Microrna expression distinguishes between germinal center B cell-like and activated B cell-like subtypes of diffuse large B cell lymphoma

Authors


Abstract

Diffuse large B cell lymphoma (DLBCL) is an aggressive malignancy that accounts for nearly 40% of all lymphoid tumors. This heterogeneous disease can be divided into germinal center B cell-like (GCB) and activated B cell-like (ABC) subtypes by gene expression and immunohistochemical profiling. Using microarray analysis on prototypic cell lines, we identified microRNAs (miR-155, miR-21 and miR-221) that were more highly expressed in ABC-type than GCB-type cell lines. These microRNAs were over-expressed in de novo DLBCL (n = 35), transformed DLBCL (n = 14) and follicular center lymphoma cases (n = 27) compared to normal B cells. Consistent with the cell line model, expression levels were higher in DLBCL cases with an ABC-type immunophenotype than those that were GCB-type (p < 0.05). Moreover, using multivariate analysis we found that expression of miR-21 was an independent prognostic indicator in de novo DLBCL (p < 0.05). Interestingly, expression levels of both miR-155 and miR-21 were also higher in nonmalignant ABC than in GCB cells. As we also demonstrate that expression of microRNAs can be measured reliably from routine paraffin-embedded biopsies of more than 8-years-old (p < 0.001), we suggest that microRNAs could be clinically useful molecular markers for DLBCL as well as other cancers. © 2007 Wiley-Liss, Inc.

MicroRNAs are a recently discovered class of naturally occurring short noncoding RNA molecules that negatively regulate eukaryotic gene expression. They function by binding to complementary target sequences in the 3′UTR of mRNA resulting in repression of translation. It is currently believed that 10–30% of all human genes are a target for microRNA regulation.1 MicroRNAs play key regulatory roles in a diverse range of pathways including control of hematopoiesis, developmental timing, cell differentiation, apoptosis, cell proliferation and organ development (reviewed by Kim2). The potential importance of microRNAs in cancer is implied by the finding that the majority of human microRNAs are located at cancer-associated genomic regions,3 and there is emerging evidence to suggest that dysfunctional expression of microRNAs is a common feature of malignancy.4 It has been suggested that microRNA expression profiling can distinguish cancers according to diagnosis and developmental stage of the tumor to a greater degree of accuracy than traditional gene expression analysis.5 The identity of lymphoma-associated microRNAs however remains poorly defined.

Diffuse large B cell lymphoma (DLBCL) is the most common form of adult lymphoma accounting for nearly 40% of all lymphoid tumors.6 Despite improvements in treatment regimes, the majority of patients succumb to this aggressive disease. DLBCL is characterized by marked clinical and pathological heterogeneity that is reflected at the molecular level. Gene expression and immunohistochemical studies have revealed the presence of at least 2 distinct molecular subtypes of DLBCL representing the postulated cell of origin; those that are germinal center B cell-like (GCB) and those that are activated B cell-like (ABC).7, 8, 9, 10 It is not known whether similar heterogeneity is also present at the microRNA level. Therefore, we used a microRNA microarray to compare microRNA expression profiles of a well-defined cell line model representative of the ABC and GCB subtypes,7, 11 along with normal lymphocyte populations. These results were then tested on a series of 76 clinical cases.

Material and methods

Cell lines, tissue and patient samples

DLBCL cell lines SU-DHL4, SU-DHL6, SU-DHL10, OCI-Ly3, Mieu and DB were routinely cultured in RPMI 1640 containing 10% fetal calf serum (Invitrogen, Paisley, UK), whilst OCI-Ly10 was cultured as previously described.7 Cell lines were obtained from the DSMZ cell collection (Braunschweig, Germany) except OCI-Ly3 and OCI-Ly10 cell lines that were kindly provided by Dr Eric Davis (NCI, Bethesda, MD).

Peripheral blood B cells (CD19+), plasma cells (CD138+) and T cells (CD4+, CD8+) were purified from pooled buffy coat samples (12 individuals) by positive immunomagnetic selection (Miltenyi Biotec, Bisley, UK). Further, B cell purifications (i.e. CD27, CD43, CD10, CD38 and IgD) were carried out using CD19 Multisort beads followed by selection with relevant magnetic beads as indicated in the text. B cell progenitors (CD19+/CD10+) and GCB (CD19+/IgD/CD39) were purified from bone marrow and tonsilar material respectively, pooled from 3 healthy individuals as described previously.12 ABCs, obtained from Yorkshire Bioscience (York, UK), were activated by incubation with pokeweed mitogen for 4 days, as previously described.12 Lymphocyte populations of >90% purity were routinely obtained as determined by FACS analysis (data not shown).

