Material availability

Unique and stable reagents generated in this study are available upon request.

Crosslinking gradient (GraFix)

The elution fractions of the freshly purified complexes were pooled and transferred to a crosslinking gradient, GraFix (Kastner et al., 2008), with the following buffer composition: 10% - 30% Glycerol, 150 mM KCl, 20 mM HEPES KOH pH 7.8; 0,05% Glutaraldehyde and spun for 14 h at 160,000 g (SW60Ti). The gradient was aliquoted in 150 μl fractions, quenched with 20 mM Tris-HCl pH 7.8 (final concentration) and the protein complex traced with a Dot-Blot against the SBP-tag. Fractions containing cross-linked complex were pooled and the glycerol was removed by multiple rounds of concentration using an 0.5 mL Amicon spin column (50 kDa cut-off).

Size exclusion chromatography

50 μL of the freshly purified complexes at 1-2 mg/ml were injected into a Superose6® Increase 3.2/300 column and eluted over 3.3 mL equilibrated in Buffer3 at the flowrate of 0.04 ml/min. Fractions of 100 μL were collected and subsequently analyzed by SDS-PAGE.

Interactions were assessed by mixing the purified protein complexes in 1:1 molar ratio and incubating them on ice for 30 min before injections.

Sample vitrification

R1.2/1.3 UltrAuFoil 300 mesh grids were glow-discharged from each side for 20 s at 25 mA at 0.3 bar using a Pelco EasyGlow device. The protein concentration was adjusted to 0.5 mg/ml as described above and 2.5 μl sample applied to each side of the grid. Excess of the sample was blotted away using a Vitrobot MARK IV at 4°C, 100% humidity for 2 s at −10 blotting force and plunge frozen in liquid ethane.

Cryo-EM data collection

The grids were loaded into a Titan Krios (FEI) electron microscope at the CM01 ESRF beamline (Kandiah et al., 2019) (dataset ESRF1 and ESRF3) or at the EMBL Heidelberg cryo-EM platform (dataset EMBL2), both equipped with a K2 Summit direct electron detector and a GIF Quantum energy filter (Gatan). Microscopes were operated at 300kV acceleration voltage in an EF-TEM mode. The cryo-EM data was acquired using Thermo Fisher EPU software (ESRF) or serialEM (Mastronarde, 2005) (EMBL) at a nominal magnification of × 165 000, resulting in 0.83Å · pixel^-1 (ESRF) and 0.81Å · pixel^-1 (EMBL). ESRF movies were acquired for 4 s at a flux of 11.7e · Å^-2· s^–1 and the total fluence of 46.8 e · Å^-2 was fractionated into 40 movie frames. EMBL movies were acquired for 6 s at a flux of 7e · Å^-2· s^–1 and the total fluence of 42 e · Å^-2 was fractionated into 40 movie frames. A total of 6182 (ESRF) and 13,086 (EMBL) movies were acquired with a defocus range from −0.5 to −3.0μm.

Cryo-EM data processing

All image processing was performed within Relion 3.0 (Zivanov et al., 2018), unless stated otherwise. For all three datasets beam-induced motion correction was performed using Relion’s implementation of MotionCorr2 (Zheng et al., 2017) with a 5x5 patch model without binning followed by CTF estimation with CTFFIND 4.1 (Rohou and Grigorieff, 2015).

The datasets ESRF1 (1.02 million particles) and EMBL2 (4.65 million particles) were processed separately using a standard Relion workflow (Scheres, 2012), yielding 4.08 Å (118 k particles) and 3.98 Å (84 k particles) resolution respectively.

Automated particle picking of the ESRF1 dataset identified 1.02 million particles which were subjected to 3 rounds of 2D classification which reduced the particle number to 691 k. Two subsequent rounds of 3D classification with 3 classes each and a subsequent refinement resulted in a 8.3 Å resolution map with 302k particles. The refined map was classified further into 3 classes, the best class refined and again split into 5 classes. The best class containing 22 k particles was refined and yielded a 4.08 Å resolution map.

