Prognostic Impact of Genetic Subgroups and Development of Gene Classifiers for Response, PFS and OS In Multiple Myeloma Patients Treated with Bortezomib or Conventional Agents In HOVON65/GMMG-HD4 Trial

In newly diagnosed myeloma patients, bortezomib treatment induces high rates of complete response (CR) and very good partial response (VGPR). Recently, we published the clustering of gene expression profiles in 320 MM patients, who were included in a large prospective, randomized, phase III transplantation trial with bortezomib (PAD) versus conventional vincristine (VAD) based induction treatment (HOVON65/GMMG-HD4). We identified 12 distinct subgroups CD-1, CD-2, MF, MS, PR, HY, LB, Myeloid, including three novel defined subgroups NFκB, CTA, and PRL3 and a subgroup with no clear gene expression profile (NP).

To look at the prognostic impact of these 12 clusters in the trial and group clusters together into a high risk (HR) and low risk (LR) group in the different treatment arms. Furthermore, to define a high risk signature to identify the patients at increased risk of disease progression.

Gene expression profiles of myeloma cells obtained at diagnosis of 320 HOVON65/GMMG-HD4 patients were available. Response, progression free survival (PFS) and overall survival (OS) data were available for the first 628 patients, resulting in analysis of gene expression in relation to prognosis in 229 patients. The prognostic impact of the genetic subgroups separately and grouped into high and low risk were evaluated using Kaplan Meier and Cox regression analysis using exhaustive search (R). For the high risk gene signature the HOVON65 gene expression data was used as training set with PFS as outcome measure. Two independent myeloma datasets with survival data were used as an external validation, UAMS (GSE2658) and APEX (GSE9782)). The signature was generated by a Cox proportional hazard model in combination with LASSO (Least Absolute Shrinkages and Selection Operator) for simultaneous parameter estimation and variable selection using the R package glmnet. ISS stage was implemented by adjusting the individual covariant penalization factors of the LASSO.

The highest CR+nCR rates were found in the PRL3 and NP clusters, i.e. 78% and 86%, respectively (VAD), and 100% (PAD). The lowest CR+nCR rate was 17% in the CD1 cluster (PAD) and 0% in the CD2, MF and PR clusters (VAD). Based on the impact of clusters on PFS and OS in the VAD arm, the MS, MF, PR and CTA clusters were included into a High Risk (HR) group. This HR group showed a median PFS of 13 months and OS of 21 months vs. the Low Risk (LR) group consisting of the remainder of clusters with a median PFS of 31 months and a median OS not reached (P<.001).In contrast, in the PAD arm, only the PR cluster conferred a poor prognosis and exclusively formed the HR group, showing a median PFS of 12 months and OS of 13 months vs. the LR groups consisting of the remainder of clusters with a median PFS of 32 months and a median OS not reached (P=.004). In the poor prognostic subgroup MF, a striking difference in outcome on PAD vs. VAD was observed, i.e. 24 vs. 3 months (PFS) and 39 vs. 4 months (OS), respectively. In the multivariate analysis, including the covariates recurrent translocations, 17p deletion, 13q loss, LDH and ISS stage, the HR group definition remained independent poor prognostic indicators for both treatment arms. A LASSO-based high-risk signature was generated (48 probe sets). External validation was performed in the data sets UAMS TT2 and APEX. In both studies, the high-risk signature identified a high-risk population of 13% with highly significant log rank p-values (7.7*10⁻⁸ and 4.2*10⁻⁹, respectively). Combination of all validation data, n=615, corrected for type of treatment and study, resulted in an overall Cox proportional hazard model's hazard ratio of 3.8 with p <2*10⁻¹⁶.

Distinctive gene expression clusters affect prognosis, which differ depending on treatment. In the conventional treatment arm (VAD), MS, MF, PR and CTA clusters confer a worse prognosis while with bortezomib based treatment, only the PR cluster affects prognosis negatively. These HR groups remain independent poor prognostic indicators. Based on the HOVON65/GMMG-HD4 study population, a high-risk signature was generated with strong and highly significant predicting ability in two independent data sets.

Mulligan:Millenium: Employment. Goldschmidt:Celgene: Membership on an entity's Board of Directors or advisory committees; Ortho Biotech: Membership on entity's Board of Directors or advisory committees; Ortho Biotech: Research Funding; Celgene: Research Funding; Chugai Pharma: Research Funding; Amgen: Research Funding. Sonneveld:Ortho Biotech: Research Funding; International Myeloma Foundation: Research Funding; Ortho Biotech: Consultancy.