Designed endocytosis-inducing proteins degrade targets and amplify signals

Computational design of sortilin minibinders as Sort_EndoTags

Using a Rosetta-based binder design protocol¹⁵, 21,000 binders were generated to each of 2 sites on the sortilin. The epitope of site1 comprises five amino acids (UniProt numbering: F92, V93, T546, T559 and T561), chosen since it provided a modest patch of exposed hydrophobicity while avoiding any sites of known interactions. The selected epitope has the added feature that a binder to this location would be pH-dependent, since this region undergoes considerable structural change at low pH¹⁸. As previously described¹⁵, we used a set of scaffold libraries to generate several million docks to each of the sites. As in the protocol, 100,000 docks were sub-selected and sequence was designed. Helical motifs were extracted, and 3,000 designs were selected, grafted and subjected to further design. Designs were filtered based on their Rosetta ddG and ContactMolecularSurface to the hydrophobic residues listed above. This resulted in 42,000 designs that were tested experimentally. The final designed sequences for Sort_EndoTag are provided in Supplementary Table 2.

Computational design of IGF2R and ASGPR minibinders

The minibinders against IGF2R domain 6 and domain 11 were computationally designed via a Rosetta-based approach as described¹⁵. In brief, the structures of IGF2R domain 6 (Preotein Data Bank (PDB) 6UM2) and IGF2R domain 11 (PDB 1GP0) were refined using Rosetta Fastrelax with coordinate constraints. The residues at the IGF2-binding site for each domain were selected as ‘hotspot’ residues. Helical protein scaffolds were docked against the hotspot residues via the Patchdock followed by the Rifdock protocol. After sequence optimization with Rosetta FastDesign and filtering with Rosetta interface metrics including ddg and contact_molecular_surface, the top candidates were then resamplered with Rosetta Motifgraft⁴⁴ and FastDesign. Candidates passing previous filters were then filtered again with exposed hydrophobicity (sap_score) and optimized with a net-charge of −7. The final designed sequences for IGF_EndoTag are provided in Supplementary Table 1.

The minibinders against ASGPR were designed with a Rosetta-based approach integrated with ProteinMPNN and AlphaFold2. The crystal structure of ASGPR (PDB 5JQ1) was refined and helical protein scaffolds were docked against the exposed hydrophobic residues via Patchdock followed by Rifdock. The sequences were optimized with protein-MPNN and interface scores were calculated with Rosetta Fastrelax. The models were then predicted by AlphaFold2 and scored after Fastrelax. Designs with pae_interaction<10 and relaxed_ddg < −40 were selected for resampling with another round of protein-MPNN prediction followed by Rosetta Fastrelax. After final round filtering with pae_interaction, relaxed_ddg and sap_score, the sequences were further optimized to have a net-charge of −7. The final designed sequences for AS_EndoTag are provided in Supplementary Table 4.

Computational design of IGF_EndoTags

To generate flexible IGF2R agonists, all combinations of GS linkers with various lengths linking D6mb and D11mb were modelled with AlphaFold2²⁵. The designs with poor monomer plddt (plddt < 85) were dropped.

To generate rigid IGF_EndoTags, the major binding helix from D11mb or the native IGF2-binding helix, and two interface helices from D6mb were extracted. Crystal structures obtained for D6mb and D11mb in complex with IGF2 and IGF2R were used as starting points for design. Domains 6 and 11 of the complex structures were aligned with the respective domains of IGF2R in the putative receptor internalizing conformation available in the Protein Data Bank (PDB: 6UM2). In this orientation, the two interface helices from D6mb and the single interface helix from D11mb were extracted and used as motifs to scaffold by protein inpainting²⁴. To increase the likelihood of design success, the D11mb structure was adjusted to form an ideal three helical bundle with the two domain 6 helices. Protein inpainting was implemented such that the interacting residues within 3 Å of the receptor maintained the same identity as in the original minibinders. To increase design diversity, the domain 11 helix motif was randomly perturbed by rigid-body translations (up to 5 Å) and rotations (up to 10 radians) for each design prior to inpainting a scaffold between the motifs. The best inpainting outputs were selected by RosettaFold LDDT metrics (>0.5) for the inpainted region and used for sequence design with ProteinMPNN. ProteinMPNN sequence design was performed on the inpainted outputs in their desired complex orientation (with both domains 6 and 11 present) while fixing the original minibinder identities of interface residues (D6mb: R4, V8, Q11, D15, V20, K24, M25, I27, I31 and E34; D11mb: M1, A4, L7, L8 and W11). After 2,000 sequences were generated for each ProteinMPNN input, designs were filtered by predicted Rosetta ddG. AlphaFold2 structure predictions of the designed sequences were filtered by the pLDDT metric (keeping those with pLDDT > 90), and designs with a sub-angstrom backbone atom root mean squared deviation to the original design models realigned to D6mb and D11mb crystal structures (in complex with the IGF2–M6PR target domains). Finally, the complexes were assessed by Rosetta FastRelax. Designs with ddG metrics less than −40 and spatial aggregation propensity scores less than 35 were selected for expression and experimental assays.

