Protein designBackbone generation

We explored two approaches to generate scaffolds with β-sheets with open slots for peptide β-strand insertion (Fig. 1 and Supplementary Fig. 1) using blueprint-based backbone building in Rosetta and PyRosetta4 (refs. 43,44,45,46,47). First, we explored using a two-domain binder approach (Supplementary Fig. 1a). We started from a scaffold, 2003285_0000, designed by FoldIt players29 (domain 1) and generated a β-sheet that extends from the C-terminal strand of the scaffold using blueprint-based backbone generation43,45. Then, this sheet was expanded into a second mixed α/β domain with three strands and one helix or four strands and two helices. The central strand of the β-sheet that encompasses both domains was split off from generating an individual peptide in β-strand conformation that can bind the designed deep cleft between domain 1 and domain 2. A connecting loop linking the helices that make up the interdomain interface was next generated using loop closure48 to yield a single-polypeptide two-domain binder that clamps the peptide on either side through β-strand backbone bonds (Supplementary Fig. 1a). The same approach was followed to generate β-hairpin-binding scaffolds.

In the second approach, a different FoldIt scaffold, 2003333_0006 (ref. 29), was modified to function as a peptide binder (Supplementary Fig. 1b). The connection between β-strands 3 and 4 was removed to create the individual peptide component. To stabilize the modified binder and ensure its solubility in the absence of peptide, we designed buttressing secondary structure elements that support the binding interface and scaffold. β-strand 3 was paired with another antiparallel strand, whereas helices 1 and 2 were backed up by either one or two supporting helices. After backbone generation, Rosetta combinatorial sequence design calculations were used to optimize the sequences of both the scaffold and the peptide for high-affinity binding. Designs with favorable interaction energy, few unsatisfied buried polar atoms and high shape complementarity, and for which Rosetta folding simulations yielded models close to the designed model, were selected for experimental characterization.

Sequence design

The amino acid sequences of the newly built polyvaline backbones were optimized using Rosetta flexible backbone enabled combinatorial side chain design followed by a second design round for the peptide-binder interface32,49. Ref2015, beta_nov16 or beta_genpot scorefunctions was used during design50. For a subset of designs, buried polar hydrogen-bond networks were designed using the HBNet mover30.

The affinity between peptide and binder was computationally improved by introducing hydrophobic interaction pairs to the solvent-exposed side of the interface. All solvent-exposed interaction pairs for which the Cα atoms were within 6 Å of each other were selected and allowed to be redesigned with the PackRotamersMover to only phenylalanine, alanine, methionine, isoleucine, leucine, tyrosine, valine and tryptophan using a fixed backbone. For computational affinity optimization of the natural target peptides, all surface exposed residues on only the binder within 6 Å of the target hydrophobic side chain were allowed to be redesigned. Residues around the redesigned interaction pairs were repacked. Single redesigned pairs and combinations of pairs were selected for experimental characterization.

To facilitate crystallization, the surface residues outside the interface were redesigned using ProteinMPNN36 for design C104. The structures of sequences obtained from ProteinMPNN were predicted using AlphaFold2 (ref. 51), and designs with rmsd ≤ 1.5 and plDDT ≥ 85 to the original designed model were selected for experimental characterization.

Design of rigid helical fusions

Rigid fusion of peptide binders and components of the LHD hetero-oligomer system was performed as described previously28,52.

