DNA microarray analysis of natural killer cell-type lymphoproliferative disease of granular lymphocytes with purified CD3CD56+ fractions


Natural killer (NK) cell-type lymphoproliferative disease of granular lymphocytes (LDGL) is characterized by the outgrowth of CD3CD16/56+ NK cells, and can be further subdivided into two distinct categories: aggressive NK cell leukemia (ANKL) and chronic NK lymphocytosis (CNKL). To gain insights into the pathophysiology of NK cell-type LDGL, we here purified CD3CD56+ fractions from healthy individuals (n=9) and those with CNKL (n=9) or ANKL (n=1), and compared the expression profiles of >12 000 genes. A total of 15 ‘LDGL-associated genes’ were identified, and a correspondence analysis on such genes could clearly indicate that LDGL samples share a ‘molecular signature’ distinct from that of normal NK cells. With a newly invented class prediction algorithm, ‘weighted distance method’, all 19 samples received a clinically matched diagnosis, and, furthermore, a detailed cross-validation trial for the prediction of normal or CNKL status could achieve a high accuracy (77.8%). By applying another statistical approach, we could extract other sets of genes, expression of which was specific to either normal or LDGL NK cells. Together with sophisticated statistical methods, gene expression profiling of a background-matched NK cell fraction thus provides us a wealth of information for the LDGL condition.


Lymphoid cells (10–15%) in peripheral blood (PB) are characterized by the presence of multiple azurophilic granules in pale blue cytoplasm, referred to as large granular lymphocytes (LGLs). Such LGLs originate either from CD3+ T cells or CD3CD16/56+ natural killer (NK) cells,1 and sustained outgrowth of LGLs has been designated as lymphoproliferative disease of granular lymphocytes (LDGL),2 granular lymphocyte-proliferative disorders (GLPD)3 or LGL leukemia (LGLL).4

NK cell-type LDGL can be further subdivided into two distinct categories, that is, aggressive NK cell leukemia (ANKL) and chronic NK lymphocytosis (CNKL).5 The former is a clonal disorder of NK cells with a very poor outcome. Mono- or oligoclonal Epstein–Barr virus (EBV) genome can be frequently found in an episomal position in these NK cells,6 suggesting a pathogenetic role of EBV in this disorder. The leukemic NK cells are often refractory to chemotherapeutic reagents, and multiple organ failure is common to ANKL patients.

In contrast, a chronic, indolent course is characteristic to CNKL. Individuals with CNKL are often symptom-free with infrequent fever, arthralgias, and cytopenia, and their NK cells are rarely positive for EBV genome.7,8 Although the clonality of CNKL cells is still obscure partly due to the limited availability of assessment procedures, one study with X chromosome-linked gene analysis failed to detect clonality in the affected NK cells,9 suggesting a reactive, rather than a neoplastic, nature of CNKL condition. This hypothesis is further supported by the fact that splenectomy can lead to a sustained elevation of PB NK cell count in vivo.10 However, the hypothesis for the reactive nature of CNKL may be challenged by the fact that some CNKL patients were proved to have a clonal proliferation in NK cells and/or to undergo transformation into NK cell leukemia/lymphoma.11,12

Making issues further complicated, the diagnostic criteria for CNKL are not clearly settled yet. Previous reports have proposed the requirement of sustained (>6 months) outgrowth of NK cells in PB (>2.0 × 109 or >0.6 × 109/l)2,8 for the diagnosis of CNKL. However, NK cell count in the PB of CNKL individuals may fluctuate, and does not always fulfill the criteria. Morice et al13 have reported that affected NK cells may have a restricted expression of a single isoform of killing inhibitory receptors (KIRs), supporting the usefulness of KIR expression as a clonality marker of NK cells.13 However, these findings yet provide little information for the nature of affected NK cells in the CNKL condition.

DNA microarray enables us to measure the expression level for thousands of genes simultaneously,14,15 and would be a promising tool to shed light from a new direction on the pathophysiology as well as diagnostic system for LDGL. Gene expression profiling with microarray has, for instance, succeeded in the differential diagnosis between acute myeloid leukemia (AML) and acute lymphoid leukemia (ALL), in extracting novel prognostic markers for prostate cancer,16 and in the identification of molecular markers for myelodysplastic syndrome (MDS)17 or chronic myeloid leukemia (CML).18

However, simple comparison of tissues or specimens may only yield pseudopositive and pseudonegative data. Although NK cells occupy 10–15% of PB mononuclear cells (MNCs) in healthy individuals, 80–90% of MNCs may be composed of affected NK cells in CNKL patients. If PB MNCs are simply compared between these two groups, any genes specific to NK cells would be considered to be activated in the latter. This misleading result may not reflect any changes in the amount of mRNA per NK cell. To minimize such pseudopositive/pseudonegative data, background-matched NK cell fractions should be purified from healthy individuals as well as LDGL patients prior to microarray analysis. Such approach, referred to as ‘background-matched population (BAMP) screening’,17 should pinpoint the gene expression alterations truly specific to each condition.

The efficacy of BAMP screening has been already demonstrated by Makishima et al19 in the analysis of CD4CD8+ T-cell type LDGL. CD4CD8+ fractions were purified from PB MNCs of such LDGL patients and age-matched healthy volunteers, and were subjected to microarray analysis, resulting in the identification of novel molecular markers for T cell-type LDGL.

Analogously, here we isolated CD3CD56+ NK cell fractions from healthy volunteers (n=9) as well as individuals with CNKL (n=9) or with ANKL (n=1). By using high-density oligonucleotide microarray, expression profiles for >12 000 human genes were obtained for these purified NK cell specimens. Analysis of the data set with sophisticated statistical methods has clarified that the affected NK cells are clearly distinct from normal ones, at least, with regard to transcriptome.

