MLL oncoprotein levels influence leukemia lineage identities

Abstract

Chromosomal translocations involving the mixed-lineage leukemia (MLL) locus generate potent oncogenic fusion proteins (oncoproteins) that disrupt regulation of developmental gene expression. By profiling the oncoprotein-target sites of 36 broadly representative MLL-rearranged leukemia samples, including three samples that underwent a lymphoid-to-myeloid lineage-switching event in response to therapy, we find the genomic enrichment of the oncoprotein is highly variable between samples and subject to dynamic regulation. At high levels of expression, the oncoproteins preferentially activate either an acute lymphoblastic leukemia (ALL) program, enriched for pro-B-cell genes, or an acute myeloid leukemia (AML) program, enriched for hematopoietic-stem-cell genes. The fusion-partner-specific-binding patterns over these gene sets are highly correlated with the prevalence of each mutation in ALL versus AML. In lineage-switching samples the oncoprotein levels are reduced and the oncoproteins preferentially activate granulocyte-monocyte progenitor (GMP) genes. In a sample that lineage switched during treatment with the menin inhibitor revumenib, the oncoprotein and menin are reduced to undetectable levels, but ENL, a transcriptional cofactor of the oncoprotein, persists on numerous oncoprotein-target loci, including genes in the GMP-like lineage-switching program. We propose MLL oncoproteins promote lineage-switching events through dynamic chromatin binding at lineage-specific target genes, and may support resistance to menin inhibitors through similar changes in chromatin occupancy.

Fig. 1. Genomic enrichment of MLL oncoproteins is highly heterogeneous.

a We performed AutoCUT&RUN on a collection of MLLr cell lines and patient samples including a range of patient ages and lineage subtypes. b Comparison of CUT&RUN profiles using antibodies targeting the MLL N-terminal (red) and C-terminal (blue) regions can discriminate the oncoprotein-binding sites from wild-type MLL binding sites. c When the same magnitude and significance thresholds are applied to the MLL N versus C terminal profiles, we identify many more oncoprotein-target sites (red dots) in the SEM MLLr cell line than the CD34+ wild type control sample. p values were computed from the mean and standard deviation of Monte Carlo sampling of the N/C scores (n = 1000 samples) of each genomic interval; p values were corrected for multiple hypothesis testing using the Benjamini/Hochberg method. d Principal component analysis and UMAP embedding splits the MLLr samples into four clusters that are outlined with dotted lines; samples are colored according to the lineage subtype; patient-matched lineage-switching samples are indicated by the Greek letters α, β, γ. e The UMAP embedding from (d) but colored according to the MLL-oncoprotein-fusion partner. f Same as (d) colored according to the average oncoprotein scores (N/C terminal magnitude x significance). g The UMAP clusters capture differences in the global oncoprotein levels. p values were computed using a two-tailed independent samples t-test; Cluster 1 n = 13 samples, Cluster 2 n = 6 samples, Cluster 3 n = 15 samples, Cluster 4 n = 5 samples; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). h The average oncoprotein scores scale with the relative oncogene expression. Samples are grouped and colored according to the minimal MLL-fusion-partner exon junctions. Two MLL::AFDN samples were identified as outliers: ML-2 is indicated by a star and A70498 by an asterisk. Dotted lines indicate the regression, error bars indicate the standard deviation of the mean of three qPCR biological replicates. Source data are provided as a Source Data file.

Fig. 2. MLL oncoprotein levels scale with target gene expression.

a Genome Browser tracks show the difference in MLL N-terminal (red) and C-terminal (blue) signals in three MLL::AF4 bearing leukemias, over the most common oncoprotein-target locus: MBNL1 (bottom black bar = oncoprotein-target peak). The USP5 promoter is an example of a wild-type MLL target site (bottom black bar = wild-type MLL peak). AF4 (brown) DOT1L (cyan) ENL (magenta) and H3K4me3 (green) signal is also shown for the p454 sample. b Oncoprotein scores over MBNL1 (quantified over the bottom-black bar in (a)) are representative of the average oncoprotein scores in each UMAP cluster; p values were computed using a two-tailed independent samples t-test; Cluster 1 n = 13 samples, Cluster 2 n = 6 samples, Cluster 3 n = 15 samples, Cluster 4 n = 5 samples; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). c The MBNL1 oncoprotein scores are correlated with the relative expression of MBNL1 as measured by qPCR; dotted lines indicate the regression and R² measures the fit to the data, error bars indicate the standard deviation of the mean of three qPCR biological replicates. d ENL is significantly enriched over the oncoprotein-target sites. p values were computed using a two-tailed independent samples t-test; n values are listed as wild-type site #, oncoprotein site #: SEM = 15611, 491; p318 = 22014, 246; p279 = 25672, 272; p454 = 18509, 438; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). Source data are provided as a Source Data file.

