Heterochromatin fidelity is a therapeutic vulnerability in lymphoma and other human cancers

Abstract

Genes involved in the regulation of chromatin structure are frequently disrupted in cancer, contributing to an aberrant transcriptome and phenotypic plasticity. Yet, therapeutics targeting mutant forms of chromatin-modifying enzymes have yielded only modest clinical utility, underscoring the difficulty of targeting the epigenomic underpinnings of aberrant gene regulatory networks. Here, we sought to identify novel epigenetic vulnerabilities in diffuse large B-cell lymphoma (DLBCL). Through phenotypic screens and biochemical analysis, we demonstrated that inhibition of the H3K9 demethylases KDM4A and KDM4C elicits potent, subtype-agnostic cytotoxicity by antagonizing transcriptional networks associated with B-cell identity and epigenetically rewiring heterochromatin. KDM4 demethylases associated with the KRAB zinc finger ZNF587, and their enzymatic inhibition led to DNA replication stress and DNA damage-einduced cGAS-STING activation. Broad surveys of transcriptional data from patients also revealed KDM4 family dysregulation in several other cancer types. To explore this potential therapeutic avenue, we performed high-throughput small molecule screens with H3K9me3 nucleosome substrates and identified novel KDM4 demethylase inhibitors. AI-guided protein-ligand binding predictions suggested diverse modes of action for various small molecule hits. Our findings underscore the relevance of targeting fundamental transcriptional and epigenetic mechanisms for anti-cancer therapy.

Figure 1.. Phenotypic small molecule screens identify novel anti-cancer agents for DLBCL.

A) Experimental scheme to identify compounds with anti-cancer activity against DLBCL via arrayed phenotypic screens. HBL-1 (ABC subtype) and OCI-Ly1 (GCB subtype) cells were treated with a library of 145 small molecules targeting chromatin factors at multiple doses (100 nM, 500 nM and 1 μM) for five days and cell proliferation was measured via CellTiter-Glo (CTG) luminescence assays. All compounds were screened with three replicates per dose. B) Fold change in CTG signal (1 μM of compounds compared to 0.01% DMSO vehicle controls) for HBL-1 and OCI-Ly1 cells after five days of treatment. Each datapoint represents the average fold-change in CTG signal per compound across all screen replicates. Compounds exhibiting a log₂(fold change) < −1 and an FDR-adjusted P value < 0.0001 were defined as hits. C) Cytotoxicity of hits across all screening doses in HBL-1 and OCI-Ly1 cells. Compounds are rank ordered by CTG signal relative to DMSO on day 5 after treatment and grouped by conserved hits across both cell lines (top), hits only in HBL-1 cells (middle), and hits only in OCI-Ly1 cells (bottom). Relative CTG signal is the average of three screening replicates.

Figure 2.. JIB-04 exhibits potent anti-cancer activity against diverse subtypes of DLBCL.

A) CellTiter-Glo time-course of OCI-Ly1 cells treated with various doses of JIB-04 over 3 days. CTG signal at each time-point is normalized to the starting day 0 baseline signal. Error bars represent standard deviations across three biological replicates per dose and time-point. 0.1% DMSO was used as a vehicle control. B) JIB-04 dose titration across a broad compendium of DLBCL cell lines. CTG signal at each dose for each cell line is normalized to DMSO vehicle controls. Error bars represent standard deviations across three biological replicates per dose. C) Dose titration comparison of JIB-04 and EZH2 inhibitor, EPZ011989 in OCI-Ly1 cells. CTG signal at each dose is normalized to the CTG signal of 0.01% DMSO vehicle control. Error bars represent standard deviations across n=3 biological replicates per dose. D) In vivo treatment of tumor xenografts with JIB-04. Luciferase-expressing OCI-Ly1 cells were transplanted to NSG mice via tail vein injection and tumor burden was assessed after 10 days. Mice were treated daily for 25 days with JIB-04 (30 mg/kg) or DMSO vehicle control. The scale range in radiance (p/sec/cm²/sr) is noted for each timepoint.

Figure 3.. JIB-04 treatment alters the genome-wide heterochromatin landscape.

