Reducing affinity as a strategy to boost immunomodulatory antibody agonism

Mice

hCD40Tg mice were provided by R. Noelle (King’s College, London)¹³. hCD40Tg FcγR-null mice (hCD40Tg;Fcer1g^−/−;FcγR2b^−/−) were generated as previously described¹⁵ by first breeding Fcer1g^−/− and FcγR2b^−/− mice to generate homozygous FcγR-null mice (Fcer1g^−/− × FcγR2b^−/−). Homozygous FcγR-null mice were subsequently crossed with hCD40Tg mice¹⁵. All mice were bred in-house and maintained on a 12 h light–dark cycle, and food and water were provided ad libitum. The temperature was maintained at 20–24 °C with 55 ± 15% humidity. Mice were checked daily to ensure healthy status. Both male and female mice were used between 8–24 weeks old, mice were randomly divided into groups based on age and sex for each experiment for which blinding was not performed. All experiments were conducted under UK Home Office licence numbers PB24EEE31, P4D9C89EA, P540CBA98 and P39FE2AA7 and according to local ethics committee guidelines.

Human samples

Human PBMCs were derived from blood cones collected from healthy donors through Southampton National Blood Services with previous informed consent. The use of human blood was approved by the East of Scotland Research Ethics Service, Tayside, UK.

Cell lines

WT Ramos, CHO-k1, EG7, A20 and Jurkat cells were obtained from the American Type Culture Collection. IIA1.6 cells were used as previously described³⁵. The WT Jurkat NF-κB–GFP reporter cell line was from System Biosciences. The WT Jurkat NFAT–Luc reporter cell line was from Promega. All cell lines were maintained in a humidified incubator at 37 °C and 5% CO₂, and cultured in RPMI medium supplemented with 10% heat-inactivated FBS, 2 mM l-glutamine, 1 mM pyruvate, 100 U ml^–1 penicillin, 100 µg ml^–1 streptomycin and 50 µM β-mercaptoethanol (complete RPMI medium, all from ThermoFisher), with the exception of CHO-k1 cells, which were cultured without β-mercaptoethanol.

Generation of affinity mutants

Models of Fab–receptor complexes were obtained from the Protein Data Bank (PDB) with the following accession codes: 6FAX for ChiLob 7/4–CD40; 6MI2 for utomilumab–4-1BB; and 5WT9 for nivolumab–PD-1. To identify potential interacting residues, the models were analysed in PISA using the QtPISA interface^36,37. Potential mutants were generated by mutating the potential interacting residues identified by PISA to alanine in PyMOL 2.5.2. using the mutation wizard. Each mutant was subsequently analysed in PISA. The resulting difference in the ΔG and binding energy values from the WT structure were recorded and used to predict effects on binding affinity. To generate double mutants, a confusion matrix of the difference in ΔG and binding energy values from WT was used to give combined mutation scores. Decisions on which mutants to produce were based on the scores in the confusion matrix and the proximity of the residue to the binding interface. Humanization was performed, and resulting sequences were provided by Glycotope.

Antibodies and reagents

All antibodies were produced using the ExpiCHO system as previously described¹⁵. The variable domain sequence for utomilumab³⁸, nivolumab³⁹, varlilumab⁴⁰, TGN1412 (ref. ⁴¹) and OKT3 (ref. ⁴²) were derived from published sequences. The light and heavy chain variable domain sequences were synthesized by GeneArt and subcloned into pEE12.4 and pEE6.4 expression vectors, respectively (Lonza), encoding the constant domain of different IgG isotypes. The CDR mutations for ChiLob 7/4, utomilumab and nivolumab were achieved by site-directed mutagenesis using a QuickChange Site-Directed Mutagenesis kit (Agilent). Plasmids encoding the heavy and light chains were transiently transfected into ExpiCHO cells for 10 days before the supernatant was collected and antibodies were purified on a MabSelect SuRe column (GE Healthcare). All antibody preparations were checked by HPLC to contain <1% aggregate and by Endosafe-PTS portable tests (Charles River Laboratories) to contain <5 endotoxin units per mg antibody.