N-linked glycan verification of epitope

To verify the epitope of the designed binders, an N-linked glycan scan was performed. This was performed to rapidly determine if the computational designed binder was interacting with the chosen interface⁵⁰. Four engineered N-linked glycan variants (NN-0975, NN-0979, NN-0981 and NN-0977) with mutation close to the Sort_EndoTag-binding site were designed and expressed. For design, the computational models were used as a starting point for the computational screen. All positions 10 Å away from the interface were screened using RosettaMatch⁵¹ followed by a design step to introduce the NXS/T motif into the protein. Computational models were minimized and filtered based on geometrical restraints, CST-score <5. Next, the four variants were used as bait in the yeast display assay against the computational designed binder, Sort_EndoTag, which was displayed on the surface of yeast.

Yeast surface display screening with FACS

The yeast surface display screening was performed as described^15,17. In brief, DNAs encoding the minibinder sequences were transformed into EBY-100 yeast strain. The yeast cells were grown in CTUG medium and induced in SGCAA medium. After washing with PBSF (PBS + 1% BSA), the cells were incubated with 1 μM biotinylated target proteins (IGF2R, ASGPR or sortilin) together with Streptavidin–phycoerythrin (SAPE, Thermo Fisher, 1:100) and anti-Myc fluorescein isothiocyanate (FITC, Miltenyi Biotech, 6.8:100) for 30 min. After washing twice with PBSF, the yeast cells were then resuspended in PBSF and screened via FACS. Only cells with PE and FITC double-positive signals were sorted for next-round screening. After another round of enrichment, the cells were titrated with biotinylated target protein at different concentrations for 30 min, washed, and further stained with both Streptavidin–phycoerythrin (SAPE, Thermo Fisher) and anti-Myc fluorescein isothiocyanate (FITC, Miltenyi Biotech) at 1:100 ratio for 30 min. After washing twice with PBSF, the yeast cells at different concentrations were sorted individually via FACS and regrown for 2 days. Next the cells from each subpool were lysated and their sequences were determined next-generation sequencing or MiSeq. FACS data were collected with the Sony SH800 software suite.

For N-linked glycan verification, yeast cells displaying Sort_EndoTag were incubated with 100nM N-glycan variants of sortilin (NN-0975, NN-0979, NN-0981 and NN-0977), separately. The percentage of yeast cells located within the pre-set gate was calculated for each N-glycan variants group and compared with the wild-type sortilin group.

Biolayer interferometry

The binding affinity for the minibinders were determined using an Octet RED96 (ForteBio). To measure the binding affinity, Streptavidin-coated biosensors (ForteBio) were first loaded with biotinylated target proteins at 50–100 nM concentration, washed with Octet buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.05% surfactant P20 and 1% BSA), and incubated with titrated concentrations of corresponding binders. To measure the off rate (K_off), the biosensors were then dipped back into the Octet buffer. The on rate (K_on), K_off and K_d were further estimated with the Octet Analysis software.

Protein production and purification

Minibinders and minibinder fusions were expressed in E. coli BL21 as previously described¹. In brief, the DNA fragments encoding the design sequences were assembled into PET-29 vectors via Gibson assembly and further transformed into BL21 strain with heat-shock. Protein expression was induced by the autoinduction system and proteins were purified with Immobilized metal affinity chromatography (IMAC) approach. Next the elutions were purified by FPLC SEC using Superdex 75 10/300 GL column (GE Healthcare). Protein concentrations were determined by NanoDrop (Thermo Scientific) and normalized by extinction coefficients.

