Multiomic profiling separates multiple myeloma response to Isa-KRd induction
In newly diagnosed multiple myeloma, the same quadruplet regimen does not produce the same biologic effect across every tumor. Integrated genomic, transcriptomic and epigenomic profiling of 423 cases from the MIDAS trial mapped a strikingly heterogeneous response to induction with isatuximab, carfilzomib, lenalidomide and dexamethasone (Isa-KRd), with measurable residual disease outcomes differing by molecular subtype, pathway activity and patterns of genomic instability.
The analysis drew on three layers of tumor biology at once: whole-genome sequencing, RNA sequencing and whole-genome enzymatic methylation sequencing. This matters because myeloma is not a single disease entity but a collection of plasma-cell neoplasms driven by distinct initiating lesions, transcriptional states and epigenetic programs. A regimen can clear one molecular state with ease and leave another comparatively intact, even when the two appear similar at diagnosis by routine clinicopathologic criteria.
Transcriptomic subtype predicted who cleared disease to MRD negativity
Unsupervised gene-expression clustering separated the cohort into eight molecular groups that aligned with known myeloma biology, including translocation-defined MS, MF and CD groups and three hyperdiploid subgroups, HP1 through HP3. The hyperdiploid classes were not interchangeable. HP1 carried a proliferative program with elevated CCND2 and E2F target activity, HP2 showed a stronger B-cell identity signal, and HP3 expressed inflammatory chemokine and interferon-related genes.
Postinduction MRD negativity at 10−6 was achieved in 44.8% of the 368 enrolled patients with available assessment, rising to 59.8% at 10−5. The response distribution was uneven across molecular groups, and the separation persisted after clinical adjustment. The lowest MRD-negativity rates clustered in the CD2 and MF groups, while the highest rates appeared in HP1, HP3 and MS disease.
The biology fits the bedside pattern. Tumors with stronger B-cell differentiation programs retained a state closer to normal lineage identity, and those cases were less likely to reach deep molecular remission. By contrast, tumors with higher MYC activity and proliferation signals were more often MRD negative after treatment, suggesting that proliferative dependence may create a vulnerability to an intensive quadruplet platform. In practical terms, the molecular wiring of the plasma cell appears to shape how fully frontline therapy can extinguish residual clones.
B-cell differentiation tracked with residual disease, while MYC-linked programs tracked with clearance
Comparing MRD-positive and MRD-negative cases identified 170 genes upregulated in residual disease, including driver genes such as CCND1 and MAFA plus multiple B-cell differentiation markers, including PAX5. The MRD-negative state showed a different signature altogether: 217 genes were more active, enriched for epigenetic modifiers, vesicle trafficking and exocytosis genes, cell-cycle regulators and MYC targets.
A key pattern ran through the transcriptomic data. B-cell differentiation was strongest in the CD1/CD2 and HP2 groups, the same groups with lower MRD-negativity rates. MYC pathway activation clustered in the highly MRD-responsive HP1 and HP3 groups, but also in MF disease, where MYC alteration frequency was also high. Proliferation scores and MYC activity were tightly linked and both ran opposite to the B-cell differentiation score, forming a biologically coherent axis of response resistance and response sensitivity.
Independent modeling showed that pathway scores added information beyond the subgroup labels themselves. Including these pathway measures improved model fit and partly weakened the effect of subgroup assignment, indicating that the gene programs explain a meaningful portion of the subtype-specific MRD differences. The subgroup effect did not disappear, which argues against a single dominant mechanism and points instead to layered biology: lineage state, proliferative circuitry and oncogenic activation all contribute to whether Isa-KRd achieves deep clearance.
Genomic instability and driver lesions also separated response classes
Whole-genome sequencing identified established and newly described genomic instability phenotypes. APOBEC mutagenesis and chromothripsis were confirmed, and new instability patterns were validated in an external cohort. Across 1,451 cases drawn from three independent cohorts, new driver alterations also held up, including associations involving CCND1 and MYC.
These lesions correlated with differential MRD outcomes. The implication is straightforward: genomic instability is not just a marker of disease aggressiveness at diagnosis; it also appears to mark tumors that respond differently to frontline immune- and proteasome-based therapy. When the tumor genome has already been shaped by catastrophic rearrangement or sustained mutagenic processes, the residual clone may carry biologic features that make it harder to eradicate completely.
One sentence deserves to stand alone. The residual disease signal was not random noise; it tracked with definable tumor biology.
Nine methylation subgroups added an epigenetic layer to MRD prediction
DNA methylation profiling identified nine MRD-linked subgroups. They were separated by the degree of hypomethylation in quiescent chromatin and by cis and trans signatures of driver translocations. This is a useful reminder that myeloma response is not governed only by DNA sequence or RNA output. Chromatin state can alter which transcriptional programs remain accessible under treatment pressure, changing whether a malignant plasma cell persists or disappears below assay detection.
The methylation data extended the subtype framework beyond the transcriptome. Tumors with different chromatin settings may respond differently even when they share a driver translocation or similar baseline clinical staging. For frontline treatment selection, that means a single diagnostic biopsy can contain clues about depth of response that are invisible to standard risk stratification.
A multimodal model outperformed single-layer prediction
A random forest model integrating genomic, transcriptomic and epigenomic features predicted MRD status after induction with an area under the curve of 0.77. The multimodal approach outperformed clinical and genomic models, while offering performance comparable to transcriptomic and epigenomic models and better calibration than those single-modality approaches.
This is the most clinically consequential part of the analysis. AUC 0.77 is not a guarantee of individual-level certainty, but it is a meaningful step toward response-adapted stratification in newly diagnosed disease. The model suggests that no single molecular layer fully captures the biology of early treatment response. Sequencing alone misses the functional state of the cell, RNA alone misses genomic architecture, and methylation alone misses much of the underlying driver biology. The integrated model gains power because it sees all three.
The cohort also included 421 evaluable whole-genome profiles, 418 RNA-seq profiles and 422 methylation profiles, with 416 samples carrying all three data types. Five normal plasma-cell samples provided reference expression and methylation landscapes. Among screened patients, 394 were ultimately enrolled, and the MRD analyses were confined to the enrolled group.
Response-adapted therapy now has a molecular framework to test
MIDAS was built around a response-adapted strategy after uniform quadruplet induction, and this ancillary study gives that strategy a biological backbone. The results do not argue that MRD should be abandoned as an endpoint. They argue the opposite: MRD becomes more informative when interpreted alongside a tumor’s molecular state.
For patients with newly diagnosed disease, the immediate clinical consequence is not a new standard of care. The data instead point toward a future in which MRD negativity may be predicted, and perhaps eventually manipulated, by combining routine clinical staging with molecular subtype, proliferative signature, translocation biology and methylation context. The current findings define a multiomic classification of newly diagnosed multiple myeloma linked to Isa-KRd response, but they do not yet establish how best to use that classification to alter treatment in real time.
The unanswered practical question is narrower and more urgent: which of these molecular features will remain reproducible enough, and actionable enough, to guide frontline decisions in the clinic rather than the research lab?