A) Heatmap of statistically significant (FDR-adjusted P value < 0.05) differences in TF motif archetypes in H3K27ac and H3K4me3 ChIP-seq peaks between 500 nM JIB-04 and DMSO treated OCI-Ly1 cells. Motif archetypes are derived from Vierstra, et al. (2020). Nature. (PMID: 32728250). The archetype identifiers from the v2.1beta release are noted in parentheses and the consensus TF in the motif cluster is labeled. Color bar represents the chromVAR deviation Z-score. The standard deviation of chromVAR deviation Z-scores across all samples in a ChIP-seq experiment is plotted along the rows of the heatmaps and indicative of the variability in ChIP-seq signal over motifs between treatment conditions. B) Immunoblot for IKZF1 and IKZF3 from OCI-Ly1 and HBL-1 cells treated with multiple doses of JIB-04 for 24 hours. C) Quantification of the number of ChIP-seq regions that significantly gain or lose H3K27ac, H3K4me3, H3K9me2, or H3K9me3 in 500 nM JIB-04 versus DMSO treated OCI-Ly1 cells. ChIP-seq regions that exhibit an FDR adjusted P-value < 0.05 are considered statistically significant. D) Quantification of differentially expressed TE families between 500 nM JIB-04 and DMSO treated OCI-Ly1 cells for RepeatMasker annotated DNA, LINE, LTR, Satellite, SINE and Unknown TE classes. Differentially expressed TEs are defined as FDR-adjusted P value < 0.05 and log₂(fold change) > 0.75 and log₂(fold change) < −0.75. E) Normalized H3K9me3 ChIP-seq signal over satellite regions transcriptionally up-regulated in 500 nM JIB-04 versus DMSO treated OCI-Ly1 cells. F) Genome sequencing tracks of the 19p12 locus (chr19:21,325,513–24,219,756 in hg38 coordinates) visualizing H3K9me3 ChIP-seq in DMSO and 500 nM JIB-04 treated OCI-Ly1 cells. RepeatMasker-annotated BSR/beta satellite elements are shown in the bottom row.

Figure 4.. KDM4 proteins are targets of JIB-04 and vulnerabilities in DLBCL.

A-B) KDM4A (A) or KDM4C (B) in vitro enzymatic activity was monitored by the conversion of H3K9me3 nucleosome substrate to H3K9me2 product across a titration of JIB-04 or PDCA (positive control). Error bars represent standard deviations across three replicates per dose. AlphaLISA luminescence signal at each dose is normalized to parallel DMSO vehicle controls. C-D) Cellular thermal shift assays indicating that JIB-04 stabilizes KDM4A (C) and KDM4C (D) within OCI-Ly1 cells. Expected molecular weights: β-Actin, 42 kDa, KDM4A, 150 kDa, KDM4C, 120 kDa. Relative KDM4A/C protein signal normalized to the β-Actin loading control is quantified for each sample below each blot. E) CellTiter-Glo time-course of OCI-Ly1 cells treated with KDM4 inhibitor QC6352 across various doses. CTG signal at each time-point is normalized to the starting day 0 baseline signal. Error bars represent standard deviations across three biological replicates per dose and time-point. 0.02% DMSO was used as a “0 nM” vehicle control. F) Colony forming potential of OCI-Ly1 and HBL-1 cells exposed to DMSO, JIB-04 or QC6352 within methylcellulose media. Each point represents an independent biological replicate and error bars reflect one standard deviation. Colonies were scored in a blinded manner after two weeks of culture. Asterisks indicate an FDR < 0.05, computed by comparing each drug treatment condition with DMSO controls (two-tailed Wilcoxon rank-sum test, corrected for testing of multiple modules). G) Fold change in CellTiter-Glo signal following 6 days of CRISPRi-mediated dual knockdown of KDM4A/C in comparison to non-targeting control sgRNAs. CRISPRi machinery is dox inducible and thus fold changes for each sgRNA condition are relative to -dox controls. Each point represents an independent biological replicate. H) In vivo tumor volume time-course of tumor xenografts in NSG mice. Cas9-expressing OCI-Ly1 cells that received sgRNAs to knock out KDM4A, KDM4C or GFP (non-targeting control) were suspended in matrigel and transplanted subcutaneously to NSG mice. Tumor volumes were measured over 17 days, with measurements relative to the first day of detectable tumor. Each point represents relative tumor volumes from independent mice.

Figure 5.. Modulation of KDM4 induces cell-intrinsic inflammation via replication stress.