Matching natural peptide sequences to scaffolds

The protein sequences of amyloid precursor protein, microtubule-associated protein tau, transthyretin and serum amyloid A1 were searched for burial patterns that were also present in the peptides of designs C34, C37, C104 and CH15. For C104, both the designed model and the crystal structure of C104, minimized with FastRelax53, were used. The burial patterns representing the relative positions of solvent-inaccessible residues versus solvent-accessible residues in the designed peptides were identified by visual inspection. For each peptide, all amyloidogenic protein sequence frames of length n, where n is the number of residues in the designed peptides mentioned above, were scanned for matching regions. Only residues phenylalanine, alanine, methionine, isoleucine, leucine, valine, serine, threonine, tyrosine or glycine residues were allowed at the solvent-inaccessible positions. At the remaining positions, all residues were allowed except for proline, which was only allowed at either terminus. The sequences against which matches were searched were DAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVVIA for Aβ42, RSFFSFLGEAFDGARDMWRAYSDMREANYIGSDKYFHARGNYDAAKRGPGGVWAAEAISDARENIQRFFGHGAEDSLADQAANEWGRSGKDPNHFRPAGLPEKY for SAA1, PGGGKVQIINKKLDLSNVQSKCGSKDNIKHVPGGGSVQIVYKPVDLSKVTSKCGSLGNIHHKPGGGQVEVKSEKLDFKDRVQSKIGSLDNITHVPGGGNKKIETHKLTFRENAKAKTDHGAEIVYK for tau and GPTGTGESKCPLMVKVLDAVRGSPAINVAVHVFRKAADDTWEPFASGKTSESGELHGLTTEEEFVEGIYKVEIDTKSYWKALGISPFHEHAEVVFTANDSGPRRYTIAALLSPYSYSTTAVVTNPKE for transthyretin. For matching sequence stretches, we verified whether the matching sequence was also participating in β-strand interaction in amyloid fibrils based on published cryo-electron microscopy structures of amyloid fibrils (see main text for references). When this was the case, the sequence of the template designed peptide was mutated to the sequence of the matched sequence of the amyloidogenic protein. The resulting peptide−binder complex was minimized, and the residues in the interface of the designed binder were redesigned to optimally match the amyloidogenic sequence by also including hydrophobic interaction pairs across the solvent-accessible area of the interface (see above). The Protein Data Bank (PDB) models of the designed proteins and example scripts can be downloaded as source data.

Protein expression and purification

Synthetic genes encoding designed proteins were purchased from Genscript or Integrated DNA Technologies (IDT) in the pET29b expression vector or as eBlocks (IDT) and cloned into customized expression vectors54 using GoldenGate cloning. A His6x tag was included either at the N terminus or at the C terminus as part of the expression vector. In some cases, a tobacco etch virus (TEV) protease recognition site was introduced at the N terminus after the histidine tag. Peptide genes were purchased as fusion proteins to either the C terminus of sfGFP or the N terminus of a ubiquitin−AviTag−His6x construct separated by a Pro-Ala-Ser linker. Bicistronic genes were ordered as described28. Detailed construct information is provided in the Supplementary Data 1.

Proteins were expressed using autoinducing medium consisting of TBII medium (Mpbio) supplemented with 50× 5052, 20 mM MgSO4 and trace metal mix in BL21 LEMO E. coli cells. Proteins were expressed under antibiotic selection at 37 °C overnight or at 18–25 °C overnight after initial growth for 6–8 h at 37 °C. Cells were harvested by centrifugation at 4,000g and resuspended in lysis buffer (100 mM Tris pH 8.0, 200 mM NaCl, 50 mM imidazole pH 8.0) containing protease inhibitors (Thermo Scientific) and bovine pancreas DNase I (Sigma-Aldrich) before lysis by sonication. The reducing agent TCEP (1 mM final concentration) was included in the lysis buffer for designs with free cysteines. Proteins were purified by IMAC. Cleared lysates were incubated with 2–4 ml nickel NTA beads (Qiagen) for 20–40 min before washing the beads with 5–10 column volumes of lysis buffer, 5–10 column volumes of high-salt buffer (10 mM Tris pH 8.0, 1 M NaCl) and 5–10 column volumes of lysis buffer. Proteins were eluted with 10 ml of elution buffer (20 mM Tris pH 8.0, 100 mM NaCl, 500 mM imidazole pH 8.0). His6x tags were cleaved by dialyzing IMAC elutions against 20 mM Tris pH 8.0, 100 mM NaCl, 1 mM TCEP overnight in the presence of His6x-tagged TEV protease followed by a second IMAC column to remove His6x−TEV and uncleaved protein.

