Cell Cycle and Immune-Related Processes Are Significantly

ORIGINAL ARTICLE Cell cycle and immune-related processes are signiﬁcantly altered in chronic GVHD

SJ Oh1,2,8, SB Cho1,3,8,, S-H Park1, CZ Piao1, SM Kwon1, I Kim1,4, SS Yoon4, BK Kim4, EK Park1,5, JJ Kang6, S-J Yang6, WJ Lee7, C-H Yoo7, S Hwang7, SH Kim7, JH Kim1,3 and S Park1,4

1Diagnostic DNA Chip Center, Seoul National University College of Medicine, Seoul, Korea; 2Department of Internal Medicine, Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul, Korea; 3Seoul National University Biomedical Informatics, Seoul, Korea; 4Department of Internal Medicine, Seoul National University College of Medicine, Seoul, Korea; 5Department of Internal Medicine, Chung-ang University College of Medicine, Seoul, Korea; 6Macrogen Inc., Seoul, Korea and 7Digital Genomics Inc., Seoul, Korea

Currently, the pathogenesis of chronic GVHDis unclear. Introduction To elucidate the molecular characteristics underlying chronic GVHD, we analyzed the gene expression profiles Chronic GVHD, which is one of the most serious of 21 mononuclear cell samples from allogeneic hemato- complications of allogeneic hematopoietic stem cell trans- poietic stem cell transplantation (HSCT) recipients. Self plantation (HSCT), occurs in 20B70% of patients who organizing map (SOM) clustering showed that the entire survive for 100 days or more following the procedure.1 The expression profiles of chronic GVHDsamples were clearly pathologic characteristics of chronic GVHD include different from those of the non-GVHDsamples, and immune dysregulation, immunodeficiency and impaired significance analysis of microarray (SAM) demonstrated organ function,2 and its treatment consists of the adminis- that 120 genes, including PTDSS1, VAV1 and CD3D, tration of immunosuppressive medications for approxi- were up-regulated, and 5 genes, including calnexin, were mately 1–3 years.1 The high mortality associated with down-regulated in GVHDpatients. Gene ontology chronic GVHD makes it one of the major causes of death annotation revealed that these genes are related to the in stem cell transplant recipients.3 However, the patho- phosphorous metabolism and lipid biosynthesis. Quanti- genesis of chronic GVHD is still not well understood.4 tative real time polymerase chain reaction (qRT-PCR) Although the true pathophysiologic mechanism of experiments validated the up-regulation of PTDSS1, GVHD has not yet been fully clarified, alloreactive T cells VAV1 and CD3D in separate samples. Pathway-wise are thought to initiate chronic GVHD.2,5,6 Furthermore, global test revealed that differential gene expression in humoral immunity-related phenomena, such as B cell cell cycle and T cell immune-associated pathways were dysfunction, a high prevalence of anti-nuclear auto- significant between GVHDpatients and non-GVHD antibodies, and clinical manifestations of autoimmune patients. Seventeen classifier genes selected using a diseases are also observed in chronic GVHD patients.7–9 PAM (prediction analysis of microarray) algorithm Although these findings have been supported by individual showed favorable performance (prediction accuracy studies, it has been difficult to perform a comprehensive 0.85) for identifying patients with chronic GVHD. In ¼ study to examine these phenomena systematically. conclusion, we identified differentially expressed genes Therefore, in this study, we performed a microarray survey and pathways in chronic GVHDpatients using microarray of peripheral blood samples obtained from chronic GVHD analysis, and we also selected diagnostic genes predicting patients to elucidate the molecular characteristics under- chronic GVHDstatus. lying the pathogenesis of GVHD. Bone Marrow Transplantation (2008) 41, 1047–1057; In the post-genomic era, microarray analysis plays a key doi:10.1038/bmt.2008.37; published online 10 March 2008 role in the evaluation of whole genome mRNA expression. Keywords: chronic GVHD; microarray; cell cycle; Due to its high throughput capacity, microarray experi- immune process ments have yielded a large amount of previously unknown findings, and they are commonly used to identify diagnostic or prognostic biomarkers. For example, Golub et al.10 identified diagnostic markers for distinguishing AML and Correspondence: Dr S Park, Department of Internal Medicine, Seoul ALL, as well as subgroups of ALL patients who had National University College of Medicine, 28 Chongno-gu, Yungon different cell lineages using only data obtained by dong, Seoul 110-744, Republic of Korea. conducting microarray analyses. Furthermore, analysis of E-mail: [email protected] 8These authors contributed equally to this work. gene expression profiles using microarray data have also Received 28 September 2007; revised 18 December 2007; accepted 20 identified global gene expression regulatory relation- December 2007; published online 10 March 2008 ships,11,12 and may provide novel and holistic views Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1048 regarding the molecular markers and mechanisms under- Table 1 Summary of patients characteristics (non-GVHD; 10, lying pathologic processes. In this study, we evaluated GVHD; 11) peripheral blood samples obtained from 21 HSCT Clinical variables Total number recipients using an oligonucleotide microarray that contained 20 142 probes to assess immunologic perturba- Total number of patients 21 tion and other abnormal biologic processes associated with Sex, No. (%) chronic GVHD. We also evaluated the results of the M 14 (66.7) microarray analysis to determine if the gene expression F 7 (33.7) profile could be used as a diagnostic tool for patients with chronic GVHD. Patient age Median (range), years 42 (22–64)

