Bath: a Bayesian approach to analyze epigenetic transitions reveals a dual role of H3K27me3 in chondrogenesis

Abstract

Background: Histone modifications are key epigenetic regulators of cell differentiation and have been intensively studied in many cell types and tissues. Nevertheless, we still lack a thorough understanding of how combinations of histone marks at the same genomic location, so-called chromatin states, are linked to gene expression, and how these states change in the process of differentiation. To receive insight into the epigenetic changes accompanying the differentiation along the chondrogenic lineage we analyzed two publicly available datasets representing (1) the early differentiation stages from embryonic stem cells into chondrogenic cells and (2) the direct differentiation of mature chondrocyte subtypes.

Results: We used ChromHMM to define chromatin states of 6 activating and repressive histone marks for each dataset and tracked the transitions between states that are associated with the progression of differentiation. As differentiation-associated state transitions are likely limited to a reduced set of genes, one challenge of such global analyses is the identification of these rare transitions within the large-scale data. To overcome this problem, we have developed a relativistic approach that quantitatively relates transitions of chromatin states on defined groups of tissue-specific genes to the background. In the early lineage, we found an increased transition rate into activating chromatin states on mesenchymal and chondrogenic genes while mature chondrocytes are mainly enriched in transition between activating states. Interestingly, we also detected a complex extension of the classical bivalent state (H3K4me3/H3K27me3) consisting of several activating promoter marks besides the repressive mark H3K27me3. Within the early lineage, mesenchymal and chondrogenic genes undergo transitions from this state into active promoter states, indicating that the initiation of gene expression utilizes this complex combination of activating and repressive marks. In contrast, at mature differentiation stages the inverse transition, the gain of H3K27me3 on active promoters, seems to be a critical parameter linked to the initiation of gene repression in the course of differentiation.

Conclusions: Our results emphasize the importance of a relative analysis of complex epigenetic data to identify chromatin state transitions associated with cell lineage progression. They further underline the importance of serial analysis of such transitions to uncover the diverse regulatory potential of distinct histone modifications like H3K27me3.

Keywords: Chondrocyte differentiation; Chromatin states; Epigenetic regulation; Histone modification.

Fig. 1

The coverage of histone marks and ChromHMM states are similar in both datasets. A-B Genomic coverage of the epigenetic marks for each cell type and replicate. Points are spread out horizontally by small displacements to reduce overplotting. C-D Emission probabilities of the 15-state ChromHMM model, based on the consensus data of the replicates. The colors of the row names categorize the states into four groups: activating (green), repressed-active (orange), repressive (red), and empty (grey)

Fig. 2

The relative coverage of ChromHMM states reveals systematic differences of epigenetic coverage between cell types. Coverage in base pairs of a given state (row) on a given gene set (column) was plotted against the coverage on all genes. A shows data of the ECL and (B) of the MCL. The black line (slope = [length of all background genes] / [length of all genes within a set], intercept= 0) delineates an equal proportional coverage of the background and the gene set. Deviation from the line indicates enrichment (below the line) or depletion (above the line) of the state in the respective gene set. See Supp. Fig. B8 for all combinations of gene sets and states

Fig. 3

Systematic enrichment of activating marks during lineage establishment. Individual effect sizes for selected transitions of housekeeping, chondrogenic, and mesenchymal genes (CLES color key at the bottom left). Left and bottom axes represent the origin and target states, respectively, with state groups empty (gray), repressive (red), and activating (green). A Exemplary transition of the ECL data on housekeeping genes from the TxGene state to the IntraEnh state. This active-to-active state transition was enriched between all analyzed cell types. The enrichment was particularly strong from ESC into more mature cells and from every cell type into bmCC. B Selection of transition with drastic shifts of the epigenetic pattern, from empty or repressive states to activating states. See Supp.Fig. B12 for all combinations of states, cell types, and gene sets

Fig. 4

The formation and breakdown of the RepActPro state are directly linked to cellular differentiation. Housekeeping genes reveal a consistent pattern in ECL and MCL data, with clear enrichments between the activating promoter states (ActGene and ActProm; top left corners in the individual sections) and depletions for the persistence of the RepActPro state (bottom right corners in the individual sections). For chondrogenic and mesenchymal genes there is an enrichment in the formation of the activating states by losing H3K27me3 in the ECL (RepActPro to ActGene/ActProm; bottom rows in the individual sections). In contrast, in the MCL the formation of the RepActPro state by gaining H3K27me3 is more pronounced (ActGene/ActProm to RepActPro; right column in the individual sections). See Supp.Fig. B12 for all combinations of states, cell types, and gene sets