Structural basis for PRC2 decoding of active histone methylation marks H3K36me2/3.

Key resources table

Reagent type (species) or resource Designation Source or reference Identifiers Additional information Strain, strain background D. melanogaster Oregon-R Flybase Strain, strain background D. melanogaster w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 12xHisGU wt /12 xHisGU wt McKay et al., 2015 Strain, strain background D. melanogaster w; Df(2L)His C FRT40A/CyO ubi-GFP; 12xHisGU H3K36R /TM6B McKay et al., 2015 Strain, strain background D. melanogaster w; Df(2L)His C FRT40A/CyO twi:Gal4 UAS:GFP; 3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb)/3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb) This study Available on request Strain, strain background D. melanogaster w; Df(2L)His C FRT40A/CyO twi:Gal4 UAS:GFP; 3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb)/3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb) This study Available on request Strain, strain background D. melanogaster w; Df(2L)His C FRT40A/CyO ubi:GFP; 3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb) Pengelly et al., 2013 Strain, strain background D. melanogaster w hs-flp; w; hs-nGFP FRT40A/hs nGFP FRT40; 3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb) Pengelly et al., 2013 Strain, strain background D. melanogaster w hs-flp; M(2)25A ubi-GFP FRT40A/CyO Müller lab stocks Available on request Strain, strain background D. melanogaster yw; esc 6 b pr/CyO, P[esc + ] Struhl laboratory Struhl, 1981 Strain, strain background D. melanogaster In(2LR) Gla/CyO, esc 2 Struhl laboratory Struhl, 1981 Strain, strain background D. melanogaster w hs-flp; hs-nGFP FRT2A/hs nGFP FRT2A Beuchle et al., 2001 Strain, strain background ( Escherichia coli ) BL21(DE3) Sigma-Aldrich CMC0016 Electrocompetent cells Cell line ( Trichoplusia ni ) HighFive cell line for expression Invitrogen Product nr.: B85502 BTI-Tn-5B1-4 (RRID: CVCL_C190 ) Protein expression Cell line ( Spodoptera frugiperda ) Sf21 cell line for Baculovirus production Invitrogen Product nr.: 1149701 (RRID: CVCL_0518 ) Baculovirus production for protein expression Antibody H3K27me3 Rabbit monoclonal antibody Cell Signaling Technology Cell Signaling Technology #9733 IF (1:50) WB (1:2000) ChIP (1:500) Antibody H3K27me3 Rabbit polyclonal antibody Millipore Millipore #07–449 WB (1:1000) Antibody H3K27me1 Rabbit polyclonal antibody Millipore Millipore #07–448 WB (1:6000) Antibody H3K36me3 Rabbit monoclonal antibody Cell Signaling Technology Cell Signaling Technology #4909 WB (1:750) Antibody H3K36me2 Rabbit monoclonal antibody Cell Signaling Technology Cell Signaling Technology #2901 WB (1:250) Antibody H3K36me2 Rabbit monoclonal antibody Abcam #9049 ChIP (1:300) Antibody H2B Rabbit polyclonal antibody This study Raised against full-length recombinant Drosophila H2B WB (1:10000) Available on request Antibody H4 Rabbit polyclonal antibody Abcam Abcam #10158 WB (1:200000) Antibody Caf1 Rabbit polyclonal antibody Gambetta et al., 2009 Müller lab WB (1:10000) Antibody Abd-B Mouse monoclonal antibody DHSB DSHB (1A2E9) IF (1:300) Antibody Antp Mouse monoclonal antibody DHSB DSHB (8C11) IF (1: 100) Recombinant DNA reagent pfC31-attB-3xHisGU.H3K36A This study See Materials and Methods Available on request Recombinant DNA reagent pfC31-attB-3xHisGU.H3K36R This study See Materials and Methods Available on request Recombinant DNA reagent nucleosome-positioning sequence 601 (147 bp + linker version) Lowary and Widom, 1998 , Nekrasov et al., 2005 Recombinant DNA reagent pFB-EZH2 Choi et al., 2017 N-terminal 6xHis-tag Recombinant DNA reagent pFB-EZH2 CXC>A This study N-terminal 6xHis-tag See Materials and Methods Available on request Recombinant DNA reagent pFB-EED Choi et al., 2017 N-terminal 6xHis-tag Recombinant DNA reagent pFB-EED EED>A This study N-terminal 6xHis-tag See Materials and Methods Available on request Recombinant DNA reagent pFB-SUZ12 Choi et al., 2017 N-terminal 6xHis-tag Recombinant DNA reagent pFB-RBBP4 Choi et al., 2017 N-terminal 6xHis-tag Recombinant DNA reagent pFB-PHF1 Choi et al., 2017 N-terminal twin-strep and 6xHis-tag (SHT) Recombinant DNA reagent pFB-PHF1 C (PHF1 515-567 ) Choi et al., 2017 N-terminal twin-strep and 6xHis-tag (SHT) Peptide, recombinant protein H3 18-42 peptide MPIB core facilty Peptide, recombinant protein H3 18-42 K36me3 peptide MPIB core facilty Protein expression and purification Human PHF1-PRC2 wild-type (wt) complex was expressed and purified as previously described ( Choi et al., 2017 ). In brief, an optimized ratio of the