Rapid Irreversible Transcriptional Reprogramming in Human Stem Cells Accompanied by Discordance between Replication Timing and Chromatin Compartment

Abstract

The temporal order of DNA replication is regulated during development and is highly correlated with gene expression, histone modifications and 3D genome architecture. We tracked changes in replication timing, gene expression, and chromatin conformation capture (Hi-C) A/B compartments over the first two cell cycles during differentiation of human embryonic stem cells to definitive endoderm. Remarkably, transcriptional programs were irreversibly reprogrammed within the first cell cycle and were largely but not universally coordinated with replication timing changes. Moreover, changes in A/B compartment and several histone modifications that normally correlate strongly with replication timing showed weak correlation during the early cell cycles of differentiation but showed increased alignment in later differentiation stages and in terminally differentiated cell lines. Thus, epigenetic cell fate transitions during early differentiation can occur despite dynamic and discordant changes in otherwise highly correlated genomic properties.

Keywords: chromatin 3D architecture; chromatin 3D organization; chromatin structure; differentiation; gene expression; lineage commitment; replication timing.

Graphical abstract

Figure 1

A Stable Transcriptional Reprogramming Occurs within One Cell Cycle (A) Schematic of cell reversal experiment. Cells were stimulated for 0, 6, 12, 24, and 48 h. At each time point half the cells were washed and returned to stem cell medium for 24 h. At the end of each time point, gene expression was profiled by microarray analysis. (B) Heatmap of K-means clustered average gene expression change (with respect to ESC 0 h state) during forward differentiation and reversal for all differentially regulated genes (FDR p < 0.05) during 48 h of differentiation. Exemplary plots on the right quantify change in average log₂ expression with respect to 0 h for forward differentiation (black) and reversal (red) at different time points for the respective K-means cluster. (C) Hierarchical clustering of forward differentiation and reversal time points. K-means clustering and hierarchical clustering show that 6 or 12 h of differentiation + 24 h of reversal has an expression signature more similar to 24 h of forward differentiation. Average of two independent differentiations and reversal experiments is plotted.

Figure 2

Changes in RT Are Detectable within the First Cell Cycle (A) Method to map genome-wide RT. (B) Percentage of 50-kb windows that change RT (change in RT > 0.5) at each time point for independent replicate 1 and 2, and reproducible by RepliPrint (Experimental Procedures). (C) Heatmap of RT in 50-kb windows clustered by K-means at 0, 6, 12, 24, and 48 h. Only replication domains that were reproducible between the two replicates (independent differentiations) were chosen (Experimental Procedures). Line graphs on the right shows average RT values within three exemplary K-means clusters for both replicates. (D) Two exemplary replication domains demonstrating a change in RT between 0 and 6 h of differentiation.

Figure 3

Average Changes in Transcription Are Coincident with RT Changes (A) Line graphs of average transcriptional change (red line, log2 scale) compared with average RT change at the transcription start site (TSS) (blue line) of all genes within exemplary K-means clusters. K-means clusters were defined with transcripts that have either an expression change (FDR p < 0.05 using two independent differentiations) and/or an RT change (dRT > 0.5). The replicate with matched RT data is plotted. The changes were calculated with respect to the ESC 0 h time point. (B) Two exemplary genes illustrating that changes in RT do not cause a corresponding change in transcription. The expression level of the gene is indicated by the red line and the RT of the gene at its TSS is indicated by the blue line. (C) Table quantifying the number of transcripts within categories defined by transcriptional regulation and RT regulation between 0 and 48 h time point. The threshold for transcriptional regulation and replication regulation were FDR p < 0.05 and 0.5, respectively. The values are also expressed as percentages within each row category. (D) Top: exemplary regions (highlighted) harboring a gene that is upregulated during 48 h of DE differentiation (FDR p < 0.05, fold change > 2), but does not exhibit a corresponding change in RT. Green vertical lines mark upregulated genes and gray lines indicate nonregulated genes. Bottom: RT in 48 different cell lines/cell types/differentiation states demonstrate that the RT at this location can be regulated.

Figure 4

Increased Correlation between Chromatin Compartments and RT in Terminally Differentiated Cells Compared to Early Stages of Differentiation (A) Left: boxplot of RT for 50-kb windows that are constitutively early replicating (yellow) or late replicating (blue) in all cell types (Experimental Procedures). Right: boxplot of compartment (PC1) values for the 50-kb bins in top panel. (B) Correlation between RT and compartments (PC1) in undifferentiated ESCs (red), early differentiation intermediates (blue), and terminally differentiated cells (gray). Left panel shows data from individual differentiation states or cell lines (average of replicates). Right panel shows all replicates grouped into undifferentiated, intermediate, or terminal differentiation. Differences were significant with ^∗∗p = 0.004, ^∗∗∗p = 2.595 × 10⁻⁶ (t test). Error bars indicate SEM. (C) Scatterplot showing correlation between changes in RT and changes in PC1 for each differentiation stage (compared with undifferentiated state). R is Pearson’s correlation. MED is mesendoderm. (D) Exemplary plots show discordance between RT and compartments in ESC and early differentiation stages, but becomes concordant in IMR90 (terminally differentiated cell line). For all panels except the individual tracks in (D), plots of samples 24 and 48 h, IMR90, K562, H1ESC, and mesendoderm are average of two independent differentiations or cell collections, IMR90 is average of four independent cell collections.

Figure 5

Changes in Chromatin Marks and RT Correlate More Strongly when Stem Cells Are Compared with More Differentiated Cell Types Heatmap of changes in chromatin modification and chromatin accessibility (from roadmap epigenomics, independent replicates) at regions that change RT (absolute change in RT > 0.5) for ESC versus mesendoderm and ESC versus IMR90. The rows are ordered according to descending order of change in RT. The values on the top of the columns indicate Pearson correlation (rounded) between changes in chromatin feature versus changes in RT. Feature score is the RPKM values binned to 1 kb, normalized with input and binned into 50 kb bins to match the resolution of the replication timing data (see Experimental Procedures).