Parallel Gene Expression Profiling of Mantle Cell Lymphoma — How Do We Transform ’Omics Data into Clinical Practice

Abstract

DNA microarray technology has been a valuable tool to provide a global view of the changes in gene expression that characterize different types of B cell lymphomas, both in relation to clinical parameters but also in comparison with the non-malignant counterparts. The number of transcripts that can be analyzed on an array has dramatically increased, and now most commercially available arrays cover the whole genome, enabling overall analysis of the transcriptome.The backside of collecting this massive amount of information is that even after strict data filtering, it is impossible to do follow-up studies on all findings. Down-stream analysis is time-consuming and when performing confirmatory experiments on the protein level, the experiments are in most cases restricted to proteins recognized by commercially available reagents. Furthermore, since gene expression data is a comparative method not only are the experimental set-up but also the characteristics of both the sample and reference crucial for our ability to answer the questions posed. Thus, initial care must be taken in the design of the experiment and the preparation of the samples.

The aim of this review is to discuss the progress in mantle cell lymphoma research enabled by gene expression analysis and to pinpoint the difficulties in making efficient use of the generated data to provide a fast and accurate clinical diagnosis, efficient stratification of patients into disease sub-groups and improved therapy.

MANTLE CELL LYMPHOMA — HISTORICAL PERSPECTIVES

B cell lymphomas are the malignant counterparts of different developmental stages of normal B lymphocytes and are divided into precursor B cell neoplasms and mature/peripheral B cell neoplasms, according to the WHO classification 1. B cell differentiation is strictly regulated by ho-meostatic controls but even so, malignant transformation occasionally proceeds unimpeded 2. The main groups of B cell neoplasms are precursor B-lymphoblastic leukaemia/lymphoma, Chronic lymphocytic leukaemia/lymphoma (CLL), plasma cell myeloma, extranodal marginal zone (MZ) B cell lymphoma of mucosa-associated lymphoid tissue (MALT) type, follicular lymphoma (FL), mantle cell lymphoma (MCL), diffuse large B cell lymphoma (DLBCL) and Burkitt lymphoma 1.

MCL was originally considered to be derived from naïve B cells and most tumors show unmutated variable heavy chain (V_H) genes. However, several studies have shown that 10–20% of MCLs have somatically mutated immunoglobu-lin (Ig) genes, indicating that a sub-group has passed through a differentiation stage involving somatic hypermutation of V_H genes [3–8]. Recently, it has been shown that using an alternative method up to 60% of the MCL were shown to have <98% homology with the germline sequence and were thus considered to carry somatically mutated Ig genes 9. In the same study it was further shown that patients with somatically hypermutated V_H genes had a better survival and that intraclonal heterogeneity was common in this group. This is in contrast to previous results, where MCL cells, including the V_H mutated, has been described to be clonally identical, indicating that the tumor cells are frozen at the developmental stage where the malignant transformation took place [2, 10]. Interestingly, it has been reported that B cells can mutate their V_H genes in the absence of a germinal center (GC), but the mechanisms involved are at present not understood [11, 12]. In a previous study, using gene expression analysis of MCL and sub-populations of normal tonsillar B cells we concluded that MCL tumor cells had a transcriptional profile similar to activated B cells and not to na-ïve B cells 13, as previously proposed. This is in accordance with the mutational studies and support the conclusion that MCL is derived from pre-activated B cells and not from naïve B cells [4, 6, 7, 9, 10]. Furthermore, a biased V_H usage has been described for MCL, which suggests that the transformation is antigen driven [4, 6, 7, 9, 10]. No antigen has been identified but V_H 3–21, which is one of the most frequently used V_H genes in MCL, is common in, e.g., rheumatoid factors and may thus define an auto-antigen [2, 14].

The main feature of MCL is over-expression of cyclin D1, which is due to the translocation of BCL1 to the heavy chain locus [15, 16]. This t(11;14)(q13;q32) was identified in the 1970s as a cytogenetic event occurring in rare examples of B cell non-Hodgkin’s lymphoma (NHL). Tsujimoto and colleagues cloned this translocation breakpoint in 1984 17. Cyclin D1 is a positive regulator of the G1/S cell-cycle restriction point and the over-expression induces increased cell cycling. Even if the over-expression of cyclin D1 is the main feature of MCL, it is probably not the sole oncogenic feature, as over-expression of the gene in mice is not lymphomagenic 18.