Materials and methods

Purification of CD3CD56+ cells

PB MNCs were isolated by Ficoll–Hypaque density gradient centrifugation from the subjects with informed consent. The cells were incubated with anti-CD3 MicroBeads (Miltenyi Biotec, Auburn, CA, USA), and loaded onto MIDI-MACS magnetic cell separation columns (Miltenyi Biotec) to remove CD3+ cells. The flow-through was then mixed with anti-CD56 MicroBeads (Miltenyi Biotec), and was subjected to a MINI-MACS column for the ‘positive selection’ of CD56+ cells. Cells bound specifically to the column were then eluted according to the manufacturer's instructions, and stored in aliquots at −80°C.

Enrichment of CD3CD56+ NK cell fraction was confirmed in every specimen by subjecting portions of the MNC and column eluates to staining with Wright–Giemsa solution and to the analysis of the cell surface expression of CD3 and CD56 by flow cytometry (FACScan; Becton Dickinson, Mountain View, CA, USA). The CD3CD56+ fraction was shown to constitute >90% of each eluate of the affinity column.

DNA microarray analysis

Total RNA was extracted from the CD3CD56+ cell preparations by the acid guanidinium method, and was subjected to two rounds of amplification with T7 RNA polymerase as described.20 High fidelity of our RNA amplification procedure has been already reported.18 The amplified cRNA (2 μg) was then converted to double-stranded cDNA, which was used to prepare biotin-labeled cRNA for hybridization with GeneChip HGU95Av2 microarrays (Affymetrix, Santa Clara, CA, USA) harboring oligonucleotides corresponding to a total of 12 625 genes. Hybridization, washing, and detection of signals on the arrays were performed with the GeneChip system (Affymetrix).

Class prediction by the ‘weighted distance method’

The fluorescence intensity for each gene was normalized relative to the median fluorescence value of all human genes on the array in each hybridization. Hierarchical clustering of the data set and isolation of genes specific to the NK cells from healthy individuals (Normal) or to those of patients with LDGL were performed with GeneSpring 5.1 software (Silicon Genetics, Redwood, CA, USA).

In the comparison of normal- and LDGL-CD3CD56+ cells, t statistic and effect size (difference in the mean of expression level between normal and LDGL classes)16 were calculated for each gene. When a gene showed t>3.966 (corresponding to a significance level of 0.001 in t-test with 17 degrees of freedom) and effect size>3, the difference in expression level between two classes was considered statistically significant. The genes showing the significant differences were called as ‘informative genes’ in this study. Correspondence analysis21 was then performed with ViSta software (http://www.visualstats.org) for all genes showing a significant difference. Each sample was plotted in three dimensions, based on the coordinates obtained from the correspondence analysis.

To examine whether the informative genes were able to predict the class of the present specimens, we performed class prediction with our ‘weighted distance method’ (RO et al, submitted). This prediction method utilizes the dimensions obtained from correspondence analysis for the informative genes.

Consider a sample X to be predicted from N samples (excluding sample X) in the data set (NA from class A and NB from class B). Each sample can be represented by three dimensions, d1, d2, and d3, where di denotes the coordinate in the ith dimension for the sample. The weighted distance from sample X to sample Y is defined as

, where vi indicates the contribution of the ith dimension from correspondence analysis, and diX and diY represent di for sample X and sample Y, respectively. Let DA be the mean value of D from sample X to NA samples belonging to class A, and DB be the mean value of D from sample X to NB samples belonging to class B. When DA/(DA+DB)<T, the sample X is assigned class A, and when DB/(DA+DB)<T, the sample X is assigned class B, where T is a threshold value. In our analysis, the T value was set to be 0.4. It should be noted that the weighted distance method could be applied to more than two classes.

In a cross-validation trial for the prediction of normal or CNKL class, the entire prediction process mentioned above was repeated for the 18 samples (nine for normal and nine for CNKL). To predict the class of every sample X, the correspondence analysis was carried out for the informative genes obtained from the remaining 17 samples. In this case, the informative genes were selected with a criteria of t>4.073 (corresponding to a significance level of 0.001 in t-test with 15 degrees of freedom) and effect size>3.

All raw array data as well as details of the genes shown in the figures are available as supplementary information at the Leukemia web site.

‘Real-time’ reverse transcription-polymerase chain reaction (RT-PCR) analysis

Portions of nonamplified cDNA were subjected to PCR with a QuantiTect SYBR Green PCR Kit (Qiagen, Valencia, CA, USA). The amplification protocol comprised incubations at 94°C for 15 s, 60°C for 30 s, and 72°C for 60 s. Incorporation of the SYBR Green dye into PCR products was monitored in real time with an ABI PRISM 7700 sequence detection system (PE Applied Biosystems, Foster City, CA, USA), thereby allowing determination of the threshold cycle (CT) at which exponential amplification of PCR products begins. The CT values for cDNAs corresponding to the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and interferon-γ (IFNG; GenBank accession number, X13274) genes were used to calculate the abundance of IFNG mRNA relative to that of GAPDH mRNA. The oligonucleotide primers for PCR were 5′-IndexTermGTCAGTGGTGGACCTGACCT-3′ and 5′-IndexTermTGAGCTTGACAAAGTGGTCG-3′ for GAPDH, and 5′-IndexTermGGGCCAACTAGGCAGCCAACTAA-3′ and 5′-IndexTermGGAAGCACCAGGCATGAAATCTCC-3′ for IFNG cDNA.

