Cancer Pathway Connectivity Resolved By Drug Perturbation and RNA Sequencing

Characterization of cancer subgroups is largely based on omics profiling of patient cohorts and studying inter-patient heterogeneity. However, linking molecular differences to a function has been limited due to their static nature at steady-state. Here, we show that gene expression profiling (GEP) of responses to targeted pathway perturbation in primary chronic lymphocytic leukemia (CLL) samples offers an additional molecular layer for disease classification.

We selected 108 CLL patient samples with genetic annotation and exposed them ex-vivo to small-molecule inhibitors used clinically (ibrutinib/BTKi, duvelisib/PI3Ki, trametinib/MEKi, everolimus/mTORi, selinexor/XPO1i) or targeting key signaling pathways (compound 26/TLRi, MK2206/AKTi, nutlin-3a/MDM2i, IBET872/BETi) for 48h. GEP was performed by 3’-end-based QuantSeq with shallow depth RNA sequencing (Do et al., ASH2023 abstract #4633) resulting in 1010 drug response profiles with a median sample coverage of 2.06 million reads. Sample viability was measured by flow cytometry before and after perturbation.

First, we performed principal component analysis (PCA) on all samples in which transcriptional variation was predominantly shaped by sample viability (PC1), which explained 12.0% of the variation. This gene signature was associated with activation of hypoxia, p53 pathway and TNFα signaling via NFκB. Additionally, we identified downregulation of MYC targets, OXPHOS and protein secretion, preferentially in drug-perturbed samples. The molecular steady-state features IGHV and methylation status (PC2) and trisomy 12 (PC3) explained 8.1% and 5.1% of the variation, respectively.

Next, we performed differential gene expression analysis (DGEA) between samples either treated with drug or control. We found most differentially expressed genes (DEGs) after inhibition of the B cell receptor (BCR) pathway by PI3Ki (1574) and BTKi (991), followed by AKTi (469), mTORi (596) and MEKi (455), and p53 pathway activation by MDM2i (253). No DEGs were found for TLRi. Perturbation of broader processes such as nuclear exportation (XPO1i) and chromatin modification (BETi) resulted in 450 and 277 DEGs, respectively (FDR < 0.5%).

We modeled the relationship between drugs and perturbed genes with directed acyclic graphs to construct networks and dissect the connection of pathways. Within the main cluster which contained the kinase inhibitors and shared mostly suppressed genes, BTKi and PI3Ki were closely connected by genes enriching in BCR signaling, and immune cell activation and response processes, followed by more downstream targets of mTORi, AKTi and MEKi. The remaining drugs were only weakly connected to the main cluster and presented with distinct sets of induced genes, e. g. p53 targets after MDM2i. Direct comparison of the BCR inhibitors identified 270 DEGs (FDR < 10%) which revealed suppression of IFN response after PI3Ki but not BTKi. We assembled the pathway connectivity of CLL by integrating individual perturbed gene expression signals across the large patient cohort.

As IGHV is a key molecular and prognostic disease dimension and the strongest modulator of gene expression in CLL, we assessed how this signature was modulated by each drug. We found 790 DEGs between unmutated and mutated cases among the control samples and decreased numbers after pathway inhibition, namely PI3Ki (664), BTKi (623) and XPO1 (532) (FDR < 5%) suggesting a dynamic component whose maintenance depends on active BCR signaling. We also identified a stable component that was not modified by any perturbation including ZAP70 and CD38.

We compared the capacity to map pathway functions to disease drivers using profiles with and without perturbation. Gene set enrichment analysis of DEGs between cases with del17p and wildtype (WT) at steady-state yielded results driven by gene dosage effects. When we tested for interactions between del17p status and drug response, and used these results for further downstream analysis, del17p cases under MDM2i pressure showed impaired activation of the p53 and TNFα pathways compared to WT.

In conclusion, the addition of targeted pathway perturbations to the GEP of primary CLL samples can greatly enhance the potential for molecular and functional classification of disease and subgroups. Our study provides a blueprint to use perturbed omics profiling and interaction testing to link disease drivers to pathway activation and function.