Chemicals

Liproxstatin-1 (Lip-1: Selleckchem, cat. no. S7699), (1S,3R)-RSL3 (RSL3: Cayman, cat. no.19288), iFSP1 (ChemDiv, cat. no. 8009-2626 or Cayman cat. no. Cay29483), viFSP1 (ChemDiv, cat. no. D715-1847 or Vitas M Laboratory, cat. no. STK626779), deferoxamine mesylate salt (DFO: Sigma, cat. no. 138-14-7), ferrostatin-1 (Fer-1: Sigma, cat. no. SML0583), zVAD-FMK (zVAD: Enzo Life Sciences, cat. no. ALX-260-02), necrostatin-1s (Nec-1s: Enzo Life Sciences, cat. no. BV-2263-5), MCC950 (Sigma, cat. no. 5381200001), olaparib (Selleckchem, cat. no. S1060) and JKE-1674 (Cayman, cat. no. Cay30784-1) were used in this study. Custom-made compounds were obtained from Intonation Research Laboratories.

Cell lines

4-hydroxytamoxifen (Tam)-inducible Gpx4–/– murine immortalized fibroblasts (referred to as Pfa1 cells) have been reported previously26. Genomic Gpx4 deletion can be achieved using Tam-inducible Cre recombinase and the CreERT2–LoxP system. HT-1080 (CCL-121), HEK293T (CRL-3216), 786-O (CRL-1932), A375 (CRL-1619), B16F10 (CRL-6475), LLC (CRL-1642), MDA-MB-436 (HTB-130), SW620 (CCL-227), NCI-H460 (HTB-177) and 4T1 (CRL-2539) cells were obtained from ATCC. SKOV3 (91091004) cells were obtained from Sigma-Aldrich. HEC151 cells (JCRB1122-A) were obtained from Tebubio. MC38 cells (available from Sigma) were a gift from P. Agostinis (KU Leuven, Belgium). Rat1 cells (available from Thermo Fisher) were a gift from Medizinische Hochschule Hannover. Huh7 cells (available from Thermo Fisher) were a gift from R. Schneider (Helmholtz Munich). Pfa1, 786-O, A375, SW620, Huh7, HT-1080, Rat1, MC38, LLC and B16F10 cells were cultured in DMEM-high glucose (4.5 g glucose L–1) with 10% fetal bovine serum (FBS), 2 mM l-glutamine and 1% penicillin–streptomycin. MDA-MB-436, HEC151, SKOV3, H460 and 4T1 cells were cultured in RPMI GlutaMax with 10% FBS and 1% penicillin−streptomycin. To generate cell lines stably overexpressing FSP1, antibiotics (puromycin 1 µg mL–1 and blasticidin 10 µg mL–1) were used for selection. For culturing GPX4-deficient cells, 1 µM Lip-1 was supplemented. All cells were cultured at 37 °C with 5% CO2 and verified to be negative for mycoplasma.

Production and isolation of FSP1 enzyme

Recombinant human and mouse FSP1 proteins and FSP1 mutants were produced in BL21 E. coli and purified by affinity chromatography with a Ni-NTA system8.

FSP1 enzyme activity and inhibition assay

For the resazurin assay, enzyme reactions in TBS buffer (50 mM Tris-HCl, 150 mM NaCl) containing 15–200 nM recombinant human or mouse FSP1 and their mutants, 200 μM NADH and the inhibitors (iFSP1 and viFSP1) were prepared. After the addition of 100 μM resazurin sodium salt (Sigma, cat. no. R7017), the fluorescence intensity (FL intensity, excitation/emission wavelengths (Ex/Em) = 540 nm/590 nm) was recorded every 30 or 60 s using SpectraMax M5 or SpectraMaxiD5 microplate reader with SoftMax Pro v7 (Molecular devices) at 37 °C. Reactions without resazurin (inhibitor) were used to normalize and calculate FSP1 enzymatic activity and IC50 values. Curve fitting and calculation of IC50 values were performed using GraphPad Prism v9.

For the NADH-consumption assay, enzyme reactions in PBS (Gibco, cat. no. 14190094) containing 25–200 nM recombinant human FSP1 (or its mutants) and 50 μM menadione (Sigma, cat. no. M5625) were prepared. After the addition of 200 μM NADH, the absorbance at 340 nm at 37 °C was measured every 30 s using SpectraMax M5 Microplate Reader (Molecular Devices). Reactions without NADH or without enzyme were used to normalize the results. Curve fitting was done using GraphPad Prism v9.