baculoviruses (produced in Sf21 cells, (Invitrogen 1149701)) for the different PHF1-PRC2 subunits was used to infect Trichoplusia ni High Five insect cells (Invitrogen B85502 ). The Sf21 and High Five cells were authenticated by genotyping (Eurofins) and tested negative for mycoplasma contamination (LookOut Mycoplasma PCR Detection Kit, Sigma-Aldrich). Cells were lysed using a glass Dounce homogenizer and the complex was purified using affinity chromatography (Ni-NTA and Strep-tag), followed by simultaneous TEV mediated protease tag cleavage and Lambda Phosphatase treatment (obtained from the MPI of Biochemistry Protein Core facility) and a final size-exclusion chromatography (SEC) step in a buffer containing 25 mM Hepes, pH 7.8, 150 mM NaCl, 10% glycerol, 2 mM DTT. PRC2 CXC>A , PRC2 EED>A and PRC2 CXC>A/EED>A mutants were generated by PCR with primers containing the desired mutations, subsequent ligation and transformation. Expression and purification were performed as above. Xenopus laevis ( X.l. ) and D. melanogaster ( D.m. ) histones were expressed in E.coli strains BL21 and purified from inclusion bodies as described in Luger et al., 1999 . To mimic the inhibitory mark H3K36me3 or the allosteric activating mark H3K27me3, the cysteine side chain of a mutated D.m. histone H3 C110A K36C or X.l. histone H3 C110A K27C was alkylated with (2-bromoethyl) trimethylammonium bromide (Sigma-Aldrich) as described previously ( Simon et al., 2007 ). Nucleosomes containing these modifications are abbreviated with e.g. H3Kc36me3. For histone octamers, equimolar amounts of histones H2A, H2B, H4 and H3 (wt, H3 K36A , H3 K36R , H3Kc27me3 or H3Kc36me3) were mixed and assembled into octamers in high salt buffer containing 10 mM Tris-HCL pH 7.5, 2 M NaCl, 1 mM EDTA, 5 mM β-mercaptoethanol. Subsequent SEC was performed to separate octamers from H3/H4 tetramers or H2A/H2B dimers ( Luger et al., 1999 ). Reconstitution of nucleosomes For X.l. and D.m mononucleosomes used in biochemical assays, 6-carboxyfluorescein (6-FAM)-labeled 215 bp 601 DNA ( Lowary and Widom, 1998 ) was PCR amplified from the p601 plasmid, purified on a MonoQ column (GE Healthcare), precipitated with ethanol and dissolved in the same high salt buffer used for octamers. Optimized ratios of octamer to DNA (usually ranging between 0.8–1.3: 1) were mixed and nucleosomes were reconstituted by gradient and stepwise dialysis against low salt buffers to a final buffer containing 25 mM Hepes, pH 7.8, 60 mM NaCl, 2 mM DTT. X.l. asymmetrical dinucleosomes for cryo-EM studies containing one unmodified substrate nucleosome and one H3Kc27me3-modified (allosteric) nucleosome connected with a 35 bp linker DNA were reconstituted using the protocol described in Poepsel et al., 2018 . In brief, substrate nucleosomes and allosteric nucleosomes were separately assembled on the respective DraIII digested nucleosomal DNA. The latter was generated by PCR with primers introducing the desired linker and DraIII recognition sites and purified as described above. The assembled nucleosomes were purified on a preparative native gel system (Biorad 491 prep cell). After ligation using T4 ligase (Thermo Fisher Scientific) the resulting dinucleosomes were purified from aberrant or non-ligated mononucleosomes by a second preparative native gel system (Biorad 491 prep cell). In contrast to Poepsel et al., 2018 , the dinucleosome DNA used in this study contained an additional 30 bp overhang on the substrate nucleosome, thus resulting in the following DNA sequence: 5′–601 binding (allosteric nucleosome) – agcgatctCACCCCGTGatgctcgatactgtcata – 601 binding (substrate nucleosome) – atgcatgcatatcattcgatctgagctcca –3’ (after DraIII digestion, assembly of substrate/allosteric nucleosome and ligation to dinucleosomes). X.l. symmetrical unmodified dinucleosomes used for the HMTase assays with the PRC2 CXC mutants were obtained by reconstituting octamers with a 377 bp DNA containing two 601 sequences connected by a 35 bp linker DNA. A vector containing the 377 bp sequence was ordered from Invitrogen GeneArt and was used for PCR resulting in: 5′– atatctcgggcttatgtgatggac – 601 binding (substrate nucleosome 1) – agcgatctcaacgagtgatgctcgatactgtcata – 601 binding (substrate nucleosome 2) – gtattgaacagcgactcgggatat –3′. The PCR products were purified as described above. Optimized ratios of octamer: DNA (usually ranging between 1.8–2.3: 1) were mixed and nucleosomes were reconstituted by gradient and stepwise dialysis against low salt buffers to a final buffer containing 25 mM Hepes, pH 7.8, 60 mM NaCl, 2 mM DTT.

