Alternative splicing microarrays reveal functional expression of neuron-specific regulators in Hodgkin lymphoma cells.

Alternative splicing provides a versatile mechanism of gene regulation, which is often subverted in disease. We have used customized oligonucleotide microarrays to interrogate simultaneously the levels of expression of splicing factors and the patterns of alternative splicing of genes involved in tumor progression. Analysis of RNAs isolated from cell lines derived from Hodgkin lymphoma tumors indicate that the relative abundance of alternatively spliced isoforms correlates with transformation and tumor grade. Changes in expression of regulators were also detected, and a subset sample was confirmed at the protein level. Ectopic expression of neuron-specific splicing regulatory proteins of the Nova family was observed in some cell lines and tumor samples, correlating with expression of a neuron-specific mRNA isoform of JNK2 kinase. This microarray design can help assess the role of alternative splicing in a variety of biological and medical problems and potentially serve as a diagnostic tool.

Microarray technology is purported as a key tool for genomewide analysis of transcriptomes in post-genomic biology (1). Early successes of this technology range from detailed genomewide analysis of gene expression during Drosophila development to the molecular classification of lymphomas (2-4) Classical microarray designs, however, largely ignore sources of transcript heterogeneity such as alternative pre-mRNA splicing, which affects more than 70% of human genes and leads to the generation of multiple mRNAs and protein products from a single primary transcript (5,6). Alternative splicing is altered in disease, notably in cancer (7).
Pioneering steps have been made to extract information about alternative transcripts from conventional microarrays (8) and to develop microarray platforms specifically designed to detect alternative splicing (reviewed in Ref. 9). These include high resolution coverage of genomic sequence by tiling arrays (10), analysis of global effects associated with defects, in particular, RNA processing factors in yeast using intron-specific and splice-junction probes (11), multiplexed analysis of cancer cell lines on fiber optic arrays (12), and screening for alternative splicing events in human genes and different tissues using splice-junction oligonucleotides (6).
Work in a variety of genes and organisms has led to the hypothesis that the relative expression of general splicing factors in different cell types establishes patterns of alternative splicing in multiple genes (13)(14)(15). Tissue-specific regulators have also been found, e.g. the nervous system-specific proteins of the Nova family, which induce neuron-specific patterns of alternative splicing on target genes (16,17).
Hodgkin lymphomas are malignancies of unknown pathogenesis. Despite their heterogeneous phenotype, cDNA microarrays have revealed the expression of a characteristic set of markers regardless of their B-or T-cell origin (18,19) There is a need, however, for additional prognostic markers. Results obtained with our microarray design suggest that general patterns of alternative splicing, as well as expression of particular splicing regulators, can be indicative of cell transformation and tumor grade.
Splicing Factors Data Base-A data base of splicing factors was built from the European Molecular Biology Laboratory (EMBL) Sequence Data base containing constitutive components of the spliceosome (snRNP 1 proteins and small nuclear RNAs), members of the SR and hnRNP families of splicing regulators, splicing factor kinases, and other RNA-binding proteins (see Fig. 1A). The genes encoding these proteins and RNAs were grouped by families or functions, together with their sequence information and links to orthologous genes in different species (eta.embl-heidelberg.de:8000/cgi/angela_restricted/cgimodel.py?funϭ displayGroupFactorPage).
Microarray Analyses-Oligonucleotide microarrays were produced using the Geniom OneR system, Febit AG, Mannheim, Germany (20). Probes for bacterial genes conferring resistance to ampicillin, kanamycin, and chloramphenicol and the Drosophila gene Sex-lethal were used as negative controls. Preparation of biotinylated cRNA was essentially as reported (20). Briefly, total cytoplasmic RNA was isolated from cells in culture using the RNeasy kit (Qiagen). 5 g of total RNA were incubated with 100 pmol of T7-T (24) -primer at 70°C for 10 min in an 11-l reaction. cDNA synthesis was carried out in a 20-l reaction containing 10 mM dithiothreitol, 500 M dNTPs, 20 units of RNase OUT (Invitrogen), and 300 units of SuperScript II (Invitrogen) for 1 h at 50°C. Second strand cDNA synthesis was carried out in a 150-l * This work was supported in part by grants from Bundesministerium fü r Bildung und Forschung, European Union FP5, and Human Frontier Science Program Organization. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. DNA polymerase treatment, phenol/chloroform extraction, and ethanol precipitation, T7 transcription was carried out for 6 h at 37°C in a -l reaction using the AMS Biotech Megascript T7 kit in the presence of 1.88 mM Biotin-11-CTP, 1.88 mM Biotin-16-UTP. The DNA template was then digested for 15 min at 37°C with 1 unit of DNase I, and the labeled cRNA was purified using the RNeasy mini kit (Qiagen). cRNA fragmentation was achieved by incubating 15 g of cRNA in 40 mM Tris, pH 8.1, 30 mM magnesium acetate, 100 mM potassium acetate in a total volume of 5 l for 35 min at 94°C. The average fragment length was 100 nucleotides.