A-B) Gene set enrichment analysis (GSEA) of Hallmark gene sets on JIB-04 treated OCI-Ly1 (A) and TMD8 (B) cells. All terms depicted are statistically significant following correction for multiple hypothesis testing (FDR adjusted P-value < 0.05). Positive NES values reflect gene sets significantly enriched within JIB-04 treated cells, whereas negative NES values reflect gene sets significantly enriched in DMSO treated cells. C) Heatmap of statistically significant (FDR adjusted P-value < 0.05) differentially expressed genes between 500 nM JIB-04 and DMSO treated cells. Each column represents independent biological replicates. log₂(TPM+1) expression values are Z-score standardized along rows. D) GSEA on genes upregulated following ZNF587 knockdown in DLBCL cell lines from Martins, et al. 2024 (PMID: 38345497). Genes were ranked based on descending DESeq2 log₂(fold change) between JIB-04 and DMSO treated cells. Positive NES scores reflect gene set enrichment within 500 nM JIB-04 treated cells compared to DMSO treated cells. GSEA P-values were FDR-corrected for multiple hypothesis testing. E) Representative microscopy of HBL-1 and OCI-Ly1 cells treated with either 0.02% DMSO or 100 nM QC6352 for 48 hours and stained for DAPI and pH2A.X (Ser139). Detection of computationally inferred pH2A.X speckles is visualized in the merged images. Scale bar: 8 μm. F) Distribution of pH2A.X speckle intensity from 0.02% DMSO or 100 nM QC6352 treated OCI-Ly1 and HBL-1 cells. G) Flow cytometry quantification of replication (EdU incorporation intensity) and DNA content (DAPI) in HBL-1 and OCI-Ly1 cells after treatment with 0.02% DMSO or 100 nM QC6352 for 48 hours. Cell cycle phases are annotated within plots. H) Quantification of the proportion of cells in each cell cycle phase from (G) across three biological replicates. Error bars represent standard deviations. Asterisks indicate P < 0.05 based on Student’s t-test for QC6352 versus DMSO. I) Flow cytometry for intracellular pSTING (Ser366) after treatment with DMSO or 100 nM QC6352 for 48 hours. J) ZNF587 immunoprecipitates (IPs) with KDM4A and KDM4C demethylases. Total protein extract from OCI-Ly1 cells (top set of blots, “Input”) was subject to IP with anti-ZNF587 or IgG isotype control followed by immunoblotting (bottom set of blots, “IP”).

Figure 6.. High-throughput small molecule screens identify novel KDM4 inhibitors.

A) Experimental schematic to identify novel KDM4 inhibitors by high-throughput small molecule screening. Left panel: Recombinant human KDM4C enzyme is incubated with biotinylated H3K9me3 histone peptide substrate; followed by sequential addition of anti-H3K9me1 and streptavidin donor / protein A acceptor beads. Excitation (680 nm) of donor beads leads to emission (615 nm) from proximal acceptor beads, providing a quantitative output to what degree all reactants are bridged by H3K9me1 enzyme product. Right panel: we then leveraged this in vitro system to screen libraries of small molecules to identify inhibitors of H3K9me3 demethylation. Small molecule ligands that inhibit the demethylation of recombinant KDM4C would attenuate the AlphaLISA signal relative to DMSO vehicle controls. B) Primary screen to identify small molecule inhibitors of KDM4C enzymatic activity on histone peptide substrates. The ChemBridge 2020 library (50,000 small molecules) was screened in an arrayed manner as in (A). Replicate reproducibility and relative signal of negative (DMSO) and positive (CPI-455) control reactions were used to confirm assay robustness and derive a hit-threshold. Screens were performed in duplicate and small molecules with a Z-score < −3 in both reactions defined as hits. Visualized is the average Z-score across both screening replicates. C) 297 small molecule hits from the primary screen in (B) were rescreened with KDM4C and H3K9me3 histone peptide or semi-synthetic nucleosome substrates in duplicate. Small molecules with a Z-score < −3 in both screening replicates and across both substrate classes were of greatest interest (subject to TruHits filtering). D) 297 small molecule hits from the primary screen in (B) were rescreened with an AlphaLISA TruHits assay to filter false-positive hits that broadly interfere with the format. Candidates with an TruHits value less than two standard deviations of the mean of DMSO controls (transparent points) were considered false-positives and removed from further investigation. E) Venn diagram of small molecule hits across all secondary screens, resulting in 56 high confidence candidates. F) Dose titration of representative high confidence candidates (chemical structure noted above each plot) on H3K9me3 nucleosome substrates. AlphaLISA demethylation signal is normalized to DMSO controls. Error bars represent standard deviations across replicates.

Figure 7.. Molecular docking simulations of small molecule ligands with KDM4C

A) Binding probability profile from AI-Bind for small molecule ligand CB43433036 and human KDM4C (UniProt: Q9H3R0). Amino acid trigrams were perturbed throughout the KDM4C amino acid sequence to determine any influence on the binding prediction. Valleys in the binding profile (marked by dashed boxes) are indicative of putative binding sites. Top: annotation of UniProt protein domains along the KDM4C amino acid sequence. B) Three-dimensional structure of small molecule ligand CB43433036 in complex with the KDM4C Jumonji domain determined through molecular docking simulations and visualized in PyMOL. Dashed arrows indicate molecular interactions between the small molecule ligand and KDM4C amino acids. C) Molecular docking simulations of CB43433036 and the KDM4C PHD domains. Left: AlphaFold3 predicted structure of human KDM4C. Protein domains are colored as in (A). Middle: CB43433036 in complex with the KDM4C PHD domain determined from AutoDock Vina. Right: Visualization of polar molecular contacts between CB43433036 and KDM4C amino acids.