Primary disease AML 9 (42.9) Materials and methods AA 5 (23.8) CML 2 (9.5) Sample population MDS 2 (9.5) MM 1 (4.8) A total of 21 patients (median age; 42 years, range; 22–64 NHL 1 (4.8) years) who received allogeneic HSCT between May 1991 RCC 1 (4.8) and June 2005 at the Seoul National University Hospital, Seoul, Korea were evaluated using gene expression Acute GVHD analysis. The characteristics of the GVHD and non-GVHD Yes 8 (38.1) No 13 (57.9) patients are shown in Table 1. All patients had received HSCT from HLA (human leukocyte antigen)-matched Conditioning regimen sibling donors with the exception of 2 patients whose HLA Myeolablative 12 (57.2) were mismatched in one A locus. All patients were in Non-myeloablative 9 (42.9) complete remission and in the complete chimerism state. Days after HSCT that last sample was collected Eleven of the 21 patients developed chronic GVHD, Median (range), months 43 (5–92) including one who received partially mismatched transplantation. Protocols for all cases in this study were Abbreviations: AA ¼ aplastic anemia; MDS ¼ myelodysplastic syndrome; approved by the Institutional Review Board at the Seoul MM ¼ multiple myeloma; NHL ¼ non-Hodgkin’s lymphoma; RCC ¼ renal National University Hospital, and informed consent was cell carcinoma. obtained from all patients. resulted in a fragmented target with a size range between 100 and 200 bases. Target preparation Two micrograms of total RNA were extracted from Array hybridization mononuclear cells of peripheral blood and added to a Ten micrograms of fragmented target cRNA in 260 mlof reaction mix with a final volume of 12 ml that contained hybridization solution was used for hybridization with each T7-(dT) oligonucleotide primer. The mixture was 24 UniSet Human 20 K I Bioarray (Amersham Biosciences, incubated for 10 min at 70 1C and then chilled on ice. Little Chalfont, Bucks, UK). The hybridization solution While the mixture was on ice, 2 mlof10Â first-strand was heated to 90 1C for 5 min to denature the cRNA, and buffer, 4 mlof5mM dNTP mix, 1 ml of RNase inhibitor then chilled on ice for 5 min. Next, the sample was vortexed (40 U/ml) and 1 ml of Superscript II RNase H– reverse for 5 s at maximum speed, and 250 ml was injected into the transcriptase (200 U/ml) was added to give a final volume of inlet port of the hybridization chamber, which was then 20 ml. The mixture was then incubated for 2 h in a 42 1C placed in a 12-slide shaker tray. The hybridization chamber water bath. Next, second-strand cDNA was synthesized in ports were then sealed with 1 cm sealing strips (Amersham a final volume of 100 ml in a mixture containing 10 mlof Biosciences), and the shaker tray containing the slides was 10 Â second-strand buffer, 4 mlof5mM dNTP mix, 2 mlof loaded into a shaking incubator. The slides were then DNA polymerase mix (20 U/ml) and 1 mlofRNaseH(2U/ml) incubated for 18 h at 37 1C while shaking at 300 r.p.m. for 2 h at 16 1C. The cDNA was then purified using a Qiagen QIAquick purification kit, dried down, and resuspended in IVT reaction mix that contained 4 mlof10Â reaction buffer, Post-hybridization processing using Cy5-streptavidin 4 mlof75mM ATP, 4 mlof75mM GTP, 4 mlof75mM CTP, The 12-slide holder was removed from the shaking 3 mlof75mM UTP, 7.5 mlof10mM Biotin 11-UTP and 4 mlof incubator and the hybridization chamber was then enzyme mix with a final volume of 40 ml. The reaction mix removed from each slide. Next, each slide was briefly was then incubated for 14 h at 37 1C and the cRNA target rinsed in TNT buffer (0.075 M Tris-HCl pH 7.6, 0.1125 M was purified using an RNeasy kit (Qiagen, Hilden, Germany). NaCl, 0.0375% Tween 20) at room temperature, and then The cRNA yield was quantified by measuring the UV washed in TNT buffer at 42 1C for 60 min. The signal was absorbance at 260 nm, and then fragmented into 40 mM then developed using a 1:500 dilution of Cy5-streptavidin Tris–acetate (TrisOAc, pH 7.9), 100 mM KOAc and (Amersham Bioscience) for 30 min at room temperature. 31.5 mM MgOAc at 94 1C for 20 min. This typically Excess dye was removed by washing the slide four times

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1049 with TNT buffer (0.1 M Tris-HCl pH 7.6, 0.15 M NaCl, numbers of genes can be identified as significantly changed, 0.05% Tween 20) for 5 min each, at room temperature. however it is cumbersome to investigate and infer the Next, the slides were rinsed in wash buffer containing 0.1 Â relationship between the DEGs and pathophysiology of SSC and 0.05% Tween 20, and then dried by centrfugation. chronic GVHD. Therefore, functional annotation with Processed slides were scanned using an Axon GenePix gene ontology (GO) is used to manipulate the problem. 4000B Scanner with the laser set to 635 nm, the photo- The DAVID (the database for annotation, visualization multiplier tube (PMT) voltage set to 600 and the scan and integrated discovery) web site was used to annotate resolution set to 10 mm. Images for each slide were then the DEGs with the GO biologic process terms to infer the analyzed using the CodeLink Expression Analysis Software tentative biological mechanisms involved in GVHD.18 v4.1 (GE Healthcare Life Science, Tempe, AZ, USA). A global test was then applied to explore the differentially expressed pathways.19 The global test is a modified generalized linear model (GLM) approach for microarray Quantitative real time polymerase chain reaction data analysis, which was developed to overcome limits of (qRT-PCR) the single gene-wise analysis of microarray data. This To validate the results of the microarray experiment, real method can utilize predefined gene sets that are extracted time PCR was performed on 19 individual patient samples under the biological context. To analyze the data generated that had not been used in the microarray experiment. Of in this study, we defined biological pathways as gene sets the differentially expressed genes, PTSSD1, VAV1 and and then tested whether each pathway was differentially CD3D were arbitrarily selected for the PCR experiment expressed in GVHD patients. An adjusted p value was because their biological implication is relevant to the obtained using the Bonferroni correction procedure to pathophysiology of GVHD. Total RNA was isolated using account for multiple testing. Biomarker genes and statis- an RNeasy Mini kit (QIAGEN) and cDNA was synthe- tical models for predicting disease status were then sized using a reverse transcription system (Promega, extracted using the PAM (prediction analysis of micro- Madison, WI, USA) with 2 mg of total RNA in a 40-ml array) algorithm.20 The selection criterion for shrinkage reaction. Quantitative PCR was carried out on an ABI- was the minimal number of genes and the highest PRISM thermal cycler (Applied Biosystems, Foster City, prediction accuracy. The R 2.4.1 package (The Compre- CA, USA) using TaqMan pobe and TaqMan Universal hensive R Archive Network (CRAN), http://cran.r-project. PCR Master Mix according to the manufacturer’s instruc- org/) was used for all statistical analyses. tions. The Mann-Whitney U test was used to determine if a significant difference existed between the PCR results obtained from GVHD positive and negative patients. Results Data analysis To remove the systemic error and bias of the microarray Global gene expression profile experiment, variable stabilizing normalization (VSN)13 and The result of SOM clustering showed that the whole the quantile normalization method14 were applied to the mRNA expression profile of mononuclear cells obtained scanned probe intensities. Missing value estimation was from GVHD patients differed from those obtained from then performed using the k-nearest neighbor (KNN) non-GVHD patients. In the result map, samples obtained method,15 and only genes that had missing values for from GVHD patients revealed a tendency to cluster less than 20% of the total samples were considered in together as well as separately from samples obtained from the estimation. Consequently, 18 356probes (91.1% of non-GVHD patients (Figure 1). the total probes) were used in the analysis. Vendor-provided probe information was used for mapping each probe to the GenBank accession number. Collection Differentially expressed genes in samples obtained from of the available pathways and microarray probe mapping GVHD and non-GVHD patients to the gene members of the each pathway were performed Among 18 356probes, the SAM method identified 125 using the ArrayXpath knowledge base,16 and a total of probes as DEGs with 1,000 permutations and a FDR of 460 pathways were used in the analysis. We tested the 0.005. Of these genes, 120 were up-regulated and 5 were hypothesis that the global gene expression pattern of down-regulated in the samples obtained from GVHD mononuclear cells was different between GVHD and patients. Phosphatidylserine synthase 1 (PTDSS1) was the non-GVHD patients using sample-wise, self organizing most up-regulated gene in the GVHD group, and the map (SOM) clustering. A linear grid based on the direction pleckstrin homology domain containing family B (evectins) of the first two principal components was used to initialize member 1(PLEKHB1), cyclin dependent kinase 6(CDK6), SOM clustering. The learning rate and neighborhood and Vav 1 oncogene (VAV1), were the highest ranked up- function used were the inverse-time and Gaussian function, regulated genes in samples obtained from the GVHD respectively. patients. Calnexin (CANX) was the most down-regulated Significance analysis of microarray (SAM) was gene in the GVHD patient samples. Eukaryotic elongation performed to identify the differentially expressed genes factor-2 kinase (EEF2 K), and transducin-like enhancer (DEGs) in GVHD patients.17 After 1000 times permuta- of split 3 (TLE3) were also highly down-regulated tions, DEGs with a false discovery rate (FDR) level of in the GVHD specimens (Table 2, Figure 2 and 0.005 were obtained. In microarray experiments, large Supplementary data).

