Background

Outcomes remain poor in patients with relapsed/refractory B-cell non-Hodgkin lymphoma (R/R B-NHL), including in those with the common subtypes of follicular lymphoma (FL) and diffuse large B-cell lymphoma (DLBCL).1 2 There is a residual unmet need for effective treatment options in these patients. Bispecific antibodies that engage T cells to kill lymphoma cells via a cell surface antigen such as CD20 are currently under investigation in patients with B-NHL, and certain molecules from this class have recently become available for the treatment of R/R FL and DLBCL.3 4

Odronextamab is an Fc-silenced, human, CD20×CD3 bispecific antibody that triggers T-cell-mediated cytotoxicity in CD20-expressing cells independent of T-cell/major histocompatibility complex (MHC) interaction.5 Odronextamab demonstrated deep and durable antitumor responses with a generally manageable safety profile in an interim analysis of ELM-1, a multicenter, Phase I dose-escalation/dose-expansion trial in patients with R/R B-NHL who had previously been treated with an anti-CD20 antibody (NCT02290951).6 At doses ≥5 mg, the objective response rate (ORR) in patients with FL was 91% (N=32). In patients with DLBCL treated with odronextamab ≥80 mg, the ORR was 53% in patients without prior chimeric antigen receptor (CAR) T-cell therapy (N=15) and 33% in patients with prior CAR T-cell therapy (N=30). Interim results from the Phase II ELM-2 study of odronextamab monotherapy confirm its clinical activity, reporting ORRs of 80% in R/R FL and 52% in R/R DLBCL.7 8

Currently, there are limited data on potential baseline biomarkers associated with response to odronextamab or other bispecific antibodies being investigated in patients with R/R B-NHL. Information on mechanisms of intrinsic or acquired resistance, including the role of the target antigen CD20, is also lacking. As part of ELM-1, patient tumor biopsies were obtained at baseline, on-treatment, and at disease progression, with the aim of identifying potential biomarkers that are associated with clinical response and resistance to odronextamab.

MethodsStudy design

Patients enrolled in ELM-1 were aged ≥18 years, diagnosed with B-NHL, had received prior treatment with a CD20-directed antibody, and had ≥1 measurable lesion. Study details have been published previously.6 In brief, patients received intravenous odronextamab, initially following a step-up dosing schedule in Cycle 1, then one time per week at target doses ranging from 0.1 mg to 320 mg during Cycles 2–4 on Days 1, 8, and 15 (each cycle was 21 days). After Cycle 4, maintenance treatment was administered every 2 weeks (Q2W) until disease progression or unacceptable toxicity. If a patient achieved a complete response (CR) that was durable for at least 9 months, then dosing frequency could be decreased from Q2W to every 4 weeks. This reduction in dosing frequency was permitted by a protocol amendment, as odronextamab treatment was initially planned to be discontinued at Week 36. Odronextamab treatment was assessed in a dose-escalation phase (all B-NHL subtypes) and in a dose-expansion phase (selected doses in aggressive B-NHL or FL grade 1–3 a). Antitumor activity was measured by ORR (per Lugano classification9), with responses also reported here by molecular subgroup and DLBCL cell of origin (determined locally according to institutional guidelines).

Patient and public involvement

Patients/the public were not involved in the design of this study or the analyses of the data reported in this manuscript.

Biopsy sample collection

Lymph node/tumor biopsies were taken at the study baseline (≤28 days before the first administration of odronextamab) and at Week 5 or 6 of treatment. An optional biopsy was obtained at disease progression, at the investigator’s discretion. Biopsies were only obtained if clinically feasible, so were not available in all cases. The data cut-off for the inclusion of patients with biopsies was December 20, 2022.

Immunohistochemistry

Fully automated multiplex immunofluorescence (IF) assays were performed on the Ventana Discovery ULTRA platform (Ventana Medical Systems). Sequential primary antibody and secondary horseradish peroxidase-conjugated antibody applications were performed, with heat denaturation between steps to completely remove bound antibodies and eliminate any downstream cross-reactivity. Details on the primary antibodies, and the fluorophores used for detection, are provided in online supplemental table 1.

Following staining, tissue was counterstained and cover-slipped with Invitrogen ProLong Gold Antifade Mountant with NucBlue. Whole-slide imaging was performed on the Zeiss Axioscan (Zeiss). Quantitative image analysis was performed using the HALO Indica Labs Hyperplex module (Indica Labs). Numbers of positive cells for each immune subset were counted and their density was measured.