Cryo-EM data acquisition

Complexes of PHF1-PRC2 and asymmetrically modified 35 bp dinucleosomes were assembled and grids were prepared as described previously, with the difference of using 0.005% NP40 instead of 0.01% ( Poepsel et al., 2018 ). Cryo-EM data were collected on an FEI Titan Krios microscope operated at 300 kV and equipped with a post-column GIF and a K2 Summit direct detector (Gatan) operated in counting mode. A total of 3467 movies were collected at a nominal magnification of 81,000x (1.746 Å/pixel) at the specimen level using a total exposure of 53 e - / Å 2 distributed over 60 frames and a target defocus range from 1.5 to 3 µm. Data acquisition was carried out with SerialEM.

Cryo-EM data processing

Movies were aligned and corrected for beam-induced motion as well as dose compensated using MotionCor2 ( Zheng et al., 2017 ). CTF estimation of the summed micrographs was performed with Gctf ( Zhang, 2016 ) and particles were picked in Gautomatch ( http://www.mrc-lmb.cam.ac.uk/kzhang/ K. Zhang, MRC LMB, Cambridge, UK) using templates created from the AEBP2-PRC2-dinucleosome cryo-EM structure (EMD-7306, Poepsel et al., 2018 ). All subsequent image processing steps were performed in Relion 3.0 ( Zivanov et al., 2018 ) as shown in Figure 1—figure supplement 2 . A total of 1,028,229 candidate particles were subjected to two rounds of initial 3D classification against a reference map (AEBP2-PRC2-dinucleosome low-pass filtered to 60 Å) and the Bayesian fudge factor (T value) set to 8. 330,482 remaining particles were subjected to two more rounds of 3D classification, this time using the best 3D model from the previous run as reference. Finally, the two best 3D models were 3D refined and further classified into 10 classes without translational and rotational sampling, using a T value of 4. From this run, the best 3D classes with the highest nominal overall resolution and rotational and translational accuracies were subjected to iterative rounds of 3D refinement, this time applying a soft mask for solvent flattening, per particle CTF refinement and Bayesian polishing. The highest nominal resolution was only achieved by combining several classes from the previous 3D run, likely due to missing particle views in one or the other individual class. The final map after postprocessing had an overall nominal resolution of 5.2 Å, as determined from the gold-standard FSC criterion of 0.143 ( Rosenthal and Henderson, 2003 ; Figure 1—figure supplement 1D ). The density (Overall PHF1-PRC2:di-Nuc) with fitted models is shown in Figure 1C and in Figure 1—figure supplement 1E using UCSF ChimeraX ( Goddard et al., 2018 ). Local resolution estimation was performed in Relion 3.0 and is shown in Figure 1—figure supplement 1B . The spherical angular distribution of all particles in the final model is shown in Figure 1—figure supplement 1C . To further improve the resolution and map details of the region around the H3 N-terminus, particle subtraction and focused 3D refinement was applied ( Bai et al., 2015 ; Zhou et al., 2015 ; Ilca et al., 2015 ). Using a mask generated with UCSF Chimera ( Pettersen et al., 2004 ) and Relion 3.0 the signal of the allosteric nucleosome as well as parts of PRC2 (EED and EZH2 allo ) was subtracted from all particle images. These signal subtracted particles were then subjected to focused 3D refinement using a soft mask around the substrate nucleosome and EZH2 sub . This yielded a 4.4 Å map (EZH2 sub -Nuc sub ) as determined from the gold-standard FSC criterion of 0.143 ( Rosenthal and Henderson, 2003 ; Figure 1—figure supplement 3B ). Local resolution estimation is shown in Figure 1—figure supplement 3A . For model building and depiction, the final density was further sharpened (applied b – factor: - 66) using the Multisharpen function in Coot ( Emsley et al., 2010 ) (e.g. in Figure 1F , Figure 1—figure supplement 3D ). To confirm the side-chain information visible in the Coot sharpened map, Phenix Resolve density modification was run on the two half maps generated from the 3D