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1050 Cluster 1 Cluster 2 phase into the S phase, were identified: the ‘Cell Cycle: G1/S Check Point’ pathway, ‘Influence of Ras and Rho proteins on G1 to induce the transition to the S phase’ and the ‘RB Tumor Suppressor/Checkpoint Signaling pathway that is activated in response to DNA damage’. Of the cell cycle associated pathways, the ‘Cell Cycle: G1/S check point’ pathway was the most significant pathway identified in the analysis (P ¼ 0.0000013). When the immune-related pathways were considered, many T cell associated pathways were identified, including the ‘T Cell Receptor Signaling’ pathway, ‘Lck and Fyn tyrosine kinases involved in the initiation of TCR Activa- Cluster 3 Cluster 4 tion’, ‘T Cytotoxic Cell Surface Molecules’, and ‘T Helper Cell Surface Molecules’ pathway. Further, the ‘IL 2 signaling’ pathway and the ‘Ras-Independent pathway in NK cell-mediated cytotoxicity’ were also related to the T cell immune process. In addition, the ‘Fc Epsilon Receptor I Signaling in Mast Cells’ pathway, which is part of B cell-mediated immunity, was also found to be significant. The motility-related pathways included the ‘S1P_Signal- ing’ and ‘Rac 1 cell motility signaling’ pathways. The ‘Signal Dependent Regulation of Myogenesis by Core- pressor MITR’, ‘ALK in cardiac myocytes’ and ‘Control of Figure 1 Result of self organizing map clustering of the whole gene skeletal myogenesis by HDAC & calcium/calmodulin- expression profile. Each cluster contains both non-GVHD and GVHD dependent kinase’ pathways were found to be significant patients. If we regard the majority disease status as a gold standard of each cluster, 76.1% (16/21) of the samples were correctly and 23.9% (4/21) of in the myogenesis related group. samples were incorrectly assigned to the corresponding disease status The remaining pathways identified included the phos- without the use of clinical information. pholipase C signaling pathway, which was related to signaling process, the estrogen-responsive protein Efp, which controls the cell cycle and breast tumor growth, and the pathway that controls androgen, estrogen and Gene ontology enrichment study was performed to sulfur metabolism. define the molecular characteristics of DEGs. The following six biological process (BP) terms were found to be Validation of microarray experiments using qRT-PCR significant: the frizzled signaling pathway, phosphorous The microarray results were confirmed in 19 samples metabolism, phosphate metabolism, phosphorylation, (10 non-GVHD and 9 GVHD samples) not included protein amino acid phosphorylation and lipid biosnythesis. in the microarray experiment using qRT-PCR. The When molecular function terms were applied, 9 terms were Mann–Whitney test revealed that the expression levels of found to be significant: transferase activity, protein serine/ PTSSD1, VAV1 and CD3D were significantly different in threonine kinase activity, kinase activity, GTPase regulator the GVHD positive and negative groups (P 0.001, activity, N-acetylglucosamine 6-O-sulfotransferase activity, o Figure 3). protein kinase activity, sulfotransferase activity, transferase activity, transferring sulfur-containing groups, transferase activity and transferring phosphorus-containing groups Classification analysis (Table 3). After the initial PAM procedure was conducted, the prediction accuracies according to the shrinkage level and corresponding gene set were determined. When a shrinkage Differentially expressed pathways in samples obtained from threshold of 3.5 was used, 17 genes yielded the best GVHD and non-GVHD patients classification performance, which was 85% prediction The global test conducted to examine the expression profile accuracy with a 10 fold cross validation (Figure 4). These of 460 pathways with Bonferroni’s multiple testing correc- genes included phosphatidylserine synthase 1, pleckstrin tion revealed that difference of gene expression of 30 homology domain containing family B (evectins) pathways were statistically significant (Table 4, Po0.00012 member 1, cyclin-dependent kinase 6, vav 1 oncogene, after adjustment). These pathways can be categorized into 5 mitochondrial ribosomal protein L19, MRNA (clone groups: cell-cycle associated pathways, cell mediated ICRFp507I1077), coiled-coil domain containing 6, immuno- immune associated pathways, cell motility associated globulin superfamily member 4B, Homo sapiens AOC3 pathways, myogenesis associated pathways and other gene for vascular adhesion protein-1 exon 2, leucine rich pathways. repeat containing 27, chromosome 10 open reading frame In the cell cycle-related group, the following pathways, 75, RCD1 required for cell differentiation1 homolog which were associated with the transition of cells in the G1 (Schizosaccharomyces pombe), phospholipase A2, group

