2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Saturday, December 12, 2026
05:30 p.m. - 07:30 p.m.
MCCNO - Hall F
Technology and Procedures, Assays, Molecular Testing, Omics Technologies, Pathology, Bioinformatics, Clinical Procedures, Computational Biology, Profiling, Treatment Considerations, Combination Therapy, Drug Development, Therapy Sequence, Biological Therapies, Transplantation, Monoclonal Antibody Therapy, Adverse Events, Artificial Intelligence (AI), Machine Learning, Biological Processes, Pathogenesis, CHIP, Genomics, Hematopoiesis, Immune Mechanism, Molecular Biology, Clinical Practice (Health Services and Quality), Patient Safety, Practice Models, Quality of Care, Diseases, Hemoglobinopathies, Sickle Cell Disease, Thalassemia, Immunology, GVHD, Lymphoid Malignancies, Myeloid Malignancies, Acute Myeloid Malignancies, Chronic Myeloid Malignancies, Research, AML, Translational Research, MDS, Clinical Research, Clinical Trials, Health Outcomes Research, Patient-Reported Outcomes, Real-World Evidence, Registries, Study Population, Human, Adult, Elderly, Pediatric, Animal Model
PAPER

2486 Unsupervised factor analysis identifies an early cytotoxic immune signature that independently predicts relapse following hematopoietic stem cell transplant.

Jennifer Vidler, MBBS, MRCP, FRCPath, Yegnasew Teferi, PhD2, Amandine Tan3, Victoria Potter, MBBS, BSc, FRACP, FRCPA4, Giorgio Napolitani5, Pramila Krishnamurthy, FRCPath, PhD6
1 King's College Hospital NHS Foundation Trust, London, United Kingdom, King's College London, London, United Kingdom
2 King's College London, London, United Kingdom
3 King's College London, London, United Kingdom
4 King's College Hospital NHS Foundation Trust, London, United Kingdom
5 King's College London, London, United Kingdom
6 King's College Hospital NHS Foundation Trust, London, United Kingdom, King's College London, London, United Kingdom

Disease relapse is the major cause of treatment failure following haematopoietic cell transplantation (HCT) for myeloid malignancies. Immune signatures predictive of future relapse following T-cell depleted HCT are lacking. Identifying early predictors of relapse could trigger therapeutic intervention to avert relapse and is a critical area of unmet clinical need.

PRO-DLI (NCT02856464) was a multi-centre, phase 2, prospective clinical trial of donor lymphocyte infusions (DLI) in patients undergoing Alemtuzumab-based, T-cell depleted HCT for myeloid malignancies in the United Kingdom. 266 longitudinal peripheral blood mononuclear cell samples were analysed from 61 trial subjects. These clinically annotated translational samples were analysed by CyTOF to allow deep immunophenotyping of the T-cell compartment over the first year post-HCT.

Unsupervised clustering of T-cells using the FlowSOM R package identified 38 phenotypically distinct clusters, which were analysed using the multi-omics factor analysis (MOFA) extension MEFISTO to identify immune signatures associated with relapse. This technique allows identification of latent immune factors (shifts in coordinated immune profiles over time) within high-dimensional data, rather than comparing changes in individual cell subsets. Differences in latent immune factors between the 19 relapse and 41 remission samples at D+28 post-transplant were interrogated. Nine relapse patients with available samples at this timepoint were MRD negative by flow-cytometry. Relapse occurred at a median of 187 days post-HCT (96–384 days).

MEFISTO identified 6 latent immune factors, of which one was differentially represented at D+28 in relapsing patients when adjusted for multiple testing (BH-adj p=0.028). This relapse-associated immune signature was characterised by a shift towards a CD8+ cytotoxic immune programme dominated by terminally differentiated cytotoxic effector cells and away from CD4+ effector, naive and memory cells.

This CD8+ cytotoxic–CD4+ axis was independently associated with relapse (univariate logistic regression: OR 2.30, 95% CI 1.23–4.70, p=0.014, AUC = 0.69) and was not associated with clinical variables such as CMV reactivation, GvHD or trial arm. Each one standard deviation increase in the CD8+ cytotoxic-CD4+ axis factor score increased the odds of subsequent relapse by ~ 2.3-fold. Model-adjusted mean factor scores of this immune axis were significantly increased at D+28 in the relapse group (p=0.020) but not later timepoints, suggesting this immune program was established early post-HCT in those who went on to relapse. At 1-year post-HCT, in recipients of DLI, this factor trended towards the more balanced CD4+ end of the axis (p=0.033, BH-adjusted p=0.066) and away from the cytotoxic profile, compared to those who did not receive DLI.

Confirming the robustness of this factor, the median abundance of 9 of top 5 T-cell subsets contributing to the positive and 5 contributing to the negative loadings of this factor changed in the direction predicted by the model.

A multivariate logistical regression model was derived to investigate the independent predictive value of the latent factor score in addition to established clinical predictive data available at D+28 (Disease Risk Index, age). Higher CD8+ cytotoxic-CD4+ axis factor score was independently associated with relapse when adjusting for age and Disease Risk Index score (adj OR 2.34, 95% CI 1.28–4.71, p=0.009) and improved the model fit compared to the clinical data alone (p=0.005). The addition of available CD3 donor chimerism data (n=41) did not improve the model and was not independently associated with relapse (p=0.44) but the association remained significant (adj OR 2.68, 95% CI 1.16–7.56, p=0.035).

We used unsupervised computational models to discover a novel T-cell immune axis, identified D+28 post-HCT, that is predictive of future relapse. This immune axis was defined by a shift towards terminally differentiated CD8+ cells and away from a more balanced CD4+ effector/memory state. This may represent an ineffective inflammatory post-HCT immune state providing a sub-optimal graft-vs-leukaemia effect. The addition of this factor score to a model of currently available clinical data improved the prediction of future relapse. Further work will focus on validating these findings and characterising an immune signature of future relapse to allow timely therapeutic intervention.

Jennifer Vidler, MBBS, MRCP, FRCPath: No relevant conflicts of interest to declare.