Antibody–EndoTag fusions were produced with a mammalian expression system. Light chain of CTX/ATZ antibody and heavy chain fused with EndoTag at C-terminal constructs were ordered in CMVR from Genscript. Antibody–EndoTag fusions were then expressed via transient co-transfection of the EndoTag-heavy and light chains into Expi293F cells (Life Technologies) via PEI-MAX (Polyscience). In brief, 800 ml cultures of Expi293F cells were transfected at a density of 3 × 10⁶ cells per millilitre of culture using 1 μg plasmid DNA and 3 μg PEI per millilitre of culture. These cultures were grown in Expi293F expression medium (Life Technologies) at 37 °C in a humidified, 8% CO₂ incubator rotating at 125 rpm.

After 6 days of expression, culture supernatants were harvested via 5 min of centrifugation at 4,000g, 5 min of incubation with PDADMAC solution (Sigma Aldrich) added to a final concentration of 0.0375%, followed by an additional 5 min of centrifugation at 4,000g. Supernatants were clarified via 0.22-μm vacuum filtration and then treated to a final concentration of 50 mM Tris-HCl (pH 8) and 350 mM NaCl for IMAC. Gravity IMAC was performed by batch binding the clarified supernatants with 10 ml of Ni Sepharose Excel resin (GE Healthcare). After 20–30 min of incubation, the resin bed was washed with 10 column volumes of 20 mM Tris-HCl (pH 8), 300 mM NaCl solution. The proteins were then eluted with 3 column volumes of 20 mM Tris-HCl (pH 8), 300 mM NaCl, 300 mM imidazole solution. The batch bind process was then repeated with half the amount of resin (5 ml) and the eluates from both batch binds were combined. SDS–PAGE was performed on the IMAC eluates to assess purity.

The purified antibody–EndoTag fusions were subsequently concentrated in a 10 K MWCO Amicon Ultra centrifugal filter unit (Millipore) and polished via SEC using a Hiload 26/600 Superose 200 column (GE Healthcare) in DPBS (Gibco). The SEC fractions were re-concentrated in the same manner as before to a final concentration of 5 mg ml^−1. Endotoxin levels were assayed via Endosafe LAL Endotoxin tests (Charles River) and analytical SEC was performed using a Superdex 200 Increase 5/150 column (GE Healthcare) to obtain a high-resolution size profile. Pre- and post-freeze stability was assessed via UV-vis spectrophotometry as well as SDS–PAGE.

Cellular uptake evaluation and receptor degradation via flow cytometry

For cellular uptake assays using suspension cell lines (K562, Jurkat), the cells were incubated with corresponding fluorescence-labelled protein constructs at 37 °C for indicated time, then spun down at 500g for 5 min, resuspended and washed with cold PBS. After three washes, the cells were resuspended and transferred to a 96-well plate. For cellular uptake assays using adherent cell lines (U-251MG, Hep3B, HeLa and H1975), the cells were incubated with corresponding fluorescence-labelled protein constructs at 37 °C for indicated time, then washed with cold PBS for three times. The cells were then treated with 50 μl trypsin and incubated at 37 °C for 10 min followed by adding 50 μl DMEM. The resuspended cells were then transferred to a 96-well plate followed by 2 PBS washes. Flow cytometry was then performed in Attune NxT flow cytometer (Thermo Fisher). The data were analysed in FlowJo v9 software.

For cell surface receptor-degradation experiments, the cells were first incubated with corresponding protein reagents for indicated time at 37 °C, then washed with cold PBS 3 times. For suspension cell lines, the cells were resuspended and transferred to the 96-well plate; for adherent cell lines, the cells were first treated with trypsin for 10 min then transferred to the 96-well plate. The cells were then stained with corresponding fluorescence-labelled antibodies against the corresponding receptor for 1 h at room temperature. After washing three times with cold PBS for flow cytometry, flow cytometry was performed in Attune NxT flow cytometer (Thermo Fisher). The data were analysed in FlowJo v9 software. Representative gating strategy for flow cytometry is provided in Supplementary Figs. 1–17.