Single-cysteine variants of DAm12, DAm14 and DAm15 were purified as described above and labeled with Alexa488-C5-maleimide (Thermo) at a concentration of between 50 and 100 μM of protein and a twofold to fivefold molar excess of label in SEC buffer supplemented with 1 mM TCEP protected from light. After 3 h at room temperature or overnight at 4 °C, the labeling reaction was quenched by the addition of 1 M dithiothreitol (DTT).

As a final step, all protein preparations were polished using SEC on either Superdex 200 Increase 10/300GL or Superdex 75 Increase 10/300GL columns (Cytiva) using 20 mM Tris pH 8.0, 100 mM NaCl. The reducing agent TCEP was included (1 mM final concentration) for designs with free cysteines. For designs where a substantial void volume peak was present in addition to the monomer peak, the monomer peak was pooled and reinjected. Only designs where, upon reinjection, the void peak was mostly absent were further pursued. SDS−PAGE and LC−MS were used to verify peak fractions. Proteins were concentrated to concentrations between 0.5–10 mg ml−1 and stored at room temperature or flash frozen in liquid nitrogen for storage at −80 °C. Thawing of flash-frozen aliquots was done at room temperature or 37 °C. All purification steps from IMAC were performed at ambient room temperature.

The C104.1 complex was prepared by incubating the binder with a threefold to fivefold molar excess of the peptide for 3 h at room temperature followed by SEC.

Peptide synthesis

All Fmoc-protected amino acids were purchased from P3 Bio. The biotinylated peptides obtained by synthesis were padded at the C terminus with SGGSGG-Kbiotin, where Kbiotin is a Fmoc-Lys(biotin)-OH building block also purchased from P3 Bio. Oxyma was purchased from CEM, and DIC was purchased from Oakwood Chemicals. Dimethylformamide was purchased from Fisher Scientific and treated with an AldraAmine trapping pack (Sigma-Aldrich) before use. Piperidine was purchased from Sigma-Aldrich. Cl-TCP(Cl) resins were purchased from CEM. The peptides were synthesized on a 0.1 mmol scale using microwave-assisted solid-phase peptide synthesis via a CEM Liberty Blue system and subsequently cleaved with a cleavage cocktail consisting of trifluoroacetic acid (TFA), TIPS, water and DODT (92.5:2.5:2.5:2.5 in order). The cleavage solution was concentrated in vacuo, precipitated into cold ether and spun down by centrifugation. The pellet was washed and spun down again with ether (2⨯) and then dried under nitrogen, resuspended in water and aceyonitrile (ACN) and purified by RP-HPLC on an Agilent 1260 Infinity semi-prep system with a gradient from 20% to 70% over 15 min (A: H2O with 0.1% TFA; B: ACN with 0.1% TFA). The purified peptide fractions were combined into one, lyophilized and massed in a tared scintillation vial for the final product. Peptides derived from transthyretin, tau and serum amyloid A1 were purchased from WuXi. Depending on the isoelectric point, lyophilized peptides were solubilized in buffers containing either 100 mM Tris pH 8.0 or 100 mM MES pH 6.5 and stored at −20 °C.

Mammalian cell culture and transfection

HeLa cells (ATCC, CCL-2) were cultured in DMEM (Gibco) supplemented with 1 mM l-glutamine (Gibco), 4.5 g l−1d-glucose (Gibco), 10% FBS and (1×) nonessential amino acids (Gibco). Cells were kept in culture at 37 °C and 5% CO2 and split twice per week by trypsinization using 0.05% trypsin EDTA (Gibco) followed by passage at 1:5 or 1:10 into a new tissue culture-treated T75 flask (Thermo Scientific, 156499). Before transfection, cells were plated at 20,000 cells per well on CELLview cell culture slides (Greiner Bio-One, 543079) for 24 h, after which transfection took place using 187.5 ng of total DNA per well and 1 μg μl−1 PEI-MAX (Polyscience) mixed with Opti-MEM medium (Gibco). Transfected cells were incubated at 37 °C and 5% CO2 for 24 to 36 h before being imaged.

