Comprehensive Characterization of the Cell States and Ecosystems in Classic Hodgkin Lymphoma Using Single-Cell RNA-Seq, Digital Deconvolution, and Machine Learning

Introduction: Classic Hodgkin lymphoma (cHL) is a unique cancer distinguished by its tumor microenvironment (TME), which is comprised by rare Hodgkin-Reed-Sternberg (HRS) cells (1-5%) surrounded by an abundance of immune cells. Comprehensive characterization of the malignant and immune cell states in cHL is challenging due to the scarcity of HRS cells and the intrinsic heterogeneity of the TME. To address this challenge, we aimed to characterize the cell states and ecosystems of cHL microenvironment using advanced experimental and bioinformatic approaches.

Methods: We identified 111 patients diagnosed with cHL from 2013-19 managed at participating institutions with either formalin-fixed, paraffin-embedded (FFPE, n = 106) or cryopreserved cell suspensions prepared from excisional biopsies (n = 5). FACS purification and single cell RNA-seq (scRNA-seq) were performed on the cryopreserved tissues, resulting in the annotation of 13 distinct cell types within the scRNA-seq data. Next, we employed CIBERSORTx to construct a cHL-specific signature matrix encompassing the features of the 13 annotated cell types. For our FFPE samples, we performed bulk RNA-seq followed by digital deconvolution using the cHL-specific signature matrix, allowing us to estimate the cellular fractions of the 13 cell types within the 106 tumors. Finally, we employed a machine-learning framework (EcoTyper) to discover transcriptionally-defined cell states and ecosystems. Validation of the EcoTyper model was performed in additional scRNA-seq data (n = 3) and Visium Spatial Platform (n = 4).

Results: Patients with FFPE biopsies had a median age of 35, were predominantly male (53%), and the majority had stage IV disease (32%). From the scRNA-seq data (n = 5), we identified 13 distinct cell phenotypes including the minute classical and plasmacytoid dendritic cells, gamma delta T cells, natural killer cells, and HRS cells; then developed a gene expression signature matrix allowing for the deconvolution of bulk transcriptomes from 106 cHL biopsies. Among patients with Variant Allele Frequencies (VAFs) data from tumor and plasma biopsies, we observed a positive correlation between the HRS cell proportion and VAFs from tumor or cell-free DNA sequencing (R = 0.44 and 0.94, respectively). From EcoTyper, we discovered 28 unique cell states and 3 conserved cellular communities or “Hodgkin Lymphoma ecotypes” (HLEs) in the 106 bulk transcriptomes. HLE2 was enriched in elderly patients (Age > 45), H2 phenotype (Alig, Nature, 2024), and EBV infection (P < 0.01). In HLE2, we discovered the enrichment of an immune-suppressing CD8 T cell state expressing LAG3 (P = 1.6 * 10^-11), as well as an HRS cell state enriched in cell cycle pathways (P = 3.41 * 10^-6). Moreover, HLE2 was associated with worse progression-free survival in our cohort (HR = 3.18, P = 0.048) after multivariable adjustment for the disease stages.

To validate our EcoTyper model, we employed additional scRNA-seq on 3 validation samples (n = 31,195 cells) and spatial transcriptomics using Visium 10X (n = 4 tumors). We recovered the established 28 cell states and 3 HLEs in these datasets and found a significant spatial autocorrelation of the HLEs (median Z = 25.4).

Conclusions: Our study represents a significant advancement in unraveling the complexities of the cHL TME. By leveraging scRNA-seq, digital deconvolution, and EcoTyper framework, we have shed light on key cell states and discovered a unique HLE associated with clinical parameters and prognostic significance. Further validation in additional cohorts may allow HLEs to risk stratify patients at diagnosis and aid in identification of poor performing subgroups within current low, intermediate, and high-risk groups.