RNA sequencing

DNA and RNA were prepared from tissues stored in RNAlater Stabilization Solution, using the Qiagen AllPrep DNA/RNA kit (Qiagen). Strand-specific RNA sequencing (RNA-seq) libraries were prepared from 100 ng of RNA using the KAPA mRNA HyperPrep Kit for Illumina Platforms (Roche). Sixteen-cycle PCR was performed to amplify libraries. Amplified libraries were size-selected to 400–600 base pairs (bp) using a 2% Agarose Gel Cassette on PippinHT (Sage Science). Sequencing was performed on the Illumina HiSeq 2500 (Illumina) by multiplexed paired-read run with 2×100 cycles.

To identify cell types within RNA-seq data, previously described gene signatures10 were used to quantify lymphocyte populations within tumor specimens before and after treatment. The scores were then generated by taking the log of the total transcripts per million of all genes comprising the signature.10 Comparisons were made with the Wilcoxon test.

Whole genomic sequencing

DNA sequencing libraries were prepared from 50 ng genomic DNA using the Twist Library Preparation Kit with Enzymatic Fragmentation and the Twist Universal Adapter System (Twist Bioscience). Sequencing was performed on the Illumina NovaSeq (Illumina) by multiplexed paired-read run with 2×100 cycles.

Somatic mutation calling

Somatic mutations were called using the Sentieon Somatic FASTQ to VCF (V.3.2.0) applet with the mark duplicates option, with further details included in the online supplemental material.

Lymph node dissociation

Lymph node tissue was stored in MACS Tissue Storage Solution (Miltenyi 130-100-008) until dissociation using a sterile syringe plunger (BD Biosciences 302995). The cell suspension was filtered (70 mM cell strainer (Biologix 15–1070)) and then spun at 300 g for 5 min at 4°C, with the resulting pellet incubated in ACK lysis buffer (Thermo Fisher A1049201) for 5 min at room temperature. Cells were washed once with phosphate-buffered saline (PBS) and resuspended in PBS for counting or to be used in subsequent assays. The remaining cells were cryopreserved in freezing media containing 90% Human AB Serum (Sigma H3667) and 10% DMSO (Sigma D2650).

Flow cytometry analysis

Peripheral blood mononuclear cells (PBMCs) and dissociated lymph node cells (cryopreserved or freshly dissociated) were resuspended in flow cytometry staining buffer (Stain Buffer BSA, BD Biosciences 554657) containing Human Seroblock Fc Blocking Reagent (Bio-Rad BUF070B), incubated for 10 min at room temperature, then stained for 30 min at 4°C with fluorescently labeled antibodies and reagents for cell surface proteins (see online supplemental material for additional details). Samples were washed three times with staining buffer prior to their acquisition on an A3 Symphony Flow Cytometer (BD Biosciences). Flow cytometry data were analyzed using the FlowJo analysis software (BD Biosciences).

Single-cell processing—single-cell RNA-seq, single-cell V(D)J-seq, and single-cell CITE-seq

Dissociated lymph node cells were incubated with Human Seroblock (Bio-Rad BUF-070B) at 4°C for 10 min prior to the staining of a custom panel TotalSeq-C antibody cocktail (BioLegend; online supplemental table 2) following the manufacturer’s instructions (except that 0.2 mg of the reconstituted cocktail was used per 1×106 cells). Cellular Indexing of Transcriptomes and Epitopes (CITE)-seq staining was performed at 4°C for 30 min. Cells were then washed three times with Stain Buffer containing bovine serum albumin (BSA; BD Biosciences 554657) to remove unbound antibodies and resuspended in PBS with 0.04% BSA.

Washed cells were loaded onto a Chromium Single Cell 5’ Chip (10x Genomics, 1000287) and processed through the Chromium Controller to generate Gel Beads in Emulsion. RNA-seq, Variable, Diversity or Joining (V[D]J) gene segments, and antibody-derived-tag libraries were prepared using the Chromium Single Cell 5’ Library, Gel Beads and Multiplex Kit (10x Genomics, 1000265) following the manufacturer’s instruction. After amplification, complementary DNA (cDNA) was split into small (<300 bp) and large (>300 bp) fragment fractions. RNA-seq and V(D)J libraries were prepared from the >300 bp fraction; cell surface antibody-derived libraries were prepared from the <300 bp fraction.

To enrich the V(D)J library aliquot for T-cell receptor a/b, the cDNA was split into two 20 ng aliquots and amplified in two rounds using primers designed in-house. Details on amplification are provided in the online supplemental material.

Paired-end sequencing was performed using the Illumina NextSeq 500 for RNA-seq and antibody-derived tag libraries (Read 126 bp for unique molecular identifier (UMI) and cell barcode, 8 bp i7 sample index, and Read 255 bp transcript read) and V(D)J libraries (Read 1,150 bp, 8 bp i7 sample index, and Read 2,150 bp read).