Fluorescence microscopy

Three-dimensional images were acquired with a commercial OMX-SR system (GE Healthcare) using a 488-nm Toptica diode laser for excitation. Emission was collected on a PCO.edge sCMOS camera using an Olympus ×60 1.42-NA PlanApochromat oil immersion lens. Images of 1,024 × 1,024 (pixel size, 6.5 μm) were captured without binning. AcquireSR acquisition control software was used for data collection. z stacks were collected with a step size of 500 nm and 15 slices per image. The images were deconvolved with an enhanced ratio using SoftWoRx 7.0.0 (GE Healthcare). Finally, cell images were sum-projected using ImageJ2 v2.1.0. and v2.3.0. Scale bars equal 10 µm.

Biolayer interferometry

BLI experiments were performed on an OctetRED96 BLI system (ForteBio) at room temperature in Octet buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.05% surfactant P20) supplemented with 1 mg ml−1 BSA (Sigma-Aldrich). Before taking measurements, streptavidin-coated biosensors were first equilibrated for at least 10 min in Octet buffer. Chemically synthesized peptides with C-terminal biotin or enzymatically biotinylated peptide fusion proteins (see Supplementary Tables 6 and 7 for details) were immobilized on the biosensors by dipping them into a solution with 100−500 nM protein until the response reached between 10% and 50% of the maximum value followed by dipping the sensors into fresh Octet buffer to establish a baseline for 60 s. Titration experiments were performed at 25 °C while rotating at 1,000g. Association with designs was allowed by dipping the biosensors in solutions containing designed protein diluted in Octet buffer until equilibrium was approached, followed by dissociation by dipping the biosensors into fresh solution and monitoring the dissociation kinetics. In the peptide-binding cross-specificity assays, each biotinylated peptide was loaded onto streptavidin biosensors in equal amounts followed by 2 min of baseline equilibration. Then, association and dissociation with all the different binders was allowed for 400 s for each step. For the designed peptide−binder pairs, binder concentrations were around the Kd of the interaction between the loaded peptide and its designed binding partner, whereas the concentrations for the amyloid binders were 10, 2.5 and 0.625 μM. Global kinetic or steady-state fits were performed on buffer-subtracted data using the manufacturer’s software (Data Analysis 9.1) assuming a 1:1 binding model. Data acquisition was performed using OctetRed96 data acquisition software 9.

Enzymatic biotinylation of proteins

Proteins with Avi tags (GLNDIFEAQKIEWHE; Supplementary Tables 6 and 7) were purified as described above and biotinylated in vitro using the BirA500 (Avidity) biotinylation kit. Protein (840 μl) from an IMAC elution was biotinylated in a 1,200-μl (final volume) reaction according to the manufacturer’s instructions. Biotinylation reactions were allowed to proceed at either 4 °C overnight or for 2−3 h at room temperature on a rotating platform. Biotinylated proteins were purified using SEC on a Superdex 200 column (Increase 10/300 GL, GE Healthcare) or an S75 Increase 10/300 GL column (GE Healthcare) using SEC buffer (20 mM Tris pH 8.0, 100 mM NaCl).

Circular dichroism spectroscopy

Circular dichroism spectra were recorded in a cuvette with a 1-mm path length at a protein concentration between 0.3–0.5 mg ml−1 on a J-1500 instrument (Jasco). For temperature melts, data were recorded at 222 nm every 2 °C between 4 and 94 °C and wavelength scans were done between 190 and 260 nm at 10 °C intervals starting from 4 °C. Experiments were performed in 20 mM Tris pH8.0, 20 mM NaCl. The high-tension voltage was monitored according to the manufacturer’s recommendation to ensure optimal signal-to-noise ratio for the wavelengths of interest.