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1051 Table 2 Differentially expressed genes in chronic GVHD patients

Reporter ID Reporter name Score Fold change

NM_014754 Phosphatidylserine synthase 1 4.467 5.109 NM_021200 Pleckstrin homology domain containing, family B (evectins) 4.402 4.323 member 1 BM542736Cyclin-dependent kinase 6 4.294 4.855 NM_005428 Vav 1 oncogene 4.281 4.940 NM_014763 Mitochondrial ribosomal protein L19 4.260 4.372 AK092450 MRNA (clone ICRFp507I1077) 4.095 4.381 NM_005436Coiled-coil domain containing 6 4.028 4.225 NM_021189 Immunoglobulin superfamily, member 4B 3.924 5.298 NM_005444 RCD1 required for cell differentiation1 homolog (S. pombe) 3.917 3.399 AB051461 Leucine rich repeat containing 27 3.917 3.556 NM_014772 KIAA0427 3.908 3.266 NM_014737 Ras association (RalGDS/AF-6) domain family 2 3.878 3.306 BM556332 Chromosome 10 open reading frame 75 3.857 3.772 BM688744 RAP1A, member of RAS oncogene family 3.848 2.976 AK093645 Phospholipase A2, group IIF 3.843 3.778 NM_021212 HCF-binding transcription factor Zhangfei 3.814 3.402 AB050501 Homo sapiens AOC3 gene for vascular adhesion protein-1, exon 2 3.809 4.644 AK094585 Hypothetical protein FLJ37266 3.754 3.042 NM_014746Ring finger protein 144 3.746 3.515 AK091150 LOC440460 3.659 3.118 NM_021062 Homo sapiens H2B histone family, member F (H2BFF), mRNA 3.623 3.404 NM_015630 Enhancer of polycomb homolog 2 (Drosophila) 3.622 3.356 AK074962 Transmembrane emp24 protein transport domain containing 7 3.587 3.467 NM_021179 Hypothetical protein LOC57821 3.540 3.386 D78576Human DNA for 14-3–3 protein eta chain, exon1 3.519 3.197 NM_005398 Protein phosphatase 1, regulatory (inhibitor) subunit 3C 3.5063.510 NM_006411 1-acylglycerol-3-phosphate O-acyltransferase 1 (lysophosphatidic 3.504 3.321 acid acyltransferase, alpha) BM475617 Tropomyosin 4 3.493 3.195 AB051503 Centaurin, beta 5 3.491 2.913 NM_021242 MID1 interacting protein 1 (gastrulation specific G12-like 3.481 2.746 (zebrafish)) BM128045 Neurocalcin delta 3.4763.473 AF219991 Homo sapiens intestinal N-acetylglucosamine-6-O- 3.439 3.101 sulfotransferase (CHST5) and corneal N-acetylglucosamine-6-O- sulfotransferase (CHST6) genes, complete cds AK095695 Hypothetical protein MGC21644 3.434 2.754 NM_005420 Sulfotransferase family 1E, estrogen-preferring, member 1 3.3963.158 NM_021168 RAB40C, member RAS oncogene family 3.386 3.249 AB046847 KIAA1627 protein 3.340 3.040 NM_019843 Eukaryotic translation initiation factor 4E nuclear import factor 1 3.329 3.235 NM_004588 Sodium channel, voltage-gated, type II, beta 3.310 2.067 AL133064 Acyl-Coenzyme A binding domain containing 5 3.271 2.989 NM_021153 Cadherin 19, type 2 3.234 3.256 AB046800 Netrin-G1 ligand 3.198 2.963 NM_020421 AarF domain containing kinase 1 3.173 2.353 NM_005421 T-cell acute lymphocytic leukemia 2 3.148 2.068 BE675486 Slingshot homolog 2 (Drosophila) 3.141 4.408 NM_032320 K+ channel tetramerization protein 3.135 3.035 NM_014790 Jak and microtubule interacting protein 2 3.069 2.603 NM_004046ATP synthase, H+ transporting, mitochondrial F1 complex, 3.059 2.263 alpha subunit, isoform 1, cardiac muscle NM_025189 Zinc finger protein 430 3.033 2.382 AK094989 Zinc finger protein 585B 3.025 2.300 NM_000732 CD3D antigen, delta polypeptide (TiT3 complex) 3.021 2.338 NM_021073 Bone morphogenetic protein 5 3.0162.710 NM_017825 ADP-ribosylhydrolase like 2 3.0162.342 NM_003288 Tumor protein D52-like 2 3.008 2.036 AI718785 Meis1, myeloid ecotropic viral integration site 1 homolog (mouse) À3.121 0.389 NM_005078 Transducin-like enhancer of split 3 (E(sp1) homolog, Drosophila) À3.124 0.411 AK056323 Homo sapiens, clone IMAGE:5303499, mRNA À3.1360.364 NM_013302 Similar to NAD(P) dependent steroid dehydrogenase-like À3.177 0.498 M94859 Calnexin À3.280 0.244

The score represents the results of a statistical test by the SAM to evaluate the difference in expression between non-GVHD and GVHD specimens. The fold change is the ratio of average expression values of non-GVHD specimens to those of GVHD specimens. A positive and negative score indicates up- regulation and down-regulation in the GVHD specimens, respectively. Full list of the SAM result is in the supplementary data.

