Design of facilitated dissociation enables timing of cytokine signalling

Abstract

Protein design has focused on the design of ground states, ensuring that they are sufficiently low energy to be highly populated¹. Designing the kinetics and dynamics of a system requires, in addition, the design of excited states that are traversed in transitions from one low-lying state to another^2,3. This is a challenging task because such states must be sufficiently strained to be poorly populated, but not so strained that they are not populated at all, and because protein design methods have focused on generating near-ideal structures^4-7. Here we describe a general approach for designing systems that use an induced-fit power stroke⁸ to generate a structurally frustrated⁹ and strained excited state, allosterically driving protein complex dissociation. X-ray crystallography, double electron-electron resonance spectroscopy and kinetic binding measurements show that incorporating excited states enables the design of effector-induced increases in dissociation rates as high as 5,700-fold. We highlight the power of this approach by designing rapid biosensors, kinetically controlled circuits and cytokine mimics that can be dissociated from their receptors within seconds, enabling dissection of the temporal dynamics of interleukin-2 signalling.

Fig. 1. Strategy for designing proteins that reconfigure through facilitated dissociation.

a–c, A high-affinity interaction can rapidly exchange through facilitated dissociation (bottom pathways), but not through mutually exclusive competition (top pathways). a, Reaction diagram. b, Energy diagram. c, Schematic of induced-fit facilitated dissociation (bottom) compared with slow mutually exclusive competition (top). The host protein (H, subscripted by conformational state X or Y) is shown in blue, the target (T) in pink and the effector (E) in orange. d, Structural models of starting components (effector-responsive switch and arbitrary binder–target pair) combined to construct facilitated dissociation systems. e, Structural models of example proteins designed to undergo a facilitated dissociation process starting from a tightly interacting state X (left) through a structurally frustrated ternary intermediate in state Y (right, solid). State X (transparent) is included to show the conformational change. Thinner arrows indicate structural features and thicker arrows indicate state changes.

Fig. 2. Kinetic characterization of facilitated dissociation in AS1.

a, Slow dissociation of the target from the host in the absence of effector (solid line) and fast dissociation in the presence of 2 μM effector (dashed line) assessed by SPR. Slow dissociation data (solid grey) fitted with a double exponential (pink). b, Kinetic model describing pathways of competition. Top, mutually exclusive competition; middle, facilitated dissociation with effector binding rate-limited by conformational selection; bottom, facilitated dissociation with induced-fit effector binding. The k labels are rate constants. c, Cartoon representations of the peptide (left) and 3hb (right) effectors; interface residues are shown in grey. d, Circular dichroism spectra of the peptide and 3hb effectors. e,f, Kinetic characterization of the formation and breakage of the ternary complex intermediate with the peptide (e) and 3hb (f) effectors. Top left, fast dissociation of the target from the ternary complex; data (grey) fitted with double exponentials (orange and green). Bottom left, effector association to form the ternary complex and extremely slow subsequent dissociation; data (grey) fitted with single exponentials (colours) in the association phase. Right, apparent effector on-rate constant plotted against effector concentration (circles) and a linear (e) or hyperbolic (f) fit. k_switch is the saturating value of the hyperbolic fit. g,h, Kinetic characterization of the full facilitated dissociation pathway with the peptide (g) and 3hb (h) effectors. Left, effector-concentration-dependent dissociation of the target; data (grey) fitted (colours) as described in the Methods. Right, rate constant of facilitated target dissociation plotted against effector concentration (circles) and fitted with a hyperbolic equation (black line). In a and in the left plots of e–h, schematics show the arrangement of proteins relative to the SPR chip (grey). In the right plots of e–h, schematics show the mechanism that can be inferred from the data.

Fig. 3. Structural characterization of AS1.

a, Crystal structure of AS1 alone (grey) overlaid with the design model of AS1 in state X (blue). Inset, detailed view of side chains in the partially open effector-binding cleft. b, Cocrystal structure of AS1 and peptide effector (grey) overlaid with the design model of the AS1–effector complex in state Y (blue and orange). Inset shows the same view as in a. c, Top, cocrystal structure of AS1 (with intact cleft) and target (grey) overlaid with the design model of the target–AS1 complex in state X (blue and pink). Bottom, cocrystal structure of AS1 (with collapsed cleft) and target (grey) overlaid with the design model of the target–AS0 complex (blue and pink) whose state X resembles this collapsed state. d, Cocrystal structure of AS1 (with hydrophobic surface mutations), target and peptide effector (grey) aligned at the switch region with the design models of the target (pink) and AS1–effector complex in state Y (blue and orange) showing the designed clash. e, Top, detailed view of the target interface side chains in the ternary complex (grey) and the target–AS1 complex (pink) interacting with AS1 (blue). Bottom, detailed view of the backbone hydrogen bonding in the interfacial strand pairing. The target–AS1 complex (pink and blue) hydrogen bonds (green) are less strained than the ternary complex (grey) hydrogen bonds (red).

