Mechanism-based traps enable protease and hydrolase substrate discovery – Nature

Plasmid construction

Standard molecular biology techniques, including PCR, restriction cloning, Gibson assembly, Golden gate assembly, and quik-change mutagenesis were applied to assemble plasmids. To generate plasmids for protein expression in E. coli, the DNA fragment encoding human HtrA2(134–458), RBBP9 or SCoV2-PLpro was synthesized as a double stranded DNA (Integrated DNA Technologies (IDT)), and UL36^USP (UL36(39–285)) and UL36^USP(C65S) were amplified from UL36^USP containing plasmids⁴⁸. The encoding sequence was cloned into pNHD vector² with a C-terminal HA–Strep-tag for HtrA2 and RBBP9, and a C-terminal twin-Strep-tag for UL36^USP and SCoV2-PLpro. To convert catalytic serine/cysteine to alanine or amber stop codon, site directed mutagenesis was completed using quik-change primers (Agilent primer design). To generate vectors for protein expression in mammalian cells, DapRST², TEV², human RHBDL4 (ref. ⁴⁹) or RBBP9 encoding sequence was amplified and introduced into the previously reported pcDNA3.1 plasmid backbone⁵⁰ with 4 × [U6-^PyltRNA_CUA] . A C-terminal HA–Strep-tag was introduced to TEV, and a twin-Strep-tag was introduced at the C-terminus of RHBDL4, RBBP9, UL36^USP or SCoV2-PLpro. The DNA fragments encoding ER-resident proteins—human CCDC47, PDIA6, CALR, GANAB, ERP44, CALU, PRKCSH, FKBP9, DNAJC3 and CANX—were amplified from HEK293T cDNA and cloned into a previously reported pcDNA3.1-based BiP expressing plasmid⁵¹, in which an ER leader peptide was followed by a V5-tag and BiP encoding sequence. An additional HA-tag was introduced at the C-terminus of CCDC47, or before the ER retention motif sequence of ERP44. The DNA fragments encoding MFN2, LEMD2, EMD, HNRNPH1 and HNRNPM were amplified from HEK293T cDNA and cloned into a pcDNA3.1 plasmid. An HA-tag was placed at the C-terminus of MFN2, LEMD2 or HNRNPM, while a GFP-tag was introduced at the C-terminus of Emerin or HNRNPH1 for better detection of the proteolytic fragments. The guide RNA (gRNA) for RHBDL4 knockout was introduced into pX330-puro plasmid with the optimized scaffold⁵² via Golden gate assembly.

Western blot

Samples were separated by SDS–PAGE (note that NuPAGE 4–12%, 10% Bis-Tris or 3–8% Tris-Acetate gels running in MES or MOPS buffer were applied to achieve the optimal separation of proteins (protein fragments) of interest) and transferred to polyvinylidene difluoride (PVDF) membrane by iBlot 2 dry blotting system (Thermo Fisher Scientific). Membrane was blocked by Odyssey blocking buffer in PBS (catalogue (cat.) no. 927-40000, Li-Cor) at room temperature for 30 min. Membrane was incubated in primary antibody solution (dilution according to manufacturer’s instructions in Odyssey T20 (PBS) antibody diluent (927-75001, Li-Cor)) at 4 °C overnight. All incubations were carried out on a platform shaker. The membrane was washed three time with PBST (PBS supplemented with 0.1% Tween-20 (v/v)), and incubated with the secondary antibody solution (1: 15,000 (v/v) in PBS blocking buffer supplemented with 0.2% Tween-20 (v/v), and 0.01% SDS) at room temperature for 1 h. After washing 3 times with PBST and once with PBS, the immunoreactive proteins were visualized by the Odyssey CLx imaging system (Li-Cor) by scanning at 700 nm and/or 800 nm channels. Revert 700 Total Protein Stain (926–11015, Li-Cor) was used for total protein staining. The data were analysed by Image Studio Lite (version 5.2.5). For primary and secondary antibodies used in this study, see ‘Antibodies’ in Supplementary Methods.