refinement of the EZH2 sub -Nuc sub map ( Terwilliger et al., 2020 ). The resolution of the map according to Phenix cryo-EM density modification output improved to 4 Å and the resulting map was used as an additional guideline for model building as well as for depiction (in Figure 1—figure supplement 4A-D ). Cryo-EM data fitting, modeling and refinement Available crystal structures were fitted into the final maps using rigid-body fitting in UCSF Chimera and all manual remodeling, morphing and building was performed in Coot. For PRC2, the crystal structure of the catalytic lobe of human PRC2 (PDB: 5HYN Justin et al., 2016 ; and comparing the fitted model to the cryo EM model of AEBP2-JARID2-PRC2 PDBs: 6C23 and 6C24 Kasinath et al., 2018 ) was used. Since the SBD helix and the SANT1 helix bundle of the crystal structure was not accommodated well by the corresponding EM density, this region was fitted separately. A model of a dinucleosome with linker DNA (Supplementary dataset one in Poepsel et al., 2018 , including crystal structures of nucleosomes, PDB 3LZ1, Vasudevan et al., 2010 , also PDB 1AOI, Luger et al., 1997 , also PDB 6T9L, Wang et al., 2020 , was fitted. The above described overall model was then used as a starting model for fitting and building EZH2 sub -Nuc sub into the focused map. Where possible, missing parts in the model were built de-novo, that is the H3 N-terminal tail (residues 30–37) between the catalytic site of PRC2 and the substrate histone. Available information from crystal/cryo EM structures was used as a guide (PRC2 with H3 peptide bound: PDB: 5HYN Justin et al., 2016 and cryo EM model of AEBP2-JARID2-PRC2 PDBs: 6C23 and 6C24 Kasinath et al., 2018 ), and high-resolution structures of nucleosomes (PDB 1AOI and PDB 6T9L) ( Luger et al., 1997 ; Wang et al., 2020 ). Parts of EZH2 sub -Nuc sub model were then fitted using the morph fit routine in Coot or manually ( Casañal et al., 2020 ). Secondary structure restraints for real-space refinement were generated automatically with phenix.secondary_structure_restraints ( Sobolev et al., 2015 ) and manually curated. Hydrogens were added and the model was real-space refined with a resolution- cutoff of 4.4 Å with Phenix ( Afonine et al., 2018 ) (phenix-1.18rc1-3777), using reference structures (PDB 6T9L, Wang et al., 2020 , and PDB 1AOI, Luger et al., 1997 for nucleosome and one copy of the human PRC2 crystal structure generated from PDB 5HYN ( Justin et al., 2016 ), applying strict secondary structure and Ramachandran restraints. Our final model includes the modeled side chains of the fitted crystal/cryo-EM structures. This is in our opinion supported by the data as the substrate nucleosome protein core is resolved to app. 4 Å ( Figure 1—figure supplement 3A ) and the map in these regions shows clear bulky side-chain information ( Figure 1—figure supplement 3D ). The EZH2 density is of worse quality however even at lower resolution side chains likely contribute to the signal in the particle images and thereby an overall good model to map fit (in our case given by the high CC values as well as FSC modelvsmap ) is arguably only ensured in the presence of side chains. However we caution readers against in interpreting our model at side-chain resolution in poorly resolved regions. Structures were visualized with UCSF ChimeraX ( Goddard et al., 2018 ) and PyMOL2 ( https://pymol.org/2/ ). Electrophoretic mobility shift assay (EMSA) EMSAs on a 1.2% agarose gel in 0.4x TBE Buffer with 45 nM 6-FAM - labeled mononucleosomes (unmodified wt X.l . for bandshifts with the PRC2 CXC mutants, unmodified wt D.m . and D.m H3Kc36me3 Simon et al., 2007 ) trimethyllysine analog containing nucleosomes) and increasing PRC2 concentrations (concentrations indicated in the figures above the gels) were performed in triplicates as described in Choi et al., 2017 . A Typhoon FLA 9500 scanner and the Fiji software was used for densitometric analysis of the 6-FAM signal ( Schindelin et al., 2012 ). Background correction and calculation of the fractions of bound