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1052 GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (-) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) GVHD (+) Figure 2 Heatmap of genes differentially expressed (n ¼ 125) in GVHD and non-GVHD samples. Row labels indicate the genes and column labels indicate the conditions of the samples. The genes are sorted according to the fold ratio.

Table 3 Functional annotation of differentially expressed genes

Category Term Count (%) P-value

GO BP Frizzled signaling pathway 3 (2.59) 4.71E-03 Phosphorus metabolism 13 (11.21) 8.59E-03 Phosphate metabolism 13 (11.21) 8.59E-03 Phosphorylation 11 (9.48) 1.56E-02 Protein amino acid phosphorylation 9 (7.76) 3.11E-02 Lipid biosynthesis 5 (4.31) 4.07E-02

GO MF Transferase activity 20 (17.24) 6.35E-03 Protein serine/threonine kinase activity 8 (6.90) 9.35E-03 Kinase activity 12 (10.34) 1.76E-02 GTPase regulator activity 6(5.17) 1.87E-02 N-acetylglucosamine 6-O-sulfotransferase activity 2 (1.72) 2.54E-02 Protein kinase activity 9 (7.76) 2.58E-02 Sulfotransferase activity 3 (2.59) 2.93E-02 Transferase activity, transferring sulfur-containing groups 3 (2.59) 3.47E-02 Transferase activity, transferring phosphorus-containing groups 12 (10.34) 4.59E-02

Abbreviations: BP ¼ biological process; GO ¼ gene ontology; MF ¼ molecular function. Count indicates the number of genes assigned to the gene ontology term.

IIF KIAA0427, HCF-binding transcription factor Discussion Zhangfei, RAP1A member of RAS oncogene family, and ring finger protein 144. The sensitivity and specificity of the In this study, we screened genes obtained from peripheral classifier genes were 0.82 and 0.90, respectively. blood mononuclear cells to elucidate the molecular

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1053 Table 4 Differentially expressed biological pathways in chronic GVHD patients

Category Pathway name Number of genes P-value

Cell cycle associated pathways Cyclins and Cell Cycle Regulation 263.00E-06 Cell Cycle: G1/S Check Point 39 1.00E-06 Cell_Cycle 113 4.00E-06 Cell Cycle: G2/M Checkpoint 34 2.50E-05 Inﬂuence of Ras and Rho proteins on G1 to S Transition 35 8.00E-06 RB Tumor Suppressor/Checkpoint Signaling in response to DNA damage 22 2.10E-05 cdc25 and chk1 Regulatory Pathway in response to DNA damage 161.70E-05 Regulation of cell cycle progression by Plk3 14 2.70E-05

Immune associated pathways T Cell Receptor Signaling Pathway 67 2.30E-05 Ras-Independent pathway in NK cell-mediated cytotoxicity 22 1.30E-05 IL 2 signaling pathway 34 6.60E-05 Fc Epsilon Receptor I Signaling in Mast Cells 52 5.50E-05 HIV Induced T Cell Apoptosis 9 7.00E-05 Lck and Fyn tyrosine kinases in initiation of TCR Activation 13 7.50E-05 T Cytotoxic Cell Surface Molecules 12 8.80E-05 CTL mediated immune response against target cells 9 9.10E-05 T Helper Cell Surface Molecules 12 3.40E-05

Cell motility associated pathways S1P_Signaling 29 4.50E-05 Rac 1 cell motility signaling pathway 29 4.20E-05

Myocytes associated pathways Signal Dependent Regulation of Myogenesis by Corepressor MITR 13 6.40E-05 ALK in cardiac myocytes 45 6.70E-05 Control of skeletal myogenesis by HDAC & calcium/calmodulin-dependent kinase 39 5.10E-05

Other pathways Sulfur_metabolism 7 9.40E-05 Stathmin and breast cancer resistance to antimicrotubule agents 27 5.00E-06 Regulation of BAD phosphorylation 28 9.00E-06 Estrogen-responsive protein Efp controls cell cycle and breast tumors growth 17 1.50E-05 Androgen_and_estrogen_metabolism 263.40E-05 Phospholipase C Signaling Pathway 14 2.20E-05 Antiarrhythmic_Drug_Pathways 60 8.40E-05

The number of genes indicates the number of member genes in each pathway. The pathways were collected from the BioCarta (www.biocarta.com), KEGG (www.genome.jp/kegg/) and PharmGKB (www.pharmgkb.org) web sites. Assignment of microarray probes to the corresponding pathways was performed via the ArrayXPath knowledge base.16

0.25 0.20 0.4 0.20 0.15 0.3 0.15 0.10 0.2 0.10 0.05 0.1 GVHD(−) GVHD(+) GVHD(−) GVHD(+) GVHD(−) GVHD(+)

Figure 3 Box plots for the qRT-PCR results from non-GVHD (n ¼ 9) and GVHD (n ¼ 10) samples. Box plots of (a) PTSSD1 (Phosphatidylserine synthase 1), (b) VAV1 (Vav 1 oncogene) and (c) CD3D (CD3D antigen, delta polypeptide). The samples were obtained independently from those of the microarray experiment. The difference in gene expression between the conditions was significant in all three genes. (Mann-Whitney test, Po0.001). signature associated with the pathogenesis of chronic SOM clustering was performed to delineate the global GVHD. To date, few microarray experiments regarding gene expression profiles of chronic GVHD patients and GVHD have been conducted, and to the best of our non-GVHD patients, and our results showed that non- knowledge, this is the first report that investigated gene GVHD and GVHD samples tended to aggregate sepa- expression profiles in chronic GVHD patients using rately. The clustering pattern indicated that the microarray microarray technology. This analysis revealed many analysis was capable of discriminating GVHD status consistent and novel findings related to the pathogenesis without any other clinical information, which may be of chronic GVHD. considered as fundamental evidence that information

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1054 3.147 NM_014754 0.714 2.397 NM_021200 0.298 3.127 BM542736 0.733 3.222 NM_005428 0.843 2.378 NM_014763 0.185 3.119 AK092450 0.896 2.345 NM_005436 0.247 3.010 NM_021189 0.523 2.769 AB050501 0.437 AB051461 2.134 0.323 BM556332 2.406 0.402 5.261 NM_005444 3.478 AK093645 2.314 0.430 5.200 NM_014772 3.470 NM_021212 5.147 3.401 5.300 BM688744 3.678 NM_014746 3.470 1.469