Rare MCL-like B cell lymphomas lack cyclin D1 expression, but still have the other morphological and immunophe-notypical characteristics of MCL 19. These cyclin D1-negative MCL-like lymphomas have been shown to have a better prognosis and should be considered as an entity separate from MCL [19, 20]. Other phenotypic characteristics of MCL are the expression of CD5, CD19, CD20, CD22, CD79a, CD79b and the lack of CD10 and CD23 21. The lack of CD10 and CD23 expression distinguish MCL from FL and B-CLL, respectively. MCL cells also over-express Bcl-2 but lack t(14;18) (q32:q21), which is characteristic of FL.

MCLs grow in three different patterns in lymph nodes: mantle zone, nodular and diffuse. These growth patterns can be divided further into typical or blastoid variants 21. The characteristics of the different clinical, phenotypic and genotypic features of MCL were agreed on in 1994 22. Morphologically, typical MCLs can have either a nodular or a diffuse growth pattern, with or without residual GCs, and are normally infiltrated with either loose or tight follicular den-dritic cell aggregates 23.

The median age for MCL patients is 60 years (range 18–86) at diagnosis, and the malignancy has a male predominance (3:1) 21. Although the median survival period is short, only 3 to 5 years, and very few patients show long-term survival, a few features indicate a more favorable prognosis. These features include a proliferative index below 10% and arguably, a nodular growth pattern 21. Most patients show disseminated disease at the time of diagnosis, with 80% of patients in stage III or IV, enlarged lymph nodes and frequent (70%) bone-marrow involvement. Involvement of the blood (25%) and gastrointestinal tract is also common 23, while infiltration of the central nervous system is rarely detected and may be related to blastoid transformation 24. Lack of lymph node involvement with isolated massive splenomegaly occurs in some patients, but this is infrequent 21.

MCLs may progress to a blastoid variant, as reviewed by Matolcsy 25 and Muller-Hermelink et al. 26, but this variant may also arise de novo 24. The more aggressive blastoid variant is associated with decreased survival [27, 28] and 26–70% of MCL patients have a morphological progression towards this variant, detected either during life or at autopsy [29, 30]. Identical clonally rearranged Ig genes suggest that blastoid variants arise from the original neoplasm by clonal selection 25. Factors that predict blastoid transformation are leukocytosis, elevated serum lactate dehydro-genase and high proliferative index 24.

Most patients with MCL who show response to treatment, or even total remission, soon relapse with new tumors. Thus, more effective treatments must be sought. It is also essential to minimize toxicity, as most patients afflicted by MCL are 60 years of age or more and are likely to experience treatment-related complications 31. There is no defined standard treatment for MCL. The most commonly used regimen is CHOP (cyclophosphamide, doxorubicin, vincris-tine and prednisone), usually in combination with immuno-therapy, such as rituximab [32–34]. Recent results show increased response to this combined therapy, but it does not improve the long-term outcome of MCL patients 35. Consolidation with high-dose chemotherapy and stem cell support has been shown to prolong progression-free survival after conventional chemotherapy, but at present, is not considered to be curative 36. In addition, maintenance therapy with rituximab after a combined treatment with fludarabine, cyclophosphamide, and mitoxantron shows promising results for both MCL and FL 37. Several other regimens are being evaluated, including bortezomib, a proteasome inhibitor 38 along with a number of other biological agents and combinations as reviewed elsewhere [39–41].

Furthermore, MCL cells are highly sensitivity to radiation 40 but since the malignancy most often occurs systemically, this is in most cases not a treatment option. However, in 10–15% of MCL patients, limited stage (I/II) disease is diagnosed. Since these patients only have a few localized tumors, regional radiation therapy has been shown to be a potentially curative treatment 34.

Thus, MCL is far from a homologous entity and even though thumors share many characteristics, they can potentially be divided into sub-groups. Novel technologies, including various microarrays, can be used to identify the molecular differences involved in morphological heterogeneity, clinical behavior and mutational status. These approaches may form the basis to define different therapies tailored to each sub-type of MCL or to find common features useful for improved therapeutic interventions.