In case of the EMBL2 dataset automated picking in Relion identified 4.65 million particles. Due to the high number of particles, extraction was performed with 2-fold binning. The particles were subjected to 6 rounds of 2D classification to remove broken and poorly aligning particles, which reduced the particle number to 241 k. Those particles were re-extracted with their full pixel size and subjected to 3D classification into 6 classes. The best class, containing 84k particles, was refined and yielded a 3.98 Å resolution map.

The ESRF3 dataset was not processed separately as it was collected with the intention of merging the data with the previously collected datasets.

The final density of the EMBL2 dataset was used for template picking on all 3 datasets (ESRF1, EMBL2, ESRF2), yielding 2.9 million particles for the ESRF datasets and 6.1 million particles for the EMBL dataset. The ESRF particles were extracted from datasets 1 and 3 with full pixel size and merged. 2 rounds of 2D classification reduced the particle number to 786 k. The EMBL dataset was extracted 4-fold binned and subjected to 4 rounds of 2D classification which reduced the particle number to 1.6 million.

Since none of these datasets achieved the resolution required for de novo model building, we decided to merge all available data to improve the overall resolution. In order to do so, we determined a relative scaling factor between the maps from two different microscopes by iterative adjustment of the pixel size in one of the maps, followed by calculation of the correlation-coefficient in ChimeraX against the reference map (Wilkinson et al., 2019). As a result, a relative pixel size for the EMBL dataset was adjusted to 0.816 (instead of 0.810) and the CTF parameters were determined again for this dataset. To achieve the best possible scaling of this dataset against the ESRF reference (0.83Å · pixel^-1), we determined optimal box sizes for the extraction and the 786 k preselected ESRF particles were re-extracted in a 348x348 pixel box, while preselected 570 k EMBL particles were re-extracted in a 354x354 pixel box, downscaled to 348x348 pixels during the extraction. Both datasets were merged and the resulting 1.35 million particles were subjected to 2 rounds of 2D and one round of 3D classification. The best 3D class containing 541 k particles was refined to 4.9 Å resolution. Further 3D classification into 5 classes gave 2 equally good reconstructions, which differed in angular distribution. As previous data processing indicated a missing angle problem, both classes (312 k particles) were used in a subsequent 3D refinement yielding a reconstruction at 3.9 Å resolution. This reconstruction was used for CTF refinement and Bayesian polishing (Zivanov et al., 2019). Although we could not identify any significant heterogeneity in our data at this stage, further 3D classification allowed us to select a subset of particles, with presumably higher signal-to-noise ratio, which refined to 3.6 Å-resolution (from 99K particles) and 3.5 Å-resolution (from 27 k particles). The latter was used for subsequent model building.

The map used for de novo model building is missing some of the peripheral regions including the INTS9/11 CTD2 dimer and INTS4^CTD. While we were performing focused classification and refinement, we realized that classes containing well-defined peripheral regions originate almost exclusively in the ESRF1 dataset. It is possible that variation in sample preparation, vitrification conditions and/or ice thickness preserved those fragile elements only in 1 out of 3 datasets. Therefore, we performed focused classification and refinement on this dataset alone. Briefly, the initially picked 1.4 million particles were subjected to 3 rounds of 2D and 3 rounds of 3D classifications in order to remove broken particles and to select well-aligning classes. The resulting 115 k particles were refined to 4.4 Å-resolution. The angles assigned in this refinement were used for subsequent signal subtraction (Bai et al., 2015) followed by 3D classification without image alignment and T = 40. The best classes were reverted to the original particles and refined. Both maps yielded a final resolution of 6.5 Å.

All reported resolutions were calculated within Relion using gold standard Fourier Shell Correlation (FSC) procedures (Scheres and Chen, 2012).

Model Building and Refinement

For INTS9 and INTS11 homology models were calculated using the I-TASSER web server(Zhang, 2008) and fitted into the cryo-EM map in ChimeraX (Goddard et al., 2018). Based on the slightly larger molecular weight of INTS9, this subunit was assigned to the larger lobe, which was later confirmed by the amino acid register and identification of a unique insertion (NAD) in the INTS9^MBL domain. Homology models were rebuilt manually in Coot (0.8.9.2-pre EL) (Emsley and Cowtan, 2004). The density of the CTD1 dimer was good enough for de novo modeling. The CTD2 dimer was modeled by rigid-body docking of the previously reported coordinates (PDB ID: 5V8W) into a low-resolution lobe protruding from the end of the CTD1 dimer. Additionally, the linker helix connecting the CTD1 and CTD2 of INTS11 (495-504), which is as well part of the previously reported crystal structure of CTD2 (PDB ID: 5V8W), associates more tightly with the CTD1 dimer in our structure and its position is in agreement with the current CTD2 assignment.