For genome mapping, Cell Ranger Single-Cell Software Suite (10x Genomics, V.3.0) was used to perform sample demultiplexing, alignment, filtering, and UMI counting of RNA-seq libraries. The human GRCh38 genome assembly and RefSeq gene model for human were used for the alignment. For V(D)J libraries, Cell Ranger VDJ Software Suite (10x Genomics, V.2.2.0) was used to perform sample demultiplexing, de novo assembly of read pairs into contigs, align and annotate contigs against all the germline segment V(D)J reference sequences from human ImMunoGeneTics (IMGT), label and locate complementary determining region 3 (CDR3) regions, and group clonotypes.

Single-cell data analysis

The initial filtered single-cell (sc) RNA-seq data obtained from Cell Ranger software were used for downstream analysis. Droplets with <500 detected genes, >20% mitochondrial reads, or a total UMI count >50,000 were excluded from downstream analysis.

Following quality control, a library-size correction method was applied to normalize the raw counts using the scanpy “normalize_total”.11 The 2,000 most variable genes were selected for downstream analysis using the “scanpy.pp.highly_variable_genes” function, with the parameter “n_top_genes=2000”. Subsequently, effects of the total count per droplet and percentage of mitochondrial gene count were removed using the “scanpy.pp.regress_out” function. Principal component analysis was then performed with the “scanpy.tl.pca” function with parameters “scd_solver=‘arpack” and “n_comps=15”.

The dimensionality of each data set was further reduced using the “scanpy.tl.umap” function with default parameters. Droplets from assessed patients were clustered based on their expression profiles using the “scanpy.tl.leiden” function with parameter “resolution=0.3”. Cluster-specific marker genes were identified using the “scnapy.tl.rank_gene_groups” function with default parameters. Clusters with high expression of PTPRC, CD3E or PTPRC, CD19, CD20 were characterized as T or B cells, respectively.

scCITE-seq data from Cell Ranger software were further processed using the R package, denoised and scaled by background normalization method,12 and merged with the processed scRNA-seq data object.

ResultsCharacteristics of biopsy patient population

Tumor biopsies from 91 patients (Biomarker cohort) were collected for this analysis, comprising 58 patients with DLBCL, 31 with FL, and 1 each with mantle cell lymphoma and marginal zone lymphoma. A total of 38 patients were included from the ELM-1 dose-escalation phase and 53 from the dose-expansion phase. Overall, baseline characteristics of the Biomarker cohort were similar to those of the total ELM-1 population,6 comprising patients who were mostly men (69%), of White race (86%), and between 60 and <70 years of age (online supplemental table 3). Patients in the Biomarker cohort had received a median (range) of 3 (1–12) prior lines of treatment, and a subgroup of patients who had received previous CAR T-cell therapy were also included. Sixteen biopsies were obtained at Week 5 of treatment (DLBCL, n=8 [five paired with baseline]; FL, n=8 [six paired with baseline]).

The number of biopsies used for each assessment varied, as in some cases results were not evaluable or the tissue sample was insufficient for assessment.

Analysis of baseline biopsiesImmunohistochemistry: B-cell markers at baseline

Multiplex IF showed variable frequencies of B cells that expressed CD19, CD20, CD22, CD79b, or PAX5 (figure 1A; B cells defined by expression of CD19, CD20, CD22, CD79b, or PAX5, alone or in any combination). However, all patients had CD20+B cells within tumor biopsies, and in biopsies of patients with FL nearly all B cells expressed CD20. Among evaluable patients with DLBCL, three had baseline biopsies with <25% of CD20+B cells.

Figure 1

Baseline expression of cell surface markers within B cells using multiplex immunofluorescence (A) and RNA-seq (B) in patients with R/R B-NHL. Relative quantification of baseline CD20 levels by IHC in patients with R/R B-NHL (C; only including patients with at least one response assessment) with a representative staining example (D). B-NHL, B-cell non-Hodgkin lymphoma; CR, complete response; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; IF, immunofluorescence; IHC, immunohistochemistry; PD, progressive disease; PR, partial response; RNA-seq, RNA sequencing; R/R, relapsed/refractory; SD, stable disease; TPM, transcripts per million.

