Q34 - Metabolic optical biomarkers reveal molecular and therapeutic heterogeneity in leukemia

Introduction: : Acute myeloid leukemia (AML) is an aggressive blood cancer that remains difficult to treat due to high relapse rates and significant heterogeneity among patients, manifesting in varied genetic and molecular profiles as well as diverse clinical presentations and outcomes. The traditional options, such as chemotherapy and stem cell transplantation, have limited success, particularly for older patients who cannot tolerate intensive therapy. The epi/genetic and transcriptomic-based biomarkers are inadequate for accurately predicting treatment response or guiding effective therapeutic strategies. This underscores the urgent need for innovative biomarkers that can enhance risk stratification and overcome drug resistance, for personalized treatment and improved patient outcomes.
Metabolic alterations are a hallmark of cancer, driving drug resistance, immune evasion, and disease progression. The metabolic reprogramming in AML cells includes rewiring of glycolysis and mitochondrial respiration, altered lipid metabolism, and changes in amino acid utilization. These metabolic alterations contribute to the uncontrolled growth of leukemic cells and resistance against conventional therapies. Despite the importance of metabolism in AML, prognostic biomarkers reflecting metabolic changes are mostly inferred from secondary transcriptomic or epigenetic information. The ability to track metabolic shifts and plasticity represents new avenues for diagnostic and therapeutic interventions in AML.
In the presented study, we utilized fluorescence lifetime imaging microscopy (FLIM) to track the intensity and lifetime changes of the autofluorescent metabolic coenzymes, NAD(P)H and FAD, among 43 human leukemia cell lines. Machine-learning-assisted analysis revealed that imaging-based metabolic biomarkers can accurately profile leukemia cell identity and state, as well as predict their response to chemo-drug treatment.

Materials and

Methods: : Leukemia cell lines were cultured in RPMI + 10% or 20% FBS. Fluorescence intensity and lifetime images of NAD(P)H and FAD in leukemia cells were acquired by a two-photon microscope (Leica SP8 FALCON) with a live cell workstation. RNA-seq and chemo-drug response data of leukemia cell lines were sourced from Cancer Cell Line Encyclopedia and Genomics of Drug Sensitivity in Cancer. Image processing and analysis were carried out in Python 3.9.

Results, Conclusions, and Discussions:: We acquired FLIM images of 60,802 single cells from 43 human leukemia cell lines. Four channels of metabolic optical biomarkers (MOBs) were derived from the raw FLIM images to represent key metabolic phenotypes: fluorescence intensity of NAD(P)H, optical redox ratio (ORR), enzyme-bound NAD(P)H fraction (bound), and fluorescence lifetime of enzyme-bound NAD(P)H (bound). Artificial intelligence (AI)-assisted analysis of the FLIM data generated 525 MOB features, including 205 statistical features and 320 AI-derived features (Fig. 1A). This comprehensive dataset enabled high-content profiling of leukemia phenotypes, effectively distinguishing various leukemia types, including different acute lymphoblastic leukemia (ALL) and AML subtypes (Fig. 1B, C). The combination of FLIM imaging and AI-assisted feature extraction allowed for extensive information capture, supporting mathematical models that predict underlying gene expression pathways and responses to chemotherapeutic agents. Of the 781 pathways analyzed, 296 were confidently predicted (Pearson r > 0.4) based on the observed MOB features (Fig. 1D). These pathways encompass critical aspects of leukemia biology, such as signal transduction, metabolism, cell cycle regulation, and cytoskeletal dynamics. Additionally, our approach accurately predicted pathways not well-characterized in leukemia, such as androgen receptor signaling (Fig. 1D). Furthermore, our model, utilizing MOB features as input, demonstrated high accuracy in predicting chemotherapeutic drug responses of different leukemia cell lines to 228 chemo-drugs, highlighting its potential for clinical application (Fig. 1E). To validate our model, we induced a Cytarabine (AraC) resistance line from the wild-type MOLM13 cell line. MOB profiling distinguishes the AraC-resistant state from wild type and correctly predicts their respective IC50 (Fig. 1F). Currently, we are applying our method to profile AML patient samples and to predict their response to approved chemo-drugs. Overall, this study underscores MOBs as novel biomarkers for diagnosis and real-time monitoring of treatment responses, and paves the way for the development of more effective and less toxic therapeutic strategies for leukemia.

Acknowledgements and/or References (Optional)::