GVHD(−) GVHD(+)

Figure 4 Classifier genes selected using the PAM algorithm. The labels on the Y-axis represent the official gene symbols of the 17 classifier genes. For each gene, upper and lower bar indicate the mean expression level of GVHD positive and negative samples, respectively. The numbers on the bar represent the mean expression value of the gene under each condition. For visualization, original mean values are shifted until negative mean values become positive. Supplementary data contain the table about official gene names and original mean values of the classifier genes.

contained in microarray data is important and useful for CD3D (SAM score ¼ 3.021, fold change ¼ 2.338) plays a the elucidation of the pathophysiology of chronic GVHD. role in the maturation of T cell development.27 Increased Among the genes differentially expressed in GVHD expression of the CD3D gene indicates that T cell samples, phosphatidylserine synthase 1 (PTDSS1), which recruitment is active in GVHD patients, which supports is known to be related to phospholipid biosynthesis, the findings of previous studies.28 was the most up-regulated (SAM score ¼ 4.467, fold Up-regulation of the above 3 DEGs, which had not been change ¼ 5.109). It is well known that apoptosis plays a studied in chronic GVHD, was validated by qRT-PCR. key role in the pathogenesis of chronic GVHD21 and that The differences in mRNA expression between groups were phosphatidylserine inhibits immune responses by mediating statistically significant for all of the genes tested (Po0.001). the recognition and engulfment of apoptotic cells.22 There- Since independent samples that were not included in the fore, increased expression of PTDSS1, as shown in our microarray analysis were used in the analysis, our data study, may indicate a protective response against the strongly indicate that the molecules are associated with the overwhelming apoptotic process in GVHD. pathogenesis of chronic GVHD. VAV1 (SAM score ¼ 4.281, fold change ¼ 4.940) is a Our study showed that the CXCL13 gene was over- notable DEG because it has been reported to be a member expressed in patients with chronic GVHD (score ¼ 2.84, of many immune-related pathways, such as B and T cell fold change ¼ 2.40). CXCL13 is a highly effective attractant receptor signaling,23,24 natural killer cell mediated cyto- of human blood B lymphocytes,29 and its overexpression toxicity,25 and leukocyte transendothelial migration.26 has been reported in many autoimmune diseases, including Since these immunologic processes are essential elements rheumatoid arthritis, inflammatory bowel disease, lupus of GVHD, our findings may indicate that VAV1 is one of nephritis, Sjogren’s syndrome and myasthenia gravis.29–33 the key signaling molecules involved in the pathogenesis Similarity of the symptoms between the aforementioned of GVHD. diseases and chronic GVHD suggests that CXCL13 may be

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1055 responsible for inducing migration of B cells to the target it did not pass the Bonferroni correction, the p value of the tissues in chronic GVHD. While the roles of various B cell receptor signaling pathway was considerably low chemokines in leukocyte trafficking and migration of donor (P ¼ 1.30E-5). These findings and the overexpression of T cells to the target tissues of GVHD have been studied, the CXCL13 may indicate the involvement of the B cell roles of B cell chemokines have not been investigated in immune process in the pathogenesis of GVHD. depth with regard to the pathogenesis of GVHD. Although Interestingly, the global test showed that the cell cycle- its expression was not validated by the PCR experiment, associated pathways were differentially expressed in con- the role of CXCL13 in the humoral immune system as it cordance with the SAM results. These novel findings relates to the pathogenesis of chronic GVHD warrants appear to be related to the activation of T cell differentia- further study. tion in response to antigenic stimulation. In this study, Although the function identification of individual DEG many cell cycle-related genes and pathways were shown to provides information about biological processes involved in be differentially expressed in chronic GVHD specimens, the pathogenesis of GVHD, the molecular signatures were especially those associated with G1 to S progression. The analyzed using gene ontology enrichment test in an attempt CDK6gene, which regulates cell cycle progression from to gain a better understanding. Among significant terms, the G1 phase to the S phase, was highly overexpressed in phosphorous and phosphate metabolism were the bio- GVHD samples (score ¼ 4.29, fold change ¼ 4.86). The logical processes term found to have the largest number association between CDK6and the G1/S transition of of annotated genes. Since the phosphorus-related metabolic T lymphocytes was reported in a previous study.37 When process is connected with many signaling processes, the the global test was conducted, the ‘Cyclins and cell cycle DEGs mapped to the phosphorus-related metabolic regulation’, ‘Cell Cycle: G1/S Check Point’ and ‘Influence process terms included diverse signaling processes mem- of Ras and Rho proteins on G1 to S transition’ pathways bers, such as cyclin dependent kinase 6(CDK6),p21 were also differentially expressed in chronic GVHD activated kinase 6and testis specific kinase 6.Although the patients. All of these pathways are associated with cell meaning of these results as they relate to chronic GVHD is cycle progression from the G1 phase to the S phase, and the not clear, these terms seem to be the consequence of cell ‘IL-2 signaling’ pathway, which promotes the G1-to-S cycle and immunologic processes.22,34,35 Additionally, transition38 was also differentially expressed in the global the molecular terms were consistent with the biological test. Based on these findings, we suggest that cell cycle process terms and included many kinase and transferase progression from the G1 phase to the S phase may be a activity terms. critical step for activation of immune related cells in the In addition to the functional annotation of DEGs, we pathogenesis of chronic GVHD. analyzed pathway-wise gene expression using a global test The ‘S1P_Signaling’ and ‘Rac 1 cell motility signaling’ to clarify the molecular characteristics underlying the pathways play a role in cell movement, and are associated pathogenesis of chronic GVHD. The global test using with the recruitment of immune cells to target organs.39 the pathway information can identify the biological path- In addition, myocyte-associated pathways are linked with ways which are differentially expressed. Moreover, MEF2 (myocyte enhancer factor 2), which mediates the group-wise testing can identify differentially expressed apoptosis of T cells.40 These pathways may be another pathways, which are not readily identified by single targets for the management of GVHD because gene-wise analysis.19 the perturbation of thymic T cell apoptosis is one of the The results of the global test indicated that many T cell- characteristic features of GVHD.41 related pathways were significantly expressed in patient In this study, we used the peripheral blood sample samples, including the T cell receptor signaling pathway, without consideration of proportion of lymphocyte sub- the T helper cell surface molecule pathway, the Ras types (T, B cell and monocyte etc.). We had tried to sample independent pathway involved in NK cell mediated the subtypes of lymphocyte separately, but we found that it cytotoxicity, and the IL-2 signaling pathway (Table 3). It required larger amount of blood sample and more cost. is well known that donor-derived alloreactive T cells play a Furthermore, it is possible that the separation procedure of major role in the pathogenesis of chronic GVHD36 and our lymphocyte subtypes increases experimental bias. Besides, results also documented the activation T cell-mediated although the proportion of lymphocyte subtypes was not immune processes in chronic GVHD development. Prior measured explicitly in this analysis, it seems to have little studies have reported the following sequence of biological effect on the result of DEGs and differentially expressed phenomena in chronic GVHD: The donor T cell receptor pathway analysis because maker genes of each subtype of recognizes the host MHC antigen and is then activated and lymphocyte were not differentially expressed (See Supple- differentiated into a T helper cell via the T cell receptor mentary Table 3). Most of the marker genes were not in our signaling pathway. Donor T helper cells are then activated DEG list, except CD3D. These findings indicate that the upon recognition of the alloantigen, and in turn secrete DEGs and differentially expressed pathways did not result IL-2 and IFN-g, which recruit other T cells, cytotoxic from difference of proportion of lymphocyte subtypes T lymphocytes, natural killer (NK) cells and monocytes. between GVHD and non-GVHD patients. To validate our The elements of this process sequence could be identified in speculation, we performed linear regression analysis the results of the global test. between each of 125 DEGs and the marker gene set It is interesting to note that the Fc Epsilon Receptor I expressions. Significant result implies that the DEG’s Signaling in Mast Cells pathway, which is involved in the B expression level can be explained by the maker genes’ cell immune process, was differentially expressed. Although expression level, which means proportion of lymphocyte

