Expression Profilingin DLBCL

The two major platforms for expression profiling have been used in DLBCL and have produced remarkable and reproducible expression signatures that reveal molecularly defined subtypes of DLBCL and predict for response to treatment and long-term outcome (12-14).

In an article by Shipp et al. (14), diagnostic, pretreatment, frozen biopsy material was studied from 58 patients with DLBCL and 19 patients with folli-cular lymphoma, a closely related germinal centre malignancy. The expression profile, or "signature," of each biopsy was created using the Affymetrix 7800 gene chip, which consisted of probe features representing 7817 known and putative genes. The resulting gene expression data was analyzed using supervised learning techniques (15). The first distinction attempted, based on expression variance, was between the DLBCL and follicular lymphoma samples. This proved possible, and the resulting 30-gene model's reproducibility was confirmed by the "leave-one-out" cross-validation technique. Next, subdivision of the DLBCL samples into different subtypes based upon outcome to treatment was examined, again using a weighted-voting algorithm (16). A recurring subset of genes was indeed discovered that were highly associated with the distinction between those patients that had a good outcome (i.e., alive and cured yr after treatment) and those with a bad outcome (i.e., dead from lymphoma or DLBCL unresponsive to treatment) A model using just 13 genes could predict outcome for the 58 patients, with 70% of the patients stratified in the good-risk group being alive at 5 yr vs just 12% of the patients placed in the bad-risk group remaining alive at 5 yr (p = 0.00004).

An interesting discovery was that the 13-gene model could subdivide the patients contained within the different IPI subgroups into "cured" and "dead/ refractory" subgroups with significantly different outcomes, revealing that the gene expression-based outcome predictor contained additional information to the clinical prognostic model, the IPI. For example, when the 37 patients defined as low/low-intermediate risk according to the IPI were sorted by the 13-gene model, those displaying a "cured" gene-expression signature had a 5-yr overall survival of 75% compared with a 5-yr overall survival of 32% for the patients displaying the "fatal/refractory" signature (p = 0.02) The combination of clinical and molecular variables suggests a possible strategy for further individualization of patient treatment decisions. Nevertheless, additional information still remains to be captured because the use of both models in series failed to produce groups with 100% and 0% 5-yr overall survivals. Finally, the 13-gene model suggested novel therapeutic targets and strategies. Whether therapeutic leads suggested by expression profiling will prove useful and result in molecularly determined individualization of treatment remains to be seen.

In the other two articles, the clustering analysis was performed using a different approach termed hierarchical clustering (17). Alizadeh et al. (12) demonstrated that molecular subclassification of DLBCL on the basis of gene expression was possible. The two subtypes had expression profiles similar to those of different putative cells of origin. This finding was confirmed and extended in the much larger series of Rosenwald et al. (13), in which DLBCL was divided into three subtypes termed germinal center-like, activated B-cell-like and type 3 DLBCL. In both articles, the patients with lymphomas subclas-sified as germinal center-like fared better than the patients with lymphoma subclassified as activated B-cell-like, suggesting clinical relevance to DLBCL subclassification according to the putative cell of origin.

Was this article helpful?

0 0

Post a comment