nucleosomes was performed with R using tidyverse ( https://www.r-project.org/ ). In detail: two parts were boxed out in each lane: 1. unbound nucleosomes (‘unbound’ box) and 2. shifted nucleosomes (‘bound’, everything above ‘unbound’). The boxed-out signals were integrated and background corrected by subtracting the respective control (‘bound’ background of lane one for ‘bound’ boxes and ‘unbound’ background of lane 10 for ‘unbound’ boxes). To calculate the fraction of bound vs. unbound nucleosomes, the value for ‘bound’ nucleosome in each lane was divided by the total signal (sum of bound and unbound) of the same lane. Hill function fitting and illustration of the plot were subsequently performed with Prism 8 (GraphPad). Histone methyltransferase (HMTase) assay For all HMTase assays, 446 nM of mononucleosomes or 223 nM of dinucleosomes were incubated with indicated amounts of the different PRC2 complexes, in a reaction buffer containing 20 mM HEPES pH 7.8, 50 mM NaCl, 2.5 mM MgCl 2 , 5% glycerol, 0.25 mM EDTA, 0.5 mM DTT and 80 μM S-adenosylmethionine (SAM). Reactions were allowed to proceed for 90 min at RT before quenching by the addition of 1x (final concentration) SDS loading buffer and heat inactivation at 95°C for 5 min. Proteins were separated by electrophoresis on a 16% (w/v) SDS gel, transferred to a nitrocellulose membrane and probed with antibodies against H3K27me3 (Millipore, 07–449), H3K27me1 (Millipore, 07–448) and H4 (Abcam, ab10158). For quantification, HMTase reactions and the corresponding western blots on D.m. unmodified, H3Kc36me3, H3 K36A/R mononucleosomes were performed in triplicates and subjected to densitometric analysis (Chemiluminescence signal, ImageQuant LAS 4000). The integrated densitometric signal (band) in each lane was background corrected against the control lane (lane 1, no PRC2 in the reaction) and normalized with respect to the lane containing the highest amount (i.e. 100%) of PRC2 on unmodified nucleosomes (lane 4). The relative amounts of trimethylation/monomethylation for all other lanes were calculated with respect to lane 4. Graphical representations were made with Prism 8 (GraphPad). Mass spectrometry (MS) 500 nM of PRC2 or PRC2 CXC>A were incubated with 2 µM of either unmodified or H3 18-42 peptide containing the K36me3 modification in HMTase reaction buffer (described above) and methyltransferase activity was allowed to proceed over night at RT. Reactions were then quenched with 1% trifluoroacetic acid (TFA). Home-made stage tips with poly(styrenedivinylbenzene) copolymer (SDB-XC) were used to remove PRC2 from the reactions ( Rappsilber et al., 2007 ). First, stage tips were washed with methanol, followed by a second wash with buffer B (0.1% (v/v) formic acid, 80% (v/v) acetonitrile). The SDB-XC material was then equilibrated with buffer A (0.1% (v/v) formic acid) and 40 µl of sample was applied and washed several times. Finally, samples were eluted using buffer B and introduced into the Bruker maXis II ETD mass spectrometer by flow injection of 20 µl sample using an Agilent HPLC at a flow rate of 250 µl/min and 0.05% TFA in 70% acetonitril:H2O as solvent for ESI-MS time-of-flight analysis. Peptides were ionized at a capillary voltage of 4500 V and an end plate offset of 500 V. Full scan MS spectra (200–1600 m/z) were acquired at a spectra rate of 1 Hz and a collision cell energy of 15 eV. Raw data files were processed using Bruker Compass DataAnalysis. The m/z spectra were deconvoluted (maximum entropy method) with an instrument resolving power of 10,000 and the resulting neutral spectra peaks were integrated. For quantification, the experiment was performed in triplicates. The sum of the monomethylation peak areas was divided by the sum of the first 4 peaks of the input peptide together with the sum of the monomethylation peak areas. Illustration of the quantification was subsequently performed with Prism 8 (GraphPad). A Welch’s t-test was calculated to show the nonsignificant difference between the activity of PRC2 on unmodified or H3K36me3 peptide.

