LGsig broadens diffuse large B-cell lymphoma classification beyond genetics alone
Diffuse large B-cell lymphoma has never fit neatly inside a single molecular box. Genetic classifiers have sharpened the view, but they still leave a substantial fraction of cases unassigned, and they miss biology that lives outside the DNA sequence itself. The new LGsig platform was built to close that gap by combining genetic subtype logic with transcriptomic state, creating a classifier that can assign all DLBCLs while preserving the biologic identity of the disease subsets most likely to matter at the bedside.
The premise is straightforward: gene expression captures what a tumor is actively doing, not just what mutations it carries. In a lymphoma such as DLBCL, where signaling, lineage programs, and microenvironmental effects can shape treatment response, transcriptional state may be the piece that explains why two tumors with similar genomics behave differently. LGsig was designed from that assumption, and its performance across large training and validation cohorts suggests that the assumption was biologically sound.
Transcriptomic signatures resolved previously unclassified cases
Model development used paired genomic and transcriptomic data from the National Cancer Institute/British Columbia Cancer Agency cohort, which included 764 cases. The classifier was restricted to tumors already defined by canonical molecular classes: MCD with MYD88L265P and CD79B mutations, BN2 with BCL6 translocation and NOTCH2 mutations, EZB with EZH2 mutations and BCL2 translocation, and ST2 with SGK1 and TET2 mutations. From those subtype-specific expression patterns, investigators selected differential gene features and arrived at an optimal 294-gene panel for a nearest shrunken centroid classifier.
The resulting labels — MCDsig, BN2sig, ST2sig, and EZBsig — were not just technical conveniences. They provided a transcriptome-based map that extended classification to cases the prior system could not place. LGsig correctly identified DLBCL subtypes from transcriptional data alone and assigned previously unclassified tumors into coherent biologic groups. In those reassigned cases, the tumors resembled their classified counterparts in both transcriptional profile and clinical behavior, supporting the idea that the new assignments reflected disease biology rather than statistical noise.
One sentence captures the practical consequence: previously unclassified lymphoma no longer stayed invisible.
A53 disease moved into more biologically coherent clusters
The model also re-sorted the LG A53 subgroup, a category marked by aneuploidy and TP53 alterations. Those tumors often occupy a difficult middle ground in molecular taxonomy, because their genomic instability can obscure more informative subtype relationships. LGsig shifted A53 cases into clusters that were more biologically and therapeutically meaningful, an adjustment that matters because TP53-altered disease often tracks with aggressive behavior and treatment resistance.
This is where the added value of a transcriptional classifier becomes more than academic. Aneuploidy and TP53 disruption can describe a tumor’s damage, but not always its lineage state or signaling dependence. Expression profiling can capture those active programs, helping distinguish whether a tumor behaves more like one subtype than another despite shared genomic instability. In practice, that kind of refinement can influence how clinicians think about expected chemosensitivity and the likelihood that a novel regimen will outperform standard therapy.
POLARIX data linked subtype assignment to pola-R-CHP benefit
The most clinically provocative test for LGsig came from the POLARIX trial, where archival RNA sequencing from 678 samples allowed the classifier to be applied to a randomized treatment dataset. The analysis assessed outcomes after polatuzumab vedotin-R-CHP versus R-CHOP within each LGsig subtype. LGsig strengthened the biomarker performance of the original LG framework by identifying distinct DLBCL groups that showed a survival advantage with pola-R-CHP over R-CHOP, including both LG-classified and LG-unclassified cases.
That result matters because it ties classification to treatment effect, not just taxonomy. A biomarker becomes clinically useful when it can move beyond prognosis and start identifying who gains from a particular regimen. By extending subtype assignment to the previously unclassified fraction, LGsig widened the pool of patients for whom a treatment-associated signal could be detected. In a disease as heterogeneous as DLBCL, that can change how trial results are interpreted and how future studies are designed.
The biology behind the approach is plausible. Polatuzumab vedotin adds targeted cytotoxic pressure through CD79b-directed delivery, while the chemotherapy backbone still depends on tumor sensitivity to immunochemotherapy. If a subtype’s transcriptional program reflects B-cell receptor dependence, survival pathways, or differentiation state, then it may be more or less vulnerable to a regimen that alters those pressures. LGsig does not prove mechanism, but it aligns subtype identity with a drug response pattern that conventional genetic classification alone did not fully resolve.
Genetic and transcriptional states now read as one disease map
LGsig’s central contribution is conceptual as much as technical. DLBCL has long been divided by mutations and rearrangements, yet tumor expression states can reveal whether those lesions are functionally dominant, partially silent, or accompanied by broader biologic programs that shape response. By integrating genetic and nongenetic dimensions, the classifier converts an incomplete genetic taxonomy into a more complete disease map.
This approach also exposes the limits of mutation-only classification. Some tumors are genetically legible but clinically ambiguous. Others are biologically active in ways that DNA sequencing alone cannot capture. The 294-gene classifier turned those gray zones into assignable subtypes and, in the POLARIX setting, linked at least some of those assignments to differential benefit from pola-R-CHP. For lymphoma clinicians and trialists, that creates a more precise framework for thinking about who should be studied, how response should be stratified, and where molecular heterogeneity may be masking a true treatment effect.
The remaining challenge is not whether molecular refinement is useful; it is how to validate LGsig prospectively, in real-world specimens and in studies specifically built around subtype-linked treatment selection. The current work shows that expression-based classification can reclassify unassigned DLBCL and improve biomarker performance, but clinical adoption will depend on reproducibility across platforms, assay practicality, and confirmation that subtype assignment consistently changes management in ways that improve patient outcomes.