Biopsy material were collected from 76 patients, 49 DLBCL cases (35 de novo and 14 transformed cases) and 27 follicular center lymphoma (FCL) cases (20 were histologically grade 1 or 2 and 7 were grade 3). Patient details and treatment regimes are given in Supplementary Table S1 and S2. Forty-three samples were from frozen material and the remainder from formalin-fixed paraffin-embedded (FFPE) material. Samples were collected at time of initial diagnosis (i.e. prior to treatment) and had >80% of tumor cells as determined by hematoxylin and eosin staining (not shown). Relevant ethical permission was obtained for the use of all samples.

MicroRNA expression profiling by microarray analysis

RNA was isolated from samples in Trizol as recommended by manufacturer (Invitrogen), except FFPE sections which were purified using the Recoverall kit from Ambion (Huntington, UK). The samples were further purified into microRNA and total RNA fractions using RNeasy columns as described by the manufacturer (Qiagen, Crawley, UK). Enriched microRNA (250 ng) samples were labeled with either Cy3 or Cy5 dyes using an Array 900microRNA direct kit from Genisphere (Hatfield, PA) and hybridized to miRMAX microarrays13 containing 225 human mature microRNA sequences, using an alternating circular dye-swap design. At least 4 biological replicates of each cell line sample were used. Image analysis was carried out with BlueFuse software (BlueGnome, Cambridge, UK) and the raw data normalized by global median centering as previously described.14, 15 ANOVA was used to identify differentially expressed microRNAs by taking into account the sample type, dye and array effects. The sample type effect p-values were adjusted using Benjamini–Hochberg correction, and those with p < 0.05 were deemed differentially expressed. Two-way cluster analysis using the Pearson correlation coefficient was carried out using Genesis16 and data analysis was implemented in R (http://www.R-project.org).

Measurement of microRNA expression levels by quantitative RT-PCR and RNase protection assay

Quantitative RT-PCR (qRT-PCR) was carried out using Taqman microRNA probes as described by the manufacturer (Applied Biosystems, Warrington, UK) using 5 ng of microRNA per reaction in a Roche LightCycler 480 machine. Triplicate samples were used throughout. Expression levels are shown relative to control gene (miR-24) (i.e. ΔCT) or for clinical samples, relative to the mean ΔCT value of 6 normal B cell samples (i.e. ΔΔCT). miR-24 was used as a control gene, as recommended by manufacturer, instead of U6, as this microRNA was found to give most consistent levels of expression across frozen and FFPE lymphoma samples (Supplementary Fig. S2)).

RNase-protection assays were carried out with 100 ng of microRNA using the mirVana detection kit as recommended (Ambion). Synthetic probes were 5′-labeled with ATP-[γ-32P] using the mirVana Probe and Marker kit (Ambion) and samples were run on 15% TBE-urea polyacrylamide gels.

Immunohistochemical staining

DLBCL cases were classified as GCB or non-GCB like/ABC by staining FFPE sections with monoclonal antibodies against CD10, BCL6 and MUM1 as described previously.8

Statistical analysis

MicroRNA expression levels were compared using Mann–Whitney independent t-test. Kaplan–Meier survival analysis was carried out on relapse-free survival (RFS) times of de novo DLBCL cases as a function of microRNA expression, using the median value as cutoff. RFS was calculated as the time of diagnosis to the date of clinical relapse, death or last contact. Patients who were relapse-free at time of last contact were censored for analysis. Curves were compared by univariate (logrank) analysis using GraphPad Prism version 4.00. Cox regression analysis of factors potentially linked to survival was preformed to identify which independent factors might jointly influence survival. MicroRNA expression levels were treated as continuous variables for these (multivariate) analyses and carried out in R.