Determination of serum level of IFNG protein

Sera were obtained from healthy volunteers and individuals with aplastic anemia (AA), systemic lupus erythematosus (SLE), virus infection-associated hemophagocytic syndrome (VAH), LDGL of αβ+ T cell, LDGL of γδ+ T cell, infectious mononucleosis (IMN), CNKL, or ANKL. The serum concentration of IFNG was determined by a flow cytometer with Human Th1/Th2 Cytokine Cytometric Bead Array Kit (BD Biosciences, San Diego, CA, USA) according to the manufacturer's protocols.


Purification of NK cells

To directly compare the transcriptome of normal and affected NK cells, we here purified CD3CD56+ fractions from PB MNCs of healthy volunteers (n=9) as well as of individuals with CNKL (n=9) or ANKL (n=1). A total of 19 specimens were thus registered into this study. The clinical characteristics of the 10 patients (CNKL-19 and ANKL-1) are summarized in Table 1. The LGL count in their PB was 14 056 cells/ml±11 695 (mean±s.d.). The proportion of CD56+ cells in PB MNC was >50% in all affected individuals, indicating a predominant outgrowth of NK cells. All CD56+ fractions in this study were negative for the surface expression of CD3.

Table 1 Clinical characteristics of the patients with NK cell-type LDGL

The mono- or oligoclonal expansion with regard to EBV infection was confirmed in the NK cells of one CNKL (CNKL-9) and the ANKL (ANKL-1) patients. Importantly, the CNKL-9 patient with monoclonal expansion of EBV+ NK cells died from leukemic transformation with infiltration into multiple organs at 24 months after the blood sampling. It is, therefore, likely that this patient might have been under a transition process toward ANKL or been at a very early stage of ANKL.

Magnetic bead-based affinity column has succeeded in a substantial enrichment of the NK cell fraction. In one healthy volunteer, for instance, PB MNCs was occupied with 12.3% of CD3CD56+ fraction, while the column eluent contained 96.0% of those cells (Figure 1, upper panel). Similar purity of CD3CD56+ fraction was also obtained for the patients with CNKL, as demonstrated in the lower panel. The purified cells exhibited a homogenous phenotype of LGL (Figure 1). Successful enrichment of NK cells (>90% purity) was confirmed in every case by flow cytometry and Wright–Giemsa staining of cytospin preparations (not shown). Cell number of the CD3CD56+ fractions obtained in each individual was 3.9 × 105±3.4 × 105 (mean±s.d.).

Figure 1

Purification of CD3CD56+ fraction. MNCs isolated from PB of a healthy volunteer (normal) and a patient with CNKL were used to purify CD3CD56+ fractions (Eluent). Cell surface expression of CD3 and CD56 was monitored in each fraction by flow cytometry, and the proportion (%) of CD3CD56+ cells is indicated. Cytospin preparation of each fraction was stained with the Wright–Giemsa solutions. Scale bar, 50 μm.

BAMP screening of NK cell fractions

Biotin-labeled cRNA was prepared from surface marker-matched NK fractions from the study subjects, and was hybridized with high-density oligonucleotide microarrays (Affymetrix HGU95Av2), providing the expression data for >12 000 human genes. To exclude genes that were virtually silent transcriptionally, we first selected genes whose expression received the ‘Present’ call from the Microarray Suite 4.0 software (Affymetrix) in at least 10% of the samples. A total of 6494 genes passed this ‘selection window,’ and their expression profiles in the 19 samples are shown in Figure 2a as a dendrogram, or ‘gene tree,’ in which genes with similar expression profiles (assessed by standard correlation) among the samples were clustered near each other. In Figure 2a, several clusters of genes that were expressed preferentially in either normal or affected NK cells (shown by arrows) were identified.

Figure 2

Expression profiles of 6494 genes in NK cell fractions. (a) Hierarchical clustering of 6494 genes on the basis of their expression profiles in CD3CD56+ fractions derived from nine healthy volunteers (Normal), nine individuals with CNKL, and one with ANKL. Each column represents a single gene on the microarray, and each row a separate patient sample. Expression level of each gene is shown color-coded, as indicated by the scale at the bottom. Arrows indicate the positions of clusters of genes that were expressed preferentially in either normal or affected NK cells. (b) Two-way clustering analysis of the healthy individuals (Normal-1–9), CNKL patients (CNKL-1–9), and the ANKL patient (ANKL-1), based on the similarities in the expression profiles of the 6494 genes demonstrated in (a).

To statistically evaluate the similarity of the overall gene expression profiles across the 19 samples, we generated another dendrogram, a ‘patient tree,’ by the two-way clustering method,22 with a separation ratio of 0.5 (Figure 2b). The samples did not clearly cluster into disease-specific branches; rather, normal and affected NK samples were mixed in several branches.

Identification of LDGL-associated genes

One of the major goals in this study was to develop expression profile-based diagnostic procedures for the NK cell disorders. For such an approach to be meaningful, an important question to be addressed would be thus to clarify whether affected NK cells share a specific gene expression profile, or ‘molecular signature’,23 clearly distinct from that of normal NK cells.

Therefore, we first tried to identify genes whose expression level may efficiently differentiate normal NK cells from LDGL ones. For this purpose, we chose genes whose expression level differed significantly between the two groups of samples (Student's t-test, P<0.001). However, most of the genes thus identified had a low level of expression throughout all samples, making their usefulness as molecular markers uncertain. From these genes, therefore, we selected those whose mean expression intensity differed by 3.0 arbitrary units (U) between the two groups. The resultant 15 ‘LDGL-associated genes’ are shown in a gene tree format (Figure 3a); five of them were specific to CNKL/ANKL cells, while the remaining 10 genes were to normal NK cells.