For analysis of enzyme kinetics, reactions in PBS (Gibco, cat. no. 14190094) containing 100 nM recombinant human FSP1 and 0.04–500 μM NADH with 100 μM resazurin sodium salt or 0.03–100 μM resazurin with 200 µM NADH were mixed; then, FL intensity (Ex/Em = 540/590 nm) was recorded every 60 s using the SpectraMaxiD5 microplate reader (Molecular Devices) at 37 °C. The initial slope of FL intensity was used for following the calculations of Km and Vmax using GraphPad Prism v9.

Cell viability assay

Cells were seeded on 96-well plates (2,000–10,000 cells per well for ferroptosis inducers and 500 cells per well for TAM) or 384-well plates (800 cells per well) and cultured overnight. The next day, the medium was changed and the following compounds were added: RSL3, iFSP1, viFSP1, Lip-1, DFO, Fer-1, zVAD, Nec-1s, MCC950 and olaparib, at the indicated concentrations in each figure. Cell viability was determined after 24–48 h (for RSL3, iFSP1, and viFSP1) or 72 h (for TAM) upon treatment, using AquaBluer (MultiTarget Pharmaceuticals, cat. no.6015) or 0.004% resazurin as an indicator of viable cells.

As readout, fluorescence was measured at Ex/Em = 540/590 nm using a SpectraMax M5 microplate reader with SoftMax Pro v7 (Molecular devices) after 4–6 h of incubation in normal cell-culture medium. Cell viability (%) was normalized and calculated using untreated conditions, which is in the absence of ferroptosis inducers or tamoxifen.

The synergistic effect was assessed using MuSyC (https://musyc.lolab.xyz)38,39.

LDH release assay

For the LDH release assay, 2,500 cells per well were seeded on 96-well plates and cultured overnight. On the following day, the medium was changed to the fresh DMEM containing inhibitors and incubated for another 24 h. Necrotic cell death was determined using the Cytotoxicity Detection kit (LDH) (Roche, cat. no. 11644793001) following the manufacturer’s protocol. In brief, cell-culture supernatant was collected as a sample of the medium, and cells were then lysed with 0.1% Triton X-100 in PBS as a lysate sample. Medium and lysate samples were individually mixed with reagents on the 96-well plate, and the reaction mixture was incubated for 15–30 min at room temperature. Then, the absorbance was measured at 492 nm using the SpectraMax M5 microplate reader (Molecular Devices). The cell death ratio was calculated by LDH release (%) as follows: (absorbance (abs) of medium sample) / ((abs of lysate) + (abs of medium samples)) × 100.

Lipid peroxidation assay

One day before the experiments, 100,000 cells per well were seeded on a 12-well plate. On the next day, cells were treated with 10 µM viFSP1 for 3 h and were then incubated with 1.5 µM C11-BODIPY 581/591 (Invitrogen, cat. no. D3861) for 30 min in a 5% CO2 atmosphere at 37 °C. Subsequently, cells were washed with PBS once and trypsinized at 37 °C. Then, cells were resuspended in 500 µL PBS and passed through a 40-μm cell strainer, followed by analysis using a flow cytometer (CytoFLEX with the software (CytExpert v2.4), Beckman Coulter) with a 488-nm laser for excitation. Data were collected from the fluorescein isothiocyanate (FITC) detector (for the oxidized form of BODIPY) with a 525/40 nm bandpass filter, or from the phycoerythrin (PE) detector (for the reduced form of BODIPY) with a 585/42 nm bandpass filter. At least 10,000 events were analyzed per sample. Data were analyzed using FlowJo Software. The ratio of fluorescence of C11-BODIPY 581/591 (lipid peroxidation) (FITC/PE ratio (oxidized/reduced)) was calculated using the median value of each channel30.