Deprotection of pc-Dap containing proteins in buffer

To activate protease(pc-Dap), proteins were illuminated (365 nm, 4 mW cm⁻²) for 1 min in Tris buffer (5 mM DTT, pH 8.0) and incubated at 4 °C or 37 °C overnight to generate protease(Dap). MIC-LED-365 (500 mA, Prizmatix collimated modular Mic-LED light source, Supplementary Fig. 2) was used for illuminating proteins in solution. This apparatus was also used for illuminating suspension cells (Expi293 cells) in tissue culture hood.

HtrA2 substrate trapping in cell lysate

Thirty micrograms HtrA2–HA–Strep variant (wild type, Ala or Dap) was added to 1 ml of Expi293 cell lysate (3 μg μl⁻¹) and incubated at 30 °C for 3 h. The reaction was shaken 10 s every 10 min. Fifty microlitres of anti-HA agarose slurry (A2095, Merck) was added to the reaction and mixed at 4 °C on an end-over-end rocker for 2 h. The mixture was transferred to a Bio-spin column. The resin was washed with RIPA buffer 3 times and PBST buffer 3 times using a vacuum pump, followed by centrifugation at 5,000g for 1 min to remove the residual buffer. Then, beads were incubated in 100 μl 1× LDS loading buffer and boiled at 95 °C for 5 min. Twenty microlitres of the eluate was analysed by SDS–PAGE or western blot. Twenty microlitres of eluate was separated in a Bolt 10% Bis-Tris Plus gel for 3 min at 200 V. Gel slices containing all proteins were cut and analysed by LC–MS/MS as described in ‘Electrospray ionization tandem mass spectrometry’.

Validation of HtrA2 substrates in cell lysate

Wild-type HtrA2 or HtrA2(S306A) (1 μM) was added to 1.2 ml Expi293 cell lysate. At indicated time points, 300 μl of reaction was quenched by mixing with 100 μl 4 × LDS loading buffer and boiled at 95 °C for 5 min. The samples were analysed by western blot with primary antibodies listed in Supplementary Tables 2, 3. GAPDH was used as a loading control.

Incorporation of pc-Dap in HEK293T cells

HEK293T cells were purchased from European Collection of Cell Cultures (authenticated by STR DNA profiling) and were tested negative for Mycoplasma contamination.

HEK293T cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (FBS) (Gibco) and Penicillin-Streptomycin (Pen/Strep, 100 IU ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin) at 37 °C in a humidified incubator supplied with 5% CO₂. Cells were passaged every 2–3 days by detaching with trypsin–EDTA solution, resuspended in DMEM with 10% FBS, and seeded into cell culture flasks.

For transfection in a 24-well plate: 0.75 μl of Lipofectamine 3000 (Thermo Fisher) was diluted in 25 μl Opti-MEM (Gibco) and vortexed briefly. DNA solution was prepared by mixing 500 ng DNA mixture (substrate:DapRST:protease, 1:1:3 or empty vector:DapRST:Protease, 1:1:3) in 25 μl Opti-MEM, followed by adding 1 μl of P3000 reagent. Then, diluted Lipofectamine was added to DNA solution (1:1 v/v). The mixture was incubated at room temperature for 10 min and the DNA–lipid complexes were added to cells. Indicated concentrations in figure legends (or 0.5 mM) of pc-Dap was added to the culture medium 30 min after transfection to achieve pc-Dap incorporation. Cells were incubated at 37 °C for 40–48 h before further analysis.

Incorporation of pc-Dap in Expi293 cells

Expi293 cells were purchased from Thermo Fisher (authenticated by STR DNA profiling) and were tested negative for Mycoplasma contamination.

Expi293 cells were cultured in Expi293 media (Gibco) and shaken at 125 rpm in incubator at 37 °C with 8% CO₂. Cells were passaged every 2–3 days, starting with the cell density around 0.5 × 10⁶ cells per ml. Transfection was performed at cell density around 2.5 × 10⁶ cells per ml.

Transfection of 100 ml Expi293 cells: 300 μl of polyethyleneimine molecular mass 40,000 (PEI, 1 mg ml⁻¹, Polysciences) was diluted in 3.3 ml Expi293 medium. 100 μg DNA mixture (substrate:DapRST:protease, 1:1:3 or empty vector:DapRST:protease, 1:1:3) was diluted in 3.3 mL Expi293 media. Diluted DNA and PEI solution were mixed and incubated at room temperature for 15 min before adding to the cell culture. 0.5 mM (or indicated concentrations in figure legends) of pc-Dap was added 30 min after transfection for pc-Dap incorporation. Forty to forty-eight hours after transfection, the cells were collected and photoactivated for further analysis.