The DNA constructs containing CD40–GFP and 4-1BB–GFP were generated as previously described¹⁷. The PD-1–GFP construct was generated by subcloning human PD-1 (NCBI GenBank accession: U64863) into a pcDNA3 plasmid containing the GFP fragment at the carboxy terminus. The OKT3–scFv–CD8α DNA construct was designed as VL-(G4S)3-VH-CD8α with an EcoRI restriction site between the VH site and CD8α and cloned into a pCIpuro vector. The variable domain sequence of the anti-human IgG Fc monoclonal antibody SB2H2 was obtained from the hybridoma by sequencing the cDNA generated using the MMLV reverse transcriptase and a universal primer as previously described⁴³. The SB2H2–scFv–CD8α DNA construct was designed as VL-(G4S)3-VH-CD8α with an EcoRI restriction site between the VH site and CD8α and cloned into a pcDNA3 vector. Human PD-L1 (NCBI GenBank accession: NM_014143) was cloned into a pcDNA3 vector.

Recombinant soluble trimeric CD40L and 4-1BBL were produced in-house as previously described¹⁷. Fab fragments of mouse IgG1 anti-CD40 monoclonal antibody was generated using a Pierce Mouse IgG1 Fab and F(ab′)₂ Preparation kit (ThermoFisher). The Fab fragment of human IgG4 anti-4-1BB monoclonal antibody was generated using immobilized papain (ThermoFisher).

SPR analysis

A Biacore T200 instrument was used throughout. Recombinant extracellular domains of human CD40 (R&D Systems), 4-1BB (R&D Systems), PD-1 (R&D systems), TNFRII (R&D systems), CD27 (in-house) and CD28 (BioLegend) were immobilized onto a CM5 chip by amine-coupling chemistry. To compare the binding affinity of various target-specific monoclonal antibody mutants and natural soluble ligands towards their respective receptor or homologue, the monoclonal antibody and ligand were injected through the flow cell at 250, 50, 10, 2, 0.4 and 0 nM in HBS-EP⁺ running buffer at a flow rate of 30 ml min^–1, allowing 300 s for association and 300 s for dissociation. Data were collected using Biacore T200 control software. The sensorgrams were fitted using the bivalent analyte model, and the k_a and k_d values were calculated using Biacore Bioevaluation software; the K_D values were calculated as k_d/k_a. All SPR reagents and software were from GE Healthcare.

Assessment of antibody cell surface receptor binding

To assess the level of antibody binding to cells expressing human CD40, 4-1BB or PD-1, relevant cells were incubated with various concentrations of anti-CD40, anti-4-1BB or anti-PD-1 monoclonal antibodies as indicated in the figure legends for 30 min at 4 °C. Unbound monoclonal antibodies were then washed off using FACS wash buffer (PBS, 1% BSA and 0.01% sodium azide), and PE-conjugated polyclonal goat F(ab′)₂ secondary anti-human Fc (1:200) or PE-conjugated polyclonal goat F(ab′)₂ secondary anti-mouse Fc (1:200, both from Abcam) were added for 30 min at 4 °C. Unbound monoclonal antibody was then washed off using FACS wash buffer. To detect the level of bound Fab fragment, FITC-conjugated anti-mouse IgG Fab (1:100) or AF647-conjugated anti-human IgGκ light chain (1:100) was used depending on the monoclonal antibody isotype. The level of bound monoclonal antibody was quantified by flow cytometry.

Competitive cell surface receptor binding

For competitive cell-binding assays, Ramos cells were incubated with a fixed concentration of AF647-labelled ChiLob 7/4 hIgG1 (0.5 μg ml^–1) and various concentrations of competing ChiLob 7/4 mIgG1 or ChiLob 7/4 hIgG2 affinity mutants for 30 min. Cells were then washed, and the level of AF647-labelled ChiLob 7/4 hIgG1 remaining bound to the cell surface was quantified by flow cytometry.

Flow cytometry

Flow cytometry experiments were conducted using FACSCalibur, FACSCanto II or FACSMelody instruments (all from BD Biosciences). Flow cytometry data were collected using BD CellQuest and BD FACSDIVA software, and data analysis was performed using FCS Express software v.3 (De Novo Software) or Flowjo (BD Biosciences).

