AUTHORS
Nick Phillips, Dasom (Caroline) Lee, MD2, Mark Hamilton, MD, PhD3, Brian Sworder, MD, PhD4, Troy Noordenbos, MD5, Xiaoman Kang6, Julia Kuehn7, Lena Osswald, MD8, Shuyu Shi, PhD9, Jordan Scott Goldstein, MD, MSc10, Takeshi Sugio, MD, PhD11, Feng Tian, PhD12, Chih Long Liu, PhD13, Mari Olsen14, Arvin Eslami, MD15, Masooma Shifa Rana, MBBS16, Yasodha Natkunam, MD, PhD17, Emma Sampietro18, Amirsaman Ashtari, PhD19, Mohammad Shahrokh Esfahani, PhD20, Stefan Alig, MD21, Crystal Mackall, MD22, Zinaida Good, PhD23, Steven M. Corsello, MD24, David Kurtz, MD, PhD25, David Miklos, MD, PhD26, Maximilian Diehn, MD, PhD27, Saurabh Dahiya, MD, FACP28, Ash A. Alizadeh, MD, PhD29
AFFILIATIONS
1 Stanford University, Stanford, California, United States
2 Stanford University, Stanford, California, United States
3 MD Anderson Cancer Center, Houston, Texas, United States
4 City of Hope, Duarte, California, United States
5 Stanford University, Stanford, California, United States
6 Stanford University, Stanford, California, United States
7 Stanford University, Stanford, California, United States
8 Stanford University, Stanford, California, United States
9 Stanford University, Stanford, California, United States
10 Stanford University, Stanford, California, United States
11 Stanford University, Stanford, California, United States
12 Stanford University, Stanford, California, United States
13 Stanford University, Stanford, California, United States
14 Stanford University, Stanford, California, United States
15 Stanford University, Stanford, California, United States
16 University of Nebraska Medical Center, Omaha, Nebraska, United States
17 Stanford University, Stanford, California, United States
18 Stanford University, Stanford, California, United States
19 Stanford University, Stanford, California, United States
20 Stanford University, Stanford, California, United States
21 Essen University Hospital, Essen, Germany
22 Stanford University, Stanford, California, United States
23 Stanford University, Stanford, California, United States
24 Stanford University, Stanford, California, United States
25 Stanford University, Stanford, California, United States
26 Stanford University, Stanford, California, United States
27 Stanford University, Stanford, California, United States
28 Stanford University, Stanford, California, United States
29 Stanford University, Stanford, California, United States
ABSTRACT
Background: Anti-CD19 chimeric antigen receptor T-cell therapy (CAR19) has transformed management of relapsed/refractory large B-cell lymphomas (LBCLs), yet 40-50% of patients ultimately relapse (Locke, NEJM 2022). CD19 loss is a central mechanism of resistance seen in ≥30% of post-CAR19 relapses, yet the associated mechanism remains unclear (Spiegel, Nat Med 2021; Sworder, Cancer Cell 2023). Because CD19 is also the target of a growing arsenal of T-cell engagers, antibody-drug conjugates, and monoclonal antibodies, the determinants of CD19 expression carry implications well beyond CAR19. We hypothesized that serial comprehensive genomic profiling of relapsed LBCL tumors before and after CAR therapy could define key determinants of CD19 antigen loss.
Methods: We assembled a cohort of 97 patients with LBCLs treated with CAR19 therapies, comprising 136 tumor biopsies taken pre- (65%) and/or post-CAR19 (35%). CD19 antigen status was quantified at the protein level by immunohistochemistry (H-score) and by surface antigen density using calibrated flow cytometry (QuantiBRITE, BD). Tumor samples were classified as CD19-high (58.8%), -medium (24.3%), or -low (16.9%) using a consensus protein level gating scheme. FFPE specimens were profiled by deep whole-exome sequencing (n=88), whole-transcriptome RNA-Seq (n=131), and deep targeted DNA genotyping (n=107). Additionally, CD19 status was non-invasively assessed from blood plasma by EPIC-seq (n=42) (Esfahani, Nat Biotech 2022), using a novel genome position-aware cfDNA fragmentomic strategy.
Results: CD19 protein measurements by H-score and flow cytometry were significantly correlated (r=0.74, p=1.1x10-5), and both protein measurements were concordant with CD19 RNA expression (H-score: r=0.5, p=2.5x10-7; flow: r=0.68, p=4.8x10-6). CD19 RNA expression declined monotonically across CD19-high, -medium, and -low tumors (Jonckheere-Terpstra p=4.7x10-10), confirming that antigen loss reflects transcriptional modulation without significant evidence for isolated cytoplasmic sequestration. Differential expression analysis of CD19-high vs -low relapse cases identified CD19 itself as the most significant differentially expressed RNA (log2FC=2.75, padj=9.8x10-7). Co-expression analysis (WGCNA) identified modules positively correlated with CD19 (B-cell signaling r=0.63; lymphocyte activation r=0.55) and an anticorrelated plasma cell module containing PRDM1 and XBP1 (r=-0.58). Univariate analysis revealed CD19 protein loss was significantly associated with loss of canonical B-cell marker transcripts such as MS4A1 (p=0.034), CD22 (p=0.034), CD79B (p=0.0085), and PAX5 (p=0.018) (CD19-high vs -low, Wilcoxon rank-sum), even after adjustment for tumor content. In contrast, genetic determinants of CD19 loss were weak predictors; no gene was significantly differentially mutated between CD19-high and CD19-low relapse after FDR correction. The strongest enrichments in CD19-low relapse cases were observed for mutant epigenetic regulators (KMT2D 46% vs 15%; CREBBP/EP300 31% vs 15%). Across 335 recurrently mutated genes, XBP1 mutant AF was most strongly anti-correlated with CD19 expression (Spearman's rho=-0.32, p=0.0022). Notably, neither rare exonic nor intronic CD19 mutations were associated with CD19 RNA or protein levels (p=1.0). Finally, we developed a machine-learning EPIC-seq model, enabling non-invasive inference of tumor CD19 gene expression (r=0.61, p=5.7x10-4) and CD19-high and -low antigen abundance classification (AUC=0.81).
Conclusions: In LBCL tumors relapsing following CAR19 therapy, antigenic loss of CD19 protein is strongly transcriptionally mediated and associated with coordinated loss of mature B-cell markers (MS4A1, CD22, CD79B) and key transcription factor perturbations (PAX5, PRDM1/BLIMP-1, XBP1). In contrast, antigen loss is not significantly associated with mutations in CD19 itself or any single gene. These findings implicate B-cell lineage plasticity and plasmacytic differentiation as mechanisms of antigen loss. Finally, we demonstrate the feasibility of non-invasive inference of tumor CD19 status from routinely collected blood draws, enabling analysis of antigen-loss burden in larger cohorts at scale. If such plasticity is validated to be phenotypically stable, these results have therapeutic implications for secondary rescue of dynamic CD19 loss using other lineage markers.
DISCLOSURES
Nick Phillips: Personalis - Current equity holder in publicly-traded company; Personalis - Stock (publicly traded)