Mutant EZH2 alters the epigenetic network and increases epigenetic heterogeneity in B cell lymphoma

Abstract

Diffuse large B cell lymphomas and follicular lymphomas show recurrent mutations in epigenetic regulators; among these are loss-of-function mutations in KMT2D and gain-of-function mutations in EZH2. To systematically explore the effects of these mutations on the wiring of the epigenetic network, we applied a single-cell approach to probe a wide array of histone modifications. We show that mutant-EZH2 elicits extensive effects on the epigenome of lymphomas, beyond alterations to H3K27 methylations, and is epistatic over KMT2D mutations. Utilizing the single-cell data, we present computational methods to measure epigenetic heterogeneity. We identify an unexpected characteristic of mutant-EZH2, but not KMT2D, in increasing heterogeneity, shedding light on a novel oncogenic mechanism mediated by this mutation. Finally, we present tools to reconstruct known interactions within the epigenetic network, as well as reveal potential novel cross talk between various modifications, supported by functional perturbations. Our work highlights novel roles for mutant-EZH2 in lymphomagenesis and establishes new concepts for measuring epigenetic heterogeneity and intra-chromatin connectivity in cancer cells.

Fig 1. Mutant-EZH2 elicits a wide array of epigenetic alterations and shows an epistatic phenotype over KMT2D mutations.

A. Scheme of the CyTOF experimental system. Human lymphoma patient-derived xenografts, or cell lines, unperturbed or perturbed genetically or chemically, were stained with an epigenetic-oriented metal-conjugated antibody panel (see also S1 Table for list of antibodies and metals). From this single cell data, we mined complex epigenetic phenotypes exhibited by these cells. Drawn with Biorender. B. CyTOF analysis of the lymphoma patient-derived cell lines OCI-Ly7, carrying WT copies of EZH2 and KMT2D, and OCI-Ly1, carrying the EZH2 GOF mutation and biallelic loss of KMT2D. Colors indicate the sample index. C. Scaled, normalized levels of the indicated histone modifications on the UMAP of the cell lines that is shown in B. D. Expression levels of the indicated modifications in OCI-Ly1 and OCI-Ly7 cell lines, as measured by the CyTOF. P values were calculated by Welch’s t test. ***p value < 0.001. E–F. OCI-Ly7 cells (WT), as well as their isogenic counterparts carrying mutant-EZH2 (EZH2 Y646N), biallelic knockout of KMT2D (KMT2D^−/−), or a combination of both mutations in EZH2 and KMT2D (dMUT), were analyzed by CyTOF. UMAP was performed based on all epigenetic marks measured in two independent biological repeats, following scaling and normalization. E. plotting of the indicated mutant cells on the joined UMAP in each of the repeats. F. Each line is shown separately. EZH2-mutant cells (alone or in combination with KMT2D loss) cluster together and separately from the WT cells, indicating an epistatic epigenetic phenotype for EZH2 GOF mutation. G. Heatmap showing the differences in the means of the indicated histone modifications in each of the indicated mutant lines compared to the WT OCI-Ly7 cells. Two biological CyTOF replicates are shown. Red: higher in the mutant cells compared to WT. Blue: lower in mutant cells. H. RNA-sequencing analysis of OCI-Ly7 cells and the indicated isogenic mutant lines. Heatmap shows the rld values of differentially expressed genes between the indicated mutant lines versus OCI-Ly7. Rows are standardized and clustered; red and blue denotes high and low levels of the modifications, respectively. The list of differentially expressed genes was obtained using DESeq2 and filtered for >1 or <−1 LogFC and p.adj ≤ 0.05. K-means clustering = 2. The data underlying this figure can be found in Raw data 1–2 at 10.17605/OSF.IO/NTGUX, under CyTOF and MARS-seq folders.

Fig 2. EZH2 Y646N mutation contributes to higher epigenetic heterogeneity.

