72 ALIZADEH ET AL. receptor to bind the antigen. The centrocytes that can still bind antigen are rescued from programmed cell death and may either re-enter the proliferative centroblast state or differentiate into memory B cells or plasma cells. The current evidence that human lymphomas are related to the germinal-center B cell derives from sequence analysis of the immunoglobulin genes in these malignant cells (for review, see Klein et al. 1998). In all types of non-Hodgkin’s B-cell lymphomas that have been studied, the immunoglobulin genes show mutations that are entirely consistent with the pattern and frequency of mutations found in normal germinal-center B cells and in postgerminal-center memory and plasma cells. The process of somatic hypermutation of immunoglobulin genes is normally confined to germinal-center B cells, although in some mutant mouse strains that lack germinal centers, somatic mutation can still occur (Matsumoto et al. 1996). Thus, the presence of somatic mutations in the immunoglobulin genes of B-cell lymphomas suggests, but does not prove, that the malignancy was derived from a cell that had passed through the germinal-center microenvironment. In some types of lymphomas, notably follicular lymphoma and mucosal-associated lymphoid tissue (MALT) lymphoma, the immunoglobulin sequences show a pattern of ongoing somatic mutation within a single tumor biopsy (Bahler and Levy 1992; Du et al. 1996; Bahler et al. 1997; Qin et al. 1997). This could be interpreted as evidence that the malignant transformation event occurred while the cell was within the germinal-center milieu. Indeed, some Burkitt’s lymphoma cell lines retain the ability to somatically hypermutate their immunoglobulin genes during in vitro culture (Denepoux et al. 1997; Sale and Neuberger 1998). Despite these considerations, the extent to which malignant human B cells resemble normal B-cell subpopulations with respect to gene expression is currently unknown. Clearly, the “inheritance” of a gene expression profile by a malignant cell from its normal B-cell counterpart could influence the clinical behavior of the malignancy. Conversely, the signaling pathways that are altered during malignant transformation will themselves yield distinctive gene expression signatures within the malignancy. To address these issues on a genomic scale, we have turned to cDNA microarray analysis of gene expression, which is capable of simultaneously quantitating the expression of more than 20,000 genes. The particular version of microarray gene expression analysis that we use starts with the robotic “spotting” of polymerase chain reaction (PCR) products derived from cDNA clones of interest in an ordered array on a glass microscope slide (Fig. 1A). The resultant microarrays are then used to measure the relative expression of each spotted gene between two mRNA samples. mRNA from each sample is used to generate total first-strand cDNA probes, with each probe incorporating a different fluorescent dye during the cDNA synthesis. The two fluorescent probes are mixed and hybridized to the same cDNA microarray under stringent hybridization conditions. Following washing steps to remove nonhybridized probe, the hybridization of the two probes to the microarray is detected using scanning confocal laser microscopy. Digital image analysis is used to determine the ratio of hybridization of the two probes to each cDNA spot. This ratio has been shown to agree well with more standard methods of gene expression analysis such as Northern blot hybridization and quantitative reverse transcriptase (RT)-PCR (DeRisi et al. 1996; Iyer et al. 1999). Faced with the …