Live-cell imaging

Pfa1 cells (5,000–10,000 cells) were seeded on a µ-Dish 35 mm, low (ibidi, cat. no. 80136), and incubated overnight. The next day, the cell culture medium was changed to fresh medium. Live-cell imaging was performed using 3D Cell Explorer (Nanolive) with the software Evev1.8.2. During imaging, the cells were maintained at 37 °C and a 5% CO2 atmosphere using a temperature-controlled incubation chamber. After recording one image, a 100-fold concentration of viFSP1 in DMEM was added to the dishes (the final concentration was 10 µM viFSP1), and then recording was continued. For ferroptosis suppression, cells were pretreated 0.5 µM Lip-1 for 15 min before recording. Images were taken every 5 min for more than 6 h.

Cell lysis and immunoblotting

Cells were lysed in LCW lysis buffer (0.5% Triton X-100, 0.5% sodium deoxycholate salt, 150 mM NaCl, 20 mM Tris-HCl, 10 mM EDTA, 30 mM Na-pyrophosphate tetrabasic decahydrate), supplemented with protease and phosphatase inhibitor cocktail (cOmplete and phoSTOP, Roche, cat. no. 04693116001 and cat. no. 4906837001), and centrifuged at 20,000g, 4 °C, for 30 min to 1 h. The cell lysate was sampled by dissolved with 6×SDS loading buffer (375 mM Tris-HCl, 9 % SDS, 50 % glycerol, 9 % β-mercaptoethanol, 0.03% bromophenol blue, pH 6.8). After boiling at 95 °C for 3 min, the samples were resolved on 12% SDS–PAGE gels (Bio-Rad, cat. no. 4568043 or cat. no. 4568046) and subsequently electroblotted onto a polyvinylidene difluoride (PVDF) membrane (Bio-Rad, cat. no. 1704156 or cat. no. 1704274), following the manufacturer’s protocol. The membranes were incubated in the blocking buffer, 5% milk (Roth, cat. no. T145.2) in TBS-T (20 mM Tris-HCl, 150 mM NaCl, and 0.1% Tween-20), then probed with the primary antibodies. The primary antibodies were diluted in antibody-dilution buffer (5% BSA, 0.1% NaN3 (Sigma, cat. no.S2002) in TBS-T) and were against GPX4 (1:1000, Abcam, cat. no. ab125066), valosin containing protein (VCP, 1:10,000, Abcam, cat. no. ab109240), HA tag (1:1,000, rat IgG1, clone 3F10, developed in-house), human FSP1 (1:1,000, Santa Cruz, cat. no. sc-377120, AMID) or human FSP1 (1:5, rat IgG2a, clone AIFM2 6D8, developed in-house), human and mouse FSP1 (1:5, rat IgG2a, clone AIFM2 1A1-1, developed in-house) and human and mouse FSP1 (1:5, rat IgG2b, clone AIFM2 14D7, developed in-house), or were diluted in 5% milk in TBS-T against horseradish-peroxidase-conjugated β-actin (1:50,000, Sigma, cat. no. A3854) overnight. After the membrane was washed once, it was probed with secondary antibodies (1:1,000–1:5,000, Cell Signaling, cat. no. 7074S for rabbit; cat. no.7076S for mouse, and 1:1,000 for anti-rat IgG1b and 2a/b, developed in-house) diluted in 5% skim milk in TBS-T. The antibody–antigen complexes were detected by the ChemiDoc Imaging System with Image Lab v6.0 (Bio-Rad). Representative images are shown after being adjusted to the appropriate brightness and angle using ImageJ/Fiji software (ver.1.53).

Construction of expression plasmids

All plasmids in this study were constructed using standard molecular biology techniques, and were verified by sequencing : human AIFM2 cDNA (NM_001198696.2, C>T:1008) and codon-optimized Mus musculus (mouse) FSP1 (NP_001034283.1), Rattus norvegicus (rat) FSP1 (NP_001132955.1) and Gallus gallus (chicken) FSP1 (XP_421597.1) were cloned or synthesized into p442-IRES-blast vectors8,17. For generating mutants or subcloning, DNA was first amplified by KOD One (Sigma, cat. no. KMM-201NV), and PCR products were purified by Wizard SV Gel&PCR Clean-up System (Promega, cat. no.A9285). Ligation reactions of PCR products with digested vectors were performed using In-Fusion cloning enzymes (Takara Bio, cat. no.639649 or 638948). Subsequently, reaction mixtures were transformed into NEB stable competent cells (NEB, cat. no.C3040H). Plasmids were isolated using the QIAprep Spin Miniprep Kit (QIAGEN, cat. no.27106), followed by sequencing.