GENE EXPRESSION IN LYMPHOMA – DOGMAS AS BASIS FOR ANALYSIS

Highly parallel gene expression analysis has during the last 6–7 years revolutionized the way we look at the different malignant lymphoma entities. It has been clearly shown that the heterogeneity seen in response to treatment and clinical outcome can be correlated to differential gene usage. Global gene expression profiling was also shown early on to be useful for facilitating diagnosis, enabling separation of similar but distinct entities such as acute myeloid leukemia and acute lymphoblastic leukemia (AML-ALL) 42, as well as morphologically similar but clinically different subgroups, such as V_H mutated vs. unmutated CLL 43. A recent gene expression study on FL, using whole tumor tissue, concluded that certain genes, such as EphA1 and Smad1, expressed by non-malignant cells in the microenvironment, including granulocytes and vascular structures, affect treatment outcome for patients given Rituximab and CHOP 44. for MCL, different studies have focused on the implications of the overexpression of cyclin D1. Fu et al. 45 used gene expression analysis to show that cyclin D1-negative MCL have similar profiles compared to cyclin D1 positive MCL. In another study, the correlation between a short variant of the cyclin D1 mRNA transcript and increased proliferation in MCL was studied 46 and a third gene expression study demonstrated that cyclin D1 levels are associated with changes in proliferative activity 47. It has further been shown that survival of MCL patients can be predicted, using gene expression analysis. Rosenwald et al. constructed a proliferative signature, based on the Lymphochip, that could predict survival in MCL patients 20. Analyses of MCL have also been performed to define the main features regulating MCL growth, i.e., the apoptotic machinery and genes involved in cell survival [13, 48, 49]. Other studies of MCL have, for example, revealed differences in gene usage, comparing proliferative indexes 50 and difference in morphological subtypes 51, as well as focusing on markers, such as chemokine usage 52 or expression of the cannabinoid receptor 47. It has further been shown that ATM and p53 mutational status can be associated with distinct gene expression signatures, although only p53 mutations could be associated with decreased survival time 53. The leukemic variant of MCL has also been analyzed 54 and MCL cell lines have been assessed as tools for in vitro studies 55. Lately, the technology has also been used in parallel with global genomic analysis to define potential therapeutic targets in MCL 56.

Analysis of B cell lymphomas has centered around the belief that the malignant cell retains most of the characteristics of its benign precursor. Thus, the stage of B cell differentiation at which the malignant transformation took place will influence the behavior of the transformed daughter cells. Consequently, the finding by Alizadeh et al. in 2000 based on gene expression analysis 57, which showed that different subtypes of DLBCL were associated with different normal B cell populations, had a significant impact, not only on the approach to sub-grouping of DLBCL but also on how other B cell lymphomas can be sub-grouped. Knowledge of the normal counterpart will therefore directly facilitate the design of possible methods of intervention, since it pinpoints the events in normal B cell differentiation where the cell is more vulnerable to escaping homeostasis and immune regulation.

Initially, many studies have been performed using focused arrays. Examples are the Lymphochip, used by Staudt and colleagues in analysis of different types of lymphoma [20, 57–60], as well as the Oncochip-CNIO, used by Martinez and colleagues when analyzing MCL 49. In these cases, genes known to be involved in the immune system or related to cancer have been used. In contrast, other arrays, such as the oligonucleotide arrays from Affymetrix, have not been focused towards any groups of genes. Today, when it is possible to scan the whole genome, including uncharacter-ized transcripts, there is no longer any real advantage to focus on a limited number of genes for the initial analysis.

EXPERIMENTAL DESIGN AND ANALYTICAL STRATEGIES – IS THE BEST THERAPEUTIC TARGET A NON-B CELL ANTIGEN

Since gene expression analysis is a comparative technology, the choice of reference material is crucial for the ability to draw strict conclusions. When correlating tumor material to different pathological or clinical parameters, the material is its own reference, as for example when (i) analyzing MCL with and without the blastic variant [51, 61], (ii) correlating gene expression to survival 20 or (iii) comparing gene expression profiles between similar lymphoma entities 62. However, when aiming at identifying potential therapeutic targets, the choice of reference material becomes more difficult. The most important parameter, when selecting a reference, is the composition of the tumor and reference material. The heterogeneity of most archived material makes analysis of the actual gene expression of the malignant cells impossible since it is the sum of the different cell types in the tumor that contributes to the identified signatures. The microenvi-ronment is a key factor for the survival of the tumor and it is clear that both cells and soluble factors influence this milieu and will have a great impact on the analysis, as exemplified by the above gene expression study on FL by Harjunpaa et al. 44. It has already been suggested that the outcome of patients afflicted by FL, in comparison with MCL and DLBCL, is more dependent on the features of the microenvi-ronment than the status of the tumor cells themselves [63, 64]. We have also previously shown that MCL display surface receptors that most likely are involved in cross-talk with neighboring immune cells, such as follicular dendritic cells and that an altered chemokine production may confer survival advantages to the tumor cells 52. Consequently, to be able to make conclusive statements regarding the nature of either the tumor cells or cells in the microenvironment, these players need to be analyzed separately, which today is too seldom performed due to lack of adequate tumor material.