Expression of B-cell markers was assessed according to prior therapy, to determine whether progression on a specific treatment (excluding CD20-directed therapies) was associated with loss of expression of the antigen that was targeted. Among patients with DLBCL who received anti-CD19 CAR T-cell therapy (n=20 [12 prior CAR T-cell therapy, 8 prior CAR T-cell therapy and CD19/CD79b-targeting treatment]), the median percentage of total B cells that were CD19+ was lower than in the overall DLBCL Biomarker cohort, and 10 (50%) patients had negligible CD19 expression (online supplemental figure 1). Among the four patients who had also received anti-CD79b treatment, two had negligible CD79b expression and two retained CD79b expression (online supplemental figure 1).

Messenger RNA expression: B-cell markers at baseline

At the messenger RNA (mRNA) level, high and consistent expression of CD20, CD19, CD22, and CD79b was seen across patients with FL, and expression levels of these markers were generally higher than in patients with DLBCL (figure 1B). While expression levels of all markers varied more within patients with DLBCL, CD20 was highly expressed in most patients with either disease, despite prior treatment with a CD20-directed antibody.

Immunohistochemistry: CD20 expression and response

Baseline CD20 expression was further investigated by analyzing the intensity of chromogenic immunohistochemistry (IHC) staining within biopsies. CD20 staining intensity was categorized as strong, moderate, or weak in patient subsets stratified by response to odronextamab (figure 1C). Representative images of these staining categories are shown in figure 1D. The dose level of odronextamab varied across samples, although all but one patient received odronextamab in the active dose range (≥5 mg for FL and ≥80 mg for DLBCL; details on dosing for patients with chromogenic IHC assessment in online supplemental table 4). The proportions of strong, moderate, and weak CD20 staining varied between patients; however, there was no association of CD20 staining intensity with response to odronextamab in patients with DLBCL. Due to the high odronextamab response rate observed in patients with FL, only one non-responder sample was available for this analysis (sequencing data were not available for this patient). Further analysis of response association was restricted to patients with DLBCL only.

Immunohistochemistry: immune cell/immune cell regulators and response

We next analyzed baseline levels of immune-cell infiltrate within tumor biopsies from patients with DLBCL and studied their association with response to odronextamab. Multiplex IF assessed the number of CD3, CD8, CD4, and regulatory T cells (Tregs), as well as CD68 cells (figure 2A). There was a statistically significant increase (p=0.019) in Tregs in responders versus non-responders, in addition to a numerical trend for increased overall baseline CD3 T cells and CD8 T cells in responders. Otherwise, no significant differences were seen between responders and non-responders.

Figure 2

T-cell infiltration density within tumors (A) and PD-L1 expression within tumors at baseline and its association with clinical responses to odronextamab (B). Only patients with at least one response assessment were included in these analyses. P values were calculated using the unpaired Wilcoxon test and paired Wilcoxon test for Parts A and B, respectively. CR, complete response; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; IF, immunofluorescence; PD, progressive disease; PD-L1, programmed cell death-ligand 1; PR, partial response; SD, stable disease; Treg, regulatory T cell.

The interaction of checkpoint inhibitor receptors with their ligands can lead to downregulation of the T-cell immune response, resulting in an immunosuppressive tumor microenvironment. Programmed cell death protein-1 (PD-1) and lymphocyte activation gene-3 (LAG-3) are two immune inhibitory checkpoint proteins expressed on the surface of antigen-experienced T cells.13 We first investigated the impact of tumor microenvironment in patients treated with odronextamab by studying baseline and on-treatment expression of programmed cell death-ligand 1 (PD-L1) through IF (figure 2B). A trend towards a higher density of PD-L1+ cells within baseline tumor biopsies was noted in patients with DLBCL who went on to respond to odronextamab versus non-responders, and an overall trend towards an increase in PD-L1+ cells from baseline to Week 5 was noted in patients with both DLBCL and FL (online supplemental figure 2). We next investigated the expression of PD-1 and LAG-3 on CD8 T cells within baseline tumor biopsies of patients with DLBCL using IF. The proportion of CD8 T cells that expressed PD-1 and/or LAG-3 varied widely among patients and was not associated with clinical response to odronextamab (online supplemental figure 3).

Gene sequencing analysis: response by MYC, BCL2, and BCL6 rearrangements

We next investigated responses to odronextamab across different subgroups of DLBCL, some of which were determined by local assessment and so included patients outside the Biomarker cohort. Patients with high-grade lymphoma, with dual or triple chromosomal rearrangements in the MYC, BCL2, and/or BCL6 genes (known as double-hit or triple-hit lymphoma), are considered a high-risk subgroup with poor outcomes to standard-of-care therapy.14 In this analysis, responses to odronextamab were observed across patients with BCL2, BCL6, or MYC gene fusions, including double-hit and triple-hit disease (figure 3A). Among the 23 patients with baseline gene rearrangement data (local assessment) who achieved a CR/partial response, 8 (35%) had MYC and BCL2/BCL6 gene fusions.

