AUTHORS
Paul Buckley, PhD, Roberta Dunn2, Julie Chan3, Cynthia Bishop4, Timothy Woo5, Katrina Todd6, Elizabeth Phillips7, Alessia Bottos, PhD8, Christopher Bolen, PhD9, Verena Passerini, PhD10, Oliver Weigert, MD11, Piers E M Patten, MB ChB PhD12
AFFILIATIONS
1 King's College London, London, United Kingdom
2 King's College London, London, United Kingdom
3 King's College London, London, United Kingdom
4 Guys & Saint Thomas Nhs Foundation Trust, London, United Kingdom
5 King's College London, London, United Kingdom
6 King's College London, London, United Kingdom
7 University of Manchester, Manchester, United Kingdom
8 F. Hoffmann-La Roche Ltd, Basel, Switzerland
9 Genetech Inc, South Sanfrancisco, United States
10 LMU Hospital, Munich, Germany
11 LMU Hospital, Munich, Germany
12 King's College London, London, United Kingdom
ABSTRACT
The variable clinical course and response to treatment of Follicular Lymphoma (FL) is strongly influenced by its tumor microenvironment (TME). Indeed, cellular interactions dictate pro- and anti-tumor processes within the TME, although these mechanisms remain incompletely understood. Emerging spatial techniques have shown features such as stromal desmoplasia and alterations to follicular growth patterns precede relapse. Nonetheless, studies characterising the FL TME in spatial detail remain limited, and how such TME features are related to molecular subgroups, treatment response and POD24 remain incompletely resolved.
Therefore, we performed a detailed characterisation of the FL TME, utilising fixed formalin paraffin embedded (FFPE) biopsies from 220 untreated FL patients enrolled on the GALLIUM trial (), assessing frontline chemo-immunotherapy regimens. TMAs were constructed in Leeds, UK (n=139) and Kiel, Germany (n=81). We optimized a 31-plex antibody panel to identify key immune lineages, functional sub-states (e.g. memory/effector phenotype) and stromal architecture. Samples were stained and imaged by Hyperion mass imaging cytometry enabling detailed single cell spatial profiling. A EZH2 gene expression signature status was available for n=134 patients.
Deep-learning cell segmentation resolved >3 million cells across 282 regions-of-interest (ROI). To inform cell clustering, batch correction using fast-mutual-nearest-neighbors was performed between blocks to mitigate site-to-site variability. We identified distinct B cell populations, CD4+ and CD8+ T cells, as well as macrophages, stromal and vascular regions. Deep phenotyping revealed T cell subsets among others, including CD4+ T follicular helper cells (Tfh), and memory, naïve and checkpoint-expressing CD4 and CD8+ T cell populations.
We next defined ten distinct cellular neighborhoods (CN) in FL LNs. The cellular composition of most CNs were dominated by B cells while cases with an EZH2-mutant gene signature showed elevated CD45RO+CD27+CD45RA- CD8+ T cells enriched within B-cell and T-cell dominated CNs. CN membership within follicular areas indicated striking heterogeneity, with some follicles composed exclusively of follicular dendritic cells and B cells, while others were diversely populated, sometimes within the same patient. To ascertain whether reproducible TME groups were observed, we performed unsupervised clustering of patient-level CN composition. Three TME architectural groups describing patient-CN makeup emerged: a CD10+CD27+ follicular architecture with minimal immune infiltration (TME1), a CD10+CD21+ architecture with immune content concentrated in discrete interfollicular neighbourhoods (TME2), and a CD45RA+ B cell architecture with dispersed immune infiltration (TME3).
Finally, we used kernel density estimation to characterise B cell structures. CD21 expression was used to define follicles, while non-follicular B cell aggregates (defined as B cell dense with absent CD21), were identified utilising B cell positions and CD21 expression. Small CD21+ regions below 8,000um2 or 100 cells were categorised separately. We also identified B cell+ CD21+ regions which appeared diffuse or contiguous, with attenuated localised interfollicular region, and these were categorised separately.
564 follicles were detected in 243 ROIs from 205 patients, ~6.3% of ROI yielded no follicles. Follicles had a median area of 37,100 um2, an equivalent diameter of 217um, though follicles generally touched ROI borders therefore these estimates are lower bounds. Inter-follicle distances were modelled (edge-to-edge nearest-neighbor distances) and found to differ between TME architectures: with TME3 carrying the widest gaps in both Leeds and Kiel cohorts Leeds follicles: n = 310, LR = 7.49, p=0.024, q=0.090; Kiel follicles n=168, LR=15.7, p=3.9x10-4,q=0.0031).
Our data indicate highly variable phenotypic and morphological tumour microenvironment arrangements among untreated follicular lymphoma tumours. Correlation of these architectural features with clinical and molecular features is ongoing and data will be presented at the ASH 2026 meeting. We envision that this characterization of spatial tumour-immune crosstalk within neighbourhoods will indicate how the TME shapes disease behaviour and importantly inform appropriate therapeutic strategies.
DISCLOSURES
Piers E M Patten, MB ChB PhD: AstraZeneca - Honoraria; BeOne - Honoraria; BeOne - Honoraria; Astra Zeneca - Honoraria; Adaptive Biotechnologies - Honoraria; Eli Lilly - Honoraria; Lilly - Honoraria; Abbvie - Honoraria; AbbVie - Honoraria; Janssen - Honoraria; BeOne - Research Funding; AstraZeneca - Research Funding; AstraZeneca - Consultancy (Includes expert testimony); AbbVie - Consultancy (Includes expert testimony); BeOne - Consultancy (Includes expert testimony); Eli Lilly - Consultancy (Includes expert testimony)