Microarrays were hybridized with 15 g of fragmented cRNA in a final volume of 20 l. Detection, processing of raw data including background correction, array-to-array normalization, determination of gene expression levels, and calculation of -fold change values were carried out as described (20,21) using data from 13 perfect match/single mismatch oligonucleotide pairs (for mRNAs and exons) or 11 pairs (for splice junctions). Tiling arrays (full coverage of a sequence by oligonucleotides offset by one nucleotide) were generated for selected sequence features. At least six different RNA isolates were used per cell line to calculate median values of -fold changes (data provided as Supplementary Materials), which were categorized as up-regulation , down-regulation, or no change based on a 2-fold threshold and the request that at least 50% of the replicates showed consistent behavior, and less than 30% of the remaining data showed disparate results. Subsequent analyses (including Spearman correlation clustering, gene trees, condition trees, and analysis of variance analysis) used the Excel and Gene Spring 6.2 (Silicon Genetics) software programs.
Western Blot-Total protein extracts were fractionated by electrophoresis on SDS-polyacrylamide gels and transferred to nitrocellulose membranes using a semi-dry blotting chamber. Membranes were incubated with 1ϫ phosphate-buffered saline-0.1% Tween supplemented with 5% nonfat milk and antibodies: rabbit polyclonal antihuman Nova-1/3000; rabbit polyclonal antihuman PSF, 1/5000 (a gift from J. Patton,

FIG. 1. Experimental design.
A, categories of splicing regulators and splicing events analyzed in this study. The number/identity of genes are indicated in brackets. TNF, tumor necrosis factor; TNFR, TNF receptor; VEGF, vascular endothelial growth factor; ARF, alternative reading frame of p16. B, distribution of oligonucleotides to study alternative splicing events. 13 oligonucleotide pairs (perfect match/ mismatch, only three are drawn) were selected for each alternative or neighboring constitutive exon, and 11 (only one is drawn) were selected for each of the splice junctions. Vanderbilt University); rabbit polyclonal antihuman SAP155 1/300 (a gift from R. Reed, Harvard Medical School); monoclonal antihuman U2AF65, 1/1000; monoclonal antihuman ␤-tubulin, 1/5000 (Sigma). Antibody reactivity was detected by chemiluminescence (Western Lighting TM , PerkinElmer Life Sciences).

RESULTS
Experimental Design-Based on previous optimization efforts (22), 25-mer oligonucleotides were designed to analyze the expression of 86 mRNAs encoding proteins implicated in regulation of pre-mRNA splicing (Fig. 1A). Each mRNA was analyzed with 13 pairs of oligonucleotides, each pair containing one perfect match and a mismatch control with a single transversion at position 13. To detect ϳ100 splicing events, mostly involving alternative cassette exons, occurring in genes related to cell signaling, proliferation, adhesion, and apoptosis (Fig.  1A), 13 oligonucleotide pairs corresponding to constitutive or alternative exons, as well as 11 oligonucleotide pairs spanning alternative splice junctions, were selected (Fig. 1B). Microarrays were generated by light-activated in situ synthesis of the selected oligonucleotides (20).
Total cytoplasmic RNA was isolated from cell lines derived from Hodgkin lymphoma tumors at stages IIIb (HD-MY-Z), IV (HDML-2), or IVb (L-540) and from a non-tumor B cell line (LCL-H0) as a reference. After reverse transcription using an oligo(dT)-T7 RNA polymerase promoter primer and second strand synthesis, biotin-labeled cRNAs were generated by T7 transcription and hybridized to the microarrays. Hybridizations corresponding to at least six independent RNA isolates were carried out for each of the cell lines analyzed, resulting in ϳ70 independent measurements per cell line and sequence feature analyzed. -Fold changes between Hodgkin and reference cell lines for each sequence feature and their statistical analyses are provided as Supplementary Data. The ratios between alternative and constitutive exons and between the relevant splice junctions represent independent sources of information for analyzing the relative use of alternative splice sites.