Figure 3

Odronextamab responses among different molecular subtypes (A). Common oncogenic variants observed in biopsy samples and outcomes in those patients (B). Odronextamab responses among different cells of origin in R/R DLBCL (C). Only patients with at least one response assessment were included in these analyses. CAR, chimeric antigen receptor; CR, complete response; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; GCB, germinal center B-cell-like; N/A, not available; PD, progressive disease; PLD, post-last dose; PR, partial response; R/R, relapsed/refractory; SD, stable disease; W, week; WT, wild-type.

Ten of 23 patients with biopsies available for analysis at screening or Week 5 had common lymphoma oncogenic gene variants, as summarized in figure 3B. Six patients were reported to have TP53 mutations. Two CD58 mutations were also identified: one splice region mutation after the first exon in Patient 2, and one at the transcription start site that is predicted to be deleterious in Patient 9. In addition, a CD80 missense mutation that is predicted to be deleterious was found in Patient 5.

The presence of common lymphoma gene variants did not preclude response to odronextamab. Patient 8 had an ongoing CR at 18 months despite a TP53 gene variant, and Patients 9 and 10 achieved CRs despite harboring two gene variants each. Of note, Patient 7 had a CD20 mutation detected at baseline, but CD20 protein expression was detected and the patient remained in CR after 5 years.

Cell of origin analysis: response by lymphoma subtype

DLBCL can also be classified by the cell of origin, as germinal center B-cell-like (GCB) and non-GCB.15 Similar to the sequencing analysis, responses were observed regardless of the cell-of-origin subtype (figure 3C). From 38 patients with baseline cell-of-origin data (local assessment) who achieved a CR/partial response, 17 (45%) were non-GCB, 16 (42%) were GCB, and 5 (13%) were of unknown subtype.

On-treatment biopsy analysis

To investigate the mode of action of odronextamab as a T-cell-redirecting bispecific antibody, we analyzed the immune infiltrate within paired tumor biopsies at baseline and at Week 5 of treatment. A multiplex IF panel examined the presence of CD20+ cells, T-cell subsets, and CD68+ cells at both time points (figure 4A). A significant decline in CD20+ cells and a significant increase in CD3+ T cells as a proportion of all immune cells was observed, thereby decreasing the target:effector (CD20:CD3) ratio (figure 4B). At Week 5, mRNA expression of the B-cell marker CD20 was decreased within whole tumor biopsies (online supplemental figure 4). Immune cell signature analysis by RNA-seq supported the IF data, including an increase in CD8 signature and cytotoxic signature from baseline to Week 5 in 4/5 evaluable sample pairs for each signature, and a concurrent decrease in naïve T-cell signature and Treg signature (seen in 4/5 and 3/5 evaluable pairs, respectively) (online supplemental figure 5). The pharmacodynamic effect of odronextamab treatment is thus consistent with an increase in intratumoral T-cell activity leading to a depletion of B cells. This pharmacodynamic effect of odronextamab was observed early in treatment regardless of clinical response, indicating the potential mechanisms of resistance were not related to the early on-target drug activity.

Figure 4

Proportion of total immune cells at baseline and Week 5 within tumor biopsies. Percentage of total immune cells from lymph node biopsies (*total [100%] immune cell number comprises CD20+ plus CD3+ plus CD68+ cells) (A). Target:effector (CD20:CD3) ratio (B). P values were calculated using the unpaired Wilcoxon test and paired Wilcoxon test for Parts A and B, respectively. CR, complete response; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; IF, immunofluorescence; PD, progressive disease; PR, partial response; SCR, screening; SD, stable disease; W, week.

Tumor tissue analysis following disease progression

To study potential mechanisms of tumor escape, CD20 protein expression on tumor cells and CD20 gene variants were examined in biopsies at progression (figure 5A). In total, six out of nine progression biopsies with evaluable data were negative for CD20 protein expression (figure 5A). Frameshift CD20 mutations were identified in three patients at progression (Patients 11, 12, and 13; figure 5B). In Patient 11, the truncating mutation Gln204ArgfsTer4 is thought to increase CD20 mRNA decay.16 17 In Patient 12, the truncating mutation Ser43LeufsTer7 removes an extracellular loop, the truncating mutation Met197IlefsTer2 increases CD20 mRNA decay, and Leu198Met is a missense mutation.16 17 In Patient 13, a paired Week 5 biopsy showed that the CD20 gene was wild-type before the appearance of the truncating mutation Met71ArgfsTer12 in the 4-month progression biopsy, which is thought to increase CD20 mRNA decay.16 17 Assessment of mutation status at baseline was not available for Patients 11 and 12. The Gln204ArgfsTer4 mutation was also detected at baseline in Patient 7 (figure 3C); however, CD20 expression was maintained in this patient and a CR was ongoing.