Construction of histone transgenes to generate H3 K36A and H3

K36R strains

Site directed mutagenesis on pENTR221-HisGU.WT, pENTRL4R1-HisGU.WT and pENTRR2L3-HisGU.WT ( Günesdogan et al., 2010 ) was used to mutate histone H3K36 to alanine or arginine. The final constructs pfC31-attB-3xHisGU.H3K36A and pfC31-attB-3xHisGU.H3K36R were generated by Gateway LR recombination of above vectors and integrated at attP sites VK33 (BDSC 9750) and 86Fb (BDSC 130437). The full genotypes of animals used in the study are described below.

Drosophila strains and genotypes

The following strains were used in this study: Oregon-R w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 12xHisGU wt /12 xHisGU wt ( McKay et al., 2015 ) w; Df(2L)His C FRT40A/CyO ubi-GFP; 12xHisGU H3K36R /TM6B ( McKay et al., 2015 ) w; Df(2L)His C FRT40A/CyO twi:Gal4 UAS:GFP; 3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb)/3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb) (generated in this study) w; Df(2L)His C FRT40A/CyO twi:Gal4 UAS:GFP; 3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb)/3xHisGU H3K36R (VK33) .3xHisGU H3K36R (86Fb) (generated in this study) w; Df(2L)His C FRT40A/CyO ubi:GFP; 3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb) ( Pengelly et al., 2013 ) w hs-flp; w; hs-nGFP FRT40A/hs nGFP FRT40; 3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb) ( Pengelly et al., 2013 ) w hs-flp; M(2)25A ubi-GFP FRT40A/CyO yw; esc 6 b pr/CyO, P[esc + ] In(2LR) Gla/CyO, esc 2 w hs-flp; hs-nGFP FRT2A/hs nGFP FRT2A The following genotypes were used for the experiments shown in: Figure 4 wt: Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU wt (VK33)/12xHisGU wt (VK33) H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B H3 K36R : w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb)/3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb) H3 K36A : Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb)/3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb) Figure 5A, C, E wt: Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU wt (VK33)/12xHisGU wt (VK33) H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B Figure 5B, D, F wt: Oregon-R H3 K36A : w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb)/3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb) H3 K36R : w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb)/3xHisGU H3K36R (VK33). 3xHisGU H3K36R (86Fb) Figure 6A wt: Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU wt (VK33)/12xHisGU wt (VK33) H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B H3 K36R : w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb)/3xHisGU H3K36R (VK33) 3xHisGU H3K36R (86Fb). H3 K36A : w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb)/3xHisGU H3K36A (VK33) 3xHisGU H3K36A (86Fb). H3 K27R : w; Df(2L)His C FRT40A/Df(2L)His C FRT40A; 3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb) esc – : esc 6 b pr/CyO, esc 2 ( esc mat- zyg- obtained as progeny from esc 6 b pr/CyO, esc 2 parents). Figure 6B wt: Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU wt (VK33)/12xHisGU wt (VK33) H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B Figure 6C wt: Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU wt (VK33)/12xHisGU wt (VK33) H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B H3 K27R : w hs-flp; Df(2L)His C FRT40A/hs-nGFP FRT40A; 3xHisGU H3K27R (68E)3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E)3xHisGU H3K27R (86Fb) Figure 6D wt: Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU wt (VK33)/12xHisGU wt (VK33) H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B H3 K36A : w hs-flp; Df(2L)HisC FRT40A/M(2)25AubiGFP FRT40; 3xHisGU H3K36A (VK33) 3xHisGU H3K36R (86Fb)/ + H3 K27R : w hs-flp; Df(2L)HisC FRT40A/M(2)25A ubi-GFP FRT40; 3xHisGU H3K27R (68E) 3xHisGU H3K27R (86Fb)/ + Figure 6—figure supplement 1 wt: w hs-flp; hs--nGFP FRT2A/hs nGFP FRT2A H3 K36R : Df(2L) HisC FRT40/Df(2L) HisC FRT40; 12xHisGU H3K36R (VK33)/TM6B H3 K27R : w hs-flp; Df(2L)His C FRT40A/hs-nGFP FRT40A; 3xHisGU H3K27R (68E)3xHisGU H3K27R (86Fb)/3xHisGU H3K27R (68E)3xHisGU H3K27R (86Fb) Comparison of the lethality of H3 K36R and H3 K36A mutants The difference in the lethality phase of the H3 K36R mutants generated in this study compared to H3 K36R mutants in the strain from Matera and colleagues was unexpected because both strains the Df(2L)HisC homozygotes carry 12 copies of the HisGU H3K36R cassette (i.e. four 3x HisGU H3K36R arrays in our strain and a single 12x HisGU H3K36R array in the strain from Matera and colleagues). A possible explanation for the poorer survival of H3 K36R mutants in the strain generated here could be that histone transgene expression from the 3xHisGU H3K36R miniarrays is for some reason less effective that in the case of the 12x HisGU H3K36R array. We also note that a recent study reported that among Df(2L)HisC homozygotes that carry 20 HisGU H3K36A copies, about 50% of the mutant animals develop up to the pupal stages ( Zhang et al., 2019 ). Zhang et al have not analyzed their H3 K36A mutants any further but it is possible that the higher copy number of the HisGU H3K36A cassette accounts for the better survival compared to the H3 K36A strain generated in this study. Further studies will be needed to explore whether H3 K36R and H3 K36A mutants show comparable phenotypes in larvae.