Results

The microRNA expression profiles of DLBCL cell lines are distinct from normal lymphocyte populations and differ between ABC and GCB subtypes of DLBCL

Cluster analysis, based on differentially expressed microRNAs (p < 0.05), classified the malignant DLBCL cell lines as distinct from normal lymphocyte samples, with the majority of differentially expressed microRNAs being up-regulated in DLBCL (Fig. 1a).

Figure 1.

(a) Heat map representation of data from 2-way cluster analysis of differentially expressed microRNAs (p < 0.05) measured by microarray in DLBCL cell lines and normal lymphocyte populations. Values shown are median of biological replicates (at least 4 per cell line). B cells (CD19+), naïve B cells (CD19+/CD27) and T cells (CD3+) were purified from peripheral blood and GC-containing B cells (CD19+/IgD) from tonsils of healthy controls. miR-21, miR-221 and miR-155 are highly expressed in ABC-type but not GCB-type cell lines as demonstrated by (b) RNase-protection assays and (c) qRT-PCR.

To identify microRNAs associated with either GCB- or ABC-type DLBCL, we used a well-defined cell line model consisting of 3 GCB- (SU-DHL4, SU-DHL6 and SU-DHL10) and 2 ABC-type cell lines (OCI-Ly3 and OCI-Ly10).7, 11 The microRNA expression profiles of the 2 subtypes clustered distinctly, with ABC-type cell lines over-expressing a number of microRNAs including miR-155, miR-221 and miR-21 (Fig. 1a). These microRNAs were shown to be more highly expressed in ABC-type cell lines than GCB-type cell lines by both RNase-protection assay and qRT-PCR (Figs. 1b and 1c).

To see whether the expression of microRNAs identified as being DLBCL subtype-specific in the cell line model, were associated with the immunophenotype of clinical samples, we measured expression levels of 35 de novo DLBCL cases by qRT-PCR. As a number of these cases (17/35) were from FFPE material, we compared the expression levels of 10 pairs of matched FFPE and frozen samples. The samples showed remarkably concordant expression levels (Supplementary Table S3, Fig. 2a). DLBCL cases over-expressed miR-21, miR-155 and miR-221 by an average of 9.3, 4.6 and 2.3-fold respectively compared to normal peripheral blood B cells. Consistent with the cell line model, those with an ABC-type immunophenotype expressed significantly higher levels of these microRNAs than those with a GCB-like immunophenotype (p < 0.05, Figs. 2b, 2c and 2d). Expression levels of miR-21, miR-155 and miR-221 were on average 4.1, 2.6 and 1.5 times more highly expressed in ABC-type than GCB-type cases, respectively. We did not find any other associations (using independent t-test) between expression levels of microRNAs and various clinical parameters including age, sex, extranodal disease or IPI stage (data not shown).

Figure 2.

(a) Expression levels of miR-155 in matched frozen and FFPE samples measured by qRT-PCR. Linear regression line shown (r2 = 0.796, p = 0.0005). (b,c,d) Expression levels of miR-21, miR-155 and miR-221 respectively in de novo DLBCL cases of ABC (n = 18) and GCB (n = 17) immunophenotype, transformed DLBCL (trans, n = 14) and FCL cases (n = 27) measured by qRT-PCR. Fold change (2ΔΔCT) are shown relative to controls (6 normal B cell samples) and data are represented as box-whisker plots. Expression levels of ABC-type cases were higher than GCB-type cases as measured by Mann–Whitney independent t-test with p values of 0.01, 0.03 and 0.04, respectively.

To look at expression of these microRNAs in related lymphomas we also measured expression levels in 14 cases of DLBCL that had undergone high grade histological transformation from a previously diagnosed FCL, as well as 27 cases of FCL (20 histological grade 1 or 2 and 7 grade 3 cases). Although miR-155, miR-221 and miR-21 were also over-expressed when compared with normal peripheral blood B cells, we found no significant differences between expression levels in these different lymphomas (Figs. 2b, 2c and 2d).