Figure 5

Summary of CD20 expression and presence of gene variants in progression biopsies (A). Summary of inactivating CD20 mutations detected (B). CR, complete response; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; IHC, immunohistochemistry; Mut, mutant; N/A, not available; PD, progressive disease; PLD, post-last dose; PR, partial response; SD, stable disease; W, week; WT, wild-type.

At progression, three patients (Patients 14, 15, and 16) had wild-type CD20 genes yet no CD20 protein expression on biopsy. CD20 protein expression at baseline was not available for these three patients, although documented CD20-positive disease was required for inclusion in the study. Patient 17 had wild-type CD20 at baseline, displayed CD20 protein expression at Week 5, and maintained CD20 protein expression in the 5-month progression biopsy. CD20 protein was also detected in the progression biopsies of Patients 19 and 20, further indicating that in some cases progression could occur without loss of the CD20 protein.

Case studies of patients with disease progression—single-cell analyses

Single-cell analyses were performed on biopsies from two patients on disease progression: one from a patient with FL and one from a patient with DLBCL.

The patient with FL had received six prior therapies but was CAR T-cell therapy-naïve prior to starting this study. This patient achieved a CR with odronextamab, but disease progression was observed at Week 36 and a biopsy was then performed. Flow cytometry analysis of the tumor biopsy at progression indicated that most CD3-negative cells were B cells, based on CD19 and CD22 cell surface expression (figure 6A,B), but no CD20 cell surface expression was detected (online supplemental figure 6A). In accordance with the known upregulation of PD-1 in multiple cancers,18 expression of this molecule on CD3+T cells was substantially greater in this patient with FL compared with PBMCs from a healthy control (online supplemental figure 6B).

Figure 6

Flow cytometry analysis of dissociated immune cells from a fresh lymph node biopsy collected from a patient with FL with disease progression, relative to a healthy donor PBMC control. Flow plot gates indicate CD45+ cells that are CD3− (A). Flow plots demonstrating the MFI of B-cell surface markers CD20, CD19, and CD22 gated on cells that are CD3−CD19+ (B). UMAP of total cells captured with scRNA-seq for the patient with FL dissociated lymph node biopsy sample. Of the six clusters, groups of T cells and B cells are indicated by dashed lines (C). Feature plots showing normalized gene expression of the immune cell markers PTPRC/CD45, CD3E, CD4, CD8A, CD20, CD19, PAX5, CD68, and NCAM1 (D). Feature plots showing CD20 and CD19 normalized cell surface protein expression by scCITE-seq analysis (E). A clonal BCR sequence was found in all B cells profiled by scBCR-seq (F). BCR, B-cell receptor; CDR3, complementary determining region 3; FL, follicular lymphoma; FMO, fluorescence minus one; MFI, mean fluorescence intensity; PBMC, peripheral blood mononuclear cell; scBCR-seq, single-cell B-cell receptor sequencing; scRNA-seq, single-cell RNA sequencing; SSC, side scatter; UMAP, Uniform Manifold Approximation and Projection.

scRNA-seq was also performed on the FL lymph node biopsy. Unsupervised clustering identified six clusters that were broadly categorized as B cells (89%) and T cells (11%) (figure 6C). Most B cells expressed CD20, CD19, and PAX5 by scRNA-seq (figure 6D), but no CD20 surface protein was detected by scCITE-seq (figure 6E). Single-cell B-cell receptor (BCR) sequencing identified 116 unique CDR3 sequences, yet all were classified as similar due to ≥85% nucleotide similarity, and all sequences were derived from the same BCR rearrangement. These data suggest that all B cells in this biopsy originated from a single B-cell clone (figure 6F).

The patient with DLBCL was heavily pretreated, having received rituximab plus CHOP, then rituximab plus polatuzumab (CD79b-directed antibody-drug conjugate), followed by CD19-directed CAR T-cell therapy with axicabtagene ciloleucel, prior to enrollment in this study. This patient did not respond to odronextamab, with disease progression recorded following 6 weeks of treatment. ScRNA-seq performed on the lymph node biopsy indicated high gene expression of CD79b and PAX5 (figure 7A,B). Previous treatment with polatuzumab had only been for 1–2 months, so an effect on the target antigen may not be expected. In contrast, CD19 gene expression was very low (figure 7A,B), indicating a potential mechanism of resistance to CAR T-cell therapy. Indeed, this patient never achieved an objective response to CAR T-cell therapy, although CD19 expression was not available at baseline. At odronextamab progression, the biopsy sample was negative for CD20 expression by IHC, and only weakly positive by flow cytometry, thus representing a similar CD20 protein expression pattern to the patient with FL who progressed (described above). Unlike the patient with FL, no CD20 expression was detected by scRNA-seq (figure 7B). There was also a lack of CD22 expression on immune cells in this patient (figure 7B), although the reason for this is not apparent based on their treatment history.

