Risk-stratified patient selection in multiple myeloma for frontline treatment decision-making based on a transcriptomic classifier
Authors
Maria Kuzmicheva, Karen Baghdasaryan, Stanislav Kurpe, Andrey Kravets, Evgenia Alekseeva, Nazar Aryutyunian, Oleg Baranov, Eduardo Shugaev-Mendosa, Konstantin Chernyshov, Nikita Kotlov
- BostonGene Corporation, Waltham, MA, USA
Abstract
Background:
The treatment paradigm for newly diagnosed multiple myeloma (NDMM) is rapidly evolving, with daratumumab-based quadruplet regimen established as standard of care and CAR-T and BiTEs therapies moving into the frontline setting. Current clinical risk assessment tools—including the International Staging System (ISS) based on biochemical and cytogenetic markers, as well as transcriptome-based classifiers such as SKY92 and CoMMpass—identify high-risk populations but lack a unified framework to guide risk-stratified frontline treatment.
Aims: To address this gap, we integrated SKY92 and CoMMpass classifications using NGS, to develop a consensus model for risk stratification. This model refines risk categories and identifies patients who may benefit from intensive frontline strategies, including BiTE combinations or CAR-T therapies.
Methods:
We used the CoMMpass (N=715) and SKY92 (N=282) cohorts to develop transcriptomic signatures that identify tumors with a highly aggressive proliferative phenotype. Genes strongly associated with poor survival and high inter-gene correlation were selected. Signature scores were computed using a modified single-sample Gene Set Enrichment Analysis (ssGSEA) approach and integrated into a consensus logistic regression model to yield a proliferative risk (PR) score. This PR-based model stratifies patients into three groups: high risk (worst survival), intermediate risk, and low risk (best survival). Validation was performed using the GSE136324 and GSE2658 cohorts. Progression-free survival (PFS) and overall survival (OS) were assessed with Kaplan–Meier estimates.
Results:
Compared with conventional classifiers (SKY92 and CoMMpass), our consensus PR-based model provides more granular risk stratification, identifying both high-risk and intermediate-risk patients. While CoMMpass and SKY92 classified 8% and 20% of patients as high-risk, respectively, our integrated model identified 10.8% (n=76) as high-risk and 20.1% (n=142) as intermediate-risk. These subgroups were significantly associated with inferior PFS and OS (p<0.001). On the test dataset, the model achieved ROC > 0.9 for predicting OS and PFS. Validation on an independent dataset confirmed platform-independence, with risk classification remaining significantly prognostic for OS and PFS. Combining the PR-based model with ISS further enhanced risk stratification, identifying distinct intermediate-risk subgroups within ISS standard-risk categories. Furthermore, we identified an adaptive Unfolded Protein Response (UPR) signature in the high-risk subgroup, derived from co-expressed genes involved in selective transcription and RNA quality control. Notably, the UPR score increased progressively across risk groups (p<0.001), linking high proliferative risk with an adaptive stress response that promotes proteasome inhibitor resistance.
Summary/Conclusion:
Our consensus PR-based model, combined with ISS, outperforms conventional classifiers by precisely defining high- and intermediate-risk NDMM subgroups. The identification of an adaptive UPR signature in high-risk patients highlights a biological mechanism linked to treatment resistance. These findings support evaluation of the PR-based model in patients receiving daratumumab-based regimens and suggest that high-risk patients may benefit from more intensive frontline strategies, such as BiTE combinations or CAR-T therapies.
The treatment paradigm for newly diagnosed multiple myeloma (NDMM) is rapidly evolving, with daratumumab-based quadruplet regimen established as standard of care and CAR-T and BiTEs therapies moving into the frontline setting. Current clinical risk assessment tools—including the International Staging System (ISS) based on biochemical and cytogenetic markers, as well as transcriptome-based classifiers such as SKY92 and CoMMpass—identify high-risk populations but lack a unified framework to guide risk-stratified frontline treatment.
Aims: To address this gap, we integrated SKY92 and CoMMpass classifications using NGS, to develop a consensus model for risk stratification. This model refines risk categories and identifies patients who may benefit from intensive frontline strategies, including BiTE combinations or CAR-T therapies.
Methods:
We used the CoMMpass (N=715) and SKY92 (N=282) cohorts to develop transcriptomic signatures that identify tumors with a highly aggressive proliferative phenotype. Genes strongly associated with poor survival and high inter-gene correlation were selected. Signature scores were computed using a modified single-sample Gene Set Enrichment Analysis (ssGSEA) approach and integrated into a consensus logistic regression model to yield a proliferative risk (PR) score. This PR-based model stratifies patients into three groups: high risk (worst survival), intermediate risk, and low risk (best survival). Validation was performed using the GSE136324 and GSE2658 cohorts. Progression-free survival (PFS) and overall survival (OS) were assessed with Kaplan–Meier estimates.
Results:
Compared with conventional classifiers (SKY92 and CoMMpass), our consensus PR-based model provides more granular risk stratification, identifying both high-risk and intermediate-risk patients. While CoMMpass and SKY92 classified 8% and 20% of patients as high-risk, respectively, our integrated model identified 10.8% (n=76) as high-risk and 20.1% (n=142) as intermediate-risk. These subgroups were significantly associated with inferior PFS and OS (p<0.001). On the test dataset, the model achieved ROC > 0.9 for predicting OS and PFS. Validation on an independent dataset confirmed platform-independence, with risk classification remaining significantly prognostic for OS and PFS. Combining the PR-based model with ISS further enhanced risk stratification, identifying distinct intermediate-risk subgroups within ISS standard-risk categories. Furthermore, we identified an adaptive Unfolded Protein Response (UPR) signature in the high-risk subgroup, derived from co-expressed genes involved in selective transcription and RNA quality control. Notably, the UPR score increased progressively across risk groups (p<0.001), linking high proliferative risk with an adaptive stress response that promotes proteasome inhibitor resistance.
Summary/Conclusion:
Our consensus PR-based model, combined with ISS, outperforms conventional classifiers by precisely defining high- and intermediate-risk NDMM subgroups. The identification of an adaptive UPR signature in high-risk patients highlights a biological mechanism linked to treatment resistance. These findings support evaluation of the PR-based model in patients receiving daratumumab-based regimens and suggest that high-risk patients may benefit from more intensive frontline strategies, such as BiTE combinations or CAR-T therapies.
Latest publications
