“Guided” Unsupervised Multiplexed Cytof Analysis Enhances Detection of Minimal/Measurable Residual Disease in Acute Myeloid Leukemia

Multiparameter flow cytometry (MFC) and molecular techniques remain the cornerstone for detecting acute myeloid leukemia (AML) minimal/measurable residual disease (MRD) with precision. However, a significant number of MRD(-) patients still relapse, indicating that current methods may miss persistent leukemia due to lack of sensitivity based on suboptimal limits of detection or the presence of distinct leukemia-associated phenotypes (LAP), underscoring the need for improved MRD detection methods.

MFC MRD assessment has several limitations: a) reliance on a limited set of aberrant markers not universally expressed in all AML cells; b) incomplete coverage of LAPs; c) clonal evolution resulting in new leukemic clones undetected by current panels; d) the need for specialized expertise; and e) restriction to surface marker assessment, excluding critical intracellular antigens. We hypothesize that single-cell proteomics using CyTOF to analyze more features per cell, combined with unsupervised analysis, will enhance AML MRD detection. This approach can better distinguish MRD cells from healthy HSCs/HSPCs, enabling the agnostic identification of aberrant features by automatically capturing these characteristics without the need for predefined markers.

To explore this notion, we conducted integrated CyTOF analysis of 60 AML patients and 12 bone marrow (BM) samples. This revealed diverse AML proteomic profiles distinct from HSCs/HSPCs showing that unsupervised methods can capture various aberrant leukemia profiles in a feature-agnostic manner. To simulate AML MRD conditions, we mixed barcoded primary AML cells (n=5) and AML cell lines (Kasumi, Molm13 and OCI-AML3) with healthy BM cells at ratios of 1:10, 1:100, 1:1,000, 1:10,000, and 1:100,000. These samples were analyzed using a 50-parameter CyTOF panel. UMAP analysis showed that primary AML cells and cell lines were well-segregated up to a 1:1,000 ratio, with gradual deterioration in segregation at higher dilutions. At a 1:100,000 dilution, AML cells were indistinguishable from healthy HSCs/HSPCs, highlighting UMAP's limitations in detecting ultra-rare populations due to its reliance on density estimation and local structure preservation. Despite strategic adjustments of UMAP parameters, including neighbors, minimum distance, type of distance metric, and epochs, segregation did not improve. Detailed UMAP plots revealed that AML cells clustered distinctly from healthy HSCs/HSPCs at low dilutions but overlapped with them at higher dilutions, suggesting that UMAP clusters rare cells with more abundant neighboring cells based on overlapping features rather than distinguishing them as a distinct cluster, as seen at lower dilutions.

This observation led us to hypothesize that the presence of a more abundant population with similar proteomic features could “guide” the clustering of AML MRD cells as a distinct entity alongside these abundant and phenotypically similar neighbors, thereby distinguishing them from healthy HSCs/HSPCs. This scenario mirrors clinical settings where longitudinally collected pre- and post-therapy AML samples are pooled and subjected to unsupervised analysis. In such analyses, the presence of baseline AML cells can facilitate and guide the clustering of AML MRD cells. To test this hypothesis, we conducted an in-silico experiment where AML cells from eight samples were mixed with healthy BM cells at various dilution. Our results demonstrated that even at a 1:100,000 dilution (23 AML cells and 2.3 million healthy BM cells), AML cells distinctly clustered from healthy HSCs/HSPCs in all eight phenotypically different samples when guided by the presence of more abundant AML cells. Furthermore, we extended these findings by performing integrated CyTOF analysis of pre- and post-therapy AML samples (n=10). The “guided” unsupervised analysis, facilitated by the inclusion of phenotypically similar baseline samples, substantially improved the detection of AML MRD.

In conclusion, our study suggests that multiplexed CyTOF analysis with a “guided” unsupervised analytic approach can significantly improve the detection and characterization of AML MRD, greatly enhancing the precision of MRD detection in clinical settings. In addition of enumerating MRD cells with very high sensitivity, CyTOF also enables the analysis of proteomic profiles of MRD, thus potentially guiding AML MRD therapy.

Andreeff:Eterna: Current holder of stock options in a privately-held company, Honoraria, Research Funding; Glycomimetics: Honoraria; Ona: Honoraria; SentiBio: Current holder of stock options in a privately-held company, Honoraria, Research Funding; Aptose: Honoraria; Syndax: Honoraria, Research Funding; Sellas: Honoraria, Research Funding; Boehringer-Ingelheim: Honoraria; Paraza: Honoraria; Roivant: Honoraria; Chimerix: Current holder of stock options in a privately-held company; Oncolyze: Current holder of stock options in a privately-held company; Daiichi-Sankyo: Research Funding; Oxford Biomedical: Research Funding; Ellipses: Research Funding; Kintor Pharmaceutical: Research Funding.