Photoactivation of protease(pc-Dap) and substrate trapping in mammalian cells

Forty to forty-eight hours after transfection, cell culture medium was replaced with fresh medium, and cells were illuminated for 2 min. The apparatus for illuminating adherent mammalian cells was built in-house (Supplementary Fig. 7). LuxiGen 365 nm UV LED Emitter (LZ4-04UV0R-0000, Mouser Electronics) was used for illumination. The UV intensity at the well plate was set at 4 mW cm⁻². After illumination, cells were incubated at 37 °C for indicated period of time (Proteasome inhibitor MG132 (2 μM) was added if needed). For adherent cells, at each time point, cells in a 6-well plate were washed with PBS and lysed in 400 μl RIPA lysis buffer (89900, Thermo) supplemented with Halt Protease Inhibitor Cocktail (78429, Thermo Fisher) and the Universal Nuclease (88702, Thermo Fisher). The lysate was cleared by centrifuging at 21,000g for 5 min and the supernatant was flash frozen and stored at −80 °C. For suspension cells, at each indicated time point, 5 ml cell culture was centrifuged at 650g for 5 min and the cell pellet was flash frozen and kept at −80 °C. Then, cell pellets were lysed in 1 ml RIPA lysis buffer supplemented with protease inhibitors and the Universal Nuclease at 4 °C. Cell lysates were centrifuged at 21,000g for 5 min. The cleared lysates were used for SDS–PAGE and western blot analysis or affinity enrichment by MagStrep type3 XT beads (2-4090-002, IBA).

Trapping endogenous substrates to RHBDL4(S144Dap)

RHBDL4 variants were expressed in Expi293 cells as described in ‘Incorporation of pc-Dap in Expi293 cells’. Forty hours after transfection, 50 ml cell culture was resuspended in fresh Expi293 media and illuminated (365 nm, 4 mW cm⁻²) for 2 min. Cells were incubated at 37 °C for 4 h in the presence of 2 μM MG132 before collection. After pelleting, cells were resuspended in HEPES buffer (50 mM HEPES, pH 7.4, 150 mM NaCl, 1 mM MgCl₂, 5% glycerol, 1 mM DTT) supplemented with protease inhibitors and the Universal Nuclease. The suspension was lysed by passing twice through an Avestin Emulsiflex C3 homogenizer (ATA Scientific) at 3,000–5,000 psi. The lysate was centrifuged at 1,000g for 5 min twice and the supernatant was further centrifuged at 100,000g for 1 h. The pellet was washed with Na₂CO₃ (100 mM, pH 11.3) at 4 °C for 20 min and then centrifuged at 140,000g for 1 h. The membrane fraction was dissolved in 2% SDS buffer (50 mM Tris, pH 8, 150 mM NaCl, 1 mM DTT and protease inhibitors). The solution was diluted by 10% NP40 to generate a final concentration of 0.1% SDS and 1% NP40. One-hundred microlitres MagStrep type3 XT beads were added to the solution and incubated at room temperature for 1 h. The beads were washed with RIPA and PBST three times each. Proteins attached to the beads were eluted in 1× LDS loading buffer by heating at 65 °C for 15 min. The eluates were separated in a Bolt 10% Bis-Tris Plus gel for 3 min. Gel slices containing proteins were cut and analysed by LC–MS/MS.

RHBDL4 cleavage assay

Empty vector, wild-type (WT) RHBDL4 or RHBDL4(S144A) plasmid was co-transfected with candidate substrate containing plasmid or empty vector (for endogenous substrates) into HEK293T or Expi293 cells. The amount of candidate substrate-containing plasmid was optimized for expression level. Forty to forty-eight hours after transfection, cells were collected and lysed in RIPA buffer supplemented with protease inhibitors and the Universal Nuclease. The lysate was cleared and analysed by western blot.

