Epigenetic regulation is mediated by protein complexes that couple recognition of chromatin marks to activity or recruitment of chromatin-modifying enzymes. Polycomb repressive complex 2 (PRC2), a gene silencer that methylates lysine 27 of histone H3, is stimulated upon recognition of its own catalytic product and has been shown to be more active on dinucleosomes than H3 tails or single nucleosomes. These properties probably facilitate local H3K27me2/3 spreading, causing heterochromatin formation and gene repression. Here, cryo-EM reconstructions of human PRC2 bound to bifunctional dinucleosomes show how a single PRC2, via interactions with nucleosomal DNA, positions the H3 tails of the activating and substrate nucleosome to interact with the EED subunit and the SET domain of EZH2, respectively. We show how the geometry of the PRC2–DNA interactions allows PRC2 to accommodate varying lengths of the linker DNA between nucleosomes. Our structures illustrate how an epigenetic regulator engages with a complex chromatin substrate.
Access optionsAccess options
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Subscribe to Journal
Get full journal access for 1 year
only $18.75 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank P. Grob and J. Fang for technical support, T. Houweling and A. Chintangal for computer support, and D. King for MS confirmation of histone alkylation. We are thankful to R.M. Glaeser for his support and helpful discussions and to S. M. Sterling, A. Patel, B. LaFrance, D. Lipscomb, R. K. Louder, T. H. D. Nguyen and B. J. Greber for helpful discussions and advice during data acquisition and processing. We also thank T. H. D. Nguyen and B. J. Greber for critical comments on the manuscript. Computational resources were provided in part by the National Energy Research Scientific Computing Center (DE-AC02-05CH11231) and LAWRENCIUM computing cluster at Lawrence Berkeley National Laboratory. This work was in part funded by Eli Lilly through the Lilly Research Award Program. S.P. was supported by the Alexander von Humboldt Foundation (Germany) Feodor-Lynen postdoctoral fellowship. V.K. was supported by a postdoctoral fellowship from the Helen Hay Whitney Foundation. E.N. is supported by the Howard Hughes Medical Institute.
Integrated supplementary information
(a) Schematic representation of the generation of substrate dinucleosomes. See methods section for details. ΦK27me3 = pseudo trimethylated lysine. (b) Dinucleosome ligation and purification. A native PAGE gel shows mononucleosomes before ligation (Nclmod and Nclsub), after ligation (Lig), and following preparative native PAGE purification (DiNcl). (c) Left, SDS-PAGE of the purified PRC2 complex used in this study. 4-20% gradient gel, stained with coomassie. Right, electrophoretic mobility shift assay of dinucleosome binding to PRC2 under the conditions used for cryo sample preparation. 1.2 µM dinucleosomes with 35 bp (DiNcl35), 40 bp (DiNcl40) or 30 bp (DiNcl30) linker lengths were incubated with 1.6 µM PRC2. Gels in (b) and (c) were stained with SYBR gold. (d) Reference-free 2D classes of negatively-stained EM samples of PRC2-DiNcl35 (left, after an initial round of 2D class averaging to include only classes with two nucleosomes) or PRC2 alone as a control (right).
(a) Picked particles were subjected to two rounds of 3D classification after which 102,743 particles were refined to 4.7 Å resolution overall. In order to improve the SUZ12 N-terminal region, focused classification after background subtraction was performed which helped achieve 4.6 Å overall resolution but with improved density for this sub-domain. (b) Fourier shell correlation (FSC) curve of the two independent half maps indicating an overall resolution of 4.6 Å according to the 0.143 FSC criterion (Rosenthal and Henderson, J. Mol. Biol. 333, 721-745, (2003), Scheres and Chen, Nat. Methods 9, 853-854, (2012)) (gold standard refinement). (c) EM map of PRC2 colored according to local resolution using the BLOCRES and BLOCFILT programs of the BSOFT package (Heymann and Belnap, J. Struct. Biol. 157, 3-18, (2007)).
