KMT2C/KMT2D-dependent H3K4me1 mediates changes in DNA replication timing and origin activity during a cell fate transition

Abstract

Mammalian genomes replicate in a cell-type-specific order during the S phase, correlated to transcriptional activity, histone modifications, and chromatin structure. The causal relationships between these features and DNA replication timing (RT), especially during cell fate changes, are largely unknown. Using machine learning, we quantify 21 chromatin features predicting local RT and RT changes during differentiation in embryonic stem cells (ESCs). About one-third of the genome shows RT changes during differentiation. Chromatin features accurately predict both steady-state RT and RT changes. Histone H3 lysine 4 monomethylation (H3K4me1), catalyzed by KMT2C and KMT2D (KMT2C/D), emerges as a top predictor. Loss of KMT2C/D or their enzymatic activities impairs RT changes during differentiation. This correlates with local H3K4me1 loss and reduced replication origin firing, while transcription remains largely unaffected. Our findings reveal KMT2C/D-dependent H3K4me1 as a key regulator of RT and replication initiation, a role that likely impacts diseases associated with KMT2C/D mutations.

Keywords: CP: Molecular biology; DNA replication domains; H3K4 monomethylation; KMT2C; KMT2D; MLL3; MLL4; cell cycle; cell fate; chromatin features; epiblast-like cells; epigenetics; formative; genomics; histone modifications; initiation zones; machine learning; naive; pluripotency; predictive modeling; replication origins; replication timing; transcription.

Figure 1.. Genome-wide changes in replication timing during pluripotency transition

(A) Scheme of BioRepli-seq method and analysis. (B) Genome track of log₂ RT displaying early/late RT domains in naive (N) and EpiLC (Epi) states. Gene exons within the genome track are depicted with arrowheads. (C) Heatmap showing genome-wide RT for all 50 kb genomic bins per replicate of naive-to-EpiLC states. Associated chromosomes (chr) are shown. Clustering was performed by K-means. Individual replicates derived from three independent cultures for each cell state are shown. (D) Sankey plot summarizing all significant RT changes (FDR < 0.05) for all 50 kb genomic bins during naive-to-EpiLC transition.

Figure 2.. Chromatin state accurately predicts global RT, with H3K4me1 emerging as a top predictor

(A) Workflow for RT segmentation, machine learning, and feature selection. (B and C) Scatterplots showing predicted versus observed naive RT (B) and ΔRT (C) during transition (difference of EpiLC and naive RT) based on fitted elastic net regression machine learning model. Spearman’s correlation coefficient (rho) is shown. (D and E) Parameter weight for each chromatin feature for naive RT (D) and ΔRT (E) during transition based on fitted elastic net regression machine learning model. A larger magnitude of parameter weight implies more predictive strength of feature. See STAR Methods for details. (F–H) Example genome tracks of regions showing earlier (F), later (G), and constant (H) RT and the associated chromatin feature signals of the top 3 predicting features (see D and E) in naive (N) and EpiLC (Epi) states. Exons are depicted with arrowheads.

Figure 3.. KMT2C/D activity shapes genome-wide RT dynamics during ESC differentiation

(A) Scheme of loss-of-function models. (B) Example genome track of KMT2C/D activity-dependent RT domain showing average RT (derived from three independently grown cultures) of WT, KMT2C^KO, KMT2C/D^dKO, and KMT2C/D^dCD ESCs in naive (N) and EpiLC (Epi) states. (C and D) Heatmap (C) and density plot (D) showing genome-wide ΔRT (difference of EpiLC and naive RT) during cell state transition for control (KMT2C^KO), KMT2C/D^dKO, and KMT2C/D^dCD ESCs. Heatmap is sorted by control ΔRT. (E–G) Volcano plots showing genome-wide ΔRT versus negative log₁₀(FDR) during naive-to-EpiLC transition for WT (E), KMT2C/D^dKO (F), and KMT2C/D^dCD (G) ESCs. The percentage of genome changing is shown for FDR < 0.05 (dashed line). (H) Correlation plot of changes in ΔRT in KMT2C/D^dKO relative to controls compared to KMT2C/D^dCD relative to controls. Spearman’s correlation coefficient (rho) is shown. (I) Heatmap showing differential (EpiLC relative to naive state) H3K4me1, RT, and transcription during transition for control (KMT2C^KO), KMT2C/D^dKO, and KMT2C/D^dCD ESCs for all 50 kb bins stratified into quartiles by changes in peak signal of H3K4me1 in controls during transition. (J) Coefficients of linear regression explaining difference in ΔRT between KMT2C/D^dKO (top) or KMT2C/D^dCD (bottom) relative control (KMT2C^KO) using respective differences in H3K4me1 and transcription.