Immunohistochemistry and immunofluorescence stainings

Embryos of the appropriate genotypes listed above were identified by the lack of GFP marked balancer chromosomes, fixed and stained with Abd-B antibody, following standard protocols. Imaginal discs from third instar larvae were stained with Antp and Cy3-labeled secondary antibodies following standard protocols. For clonal analysis ( Figure 3D ), clones were induced 96 hr before analyses by heat-shocked induced expression of Flp recombinase in the genotypes listed above. ChIP-seq analysis in Drosophila embryos and in larval tissues Embryo collection, chromatin preparation, and ChIP 21-24 hr old wt , H3 K36A embryos (see above for details of genotypes) were dechorionated, quick-frozen in liquid N2 and stored at -80°C. 5 µl of thawed embryos were homogenized in 5 mL of fixing solution (60 mM KCl, 15 mM NaCl, 4 mM MgCl 2 , 15 mM Hepes pH 7.6, 0.5% Triton X-100, 0.5 mM DTT, protease inhibitors, 0.9% Formaldehyde) at r.t. The homogenate was filtered through a strainer (Greiner Bio-One, EASYstrainer 100 µm, #542 000) and incubated for 10 min with frequent gentle shaking. Cross-linking was stopped by the addition of 450 µl of 2.5 M Glycine. Fixed nuclei were washed with 1 ml of buffer A1 (60 mM KCl, 15 mM NaCl, 4 mM MgCl 2 , 15 mM Hepes pH 7.6, 0.5% Triton X-100, 0.5 mM DTT, protease inhibitors), washed with 1 ml of pre-lysis buffer (140 mM NaCl, 15 mM Hepes pH 7.6, 1 mM EDTA, 0.5 mM EGTA, 1% Triton X-100, 0.5 mM DTT, 0.1% Na Deoxycholate, protease inhibitors), resuspended in 1 ml of lysis buffer (140 mM NaCl, 15 mM Hepes pH 7.6, 1 mM EDTA, 0.5 mM EGTA, 1% Triton X-100, 0.5 mM DTT, 0.1% Na Deoxycholate, protease inhibitors, 0.1% SDS, 0.5% N-laurylsarcosine), incubated at least 10 min at 4°C with shaking, and transferred into milliTUBES 1 ml AFA Fiber (100) (Covaris, #520130) for sonication. Sonication was performed in a Covaris S220 AFA instrument using the following setup: 140W (peak incident power) / 5% (duty cycle) / 200 (cycle per burst) / 15 min. Insoluble material was removed by centrifugation in an Eppendorf centrifuge at 14000 rpm (10 min at 4°C). Input chromatin was quantified by measuring DNA concentration after decrosslinking using Qubit (Thermo Scientific) and 250 ng of chromatin were used for each ChIP experiment. 250 ng of an independently prepared batch of D. pseudoobscura chromatin were spiked-in in each ChIP experiment for subsequent normalization of the ChIP-seq datasets. The rest of the ChIP protocol was performed as described in Bonnet et al., 2019 . For each condition, the ChIP experiment was performed in duplicates from two biologically independent chromatins. ChIP on hand-dissected CNS and imaginal disc tissues from 3 rd instar wt or H3 K36R homozygous larvae (see above for details on genotypes) was performed as described Laprell et al., 2017 with the difference D. pseudoobscura chromatin was spiked in at a 1:1 ratio of dm / dp chromatin.

Library preparation and sequencing

Library preparation for sequencing was performed with TruSeq kits from Illumina. Illumina systems (NextSeq 500) were used for paired-end DNA sequencing. All reads were aligned using STAR ( Dobin et al., 2013 ) to the D. melanogaster dm6 genome assembly ( dos Santos et al., 2015 ) and to the D. pseudoobscura dp3 genome assembly (Nov. 2004, FlyBase Release 1.03). Only sequences that mapped uniquely to the genome with a maximum of two mismatches were considered for further analyses. Identification of H3K36me2 and H3K27me3 enriched regions The Bioconductor STAN-package ( Zacher et al., 2017 ) was used to define the location of H3K36me2-enriched regions. The seven chromosome arms (X, 2L, 2R, 3L, 3R, 4 and Y) defined in the dm6 genome assembly were segmented in 200 bp bins. STAN annotated each of these bins into 1 of 3 ‘genomic states’ based on the number of H3K36me2 ChIP-seq reads and the number of input reads overlapping with each bin in wildtype embryos or larvae. These 3 ‘genomic states’ corresponded to: ‘H3K36me2 enriched’ regions; ‘low or no H3K36me2’ regions and ‘no input’ regions. The Poisson Lognormal distribution was selected and fitting of hidden Markov models was performed with a maximum number of 100 iterations. Stretches of consecutive bins annotated as ‘H3K36me2 enriched’ regions were sometimes separated by a few bins showing another type of annotation (i.e. ‘no input’). To define a relevant set of H3K36me2 enriched regions, we considered that if stretches of consecutive bins annotated as ‘H3K36me2 enriched’ regions are not separated by more than 7 Kb, they can be fused. High-level H3K27me3 domains previously defined using the same Bioconductor STAN-package in Bonnet et al., 2019 were used in this study.

Normalization and visualisation of H3K27me3 and H3K36me2 ChIP-Seq datasets

The proportion of D. pseudoobscura reads as compared to D. melanogaster reads in input and in samples was used to normalize the H3K36me2 and H3K27me3 ChIP-seq datasets from H3 K36A and H3 K36R mutants to the corresponding wildtype H3K36me2 and H3K27me3 ChIP-seq datasets respectively (see Supplementary file 2 ). Chip-seq tracks shown in Figure 4 show the average of the two replicates that were performed for each condition. Y-axes of ChIP-seq tracks correspond to normalized numbers of mapped reads per million reads per 200 bp bin.