Monitoring protein degradation via western blot

Cells were cultured in T75 flasks at 37 °C in a 5% CO₂ atmosphere. HEP3B (ATCC), HeLa (ATCC), and MDA-MB-231 were cultured in DMEM supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 1% penicillin/streptomycin. Jurkat-CTLA4 (Promega, JA3001) and H1975 were cultured in RPMI supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 1% penicillin/streptomycin. Adherent cells were plated (100,000 cells per well in a 24-well plate) one day before the experiment, whereas suspension cells were plated on the day of the treatment. Cells were incubated with 250 µl of complete growth media with pLYTAC or untreated controls for indicated time. Cells were then washed with PBS 3 times and lysed with RIPA buffer supplemented with protease inhibitor cocktail (Roche), 0.1% Benzonase (Millipore-Sigma), and phosphatase inhibitor cocktail (Roche) on ice for 30 min. The cells were scraped, transferred to Eppendorf tubes, and spun down at 21,000g for 15 min at 4 °C. The supernatant was collected and the protein concentration was determined by BCA assay (Pierce). Equal amounts of lysates were loaded onto 4–12% Bis-Tris gel and separated by SDS–PAGE. Then, the gel was transferred onto a nitrocellulose membrane and stained with REVERT Total Protein Stain (LI-COR), then blocked with Odyssey Blocking Buffer (TBS) (LI-COR) for 1 h at room temperature. The membrane was incubated with primary antibodies (rabbit anti-EGFR D38B1 Cell Signaling Technologies, rabbit anti-HER2 2242 Cell Signaling Technologies, rabbit anti-PD-L1 E1L3N Cell Signaling Technologies, rabbit anti-CTAL4 E1V6T Cell Signaling Technologies, mouse anti-vinculin V284 Bio-Rad) overnight at 4 °C, washed 3 times with TBST. Subsequently, the membrane was incubated with secondary antibody (800CW goat anti-mouse or goat anti-rabbit LI-COR 926-32211) for 1 h at room temperature, and washed 3 times with TBST for visualization with an Odyssey CLx Imager (LI-COR). Image Studio (LI-COR) was used to quantify band intensities. Full scans of western blot gels are provided in Supplementary Figs. 1–16.

Fluorescence imaging

Wild-type HeLa (ATCC CCL-2) were cultured at 37 °C with 5% CO₂ in flasks with Dulbecco’s modified Eagle medium (DMEM) (Gibco) supplemented with 1 mM l-glutamine (Gibco), 4.5 g l^−1 d-glucose (Gibco), 10% fetal bovine serum (FBS) (Hyclone) and 1% penicillin-streptomycin (PenStrep) (Gibco). To passage, cells were dissociated using 0.05% trypsin EDTA (Gibco) and split 1:5 or 1:10 into a new tissue culture-treated T75 flask (Thermo Scientific ref 156499).

For imaging 35-mm glass bottom dishes were seeded at a density of 20,000 cells per dish. A final monomeric concentration of 100 nM of ligands were incubated with cultured cells. Cells were fixed 4% paraformaldehyde, permeabilized with 100% methanol, and blocked with PBS + 1% BSA. Cells were immunostained with LAMP2A antibody (Abcam ab18528) followed by goat anti-rabbit IgG Alexa Fluor 488 secondary antibody (Thermo Fisher A-11034) and 4′,6-diamidino-2-phenylindole (DAPI) (Thermo Fisher D1306) and stored in the dark at 4 °C until imaging.

Cells were washed twice with HBSS and subsequently imaged in HBSS in the dark at 37 °C. Right before imaging, cells were incubated with 25 µM DTZ. Epifluorescence imaging was conducted on a Yokogawa CSU-X1 microscope equipped with a Hamamatsu ORCA-Fusion scientific CMOS camera and Lumencor Celesta light engine. Objectives used were: 10×, NA 0.45, WD 4.0 mm, 20×, NA 1.4, WD 0.13 mm, and 40×, NA 0.95, WD 0.17–0.25 mm with correction collar for cover glass thickness (0.11 mm to 0.23 mm) (Plan Apochromat Lambda). All epifluorescence experiments were subsequently analysed using NIS Elements software.

Generation of knockout lines

IGF2R-knockout HeLa cells were a generous gift form S. Banik. SORT1 and TfR KO cells were generated using Gene Knockout Kit v2 (Synthego) using the manufacturer’s protocols.

Confocal microscopy

Indicated cells were seeded in 18-well glass bottom µ-Slides (Ibidi, 81817) at a density of 15,000 cells per well. Fluorescently labelled ligands were incubated with the cultured cells for 0.25, 3, 6 or 24 h. Thirty minutes before image acquisition, cells were additionally incubated with LysoTracker (Thermo Fisher Scientific, L7528, L7526, L12492) was added for 30 min. Fluorescently labelled anti-IGF2R (Novus Biological, NB300-514AF647) was added for 30 min. Cells were washed 3× in PBS and immediately proceeded to imaging.