SEC binding assays

SEC binding assays between purified designs and GFP−peptide fusions were performed on a Superdex 75 Increase 10/300 GL column (Cytiva) in 20 mM Tris pH 8.0, 100 mM NaCl using 500-μl injections containing a 15 or 20 μM final concentration of each component. Binding reactions were allowed to equilibrate for at least 45 min before injection. For the subunit exchange experiment, the disulfide-stabilized complex between C104.2 and ubiquitin−pep104.2 as well as the control base noncovalent complex was allowed to form overnight at a 20 μM equimolar concentration under oxidizing conditions, after which competing GFP−pep104 was added to the preformed complexes to a final concentration of 20 μM. After at least 45 min, the reaction was injected onto an SEC column. Elution profiles were collected by monitoring absorbance at 230 nm and 395 nm (absorbance of GFP). All experiments were performed at room temperature. Data were analyzed and acquired using the manufacturer’s Unicorn 7.3 software.

Disulfide formation assay

Individual protein components were purified as described above in the presence of 1 mM TCEP except for in the last SEC step, where no reducing agent was present. Reactions were incubated at room temperature using 50 μM of each component in 20 mM Tris pH 8.0, 100 mM NaCl. Reactions were stopped by adding an equal volume of 2⨯ nonreducing SDS protein-loading buffer at the indicated time points.

NMR

All NMR experiments for C34 were performed on Bruker Avance III HD 14.1 T or 18.8 T spectrometers equipped with cryogenically cooled x,y,z pulse-field gradient triple resonance probes. Resonance assignments were obtained by triple resonance (HB)CBCA(CO)NNH, HNCACB, HNCO, HN(CA)CO and HNN experiments55 acquired using U-13C,15N-labeled samples. Note that the spectra shown in Fig. 3a were recorded at 25 °C, but the resonance assignment for free C34 was done at 50 °C to reduce the line broadening arising from conformational exchange. Information from 13Cα, 13Cβ, 13CO, 15N and 1HN chemical shifts was combined into a single secondary structure propensity (SSP) score representing the expected fraction of α-structure or β-structure 15N R2 rates for the bound state of C34 were measured using the in-phase Carr−Purcell−Meiboom−Gill (CPMG) experiment56 with νCPMG = 1 kHz, Trelax = 30 ms and CPMG refocusing pulses applied at a γB1/2π = 5.7 kHz field and phase-modulated according to the (x,x,y,-y) cycling scheme57. A NOESY dataset for recording intermolecular NOEs was acquired as previously described58 with a mixing time of 150 ms, using 450 μM of U-13C,15N-labeled C34 and 450 μM of unlabeled peptide at natural isotopic abundance. Data were acquired using topspin3.2 and topspin3.5. Data were analyzed using nmrPipe 11.0 and CARA 1.9.1.7.

Aβ expression, purification and labeling

Aβ42 peptide was expressed and purified as reported previously59. In short, the synthetic gene encoding NT*FlSp was purchased from Genscript (Genscript Biotech), ligated into pT7 plasmid containing a TEV recognition site (TRS) for Aβ42 (ref. 60) and transformed into chemically competent E. coli BL21 (DE3) cells and expressed as described previously61. Upon cleavage of the fusion protein with TEV protease, the sample was dissolved in 15 ml of 8 M guanidine-hydrochloride (GuHCl) and monomeric Aβ was purified on a Superdex 26/600 30-pg size exclusion column and lyophilized as aliquots until further use. To generate fibrils, several aliquots of lyophilized Aβ were combined for an increased protein concentration by dissolving aliquots in 1 ml of 8 M GuHCl and subjecting them to SEC on a Superdex 75 10/300 Increase column in 20 mM sodium phosphate, 0.2 mM EDTA buffer at pH 8.0. Subsequently, collected monomeric peptide, typically at a concentration of 30 µM, was pipetted into PEGylated plates (Corning, 3881) and incubated at 37 °C in a plate reader, with 100 rpm orbital shaking. To track the degree of monomer conversion into fibrils, ThT was added exclusively to control wells, and after the plateau was reached, the fibrils were harvested from ThT-free sample wells. To perform binding experiments of monomeric Aβ with the binders, a cysteine-carrying Aβ mutant (S8C) was expressed and purified as described previously62. Briefly, the plasmid carrying synthetic genes with E. coli optimized codons for the S8C mutant (developed by Thacker and colleagues and purchased from Genscript) were transformed into the BL21 DE3 pLysS star E. coli strain and the protein was expressed in auto-induction medium63. Upon purification using IEX and subsequent SEC on a 26 × 600 mm Superdex 75 column, the S8C monomer was eluted in sodium phosphate buffer supplemented with 3 mM DTT to prevent its dimerization and then lyophilized. For conjugation of the protein with a fluorescent dye, the lyophilized fractions were dissolved in 8 M GdnHCl and subjected to SEC in buffer without DTT before adding Alexa Fluor 488 dye (ThermoFisher) in at least 5× molar excess. The protein−dye mixture was incubated overnight at 4 °C, the free dye was removed via column chromatography and the protein was used immediately.