Supplementary Figure 3 Structural characterization of dinucleosome binding by the full PRC2–AEBP2 complex
(a) Visualization of the movement of nucleosomes in the group 2 of PRC2 dinucleosome complexes (Fig. 1c). Marked in red, the nucleosome proximal to the CXC and SET domains of EZH2 on the top lobe, moves by about 40 Å across the different configurations observed. Marked in green, the nucleosome proximal to the bottom lobe of PRC2, moves very dramatically with respect to the complex and its structure is poorly defined in most of the class averages, pointing to both translation and rotation and thus, a very flexible tethering. (b) Further quantitation of the variability in position of the nucleosome distal from the active site within group 2 of PRC2 dinucleosome complexes (Fig. 1c). The nucleosome near the N-terminal part of SUZ12 moves up to >130 Å, whereas the potential substrate nucleosome near the active site of EZH2 is less flexible. Left, compilation of all classes in group 2 with red arrows marking the distance from the RBAP48 WD40 domain to the nucleosome as well as the angle. Middle panel, same arrows as in the left panel, without the 2D class averages for clearer visualization. Right, all arrows indicating all observed positions of this nucleosome, superimposed with their origin at the RBAP48 WD40 domain. The range of possible distances and orientations becomes clear. The average distance from RBAP48 to the nucleosome (shortest distance) is 83 Å ± 20 Å.
(a) For the first dataset, a subset of micrographs (top left: example micrograph showing PRC2-dinucleosome complexes in vitreous ice; green circles represent example particle picks) was used for manual picking and reference-free 2D classification (top right: example classes) to generate templates for automated particle picking using Gautomatch (Kay Zhang, MRC LMB, Cambridge, UK). Initial particle picks were reviewed and edited manually to exclude ice and aggregates and increase the number of good picks. A final set of 104,836 particles were subjected to 3D classification using an initial model that was manually created based on the 2D class averages obtained after manual picking. (b) Particle images from automated picking were used for 3D classification of the second dataset, bad picks were sorted out by two rounds of 3D classification, resulting in a subset of 38,400 particles from a total of ~2,000 micrographs after sorting out bad micrographs. (c) The smaller, third dataset was processed analogously to the second dataset.
(a) The best classes from three datasets (Supplementary Fig. 2) were combined and 3D auto-refined, resulting in the global PRC2-DiNcl35 reconstruction used for initial characterization (top right). (b) Fourier shell correlation (FSC) curve of the two independent half maps indicating an overall resolution of 6.2 Å according to the 0.143 FSC criterion (Rosenthal and Henderson, J. Mol. Biol. 333, 721-745, (2003), Scheres and Chen, Nat. Methods 9, 853-854, (2012)) (gold standard refinement). (c) Angular distribution of particle orientations indicates some preferred orientation biased towards ‘top views’ of the PRC2-DiNcl35 complex. Effects of anisotropy were minor and did not affect the interpretability of the map. (d) Local resolution estimation. PRC2-DiNcl35 EM map colored and filtered according to local resolution using the BLOCRES and BLOCFILT programs of the BSOFT package (Heymann and Belnap, J. Struct. Biol. 157, 3-18, (2007)). Color range: 6-10 Å. Regions of low resolution, indicating pronounced flexibility, correspond to the SANT domains, the peripheral regions of Nclmod, and the linker DNA. (e) 3D classification of the refined reconstruction to identify and compare subpopulations of PRC2-DiNcl35. The two maps shown for comparison in Fig. 2 are marked by boxes. (f) Superposition of all classes as outlines, and the dyad axes according to the coloring of the classes above. The variability of the orientation of dyad axes between classes is larger for Nclmod than for Nclsub.
Supplementary Figure 6 SBD, EBD and SANT regions of EZH2 and flexible fitting of the crystal structure of human PRC2
(a) Alternative coloring scheme to illustrate more clearly the location of the SBD, EBD and SANT1 regions of EZH2. The long, N-terminal helix of EZH2 has been formally subdivided and into the most N-terminal part contacting the SANT1 helix bundle (i.e., SBD, aa 10-33, orange) and the C-terminal part traversing EED (i.e., EBD, aa 34-72, pink) . For clarity, all other figures have a uniform color for this region. The EBD is of no functional relevance for our structural analysis, and the SBD helix is clear to be an extension of the EBD contacting the nucleosomal DNA. As mentioned above, the SANT1 domain is not well resolved in our reconstructions due to its flexibility (b) The PRC2 crystal structure was fitted into the EM map by flexible fitting using iMODFIT (Lopez-Blanco and Chacon, J. Struct. Biol. 184, 261-270, (2013)) (red, model before flexible fitting, blue, model after flexible fitting into the EM map). The overall fit deviates only slightly from the crystal structure. The SBD helix of EZH2 is an exception (top right), as it is tilted upwards away from EED by approximately 36˚ with respect to the reported crystal structure.