Expression of miR-21 is an independent prognostic indicator in de novo DLBCL

To assess the potential prognostic impact of these microRNAs in de novo DLBCL we preformed a retrospective analysis. Using Kaplan–Meier survival (univariate) analysis, we found that high miR-21 expression was associated with longer RFS in de novo DLBCL cases (p = 0.04, Fig. 3a). We found no significant correlation between expression of either miR-155 or miR-221 and prognosis in this series of cases (data not shown). We further examined the affect of various factors that might affect prognostic outcome in this cohort (i.e. sex, age, immunophenotype, stage, LDH, extranodal disease and IPI stage), as well as microRNA expression levels, by multivariate Cox proportional hazard regression analysis. We found that only expression of miR-21 and IPI status were statistically significant independent indicators of prognosis in this cohort (p < 0.05, Table I). We found no association between expression levels of any of the identified microRNAs and survival in FCL patients (data not shown).

Figure 3.

(a) Kaplan–Meier survival curves of de novo DLBCL patients (n = 35) based on expression levels of miR-21. Curves were compared by univariate (logrank) analysis. High expression of miR-21 was associated with longer RFS period (p = 0.04). Expression levels were defined as high or low relative to the median. (b,c,d) Expression levels of miR-21, miR-155 and miR-221 respectively in different T and B cell developmental stages measured by qRT-PCR.

Table I. Cox Regression analysis Of Prognostic Factors Associated (p < 0.1) with Relapse-Free Survival of DLBCL Patients
VariablesHazard ratio (95% CI)Favorable/ unfavorablep
miR-210.77 (0.61–0.97)High/low0.025
IPI stage2.55 (1.39–4.67)0–2/3–40.003
Immunophenotype0.27 (0.07–1.06)GCB/ABC0.060

Expression levels of miR-155 and miR-21 are higher in nonmalignant ABC than GCB cells

We measured expression levels of miR-21, miR-155 and miR-221 in different B-cell and T-cell developmental stages obtained from healthy individuals (Figs. 3b, 3c and 3d). Interestingly miR-155 and miR-21 were most highly expressed in ABC and less so in GCB cells. In contrast, resting B cells had the lowest levels of expression of miR-155 and miR-221. Generally miR-221 was expressed at lower levels in lymphocytes than either miR-155 or miR-21.

Discussion

DLBCL is the most common form of non-Hodgkin's lymphoma with an annual incidence of over 25,000 cases in the United States alone. Although combination chemotherapy has transformed DLBCL from a fatal disease to a potentially curable one, some patients have a highly variable response to treatment, and fewer than half of patients are cured. Considerable effort has been made to identify which patients do not respond to current therapeutic regimes and hence have a poorer prognostic outcome. Gene expression studies have identified several prognostically distinct subtypes within DLBCL including the ABC- and GCB-like molecular subtypes.7 These subtypes can also be classified by immunophenotype, with an accuracy of 88 and 71% of cases, respectively.8 We reasoned that if microRNAs could also distinguish between these subtypes they could be used to provide additional prognostic information for this heterogeneous disease.

Using microarrays to investigate the microRNA expression profile of DLBCL cell lines in comparison with normal lymphocyte populations, we found that most microRNAs that were over-expressed in malignant samples were either members of the miR-17-92 cluster or of homologous clusters miR-106a-92 and miR-106b-25. This data is consistent with other studies17 and over-expression of the miR-17-92 cluster has been observed in lung cancer18 and solid tumors19 as well as in lymphoma. The miR-17-92 cluster is upregulated directly through c-myc binding, as well as being associated with the 13q31 amplification that harbors its parental transcript, c13orf25.17, 18, 20 Furthermore, expression of this cluster has been shown to increase lymphogenesis in a murine model.17 The only microRNA identified by our analysis as being consistently more highly expressed in normal lymphocyte samples than DLBCL was miR-150. This microRNA was previously shown to be highly expressed in during murine lymphocyte development.21

Using a prototypic cell line model of the GCB and ABC subtypes of DLBCL,7, 11 we identified microRNAs (i.e. miR-155, miR-221 and miR-21) that were over-expressed in ABC-type but not GCB-type cell lines. These microRNAs were found to be over-expressed in clinical cases of DLBCL and FCL as well as DLBCL cases that had undergone high grade transformation from previously diagnosed FCL. Consistent with the cell line model, these microRNAs were more highly expressed in de novo DLBCL cases that were immunophenotypically classified as ABC-type than those that were GCB-type (p < 0.05).