Confocal laser scanning microscopy was performed on a Nikon A1R HD25 system equipped with a LU-N4 laser unit (Lasers used: 488 nm, 561 nm, 640 nm). Data were acquired using a 20×, NA 0.75, WD 1.00 mm air objective (Plan Apochromat Lambda) in combination with 1 multialkaline (EM 650 LP) and 2 GaAsP detectors (DM 560 LP EM 524/42 (503-545) and DM 652 EM 600/45 (578-623)). Acquisition was controlled via NIS Elements software and data were analysed via Fiji and custom-written Python scripts.

Mass spectrometry and proteomics

Cell pellets were thawed on ice and lysed in a lysis buffer (400 μl, 1 tablet of Pierce EDTA-free Protease Inhibitor Tablets dissolved in 50 ml of PBS) using a probe sonicator (3× 3 pulses). Protein concentration was adjusted to 2.0 mg ml^−1 and the samples (100 μl, 200 μg protein) were transferred to new Eppendorf tubes (1.5 ml) containing urea (48 mg per tube, final urea concentration: 8 M). DTT (5 μl, 200 mM fresh stock in H₂O, final DTT concentration: 10 mM) was then added to the tubes and the samples were incubated at 65 °C for 15 min. Following this incubation, iodoacetamide (5 μl, 400 mM fresh stock in H₂O, final iodoacetamide concentration: 20 mM) was added and the samples were incubated in the dark at 37 °C with shaking for 30 min. Ice-cold methanol (600 μl), CHCl₃ (200 μl), and H₂O (500 μl) were then added, and the mixture was vortexed and centrifuged (10,000g, 10 min, 4 °C) to afford a protein disc at the interface between CHCl₃ and aqueous layers. The top layer was aspirated without perturbing the disk, additional methanol (600 μl) was added, and the proteins were pelleted (10,000g, 10 min, 4 °C) and used in the next step or stored at −80 °C overnight.

The resulting protein pellets were resuspended in EPPS buffer (160 μl, 200 mM, pH 8) using probe sonicator (3× 3 pulses). Trypsin (10 μl, 0.5 μg μl^−1 in trypsin reconstitute buffer) and CaCl₂ (1.8 μl, 100 mM in H₂O) were added and the samples were incubated at 37 °C with shaking overnight.

Peptide concentration was determined using the microBCA assay (Thermo Scientific) according to the manufacturer’s instructions. For each sample, a volume corresponding to 25 μg of peptides was transferred to a new Eppendorf tube and the total volume was brought up to 35 μl with EPPS buffer (200 mM, pH 8). The samples were diluted with CH₃CN (9 μl) and incubated with the corresponding TMT tags (3 μl per channel, 20 μg μl^−1) at room temperature for 30 min. An additional TMT tag (3 μl per channel, 20 μg μl^−1, 30 min) was added and the samples were incubated for another 30 min. Labeling was quenched by the addition of hydroxylamine (6 μl, 5% in H₂O). Following a 15 min incubation at room temperature, formic acid was added (2.5 μl, final formic acid concentration: 5%). Twenty microlitres of labelled peptides for each channel were combined into a 2.0 ml low-binding Eppendorf tube, and 25 μl of 20% formic acid was added. The resulting mixture was lyophilized to remove the solvents before high pH fractionation.

The spin columns from Pierce High pH Reversed-Phase Peptide Fractionation Kit were pre-equilibrated prior to use. In brief, the columns were placed in Eppendorf tubes (2 ml), spun down to remove the storage solution (5,000g, 2 min), and washed with CH₃CN (2× 300 μl, 5,000g, 2 min) and buffer A (2× 300 μl, 95% H₂O, 5% CH₃CN, 0.1% formic acid, 5,000g, 2 min). TMT-labelled peptides were re-dissolved in buffer A (300 μl, 95% H₂O, 5% CH₃CN, 0.1% formic acid) and loaded onto pre-equilibrated spin columns for high pH fractionation. The columns were spun down (2,000g, 2 min) and the flow through was used to wash the original Eppendorf tube and passed through the spin column again (2,000g, 2 min). The column was then washed with buffer A (300 μl, 2,000g, 2 min) and 10 mM aqueous NH₄HCO₃ containing 5% CH₃CN (300 μl, 2,000g, 2 min), and the flow through was discarded. The peptides were eluted from the spin column into fresh Eppendorf tubes (2.0 ml) with a series of 10 mM NH₄HCO₃/CH₃CN buffers (2,000g, 2 min). The following buffers were used for peptide elution (CH₃CN (%)): 7.5, 10, 12.5, 15, 17.5, 20, 22.5, 25, 27.5, 30, 32.5, 35, 37.5, 40, 42.5, 45, 47.5, 50, 52.5, 55, 57.5, 60, 62.5, 65, 67.5, 70, 72.5, 75, 80 and 95. Every tenth fraction was combined into a new clean Eppendorf tube (2 ml) and the solvent was removed using a benchtop lyophilizer and stored at −20 °C before analysis.