Lentiviral production and transduction

HEK293T cells were used to produce lentiviral particles. The ecotropic envelope protein of the murine leukemia virus (MLV) was used for murine-derived cells. A third-generation lentiviral packaging system consisting of transfer plasmids, envelope plasmids (pEcoEnv-IRES-puro or pHCMV-EcoEnv (Addgene, cat. no.15802) (ecotropic particles)) and packaging plasmids ((pMDLg_pRRE and pRSV_Rev, or psPAX2 (Addgene, cat. no.12260)) were co-lipofected into HEK293T cells using transfection reagents (PEI MAX (Polysciences, cat. no. 24765) or X-tremeGENE HP agent (Roche, cat. no. 06366236001)). Cell culture supernatants containing viral particles were collected at 2–3 d post-transfection and filtered through a 0.45-µm PVDF filter (Millipore, cat. no. SLHV033RS), and were then stored at −80 °C.

Cells were seeded on 12- or 6-well plates with lentivirus supplemented with 10 µg mL–1 protamine sulfate overnight. On the next day, the cell-culture medium was replaced with fresh medium containing appropriate antibiotics, such as puromycin (Gibco, cat. no. A11138-03; 1 µg mL–1) and blasticidin (Invitrogen, cat. no. A1113903; 10 µg mL–1), and transduced cells were selected until non-transduced cells were completely dead.

Mutational screens

Mutated AIFM2 cDNA (NM_001198696.2, C>T:1008) with the adapter sequence for p442 was amplified from the virus expression vector p442-hFSP1-IRES-Blast8 by error-prone PCR using the GeneMorph II Random Mutagenesis Kits (Agilent cat. no. 200550). The optimal mutagenesis rate (3–5 DNA mutations per gene) was achieved using PCR under the following conditions: (1) 95 °C for 120 s, (2) 95 °C 30 s, (3) 60 °C 30 s, (4) 72 °C for 75 s, (5) 72°C 10 min; the cycles in steps 2–4 were repeated 25 times, and all other steps were performed once. After the isolation of PCR products using agarose electrophoresis and clean-up using the Wizard SV Gel&PCR Clean-up System, the FSP1 fragment was ligated into a digested p442-IRES-blast vector using In-Fusion enzyme (Takara, cat. no. 639649) under the following conditions: 400 ng DNA insert, 200 ng vector and 8 µL enzyme in 40 µL reaction volume, at 50 °C for 15 min. Then, the insert–vector mixture was transformed into NEB stable cells, which were incubated at 37°C for 1 h. After the induction of antibiotic-resistant genes, cells were applied to 8 plates (Thermo Fisher, cat. no. 240835) and cultured with ampicillin selection growth medium at 30 °C overnight. On the next day, colonies were collected, and DNA was isolated using the QIAGEN Plasmid Maxi Kit (Qiagen, cat. no. 12163).

As described above, a third-generation lentiviral packaging system consisting of FSP1 mutant library plasmids, envelope plasmids (pEcoEnv-IRES-puro) and packaging plasmids (pMDLg_pRRE and pRSV_Rev) were co-lipofected into HEK293T cells using transfection reagents (X-tremeGENE HP agent). Cell culture supernatants containing viral particles were collected at 2 d post-transfection and filtered through a 0.45-µm PVDF filter, and were then stored at −80°C.

Pfa1 cells were seeded into ten T-175 flasks (1.0 × 106 cells per flask, 2.0 × 107 cells in total) with medium containing 10 μg mL–1 protamine and lentivirus containing FSP1 mutants and were transduced with extremely low infection efficiency (MOI = approximately 0.1, calculated as previously described43). The next day, the medium was replaced with the fresh medium containing blasticidin (12.5 µg mL–1) and puromycin (1 µg mL–1). After selection using blasticidin for 3 d, 10 million cells were collected as a control group; at the same time, 20 million cells were seeded into five flasks (2.0 × 106 cells per flask, 2.0 × 107 cells in total) with 500 nM RSL3. After the induction of ferroptosis by RSL3 for 2 d, 10 million cells were collected as the RSL3-treated group. Then, 10 million cells per condition were treated with individual FSP1 inhibitors (5 µM for the first 5 d and 10 µM thereafter) and with 500 nM RSL3 and 1 µM 4-OH Tam so that the Pfa1 cells became fully resistant to FSP1 inhibitors while still expressing functional FSP1. Then, surviving Pfa1 Gpx4-KO cells overexpressing resistant FSP1 mutants were expanded, and 10 million cells per condition were collected and stored at −80°C.