The directionality of the INTS4 helical repeat was identified by crosslinking and mass spectrometry. The model for HEAT repeats 1-9 was built de novo into the high-resolution map and the register assignment was based on visible side-chain densities and secondary structure predictions (Kelley et al., 2015). The length of predicted helices is in good agreement with the visible density. Additionally, residues 264-280 are predicted to form a short extended loop between repeats H6 and H7, which is indeed visible in our map and provides a landmark verifying a correct sequence assignment.

A medium resolution map obtained from focused classification and refinement was used to extend the model with 3 additional HEAT repeats, whose register was tentatively assigned based on the secondary structure predictions.

The atomic model was refined in reciprocal space with Refmac5 (Murshudov et al., 2011) with the secondary structure restrains generated in ProSMART (Nicholls et al., 2014) within the CCP-EM (1.3.0) software suit (Burnley et al., 2017) . Half-map validation was performed as previously described (Brown et al., 2015). The atomic model was visualized in ChimeraX (Goddard et al., 2018) and PyMol and the electrostatic calculations were performed with the APBS plugin in PyMol (Schrödinger).

Cross-linking Mass spectrometry

Protein complexes were purified as previously described and adjusted to a concentration of 1 mg/ml in a 50 μL volume. DSS (disuccinimidyl suberate) was dissolved in Dimethylformamide with a final concentration of 50 mM. The dissolved crosslinker was added to the protein aliquot with a final concentration of 1 mM and 10 mM respectively and incubated at 35°C for 30 min. The reaction was quenched with 0.1 volumes of 1M Ammoniumbicarbonate and again incubated for 10 min at 35°C. Subsequently the sample was treated with 0.8 volumes of 10 M Urea and 250 mM Ammoniumbicarbonate as well as 0.05 volumes RapiGest SF Surfactant (Waters, cat. No. 186008090) and sonicated for 1 min in an ultrasound bath. Later, DTT was added with a final concentration of 10 mM, incubated for 30 min at 37°C and freshly prepared Iodoacetamide was added with a final concentration of 15 mM and again incubated for 30 min at room temperature (protected from light).

Subsequently, the sample was treated with different proteases beginning with Endoproteinase Lys-C (5 μL of an 0.1 mg/ml stock in Ammoniumbicarbonate) and incubated for 4 h at 37°C. The Urea concentration was adjusted to 1.5 M with HPLC grade water and Trypsin was added (1 μL of an 1 mg/ml stock) and incubated over night at 37°C. Subsequently the sample was acidified with Trifluoroacetic acid (1% v/v final concentration), incubated at 37°C for 30 min and spun down at 17,000 g for 5 min. The supernatant was discarded and the pellet was frozen in liquid nitrogen and stored at −80°C or dry ice. The mass spectrometry experiment and cross-links identification was performed by the EMBL Proteomics Core Facility in Heidelberg.

Analytical pull-down and quantitative mass spectrometry

The cell lines expressing Protein A – TEV – SBP N-terminal tagged INTS4, INTS5, INTS7, INTS10 or INTS14 were grown in 150 mL Freestyle media to a density of 1x10⁶ cells/ml and harvested at 300 g (JA14) for 10 min and 4°C. The cells were disrupted and fractionated following Dignam’s nuclear extract preparation protocol (Dignam et al., 1983) Nuclear extract and S100 fractions were frozen in liquid nitrogen and stored at −80°C until further use. The S100 or the nuclear extract fractions were thawed, and incubated with IgG beads for 2 h on a turning wheel at 4°C. Subsequently the beads were washed with Buffer3 (150 mM KCl, 20 mM HEPES-KOH pH 7.8) and 200 μL Buffer3 containing 12.5ug of TEV protease was added to the IgG beads to elute the proteins. The digestion was performed at 20°C for 90 min and the resin was additionally eluted 4 times with 200 μL of Buffer3. The elution fractions were pooled, added to Steptavidin beads and incubated again for 90 min on a turning wheel at 4°C. Subsequently, the beads were washed and eluted in 70 μL Buffer4 (150 mM KCl, 20 mM HEPES-KOH pH 7.8, 10 mM desthiobiotin). The eluate was spun down at 17,000 g and 4°C or 5 min, the supernatant collected and flash frozen in liquid nitrogen and stored at −80°C or on dry ice until further use.