Kinetic assays of fibril inhibition

Aliquots of purified lyophilized Aβ were dissolved in 8 M GuHCl and the monomeric protein was isolated by gel filtration on a Superdex 75 10/300 Increase column in 20 mM sodium phosphate, 0.2 mM EDTA buffer at pH 8.0. Samples were prepared on ice, using careful pipetting to avoid the introduction of air bubbles, and pipetted into a 96-well half-area plate of PEGylated black polystyrene with a clear bottom (Corning 3881), 100 μl per well, with three to four replicates per sample. All samples diluted with buffer to the final concentration of 2 μM Aβ were supplemented with 6 μM ThT (Sigma), with a range of concentrations of the binders per experiment. The kinetic assays were initiated by placing the 96-well plate at 37 °C under quiescent conditions in a plate reader (FLUOstar Optima, BMGLabtech). The ThT fluorescence was measured through the bottom of the plate every 165 s with a 440-nm excitation filter and a 480-nm emission filter.

Analysis of aggregation kinetics

Integrated rate laws describing the aggregation of Aβ42 were derived previously64. They reproduce well the kinetic curves obtained in ThT assays and can be used to quantify inhibitory effects. Here, we used the amylofit platform39 to determine the rate constants of aggregation in the absence of an inhibitor. Using the affinities of binder to monomer determined by MDS, we then calculated the concentrations of monomer expected to be bound at each binder concentration. Assuming all monomer bound is completely removed from the aggregation reaction (that is, ignoring dissociation of the monomer−binder complex over the timescale of aggregation), the effect of binders on the aggregation reaction is the same as a lowering of the monomer concentration. The kinetic curves resulting from this effective reduction of the monomer concentration were then computed using the amylofit platform (Extended Data Fig. 9), and the effect was found to be insufficient to explain the observed degree of inhibition. We then explored whether the presence of an additional mechanism of inhibition, by interaction with aggregated species, was able to describe the observed aggregation. To model this additional inhibition, we allowed the rate of secondary nucleation to vary with binder concentration, as detailed previously39. These results are shown as solid lines in Fig. 6 and effectively describe the inhibition at substoichiometric binder concentrations. At higher binder concentrations, when the majority of monomer is expected to be bound, these fits perform less well and thus only the experimental measurements, not the fits, are shown at the highest binder concentrations.

Cell viability assay

Cell viability assays were performed on SHSY-5Y human neuroblastoma cells cultured under standard conditions at 37 °C in a humidified incubator with 5% CO2. Cells were seeded at a density of 25,000 cells per well in a white-walled, clear-bottomed 96-well plate and cultured for 24 h in DMEM supplemented with 10% FBS. The culture medium was then replaced with phenol red-free DMEM without serum, supplemented with an antibiotic-antimycotic agent. Aβ monomer was isolated by gel filtration in 20 mM sodium phosphate buffer at pH 8.0 (without EDTA), mixed with the binders at a ratio of 10 µM:20 µM Aβ to binder, and stored on ice until further use. Samples used in the treatment of the cells were prepared by incubation in a 96-well nonbinding plate (Corning, 3881) at 37 °C so that the progress of aggregation could be tracked by ThT fluorescence in the control wells. Aliquots of corresponding ThT-free samples were taken when the reaction reached t½ (when 50% of full aggregation was reached, corresponding to the highest concentration of cytotoxic oligomeric species64,65) and immediately diluted tenfold in medium and applied to cells. The cells were then cultured in the presence of the peptides or buffer for an additional 24 h before the viability assays were performed. Cell viability was measured with CellTiter 96 AQueous One MTS reagent from Promega. The MTS reagent was added to the cell culture medium and incubated with the cells at 37 °C in a humidified incubator with 5% CO2 for 1 h before the absorbance at 495 nm was measured in an Optima FLUOstar plate reader. All values given for the assay account for the positive control (2% Triton X-100) values as a baseline readout and are normalized relative to the untreated cells.

