Gene Expression Profiling (GEP) of CD138-Purified Plasma Cells: Modeling for High LDH and Cytogenetic Abnormalities as Independently Adverse Features of Multiple Myeloma (MM) In Total Therapy (TT) Protocols

The prognosis of patients with MM is best captured by GEP-defined risk, distinguishing ~15% of patients with a median overall survival (OS) in TT2 of ~2yr as opposed to >10yr in low-risk disease. Serum LDH elevation has remained an independent adverse feature, along with the presence of metaphase cytogenetic abnormalities (CA). Here we examined whether GEP features within CD138-selected plasma cells can identify MM-associated serum LDH elevation and CA.

The training set consisted of 621 cases and the test set consisted of 325 TT patients who had both baseline clinical and GEP data from whole genome Affymetrix U133Plus2.0 microarrays on CD138-enriched plasma cells. Using the training data and the scoring approach described in Shaughnessy et al (2007), we defined a GEP score based on 50 genes to predict serum LDH >190U/L and a GEP score with 15 genes to predict the presence of CA. The gene scores were then tested in univariate and multivariate Cox regression models to assess their clinical utilities in both training and test sets.

In the training set, sensitivity and specificity of the GEP 50-gene score for serum LDH >190 were 65% and 76%, respectively; those of the GEP 15-gene score predicting CA were 75% and 78%, respectively. While both GEP scores were significantly associated with overall and event-free survival (p < .001), they were not selected in multivariate stepwise Cox regression analysis. However, the gene scores seem comparable to the clinical variables they try to predict. When serum LDH >190 and CA were replaced by the GEP scores in multivariate models, both gene scores were significant at the .1 level, and the R-squared (R2) statistic decreased by only 2.4% (Table 1). In the test set, sensitivity and specificity of the GEP 50-gene score for serum LDH >190 were 59% and 71%, respectively. Those of the GEP 15-gene score for CA were 58% and 70%, respectively. The GEP scores behaved similarly overall as in the training set with the LDH gene score showing a little higher association than the CA gene score (p<.01 for the GEP CA score and p<.001 for the GEP LDH score). In multivariate stepwise Cox regression, only CA was selected, and when CA was replaced with the GEP CA score (or GEP LDH score), the R-squared statistic decreased by 4.6% (or 1.9%).

In summary, we conclude that the GEP scores for LDH and CA are highly comparable to their corresponding clinical variables in terms of survival prediction capability. Our future research will focus on refining our gene selection procedure to achieve higher prediction accuracy, studying the molecular features of selected genes and, based on GEP data, building a predictive model that is at least as powerful as if only clinical variables were used in the model.

No relevant conflicts of interest to declare.