Figure 3

Identification of LDGL-associated genes. (a) Expression profiles of 15 LDGL-associated genes are shown in a dendrogram, color-coded as indicated by the scale in Figure 2a. Each row corresponds to a single gene, and each column to NK cells from healthy individuals (normal) and patients with CNKL or ANKL. The gene symbols are indicated on the right. The names, accession numbers, and expression intensity data for these genes are available at the Leukemia web site. (b) Two-way clustering analysis of the 19 samples based on the expression levels of the LDGL-associated genes. (c) Correspondence analysis of the LDGL-associated genes identified three major dimensions in their expression profiles. Projection of the specimens into a virtual space with these three dimensions revealed that the specimens from healthy individuals (normal) were clearly separated from those from the patients with CNKL or ANKL. The position of EBV+ CNKL-9 sample is indicated.

The former group of the genes included those for transcriptional factors involved in the regulation of cell growth and/or apoptosis. B lymphoma Mo-MLV insertion region (BMI1; GenBank accession number, L13689), for example, belongs to the Polycomb type of DNA-binding proteins.24 Intriguingly, BMI1 is expressed in hematopoietic stem cells (HSCs), and plays an indispensable role in the self-renewal process of HSCs.25,26 Therefore, abundant expression of BMI1 gene only in the affected NK cells may be involved in the deregulated outgrowth of the NK cells. Similarly, a zinc-finger protein ZFR (GenBank accession number, AI743507) was shown to protect embryonic cells from apoptosis and provide mitotic activity.27

Class prediction by our ‘weighted distance method’

We next performed two-way clustering analysis, with a separation ratio of 0.5,. of the 19 specimens based on the expression levels of such 15 LDGL-associated genes. As shown in Figure 3b, the samples clustered into three major branches; one contains mainly normal NK specimens (with an addition of CNKL-1), another is composed solely of two CNKL patients (CNKL-2 and -7), and the other contains only affected NK cells. It should be noted that the ANKL sample was clustered closely with CNKL ones in the third branch.

Do the gene-expression profiles of NK cells differ between healthy individuals and those with NK cell disorders, and, if so, how different? Is such difference large enough to develop an expression profile-based diagnosis system? To address these issues, we performed correspondence analysis21 to extract three major dimensions from the expression patterns of the 15 LDGL-associated genes. On the basis of the calculated three-dimensional coordinates for each sample, the specimens were then projected into a virtual space (Figure 3c). All normal samples were placed at a position clearly far from that of the affected NK cells, indicating that all affected NK cells possessed a common molecular signature which was distinct from that of the normal NK cells. Again, here the two samples with clonal EBV infection (CNKL-9 and ANKL-1) were placed closely with the other CNKL specimens.

The clear separation of affected NK specimens from normal ones in Figure 3c also supported the feasibility of an expression profile-based prediction for NK cell disorders. We therefore tried class prediction (normal or LDGL) for each specimen on the basis of the coordinates calculated by the correspondence analysis. The relative ‘weighted distances’ of a given specimen to the normal or LDGL group (excluding the specimen for the prediction) were calculated, and the specimen was assigned a class when the relative distance to the class was <0.4. As demonstrated in Table 2, our weighted-distance method could correctly predict the class of every sample examined, making the array-based diagnostic procedure of NK cell-type LDGL into reality.

Table 2 Diagnosis by the ‘weighted-distance method’

Comparison of ‘Normal vs CNKL’ by the weighted-distance method

Given the large difference in the clinical course between CNKL and ANKL, there may be gene-expression alterations specific to the latter condition, which characterize its aggressive clinical course. Therefore, it might be appropriate to investigate these two conditions separately. We thus focused on the comparison between normal individuals and those with CNKL, and tried to assign, by the weighted-distance method, either normal or CNKL class to every specimen among nine healthy individuals and nine patients with CNKL.

To accurately measure the prediction power of our weighted-distance method, we conducted a cross-validation trial (i.e., ‘drop-one-out’ format) for the diagnosis of normal or CNKL class. To predict the class of sample X, ‘CNKL-associated genes’ were extracted from the comparison of remaining 17 samples according to the criteria used in Figure 3a (P<0.001 in Student's t-test, and effect size>3). The number of such CNKL-associated genes ranged from 4 to 10. Correspondence analysis was carried out for the expression profiles of the CNKL-associated genes, and was used to calculate the relative weighted distance of the ‘dropped’ sample X to either normal or CNKL class. As shown in Table 3, with a T-value of 0.4, a clinically matched prediction was obtained for 14 (77.8%) out of 18 cases, while one case (CNKL-2) was unpredictable and three cases (normal-6, normal-8 and CNKL-1) received a prediction incompatible with the clinical diagnosis. Therefore, even in a cross-validation assay, the weighted-distance method could achieve a high accuracy.

Table 3 Cross-validation of disease prediction

For comparison, we also conducted a cross-validation trial of class prediction by using a known prediction algorithm, the ‘k-nearest neighbor method’ (http://www.silicongenetics.com/Support/GeneSpring/GSnotes/class_prediction.pdf). Among the 18 samples tested, only 10 samples (55.6%) received correct prediction, indicating the superiority of our weighted distance method.

Since CNKL-9 patient had NK cells with EBV in a clonal episomal form, and had progressed into an ANKL phase in a relatively short period, we questioned if this patient had an atypical molecular signature for CNKL. To visualize the similarity of transcriptome of CNKL-9 sample with that of the other CNKL ones, the result of the cross-validation trial for CNKL-9 is demonstrated as a virtual-space format in Figure 4a. Correspondence analysis of nine genes that most efficiently differentiated normal-1–9 from CNKL-1–8 has identified three major dimensions in their expression pattern, and projection of the CNKL-9 patient together with the other samples in a 3D space indicated that CNKL-9 had an expression profile highly similar to that of the other CNKL subjects at least in the space of these nine highly informative genes.

