Summary

An analysis of the MIDAS trial linked molecular features of newly diagnosed multiple myeloma to measurable residual disease after quadruplet induction. A model combining genomic, gene-expression and DNA-methylation data achieved an AUC of 0.77 in a held-out group from the same trial.

Researchers analysing the phase 3 MIDAS trial found that molecular features of newly diagnosed multiple myeloma were associated with whether cancer cells remained detectable after quadruplet induction therapy. Their model combined whole-genome sequencing, RNA sequencing and DNA-methylation data and achieved an area under the curve (AUC) of 0.77 in a held-out group of 120 patients from the same trial.

The study, published in Nature Medicine, examined response to Isa-KRd: isatuximab, carfilzomib, lenalidomide and dexamethasone. Measurable residual disease (MRD) is a sensitive assessment of remaining myeloma cells after treatment. Here, MRD negativity was assessed at a sensitivity of one cancer cell in a million (10⁻⁶).

Contents

Molecular patterns tracked with MRD response

The researchers profiled 423 tumour samples using genomic, transcriptomic and epigenomic methods. The MRD analyses were limited to enrolled trial participants with an available assessment: 368 of 394 enrolled patients. After induction, 44.8% of those assessed were MRD-negative at 10⁻⁶.

Response rates varied across gene-expression groups. HP1, HP3 and MS groups had higher MRD-negativity rates, while CD2 and MF groups had lower rates. The analysis also linked a B-cell differentiation programme with MRD positivity, and proliferation and MYC-related programmes with MRD negativity. These are associations within the study, not evidence that any one molecular programme causes a treatment response.

Several transcriptomic findings were reproduced in RNA-sequencing data from 257 patients in the CASSIOPEIA trial. That cohort received a different quadruplet regimen, D-VTd, so the replication supports the relevance of the molecular patterns across trial settings while involving a different treatment combination.

A model combining three data layers

For the prediction analysis, the team used 361 patients with complete multiomic data and MRD status. They assigned 241 to a training set and reserved 120 as a held-out validation set. A random-forest model integrated genomic alterations and mutational signatures with gene-expression and methylation features.

In the held-out set, the combined model achieved an AUC of 0.77 (95% confidence interval 0.67–0.85) and 71% accuracy. AUC measures how well a model distinguishes between two outcomes across possible decision thresholds; 0.5 corresponds to chance-level discrimination and 1 to perfect discrimination. The multiomic model had higher discrimination than the clinical and genomic models tested. Its AUC was not significantly different from the transcriptomic model (0.73) or the epigenomic model (0.75), although its calibration was better than either of those single-layer models.

What the results mean for treatment research

The findings show how combining molecular measurements can capture differences in early response that are not represented by clinical features or genomic data alone. The study also identified nine methylation groups associated with MRD response, with differences in the extent of DNA hypomethylation, including in regions of largely inactive chromatin.

The prediction results are internal to MIDAS: the held-out patients came from the same trial cohort used to develop the model. The authors say validation in independent cohorts is needed to assess whether its performance generalises. The analysis concerns MRD after induction; whether these early differences correspond to long-term outcomes in MIDAS will require follow-up. MRD is an early response measure associated with later outcomes in multiple myeloma, but this study did not report those long-term results.

Sources