B cell activation assay

Human B cells were purified from human PBMCs using a MojoSort Human B Cell Isolation kit (BioLegend). Mouse B cells were purified from splenocytes using a MojoSort Mouse Pan B Cell Isolation kit (BioLegend). Purified B cells were incubated with various anti-CD40 monoclonal antibodies as indicated in the figure legends for 2 days and imaged using a conventional light microscope (Olympus CKX41). CD23 (anti-CD23, 1:160) and CD86 (anti-CD86, 1:100) expression was assessed by flow cytometry. To assess B cell proliferation, ³H-thymidine (Perkin Elmer) was added at 1 µCi per well on day 3 for an additional 18 h as previously described¹⁵.

OTI expansion assay

OTI expansion assays were performed as previously described¹⁵. To assess the ability of anti-CD40 monoclonal antibodies to induce OTI T cell expansion, 1 × 10⁵ OTI cells were intravenously injected into CD40KOTg mice 1 day before the intravenous injection of 100 µg OVA in combination with 25 µg or 500 µg of various anti-CD40 monoclonal antibodies. Mice were then bled 4–5 days later as indicated in the figure legends, and the level of OTI expansion was assessed on the basis of the proportion of CD8⁺ SIINFEKL tetramer-positive cells by flow cytometry.

EG7 tumour therapy

The EG7 model was generated as previously described¹⁵. In brief, mice were intravenously inoculated with 5 × 10⁵ EG7 cells and then treated with 25 µg anti-CD40 monoclonal antibody and 100 µg OVA intravenously when the sum of tumour length and width reached approximately 10 mm. Tumour size was measured three times per week using digital calipers, and mice were culled when the sum of the tumour length and width reached 30 mm or when the general health of the mice reached humane end-point criteria. For re-challenge, tumour-free mice were intravenously inoculated with 5 × 10⁵ EG7 cells and monitored for tumour growth as described above. Tumour volume was calculated using the following formula: V = (W² × L)/2, where W is the tumour width and L is the tumour length.

Human DC activation and mixed leukocyte reaction

Human immature DCs were generated as previously described⁴⁴. In brief, CD14⁺ monocytes were isolated from human PBMCs using a magnetic negative selection kit (Miltenyi Biotech) and then cultured in the presence of 500 IU ml^–1 IL-4 and 1,000 IU ml^–1 GM-CSF (both cytokines produced in-house) for 5–6 days. The identity of DCs was confirmed by CD11c (anti-CD11c, 1:20) and DC-SIGN (anti-CD209, 1:20) expression. For direct stimulation, immature DCs were treated with 50 µg ml^–1 anti-CD40 monoclonal antibody for 2 days, and the level of CD86 expression (anti-CD86, 1:20) was quantified by flow cytometry. A mixed leukocyte reaction was performed as previously described¹⁵. In brief, varying numbers of immature DCs were first treated with 50 µg ml^–1 anti-CD40 monoclonal antibody for 2 days and then washed and further incubated with 0.1 × 10⁶ purified human allogeneic CD4⁺ T cells (MojoSort Human CD4 T Cell Isolation kit, BioLegend) for 4 days. ³H-thymidine was added at 1 µCi per well on day 4 for an additional 18 h to assess T cell proliferation.

Human PBMC peptide recall assay

The expansion of antigen-specific T cells within PBMCs was achieved using a CEFX Ultra SuperStim Pool (JPT Peptide Technologies), which contains a pool of 176 known peptides based on different infectious agents that have been shown to induce antigen-specific T cell expansion^45,46. In brief, fresh human PBMCs were labelled with 2 µM CFSE (ThermoFisher) in PBS and then 0.2 × 10⁶ PBMCs were incubated with 50 µg ml^–1 anti-CD40 monoclonal antibody and 0.6 µM CEFX Ultra SuperStim Pool for 5 days to recall antigen-responsive CD8⁺ T cells. The proliferating cells expressing CD3 (anti-CD3, 1:20) and CD8 (anti-CD8, 1:20) were regarded as antigen-responsive cells, and their level of activation was measured on the basis of CD25 expression (anti-CD25, 1:20) on day 5 by flow cytometry.

NF-κB assay

A pCIpuro vector encoding CD40, 4-1BB (expressing the hCD40 intracellular signalling domain) or PD-1 (expressing the CD40 transmembrane and intracellular signalling domain) was transfected into Jurkat NF-κB–GFP cells, and stable clones were selected using 1 µg ml^–1 puromycin. To examine NF-κB activation, cells stably transfected with each receptor were incubated with the relevant monoclonal antibody as indicated in each legend for 6 h at 37 °C. The level of NF-κB activation was subsequently quantified by GFP fluorescence, which was assessed by flow cytometry.