Figure 4

Investigation of the EBV+ samples. (a) We could isolate nine genes, expression of which differentiated normal NK cells (normal) from indolent CNKL ones (CNKL-1–8). The EBV+ CNKL-9 was projected into a virtual space together with the other normal and CNKL specimens, based on the coordinates calculated by the correspondence analysis of such nine genes. (b) A total of seven genes were identified to be differentially expressed between normal NK cells (normal) and CNKL cells. The ANKL-1 sample was projected into the virtual space as in (a).

To confirm the similarity in the gene-expression profile of EBV+ ANKL cells to the CNKL ones, we next carried out correspondence analysis for the ANKL-1 patient. Statistical comparison of transcriptome between Normal-1–9 and CNKL-1–9 subjects identified a total of seven genes, which contrasted the expression profile of normal NK cells from that of CNKL NK cells. As shown in Figure 4b, projection of the ANKL-1 patient into a 3D space constructed from the data of such seven genes demonstrated that the EBV+ ANKL-1 sample was plotted at a neighbor position to those of the CNKL samples. In accordance with the 3D view, the weighted-distance method also concluded that the ANKL-1 sample belonged to the same class with the CNKL ones (data not shown). These analyses unexpectedly suggested that the gene expression profile characteristic to CNKL NK cells is also shared with EBV+ NK cells. It should be noted, however, that additional genetic changes specific to EBV infection may exist, and account for the fulminant clinical character of EBV+ LDGL.

Isolation of single-gene markers for LDGL diagnosis

The gene set identified in Figure 3a may potentially be the candidate genes to construct custom-made DNA microarrays for disease diagnosis of NK cell disorders. Since availability of DNA microarray systems is still restricted in current hospitals, however, it would be valuable if a high expression of single gene or its product can be used as a reliable marker for such purposes. For instance, it would be highly useful if the serum level of a protein can help to diagnose NK cell disorders. Given the presence of false data even with DNA microarray, it is unlikely that an expression of any single gene can correctly diagnose all samples. Therefore, here we have tried to isolate genes whose high expression may be ‘sufficient’ to predict the presence of NK cell-type LDGL, but the absence of its expression may not necessarily mean that the NK cells are normal. Such type of predictor genes should be strictly inactivated in all normal NK cells, but become activated in, at least, a part of the NK cells in the LDGL group.

To screen such type of predictors, first, the mean expression value of each gene was calculated for the normal or LDGL group. Then, with the use of GeneSpring software, we searched for genes whose expression profiles were statistically similar, with a minimum correlation of 0.95, to that of a hypothetical ‘LDGL-specific gene’ that exhibits a mean expression level of 0.0 U in the normal group and 100.0 U in the LDGL group. From such 652 genes identified, we then applied another criteria that gene-expression value should be (i) 60.0 U in, at least, one of the LDGL samples, and (ii) <25.0 U in all normal samples. A total of six genes were finally identified to be ‘LDGL-specific’ (Figure 5a). Here we have tried to extract LDGL-specific genes with minimum false-positive results, while allowing false-negative ones. Therefore, we should confidently tell that the given NK cells are of LDGL if one of the ‘LDGL-specific genes’ is highly expressed in the specimens.

Figure 5

Identification of single-gene markers for LDGL. (a) Dendrogram showing the expression profiles of six genes whose expression intensity was kept suppressed in normal NK cells, but became activated in a part of affected NK cells. Each row represents a single gene, and each column a separate patient sample. Expression level of the genes is shown colored according to the scale in Figure 2a. (b) Expression profiles of 22 ‘normal-specific genes’ are demonstrated as in (a). Two different oligonucleotide sets for the alphaE-catenin (CTNNA1) gene are present on an HGU95Av2 array.

Conversely, we also tried to identify ‘normal-specific genes’ through the same approach. Firstly, a total of 1424 genes were identified to be statistically similar to a hypothesized ‘normal-specific gene’ that has a mean expression value of 100.0 U in the normal group, but of 0.0 U in the LDGL group. Among these genes, those whose expression was kept below 25.0 U throughout the samples in the LDGL group, but became activated at 60.0 U in, at least, one sample in the normal group were selected. We could thus extract a set of 22 genes, expression of which was specific to normal NK cells (Figure 5b).

Confirmation of overproduction of IFNG

NK cells become activated and produce IFNG in response to the stimulation with IL-2,28 IL-12,29 and IL-15.30 Under physiological circumstances, however, activated NK cells eventually undergo apoptotic changes to prevent overactivation of the immune system. Interestingly, IFNG itself provides a survival signal onto NK cells, and, therefore, sustained incubation with IFNG of NK cells protects efficiently them from cell death.31 It was thus provocative to find IFNG in our LDGL-specific genes (Figure 5a), indicating a potential role of IFNG in the outgrowth mechanism of NK cells in the LDGL condition.

Here we have confirmed the disease-specific expression of IFNG gene by a quantitative ‘real-time’ RT-PCR assay. As shown in Figure 6a, abundant expression of IFNG mRNA was only detected in the purified NK cell fraction of LDGL patients, but not of the normal controls. Furthermore, a high concentration of IFNG protein was also observed in the sera of CNKL/ANKL patients (Figure 6b), proving that overexpression of IFNG mRNA in NK cell disorders leads to the systemic elevation of IFNG protein level. Interestingly, overexpression of IFNG protein was also noticed in the patients with IMN. High expression of IFNG in IMN individuals may result from the infection of EBV associated with IMN, or may indicate the activated status of T or NK cells in the condition of IMN.

