Background. Achievement of CR is generally associated with improved clinical outcomes for patients (pts) with MM and represents a primary endpoint of current clinical trials. The GIMEMA Italian Myeloma Network designed a phase 3 study to demonstrate that the triplet VTD regimen was superior over a doublet such as thalidomide-dexamethasone (TD) as induction therapy prior to double ASCT for newly diagnosed MM. On an intention-to-treat basis, the rate of complete or near complete response (CR/nCR) was 31% for the 236 pts on VTD induction therapy, while it was 11% (p<0.0001) for the 238 pts on TD induction therapy. Since enhanced rates of CR/nCR affected by VTD incorporated into ASCT resulted in extended progression-free survival, prediction of CR by pharmacogenomic tools is likely to be an important goal to prospectively select those pts who are more likely to benefit from a given therapy.
Methods. For this purpose, in a molecular substudy to the main clinical study we assessed the ability of gene expression profile (GEP) to predict attainment of CR/nCR in 122 pts enrolled in the VTD arm of the study. Their characteristics at baseline, including cytogenetic abnormalities, were comparable with those of the whole population of 236 pts. Highly purified CD138+ plasma cells were obtained at diagnosis from each of these pts and were profiled for gene expression using the Affymetrix U133 Plus2.0 platform. In order to build a low-dimensional signature with optimal performance, genomic data were analyzed with an original algorithm that exploits quadratic discriminant analysis with a bottom-up approach that builds N-gene signatures starting from two-dimensional signatures. Gene models were applied to test datasets to predict achievement of either CR/nCR or less than nCR, and classification performances were validated by a leave-one-out crossvalidation procedure.
Results. Thirty four pts out of the 122 (28%) who were included in the present analysis achieved a CR/nCR, while the remaining 88 patients failed this objective. The molecular approach described above allowed to identify several gene signatures among which we choose a 163-gene signature that provided a predictive capability of 79% sensitivity, 87% specificity, 71% positive predictive value (PPV) and 92% negative predictive value (NPV). These expression values were used in an unsupervised hierarchical clustering to stratify the population of 122 profilated pts into 3 well defined subgroups. Seventy nine pts were included in subgroup A, while the remaining 43 pts were included in either subgroup B (n=22) or subgroup C (n=21). Notably, 19 out the 34 CR/nCR pts (56%) clustered in subgroup B, whereas the remaining 15 pts were randomly distributed within subgroup A. Analysis of demographic and disease characteristics of the pts belonging to the 3 major subgroups, revealed that in subgroup B the frequencies of pts carrying del(13q) (78%) or del(17p) (22%) or with an IgA isotype (54%) were significantly higher in comparison with the corresponding values found in subgroup A (47%, 4%, and 10%, respectively) and subgroup C (38%, 10%, and 5%, respectively).
In order to obtain a more feasible set of genes predictive of CR/nCR, several smaller signatures originating from the 163-gene signature were further analyzed by means of the same algorithm described above. The best predictive capability was obtained with a 41-gene signature that provided 88% sensitivity, 97% specificity, 91% PPV and 95% NPV.
A GeneGo ® network analysis of genes included in the signatures showed that the most relevant network nodes included tumour suppressor genes (FBXW7 and MAD), genes involved in inflammatory response (TREM1 and TLR4) and genes involved in B cell development (IKZF1, IL10 and NFAM1). Genes included in the signatures do not gather in specific chromosomes, thus confirming the absence of bias on selection of signatures genes, potentially due to prevalence of MM typical chromosomal aberrations.
Conclusions. GEP analysis of a subgroup of pts who received VTD induction therapy allowed to provide a 41-gene signature that was able to predict attainment of CR/nCR and, conversely, failure to achieve at least nCR in 91% and 95% of cases, respectively. These favorable results might represent a first step towards the possible application of a tailored therapy based on the single patient’s genetic background.