Assays to evaluate the impact of receptor density on monoclonal antibody agonism

Jurkat NF-κB–GFP cells stably transfected with CD40, 4-1BB or PD-1 were first sorted into populations expressing low, medium or high levels of the respective receptor using a FACSMelody instrument (BD). Cells were then treated with various monoclonal antibodies as indicated for 6 h at 37 °C, and the level of NF-κB activation was subsequently quantified by GFP fluorescence, which was assessed by flow cytometry. Receptor quantification was performed using a Quantum Alexa Fluor 647 MESF kit (Bangs Laboratories). To quantify the level of PD-1 expression on human primary T cells, human PBMCs were activated with Immunocult (Stem Cell Technologies) for 2 days, and then CD3⁺ T cells were analysed for PD-1 expression by flow cytometry. To assess the level of 4-1BB expression on human primary CD8⁺ T cells, CD8⁺ T cells were purified from human PBMCs using a MojoSort Human CD8 T Cell Isolation kit (BioLegend) and then activated with plate-bound anti-CD3 (clone OKT3) and anti-CD28 (clone TGN1412) (both produced in-house) for 24 h before 4-1BB quantification by flow cytometry.

Assays to evaluate the effect of antibody concentration on monoclonal antibody agonism

Jurkat NF-κB–GFP cells stably transfected with CD40, 4-1BB or PD-1 were treated with 50 µg ml^–1 monoclonal antibody as indicated for 30 min at room temperature, and then excess unbound monoclonal antibody was washed off. Cells were then incubated at 37 °C for various periods as indicated, and the level of monoclonal antibody remaining bound to the cell surface was quantified by DL650-conjugated goat F(ab′)₂ secondary anti-mouse Fc (1:200) or by DL650-conjugated goat F(ab′)₂ secondary anti-human Fc (1:200, both from Abcam) using flow cytometry. The level of NF-κB activation was concurrently quantified by GFP fluorescence, which was assessed by flow cytometry.

Confocal microscopy

DNA encoding CD40ECD–GFP and 4-1BBECD–GFP were subcloned into a pCIpuro vector and transfected into Jurkat cells using Nucleofector kit V (Lonza). Stable Jurkat clones were selected using 1 µg ml^–1 puromycin. IIA1.6 cells stably transfected with full length PD-1–GFP (IIA1.6 PD-1–GFP cells) were generated by transfecting IIA1.6 cells with a pCIpuro plasmid encoding PD-1–GFP using nucleofection kit V (Lonza), and stable clones were selected using 4 µg ml^–1 puromycin. Confocal microscopy was performed as previously described¹⁷. Jurkat cells were incubated with 50 µg ml^–1 of the indicated monoclonal antibody for 3 h at 37 °C, and then fixed with cold methanol on ice for 10 min before the nucleus was stained with DAPI (ThermoFisher). Alternatively, cells were fixed with 2% paraformaldehyde (ThermoFisher) at room temperature for 10 min before the nucleus was stained with DAPI. For live-cell imaging, cells were imaged directly without fixation. Confocal images were acquired using a Leica SP8 confocal microscope, and data were analysed using Leica Application Suite X (all from Leica). To measure receptor clustering at the cell–cell junctions in relation to the periphery of the cells, a clustering index was calculated (Extended Data Fig. 7a). Confocal images through the centre of the cells were opened in Leica Application Suite X software (Leica), and fluorescence intensity measurements were taken for regions of interest (ROIs) at the cell–cell junctions or at the periphery of the cells (non-contacting membrane). These were determined by eye, with cell periphery measurements taken for each corresponding cell–cell junction; that is, two cell periphery ROIs would be collected for a cell that contained two cell–cell cluster ROIs. A clustering index was calculated for five confocal images, with up to five cell–cell clusters per image for each treatment. An average clustering index was calculated from five confocal images, with up to five cell–cell clusters per image for each treatment. The clustering index designates the ratio of fluorescence intensity at the cell–cell junction over the fluorescence intensity at the cell periphery, which was calculated using Leica Application Suite X software (Leica). A larger clustering index denotes higher levels of receptor clustering. The circularity of the cells was measured in ImageJ, whereby a circularity value of 1.0 indicates a perfect circle (circularity = 4π(area/perimeter²). Confocal images through the centre of the cells were opened in ImageJ, a ROI was manually drawn around individual cells following the membrane and the circularity measured. Cell circularity was measured for five confocal images for each treatment, and the results of three independent experiments were pooled.