Fig. 3. At elevated levels the MLL oncoproteins are instructive of lineage-specific genomic binding sites.

a Principal Components 1 and 2 organize MLLr leukemia samples according to the oncoprotein-fusion partners. Dotted regression lines are included for samples bearing related translocations. Pie charts indicate the frequency of each MLL-fusion partner found in ALL versus AML. b The slope of the MLL-oncoprotein regression lines from (a) are highly correlated with the frequency that each fusion partner is found in AML versus ALL. c The top 5% of sites (indicated by the red line) that contribute most to Principal Component 1 were identified by rank ordering the 1oading values. d same as (c) but for Principal Component 2. e UMAP from Otto et al. ref. of single-cell RNA-seq data from 7439 cells from lineage-depleted bone marrow; HSC/MPP = Hematopoietic Stem Cell/Multipotent Progenitor, MEP = Megakaryocyte Erythroid Progenitor, pDC = pre-Dendritic Cell, proB = pro-B Cell, LMPP = Lymphoid-Myeloid-Primed Progenitor, GMP = Granulocyte Monocyte Progenitor. f UMAP embedding from (e) colored by the average z-score of genes from PC1. g UMAP embedding from (e) colored by the average z-score of genes from PC2.

Fig. 4. MLL oncoproteins bind to lineage-specific and fusion-partner specific genomic sites.

a Genome Browser tracks show the MLL::AF4 oncoprotein binds the FLT3 locus while the MLL::AFDN oncoprotein does not (bottom black bar = oncoprotein-target peak). b Scatterplot showing the FLT3-oncoprotein scores (quantified over the bottom black bar in (a)) relative to the average oncoprotein score in each sample. Samples are colored by the MLL-fusion partner and dotted-lines indicate the regression. c Same as (b) but showing TAPT1 oncoprotein scores. d The MLL::ENL oncoprotein scores on FLT3 are correlated with the relative expression of FLT3 as measured by qPCR, while MLL::AF4 and MLL::AFDN levels are not correlated with expression of FLT3; dotted-lines indicate the regression and R² measures the fit to the data, error bars indicate the standard deviation of the mean of three qPCR biological replicates. e Same as (a) showing the ZNF521 locus. f Same as (c) but showing ZNF521 oncoprotein scores. g The MLL::AFDN oncoprotein scores on ZNF521 are correlated with the relative expression of ZNF521 as measured by qPCR. Only one MLL::ENL sample expresses ZNF521. h Same as (a) showing the HOXA9 locus, and a different MLL::AF4-rearranged sample in which HOXA9 is not called as an oncoprotein target gene. i Same as (b) but showing HOXA9 oncoprotein scores. j HOXA9 and IRX1 show mutually exclusive expression in MLL::AF4 samples, except for the p279 sample that went on to lineage switch and give rise to the p318 AML sample. The Ct values from three qPCR biological replicates are shown. k The MLL-fusion partners direct the genomic occupancy of the oncoproteins to preferentially activate either lymphoid or myeloid lineage programs. Source data are provided as a Source Data file.

Fig. 5. The oncoprotein-binding sites change during B-ALL-to-AML lineage-switching events.

