Applications of data science to investigate risk factors in allogeneic stem cell transplantation

Greenwood, David Rhys (2023). Applications of data science to investigate risk factors in allogeneic stem cell transplantation. University of Birmingham. Ph.D.

Greenwood2023PhD.pdf
Text - Accepted Version
Restricted to Repository staff only until 1 December 2025.
Available under License All rights reserved.
Download (10MB) | Request a copy

Abstract

Allogeneic haematopoietic stem cell transplantation (allo-HSCT) from healthy donors can provide curative therapy for a variety of life-threatening haematological malignancies. However, the risk of relapse, graft-vs-host disease (GvHD), infection, and toxicity remain major causes of mortality.

Mature immune cell populations are also infused as a component of the haematopoietic stem cell (HSC) graft. We aimed to investigate how the cellular composition of grafts varied between donors and to study associations of graft composition with recipient prognosis. Moreover, we intended to examine proteins measurable in the serum post-transplant as potential prognostic biomarkers.

Cell populations were identified in graft samples through the application of unsupervised clustering to single-cell mass cytometry (CyTOF) data. Cluster analysis for compositional data was used to define distinct types of graft based on cellular composition. By applying supervised variable selection algorithms to proteomics data, prognostic biomarkers were identified in serum samples taken on the 14\(^{th}\) day post-transplant. Lastly, survival and competing risks regression were used to test if graft composition and potential biomarkers were associated with prognosis following allo-HSCT.

The cellular composition of stem cell grafts was shown to be highly variable between donors with HSC representing a small fraction of cellular content. We found evidence that differences in graft composition were associated with significant changes in prognosis beyond that attributable to other established risk factors. Based on cellular composition, we were able to define three distinct types of grafts that were linked with differences in overall survival (OS), GvHD-free relapse-free (GRFS) survival, and transplant-related mortality (TRM). Concerning the contribution of individual cell types, a greater proportion of T cells (relative to HSCs) was predictive of an increased incidence of chronic GvHD. Furthermore, monocytes and a population of DNAM-1\(^{+}\) lymphocyte progenitors had a protective effect on the rate of GRFS and incidence of TRM, respectively.

Several serum proteins were identified as potential prognostic biomarkers when measured on the 14\(^{th}\) day following transplantation. High concentrations of PD-L1, CCL23, and CDCP1 expression were predictive for lower rates of OS, GRFS, and increased incidence of relapse-related morality, respectively. Moreover, IL-6 concentration was associated with a lower incidence of relapse-related mortality but an increase in deaths from other causes.

These findings reveal that graft composition and proteins expressed in the serum after allo-HSCT are important predictors of clinical prognosis. This work demonstrates the benefit of applied data science for biomarker discovery and highlights therapeutic opportunities for graft modification in the treatment of haematological malignancies with allo-HSCT.

During the development of this thesis, I spent a period of time on secondment to COVID-19 research. A paper published on this topic is integrated as Appendix A.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):