AUTHORS
Steven Wang, MD, Mischa F. B. Steketee, MSc2, Cristina Gómez-Martín, PhD3, Monique A.J. van Eijndhoven4, Esther E. E. Drees, MD, PhD5, Clara Klerk, MD6, Leonie Strobbe, MD7, Yorick Sandberg8, Margriet Oosterveld, MD9, Rozemarijn van Rijn, MD10, Marjolein van der Poel, MD, PhD11, Lara Bohmer, MD12, Nicole de Graauw, MD13, Inge Ludwig, MD14, Pim G.N.J. Mutsaers, MD15, Yvonne Tromp, MD16, Koen Heer, MD17, Ad Koster, MD18, Marie Jose Kersten, MD19, Ilse Houtenbos, MD, PhD20, Gert-Jan Timmers, MD21, Daphne De Jong, MD, PhD22, Josée Zijlstra23, Xiaofei Ye, PhD24, Qiang Pan-Hammarstrom25, Bauke Ylstra, PhD26, Florent Mouliere, PhD27, Martine E.D. Chamuleau, MD PhD28, Yongsoo Kim, PhD29, Dirk Michiel Pegtel30
AFFILIATIONS
1 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
2 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
3 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
4 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
5 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
6 Dijklander Hospital, Hoorn, Netherlands
7 Gelre Hospital, Zutphen, Netherlands
8 Maasstad Hospital, Rotterdam, Netherlands
9 Canisius Wilhelmina Hospital, Nijmegen, Netherlands
10 Medical Center Leeuwarden, Leeuwarden, Netherlands
11 Maastricht UMC, Maastricht, Netherlands
12 Haga Teaching Hospital, The Hague, Netherlands
13 Bravis Hospital, Roosendaal, Netherlands
14 Bernhoven, Uden, Netherlands
15 Erasmus MC, Rotterdam, Netherlands
16 Saxenburgh Medical Center, Hardenberg, Netherlands
17 Flevoziekenhuis, Almere, Netherlands
18 VieCuri Medical Center, Venlo, Netherlands
19 Amsterdam UMC - Location University of Amsterdam, Amsterdam, Netherlands
20 Spaarne Hospital, Haarlem, Netherlands
21 Amstelland Hospital, Amstelveen, Netherlands
22 HOVON Pathology Facility and Biobank, Amsterdam, Netherlands, Netherlands Cancer Institute, Amsterdam, Netherlands
23 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
24 Karolinska Institutet, Stockholm, Sweden
25 Karolinska Institutet, Stockholm, Sweden
26 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
27 Cancer Research UK National Biomarker Centre, University of Manchester, Manchester, United Kingdom
28 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
29 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
30 Amsterdam UMC - Location Vrije Universiteit, Amsterdam, Netherlands
ABSTRACT
Introduction
The tumor microenvironment (TME) evolves under therapeutic pressure, yet capturing this evolution requires repeat biopsies that are rarely obtained in routine practice. Circulating extracellular vesicles (EVs) are actively secreted by living cells and carry RNA from malignant cells, the surrounding TME, and systemic organs, offering a minimally invasive window into host-tumor biology. We hypothesized that serial plasma EV-RNA can resolve TME and systemic macroenvironment (SME) cell states and their higher-order ecotypes, and that their on-treatment dynamics may inform prognosis in large B-cell lymphoma (LBCL).
Methods
We generated serial plasma EV-RNA data from 162 LBCL patients receiving first-line immunochemotherapy in two HOVON studies, including 58 patients with paired baseline and post-cycle-one samples and 23 healthy donors. We developed TMEV-Statescope, a Bayesian deconvolution framework that integrates ichorCNA cell-free DNA tumor fractions from shallow whole-genome sequencing as a per-sample to inform deconvolution. A single-cell RNA-seq atlas of 81,353 cells across 15 cell types from 17 DLBCL tumors served as another input for the deconvolution (Ye et al., Cell Reports 2022). Convex non-negative matrix factorization identified recurrent EV-inferred cell state combinations as ecotypes. EV-inferred cell states and ecotypes are compared against DLBCL tissue-derived cell states and ecotypes by EcoTyper (Steen et al., Cancer Cell 2021). Finally, Elastic Net classifier benchmarked prognostic value against Statescope without tumor priors, EV-mRNA alone, and the International Prognostic Index (IPI).
Results
Baseline LBCL EVs were enriched for 597 transcripts versus healthy donors, including B-cell programs (PAX5) and MYC, E2F, and PI3K/AKT/mTOR pathways. Tumor-burden integration in TMEV-Statescope markedly improved concordance between inferred malignant fractions and measured tumor burden. TMEV-Statescope resolved 56 EV-inferred cell states across 15 cell types and 7 recurrent ecotypes spanning malignant, TME, and systemic origins. A ribosomal, MYC-driven malignant state characterized 82 percent of high-grade B-cell lymphoma (HGBL) with MYC and BCL2 rearrangements and matched the poor prognostic EcoTyper tissue state S05. After one cycle, EV fractions from B, T, and NK cells rose in non-progressors, while a systemic baseline ecotype (Ecotype 5) was cleared by therapy and TME-associated ecotypes became abundant. Baseline ecotypes were not prognostic, but on-treatment ecotypes were. Ecotype 3 marked progressors (P<0.01) and Ecotype 7 marked sustained remission (P<0.05). The Elastic Net model reached a cross-validated AUC of 0.75, exceeding Statescope without priors (0.62), EV-mRNA alone (0.64), and IPI (0.60). TMEV-Statescope risk added independent prognostic value on top of ctDNA detection for time-to-progression and overall survival by multivariate Cox regression (P<0.05 and P<0.01). Patients with detectable ctDNA and high-risk EV features after one cycle had the poorest survival, with only 25% event-free beyond 6 months versus above 50% for other ctDNA-positive groups.
Conclusions
TMEV-Statescope decodes TME and systemic evolution from plasma EV-RNA and identifies early on-treatment ecotype shifts that predict outcome beyond ctDNA and IPI in LBCL. This novel minimally invasive framework enables serial, individualized monitoring of tumor-immune dynamics and supports early treatment-response prognostication.
DISCLOSURES
Steven Wang, MD: Amsterdam UMC - Patents & Royalties