The resulting 10 combined fractions were resuspended in buffer A (25 μl) and analysed on the Orbitrap Fusion mass-spectrometer (4 μl injection volume) coupled to a Thermo Scientific EASY-nLC 1200 LC system and autosampler. The peptides were eluted onto a capillary column (75 μm inner diameter fused silica, packed with C18) and separated at a flow rate of 0.3 μl/min^−1 using the following gradient: 5% buffer B in buffer A from 0–10 min, 5%–35% buffer B from 10–129 min, 35%–100% buffer B from 129–130 min, 100% buffer B from 130–139 min, 100%–5% buffer B from 139–140 min, and 5% buffer B from 140–150 min (buffer A: 100% H₂O, 0.1% formic acid; buffer B: 20% H₂O, 80% CH₃CN, 0.1% formic acid). Data were acquired using an MS3-based TMT method. In brief, the scan sequence began with an MS1 master scan (Orbitrap analysis, resolution 120,000, 375 − 1,600 m/z, cycle time 3 s) with dynamic exclusion enabled (repeat count 1, duration 30 s). The top precursors were then selected for MS2/MS3 analysis. MS2 analysis consisted of: quadrupole isolation (isolation window was set to 1.2 for charge state z = 2; 0.7 for charge state z = 3; 0.5 for charge states z = 4–6) of precursor ion followed by collision-induced dissociation (CID) in the ion trap (normalized collision energy 35%, maximum injection time 50 ms, MS2 resolution was set to turbo). Following the acquisition of each MS2 spectrum, synchronous precursor selection (SPS) enabled the selection of MS2 fragment ions for MS3 analysis (SPS isolation window was set to 1.3 for charge state z = 2; 0.7 for charge state z = 3; 0.5 for charge states z = 4–6). MS3 precursors were fragmented by HCD and analysed using the Orbitrap (collision energy 65%, maximum injection time 120 ms). The raw files were converted to mzML files using the MSConvert tool from ProteoWizard (version 3.0.22088). A reverse concatenated, non-redundant variant of the Human UniProt database (29 November 2022) was searched using FragPipe (version 18.0) with the built-in TMT10-MS3 workflow. The virtual references were used for the data sets due to the lack of a pooled sample. The quantified proteins were filtered with false discovery rate < 1% with median centreing normalization. Data are presented as the mean fold change to DMSO-treated controls. n = 3 per group. P values were calculated by a two-tailed unpaired t-test with Welch’s correction.

IgG and LHDB supernatant clearance assays

Jurkat or K562 cells seeded in 96-well culture plates in 300 μl medium were incubated with AF647-conjugated IgG (Novusbio) or LHDB alone or together with protein G-EndoTag reagents. At various timepoints, the cells were pelleted down and 30 μl of supernatants were extracted and further diluted to 45 μl by using a PBS buffer. After shaking in an orbital shaker for 5 min, the fluorescence intensity was measured using a Neo2 plate reader (BioTek) at wavelength 647 nm. The percentage clearance was measured by normalizing the control group without adding protein G-EndoTag reagent.

SEC binding assay

ASGPR protein (28.4 kDa) at 1 μM (diluted in PBS) was incubated with 3 μM AS_EndoTag-3C for 30 min and run through an ÄKTA SEC protein purification system using a S200 16/90 column. The absorbance at 230 nm was used as a readout for binding. The SEC traces of the complex was compared to the traces of individual ASGPR or AS_EndoTag-3C at same concentration.

In vivo mouse study

Mouse lymphoma cell line A20 cell was purchased from American Type Culture Collection (ATCC, TIB-208). The cell was cultured in RPMI-1640 medium (Gibco, Thermo Scientific, 31870074), supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco, Thermo Scientific), 1× GlutaMAX (Gibco, Thermo Scientific), 1× penicillin/streptomycin solution (Gibco, Thermo Scientific). The cell was cultured at 37 °C in humidified condition with 5% CO₂.

