AUTHORS
Shady Adnan-Awad, PhD, Avery Buchanan, MRes2, Kristian Gurashi, PhD3, Yuqi Shen, PhD4, Daniel North, Md PhD5, Sophie Kellaway, PhD6, Adam Mead, MBBChir7, Sarah Gooding, MD PhD8, Abdullah Obaid Khan, PhD9
AFFILIATIONS
1 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
2 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
3 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
4 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
5 ACRF Translational Research Laboratory, Melbourne, Victoria, Australia, Peter MacCallum Cancer Centre, Melbourne, Victoria, Australia
6 School of Medicine, Nottingham, Nottinghamshire, United Kingdom
7 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
8 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
9 MRC Molecular Haematology Unit, Weatherall Institute of Molecular Medicine, Oxford, United Kingdom
ABSTRACT
Introduction. Remodelling of the tumour microenvironment (TME) drives progression and treatment resistance in blood cancer, yet resolving how a given marrow niche is altered in disease remains a major challenge. Human marrow sampling is destructive to niche cells and species-specific biology remains a significant translational barrier. As such, understanding disease-specific remodelling of the TME, and how those drivers can be targeted, remains an unmet clinical need. Here we present a novel paradigm wherein a bioengineered human platform is used to generate a computational framework through which niche states can be imputed from tumour-only patient data, nominating TME targeting combination approaches.
Methods. We engrafted primary cells from patients with blood cancers (JAK2 myelofibrosis (MF) n=4, post-MPN sAML n=4, multiple myeloma n=3, de novo AML (TP53 mutant) n=2) and healthy donors (n=5)) into combined bone and bone marrow organoids for 14 days under identical experimental conditions. The resulting single-cell RNA sequencing (scRNAseq) data was used to build a virtual tumour microenvironment (vTME) model of innate immune and stromal remodelling and develop a latent-variable framework to impute niche states from tumour-only patient data.
Results. Integration of 24 chimeroid samples (18 engrafted, 6 unengrafted comBO controls; 327,634 cells) resolved a 16-dimensional vTME (eight curated, eight data-driven cNMF modules). We then trained OLiMPUS (Organoid Learned imputation of niche states, partial least squares with k-nearest-neighbour transfer) on covariance between the engrafted patient cells and the organoid niche (iPSC-derived stroma and immune cells). In effect, OLiMPUS learns which tumour programmes travel with which stromal states in the organoid, then reads that relationship backwards in patients where only the tumour was sequenced.
We then used this framework on publicly available, haematopoietic/tumour only cells to establish whether niche states could be reliably inferred. Applied to scRNAseq from 241 patients across three independent cohorts (MF n=21; AML n=188; myeloma n=32), inferred niche states recovered their chimeroid-derived disease direction in 6 of 10 testable modules at the patient level: in myeloma, inflammatory (Cohen’s d = 0.61), endothelial (0.66), acute activation (0.54) and mast (0.49); in subtype-matched TP53 de novo AML, macrophage (0.24) and metabolic (0.43). Leave-one-sample-out testing over 18 held-out samples recovered niche states at Spearman ρ = 0.47 to 0.68 across modules (permuted null ρ ≈ 0). The four modules that failed to transfer were stroma-autonomous (osteogenic and HSC niche), marking the limit of haematopoiesis-based inference. OLiMPUS imputed clinical axes absent from the training set: acute activation rose monotonically across the myeloma continuum (normal marrow through MGUS, smouldering, and MM), and metabolic remodelling increased with AML genetic risk (ρ = 0.21, p = 0.024).
OLiMPUS resolved a tumour-extrinsic PTGS2/COX-2 programme to a myeloma-specific mast-cell prostaglandin axis (HPGDS/PGD2) regulated by NFATC1/STAT1 and imputed from a plasma-cell immunoglobulin/IL32 and lymphoid programme. This nominates the mast-cell inflammatory node as a candidate TME driver in myeloma. To test whether this predicted node is druggable, we treated primary myeloma patient-chimeroids with celecoxib (COX-2 inhibitor, 20 µM), bortezomib (5 nM) and the JAK-STAT1 inhibitor ruxolitinib (250 nM) and profiled responses with flow cytometry. Bortezomib was broadly cytotoxic; ruxolitinib reduced myeloma proliferation but was associated with substantially greater depletion of non-malignant haematopoietic cells. Celecoxib selectively reduced myeloma proliferation in 3 patient chimeroids without cytotoxicity while sparing haematopoiesis, consistent with a niche origin for a prostaglandin axis previously attributed to the tumour.
Conclusions. To our knowledge this is the first framework to learn tumour-stroma covariance in an engineered human niche and transfer it to tumour-only patient data. Any tumour-only cohort already sequenced can therefore be re-read for niche state without new tissue, and the drivers it nominates can be tested in a patient relevant experimental system: here, a mast-cell inflammatory node whose modulation suppresses myeloma while sparing the healthy marrow.
DISCLOSURES
Abdullah Obaid Khan, PhD: Alethio Therapeutics - Honoraria; Sanofi - Honoraria; AstraZeneca - Research Funding; GB2202025.9 - Patents & Royalties; GB 2415266.2 - Patents & Royalties; WO 2023/156774 - Patents & Royalties; GB2402478.8 - Patents & Royalties; GB2216647.4 - Patents & Royalties; Alethio Therapeutics - Consultancy (Includes expert testimony); Sanofi - Consultancy (Includes expert testimony)