After collecting cells from all conditions, cells were lysed in lysis buffer (50 mM Tris, 50 mM EDTA, 1% SDS, pH 8) with proteinase K (100 μg mL–1) at 55 °C overnight. On the next day, RNase A (50 µg mL–1) was added and incubated for 30 min at 37 °C to digest RNA. Then, the equivalent volume of phenol:chloroform:isoamyl alcohol (25:24:1) (Roth, cat. no. A156.2) was added. The solution was briefly vortexed and centrifuged for 10 min at 16,000g to separate DNA from RNA and proteins. The top phase of the DNA-containing solution was carefully collected in the new tubes and DNA was precipitated by the 2× volumes of 75 mM NaCl in ethanol by centrifugation for 10 min at 16,000g. The pellet was then washed with 70% ethanol, followed by centrifugation. After drying the remaining ethanol, the pellet was dissolved in 200 μL TE buffer per condition and incubated at 65 ̊C for 1 h. Finally, the DNA of the AIFM2 region (approximately 1,500 base pairs) was amplified by KOD One (Sigma, Cat. no. KMM-201NV) and purified as described above.

NGS library preparation was performed using ThruPLEX DNA-Seq HV PLUS kit (Takara, cat. no. R400782) with minor optimization. After the preparation of the library, the DNA was purified using NucleoMag NGS Clean-up and Size Select (Th. Geyer, cat. no. 11833159) for subsequent NGS. NGS was performed by the core facility in Helmholtz Munich.

NGS data analysis of human AIFM2 cDNA

Paired-end sequencing was performed in different conditions using an Illumina NovaSeq 6000 instrument using ThruPLEX DNA-Seq HV PLUS kit (see Supplementary Table 1). The program FastQC (v0.11.7) (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) was applied to the resulting FASTQ files to identify sequences that were over-represented (Illumina adapter) and to exclude them from further analysis. We used the Trimmomatic V.039 tool44 with the following options (ILLUMINACLIP:TruSeq2-PE_extended.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36) to trim paired-end data (see Supplementary Table 1). Trimmed paired-end reads (original length, 151 bp) were aligned to the reference AIFM2 cDNA sequence of 1,210 bp with the Burrows-Wheeler Alignment Tool (BWA), version 0.7.17-r1188 (ref. 45). First, an index was generated for the reference sequence using the command ‘bwa index’. Then, by applying the subcommand ‘mem’, BWA outputs the final alignment in the SAM (Sequence Alignment/Map) format. Aligned reads were converted to the BAM (Binary Alignment Map) format and sorted by leftmost chromosomal coordinates with the program SAMtools version 1.2 (ref. 46) using the commands ‘samtools view’ and ‘samtools sort’, respectively. From sorted BAM files, the coverage information per base for the FSP1 reference sequence was extracted by applying IGVtools of the Integrative Genomics Viewer (IGV), version 2.11.9 (ref. 47). The command ‘igvtools count’ with the options (-w 1 and–bases) was used to produce an output file in the WIG (wiggle) format for each sorted BAM file. The mutation frequency at each position (Xi) of the FSP1 reference sequence was calculated from the sum of mutated nucleotides at the corresponding position of the reference sequence in the alignment divided by the sum of all nucleotides (A,C,G,T,N) at this location. The count information per base stored in the wiggle files served as input for a custom written R script for the calculation of the mutation frequency. The tab-delimited output file of the R script contains the number of each nucleotide and deletions and insertions, as well as the sum of all nucleotides and the mutation frequency from the alignment (columns) at each position of the FSP1 sequence (rows).

$$}\,}\,(})=}}_}}(})/}}_}}(})$$

Where Xi is the position of AIFM2, Nmut the sum of mutated nucleotides at Xi and Nall is the sum of all nucleotides at Xi. The Z score was calculated using mean (µ) and standard deviation (σ) as follows:

$$}=\left[}-}}_}3\,}\,}1}}\right]/}}_}3\,}\,}1}}$$

$$_}3}=}(}}_}3}/}}}_}})$$

$$}}_}1}}=}\,(}}}_}1}}/}}}_}3})$$

$$\begin}}_}3}\\=}\,}\,(}}}_}3}/}}}_}})\end$$

$$\begin}}_}1}}\\=}\,}\,(}}}_}1}}/}}}_}3})\end$$

Where iFSP1s represents iFSP1 or viFSP1, and ctrl is the control. Then, amino acid substitutions were investigated in light of the possible DNA alternation using the codon table (Supplementary Table 1).