a Boxplot comparing the oncoprotein scores in a patient matched B-ALL and AML sample that lineage switched in response to B-ALL directed therapy. The p value was computed using a two-tailed independent samples t-test; n = 1692 oncoprotein-target sites; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). b Oncogene expression is significantly reduced in the AML after lineage switching. Bar height is the average of three qPCR biological replicates; Error bars = standard deviation, and the p value was computed using a two-tailed independent samples t-test; n = 3 qPCR biological replicates. c Genome Browser tracks showing the oncoprotein levels are elevated over the BANK1 locus in the B-ALL sample prior to lineage switching. Bottom black bars indicate B-ALL specific oncoprotein target regions. d B-ALL specific oncoprotein target genes are expressed at significantly higher levels in the B-ALL sample (gold) prior to lineage switching than the AML sample (pink). Bar height is the average of three qPCR biological replicates; Error bars = standard deviation; p values computed using a two-tailed independent samples t-test; n = 3 qPCR biological replicates. e Genome Browser tracks showing the oncoprotein levels are elevated over the FNDC3B locus in the AML sample after lineage switching. Bottom black bars indicate AML specific oncoprotein target regions. f Same as (d) but for AML-specific oncoprotein target genes. g A Venn Diagram comparing the lineage-switching B-ALL-specific and AML-specific oncoprotein-target genes to all other oncoprotein-target genes. The AML group includes 51 non-overlapping genes referred to as the GMP-like program. h The UMAP embedding of healthy lineage-depleted bone marrow from Otto et al. ref. , colored by the average z-scores of the 51 non-overlapping AML-specific oncoprotein target genes from (g). Source data are provided as a Source Data file.

Fig. 6. MLL oncoprotein dynamics influence target gene expression during B-ALL-to-AML lineage-switching events.

a Boxplot comparing the relative expression of the B-ALL-specific oncoprotein target genes in a cohort of 6 B-ALL (gold) and AML (pink) patient-matched samples that underwent a lineage-switching event as described in Tirtakusuma et al. ref. . p values were calculated using a one-sided Wilcoxon sign-rank test, n = 66 genes; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). b Same as (a) but comparing the relative expression of the 51 genes in the GMP-like program, n = 51; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). c At high levels the MLL oncoprotein reinforces the B-ALL identify by activating a lymphoid program. During lineage switching the oncoprotein levels are reduced, and the oncoprotein activates a GMP-like program.

Fig. 7. ENL remains bound to MLL oncoprotein target genes during treatment with the menin inhibitors.

a Genome Browser tracks showing MLL, menin, DOT1L and ENL bound to the MBNL1 locus in the B-ALL sample (top). In the patient-matched AML sample treated with the menin inhibitor only ENL remains bound (bottom). b Oncogene expression is significantly reduced in the AML after lineage switching. Bar height is the average of three qPCR biological replicates; Error bars = standard deviation, and the p value was computed using a two-tailed independent samples t-test; n = 3 qPCR biological replicates. c Reverse transcriptase PCR with two primer sets confirms the MLL::ENL oncogene is expressed in the AML treated with the menin inhibitor (expected size for 1 = 159 bp, and 2 = 150 bp). d Scatterplot comparing the normalized menin levels in the B-ALL and the AML sample treated with the menin inhibitor over an internal control group of oncoprotein-target sites (gray) the B-ALL-specific program (gold) and the GMP-like program (pink). e Same as (d) but showing the ENL signal persists over many of the oncoprotein-target genes in the AML sample. f Boxplot comparing ENL signal over an internal control group of oncoprotein-target genes (gray, n = 1595) the B-ALL-specific program (gold, n = 52) and the GMP-like program (pink, n = 45). p values were computed using a two-sided Mann-Whitney U Test with a Bonferroni correction for multiple hypothesis testing; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). g The menin inhibitor disrupts MLL oncoproteins, but ENL accumulation persists on oncoprotein targets in the GMP-like program. Source data are provided as a Source Data file.

Fig. 8. KMT2C and GMP gene expression changes during menin inhibitor treatment.

a Expression of the KMT2C gene is significantly reduced in the lineage-switching sample treated with the menin inhibitor. Bar height is the average of three qPCR biological replicates; Error bars = standard deviation; p value computed using a two-tailed independent samples t-test; n = 3 qPCR biological replicates. b The menin-inhibitor-resistant AML patient P9 sample has significantly higher expression of the GMP-like program than two menin-inhibitor responder samples and one additional non responder profiled in Perner et al. 2023 (ref. ). p values were computed using the two-sample t-test for independent samples. For comparisons of samples classified as non-responders to responders we used a one-sided test, otherwise we used a two-sided test. n is the number of GMP-like genes that had a significant fold change between the menin-inhibitor treated and untreated samples (padj. < 0.05): MV4;11 = 30, PDX3 = 28, PDX4 = 32, P9 = 39; boxplot center lines = median, box limits = first and third quartiles, whiskers = 1.5 times the interquartile range (IQR). Source data are provided as a Source Data file.