For the mass spectrometric analysis, 40 μL of SBP elution fractions obtained from nuclear extract (INTS4 and INTS5, shown in Figures 1B and 1C) or S100 fraction (INTS4 and INTS10, shown in Figures 1D and 1E) were subjected to an in-solution tryptic digest using a modified version of the Single-Pot Solid-Phase-enhanced Sample Preparation (SP3) protocol (Hughes et al., 2014; Moggridge et al., 2018). Samples were added to Sera-Mag Beads (Thermo Scientific, #4515-2105-050250, 6515-2105-050250) in 20 μl 15% formic acid and 60 μl of ethanol. Binding of proteins was achieved by shaking for 15 min at room temperature. SDS was removed by 4 subsequent washes with 200 μl of 70% ethanol. Proteins were digested with 0.4 μg of sequencing grade modified trypsin (Promega, #V5111) in 40 μl Na-HEPES, pH 8.4 in the presence of 1.25 mM TCEP and 5 mM chloroacetamide (Sigma-Aldrich, #C0267) overnight at room temperature. Beads were separated, washed with 10 μl of an aqueous solution of 2% DMSO and the combined eluates were dried down. Peptides were reconstituted in 10 μl of H₂O and reacted with 80 μg of TMT10plex (Werner et al., 2014) (Thermo Scientific, #90111) label reagent dissolved in 4 μl of acetonitrile for 1 h at room temperature. Excess TMT reagent was quenched by the addition of 4 μl of an aqueous solution of 5% hydroxylamine (Sigma, 438227). Peptides were mixed to achieve a 1:1 ratio across all TMT-channels. Mixed peptides were subjected to a reverse phase clean-up step (OASIS HLB 96-well μElution Plate, Waters #186001828BA) and analyzed by LC-MS/MS on a Q Exactive Plus (Thermo Scentific) as previously described (Becher et al., 2018).

Briefly, peptides were separated using an UltiMate 3000 RSLC (Thermo Scientific) equipped with a trapping cartridge (Precolumn; C18 PepMap 100, 5 lm, 300 lm i.d. × 5 mm, 100Å) and an analytical column (Waters nanoEase HSS C18 T3, 75 lm × 25 cm, 1.8 lm, 100Å). Solvent A: aqueous 0.1% formic acid; Solvent B: 0.1% formic acid in acetonitrile (all solvents were of LC-MS grade). Peptides were loaded on the trapping cartridge using solvent A for 3 min with a flow of 30 μl/min. Peptides were separated on the analytical column with a constant flow of 0.3 μl/min applying a 2 h gradient of 2 – 28% of solvent B in A, followed by an increase to 40% B. Peptides were directly analyzed in positive ion mode applying with a spray voltage of 2.3 kV and a capillary temperature of 320°C using a Nanospray-Flex ion source and a Pico-Tip Emitter 360 lm OD × 20 lm ID; 10 lm tip (New Objective). MS spectra with a mass range of 375–1.200 m/z were acquired in profile mode using a resolution of 70.000 [maximum fill time of 250 ms or a maximum of 3e6 ions (automatic gain control, AGC)]. Fragmentation was triggered for the top 10 peaks with charge 2–4 on the MS scan (data-dependent acquisition) with a 30 s dynamic exclusion window (normalized collision energy was 32). Precursors were isolated with a 0.7 m/z window and MS/MS spectra were acquired in profile mode with a resolution of 35,000 (maximum fill time of 120 ms or an AGC target of 2e5 ions).