The association of miR-155 expression with DLBCL displaying an ABC-type immunophenotype is in agreement with previous research, although fewer cases were examined in those studies.9, 10 Recently, transgenic mice expressing miR-155 targeted to B cells were shown to spontaneously develop high grade lymphomas.22 We did not find any association between expression levels of miR-155 and survival in the patient cohorts we examined (data not shown), although overall miR-155 was more highly expressed in all lymphoma cases, with an average expression 4.8-fold higher than in normal B cells. Over-expression of miR-155 appears to be a feature of many malignancies including breast cancer,14 thyroid carcinoma23 and a range of solid tumors.19

miR-221 has been shown to inhibit normal erythropoiesis through downregulation of c-kit expression.24 Therefore, we looked at levels of this protein in DLBCL and FCL cases by immunohistochemistry. We found no evidence of expression of c-kit in any of the cases (data not shown). It might be tempting to believe therefore that miR-221 suppresses expression of c-kit in DLBCL and FCL, however evidence implies that mRNA expression levels, in DLBCL at least, are not elevated to the extent found in other hematological malignancies,25 suggesting that this gene is not a significant target for miR-221 in these lymphomas. High miR-221 expression levels have also been reported in thyroid carcinoma23 and glioblastoma.15

In addition to miR-155 and miR-221, ABC-type DLBCL was also associated with high expression of miR-21. Moreover, univariate and multivariate analysis suggested that miR-21 expression was an independent prognostic indicator from IPI status in de novo DLBCL and that high expression was associated with a better prognostic outcome (p < 0.05). Inhibition of this microRNA in HeLa cells resulted in a profound increase in cell proliferation.26 Whilst the ABC-immunophenotype has been associated with an adverse prognostic significance,8 other studies do not reach similar conclusions.27, 28 We did not find a statistical association between immunophenotype and prognosis in the cohort we examined (p > 0.05, Table I), probably due to the majority of these patients having been treated with rituximab which has been found to overcome the negative prognosis associated with ABC-type DLBCL.29

We also looked at the expression of these microRNAs in different developmental stages of B and T cells from healthy individuals, and found that both miR-155 and miR-21 were more highly expressed in ABC than GCB cells. This suggests that high levels of these microRNAs in DLBCL are not necessarily malignancy-associated but may instead reflect a characteristic of their postulated cell of origin.

The biological relevance of expression of these miRNAs in DLBCL is not yet known as only very few miRNA target genes have been validated experimentally.30 This is further complicated by the fact that a single miRNA can target many hundreds of genes and conversely a single target mRNA can cooperatively bind multiple miRNAs.1, 31 Putative target genes are currently predicted using algorithms such as miRanda,30 PicTar,32 TargetScan1 and miRBase.33 As these algorithms predict differing target genes, we identified target genes that were predicted by 2 or more algorithms, highlighting genes that were previously shown to be cancer-associated (36% of targets) (Supplementary Table S4). It might be expected that targets for upregulated miRNAs include tumor-suppressor and pro-apoptotic genes. Indeed, many such genes were identified as being putative targets for the identified miRNAs (Supplementary Table S5).

Clinical utility of microRNA expression studies in cancer

Although traditional gene expression studies have been widely used to identify genes that could have diagnostic and/or prognostic utility in cancer, the need for fresh or optimally cryo-preserved samples, currently precludes their widespread application in a clinical setting. In contrast, microRNAs due to their small size are relatively resistant to RNase degradation and can be successfully isolated from routinely processed FFPE tissue. Indeed, we found that such samples gave remarkably similar results to matched frozen tissue by qRT-PCR, even when over 8-years old. Moreover, we found that FFPE-isolated microRNA gives comparable data to frozen material in microarray analysis (Supplementary Fig. S1). These properties, in addition to the potential clinical significance of microRNA expression, suggest that microRNAs are excellent tools for future molecular diagnostics in cancer.

Acknowledgements

We would like to especially thank Dr. Nigel Saunders (Sir William Dunn School of Pathology, University of Oxford, Oxford, UK) for all his advice and support with the microarray experiments. We are grateful to Dr. Russell Leek (NDCLS, Oxford) for his help with statistical analysis, and to Ms. Helen Roberton, Ms. Leticia Campo and Dr. Amanda Liggins (NDCLS, Oxford) for providing cell-line material. We are also thankful to Dr. Francesco Pezzella (NDCLS, Oxford) for critically reviewing and discussing this manuscript.

Ancillary