With this in mind, which is then the best choice of reference material, when aiming to identify therapeutic targets? Most studies use normal B cells, which is a natural choice, they being the ancestor cell. Other examples of different material used for gene expression analysis are (i) hyperplastic lymph nodes, used as reference material in comparison with lymph nodes from patients with MCL 48, (ii) sorted tonsillar B cell subpopulations 13 or (iii) naïve B cells 49. These analyses, comparing malignant and non-malignant B cells, aim at defining differentially expressed gene products. Does it make a difference if the target then is a non-B cell antigen? Rituximab, targeting CD20, has successfully been used in different B cell malignancies, although the effect on MCL patients has been more limited compared to, for example, patients with FL 37. Other examples of Food and Drug Administration (FDA) approved antibodies, targeting both malignant and non-malignant lymphoid cells, include Bexxar (Tositumomab and I-131 Tositumomab), which like Rituxi-mab targets CD20, Campath (Alemtuzumab), targeting CD52 on B cell chronic lymphocytic leukemia and Mylotarg (gemtuzumab ozogamicin), targeting CD33 positive cells in acute myeloid leukemia (www.accessdata.fda.gov/scripts/ cder/drgsatfda/index.cfm). Although effective to different degrees, it is obvious that it is more beneficial for the patient to retain an intact B cell pool. This is especially crucial for immunodeficient patients. B cell malignancies, like other types of cancer, frequently affect individuals with an impaired immune system, such as patients suffering from acquired immunodeficiency syndrome (AIDS) or patients treated with immunosuppressive agents. The role of Epstein Barr virus infection for development of NHL in immunosup-pressed individuals, for example after transplantation, and in individuals infected by human immunodeficiency virus (HIV), is particularly well documented 65. Other biological regimens like bortezomib, a proteasome inhibitor, are not specific for B cells but still show promising results 38.

Consequently, it remains to be seen whether B cell lym-phoma-associated antigens, not present on non-malignant B cells, will be more efficient as targets for new antibody-mediated therapies. However, a larger variety of reference material and strategies must be considered to enable a focused analysis for the identification of suitable targets on the malignant cells or on surrounding cells that facilitate tumor growth and survival.

FROM GENOMIC TO PROTEOMIC ANALYSIS IN MANTLE CELL LYMPHOMA – HOW FAR ARE WE FROM DEFINING NEW DIAGNOSTIC OR THERAPEUTIC TARGETS

Conventional chromosomal comparative genomic hybridization (CGH) has been used to identify common MCL-associated cytogenetic abnormalities [66–69]. Recently, high-resolution microarray-based CGH assays have been performed on a number of both primary MCL primary samples, as well as MCL cell lines [70–74]. Similar to gene expression analysis, genetic abnormalities were shown to be associated with the clinical outcome of MCL patients [20, 72] and were correlated to clinical sub-groups and parameters. Only a few genomic studies of MCL have been correlated with gene expression data [56, 73] and conversely, gene expression studies most often lack parallel analysis of the underlying genomic alterations. However, for gene expression analysis aimed at defining new therapeutic antigens, correlation at the genomic level may be of less importance than correlation at the proteome level. Despite this, the bulk of MCL data generated on expression arrays remains to be converted into information on the corresponding proteins.

The standard methods for analyzing proteins which correspond to known differentially regulated genes are immu-nohistochemistry (IHC) and Western blot, using commercially available antibodies. Since paraffin-embedded tumor tissues are often readily available in biobanks IHC analysis is thus the most frequently used method. However, IHC requires that the antigen is resistant to the harsh fixation and treatments used in histotechnology and consequently some antigens are difficult or impossible to analyze using this method 75. Western blot, on the other hand, suffers from other limitations. In Western blot analysis, separation of the different cell types in the tumor tissue needs initially to be performed, to be able to pinpoint the cell type carrying the protein of interest. In most cases, such fresh tumor samples are difficult to collect. Furthermore, both techniques are limited by the availability of suitable commercial antibodies. In a recent study, we have used tailor-made antibodies generated within the Human Proteome Resource (HPR) project [76, 77] to transfer our gene expression data to the proteome level 78. In short, gene transcripts upregulated in MCL tumors were used to generate antibodies. In subsequent IHC analysis, a large fraction of these antibodies recognized MCL tissue but not benign B cells. Several such antibodies are being further characterized using large scale tissue mi-coarrays with an aim to define antigens useful for improved diagnostics or therapy (Ek et al., manuscript in preparation).