Acquired data were analyzed using IsobarQuant (PMID: 26379230) and Mascot V2.4 (Matrix Science) using a reverse UniProt FASTA Homo sapiens database (UP000000589) including common contaminants. The following modifications were taken into account: Carbamidomethyl (C, fixed), TMT10plex (K, fixed), Acetyl (N-term, variable), Oxidation (M, variable) and TMT10plex (N-term, variable). The mass error tolerance for full scan MS spectra was set to 10 ppm and for MS/MS spectra to 0.02 Da. A maximum of 2 missed cleavages were allowed. A minimum of 2 unique peptides with a peptide length of at least seven amino acids and a false discovery rate below 0.01 were required on the peptide and protein level (Savitski et al., 2015).

The raw output files of IsobarQuant were processed using the R programming language (ISBN 3-900051-07-0). Only proteins that were quantified with at least two unique peptides were considered for the analysis. Raw signal-sums (signal_sum columns) were normalized using variance stabilization normalization (Huber et al., 2002). Ratios were computed using these normalized TMT reporter ion signals. The top3 value is the average log10 MS1 intensity of the three most abundant peptides for each protein and serves as an estimator for the average abundance of a protein in the multiplexed mass spec run.

The mass spectrometry experiment and data analysis was conducted by the EMBL Proteomic Core Facility in Heidelberg.

RNA in vitro transcription and labeling

RNA substrates for binding studies were generated by T7 run-off transcription using the Milligan transcription method from annealed template DNA oligonucleotides (Milligan et al., 1987). The U1 stem-loop 4 substrate comprises human U1 snRNA nucleotides 137-164 followed by 34 downstream nucleotides, including the 3′-box sequence. The control RNA is a scrambled sequence of the same size. Transcription products were gel purified, enzymatically capped with the vaccinia capping enzyme (NEB) following manufacturer recommendations and labeled with fluorescein at the 3′ end. Briefly, the 3′ vicinal diol was oxidized by re-suspending the RNA in 100 μL of freshly made oxidation solution (0.1 M sodium periodate and 0.1 M sodium acetate (pH 5.0)) and incubated at room temperature for 1.5 h in the dark. The reaction was quenched by the addition of 10 μL of 3 M KCl, placed on ice for 10 min, and the resulting insoluble KIO₄ was pelleted by a brief centrifugation. A thiosemicarbazide derivative of fluorescein (100 mM in DMSO) was added to a final concentration of 50 mM and incubated at room temperature for 4 h, as previously described (Hardin et al., 2015). Labeled products were purified using a denaturing PAGE.

EMSA

Fluorescein labeled RNA was adjusted to a final concentration of 10 nM, mixed with a 2-fold dilution series of purified INTS4/9/11 with a maximum concentration of 10 μM and incubated on ice for 2h. Subsequently the sample was loaded onto a 1% TBE-Agarose gel and run for 120 min at 50 V and 4°C. Gels were visualized with the Chemidoc imaging system (Bio-Rad).

Dot blot

Dot-Blots were used to trace the crosslinked protein samples after the GraFix. A PVDF membrane was incubated for 5 min in PBS and installed on a Dot blot rack (BioRad) and washed twice with 100 μL PBS. 10 μL of the crosslinked and fractionized GraFix were mixed with 100 μL PBS and blotted onto the PVDF membrane. Subsequently, the membrane was washed twice with 100 μL PBS and transferred to 5% milk in PBST. After 1 h incubation the milk was replaced by Anti-SBP antibody (Merck, cat. MAB10764), 1:5000 dilution) in 5% milk in PBST. 1 hour later, the antibody was removed, the membrane washed twice for 5 min with PBST and incubated again for 1 h with goat Anti-Mouse-IgG-HRP conjugate (Thermo Fisher, 3cat. 1430, 1:2000 dilution). Subsequently, the membrane was washed 4 times with PBST, developed with Pirece™ ECL Western Blotting Substrate (Thermo Fisher, cat. 32106, 1ml) and imaged with a Bio-Rad Chemidoc system.

GFP-based in vivo reporter assay

The reporter plasmid was cloned by inserting U7 snRNA gene (including 200 nt of the promoter and 70 nt of the downstream regions) into a modified backbone of the pFLAG_CMV10 vector (SIGMA) followed by EGFP ORF, based on the previously described design (Albrecht and Wagner, 2012).