Screening of mouse FSP1 inhibitors

Pfa1 and Pfa1 Gpx4-KO cells stably overexpressing murine FSP1 were seeded on separate 384-well plates (500 cells per well), and screened with a library of small-molecule inhibitor compounds, as reported previously8. Viability of the different cell lines was assessed 48 h after treatment using AquaBluer. Compounds showing selective lethality in Pfa1 Gpx4-KO cells stably overexpressing murine FSP1 were then validated in cell viability and in vitro FSP1 enzymatic assays, as described above.

CRISPR–Cas9-mediated gene knockout

Single guide RNAs (sgRNA) vectors for human GPX4, mouse Gpx4 and mouse Fsp1 were established, and KO cells were generated as previously reported10,17. SW620 GPX4-KO cells were transiently co-transfected with the desired sgRNAs expressing lentiCRISPRv2-blast or lentiCRISPRv2-puro using the X-tremeGENE HP. One day after transfection, cells were selected with puromycin (1 µg mL–1) and blasticidin (10 µg mL–1) until non-transfected cells were dead. Single-cell clones were isolated in serial dilutions, and KO clones were confirmed by immunoblotting.

Stable expression of genes by transfection

4T1 Gpx4-KO cells and B16F10 Gpx4-KO Fsp1-KO (DKO) cells were transfected with 141-IRES-puro, 141-hFSP1-IRES-puro,141-mFsp1-IRES-puro, 141-mGpx4-IRES-puro or 141-mGpx4 U46C-IRES-puro vectors using the X-tremeGENE HP agent. One day after transfection, cells were selected and cultured in the presence of puromycin (1 µg mL–1) and absence of Lip-1 to select for stable FSP1- or GPX4-expressing cells.

In silico modeling and structural analysis

A predicted human FSP1 structure was obtained from the AlphaFold2 database (https://alphafold.ebi.ac.uk)28. To yield the superimposed structure of FSP1 with its cofactor flavin adenine dinucleotide (FAD), NADH and ubiquinone, the structure of the FSP1 ortholog NDH-2 (PDB: 4G73 ref. 29 or 5NA1 ref. 48) was aligned to FSP1 using Pymol v2.5.2 (Schrödinger), and the positions of FAD, NADH and ubiquinone were extracted and embedded into FSP1 structure19 or another NDH-2 (PDB: 4NWZ ref. 49). The in silico modeling for iFSP1 and viFSP1 were conducted by modeling software SeeSAR v12.1 (BioSoveIT)19. After visual inspection, the most viable poses were selected, and then docking molecules were exported as PDB files. Depictions of all docking or superimposed structures were made using Pymol.

Protein alignment of FSP1 orthologues

The human FSP1 sequence and its orthologs were obtained from UniProt (https://www.uniprot.org), NCBI (https://www.ncbi.nlm.nih.gov/gene/) and the PDB (https://www.rcsb.org), then aligned and visualized using JalView50 (v2.11.2.6).

Genome DNA extraction and sequencing

Genomic DNA from SKOV3 was extracted using DNAzol (Fisher Scientific, cat. no.15413379), according to the manufacturer’s instructions. Sequencing was performed by Eurofins genomics.

Statistics and reproducibility

All data shown are the mean ± s.e.m. or mean ± s.d., and the numbers (n) in each figure legend represents biological replicates or technical replicates. All experiments (except for the mutational analysis) were performed independently at least twice. One-way or two-way ANOVA followed by Dunnett’s or Tukey’s multiple-comparisons test were performed using GraphPad Prism 9 (GraphPad Prism) (see the figure legends for more details). The results of the statistical analyses are represented in each figure. P < 0.05 was considered statistically significant. No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments or outcome assessment.

Reporting summary

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