To analyse proteins in the extracellular medium, FBS-containing medium for HEK293T cells was replaced with hybridoma serum free medium (12045076, Thermo Fisher) 24 h before collection. Expi293 medium, which is serum-free and protein-free, can be directly collected for further analysis. The medium was collected and filtered through a 0.22 μm polyethersulfone membrane. To obtain total proteins in the medium, 1/10 volume of 100% ice-cold TCA solution (T0699, Sigma) was added at 4 °C for protein precipitation. To obtain proteins in the supernatant, the medium was centrifuged at 200,000g for 1 h to separate supernatant from microvesicles. The SN was collected and added 1/10 volume of ice-cold TCA to precipitate proteins. After centrifuging at 21,000g for 10 min, the precipitate was washed once with acetone, and dissolved in 1× LDS loading buffer. The microvesicles pellet after ultra-centrifugation was washed with PBS once and dissolved in equal volume of 1× LDS loading buffer.

Deglycosylation was performed by adding 1/10 volume of 10% NP40 and PNGase (P0704S, NEB) or DeGlycosylation mix II (P6044S, NEB) to proteins dissolved in the 1× LDS loading buffer. The reaction was incubated at 37 °C for 1 h before analysis.

Brefeldin A inhibitory assay

Twenty-four hours after transfection, Expi293 cells were split into two halves treated separately with DMSO or BFA (5 μg ml⁻¹). The BFA treatment was performed in two ways: (1) BFA was directly added into medium culture; (2) the medium culture was replaced with fresh medium supplemented with BFA. Sixteen hours after BFA treatment, the cells and extracellular medium were collected and analysed as described in ‘RHBDL4 cleavage assay’.

Knockout of RHBDL4

HCT116 cells were purchased from American Type Culture Collection (authenticated by STR DNA profiling) and were tested negative for Mycoplasma contamination.

HCT116 cells were cultured in McCoy’s 5A (modified) Media (Gibco) supplemented with 10% fetal bovine serum (FBS) (Gibco) and Pen/Strep at 37 °C in humidified incubator with 5% CO₂. Cells were passaged every 2–3 days.

HCT116 cells in 6-well plates were transfected by Lipofectamine LTX (15338100, Thermo Fisher) with 2.5 μg of pX330-puro plasmid containing gRNA (5′-TCCAGTAAGTACAGAAAATG-3′) and Cas9 for RHBDL4 knockout. Twenty-four hours after transfection, cells were trypsinized and plated in a 10 cm petri dish. After 24 h, the cells were treated with puromycin (1 μg ml⁻¹). The puromycin selection stopped after 48 h. Cells were trypsinized and limited dilution was performed to generate single clones, which were expanded and analysed by western blot (anti-RHBDL4) and genotyped by sequencing the genomic DNA region targeted by the gRNA.

To detect the proteolytic fragments from endogenous substrates generated by endogenous RHBDL4, 10 million wild-type or RHBDL4 knockout HCT116 cells were cultured in hybridoma serum free medium for 40 h. The medium was collected, filtered and concentrated by a 30 kDa cut-off concentrator. Proteins were precipitated by TCA and dissolved in 1× LDS loading buffer for immunoblotting analysis.

Trapping endogenous substrates to RBBP9(S75Dap)

To identify X attached to Pept(Dap), RBBP9 variants were expressed in HEK293T cells as described in ‘Incorporation of pc-Dap in HEK293T cells’. pc-Dap (0.1 mM) was added to cells to produce RBBP9(S75pc-Dap). To characterize the entire masses of RBBP9 variants, RBBP9 variants were produced in 100 ml Expi293 cells as described in ‘Incorporation of pc-Dap in Expi293 cells’. RBBP9(S75pc-Dap) was expressed in the presence of 0.5 mM pc-Dap. 40 h after transfection, cells were illuminated (365 nm, 4 mW cm⁻²) for 2 min, and incubated at 37 °C for 3 h. Cells were then collected and lysed in Tris buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 1 mM EDTA and Universal Nuclease) by sonication. Note that protease inhibitors were not added in lysis buffer. The lysate was cleared by centrifuging at 21,000g for 20 min. RBBP9 species in the supernatant were enriched using MagStrep type3 XT beads. Proteins attached to beads were eluted in 50 mM Biotin in Tris buffer for mass characterization.