A. OCI-Ly7 cells and the indicated isogenic mutants (KMT2D^−/−, EZH2 Y646N, and the double mutant ‘dMUT’) were analyzed by CyTOF. Joint UMAP was performed for the WT OCI-Ly7 cells versus each of the three mutants, based on all epigenetic marks measured, following scaling and normalization. Colors indicate the sample index. B. Illustration of the various measurements of epigenetic heterogeneity. Shown is a hypothetical UMAP of two populations, marked in red and blue. The global heterogeneity of each population is reflected by: (i) Global area, as measured by the area enclosed by the bounding convex hull of all points in the distribution, and (ii) Global distance, measured by the distance between each two cells in the population, defined by the 95th quantile of the distribution of distances between any two cells. The local heterogeneity (iii) is measured by the net area of the clusters. If clusters are more dispersed, they would occupy a larger area. (iv) Nearest-neighbor distance (NND), which refers to the distance between each point and its nearest neighbor. For a population that is more homogenous, the distribution function for the NND will peak at shorter distances. C. The indicated heterogeneity measurements that were defined in panel C were calculated for OCI-Ly7 cells versus each of the mutant lines, across a wide range of UMAP parameters. Plotted are the ratios of the values of each mutant line versus OCI-Ly7 (WT). Thus, values over 1 indicate excess heterogeneity in the mutant over that of the WT. EZH2 mutant cells showed higher heterogeneity in all measurements. D. Ratio of the cumulative distribution functions (CDFs) of the NND values in the indicated mutant lines versus the WT cells. Values below 1 indicate higher heterogeneity of EZH2-mutant cells. E. Heterogeneity measurements calculated as in panel C for OCI-Ly7 cells carrying WT EZH2 versus OCI-Ly1 cells expressing the mutant form. Overall, OCI-Ly1 cells carrying EZH2 GOF show higher heterogeneity. F. Transcriptional heterogeneity of GCB cells from three mice expressing WT EZH2, versus two mice expressing the EZH2 Y646F GOF mutation. Sc-RNA-seq data is from Béguelin and colleagues [34] Heterogeneity was measured as the distribution of the mean distance between cells in each genotype (EZH2 GOF and WT), each color indicates a different mouse. To calculate the distribution, 300 cells were sampled from each sample and the mean Euclidean distance (based on the gene expression matrix) was calculated. The sampling was repeated 200 times for each sample. P value = 9e−132, as calculated by Welch’s t test between the two distributions (from the EZH2 GOF and the WT mice). G. Transcriptional heterogeneity for two WT EZH2 and two mutant-EZH2 human FL tumors, showing increased heterogeneity in cells carrying EZH2 mutation. Heterogeneity was measured by the Euclidean pairwise distances, using the gene expression matrix, between each pair of cells in the sample. EZH2 WT and mutant tumors were matched according to grade (grades 2–3). All tumors selected for this analysis contain WT KMT2D and CREBBP. Sc-RNA-seq data is from Han and colleagues [48] P value was calculated by Welch’s t test as in F. The data underlying this figure can be found in Raw data 1 at 10.17605/OSF.IO/NTGUX, under CyTOF folder.

Fig 3. EZH2-mutants exhibit two distinct epigenetic populations.

A. Top: Joint UMAP of OCI-Ly7 WT cells and the isogenic EZH2 Y646N cells (Left) or dMUT cells (Right). Colors indicate the sample index. Bottom: Only mutant-EZH2 cells are plotted, to highlight the subpopulation of mutant cells that cluster with WT cells, referred to as ‘WT-like’. The subpopulation of cells that cluster separately from WT is referred to as EZH2 Y646N ‘extreme’ or dMUT ‘extreme’, respectively. B. Heatmap showing standardized mean values measured by CyTOF for the indicated histone modifications in OCI-Ly7 cells expressing WT EZH2 (‘WT’), and in the two populations identified in the EZH2 Y646N cells (marked in panel F), which either manifest robustly the GOF phenotype and form a distinct cluster (EZH2 Y646N ‘Extreme’), or clusters with OCI-Ly7 WT cells (EZH2 Y646N ‘WT-like’). C. Heatmap showing the differences in the means of the indicated histone modifications of the ‘extreme’ population compared to the ‘WT-like’ sub-population’. Shown are three biological CyTOF replicates for the single EZH2 Y646N mutant line, and two biological replicates for the double mutant. ‘WT-like’ cells show higher levels of the heterochromatin marks H3K9me2 and H3K9me3, and reduced levels of histone acetylations. D. Histogram of scaled and normalized H3K27me3 levels, as measured by CyTOF, in OCI-Ly7 cells (‘WT’) or the indicated isogenic mutant lines. E. Histogram of scaled and normalized H3K27me3 levels, as measured by CyTOF, in the PDX^MUT from mice treated with EZH2 inhibitor for two weeks or with vehicle as control. The two H3K27me3 subpopulations are present in both control and treated mice. EZH2 inhibition shifted the fraction of cells in each population and reduced H3K27me3 levels in the ‘high’ population. F. OCI-Ly1 cells expressing mutant-EZH2 were treated with EZH2 inhibitor at a concentration of 10 µM for one week to completely deplete H3K27me3 (see S4K Fig). Next, the inhibitor was washed, and the cells were allowed to recover for one or two weeks. Shown is a histogram of H3K27me3 levels in untreated OCI-Ly1 cells, as well as the cells after one or two weeks of recovery. The two subpopulations of H3K27me3 are maintained throughout the experiment, yet the fraction of cells in each subpopulation shifts over time. G. EZH2 Y646N were sorted into single cells and allowed to proliferate to generate clones. Histogram of scaled and normalized H3K27me3 levels, as measured by CyTOF, in OCI-Ly7 cells (WT) or the indicated EZH2 Y646N single-cell-derived clones. The percentage of ‘WT-like’ cells is shown per clone. H. Bar plot depicting percentages of ‘WT-like’ cells in 16 EZH2 Y646N clones as in G. The ‘WT-like’ percentage was calculated based on H3K27me3 values of EZH2 Y646N overlapping with OCI-Ly7 H3K27me3 values. I. Heatmap showing the differences in the means of the indicated histone modifications in each of the indicated clones compared to the WT OCI-Ly7 cells. The data underlying this figure can be found in Raw data 1 at 10.17605/OSF.IO/NTGUX, under CyTOF folder.