Microfluidic diffusional sizing

The binding affinity of the binders and monomeric Aβ was measured on a Fluidity One-M instrument (Fluidic Analytics). Fluorescently labeled Aβ mutant was mixed with unlabeled binders at a range of concentrations and incubated on ice for at least 30 min. Before the measurements, microfluidic circuits of the Fluidity One-M chip plate were primed using sample buffer. To create a binding curve for individual designs, each one of the different Aβ−binder mixtures was measured in triplicate. Kd values were determined by nonlinear least squares fitting as described previously66 using Prism (GraphPad Software). For MDS experiments concerning interactions of binders with Aβ fibrils, microfluidic devices were fabricated and operated as described previously67,68. In brief, the microfluidic devices were fabricated in PDMS using standard soft-lithography techniques and bonded onto a glass coverslip after activation with oxygen plasma. Sample loading from reservoirs connected to the respective inlets and control of flow rate were achieved by applying negative pressure at the outlet using a glass syringe (Hamilton) and a syringe pump (neMESYS, Cetoni). Images were recorded using a custom-built inverted epifluorescence microscope fitted with a fluorescence filter set with an excitation filter at 475 ± 35 nm, emission filter at 525 ± 30 nm and dichroic mirror for 506 nm (Laser, 2000) for detection of Alexa 488-labeled binders. Images were acquired using Micro Manager, typically at flow rates of 60 and 100 μl h−1, and lateral diffusion profiles were recorded at four different positions along the microfluidic channels. Diffusion profiles extracted from fluorescence images and confocal recordings were fitted using a custom-written analysis software by numerical model simulations solving the diffusion–advection equations for mass transport under flow69.

Crystal structure determination

The C104.1 complex (19 mg ml−1) was crystallized using the vapor diffusion method at room temperature in 0.1 M Tris pH 7.8, poly(-γ-glutamic acid) low-molecular-weight polymer, 15% PEG 4000 (Molecular Dimensions) and the crystals were harvested in 25% glycerol as a cryoprotectant. Data were collected from a single crystal at 100 K and 0.97918 Å at the Advanced Photon Source at Argonne National Laboratory. Diffraction images were integrated using XDS70 or HKL3000 (ref. 71) and merged/scaled using the AIMLESS application from the CCP4-7.0.076 software suite72. Starting phases were obtained by molecular replacement using Phaser73 from within CCP4-7.0.076, using the computational design models of the individual N-terminal and C-terminal domains of C104.1 as search models. Structures were refined using either phenix.refine74 or Refmac75 and PDB-REDO76. Model building was performed using COOT77. The lack of density at the C terminus of the peptide prompted us to examine the possibility of a β-strand register shift for peptide binding. OMIT maps were used to decrease the model bias. In addition, the peptide was modeled in several off-target β-strand registers. Overall, refinement statistics and B factors were better for the model where the peptide was modeled in the designed on-target β-strand register. The final model had 96.8% of residues in the favorable region of the Ramachandran plot and no outliers. The model was evaluated using MolProbity78. Data collection and refinement statistics are recorded in Supplementary Table 3. Data deposition, atomic coordinates and structure factors reported in this paper have been deposited in the PDB (http://www.rcsb.org/) with accession code 8FG6.

Statistics and reproducibility

Unless stated otherwise, all experimental results were reproduced at least two times with two different preparations of protein reagents. Many of the BLI binding experiments were performed three or more times with three to five protein preparations that were purified independently.

Reporting summary

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