Expression of Alternative mRNA Isoforms-The ability of the microarrays to detect different mRNA isoforms was first evaluated by analyzing the expression of p16 and p14 ARF mRNAs, which encode pivotal regulators of the tumor suppressors p53 and Rb (23). p16 and p14 ARF mRNAs are generated from a single locus through alternative sites of transcription initiation and the use of alternative first exons. It was previously reported (24) that expression of p16 is increased in HDML-2 cells, expression of p14 ARF is increased in HD-MY-Z cells, and expression of either gene is undetectable in L-540 cells. The microarray results were fully consistent with these previous observations ( Fig. 2A), indicating that our microarray design is able to distinguish between the expression of alternatively spliced mRNA isoforms.
Analysis of ϳ100 splicing events documented 20 -30% changes in splicing between the different Hodgkin cell lines and the non-tumor B cell line, which showed various levels of concordance among the different tumor lines. This is illustrated in Fig. 2B by the complex pattern of alternative splicing of the gene encoding the cell adhesion molecule CD44, which has been implicated in tumor metastasis (25). Unsupervised clustering analysis of data obtained from 16 RNA preparations corresponding to four independent RNA isolates/cell line showed appropriate grouping of the samples according to the cell line and closer clustering of cell lines derived from grade IV tumors versus grade III and non-tumor B cell lines (Fig. 3). Table I summarizes the statistical analysis of the data. Taken together, the results indicate that the analysis of splicing patterns allows us to distinguish Hodgkin cell lines from the non-tumor B cell line and to group together cell lines derived from the same tumor grade.

Expression of Splicing Factors and Regulators-Changes in
the expression of spliceosomal components and other proteins related to splicing regulation were also detected (Table II). These include members of the hnRNP and SR protein families, snRNP components such as SAP155, SAP62, and PSF, splicing co-activators such as SRm 160 and 300, and neuron-specific splicing regulators such as Nova-1 and -2. A subset sample of these changes was verified at the protein level (Figs. 4 and 5). 90% of the microarray data that were tested at the RNA and/or protein level were consistent with the results obtained by independent experiments and/or literature reports. Not unexpectedly, microarray and other expression data were not always comparable. For example, microarray data detected only a putative increase in expression of the U5-associated factor PSF (ptbasf in the microarray) in HD-MY-Z cells, whereas increases were experimentally observed in the three Hodgkin cell lines (Fig. 4C).
Ectopic Expression of Nova Proteins-The neuron splicing regulator Nova-1 was initially discovered as a self-reactive antigen expressed in lung and breast tumors (26), and the presence of antibodies in serum correlates with the development of paraneoplastic opsoclonus myoclonus ataxia, a motor control disorder (27). A related protein, Nova-2, is also recognized by the sera of paraneoplastic opsoclonus myoclonus ataxia patients and is normally expressed in the central nervous system with a distribution distinct from that of Nova-1 (28). The microarray data indicated expression of Nova mRNAs in the grade III HD-MY-Z cell line and expression of Nova-2 in grade IV HDML-2 cells (Table II). These patterns of expression were confirmed at the transcript and protein level (Fig. 5, A   FIG. 3. Unsupervised clustering analyses of microarray data. Microarray results for splicing events (exons/splice junctions) for 16 RNA preparations were clustered using Gene Spring 6.2 software. Each column represents an RNA sample, and each row represents an exon/ splicing event. Colors vary from blue to red with increasing signal intensity and from dark to bright with the confidence on the value observed in multiple measurements. Colors at the bottom row and in dendrograms represent the different cell lines, as indicated. and B). Analysis of other Hodgkin cell lines and tumor samples confirmed expression of Nova-2 in another grade IV cell line and in malignant Hodgkin and non-Hodgkin lymphoma tumors (Fig. 5, B and C).
Recent results have identified pre-mRNAs containing Nova-2 binding sites, the alternative splicing of which was altered in Nova-2 knock-out mice (17). One of these, the JNK2 pre-mRNA, is alternatively spliced to generate mRNA isoforms that encode kinases that play key roles in cell signaling, pro-liferation, and apoptosis and that have different affinities for their transcription factor substrates (29).
A prediction of our results is that expression of Nova-2 in the HDML-2 cell line could alter JNK2 alternative splicing. The results of Fig. 5D verify this prediction as they show a significant enrichment of the Nova-2-induced JNK2 exon 6a in HDML-2 cells as compared with the reference cell line. Quantification of the results from five different experiments indicated a 2-3-fold increase in exon 6a utilization in Nova-2- I Statistical analysis of -fold changes in splicing in Hodgkin cell lines For each pair of Hodgkin lines, the fold differences in the use of exon/exon junctions between each cell line and the reference B cell line are compared. Similar behavior, fraction of instances in which changes, or absence of changes, are similar between the compared cell lines. Similar discriminatory behavior, fraction of instances in which similar behavior is found for the two cell lines being compared, whereas different behavior is found for the third.   expressing HDML-2 cells, a slight increase in Nova-1expressing HD-MY-Z cells, and no effect in L-540 cells, which do not express either protein (Fig. 5E). These results further argue that expression of Nova proteins is sufficient to induce neuron-enriched patterns of splicing in non-neural cells. DISCUSSION Given the prevalence and relevance of alternative splicing in higher eukaryotes, it seems likely that many aspects of the phenotype of a cell, including those leading to tumor progression, are controlled by the relative expression of alternatively spliced isoforms of multiple genes. Our results indeed suggest that general patterns of alternative splicing, as well as expression of particular splicing regulators, can be indicative of cell transformation and of tumor grade. If this is the case, microarrays able to discriminate between alternatively spliced isoforms will obviously have important medical applications.