Calculation of read coverage

In wildtype and H3 K36A and H3 K36R mutant conditions, H3K36me2 and H3K27me3 ChIP-seq read coverages across gene bodies were computed on genomic intervals starting 750 bp upstream transcription start sites and ending 750 bp downstream transcription termination sites. Read coverage is defined as the normalized number of mapped reads per million reads from a ChIP-seq dataset divided by the number of mapped reads per million reads from the corresponding input dataset across a genomic region. Among the D. melanogaster Refseq genes, approximately 10800 and 9200 are overlapping with H3K36me2 enriched regions, approximately 1030 and 1030 genes are overlapping with high-level H3K27me3 domains and 5400 and 6300 are localized in other genomic regions in embryos and larvae, respectively.

Drosophila nuclear and cell extracts for western blot analysis

For embryonic total nuclear extracts, nuclei from 21 to 24 hr old wt , H3 K36A or H3 K36A mutant embryos were purified and quantified as described in Bonnet et al., 2019 . Pellets of nuclei were resuspended in appropriate volumes of SDS sample buffer proportional to the number of nuclei in each pellet. Extracts were then sonicated in a Bioruptor instrument (Diagenode) (eight cycles (30 s ON/30 s OFF), high power mode), incubated at 75 °C for 5 min and insoluble material was removed by centrifugation at 14000 rpm for one mn at r.t. Total cell extracts from imaginal disc tissues were prepared by resuspending hand-dissected disc tissues in SDS sample buffer. Extracts were then sonicated, incubated at 75 °C for 5 min and insoluble material was removed by centrifugation. Antibodies For ChIP analysis: Rabbit monoclonal anti-H3K27me3 Cell Signaling Technology #9733 Rabbit polyclonal anti-H3K36me2 Abcam #9049 For western blot analysis on embryonic and larval extracts: Rabbit monoclonal anti-H3K27me3 Cell Signaling Technology #9733 Rabbit polyclonal anti- H3K27me3 Millipore #07-449 Rabbit polyclonal anti-H3K27me1 Millipore #07-448 Rabbit monoclonal anti-H3K36me3 Cell Signaling Technology #4909 Rabbit monoclonal anti-H3K36me2 Cell Signaling Technology #2901 Rabbit polyclonal anti-H2B (against full-length recombinant D.m. H2B) Rabbit polyclonal anti-H4 Abcam #10158 Rabbit polyclonal anti-Caf1 Gambetta et al., 2009 For immunohistochemistry and immunofluorescence analysis: Mouse monoclonal anti-Abd-B DSHB (1A2E9) Mouse monoclonal anti-Antp DSHB (8C11) Rabbit monoclonal anti-H3K27me3 Cell Signaling Technology #9733

Additional files Source code 1. PDB model of PHF1-PRC2:di-Nuc complex (related to Figure 1 ) with the corresponding description text file. A pseudoatomic pdb model of the overall PHF1-PRC2:di-Nuc structure. Available crystal structures were fitted into the final map: for PRC2, the crystal structure of the catalytic lobe of human PRC2 (PDB: 5HYN, Justin et al., 2016 ) was fitted. A model of a dinucleosome with linker DNA (Supplementary dataset one in Poepsel et al., 2018 ), including crystal structures of nucleosomes, (PDB 3LZ1, Vasudevan et al., 2010 , also PDB 1AOI, Luger et al., 1997 , and PDB 6T9L, Wang et al., 2020 ) was used. We caution the reader that this model did not undergo real-space-refinement. PRC2: Chain A: EZH2 Chain B: EED Chain C: SUZ12 Chain D – Chain K = substrate nucleosome histones Chain D: Histone H3 #1 Chain E: Histone H4 #1 Chain F = Histone H2A #1 Chain G = Histone H2B #1 Chain H: Histone H3 #2 Chain I: Histone H4 #2 Chain J = Histone H2A #2 Chain K = Histone H2B #2 Chain L – Chain S = allosteric nucleosome histones Chain L: Histone H3 #1 Chain M: Histone H4 #1 Chain N = Histone H2A #1 Chain O = Histone H2B #1 Chain P: Histone H3 #2 Chain Q: Histone H4 #2 Chain R = Histone H2A #2 Chain S = Histone H2B #2 Chain T and chain U = nucleosomal DNA Widom601+linker DNA Supplementary file 1. Cryo-electron microscopy data collection summary, processing statistics and model. Supplementary file 2. Number of aligned reads to the D. melanogaster and D. pseudoobscura genomes from ChIP and input datasets and normalization process (related to Figure 5 and Figure 5—figure supplement 1 ). Transparent reporting form