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1056 affects the DEG’s expression. All the DEGs’ expression 9 Sullivan KM, Agura E, Anasetti C, Appelbaum F, Badger C, could not be explained by the maker gene expression levels Bearman S et al. Chronic graft-versus-host disease and other (Supplementary Table 4). Based on these findings, we late complications of bone marrow transplantation. Semin concluded that our analysis result was independent on the Hematol 1991; 28: 250–259. proportion of lymphocyte subtypes. To find which lym- 10 Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, et al. phocyte subtype has a role in expression of the DEGs is Mesirov JP Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. main goal of our next research. Science 1999; 286: 531–537. The classifier gene set comprised of 17 genes which were 11 Segal E, Friedman N, Koller D, Regev A. A module map selected using the PAM algorithm showed a favorable showing conditional activity of expression modules in cancer. classification performance. The results indicate that chronic Nat Genet 2004; 36: 1090–1098. GVHD can be diagnosed using gene expression values of 12 Segal E, Friedman N, Kaminski N, Regev A, Koller D. these classifier genes in peripheral blood mononuclear cells. From signatures to models: understanding cancer using The value of these classifier genes for the early diagnosis microarrays. Nat Genet 2005; 37 (Suppl): S38–S45. and management of chronic GVHD needs to be evaluated 13 Huber W, von Heydebreck A, Sultmann H, Poustka A, in a larger number of patients, preferably in a prospective Vingron M. Variance stabilization applied to microarray data study. calibration and to the quantification of differential expression. Bioinformatics 2002; 18 (Suppl 1): S96–S104. In conclusion, we have identified molecular character- 14 Bolstad BM, Irizarry RA, Astrand M, Speed TP. istics and diagnostic markers for chronic GVHD using an A comparison of normalization methods for high density oligonucleotide microarray. Many immune and cell cycle oligonucleotide array data based on variance and bias. related genes and pathways were differentially expressed in Bioinformatics 2003; 19: 185–193. chronic GVHD patients. These results appear to indicate 15 Troyanskaya O, Cantor M, Sherlock G, Brown P, Hastie T, that alloreactive T cell activation through the T cell Tibshirani R et al. Missing value estimation methods for DNA receptor signaling pathway and cell cycle progression are microarrays. Bioinformatics 2001; 17: 520–525. the main mechanisms responsible for chronic GVHD 16Chung HJ, Park CH, Han MR, Lee S, Ohn JH, Kim J et al. induction and maintenance. Our results also indicated ArrayXPath II: mapping and visualizing microarray gene- the possibility of B cell involvement in the manifestation of expression data with biomedical ontologies and integrated biological pathway resources using Scalable Vector Graphics. the autoimmune disorder-like symptoms associated with Nucleic Acids Res 2005; 33: W621–W626. chronic GVHD. 17 Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 2001; 98: 5116–5121. Acknowledgements 18 Dennis Jr G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC et al. DAVID: Database for annotation, visualization, and integrated discovery. Genome Biol 2003; 4 (5): P3. This study was supported by a grant from the Korea Health 21 19 Goeman JJ, van de Geer SA, de Kort F, van Houwelingen HC. R&D Project, Ministry of Health & Welfare, Republic of Korea A global test for groups of genes: testing association with a (0405-BC02-0604-0004). clinical outcome. Bioinformatics 2004; 20: 93–99. 20 Tibshirani R, Hastie T, Narasimhan B, Chu G. Diagnosis of multiple cancer types by shrunken centroids of gene expres- References sion. Proc Natl Acad Sci USA 2002; 99: 6567–6572. 21 Socie´ G, Mary JY, Lemann M, Daneshpouy M, Guardiola P, 1 Lee SJ. New approaches for preventing and treating chronic Meignin V et al. Prognostic value of apoptotic cells and graft-versus-host disease. Blood 2005; 105: 4200–4206. infiltrating neutrophils in graft-versus-host disease of the 2 Baird K, Pavletic SZ. Chronic graft versus host disease. Curr gastrointestinal tract in humans: TNF and Fas expression. Opin Hematol 2006; 13: 426–435. Blood 2004; 103: 50–57. 3 Socie´ G, Stone JV, Wingard JR, Weisdorf D, Henslee-Downey PJ, 22 Cailhier JF, Laplante P, Hebert MJ. Endothelial apoptosis and Bredeson C et al. Long-term survival and late deaths after chronic transplant vasculopathy: recent results, novel mecha- allogeneic bone marrow transplantation. Late effects working nisms. Am J Transplant 2006; 6: 247–253. committee of the international bone marrow transplant 23 Farag SS. Chronic graft-versus-host disease: where do we go registry. N Engl J Med 1999; 341: 14–21. from here? Bone Marrow Transplant 2004; 33: 569–577. 4 Lee SJ, Vogelsang G, Flowers ME. Chronic graft-versus-host 24 Rouquette-Gally AM, Boyeldieu D, Prost AC, Gluckman E. disease. Biol Blood Marrow Transplant 2003; 9: 215–233. Autoimmunity after allogeneic bone marrow transplantation. 5 Hess AD. Modulation of graft-versus-host disease: role of A study of 53 long-term-surviving patients. Transplantation regulatory T lymphocytes. Biol Blood Marrow Transplant 2006; 1988; 46: 238–240. 12: 13–21. 25 Lister J, Messner H, Keystone E, Miller R, Fritzler MJ. 6Hess AD, Thoburn CJ. Immune tolerance to self-major Autoantibody analysis of patients with graft versus host histocompatibility complex class II antigens after bone marrow disease. J Clin Lab Immunol 1987; 24: 19–23. transplantation: role of regulatory T cells. Biol Blood Marrow 26Canninga-van Dijk MR, van der Straaten HM, Fijnheer R, Transplant 2006; 12: 518–529. Sanders CJ, van den Tweel JG, Verdonck LF. Anti-CD20 7 Ferrara JL, Deeg HJ. Graft-versus-host disease. N Engl J Med monoclonal antibody treatment in 6patients with therapy- 1991; 324: 667–674. refractory chronic graft-versus-host disease. Blood 2004; 104: 8 Kier P, Penner E, Bakos S, Kalhs P, Lechner K, Volc-Platzer B 2603–2606. et al. Autoantibodies in chronic GVHD: high prevalence 27 Fernandez-Malave E, Wang N, Pulgar M, Schamel WW, of antinucleolar antibodies. Bone Marrow Transplant 1990; 6: Alarcon B, Terhorst C. Overlapping functions of human 93–96. CD3delta and mouse CD3gamma in alphabeta T-cell