Figure 6

Confirmation of overexpression of IFNG. (a) Quantitation of IFNG mRNA in NK cell fractions. Complementary DNA was prepared from the NK cell fractions, and was subjected to real-time RT-PCR analysis with primers specific for the IFNG or GAPDH genes. The ratio of the abundance of IFNG mRNA to that of GAPDH mRNA was calculated as 2n, where n is the CT value for GAPDH cDNA minus the CT value for IFNG cDNA. (b) Sera were obtained from healthy volunteers (healthy) and individuals with aplastic anemia (AA), systemic lupus erythematosus (SLE), virus infection-associated hemophagocytic syndrome (VAH), LDGL of αβ+ T cell, LDGL of γδ+ T cell, infectious mononucleosis (IMN), CNKL, or ANKL. The expression level of IFNG protein in the sera was determined by flow cytometry, and shown as pg/ml.


In this manuscript, we tried to clarify whether gene-expression profiling can help to differentiate NK cells of LDGL individuals from those of healthy ones. Toward this goal, we first purified NK cell fractions from study subjects, which are characterized by the absence of cell surface CD3 molecule and the presence of CD56 antigen. Analysis with these isolated NK cells should be intrinsic for an accurate comparison of the disease status, since simple comparison with PB MNCs would be severely influenced by any changes in the cell composition of PB MNCs in each individual.

However, even by using expression profiles of the purified fractions, similarity of the expression pattern of all gene set failed to clearly separate the affected NK cells from normal ones (Figure 2b), indicating the necessity of a diagnostic system with a ‘supervised’ algorithm. For this aim, we first extracted gene clusters, expression of which was specific to either normal or affected NK cells. Correspondence analysis on the expression patterns of such ‘LDGL-associated genes’ has succeeded in the reduction of the number of pattern dimensions into three. To our surprise, projection of all samples into this 3D space clearly demonstrated that the affected NK cells (CNKL/ANKL) were placed clustered at a position separate from that of normal NK cells (Figure 3c). With coordinates in such decomposited dimensions, we then invented a novel class prediction means, ‘weighted-distance method’. As expected from the clear separation in the 3D view, the weighted-distance method provided correct prediction for all samples studied. A cross-validation trial for the disease diagnosis also gave a highly accurate prediction rate (77.8%).

Given a high incidence of clonal EBV infection in ANKL cells, EBV is believed to play an essential role in the pathogenesis of this fulminant disorder. From the point of view of gene-expression profile, the NK cells positive for EBV infection (CNKL-9 and ANKL-1) shared gene-expression patterns characteristic to other indolent CNKL cells. For instance, a comparison of normal NK cells and CNKL samples (except EBV+ CNKL-9) has identified a total of nine CNKL-related genes including those for nuclear matrix protein-2 (GenBank accession number, D50926), cytochrome C (D00265), tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, theta polypeptide or 14-3-3 protein tau (X56468), O-linked GlcNAc transferase (AL050366), IFNG, chemokine C–C motif receptor 1 (CCR1; D10925), chondroitin sulfate proteoglycan 2 (X15998), and fibrinogen-like 2 (AI432401). The 3D view of the correspondence analysis for these ‘CNKL-associated’ genes clearly demonstrated that the EBV+ CNKL-9 subject was included in the other indolent CNKL group (Figure 4a). Similarly, another EBV+ subject (ANKL-1) was placed closely with the CNKL samples, according to the expression profiles of the genes which differentiated CNKL cells from normal NK cells (Figure 4b).

These data propose a hypothesis that gene-expression alterations characteristic to an activated, yet indolent, proliferation of NK cells found in CNKL patients also take place in the highly proliferating NK cells in EBV+ individuals. In other words, a similar mechanism may be utilized for the sustained outgrowth of NK cells both in the individuals with CNKL and ANKL. It should be emphasized, however, that the number of EBV+ samples (n=2) in these analyses was too small to extract any conclusive remarks on the pathophysiology of proliferating EBV+ NK cells. Nevertheless, we believe that it was interesting to find that EBV+ NK cells may share a molecular signature with EBV LDGL cells.

We finally tried to identify single-gene markers, the presence or absence of which helps the diagnosis of LDGL. A total of 22 genes were shown to be specific to normal NK cells, including those for cyclin-dependent kinase inhibitor 1A (CDKN1A; GenBank accession number U03106) or CIP1, dual-specificity phosphatase 6 (DUSP6; AB013382), toll-like receptor 2 (TLR2; AF051152), tissue inhibitor of metalloproteinase 1 (TIMP1; D11139), and aldo-keto reductase family 1 member C3 (AKR1C3; D17793) (Figure 5b).

CDKN1A is transcriptionally regulated by the activity of p53, and functions as a major effector for antitumor activity of p53, through the suppression of cyclin-dependent kinase activities.32 Therefore, loss of expression of CDKN1A may allow uncontrolled transition at the G2–M boundary in the cell cycle, and may partially account for the overgrowth of NK cells in LDGL individuals. Similarly, DUSP6 antagonizes MAPK activities via dephosphorylation of the latter kinases.33 Decrease of DUSP6 expression may therefore contribute to overactivation of MAPK and to enhanced mitogenesis.

AKR1C3 catalyzes conversion of aldehydes and ketones to alcohols in vivo. Although its role in NK cells is unknown yet, downregulation of its transcription has been also reported in the LGLs of T cell-type LDGL.19 Comparison with DNA microarray of CD4CD8+ T-cells between healthy individuals and those with T cell-type LDGL has identified the AKR1C3 gene as the specific marker to the former. Decrease of AKR1C3 message was also confirmed by quantitative ‘real-time’ RT-PCR method in those patients. Transcriptional suppression of AKR1C3 was thus revealed in affected LGLs for both NK cell- and T cell-type LDGL, and may be a common marker for the diagnosis of LDGL condition.