Aminopeptidase assay of RBBP9

Fluorescence-based assay

Two micromolar RBBP9 was incubated with each AA–AMC over a range of different substrate concentrations in Tris buffer (100 mM Tris, 150 mM NaCl, pH 7.3). Fluorescence intensity (due to the release of the AMC fluorophore by hydrolysis of AA–AMC by RBBP9) was measured every 20 s over a 10-min period (MARS Data Analysis Software (version 3.20 R2)). For each substrate, the rate of fluorescence increase was converted to rate of product formation using standard curves. At substrate concentrations of greater than 10 μM, intermolecular quenching of AMC fluorescence by AA–AMC was found to be significant. Therefore, for all substrates other than Phe-AMC and Tyr-AMC, AA-AMC concentrations between 0 and 4 μM were used, and pseudo-first order kinetics were employed to calculate specificity constants. For Phe-AMC and Tyr-AMC, which showed significantly faster rates of hydrolysis when compared to the other substrates, a concentration range of 0 to 160 μM was used and converted rates were fitted to Michaelis–Menten kinetics in order to obtain specificity constants.

Peptide-based assay

Peptides (100 μM) were dissolved in Tris buffer (100 mM Tris pH 7.4, 150 mM NaCl, 1 mM EDTA). Two micromolar wild-type RBBP9 or RBBP9(S75A) was added to start the hydrolysis reaction. The reaction was stopped by quenching with acetic acid and monitored by mass spectrometry. Selected ion mass (SIM) mode was applied for detection of peptide substrates and the desired products.

Protein crystallization and data collection

Human RBBP9 with a C-terminal His-tag⁴⁰ (LEHHHHHH) was expressed in BL21(DE3) cells and purified by a two-step protocol consisting of HisTrap enrichment and gel filtration (Superdex 75) chromatography. Pure fractions of RBBP9 (> 98% purity determined by SDS–PAGE) were concentrated with a 10 kD MWCO Vivaspin 20 concentrator (Sartorius) to 10 mg ml⁻¹ in buffer containing 10 mM Tris (pH 7.5), 100 mM NaCl, 5 mM DTT and 5 mM Phe. Prior to crystallization, samples were cleared by centrifugation for 15 min at 10,000g. Crystallization trials with multiple commercial crystallization kits were performed in 96-well sitting-drop vapor diffusion plates (Molecular Dimensions) at 18 °C and set up with a mosquito HTS robot (TTP Labtech). Drop ratios of 0.2 μl protein solution plus 0.2 μl reservoir solution were used for coarse and fine screening. Initial hits were obtained under multiple conditions and required no further optimization. Data was collected from crystals collected from following conditions: 30% w/v PEG 4K, 0.1 M MES sodium salt, pH 6.5.

To ensure cryo-protection, crystal-containing drops were mixed with 25% glycerol in reservoir solution prior to picking and flash frozen in liquid nitrogen. Diffraction data was collected at the Diamond Light Source (DLS, UK) on beamline I04. Datasets were auto-processed with XIA2 DIALS (version 0.7.90), scaled using Aimless and Refmac5 (version 5.8.0258) in the CCP4 suite (version 7.0.078) of programs. Structure refinement and manual model building were performed with Refmac5 and COOT (version 0.8.9.2). Colour figures were prepared with PyMol (version 2.5).

Mass characterization

Electrospray ionization mass spectrometry

Mass spectra of all protein samples were acquired on an Agilent 1200 LC-MS system equipped with a 6130 Quadrupole spectrometer. A Phenomenex Jupiter C4 column (150 × 2 mm, 5 μm) was used to elute proteins. Buffer A (0.2% formic acid in H₂O) and buffer B (0.2% formic acid in acetonitrile) was used for RP-HPLC. Mass spectra were acquired in the positive mode and analysed by the MS Chemstation software (Rev.C.01.06[61], Agilent Technologies). The deconvolution program provided in the software was used to obtain the entire mass spectra. Theoretical molecular mass of proteins with non-canonical amino acids was calculated by correcting the calculated molecular mass of wild-type protein (http://www.peptidesynthetics.co.uk/tools/) with the molecular mass of non-canonical amino acids.