Supplementary Figure 7 Signal subtraction and masked refinement of the PRC2–Nclsub part of the complex
(a) To reduce the influence of the signal from the more flexible Nclmod and linker DNA on the alignment accuracy and better resolve the substrate nucleosome binding interface of PRC2, signal subtraction of the raw particle images was performed to better align this portion (green mask). This procedure resulted in a better resolved map (right panels). (b) The FSC curve of independently refined half maps (gold standard refinement) indicates an overall resolution of 4.9 Å according to the FSC = 0.143 criterion (Rosenthal and Henderson, J. Mol. Biol. 333, 721-745, (2003), Scheres and Chen, Nat. Methods 9, 853-854, (2012)) (left). (c) Local resolution estimation (BLOCRES, BSOFT package (Heymann and Belnap, J. Struct. Biol. 157, 3-18, (2007)) indicates that the best resolved parts (core of Nclsub and the SET domain of EZH2) reach 4.5 Angstroms resolution. (d) Flexible fitting of the published atomic model of human PRC2 (PDB ID 5HYN) indicates very close agreement with the EM density map, except for the SBD/SANT1 domains of EZH2. (e) Superposition of the crystal structure (purple) and fitted (blue) maps. Top, back view of the PRC2 catalytic lobe. Right panel, enlarged superposition of the SET and CXC domains (black box), indicating a slight relative tilt of the CXC domain (dashed circle). (f) Residual density after subtracting the density accounted for by the fitted Nclsub model with the extended H3 tail (as described in the text, Fig. 3e) and the PRC2 model flexibly fitted into the density. All unassigned density shown in purple. Red asterisks, additional unassigned densities, potentially representing parts of histone tails.
(a) To better resolve the interface between PRC2 and Nclmod, Nclsub and linker DNA signal was subtracted from the raw particle images23 and alignment-free 3D classification was performed, yielding classes of varying Nclmod orientations relative to PRC2. For closer assessment of the contacts between the nucleosome and PRC2, classes 1 and 3 were chosen, since these classes showed reasonable density in the region of interest and are examples of distinct Nclmod orientations. (b) Back views of the EED-Nclmod interface comparing classes 1 (left) and 3 (right) as described in Fig. 4. Asterisks according to Fig. 4 c. Bottom panel, same view as top panel, but displayed at lower threshold. For class1, a continuous density connects the peptide density at the EED WD40 domain and the exit site of the H3 tail from Nclmod (c) PRC2-Nclmod contacts mediated by the SBD and SANT1 domains of EZH2 vary with nucleosome orientation. Dots indicate possible contacts of SANT1 with either EED or DNA. Red: possible SANT1-EED contact; green: SANT1-SRM linker density; purple: putative contact between SANT1 and the nucleosomal DNA.
(a) Template-based particle picking from 1,532 micrographs resulted in selection of 190,479 particles. 3D classification and alignment yielded a major class of good particles (green box), which was further refined to obtain the final 3D reconstruction. (b) FSC curve of the PRC2-DiNcl30 refined dataset, indicating an overall resolution of 8.4 Å according to the FSC = 0.143 criterion (Rosenthal and Henderson, J. Mol. Biol. 333, 721-745, (2003), Scheres and Chen, Nat. Methods 9, 853-854, (2012)). (c) Angular distribution of particle orientations after refinement. (d). Map of the PRC2-DiNcl30 complex colored according to local resolution ranging from 9-13 Å. Local resolution estimation determined with BLOCRES (BSOFT package (Heymann and Belnap, J. Struct. Biol. 157, 3-18, (2007)). (e) Detailed views of the SANT1/SBD and the EED/SBD contacts with the nucleosomal DNA in the PRC2-DiNcl30 complex viewed from the front (left) and back (right) of the complex. Red dot, possible SANT1-EED contact as in Fig. 4d. Asterisks, EED-DNA contacts, coloring according to Fig. 4. Green dot, linker segment connecting SANT1 and the SRM.
