AFFILIATIONS
1 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
2 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
3 Stanford University, Stanford, California, United States
4 Stanford University, Stanford, California, United States
5 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
6 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
7 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, Stanford, California, United States
8 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
9 Stanford University, Stanford, California, United States
10 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
11 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
12 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States
13 Stanford University, Stanford, California, United States
14 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Weill Cancer Hub West, San Francisco, California, United States
15 Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Stanford University, Stanford, California, United States, Weill Cancer Hub West, San Francisco, California, United States
ABSTRACT
Introduction: Most patients (pts) with relapsed or refractory (r/r) large B-cell lymphoma (LBCL) treated with chimeric antigen receptor (CAR) T-cell therapy do not achieve durable remission, and the myeloid tumor microenvironment (TME) has been repeatedly implicated in resistance. Multiple tumor-associated macrophage (TAM) subsets have been reported as resistance-associated, but in different cohorts and platforms, mostly without spatial context or cross-cohort comparison. Here we systematically evaluated the relative effects of reported myeloid subsets on CAR T outcomes in r/r LBCL and defined their spatial niche localization and neighborhoods.
Methods: We studied 281 patients in a cross-platform meta-analysis across 6 cohorts. We first constructed a tissue microarray (TMA) comprising 24 selected regions from tumor biopsies of 13 LBCL pts treated with axicabtagene ciloleucel (axi-cel; n=11), bispecific CD19/CD22-CAR (NCT03233854; n=1), or CD22-CAR (NCT04088890; n=1) therapy, using biopsies taken before (n=10 pts) or at progression (n=3 pts). We profiled this TMA using 1,000-plex CosMx spatial molecular imaging and developed methods for image registration, segmentation, malignant cell annotation, and TME analysis. CosMx coordinates were registered to post-run histology, and poor-quality cores were excluded on pathology review. Spatial niches were defined by spatial latent Dirichlet allocation (scimap; 80-µm neighborhoods, 10 motifs) followed by k-means clustering (k=5 based on Silhouette scores). Cores were assigned by nearest-centroid matching to LymphoMAP reference archetypes (Li et al. Cancer Cell, 2025): lymph node (LN), T-cell-exhausted (TEX), and fibroblast-macrophage (FMAC).
For meta-analysis, we curated six CAR T-treated LBCL cohorts across spatial (this study, n=10), single-cell (Mo et al. SSRN, 2025, n=17; Li et al. Cancer Cell, 2025, n=22 CAR-T arm), and bulk (Locke et al. Nat Med, 2024, n=134; Stahl et al. Cancer Cell, 2025, n=22; Sworder et al. Cancer Cell, 2023, n=76) transcriptomics platforms. Nine literature-defined myeloid programs were scored per pt: CCL8⁺, CCL13⁺, CCL8⁺CCL13⁺, and CCL18⁺ (Mo et al. SSRN, 2025); PD-L1⁺Ki-67⁺ (Hirayama et al. HemaSphere 2024); CSF1R⁺CD14⁺CD68⁺ LAMM (Stahl et al. Cancer Cell, 2025); C1QB⁺APOE⁺CD163⁺ and IL1B⁺ (Yan et al. Nat Commun, 2024); and CD68⁺CD163⁺ M2 (Marchesi et al. Hematol Oncol, 2015). Each myeloid program was normalized within each cohort, assessed by univariable Cox regression for progression-free or event-free survival (PFS/EFS), and pooled by DerSimonian-Laird random-effects meta-analysis.
Results: Myeloid phenotypes nominally associated with inferior PFS/EFS were CCL8⁺CCL13⁺ (pooled HR 1.31, 95% CI 1.04-1.65, nominal P=0.022, I²=34%) and C1QB⁺ (HR 1.18, 95% CI 1.00-1.38, P=0.045, I²=0%), with directional concordance across spatial, single-cell, and bulk platforms, and HR>1 for the remaining 6 of 7 programs.
Spatial LDA-k-means analysis resolved 5 niches confirmed by pathology review and concordant with LymphoMAP archetypes: TEX cores contained T-infiltrated, myeloid-infiltrated, and myeloid-enriched niches (mean: 29%, 25%, and 31%), whereas LN and FMAC cores were dense-B-rich (70% and 75%), with secondary T-infiltrated (LN, 20%) and stromal (FMAC, 22%) components. Comparison of unpaired pretreatment and progression samples trended towards expansion of myeloid-infiltrated niche at progression (2% to 36%, exact permutation test P=0.052).
All 9 myeloid subsets had more TAM and fewer B-cell and fibroblast neighbors than expected (≤25μm; 300 permutations, FDR≤0.02). CCL8⁺CCL13⁺, CCL13⁺, CCL18⁺, CSF1R⁺ LAMM, C1QB⁺ and CD68⁺CD163⁺ M2 were enriched in the myeloid-infiltrated niche (odds ratio/OR 1.81-4.94, Fisher exact test, FDR<10⁻³), while CCL8⁺, PD-L1⁺Ki-67⁺, and IL1B⁺ were elevated in the myeloid-enriched niche (OR 1.36-2.03, FDR<10⁻³). Notably, CCL8⁺CCL13⁺ TAMs were the most stromal-excluded (log₂ observed/expected -0.27, FDR<0.05), and only C1QB⁺ TAMs showed an excess of T-cell neighbors (log₂ observed/expected = +0.06, FDR<0.05).
Conclusion: Cross-platform meta-analysis and spatial profiling defined CCL8⁺CCL13⁺ and C1QB⁺ TAM cell states as associated with inferior CAR T outcomes in r/r LBCL. Their localization to distinct myeloid-rich niches further nominates these TAM states and their spatial interactions as potential therapeutic targets. Prospective and functional validation to establish clinical utility and causal targetability is ongoing.