Integrator subunit knock-downs were performed in HEK293T cells (ATCC) using Lipofectamine RNAiMAX (Invitrogen) and the following siRNAs: scrambled control (UGCACCGAGUGGCGACACCUU), INTS4 (GUAGGCUUAAGGAGUAUGUGAUU) and INTS11 (CAGACUUCCUGGACUGUGUUU) used in a previous study (Albrecht et al., 2018).

Integrator subunits for the rescue experiments were cloned into a modified pFLAG_CMV10 vector with N-terminal affinity tags: 3xFLAG for INTS11, 3xHA for INTS9 and SBP for INTS4. 3xFLAG_INTS11 was carrying silent mutations providing resistance to the siRNA treatment.

24 h before transfection, cells were seeded into 24-well plates to reach 60%–80% confluency at the time of transfection. For each condition 0.8 pmol of siRNA was diluted in 50 μL of opti-MEM and mixed with 50ul of opti-MEM containing 2.4ul of RNAiMAX (Invitrogen). After 5 min incubation, the transfection mixture was added dropwise to each well, containing 500 μL of DMEM/10% FBS (GIBCO). After 24 h incubations the cells were split and re-seeded in a 1:3 ratio. After 24 h incubation the cells were transfected again with siRNA as described above. The following day, cells were split in 1:2 ratio, 12h before transfection of the reporter and rescue plasmids. For each well, 300 ng of the U7_GFP reporter plasmid was mixed with 50 ng of P2A-mCherry-N1 (Addgene #84329) and 300 ng of the rescue plasmids (or pFLAG_CMV10 negative control) in a total volume of 50 μL with DMEM (GIBCO), without FBS. The reporter/rescue/DMEM solution was mixed with 50 μL DMEM (no FBS) containing 1.95ul of LipoD293 (Sinagen). After 15 min incubation at room temperature, the mixture was added dropwise to each well.

Cells were harvested 36 h later by removing the media, adding 500 μL of cold PBS and gentle pipetting until the cells were detached. The harvested cells were spun down at 300 g for 2 min in a cooled table top centrifuge, resuspended in 50 ul RIPA buffer (150mM NaCl, 50mM Tris-Cl pH 8.0, 1%NP-40, 0.5% deoxycholate, 0.1% SDS) and transferred to a black, 96-well plate (Corning) for the fluorescence readout using a Clariostar plate reader.

The raw GFP fluorescence intensity (FI) was divided by mCherry FI (transfection control) and normalized against the mock-treated control. Each condition was done in biological triplicates. The error bars in the figures correspond to the standard deviation of the measurements.

After FI readout, samples were analyzed for transgene expression by western blot.

Immunoprecipitation assay

HEK293T cells were seeded into 6-well plates with a density of 500 000 cells per well in 1.5 mL D-MEM medium supplemented with 10% FBS. Cells were transfected 24 h later by preparing 50 μL Opti-MEM with 1 μg of each transfected plasmid and 50 μL Opti-MEM with 3 μg PEI25k per μg plasmid for each well. Subsequently, PEI and Plasmid were mixed, incubated at room temperature for 30 min and dropwise added into the cells. After 24 h of transfection 1.5 mL D-MEM/FBS was added to each well. 48 h after transfection the cells were harvested by removing the media, adding PBS and detaching the cells by gentle pipetting. The cells were spun down, resuspended in 400 μL lysis buffer (150 mM KCl, 20 mM HEPES-KOH pH 7.8 and 0.1% Triton X-100) and sonicated for 10 s at 30% amplitude. Next, the sonicated cells were spun down using a table top centrifuge at 20 000 g at 4°C for 30 min and the supernatant was added onto affinity resin (HA agarose) to capture the bait protein. The beads and the lysate were incubated for 2h on a turning wheel. Subsequently, the beads were spun down, the supernatant removed and the beads were washed 3 times with 150 mM KCl and 20 mM HEPES-KOH pH 7.8. Finally, the beads were resuspended in SDS sample buffer and heated up to 85°C for 5 min to release bound proteins. Input and elution fractions were analyzed by western blotting.

Author contributions

M.M.P. identified, expressed, and purified stable Integrator sub-complexes, and performed cryo-EM data analysis, model building and refinement, EMSA assays, and pull-down validation. W.P.G. initiated and coordinated the project and provided support with experimental procedures. M.M.P. and W.P.G. analyzed the data and wrote the manuscript.