dSTORM

IBIDI glass-bottom chambers were first coated with poly-d-lysine (Sigma). Jurkat cells expressing CD40ECD–GFP were incubated with 25 µg ml^–1 anti-CD40 or CD40L at 37 °C for 1 h and then washed with PBS and fixed with 4% paraformaldehyde. GFP was detected using AF647-conjugated anti-GFP nanobodies (Proteintech Europe, 1:500) following the manufacturer’s instructions. TCEP STORM buffer comprises three solutions (A, B and C). Solution A contains 1 µg ml^–1 catalase, 0.2 mM TCEP, 2.5% glycerol, 1.25 mM KCl, 1 mM Tris-HCl and 50 µg ml^–1 glucose oxidase. Solution B contains 40 mg ml^–1 glucose and 4% glycerol. Solution C contains 0.1 M MEA-HCl. Immediately before dSTORM collection, the TCEP STORM buffer solutions A (50 μl), B (400 μl), C (100 μl) and PBS (450 μl) were mixed and then added to the well. A wide-field fluorescence reference image was acquired before dSTORM images (10,000 frames, 30 ms of exposure) were collected using an ONI Nanoimager equipped with a 640 nm laser and NimOS1.6 software (ONI). Analyses of dSTORM data were carried out using the CODI cloud analysis platform (beta version, ONI). Images were subjected to drift correction and filtering before ROIs were drawn around the cell–cell junctions. For CD40L, ROIs were also drawn around large clusters present outside the cell–cell junctions. Localizations within the ROIs were identified and grouped into subclusters using HDBSCAN81. The following features were extracted for each individual subcluster: number of localizations; density (localizations/area); and area (computed from the convex hull of the cluster).

In vitro assessment of receptor internalization

The level of CD40 and PD-1 internalization was quantified using a fluorescence quenching assay as previously described⁴⁷. To assess CD40 internalization, AF488-labelled anti-CD40 monoclonal antibody or AF488-labelled anti-CD20 rituximab hIgG2 were added to Ramos cells as indicated for 10, 30, 60, 120 or 180 min at 4 °C or 37 °C. To assess PD-1 internalization, AF488-labelled anti-PD-1 monoclonal antibody or the anti-CD3 OKT3 hIgG1 pre-opsonized with AF488-labelled anti-human IgG Fc monoclonal antibody SB2H2 (produced in-house) were added to Jurkat NFAT–Luc–PD-1 cells as indicated for 10, 30, 60, 120 or 180 min at 4 °C or 37 °C. Subsequently, Ramos or Jurkat NFAT–Luc–PD-1 cells were washed, and half the cells treated with anti-AF488 antibody (ThermoFisher, 1:100) at 4 °C that quenches AF488 fluorescence. The remaining unquenched AF488 fluorescence analysed by flow cytometry correlates to internalized CD40 or PD-1. The percentage of total expression quantifies remaining cell-surface-bound receptor and was calculated as a percentage (unquenched fluorescence – quenched fluorescence)/(unquenched fluorescence).

ADCP

Ramos cells stably transfected with 4-1BBECD-Tm (Ramos 4-1BB cells) were generated by transfecting Ramos cells with pcDNA3 plasmid encoding 4-1BBECD-Tm using nucleofection kit V (Lonza), and stable clones were selected using 1 mg ml^–1 geneticin (Extended Data Fig. 10c). ADCP was performed as previously described¹⁷, using Ramos 4-1BB cells as target cells and human monocyte-derived macrophages (hMDMs) as the effector cells. The hMDMs were derived by culturing monocytes in the presence of 100 ng ml^–1 M-CSF (in-house) for 6 days. The day before the phagocytosis assay, 1 × 10⁵ hMDMs were plated onto a 96-well flat-bottom plate (ThermoFisher). The next day, target Ramos 4-1BB cells were labelled with CFSE followed by opsonization with various anti-4-1BB monoclonal antibodies as indicated in the figure legends for 30 min at 4 °C. In total, 5 × 10⁵ target cells were then added to each well and incubated at 37 °C for 30 min for phagocytosis to occur. The samples were subsequently stained with anti-CD14-APC (1:20) to identify hMDMs, and cells positive for both CFSE and CD14 as assessed by flow cytometry were classified as hMDMs that had undergone phagocytosis. The percentage of ADCP was calculated as follows: (CFSE⁺CD14⁺ cells)/(total CD14⁺ cells) × 100.