A number of growth-promoting factors were found in the ‘LDGL-specific genes’ in Figure 5a. IFNG and BIRC334 are, for instance, known to protect NK cells from apoptosis, and IRF4 has an oncogenic activity in vivo.35 Additionally, CHD1 contains an SNF2-related helicase/ATPase domain, and is presumed to be involved in the regulation of chromatin structure and gene transcription as well.36

Our data, together with that by Mizuno et al31 suggest that the serum concentration of IFNG protein may be an indicator of NK cell-type LDGL. Direct production of IFNG by affected NK cells may also imply the presence of an autocrine loop for the NK cell growth.

The mechanism by which affected NK cells produce IFNG is still to be revealed. It is known that IL-2, IL-12, IL-15, and IL-18 all activate production of IFNG in NK cells. In our microarray data set, however, none of IL-2, IL-12, and IL-18 were found to be significantly expressed in the subjects (not shown). Although IL-15 was moderately expressed in our NK samples, its expression level did not differ between normal and affected NK cells. In support of this notion, we could not detect significant level of IL-2 protein in the examination of serum level of cytokines (not shown). Therefore, it is currently an open question as to whether activation of IFNG transcription in the affected NK cells is a secondary event from the stimulation by other cells such as T cells, or intracellular mechanism of IFNG expression is deregulated in the affected NK cells.


We have characterized the transcriptome of a relatively uncommon disorder, NK cell-type LDGL. Comparison of purified NK cells between healthy and CNKL individuals led to the identification of gene sets which are useful in the expression profile-based differential diagnosis of the disorder. Such disease-associated genes have also provided us insights into the molecular pathogenesis of NK cell-type LDGL. Together with further optimization of statistical methods, increase in the number of both genes and subjects for the analysis would help to define and clarify the clinical entities of NK cell disorders.


  1. 1

    Loughran Jr TP . Clonal diseases of large granular lymphocytes. Blood 1993; 82: 1–14.

  2. 2

    Semenzato G, Zambello R, Starkebaum G, Oshimi K, Loughran Jr TP . The lymphoproliferative disease of granular lymphocytes: updated criteria for diagnosis. Blood 1997; 89: 256–260.

  3. 3

    Oshimi K . Granular lymphocyte proliferative disorders: report of 12 cases and review of the literature. Leukemia 1988; 2: 617–627.

  4. 4

    Loughran Jr TP, Starkebaum G . Large granular lymphocyte leukemia. Report of 38 cases and review of the literature. Medicine (Baltimore) 1987; 66: 397–405.

  5. 5

    Jaffe ES, Harris NL, Stein H, Vardiman JW (eds). Pathology and Genetics of Tumours of Haematopoietic and Lymphoid Tissues. Lyon: IARC Press, 2001.

  6. 6

    Gelb AB, van de Rijn M, Regula Jr DP, Cornbleet JP, Kamel OW, Horoupian DS et al. Epstein–Barr virus-associated natural killer-large granular lymphocyte leukemia. Hum Pathol 1994; 25: 953–960.

  7. 7

    Tefferi A . Chronic natural killer cell lymphocytosis. Leuk Lymphoma 1996; 20: 245–248.

  8. 8

    Rabbani GR, Phyliky RL, Tefferi A . A long-term study of patients with chronic natural killer cell lymphocytosis. Br J Haematol 1999; 106: 960–966.

  9. 9

    Nash R, McSweeney P, Zambello R, Semenzato G, Loughran Jr TP . Clonal studies of CD3-lymphoproliferative disease of granular lymphocytes. Blood 1993; 81: 2363–2368.

  10. 10

    Garcia-Suarez J, Prieto A, Reyes E, Arribalzaga K, Perez-Machado MA, Lopez-Rubio M et al. Persistent lymphocytosis of natural killer cells in autoimmune thrombocytopenic purpura (ATP) patients after splenectomy. Br J Haematol 1995; 89: 653–655.

  11. 11

    Ohno Y, Amakawa R, Fukuhara S, Huang CR, Kamesaki H, Amano H et al. Acute transformation of chronic large granular lymphocyte leukemia associated with additional chromosome abnormality. Cancer 1989; 64: 63–67.

  12. 12

    Tefferi A, Greipp PR, Leibson PJ, Thibodeau SN . Demonstration of clonality, by X-linked DNA analysis, in chronic natural killer cell lymphocytosis and successful therapy with oral cyclophosphamide. Leukemia 1992; 6: 477–480.

  13. 13

    Morice WG, Kurtin PJ, Leibson PJ, Tefferi A, Hanson CA . Demonstration of aberrant T-cell and natural killer-cell antigen expression in all cases of granular lymphocytic leukaemia. Br J Haematol 2003; 120: 1026–1036.

  14. 14

    Cheung VG, Morley M, Aguilar F, Massimi A, Kucherlapati R, Childs G . Making and reading microarrays. Nat Genet 1999; 21: 15–19.

  15. 15

    Duggan DJ, Bittner M, Chen Y, Meltzer P, Trent JM . Expression profiling using cDNA microarrays. Nat Genet 1999; 21: 10–14.

  16. 16

    Dhanasekaran SM, Barrette TR, Ghosh D, Shah R, Varambally S, Kurachi K et al. Delineation of prognostic biomarkers in prostate cancer. Nature 2001; 412: 822–826.

