Fig 1.
Schematic sequences of wild type human linker histone H1.2 (A) and the globular domain (dark grey) of chicken histone H1 (PDB 1GHC [38] B).
The numbers represent aligned lysine residues that have been found to undergo ubiquitylation. [14–16] The red numbers indicate lysines considered in this study that have not been found to be ubiquitylated in nature (K47 in 1GHC and K81 in human H1.2). Lysines within the orange outline are investigated in this study. Abbreviations: NTD (N-terminal domain), GD (globular domain), CTD (C-terminal domain).
Table 1.
Five chromatosome structures used as reference in this study: PDBs 4QLC, 5NL0, and 5WCU (rows 1–3) exhibit an on-dyad binding mode of the linker histone.
The reference structure of the off-dyad binding mode (row 5) was obtained via HADDOCK docking using a nucleosome from 1ZBB and PDB 1HST as the histone. [10, 50]. PDB 1ZBB is used as a reference for a an array of multiple nucleosomes. The last row represents the present study. We use PDB 1GHC as the linker histone (also see Fig 1 for a sequence alignment of the linker histones).
Fig 2.
Illustration of data-generation (top half) and analysis (lower half) workflow to identify possible structures of ubiquitylated linker histone in a chromatosome.
Data was produced by MD simulations driven by an expansion scheme which selects new starting structures from sparsely populated regions of phase space. The aggregated MD data is projected into a combined low-dimensional space and clustered iteratively. Clusters (i.e. characteristic conformational states) are then fitted into different chromatosome structures using a new geometric scoring method.
Fig 3.
Sketch-map projections of all conducted simulations colored by various parameters and exemplary structures.
A contour plot with transparent patches colored according to ubiquitylation site (A) shows that the most distinct feature of the six HUb variants is the ubiquitylation site. A density map of the combined projections (C) shows that the low-dimensional representation is mostly shallow (blue) with only small regions of higher density (red). Coloring the combined projection according to the starting χ3 angles ((D) and (E)) shows a finer structure of the projection. Four exemplary structures from (E) have been chosen and their RMSD centroids are visualized with cartoon representation and colored according to secondary structure features (B). The blue transparent sphere is focused on the histone subunit, the grey one on the Ub subunit. Closeness in sketch-map space is loosely related to structural similarity, due to sketch-map’s non-linear dimensionality reduction. In (A) regions of high density and high RMSD variance were excluded, to not bias the density towards these regions. Subfigure (C) was rendered after removing the high-density, high-RMSD variance region (Sec. S2.3, S2 Text, S5 Fig).
Fig 4.
Applying an iterative HDBSCAN clustering, a total of 496 clusters were obtained from the complete sketch-map projection.
A selection of a few clusters from a section of the joint projection (compare Fig 3A, 3C and 3D for overview) is visualized as colored patches, whereas the outliers are grey and slightly transparent (A). Structure renders of these clusters (blue sphere centered on H1 subunit, grey sphere centered on Ub subunit) show that a satisfying degree of structural cohesion could be achieved. Subfigures (B) and (C) show the sample density and the dominant ubiquitylation site of the region in (A), respectively.
Fig 5.
Results of the interpenetration and scoring algorithm and example structures for scoring K30HUb into 5NL0.
The sketch-map projection (A) colored according to ISA score exhibits distinct regions with fitting (green, low score) and non-fitting (red, high score) conformations. The histogram of scores has a skew to lower scores (B). In (C) and (D) exemplary HUb clusters are rendered within the nucleosome of 5NL0, their location in sketch-space is annotated in (A). Both (C) and (D) visualize the DNA as tubes and the core histones as a transparent blue surface. The HUb structure bundles are rendered using their secondary structure and colored accordingly. In (C) the K30HUb cluster with the lowest ISA score is shown. Using this cluster the HUb chromatosome complex could easily be used as input for further simulations. In (D) a cluster with an ISA score of 150 is shown. This cluster can be taken as an example of a high-score cluster, that does only slightly intersects with the DNA linkers.
Fig 6.
Scores and renders of the six HUb variants in the four chromatosome structures.
In (A) the sketch-map projection is colored according to the ISA score where a specific region exhibits high scores and contains non-fitting HUb conformations. The distribution of scores can be seen in (B), (C), and (E). The clusters with the lowest scores of the parent chromatosomes 5NL0, 5WCU, 4QLC, and the off-dyad structure are shown in (D, F-H), their location annotated in (A). Coloration of renders is in line with Fig 5. Note, that the lowest scoring cluster for 5WCU (F) and 4QLC (G) is the same cluster. Both chromatosomes are structurally very similar, missing the DNA linkers that are present in 5NL0. Also note, how the Ub subunit of the lowest scoring cluster fitted into the off-dyad chromatosome points into a different direction. The off-dyad structure in (H) contains the tail domains of the core histones. In (I), the crystal structure of 1UBQ was colored according to the average per-residue minimal distance to DNA after placing HUb into 5NL0. The possible values range from red (closest) to blue (farthest).
Fig 7.
Using ISA to gauge the positions of HUb in a tetranucleosome array.
The median ISA score over the 8 possible HUb positions (4 nucleosomes in the array using on-dyad and off-dyad binding motifs) exhibit large regions of higher ISA scores (A). In (B-E) the distribution of ISA scores is visualized for all 8 positions combined (B), differentiated by ubiquitylation site (C), and divided by the 4 off-dyad and 4 on-dyad HUb positions (E) and (D), respectively. The render in (F) uses the on-dyad 5NL0 chromatosome as reference structure for the linker histone positions 1 and 4 and the off-dyad reference structure for the positions 2 and 3. The geometric cluster centroids are annotated in (A) accordingly. They exhibit mean ISA scores between 12 and 86. These clusters are chosen for illustrative purposes to exhibit how certain sub-states in the conformational landscape of all HUb variants can fit into the more restrictive tetranucleosome array. For the four positions 1, 2, 3, and 4, the chosen clusters contain predominantly the linkage types K51, K30, K51, and K41, respectively. The DNA is rendered as light blue tubes, core histones are blue translucent. Secondary structure elements are colored as default in VMD. [56]
Fig 8.
Refinement of non-uniform clusters via iterative HDBSCAN.
An initial run of HDBSCAN combines all points in (A) into a single cluster. Rendering this cluster would result in a non-uniform bundle of structures (D). Furthermore, the spatial distribution in the low-dimensional space (B) and the distribution of RMSD values (C) indicate, that there are multiple sub-clusters that could be resolved by subsequent clustering. Using a lower minimal cluster size parameter the larger cluster is decomposed into smaller clusters (1–4) and the renderings of these regions (E 1–4) also show greater uniformity. Renders show the histone subunit (blue translucent sphere) on the right side and the ubiquitin subunit (grey translucent sphere) on the left side.
Fig 9.
Comparison of scores obtained by our interpenetration and scoring algorithm (ISA) and ROSETTA for a subset of structures.
We chose a random subset of structures from the expansion scheme simulations to score using ROSETTA and ISA. Rosetta scores range from −250 to ≈ 18 000 Rosetta Energy Units (REU). ROSETTA’s scoring functions with default weights support our claim that many structures exhibit lower scores, as the most structures can be found for lower scores where a high bin count can be observed (A). Our ISA algorithm correlates with the much more sophisticated ROSETTA algorithm with a Pearson correlation coefficient of 0.88 (B). However, our algorithm was 1000 times faster than pyROSETTA (both algorithms have been parallelized on a per-structure basis using the Python package joblib [69]).