The fraction of changes observed between Hodgkin cell lines and the reference cell line is substantial, affecting 20 -30% of the splicing events analyzed and showing a 2-3-fold better correlation for cell lines derived from tumors of the same grade (Table I). In the vast majority of cases, these changes are exonor splice junction-specific, indicating that they do not correspond to a general increase or decrease in the output of a gene but that they represent bona fide changes in splicing rather than transcription. Intriguingly, some of the changes detected affect exons described previously as constitutive (e.g. in the CD21 gene), even when no changes were observed in other regions of the pre-mRNA described previously as alternatively spliced. This is consistent with the suggestion that significant, albeit not general, changes in the mechanisms of exon definition occur in tumor cells (7) Given that 1) changes in the relative expression of general splicing factors can modulate alternative splicing of multiple genes (15) and 2) changes in the relative levels of general splicing factors are observed in different cell types or disease states (30,31), it has been postulated that the relative expression of splicing factors acts as a "cellular code" that establishes patterns of alternative splicing in multiple genes (13,14). Clustering analyses similar to those shown in Fig. 3, however, did not show the same degree of correlation between tumor grade and overall changes in the expression of splicing factors as they did with changes in splicing (data not shown). Similarly, splicing events were 5-fold more frequent than splicing factors in the list of array features that provided the best discrimination between cell lines (GeneSpring 6.2 analysis of variance software, data not shown). It is possible that global changes in splicing factor concentrations are not as predictive of tumor grade as the expression of isoforms of genes involved in cell cycle control, signaling, adhesion, etc., and therefore, that changes in a relatively small number of splicing regulators, or in their activities via post-translational modifications, are responsible for changes in a significant number of alternative splicing events. Alternatively, because a large fraction of splicing factor genes are themselves alternatively spliced, mRNA-based analysis of their expression may not be as predictive as changes in the expression of individual isoforms. Microarrays designed to detect expression of alternatively spliced isoforms of splicing factors will help test this possibility.
Nevertheless, changes in splicing regulatory proteins were also detected. An interesting observation was that individual components of splicing complexes appeared to be regulated independently of other partners. This was the case for the 65-kDa subunit of the splicing factor U2AF (Fig. 4A) and for the U2 snRNP component SAP155 (Fig. 4B), the levels of which were increased without parallel increases in U2AF 35 or some other components of U2. It is conceivable that a U2AF complex devoid of its 35 K d subunit causes changes in 3Ј splice site recognition (32) and that an excess of SAP155, which has repeats that are targets of cdk2 phosphorylation (33), can titrate the activity of this kinase and influence cell cycle progression.
Ectopic expression of Nova proteins was originally reported in lung and breast tumors, leading to the production of autoantibodies that are thought to be the cause for the subsequent development of motor disorders in these patients (26,27). Our microarray analyses revealed expression of Nova-1 and -2 in some Hodgkin cell lines, observations that were independently confirmed and extended to other Hodgkin cell lines and tumor samples (Fig. 5). These findings could provide pathophysiological basis for explaining the clinical development of paraneoplastic opsoclonus-myoclonus in some Hodgkin patients (34).
Although Nova proteins have been implicated in neuronspecific splicing regulation (16,17), changes in alternative splicing caused by ectopic expression of these proteins in tumors or tumor cell lines, with potential effects in the regulation of cell proliferation, had not been reported. Our observation that expression of Nova-2 correlated with increased expression of neuronal isoforms of JNK2, together with the key roles that these proteins play in cell signaling, proliferation, and apoptosis (29,35), are compatible with the possibility that Nova-2 expression causes changes in alternative splicing that influence tumor progression.
In summary, although larger data sets and powerful computational tools will be required for full exploitation of this microarray design, the results reported in this study have provided new insights of potential medical relevance, including an explanation for clinical reports of motor disorders in Hodgkin patients and a correlation between global changes in alternative splicing and cell transformation and tumor grade.