AUTHORS
Jordan Scott Goldstein, MD, MSc, Steven Wang, MD2, Benoit Tessoulin, MD, PhD3, Cedric Rossi, MD, PhD4, Jan Boegeholz, MD5, Eric Y. Stutheit-Zhao, MD, PhD6, Julia Christina Kuehn, MD7, Chih Long Liu, PhD8, Florian Scherer, MD, PhD9, Scott Bratman10, Steven Le Gouill, MD PhD11, Martine E.D. Chamuleau, MD PhD12, Marcel Nijland, MD, PhD13, David Kurtz, MD, PhD14, Maximilian Diehn, MD, PhD15, Ash A. Alizadeh, MD, PhD16
AFFILIATIONS
1 Stanford University, Palo Alto, California, United States
2 Stanford University, Palo Alto, California, United States
3 University Hospital of Nantes, Nantes, France
4 CHU Dijon, Dijon, France
5 University Hospital Zurich, Zurich, Switzerland
6 University Health Network, Toronto, Ontario, Canada
7 Stanford University, Palo Alto, California, United States
8 Stanford University, Palo Alto, California, United States
9 University Hospital Freiburg, Freiburg im Breisgau, Germany
10 University Health Network, Toronto, Ontario, Canada
11 Institut Curie, Paris, France
12 Vrije Universiteit Amsterdam, Amsterdam, Netherlands
13 Universitair Medisch Centrum Groningen, Groningen, Netherlands
14 Stanford University, Palo Alto, California, United States
15 Stanford University, Palo Alto, California, United States
16 Stanford University, Palo Alto, California, United States
ABSTRACT
Introduction: Large B-cell lymphomas (LBCL) exhibit substantial intra-patient biologic heterogeneity; however, single site biopsies limit assessment of clonal evolution across anatomic sites. Circulating tumor DNA (ctDNA) overcomes this by detecting mutations shed across lymphoma sites. We previously showed that by integrating pre-treatment ctDNA and tissue biopsies, the Metric of Spatial Anatomical Intra-tumoral Genetic Complexity (MOSAIC) quantifies anatomic genetic heterogeneity, and is associated with survival, clonal evolution, and TP53 and CREBBP mutations in LBCL (Goldstein, ASH 2025). Here, we extend MOSAIC temporally to longitudinal molecular data in LBCL to predict molecular response and capture anatomic evolution during treatment.
Methods: We analyzed 278 specimens from 66 patients with baseline plasma, germline and tumor tissue genotyping, plus longitudinal plasma at cycle 2 (C2D1) and/or 3 (C3D1) (Kurtz, JCO 2018). At C2 and C3, 38 and 42 samples were analyzed respectively. Plasma samples were profiled by CAPP-Seq. Early molecular response (EMR) was defined as a 2 log-fold change (LFC) between C1 and C2, and major molecular response (MMR) as a 2.5 LFC between C1 and C3 (Kurtz, JCO 2018). We performed more extensive longitudinal profiling for two LBCL patients with high pre-treatment MOSAIC scores to explore temporal changes in anatomic heterogeneity during treatment. Longitudinal MOSAIC is calculated as previously described (Goldstein, ASH 2025), substituting plasma from a new timepoint anchored to the initial tissue biopsy.
Results: Median pre-treatment MOSAIC was 0.235, stratifying high vs low anatomic heterogeneity. High MOSAIC patients had significantly worse PFS (HR=4.3, p=0.005) and OS (HR=12.7, p=0.002). MOSAIC was associated with EOT PET CR (p=0.03): 31 (94%) low MOSAIC patients reached CR at EOT vs. 23 (70%) high MOSAIC. At C2, 14 (37%) had undetectable MRD and 26 (68%) had an EMR. At C3, 24 (57%) had undetectable MRD and 34 (81%) had an MMR.
Low MOSAIC correlated with greater LFC in ctDNA from C1 to C2 (r=0.41, p=0.01) as well as EMR (p=0.05) and C2 MRD status (p=0.04). Of low MOSAIC patients, 17 (85%) had an EMR and 11 (55%) had undetectable MRD; of high MOSAIC patients, 9 (50%) had an EMR and just 3 (15%) achieved undetectable MRD. Low MOSAIC also correlated with greater LFC in ctDNA from C1 to C3 (r=0.34, p=0.03), but not MMR (p=0.31) or C3 MRD status (p=0.16), likely reflecting compressed dynamic range near the CAPP-Seq detection limit at C3. Of low MOSAIC patients, 22 (88%) had an MMR and 17 (68%) had undetectable MRD; of high MOSAIC patients, 12 (71%) had an MMR and 7 (41%) had undetectable MRD.
Combining MOSAIC with EMR added prognostic value. Patients with high MOSAIC and no EMR at C2 had poor outcomes with 33% 2-year PFS (HR=9.6, p=0.0004). Notably, none of the 3 patients with low MOSAIC but no EMR progressed. Low MOSAIC patients who had an EMR had 2-year PFS of 87%, while high MOSAIC patients who had an EMR had 2-year PFS of 78%.
To understand temporal evolution of MOSAIC, we performed longitudinal plasma profiling of 2 high MOSAIC patients who ultimately relapsed. Patient 1 presented with stage 4 MYC-rearranged transformed LBCL with bulky disease: high MOSAIC score (0.25), high ctDNA VAF (18%), 4 plasma clones (determined by sciClone), and TP53 mutations present below 1%. The patient had a partial response at C3 of DA-EPOCH-R and no MMR (ctDNA LFC of 0.85). A TP53 clone rose dramatically, and MOSAIC increased to 0.44. At C5, she progressed clinically; the rising TP53 clone became dominant at 30% VAF and MOSAIC was 0.36. Patient 2 presented with stage 2 bulky abdominal LBCL: high MOSAIC score (0.35), 2 plasma clones, and TP53, CREBBP, and STAT6 mutations. R-CHOP decreased ctDNA levels to 0.2% (10-2.76) at EOT, though MRD remained detectable and MOSAIC increased to 0.38. She progressed after R-DHAP and ASCT with rapid rise in ctDNA level, and emergence of a STAT6 alteration and new CSMD3 mutation. MOSAIC rose markedly, peaking at 0.51 during subsequent DeVIC & radiation.
Conclusions: Baseline anatomic spatial heterogeneity predicts temporal response dynamics in LBCL, and integrating MOSAIC with early molecular response improves risk stratification. Longitudinal MOSAIC assessment can detect evolving anatomic heterogeneity during treatment. Together, combined monitoring of molecular response and evolution can unlock precise, dynamic risk- and evolution-adapted trial designs in lymphoma.
DISCLOSURES
Jordan Scott Goldstein, MD, MSc: Foundation Medicine - Consultancy (Includes expert testimony); AstraZeneca - Consultancy (Includes expert testimony)