Figure 7

Analysis of immune cells from the biopsy of a patient with DLBCL with disease progression. Distribution of normalized gene expression by scRNA-seq of B-cell markers CD19, CD20, CD79b, CD22, and PAX5 (A). Feature plots showing normalized gene expression of B-cell markers CD19, CD20, CD79b, CD22, and PAX5 (B). DLBCL, diffuse large B-cell lymphoma; mRNA, messenger RNA; scRNA-seq, single-cell RNA sequencing; scCITE-seq, single-cell Cellular Indexing of Transcriptomes and Epitopes sequencing; UMAP, Uniform Manifold Approximation and Projection.

Discussion

In this biomarker study, we found that CD20 was expressed in almost all tumors from heavily pretreated patients with R/R B-NHL, all of whom had received prior anti-CD20 antibody therapy before enrollment into this trial. Biopsy analyses showed that most B cells within tumors expressed CD20, whereas expression of CD19, CD22, and CD79b was more heterogeneous. Baseline levels of CD20 (IHC quantification) were not associated with odronextamab responses in patients with DLBCL. Furthermore, patients responded to odronextamab regardless of the molecular subtype of lymphoma, cell of origin, or the presence of some common oncogenic gene variants. Overall, these results suggest that testing for CD20 expression at baseline may not be useful for identifying patients who can benefit from odronextamab treatment.

The potential mode of action of odronextamab, as a T-cell-redirecting bispecific antibody that induces cytotoxicity of CD20+ lymphoma cells, has previously been demonstrated in preclinical studies.19 Our clinical biomarker analyses of paired biopsies taken before odronextamab treatment and after 5 weeks of treatment further support this mode of action, through a pharmacodynamic shift of immune infiltrating cells. The dramatic reduction in the proportion of CD20+ cells and the increase in the proportion of T cells within the tumor is in line with the proposed mode of action. This pharmacodynamic effect was even seen in patients who did not achieve clinical response, suggesting that resistance/refractoriness is not due to a lack of initial on-target activity.

Instead, following the initiation of odronextamab treatment, we identified inactivating CD20 gene mutations and loss of CD20 expression as a potential mechanism of resistance in some patients with progressive disease. Indeed, in the biopsy of one patient with disease progression 4 months after completion of treatment, all B cells appeared to be derived from a single CD20- clone. However, alternative mechanisms of resistance may exist, as CD20 protein expression was maintained in a minority of cases of progressive disease.

Loss of CD20 expression and associated CD20 mutations have also been reported with the CD20×CD3 bispecific antibody mosunetuzumab.20 Eight variants of CD20 were identified in six patients with pretreatment and post-treatment paired biopsies.20 Variants were observed in the transmembrane, splice site, and extracellular domains, and in most cases were associated with loss of CD20 expression. Of interest, none of the reported CD20 mutations identified post-mosunetuzumab treatment matched the inactivating mutations identified in this study. Although mosunetuzumab and odronextamab likely target different regions of CD20, the majority of mutations were found outside the extracellular domain which would contain the binding sites, suggesting that variations in drug design do not drive the differences in mutations among progressors. Hence, while in vitro studies have suggested that interference at the binding site may lead to resistance to CD20×CD3 bispecific antibodies,21 these clinical data indicate the primary driver of mutant selection during treatment appears to be the loss of CD20 protein expression. Although there is no evidence to suggest that particular regions of CD20 are more susceptible to mutation, it is interesting to note that most of the inactivating mutations in both studies target exon 7 of the canonical CD20 transcript. This suggests that post-translational processing (splicing) and surveillance (non-sense-mediated decay) of that region might be a common mechanism in the loss of CD20 expression. Taken together, these data indicate that there is no single CD20 variant that drives resistance to CD20×CD3 bispecific antibodies.

This common finding of loss of CD20 expression following treatment with CD20×CD3 bispecific antibodies might be linked to the T-cell-mediated killing of CD20 target cells, leading to strong immune selection. The loss of CD20 following bispecific antibody treatment contrasts with resistance mechanisms to anti-CD20 monoclonal antibodies (eg, rituximab), and might be due to more stringent selection for loss of CD20 with bispecific antibodies. Indeed, complete loss of CD20 expression is relatively uncommon with anti-CD20 monoclonal antibody therapy.22 Data from the current study support this finding; while all patients in this study received prior anti-CD20 antibody therapy, nearly all who were tested had CD20+ B cells within tumor biopsies.