ADCC

IIA1.6 cells stably transfected with full-length human 4-1BB (IIA1.6-4 1BB cells) were generated by transfecting IIA1.6 cells with pCIpuro plasmid encoding human 4-1BB using nucleofection kit V (Lonza), and stable clones were selected using 4 µg ml^–1 puromycin (Extended Data Fig. 10d). ADCC was performed as previously described⁴⁸, using IIA1.6-4 1BB cells as targets and human PBMCs as effector cells. In brief, target cells were labelled with calcein-AM (ThermoFisher), added to wells of a 96-well round-bottom plate (ThermoFisher), at 8 × 10⁵ per well and then incubated with various anti-4-1BB monoclonal antibodies as specified in the figure legends for 30 min at 4 °C. In total, 4 × 10⁶ effector cells were then added to each well and incubated at 37 °C for 4 h before cells were centrifuged and the supernatant quantified for calcein-AM fluorescence using a Varioskan Flash plate reader (ThermoFisher). The control wells for maximal lysis contained 4% Triton X-100 (Sigma). The level of ADCC was expressed as the percentage of maximal lysis = (experimental fluorescence – background fluorescence)/(maximal lysis – background fluorescence) × 100.

PD-1 blockade assay

Jurkat NFAT–Luc–PD-1 cells were generated by transfecting WT Jurkat NFAT–Luc cells (Promega) with pcDNA3 plasmid encoding human PD-1 using nucleofection kit V (Lonza), and stable clones were selected using 1 mg ml^–1 geneticin. CHO OKT3–scFv–CD8α–PD-L1 cells were generated by transfecting CHO-k1 cells with pCIpuro vector encoding OKT3–scFv–CD8α and pcDNA3 vector encoding human PD-L1 and selecting stable clones using 10 µg ml^–1 puromycin and 1 mg ml^–1 geneticin. To assess the ability of anti-PD-1 monoclonal antibodies to block the PD-1–PD-L1 interaction, 5 × 10⁴ CHO OKT3–scFv–CD8α–PD-L1 cells were plated onto wells of sterile white opaque 96-well microplates (Perkin Elmer) overnight. The next day, 5 × 10⁴ Jurkat NFAT–Luc–PD-1 cells were added to each well along with the various anti-PD-1 monoclonal antibodies as indicated in each figure legend for 6 h before ONE-Glo reagent (Promega) was added. Luciferase activity was read using a Varioskan Flash plate reader (ThermoFisher).

PD-1-mediated T cell suppression assay

CHO-k1 cells stably transfected with SB2H2–scFv–CD8α (CHO SB2H2–scFv–CD8α cells) were generated by transfecting WT CHO-k1 cells with pcDNA3 plasmid encoding SB2H2–scFv–CD8α using GenePorter (Amsbio). Stable clones were selected using 1 mg ml^–1 geneticin. For the PD-1-mediated T cell suppression assay, CHO SB2H2–scFv–CD8α cells were first incubated with 20 µg ml^–1 nivolumab variants and 5 µg ml^–1 OKT3 hIgG1 for 30 min before excess unbound monoclonal antibody was washed off using complete RPMI medium. Jurkat NFAT–Luc–PD-1 cells were then added to the opsonized CHO SB2H2–scFv–CD8α cells and incubated for 6 h. CD69 expression (anti-CD69, 1:20) on Jurkat NFAT–Luc–PD-1 cells was measured by flow cytometry, and the level of NFAT activation was quantified by measuring the luciferase activity using ONE-Glo Reagent (Promega) as described above.

Statistics and reproducibility

Data analysis was performed using GraphPad Prism 9.2.0 (GraphPad Software). Two-tailed, non-paired Student t-test was used for pairwise comparisons. One-way analysis of variance followed by Kruskal–Wallis test was used for multiple comparisons as specified in the figure legends. Throughout, *P < 0.05, **P < 0.01, ***P < 0.001 and NS, not significant. Reproducibility, including technical replicates and independent biological experiments, is stated in each figure legend.

Reporting summary

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