(a) Image processing for the PRC2-DiNcl40 dataset collected at 120 kV with a 4k x 4k CCD camera. Left, representative micrograph. Middle, reference-free 2D classes to initially sort particles. Right, 3D classification resulting in a particle set of 16,333 particles used for auto-refinement. (b) Final reconstruction of the PRC2-DiNcl40 complex as top (left), front (middle) and back (right) views. (c) FSC plot indicating a final resolution of 13.3 Å according to the FSC = 0.143 criterion (Rosenthal and Henderson, J. Mol. Biol. 333, 721-745, (2003), Scheres and Chen, Nat. Methods 9, 853-854, (2012)). (d) Final EM reconstruction of PRC2-DiNcl40 with fitted atomic models for PRC2 and the DiNcl40 (for dinucleosome model generation, see methods).
Atomic models of mononucleosomes (PDB ID 3LZ1) were rigid-body fitted into the EM maps of PRC2-DiNcl30 (left) and PRC2-DiNcl40 (middle). Middle left, superposition of the nucleosome DNA backbones within the PRC2-DiNcl30 (blue) and PRC2-DiNcl40 (red) complexes. Right, the distance between the DNA exit base pairs from the nucleosome was measured as a straight axis in Chimera, using corresponding terminal base pairs in the nucleosome models as anchor points. The difference between the two distances, was ~ 28 Å, compatible with an expected linker DNA length difference of ~ 33 Å, taking into account differences in the linker DNA paths due to bending.
Supplementary Figure 12 Masked refinement after signal subtraction of PRC2–DiNcl30 and comparison with PRC2–DiNcl35 map
(a) Reconstruction of PRC2-DiNcl30 after subtracting the signal of Nclmod. Coloring according to the scheme in Fig. 1. (b) FSC plot indicating an improved overall resolution of 7.7 Å of the EM map after signal subtraction, according to the FSC = 0.143 gold standard criterion (Rosenthal and Henderson, J. Mol. Biol. 333, 721-745, (2003), Scheres and Chen, Nat. Methods 9, 853-854, (2012)). (c) Map of the PRC2-DiNcl30 complex after signal subtraction colored according to local resolution ranging from 8-12 Å. Right, inset showing the CXC domain as an enlarged view. Local resolution estimation determined with BLOCRES (BSOFT package (Heymann and Belnap, J. Struct. Biol. 157, 3-18, (2007)). (d) Comparison of the overall fits of the PRC2 atomic model as shown in Fig. 2 into the density map of PRC2-DiNcl35 (light yellow, map low-pass filtered to 7 Å) and PRC2-DiNcl30 (light gray). The two maps have been aligned based on their EED and EBD densities. (e) EM densities of the CXC domain of EZH2 within the PRC2-DiNcl35 (light yellow, left panels) and PRC2-DiNcl30 (light gray, right panels) complexes. The model (see methods and Supplementary Fig. 6d) was fitted into the PRC2-DiNcl35 map (low-pass filtered to 7 Å). The CXC domain deviates from the otherwise good fit of the atomic model of PRC2 in the PRC2-DiNcl30 complex. The first zinc cluster of CXC is potentially more flexible in PRC2-DiNcl30 and not as well resolved (top panel, back view). The DNA binding site of the CXC is tilted towards the back of PRC2 and the SET domain (red arrow, lower panels, bottom view).
Distances of H3 tails within the PRC2-DiNcl35 cryo-EM structure discussed in this study and the H3K4 binding site within RBAP48 (yellow circle). Only the direct connections made by the H3 tails from the nucleosome to the active site (H3sub) or EED (H3mod) are in good agreement with the distance that can be bridged by the respective number of residues without major clashes (green lines). Red indicates distances too large to be bridged by the histone tails. Orange, theoretically possible connections to RBAP48, but ruled out due to major clashes within PRC2.