Electrospray ionization tandem mass spectrometry

Proteins (including TEV-GFP conjugate, substrates trapped to HtrA2 or RHBDL4) in polyacrylamide gel slices (1–2 mm) were enzymatically digested in situ for LC–MS/MS analysis. In brief, the excised protein gel pieces were placed in a 96-well microtitre plate and destained with 50% v/v acetonitrile and 50 mM ammonium bicarbonate, followed by reduction with 10 mM DTT and alkylation with 55 mM iodoacetamide. RBBP9 eluates in solution were treated in two ways: (1) incubation at room temperature overnight in the presence of 10 mM DTT without alkylation; (2) reduction with 10 mM DTT for 30 min and alkylation with 55 mM iodoacetamide. Then, proteins were digested with trypsin/LysC (Promega) overnight at 37 °C. The resulting peptides were extracted in 2% v/v formic acid, 2% v/v acetonitrile and analyzed by nanoscale capillary LC-MS/MS, which uses an Ultimate U3000 HPLC (ThermoScientific Dionex) with a flow rate of 300 nl min⁻¹. A C18 Acclaim PepMap100 5 μm, 100 μm × 20 mm nanoViper (ThermoScientific Dionex) was used to trap the peptides before separation on a C18 Acclaim PepMap100 3 μm, 75 μm × 150 mm nanoViper (ThermoScientific Dionex). Peptides were eluted with a gradient of acetonitrile. The eluate was directly introduced to a modified nanoflow ESI source with a hybrid dual pressure linear ion trap mass spectrometer (Orbitrap Velos, ThermoScientific). Data-dependent analysis was carried out using a resolution of 30,000 for the full MS spectrum, followed by ten MS/MS spectra in the linear ion trap. MS spectra were collected over an m/z range of 100–2,000.

LC–MS/MS data analysis by Venn diagram

LC–MS/MS data were searched against an in-house protein sequence database containing Swiss-Prot and the protein constructs specific to the experiment, using the Mascot search engine program (Matrix Science, version 2.4). Database search parameters were set with a precursor tolerance of 5 p.p.m. and a fragment ion mass tolerance of 0.8 Da. Variable modifications for oxidized methionine, carbamidomethyl cysteine, pyroglutamic acid, and deamination of glutamine/asparagine were included. MS/MS data were validated using the Scaffold program (version 5, Proteome Software Inc.).

LC-MS/MS data analysis by volcano plot

For quantitative analysis, MS raw files were processed by MaxQuant software (version 1.6.3.4) and searched with the embedded Andromeda search engine against the corresponding database (Uniprot). The required FDR was set to 1% or 5% at peptide and protein levels. The maximum number of allowed missed cleavages was set to two. Protein quantification was done by LFQ with default settings. The MaxQuant ProteinGroups output file was further processed with Perseus (version 1.6.2.3)⁵³. Contaminations and reverse hits were removed by filtering. The remaining protein quantifications were log₂-transformed.

Determination of X attached to Dap in RBBP9

LC–MS/MS files (in RAW format) were first converted to mzML format⁵⁴ using ProteoWizard (version 3.0.11252)⁵⁵. Data preparation and processing were then performed using custom Python (version 3.8.1) scripts written with the pyOpenMS package (version 2.4.0)⁵⁶. In brief, collected spectra were centroided and all MS2 spectra with a precursor mass lower than that of the unconjugated Dap-containing tryptic peptide from RBBP9 (Pept(Dap)) were filtered out. For each filtered MS2 spectrum, the ten most abundant peaks in each 100 Th mass interval were extracted.

Based on the peptide sequence of Pept(Dap) and the precursor mass for each MS2 spectrum, a list of theoretical ion masses was calculated; these corresponded to the MS2 fragmentation of a substrate-conjugated Pept(Dap), (Pept(Dap-X)). This list contained the monocationic b- and y-ions, the dicatonic b- and y- ions, and ions corresponding to water or ammonium losses from the side-chains of b- or y- ions. Peaks in the MS2 spectrum were matched against this list, and a score for this matching was calculated as previously described⁵⁷. This score was ten times the negative logarithm of the approximate probability that at least k out of n masses have been matched by chance, where k is the number of matches and n is the number of masses in the list.

To extract the top-scoring spectra, the family-wise error rate for the probability values was controlled at 0.05 using the Bonferroni correction. The mass difference between Pept(Dap) and the precursor ion for each Pept(Dap-X) spectrum was calculated to determine the molecular mass of each conjugate. For each mass shift, representative top-scoring spectra were manually interrogated to verify the assignment.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.