A few de novo parallel proteome analyses of MCL have been performed, for example using commercial antibody microarrays 79 or two-dimensional gel electrophoresis, in MCL cell lines (GRANTA-519 and MAVER-1) or excised tissue, respectively [80, 81]. However, the number of proteins that can be analyzed is limited compared to parallel gene expression profiling. The antibody microarray analyzed 512 antibodies, whereas the 2D gel electrophoresis could identify 750 spots in the MCL tumor material. High-density antibody-based microarrays may offer a way to efficiently transfer gene expression data into proteomics, but this technique like IHC and Western blot requires the availability of suitable antibodies. The generation of tailor-made antibodies designed for microarray applications is on-going and uses large recombinant antibody libraries, which will eventually offer an unbiased analysis of novel gene products 82. A summary of the process where gene expression data is used to focus interest on antigens involved in a specific malignancy and how these targets can be validated at the protein level and in functional analysis is shown in Fig. (1). As discussed above, there is a clear need for specific tailor-made reagents capable of validating all findings without discrimination.

Fig. (1).

Bioinformatic analysis of DNA* microarray data from malignant and normal cells identifies differentially expressed target genes, as illustrated by (a) the heat map, where genes upregulated in malignant cells are shown in red and downregulated genes are shown in green. (b) Downstream analysis is often focused on a limited number of targets for which commercially antibodies (abs) are available. (c) Tailor-made antibodies allow downstream analysis of all identified gene products. Antibodies or antibody-fragments (scFvs) can be used in various applications, for example (d) antibody microarrays, where different materials such as plasma, whole cells or tissue extracts can be analyzed, or on (e) tissue microarrays, where a large number of different tissues is analyzed on a single slide or (f) for functional analysis, where the therapeutic potential can be assessed.

*The DNA illustration is adapted from http://genomics.energy.gov with permission from the U.S. Department of Energy Genome Program. The antibody illustration was created by Mats Ohlin from the B12 antibody (1HZH, www.pdb.org). The antibody microarray illustration was kindly provided by Christer Wingren.

To finally make sense of gene and protein expression data, relevant in vivo models for functional analysis of the identified potential targets are needed. Murine models have until recently been lacking, except for a xenograft using the UPN1 cell line 83, but in 2006 two separate models were presented in which MCL-like tumors could be induced using mitogenic stimulation in Eμ-cyclin D1 carrying transgenic mice 84. In a second study, a mutant form of cyclin D1 prevented export from the nucleus, thus showing that constitutively nuclear cyclin D1 induces B-cell lymphomas in transgenic mice 85. Further evaluation of these models and there utility in MCL research is needed.

In summary, we conclude that although extensive research on MCL using gene expression analysis has been performed, parallel, unbiased tools to transfer the data to the protein level needs to be implemented to facilitate the transfer of the new findings into the clinic. These tools include combinations of high-throughput affinity proteomics, such as antibody and tissue microarrays, which recently have shown great promise in defining new and improved approaches to facilitate diagnosis, treatment decisions and in the long term may potentially contribute to improved treatment outcome for patients.

ABBREVIATIONS

AIDS

Acquired immunodeficiency syndrome

ALL

Acute lymphoblastic leukemia

AML

Acute myeloid leukemia

CHOP

Cyclophosphamide, doxorubicin, vincristine and prednisone

CLL

Chronic Lymphocytic Leukemia

CGH

Comparative genomic hybridization

DLBCL

Diffuse large B cell lymphoma

FDA

Food and Drug Administration

FL

Follicular lymphoma

GC

Germinal center

HIV

Human immunodeficiency virus

HPR

Human Proteome Resource

Ig

Immunoglobulin

IHC

Immunohistochemistry

MCL

Mantle cell lymphoma

MZ

Marginal zone

MALT

Mucosa associated lymphoid tissue

NHL

Non Hodgkin’s lymphoma

V_H

Variable heavy chain