Fig 4. Single-cell analysis uncovers potential interactions within the epigenetic network.

A–C. Models to decipher interactions within the epigenetic network, applied to unperturbed OCI-Ly7 single-cell CyTOF data. A. XGBoost analysis. B. ‘Probability of link’ analysis. For both A and B, the Y axis indicates ‘source’ modification and X axis indicates its ‘target’ modification affected by the Y axis. C. Partial correlations between histone modifications. Yellow blocks and arrows highlight connections that are discussed in the text. D. Left – Heatmap showing standardized mean values measured by CyTOF, for the indicated histone modifications, of unperturbed OCI-Ly7 cells compared to each of the following perturbations: EZH2 inhibition (‘EZH2i’), over expression of WT EZH2 (‘EZH2 OE’), or expression of EZH2 GOF (‘EZH2 Y646N’). For EZH2i, cells were incubated with DMSO (control) or 10 µM of the EZH2 inhibitor Tazemetostat for 48 h. For EZH2 OE, cells were nucleofected with PBS (control) or 1 µg of plasmid expressing WT EZH2. For EZH2 Y646N, cells underwent mutagenesis for EZH2 using CRISPR/Cas9. Right – heatmap scoring the effect of the perturbation by showing the mean of the absolute change per each mark. Effect shown in descending order from high (orange) to low (white). E. Average coverage of Cut&Run reads of the indicated histone modifications, over H3K27me2 and H3K27me3 peaks, in OCI-Ly7 cells and the isogenic EZH2-mutant cells (EZH2 Y646N). F–G. XGBoost analysis done on CyTOF data of unperturbed germinal-center B cells (CD20⁺, CD38⁺), derived from tonsils, for two independent samples derived from different patients. Tonsils were dissociated to single cells followed by staining with the panel of metal-conjugated antibodies and CyTOF analysis. Yellow blocks and arrows highlight connections that are discussed in the text. The data underlying this figure can be found in Raw data 1, 3 at 10.17605/OSF.IO/NTGUX, under CyTOF and Cut and Run folders.

Fig 5. Single-cell analysis reveals non-linear relationship between H3K27me3 and BCL6.

A. Histogram of BCL6 levels in OCI-Ly7 WT, EZH2 Y646N cells, and the double mutant of EZH2 Y646N and KMT2D^−/−, as measured by CyTOF. B. Normalized coverage values of H3K27me3 on the promoter of BCL6 gene, as measured by Cut&Run, in OCI-Ly7 and EZH2 Y646N. Chr3:187742009-187750134. C–D. BCL6 mean levels in cells binned according to H3K27me3 levels. Red hue represents standard error of the mean. Number of cells in each bin is shown. C. OCI-Ly7 cells with WT EZH2. D. OCI-Ly7 cells with mutant-EZH2 (isogenic cell line). For this analysis measurements from four CyTOF experiments were pooled. H3K27me3 and BCL6 show a non-linear relationship. E. Model for the non-linear ‘bell-shaped’ relationship between H3K27me3 and BCL6. For low H3K27me3 levels, as in most OCI-Ly7 cells, it is positively correlated with BCL6. For high H3K27me3 levels, as in most EZH2-mutant cells, it is negatively correlated with BCL6. Thus, inhibiting EZH2 in the context of WT or mutant-EZH2 generates opposite effects on BCL6 levels. Generated with Biorender.com. F. Expression levels of H3K27me3 and BCL6, as measured by CyTOF, in the isogenic OCI-Ly7 WT and EZH2-mutant cells treated with EZH2 inhibitor for 48 h at a concentration of 10 µM. In OCI-Ly7 with low H3K27me3 levels, EZH2 inhibition downregulated BCL6. In the same cells expressing mutant-EZH2 and high H3K27me3, EZH2 inhibition led to upregulation of BCL6, validating a non-linear relationship. P values were calculated by Welch’s t test. ***p value < 0.001. G. Heatmap showing mean values of OCI-Ly7 control and BCL6 overexpressing cells, for the indicated histone modifications. BCL6 overexpression results in an increase in H3K27me3 as well as additional marks of heterochromatin, and a decrease in histone acetylation levels. H. H3K27me3 mean levels in cells binned according to BCL6 levels. For this analysis measurements from four CyTOF experiments were pooled. Red hue represents standard error of the mean. Number of cells in each bin is shown. I–J. Quantitative RT-PCR analysis of EHZ2 expression. ΔΔC_T values relative to OCI-Ly7 ± s.d (n = 3) are shown. I. OCI-Ly7 cells were treated with 50 µM of the BCL6 inhibitor (BCL6i) FX1 for the indicated time points. J. OCI-Ly7 control and BCL6 over expression. Both treatments show an increase in EZH2 expression. The data underlying this figure can be found in Raw data 1, 6 at 10.17605/OSF.IO/NTGUX, under CyTOF and RT-PCR folders.