CD20×CD3 Bispecific Antibody Biomarker Discovery

What this research found

Which molecular markers should decide who receives a CD20×CD3 bispecific antibody? Six candidate genes were scored against pooled diffuse large B-cell lymphoma (DLBCL) genomics, CRISPR dependency screens, target tractability, published trial efficacy, and post-marketing safety reports, then combined into a single weighted evidence ranking. CREBBP (0.682) and EZH2 (0.675) finished on top, and because their mutations co-occur more often than chance, the recommended Phase III design uses them jointly to define an enriched population.

• CREBBP and EZH2 led the composite ranking at 0.682 and 0.675, ahead of TP53 (0.545), MS4A1 (0.530), B2M (0.397), and CD58 (0.206). CREBBP led on prevalence — 12.4%, or 147 of 1,184 pooled DLBCL samples — while EZH2 led on functional essentiality with a mean Chronos score of -0.387.
• CREBBP and EZH2 mutations co-occur: odds ratio 3.04 at FDR q = 0.0036, the only one of 15 gene pairs to survive Benjamini-Hochberg correction. That dependence is what justifies treating the two as one enrichment criterion rather than two independent strata.
• The drug target itself is invisible to mutation calling. MS4A1 showed no somatic point mutations across the DLBCL cohorts, and splitting 113 B-cell lymphoma lines at the median MS4A1 expression (log2(TPM+1) = 7.37) produced no CRISPR dependency difference that survived correction — consistent with CD20 loss occurring through epigenetic and post-translational routes.
• Response and toxicity both differ by agent and setting. Glofitamab reached 52.0% overall response and 39.4% complete response in relapsed/refractory DLBCL, mosunetuzumab 80.0% and 60.0% in relapsed/refractory follicular lymphoma. Cytokine release syndrome appears in 31.4% of glofitamab adverse event reports (578 of 1,839) against 21.1% for mosunetuzumab (165 of 783).
• The resulting trial design defines an enriched subgroup by CREBBP or EZH2 mutation covering roughly 17.7% of relapsed/refractory DLBCL, about 2,207 eligible U.S. patients a year, with roughly 300 intention-to-treat patients giving about 80% power at a hazard ratio of 0.70 and alpha of 0.025 for each co-primary endpoint.

How it was done

Six candidate genes — MS4A1, TP53, CREBBP, EZH2, B2M, and CD58 — were profiled across six independent sources: mutation and copy-number calls for 1,184 DLBCL samples from three cBioPortal studies, Chronos CRISPR gene effects from DepMap 26Q1, Open Targets tractability and clinical-phase data, three registered bispecific antibody trials, 130 PubMed articles, and 2,622 adverse event reports. Pairwise co-occurrence was tested with Fisher's exact test under Benjamini-Hochberg correction, and cell lines were split at median CD20 expression and compared with Mann-Whitney tests. The six genes were then ranked by a transparent weighted score — 30% genomic prevalence, 25% functional dependency, 30% tractability and disease association, 15% literature volume, each min-max normalised — and the ranking was translated into a Phase III statistical analysis plan covering stratification roles, companion diagnostic tiers, powering, and FDA and EMA alignment. The output is a 34-page white paper with 48 citations, five figures, and four tables.

Data sources

• cBioPortal — 1,184 DLBCL samples across three studies (dlbcl_duke_2017, dlbclnos_tcga_gdc, dlbcl_dfci_2018), accessed April 2026
• DepMap Public 26Q1 — Chronos CRISPR gene effects for 56 B-cell lymphoma lines and expression data for 113
• Open Targets Platform v4 — association, tractability, and clinical-phase data for six genes
• ClinicalTrials.gov — NCT04408638, NCT04676360, NCT03677141
• openFDA FAERS Q4 2025 release — 2,622 reports, comprising 1,839 for glofitamab and 783 for mosunetuzumab
• PubMed via NCBI Entrez — 130 unique publications, 2020–2026
• Dickinson et al., New England Journal of Medicine 2022 (glofitamab); Budde et al., Lancet Oncology 2022 (mosunetuzumab)

Figures from this analysis

Outputs produced

• cbioportal_mutations_cnas.csv

  Mutation and CNA data for MS4A1, TP53, CREBBP, EZH2, B2M, CD58 from 3 DLBCL cBioPortal studies (n=1184 total samples, 496 alteration records)

• cbioportal_co_occurrence.csv

  Pairwise co-alteration matrix for 6 target genes with Fisher exact test + BH FDR correction (15 gene pairs). Columns: pval_fisher, pval_adj_fdr, significant_fdr05, significant_raw_p05

• depmap_lymphoma_dependencies.csv

  DepMap 26Q1 CRISPR Chronos gene effect scores for 56 B-cell lymphoma cell lines, annotated with disease metadata and CD20 expression cohort

• depmap_cd20_stratified_lines.csv

  113 B-cell lymphoma cell lines stratified into CD20-high (n=57) and CD20-low (n=56) cohorts by MS4A1 log2(TPM+1) expression (median=7.37), with per-gene CRISPR effect scores

• depmap_mw_test_summary.csv

  Mann-Whitney U test summary for CD20-high vs CD20-low CRISPR dependency comparison (6 genes), with raw and BH FDR-adjusted p-values

• opentargets_biomarkers.csv

  Open Targets Platform v4 data for 6 target genes: disease association scores (top 100 per gene), tractability (SM/AB modalities), drug/clinical candidates, priority score. All 6 genes have DLBCL association scores.

• clinicaltrials_bispecifics.csv

  Clinical trial data for NCT04408638 (glofitamab, R/R DLBCL), NCT04676360 (mosunetuzumab, R/R FL), NCT03677141 (mosunetuzumab, R/R NHL). Contains trial metadata, status, enrollment, and published efficacy metrics (ORR, CR).

• pubmed_biomarker_literature.csv

  130 unique PubMed articles (2020-2026) on glofitamab/mosunetuzumab biomarker correlations. Fields: PMID, title, authors, journal, pub_date, abstract (up to 2000 chars), MeSH terms, relevance tags.