All animal experiments were conducted at the Instituto de Medicina Molecular João Lobo Antunes (IMM), Lisbon. Animal work was performed in strict accordance with Portuguese Law (Portaria 1005/92) and the European Guideline 86/609/EEC and follow the Federation of European Laboratory Animal Science Associations guidelines and recommendations concerning laboratory animal welfare. All animal experiments were approved by the Portuguese official veterinary department for welfare licensing (Direção Geral de Alimentação e Veterinária) and the IMM Animal Ethics Committee (authorization AWB_2021_03_GB_Targ CancerDrugs). Eight-week-old female BALB/c mice (purchased from Charles River) were used in this study, with 5 × 10⁶ A20 cells inoculated subcutaneously in the flank. Tumour growth was monitored over time, by performing bilateral vernier caliper measurements every day and mean tumour volumes were calculated using the formula (length × width²)/2. Treatments were initiated when tumours reached approximately 100 mm³ (approximately 10 days after tumour induction), with the mice been randomly assigned to receive ATZ, ATZ–IGF_EndoTag1 or ATZ–IGF_EndoTag4 and isotype as controls (n = 6 mice per group). Treatments were administered intratumourally in a total of three injections for every three days. Animals were monitored every day; tumours were measured as described before and mouse weight was evaluated throughout the study. Animals were killed whenever reaching humane endpoints: loss of 20% of body weight, breathing impairment, or poor reaction to external stimuli. No signs of animal suffering or discomfort were observed during the experiment. For efficacy study, once control (isotype-treated mice) tumours reached 1,000 mm³, all mice were killed (by isoflurane overdose), and the tumours were removed for western blot analysis. For survival monitor, each mouse was killed respectively when tumours reached 1,000 mm³. The light/dark cycle was 14 h light/10 h dark (lights on at 07:00; lights off at 21:00). The temperature was 20–24 °C and the relative humidity was 55 ± 10%, with controlled supply of HEPA-filtered air provided to individually ventilated cages. Maximum number of animals per cage was five. Social isolation was avoided whenever possible. The type of food was autoclaved diet pellets RM3A (P), from SDS Special Diets Services (801030). Food was placed in a grid inside the cage and provided ad libitum to animals. The type of water was sterile water treated by reverse osmosis. Water was provided ad libitum to animals through bottles with a capillary hole. The data collected was analysed using GraphPad Prism9.

In vivo PD-L1 degradation of tumour samples by western blot

Tumour samples isolated from the mice were homogenized, lysed in RIPA buffer containing protease inhibitor (Roche), phosphatase inhibitors (Sigma) and 0.1% Benzonase (Sigma) on ice for 30 min. The lysates were spun at 21,000g for 15 min at 4 °C. The supernatant was collected, and the protein concentrations were quantified using BCA assay (Sigma). Fifty micrograms of protein were loaded per lane and separated on 12% SDS–PAGE gels, and then transferred onto polyvinylidene difluoride (PVDF) membranes (GE Healthcare). Membranes were then blocked with 5% BSA in TBS supplemented with 0.1% Tween-20 (TBST) for 1 h at room temperature, and then probed with following specific primary antibodies at 4 °C overnight. After three times of washing with TBST, secondary antibodies were added to the membrane for 1 h at room temperature. All membranes were washed three times and exposed using ECL substrate (Bio-Rad, 170–5060) and Amersham 800 Imaging System (Cytiva). The primary antibodies used included PD-L1 (sc-518027) and beta-actin (sc-47778), the secondary antibody was goat anti-mouse IgG H&L (HRP) (Abcam, ab205719). The intensities of the bands were quantified by ImageJ.

Statistical analysis

No statistical analysis was used to determine the sample size. The sample size was determined by our ability to detect meaningful differences between treatments. Western blot experiments in vitro and BLI binding assays were conducted with sample size of one based on low variance from previous experience. Western blot experiments in mice were performed with sample size of 3 to reduce the variance of protein level across animals from our previous best practice. The data were collected as biological replicates as indicated in the figure legend. All cell experiments were done multiple times to ensure reproducibility. All images were representative of three independently replicated samples. Statistical analyses are specified in figure legends.

Reporting summary

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