that is known to be involved in thymocyte differentiation.23 but not upregulated in SS PBMC (data not shown). Taurine itself In summary, we have identified and validated marker genes is highly protective against apoptosis of T cells induced by CD95 differentially expressed by leukemic T cells in SS, including engagement or by AICD15 and inhibits apoptosome formation.16 PLS3, DNM3, IGFL2, CDO1, TNFSF11, NEDD4L and KLHDC5. Taurine is also chemoprotective against methotrexate and These genes well discriminate between peripheral blood vincristine.17 In case of vincristine, taurine inhibits the mononuclear cells in SS as compared to healthy controls and degradation of vincristine by hypochloric acid (HOCl), a process ID. In addition, CDO1 and DNM3 distinguish between PBMCs

Leukemia Se´zary syndrome gene expression N Booken et al 399 and lesional skin of SS versus MF. Combined analysis of CDO1 10 Wald A. Tests of statistical hypotheses concerning several and DNM3 in PBMC samples may be used with high sensitivity parameters when the number of observations is large. Trans Amer and specificity for the diagnosis of SS and may be complemen- Math Soc 1943; 54: 426–482. ted by analysis of STAT4 and PLS3 to reduce misclassifications 11 Kari L, Loboda A, Nebozhyn M, Rook AH, Vonderheid EC, Nichols C et al. Classification and prediction of survival in patients with the of patient samples. Further studies on the functional level are leukemic phase of cutaneous T cell lymphoma. J Exp Med 2003; needed to evaluate the potential biological importance of these 197: 1477–1488. proteins and to validate them as potential therapeutic targets for 12 Su MW, Dorocicz I, Dragowska WH, Ho V, Li G, Voss N et al. this hitherto intractable disease. Aberrant expression of T-plastin in Sezary cells. Cancer Res 2003; 63: 7122–7127. 13 Van Doorn R, Dijkman R, Vermeer MH, Out-Luiting JJ, van der Acknowledgements Raaij-Helmer EM, Willemze R et al. Aberrant expression of the tyrosine kinase receptor EphA4 and the transcription factor twist in Sezary syndrome identified by gene expression analysis. Cancer We thank Dr M Goebeler for critically reading the manuscript and Res 2004; 64: 5578–5586. Dr K Schledzewski, Mrs Arif-Said, Mrs Schmuttermaier and Mrs 14 Hahtola S, Tuomela S, Elo L, Hakkinen T, Karenko L, Nedoszytko Demory for excellent technical assistance. B et al. Th1 response and cytotoxicity genes are down-regulated in cutaneous T-cell lymphoma. Clin Cancer Res 2006; 12: 4812–4821. References 15 Maher SG, Condron CE, Bouchier-Hayes DJ, Toomey DM. Taurine attenuates CD3/interleukin-2-induced T cell apoptosis in an in 1 Dippel E, Goerdt S, Assaf C, Stein H, Orfanos CE. Cutaneous T-cell vitro model of activation-induced cell death (AICD). Clin Exp lymphoma severity index and T-cell gene rearrangement. Lancet Immunol 2005; 139: 279–286. 1997; 350: 1776–1777. 16 Takatani T, Takahashi K, Uozumi Y, Shikata E, Yamamoto Y, Ito T 2 Klemke CD, Mannsmann U, Poenitz N, Dippel E, Goerdt S. et al. Taurine inhibits apoptosis by preventing formation of the Prognostic factors and prediction of prognosis by the CTCL Severity Apaf-1/caspase-9 apoptosome. Am J Physiol Cell Physiol 2004; Index in mycosis fungoides and Se´zary syndrome. Br J Dermatol 287: C949–C953. 2005; 153: 118–124. 17 Ozgen U, Savasan S, Stout M, Buck S, Ravindranath Y. Further 3 Dippel E, Assaf C, Hummel M, Schrag HJ, Stein H, Goerdt S et al. elucidation of mechanism of resistance to vincristine in myeloid Clonal T-cell receptor gamma-chain gene rearrangement by PCR- cells: role of hypochlorous acid in degradation of vincristine by based GeneScan analysis in advanced cutaneous T-cell lympho- myeloperoxidase. Leukemia 2000; 14: 47–51. ma: A critical evaluation. J Pathol 1999; 188: 146–154. 18 Ramsden DB, Kapadi A, Fitch NJ, Farmer MJ, Bennett P, Williams 4 Dippel E, Klemke D, Hummel M, Stein H, Goerdt S. T-cell AC. Human cysteine dioxygenase type I (CDO-I; EC 1.13.11.20): clonality of undetermined significance. Blood 2001; 98: 247–248. 50 flanking region and intron-exon structure of the gene. Mol 5 Klemke CD, Fritzsching B, Franz B, Kleinmann EV, Oberle N, Pathol 1997; 50: 269–271. Poenitz N et al. Paucity of FOXP3+ cells in skin and peripheral 19 Poenitz N, Simon-Ackermann J, Gratchev A, Qadoumi M, Klemke blood distinguishes Sezary syndrome from other cutaneous T-cell CD, Stadler R et al. Overexpression of c-myb in leukaemic and lymphomas. Leukemia 2006; 20: 1123–1129. non-leukaemic variants of cutaneous T-cell lymphoma. Dermatol- 6 Poszepczynska-Guigne E, Schiavon V, D’Incan M, Echchakir H, ogy 2005; 211: 84–92. Musette P, Ortonne N et al. CD158k/KIR3DL2 is a new phenotypic 20 Loebel DA, Tsoi B, Wong N, Tam PP. A conserved noncoding marker of Sezary cells: relevance for the diagnosis and follow-up of intronic transcript at the mouse Dnm3 locus. Genomics 2005; 85: Sezary syndrome. J Invest Dermatol 2004; 122: 820–823. 782–789. 7 Nebozhyn M, Loboda A, Kari L, Rook AH, Vonderheid EC, Lessin S 21 Li CM, Margolin AA, Salas M, Memeo L, Mansukhani M, et al. Quantitative PCR on 5 genes reliably identifies CTCL patients Hibshoosh H et al. PEG10 is a c-MYC target gene in cancer cells. with 5–99% circulating tumor cells with 90% accuracy. Blood Cancer Res 2006; 66: 665–672. 2006; 107: 3189–3196. 22 Lin CS, Lau A, Huynh T, Lue TF. Differential regulation of human 8 Chu TM, Weir B, Wolfinger R. A systematic statistical linear T-plastin gene in leukocytes and non-leukocytes: identification of modeling approach to oligonucleotide array experiments. Math the promoter, enhancer, and CpG island. DNA Cell Biol 1999; 18: Biosci 2002; 176: 35–51. 27–37. 9 Allison PD. Logistic Regression Using SAS System: Theory 23 Aliahmad P, O’Flaherty E, Han P, Goularte OD, Wilkinson B, and Application. SAS Institute Inc.: Cary, NC, 1999, Chapter 2 Satake M et al. TOX provides a link between calcineurin activation and 3:4–78. and CD8 lineage commitment. J Exp Med 2004; 199: 1089–1099.

Supplementary Information accompanies the paper on the Leukemia website (http://www.nature.com/leu)

Leukemia