Unraveling Molecular Subgroup-Specific Pathway Dependencies in Chronic Lymphocytic Leukemia through Drug-Perturbed Transcriptomic Profiling and Statistical Modeling

Gene expression profiling following targeted pathway perturbations on primary patient samples provides important information for understanding heterogeneous pathway dependencies in cancer, thereby supporting the design of personalized therapies. However, analyzing such high-dimensional data poses a challenge. In this study, we developed a novel computational framework utilizing semi-supervised factor analysis, regularized multiple linear regression, and causal inference to analyze and interpret drug-perturbed transcriptomic profiles from patient samples. We applied our framework to RNA sequencing data obtained after ex-vivo perturbation with ten different small-molecule inhibitors in primary tumor cells from 108 chronic lymphocytic leukemia (CLL) patients, yielding 1010 transcriptional drug response profiles (Do et al., ASH2023 abstract 4633).

To capture and quantify the shared and unique dimensions of gene expression changes induced by different pathway inhibitors, we employed guided sparse factor analysis (GSFA), originally designed for analyzing single-cell CRISPR screening (Zhou et al., Nat Methods, 2023). We identified perturbations uniquely associated with single factors, such as XPO1 inhibition by selinexor, MDM2 inhibition by nutlin-3a, and MEK1/2 inhibition by trametinib, suggesting the induction of distinct downstream signatures not shared with other perturbations. Conversely, perturbations like BTK inhibition by ibrutinib, PI3K inhibition by duvelisib, and mTOR inhibition by everolimus were linked to multiple factors shared across perturbations, suggesting extensive downstream pathway cross-talks. Notably, PI3K inhibition was significantly associated with two factors, one shared with BTK inhibition, and the other with PI3K and mTOR inhibition. Pathway enrichment analysis revealed that the factor shared between BTK and PI3K was enriched for the B-cell receptor signaling pathway, while the factor shared among PI3K, ATK, and mTOR inhibitions was enriched for MAP kinase pathway genes. This highlights the complex role of PI3K signaling in CLL and provides novel insights into the hierarchical pathway structure in CLL biology.

We further examined the heterogeneity of pathway activity changes upon perturbation among patients with different driver mutations. Based on the GSFA model, we derived a per-patient perturbation score for each factor and associated them with major molecular subgroups in CLL using regularized multiple linear regression. We observed a strong association between TP53 mutational status and the factor representing MDM2 inhibition, reflecting the known dependencies of DNA damage response signaling on TP53 mutational status. Furthermore, this approach uncovered that IGHV mutational status strongly determines the downstream MAP kinase activity change upon PI3K/mTOR/AKT inhibition but not the BCR signaling activity change upon PI3K and BTK inhibition. By testing for interaction in linear models, we further investigated the dependence of individual gene expression changes upon pathway perturbation and the presence of driver mutations. Thanks to the high-dimensional nature of gene expression profiling, we could infer causal interactions between mutations and targets of perturbation using the method developed by Fischer et al., eLife, 2015. We successfully recaptured the known signaling structure of the BCR-PI3K cascade and TP53-MDM2 interaction. Moreover, we unveiled other novel interactions, such as the activation of BTK and MEK signaling by trisomy12, explaining our previous observations of increased sensitivity to BTK and MEK inhibitors in CLL samples with trisomy12 (Dietrich et al., J Clin Invest, 2018).

This study underscores the value of perturbed transcriptomic profiling of primary cancer patient samples for mapping detailed pathway structures and dependencies on disease drivers. Our comprehensive computational workflow serves as a guide for mining and interpreting complex high-dimensional data, advancing our understanding of disease biology, and facilitating the rational design of novel personalized therapies, including combinational treatments.