  17. 17

    Miyazato A, Ueno S, Ohmine K, Ueda M, Yoshida K, Yamashita Y et al. Identification of myelodysplastic syndrome-specific genes by DNA microarray analysis with purified hematopoietic stem cell fraction. Blood 2001; 98: 422–427.

  18. 18

    Ohmine K, Ota J, Ueda M, Ueno S-i, Yoshida K, Yamashita Y et al. Characterization of stage progression in chronic myeloid leukemia by DNA microarray with purified hematopoietic stem cells. Oncogene 2001; 20: 8249–8257.

  19. 19

    Makishima H, Ishida F, Ito T, Kitano K, Ueno S, Ohmine K et al. DNA microarray analysis of T cell-type lymphoproliferative disease of granular lymphocytes. Br J Haematol 2002; 118: 462–469.

  20. 20

    Van Gelder RN, von Zastrow ME, Yool A, Dement WC, Barchas JD, Eberwine JH . Amplified RNA synthesized from limited quantities of heterogeneous cDNA. Proc Natl Acad Sci USA 1990; 87: 1663–1667.

  21. 21

    Fellenberg K, Hauser NC, Brors B, Neutzner A, Hoheisel JD, Vingron M . Correspondence analysis applied to microarray data. Proc Natl Acad Sci USA 2001; 98: 10781–10786.

  22. 22

    Alon U, Barkai N, Notterman DA, Gish K, Ybarra S, Mack D et al. Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Proc Natl Acad Sci USA 1999; 96: 6745–6750.

  23. 23

    Ramaswamy S, Ross KN, Lander ES, Golub TR . A molecular signature of metastasis in primary solid tumors. Nat Genet 2003; 33: 49–54.

  24. 24

    Alkema MJ, Wiegant J, Raap AK, Berns A, van Lohuizen M . Characterization and chromosomal localization of the human proto-oncogene BMI-1. Hum Mol Genet 1993; 2: 1597–1603.

  25. 25

    Park IK, Qian D, Kiel M, Becker MW, Pihalja M, Weissman IL et al. Bmi-1 is required for maintenance of adult self-renewing haematopoietic stem cells. Nature 2003; 423: 302–305.

  26. 26

    Lessard J, Sauvageau G . Bmi-1 determines the proliferative capacity of normal and leukaemic stem cells. Nature 2003; 423: 255–260.

  27. 27

    Meagher MJ, Braun RE . Requirement for the murine zinc finger protein ZFR in perigastrulation growth and survival. Mol Cell Biol 2001; 21: 2880–2890.

  28. 28

    Thornton S, Kuhn KA, Finkelman FD, Hirsch R . NK cells secrete high levels of IFN-gamma in response to in vivo administration of IL-2. Eur J Immunol 2001; 31: 3355–3360.

  29. 29

    Ross ME, Caligiuri MA . Cytokine-induced apoptosis of human natural killer cells identifies a novel mechanism to regulate the innate immune response. Blood 1997; 89: 910–918.

  30. 30

    Carson WE, Giri JG, Lindemann MJ, Linett ML, Ahdieh M, Paxton R et al. Interleukin (IL) 15 is a novel cytokine that activates human natural killer cells via components of the IL-2 receptor. J Exp Med 1994; 180: 1395–1403.

  31. 31

    Mizuno S, Akashi K, Ohshima K, Iwasaki H, Miyamoto T, Uchida N et al. Interferon-gamma prevents apoptosis in Epstein–Barr virus-infected natural killer cell leukemia in an autocrine fashion. Blood 1999; 93: 3494–3504.

  32. 32

    El-Deiry WS, Tokino T, Velculescu VE, Levy DB, Parsons R, Trent JM et al. WAF1, a potential mediator of p53 tumor suppression. Cell 1993; 75: 817–825.

  33. 33

    Muda M, Boschert U, Dickinson R, Martinou JC, Martinou I, Camps M et al. MKP-3, a novel cytosolic proteintyrosine phosphatase that exemplifies a new class of mitogen-activated protein kinase phosphatase. J Biol Chem 1996; 271: 4319–4326.

  34. 34

    Liston P, Roy N, Tamai K, Lefebvre C, Baird S, Cherton-Horvat G et al. Suppression of apoptosis in mammalian cells by NAIP and a related family of IAP genes. Nature 1996; 379: 349–353.

  35. 35

    Iida S, Rao PH, Butler M, Corradini P, Boccadoro M, Klein B et al. Deregulation of MUM1/IRF4 by chromosomal translocation in multiple myeloma. Nat Genet 1997; 17: 226–230.

  36. 36

    Woodage T, Basrai MA, Baxevanis AD, Hieter P, Collins FS . Characterization of the CHD family of proteins. Proc Natl Acad Sci USA 1997; 94: 11472–11477.

Download references


We thank all patients, healthy volunteers, and physicians who participated in the collection of NK cell depository. We are also grateful to Dr T Miwa for his helpful suggestions.

Author information

Correspondence to H Mano.

Additional information

This work was supported in part by a grant-in-aid for research on the second-term comprehensive 10-year strategy for cancer control from the Ministry of Health, Labor, and Welfare of Japan, by a grant from Mitsubishi Pharma Research Foundation, by a grant from Takeda Science Foundation, and by a grant from Sankyo Foundation of Life Science.

b >Supplementary Information

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

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Choi, Y., Makishima, H., Ohashi, J. et al. DNA microarray analysis of natural killer cell-type lymphoproliferative disease of granular lymphocytes with purified CD3CD56+ fractions. Leukemia 18, 556–565 (2004) doi:10.1038/sj.leu.2403261

Download citation


  • LDGL
  • DNA microarray
  • correspondence analysis

Further reading