Functional Dissection of Cellular Programs to Uncover Novel Gene Dependencies in AML

Genome-scale CRISPR-Cas9 screens have the power to unveil the Achilles' heel of neoplastic cells. Typically, analyses of such large-scale data sets focus on single gene dependencies. An alternative strategy is to evaluate functional networks enriched in a disease or molecular subset of interest. We applied this strategy to the Broad Institute's Cancer Dependency Map (DepMap) data set consisting of genome-scale CRISPR-Cas9 screens in over 1000 cancer cell line models. We hypothesized that interrogating enriched pathways with context-specific dependencies leads to the discovery of functionally informed gene dependencies.

Single sample gene set enrichment analysis (ssGSEA) of over 1000 cancer cell lines in the DepMap identified hematologic malignancies to be highly enriched for signatures associated with elevated transcriptional activity. Notably, within hematologic malignancies, acute myeloid leukemia (AML) with a rearrangement in the KMT2A gene ( KMT2Ar-AML) was the most significantly enriched for dependency on transcriptional activity-associated gene signatures. We next interrogated KMT2Ar-AML gene dependencies (excluding common essential genes) with the human kinase database KinMap (http://www.kinhub.org/kinmap/) to identify readily druggable genes involved in transcriptional activity. Cyclin-dependent kinase 13 ( CDK13) was the only gene meeting these criteria. Further DepMap data analysis confirmed that KMT2Ar-AML cell lines were indeed more strongly dependent on CDK13 than all other cancer cell lines screened. Moreover, when we performed GSEA on the gene dependencies observed in CDK13-dependent versus non-dependent cell lines, we found transcriptional activity and KMT2A-target gene sets as top hits.

We next validated AML cell dependency on CDK13 with orthogonal genetic approaches (CRISPR-Cas9 knock-out and shRNA) in vitro in human AML cell lines and cells from patient-derived xenograft (PDX) models of KMT2Ar-AML. Perturbation of CDK13 induced cell death with hallmarks of apoptosis. Doxycycline-inducible shRNA directed against CDK13 also impaired AML progression in the peripheral blood and bone marrow in an MV4-11- KMT2Ar xenograft. Moreover, in an orthotopic PDX model, doxycycline inducible knock-out of CDK13 in mice with established disease significantly reduced leukemia burden in the peripheral blood and bone marrow.

CDK13 is reported to regulate transcriptional elongation and the clearance of prematurely terminated RNAs. To decipher why AML cells, particularly KMT2Ar, need CDK13 to survive, we identified the CDK13 binding sites and histone marks involved in transcriptional regulation in AML using CUT and RUN and ChIP-seq: We determined that CDK13 co-localizes with H3K4me3 and H3K27ac at promoter sites. Global gene expression studies following the knock-out of CDK13 revealed that the majority of CDK13 bound genes were also differentially expressed, with significantly more genes decreased than increased in expression after CDK13 knock-out. GSEA of CDK13-bound genes with decreased expression upon CDK13 knock-out identified gene signatures associated with transcriptional activation and KMT2A-targets. We next studied the characteristics of CDK13 target genes and observed a predominance of long genes (total length of gene exons > 10 kb, GRCh38.p12). GSEA of genes ranked by size, confirmed that longer genes were enriched for CDK13 target genes, CDK13 differential AML dependencies, and for KMT2A-target transcriptional signatures. Additionally, following CDK13 perturbation in KMT2Ar-AML we observed a decrease of Pol2/pPol2Ser2 occupancy at the transcription end site of long genes.

In summary, using functional pathway analyses, we identified CDK13 as a candidate gene dependency enriched in KMT2Ar-AML, a target not immediately prioritized with single gene analysis. Mechanistically, our data support that CDK13 controls Pol2 processivity, with CDK13 perturbation leading to a decreased occupancy at the 3' end of long genes. KMT2Ar, CDK13 dependent cells are also enriched for dependency on long genes, thereby suggesting a molecular basis for the CDK13 dependency in KMT2Ar-AML.