Figure 4.. KMT2C/D activity locally associates with replication origin firing

(A) Example genome track of KMT2C/D-dependent naive RT domain for KMT2C^KO (Control), KMT2C/D^dKO, and KMT2C/D^dCD highlighting the loss of early peaks within larger later domains. (B) Metagene analysis of naive WT KMT2C/D ChIP-seq (counts per million [cpm] of WT subtracted by KMT2C/D^dKO negative control), naive WT OK-seq replication fork direction (RFD), and naive WT NAIL-seq (RPKM) centered at all KMT2C/D-dependent naive RT domains (changing more than 2-fold on the log₂ in mutants versus control) that also show loss of H3K4me1 (lost in mutants by more than 1-fold on log₂ scale). (C and D) Metagene analysis of H3K4me1 CUT& RUN (C) and OK-seq (D) data at H3K4me1 peaks stratified into quartiles (Q1–Q4) based on their KMT2C/D-dependent H3K4me1 read countchange in control versus mutants (average of changes in KMT2C/D^dKO and KMT2C/D^dCD) during naive-to-EpiLC differentiation. Q1 and Q4 represent sites with the greatest gain and loss of KMT2C/D-dependent H3K4me1, respectively.

Figure 5.. KMT2C/D activity locally promotes replication origin firing

(A) Metagene analysis of OK-seq data from control (KMT2C^KO), KMT2C/D^dKO, and KMT2C/D^dCD cells at KMT2C/D-dependent H3K4me1 peaks in naive state. Peaks were stratified into quartiles (Qs) based on their H3K4me1 read count change in mutants (average of KMT2C/D^dKO and KMT2C/D^dCD) versus control. KMT2C/D-dependent peaks showing more than 2-fold reduction in H3K4me1 in mutants versus control were used as input. (B) Metagene analysis of OK-seq data from control (KMT2C^KO), KMT2C/D^dKO, and KMT2C/D^dCD cells at gained H3K4me1 peaks during differentiation. Peaks were stratified into 8 Qs based on their H3K4me1 read count change in controls during the naive-to-EpiLC differentiation (see Figure S5F for Q1–Q4). Only gained peaks showing more than a 2-fold gain in H3K4me1 in controls were used as input. (C) Metagene analysis of OK-seq data from control (KMT2C^KO), KMT2C/D^dKO, and KMT2C/D^dCD cells at KMT2C/D-dependent H3K4me1 peaks in naive state. Peaks were stratified into Qs based on their RT change in mutants (average of KMT2C/D^dKO and KMT2C/D^dCD) versus control. KMT2C/D-dependent peaks showing more than a 2-fold reduction in H3K4me1 in mutants versus control were used as input. (D) Metagene analysis of OK-seq data from control (KMT2C^KO), KMT2C/D^dKO, and KMT2C/D^dCD cells at gained H3K4me1 peaks during differentiation. Peaks were stratified into Qs based on their difference in RT changes in controls versus mutants (average of KMT2C/D^dKO and KMT2C/D^dCD) during the naive-to-EpiLC differentiation. Only gained peaks showing more than a 2-fold gain in H3K4me1 in controls were used as input.