Chromatin complex dependencies reveal targeting opportunities in leukemia

Abstract

Chromatin regulators are frequently mutated in human cancer and are attractive drug targets. They include diverse proteins that share functional domains and assemble into related multi-subunit complexes. To investigate functional relationships among these regulators, here we apply combinatorial CRISPR knockouts (KOs) to test over 35,000 gene-gene pairings in leukemia cells, using a library of over 300,000 constructs. Top pairs that demonstrate either compensatory non-lethal interactions or synergistic lethality enrich for paralogs and targets that occupy the same protein complex. The screen highlights protein complex dependencies not apparent in single KO screens, for example MCM histone exchange, the nucleosome remodeling and deacetylase (NuRD) complex, and HBO1 (KAT7) complex. We explore two approaches to NuRD complex inactivation. Paralog and non-paralog combinations of the KAT7 complex emerge as synergistic lethal and specifically nominate the ING5 PHD domain as a potential therapeutic target when paired with other KAT7 complex member losses. These findings highlight the power of combinatorial screening to provide mechanistic insight and identify therapeutic targets within redundant networks.

Fig. 1. Systematic screening of chromatin regulator pairs.

a Combinatorial screening workflow with S. pyogenes (Sp) and S. aureus (Sa) Cas9s (generated using BioRender). b, c Knockout data from the 300k library screen (n = 35,684 gene pairs) with non-targeting combinations (black circles), single knockouts (dark gray circles), depleting combinations labeled for Reh in b (red circles) and THP-1 in c (blue circles), KAT7 complex members (blue triangles), and compensatory non-lethal pairs (olive circles). Dotted lines at z-score of −4 and average log2 fold change, −0.36 for Reh and −0.27 for THP-1. Data are from duplicate screens. Screening data available in Supplementary Data 3. d–g Primary screening data for a gene-gene combination (n = 8 sgRNA pairs) or gene-non-targeting combination (n = 56 sgRNA pairs) in THP-1 for: ASF1A and ASF1B in d, KAT7 complex members BRPF1 and ING5 in e (**P = 0.002), ING1 and ING2 in f (**P = 0.0059), and BRD2 and BRD3 in g (**P = 0.0015). Non-lethal pairings are highlighted (olive circles). Solid black line indicates mean. Data are from duplicates with example replicate shown. P-values are based on the two-tailed Mann–Whitney test, **P < 0.01; ***P < 0.001; ****P < 0.0001. Source data are provided as a Source Data file.

Fig. 2. Synergistic pairs converge on protein complexes and redundant paralogs.

a Validation screen workflow in 7 leukemia cell lines (generated using BioRender). ALL, acute lymphocytic leukemia. b Single knockout for top performing gene pairs tested in validation library. Data are the mean of duplicate screens. c z-score and log fold-change heatmaps of paralog combinations meeting a 1 × 10⁻⁵ FDR threshold. Rearrangements denoted with an “r” for TEL-r or MLL-r carrying cell lines. THP-1 and MV4-11 are susceptible to ARID1A knockout alone, denoted with asterisk (*). Data are derived from duplicates. d Bar plot depicting overlap of the top 5 synergistic lethal hits from each of the 7 indicated cell lines as determined by LFC and z-score. Paralog pairs highlighted by complex color denoted in e and non-paralogs in black. e Top scoring gene pairs from the 300k library screen organized by protein complexes they comprise. Asterisk (*) indicates MEAF and BRPF1 were not tested in the validation screen. Source data are provided as a Source Data file.

Fig. 3. Two mechanisms for NuRD loss.

a NuRD complex schematic. b Heatmaps of expression and single knockout data for select members of the NuRD complex. Rearrangements were denoted with an “r” for TEL-r or MLL-r carrying cell lines. DepMap 21q1 release. c Heatmap of combinatorial knockouts of top NuRD complex pairings tested in follow up screen. d Gene expression heatmaps of DEseq differential genes in indicated knockout conditions for 7 days in Reh (75 genes), THP-1 (132 genes) and MV4-11 (108 genes). Data are normalized to pairwise non-targeting sgRNA transduced cells and average of 2 replicates per condition, TPM + 1. Flow cytometry gating strategy found in Supplementary Fig. 10. e Scatter plot depicting log2-fold change expression in MTA1;MTA2 KO vs. CHD4 KO for THP-1, Reh and MV4-11. Pearson correlations denoted on each plot. RNAseq data provided in Supplementary Data 5. Source data are provided as a Source Data file.

Fig. 4. ING5 synergy with KAT7 complex partners.

a KAT7 complex schematic. b Expression and single knockout heatmaps for select members of the KAT7 complex. DepMap 21q1 release. c Heatmap of combinatorial knockouts of top KAT7 complex pairings tested in validation screen. Data are repeated from Fig. 2c. d ING5 protein annotated with ING (purple) and PHD (yellow) Pfam domains and sgRNA targeting sites. e Competition experiment of ING5 knockout single sgRNAs delivered by mCherry vector and ING4 KO or safe harbor targeting (Safe) sgRNAs delivered by GFP vector to THP-1 cells and analyzed by flow cytometry after 6 and 14 days. Sample represents 3 replicates collected at each respective timepoint. Data are presented as mean values ± standard deviation. See Supplementary Fig. 10 for flow gating strategy and Supplementary Data 6 for values. P-values are based on the two-tailed Welch’s t test, *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 f, Log₂ fold change differences for all tested ING5 ING and PHD targeting sgs in ING4 KO cells. Triangles represent sg7. Data presented as median, boxes denoting upper and lower quartiles and whiskers the range. g Relative read frequency of ING5 PHD knockout when paired with non-targeting (NT, 14 sgRNAs), ING4 (2 sgRNAs), BRPF1 (2 sgRNAs), BRPF1 (2 sgRNAs), and MEAF6 (2 sgRNAs) after 21 days in THP-1. Combinations were tested in duplicate, solid black bars at mean. Source data are provided as a Source Data file.