Bone Marrow Transplantation Cell cycle and immune alteration in chronic GVHD SJ Oh et al 1057 development revealed in a humanized CD3gamma-mouse. the role of CC chemokine ligand 21 in thymic hyperplasia. Blood 2006; 108: 3420–3427. J Immunol 2006; 177: 7868–7879. 28 Perez-Simon JA, Sanchez-Abarca I, Diez-Campelo M, 34 Zheng Y, Kong Y, Goetzl EJ. Lysophosphatidic acid receptor- Caballero D, San Miguel J. Chronic graft-versus-host disease: selective effects on Jurkat T cell migration through a Matrigel Pathogenesis and clinical management. Drugs 2006; 66: model basement membrane. J Immunol 2001; 166: 2317–2322. 1041–1057. 35 Gizard F, Robillard R, Barbier O, Quatannens B, Faucompre´ 29 Ishikawa S, Sato T, Abe M, Nagai S, Onai N, Yoneyama H A, Re´ villion F et al. TReP-132 controls cell proliferation by et al. Aberrant high expression of B lymphocyte chemokine regulating the expression of the cyclin-dependent kinase (BLC/CXCL13) by C11b+CD11c+ dendritic cells in murine inhibitors p21WAF1/Cip1 and p27Kip1. Mol Cell Biol 2005; lupus and preferential chemotaxis of B1 cells towards BLC. 25: 4335–4348. J Exp Med 2001; 193: 1393–1402. 36Gilman AL, Serody J. Diagnosis and treatment of chronic 30 Shi K, Hayashida K, Kaneko M, Hashimoto J, Tomita T, graft-versus-host disease. Semin Hematol 2006; 43: 70–80. Lipsky PE et al. Lymphoid chemokine B cell-attracting 37 Arbogast A, Boutet S, Phelouzat MA, Plastre O, Quadri R, chemokine-1 (CXCL13) is expressed in germinal center of Proust JJ. Failure of T lymphocytes from elderly humans to enter ectopic lymphoid follicles within the synovium of chronic the cell cycle is associated with low Cdk6activity and impaired arthritis patients. J Immunol 2001; 166: 650–655. phosphorylation of Rb protein. Cell Immunol 1999; 197: 46–54. 31 Abad C, Juarranz Y, Martinez C, Arranz A, Rosignoli F, 38 Rudd CE. Cell cycle ‘check points’ T cell anergy. Nat Immunol Garcı´ a-Go´ mez M et al. cDNA array analysis of cytokines, 2006; 7: 1130–1132. chemokines, and receptors involved in the development of 39 Halin C, Scimone ML, Bonasio R, Gauguet JM, Mempel TR, TNBS-induced colitis: homeostatic role of VIP. Inflamm Bowel Quackenbush E et al. The S1P-analog FTY720 differentially Dis 2005; 11: 674–684. modulates T-cell homing via HEV: T-cell-expressed S1P1 32 Hjelmervik TO, Petersen K, Jonassen I, Jonsson R, amplifies integrin activation in peripheral lymph nodes but not Bolstad AI. Gene expression profiling of minor salivary glands in Peyer patches. Blood 2005; 106: 1314–1322. clearly distinguishes primary Sjogren’s syndrome patients from 40 Youn HD, Chatila TA, Liu JO. Integration of calcineurin and healthy control subjects. Arthritis Rheum 2005; 52: 1534–1544. MEF2 signals by the coactivator p300 during T-cell apoptosis. 33 Le Panse R, Cizeron-Clairac G, Bismuth J, Berrih-Aknin S. EMBO J 2000; 19: 4323–4331. Microarrays reveal distinct gene signatures in the thymus of 41 Kansu E. The pathophysiology of chronic graft-versus-host seropositive and seronegative myasthenia gravis patients and disease. Int J Hematol 2004; 79: 209–215.

Supplementary Information accompanies the paper on Bone Marrow Transplantation website (http://www.nature.com/bmt)

Bone Marrow Transplantation