CAR T-cell therapy studies have shown similar relapse and resistance data to bispecific antibodies, with a loss of CD19 expression.23 A single-center study of patients with large B-cell lymphoma (LBCL) treated with the anti-CD19 CAR T-cell therapy axicabtagene ciloleucel reported that 10 of 16 patient biopsies had either lost or had diminished CD19 expression at progression.24 In accordance with these findings, CD19 expression was barely detectable in our single-cell biopsy analysis of a patient with DLBCL who had progressed after anti-CD19 CAR T-cell therapy. However, the variable CD19 expression reported in our study was not specifically related to prior CD19 CAR T-cell therapy, as low or negative CD19 lymphoma was also observed in CAR T-cell-naïve patients in this study (data not shown). In addition, the single-cell biopsy analysis from a patient with relapsed FL found that while CD20 protein expression was lost on progression, CD19 expression was maintained. Further analyses are required to investigate treatment sequencing, particularly with CD19 CAR T-cell therapy, following progression with CD20×CD3 bispecific antibody treatment.

Common lymphoma gene variants were observed in 10 of 23 biopsy samples, including mutations within TP53 and CD79b that are recurrent within DLBCL.25 Other mutations were in genes for CD58, CD80, CD86, and B2M, which have been associated with immune escape.26 CD58 is a ligand for CD2 expressed on natural killer cells and cytolytic T cells, with a role in adhesion and activation of these cells; B2M forms an essential part of MHC class I molecules; and CD80/86 are co-stimulatory ligands with a role in T-cell activation. Crucially, despite the potential for some of these gene variants to suppress antitumor responses, their presence did not necessarily prevent response to odronextamab. The mode of action of CD20×CD3 bispecific antibodies such as odronextamab may allow the immune system to overcome many immune escape mechanisms, including the loss of MHC class I and CD58 expression. Simultaneous engagement of both tumor cells and T cells by bispecific antibodies can crosslink these cell types,27 permitting tumor cell detection by T cells without MHC engagement, and thus T-cell mediated cytotoxicity independent of T-cell receptor recognition. CD58 variants have been associated with poor outcomes in patients with DLBCL,28 29 including in those who received CD19 CAR T-cell therapy consistent with the requirement of CD58 for CAR T-cell activation.30 In contrast, as the immunological synapse formed between the tumor cell and T cell by a CD20×CD3 bispecific antibody delivers a T-cell activation signal,27 reliance on co-stimulation from the CD58−CD2 interaction may be negated. While data in the current study are limited, and further investigation of the CD58−CD2 pathway for CD20×CD3 bispecific antibodies is warranted, the identification of a CD58 deleterious mutation in a patient who achieved CR indicates that CD58 is not required for odronextamab response.

Our biomarker study builds on findings reported with the CD20×CD3 bispecific antibody glofitamab from a smaller cohort of patients.31 For example, the response to glofitamab was also not related to cell of origin or baseline CD20 expression levels in this population of patients with DLBCL. In addition, the glofitamab study reported a trend for a higher percentage of CD8 T cells in baseline tumor biopsies among patients who achieved a CR. Here, we observed a similar trend between response to odronextamab and increased CD8 T cells at baseline. Among the other T-cell types assessed, our data suggest that baseline levels of Tregs may be associated with response to odronextamab. This finding has not been previously reported for other bispecific antibodies in B-NHL, although some studies in patients with non-small-cell lung cancer suggest that a higher frequency of Tregs at baseline is linked to an increased likelihood of clinical response to anti-PD-1 agents.32 33 As expansion of Tregs is dependent on interleukin-2 produced by activated CD4 or CD8+ T cells, higher amounts of Tregs might reflect higher overall immune activity, and thus translate into a greater probability of response.34 35

When considering our results in the context of data presented for other T-cell-engaging therapies, it is noteworthy that the glofitamab study31 reported that overexpression of MYC, downregulation of TP53 targets, and the presence of TP53 mutations within baseline tumor biopsies were all significantly associated with lack of CR to glofitamab. However, in our study, three out of six patients with TP53 mutations at baseline/Week 5 achieved CRs with odronextamab.

Expression of checkpoint inhibitor receptors PD-1 and LAG-3 among CD8 T cells did not differ between responders and non-responders in the current study, although there was a greater number of PD-L1+ cells in responders. This latter point suggests that combining odronextamab with PD-1/PD-L1 inh