Fig. 4. Modulation and applications of facilitated dissociation.

a,b, Comparison of three representative designs with different facilitated dissociation kinetics. a, For each design: left, model of host in state X (blue) aligned to the target (pink) to show any clash influencing the target off-rate in the absence of effector; right, model of host–effector complex in state Y (blue and orange) aligned to the target (pink) to show the designed clash, and (grey) AF2 prediction of the target position relative to the switch in the ternary complex to show how the clash resolves through global strain. b, Forward and reverse facilitated dissociation rates: target off-rate constants versus effector concentration (orange circles) and effector off-rate constants versus target concentration (pink circles) fitted with hyperbolics (black lines). Cartoons illustrate the forward and reverse facilitated dissociation pathways. Dashed lines mark the base and accelerated off-rate constants for forward (orange) and reverse (pink) facilitated dissociation. c, Chain reactions. FRET time courses showing slow transfer of a kinetically trapped effector (blue, top schematic) and accelerated transfer through facilitated dissociation (orange, bottom schematic). d, Breaking split enzymes. Luminescence time courses showing breakage of a reversible split luciferase through slow direct competition (blue, top schematic) and faster facilitated dissociation (orange, bottom schematic). e, Rapid sensing. Luminescence time course (orange) showing rapid sensing of SARS-CoV-2 through facilitated dissociation (schematic). c–e, Data fitted with single exponentials (black lines).

Fig. 5. Characterization of a rapidly switchable IL-2 mimic.

a, Natural pathways for terminating IL-2 signalling are slow. b, Through facilitated dissociation, signalling could be rapidly terminated. c, Model of ASNeo2 binding IL-2Rβγ_c to activate signalling (left), which quickly terminates after adding effector (right). d, Left, accelerated dissociation of γ_c; data (grey) fitted (colours) as described in the Methods. Right, γ_c dissociation rate constant versus effector concentration (circles) fitted with a hyperbolic (black line). e, Relative IL-2Rβ/γ_c dimerization on the cell surface at first (grey; n = 37), after adding ASNeo2 (blue; n = 32) and after subsequently adding effector (orange; n = 33). f, Time courses of IL-2Rβ/γ_c dimerization after pre-stimulation with ASNeo2 then adding nothing (blue) or effector (orange), fitted with an exponential (black) yielding the rate constant k_app. g, Dose–response of STAT5 phosphorylation from stimulation with ASNeo2 alone (blue) or with effector (orange) (n = 1). h, Time courses of STAT5 phosphorylation after stimulation with ASNeo2 for 5 min then adding nothing (blue), effector (orange) or ruxolitinib (green) (n = 3). i–n, Human T cells were stimulated with ASNeo2 or left untreated as a control (grey). Signalling was sustained (blue) or terminated with effector (orange) after the indicated duration. i,j, Cell division (i, by carboxyfluorescein succinimidyl ester (CFSE) staining) and survival (j) three days after stimulation (n = 4); statistics from ANOVA with two-sided Tukey’s post-test. NS, not significant. k, Time courses of BCL2 expression (by quantitative PCR (qPCR); n = 3). l–n, RNA-seq analysis six hours after stimulation (n = 3). l, Principal component (PC) analysis. m, Changes in gene expression after transient stimulation. Points denote differentially expressed genes; the most significant are labelled. n, Heat map of differentially expressed genes from hallmark gene sets with high gene correlation. In e,h–k, lines and bars represent medians (e) or means (h–k); error bars and shaded areas represent 95% confidence intervals. n refers to biologically independent samples.

Extended Data Fig. 1. Structural characterization of AS5 and structural frustration of AS1 state TH_X.

a, Crystal structure of AS5 alone (grey) overlaid with the design model of AS5 in state X (blue). Inset shows a detailed view of side chains in the partially open effector-binding cleft. b, Cocrystal structure of AS5 and peptide effector (grey) overlaid with the design model of the AS5–effector complex in state Y (AS5 in blue, effector in orange). Inset shows the same view of the side chains in the effector-binding cleft as in a. c, Design model of AS1 in state X (blue) aligned to the target (pink), showing a minor clash. d, Three cocrystal structures of AS1 (with intact cleft) and target with methylated lysines (grey) overlaid with the AF2 model of the target–AS1 complex in state X (AS1 in blue, target in pink), showing fluctuation in the target binding conformation.

Extended Data Fig. 2. DEER characterization of AS1 and AS114.

a, Raw DEER traces (black), foreground fits (colours), and background fits (grey) for AS1 and AS114 with all combinations of target and effector. Experiments on complexes included the target, effector, or both in excess over the host at concentrations higher than required to fully form the complex (Supplementary Fig. 9), and spin labels were placed far from the target and effector binding sites. Thus, changes in the DEER distance distributions with different combinations of target and effector should only reflect changes in host conformation. b, Distance distributions from experiment (colours) and simulated from the structural state represented by the cartoons (black). For AS1, the simulated and experimental distance distributions agree well, further validating that each state adopts its designed conformation. For AS114, the simulations consistently overestimate the experimental distribution by ~5 Å, but the shift in the distance distributions with the effector compared to those without validates the designed conformational change. c, Experimental distance distributions of all states, coloured corresponding to b. For both AS1 and AS114, the ternary complex distribution (green) aligns with the host–effector complex distribution (orange) and not with the host alone (blue) or target–host complex (pink) distributions, confirming that the ternary complex is primarily in state Y. b,c, Lines represent the distance distribution which best fits the time domain data; shaded regions represent 95% confidence intervals from bootstrapping (Methods).

Extended Data Fig. 3. MD analysis of the AS1 ternary complex.

a, The Cα RMSD of the MD trajectories from the crystal structure (grey) is lower than from the aligned clashing design models of target and host–effector complex (black), showing that the MD simulations strain away from the clashing state in a manner similar to the crystal structure. b, Cα RMSD of the switch (blue) and target (pink) from their position in the crystal structure when the entire structures are aligned. Compared to trajectories 1 and 2, trajectory 3 shows reduced target deformation and increased switch deformation, showing that these trajectories differ in where they localize strain to resolve the clash. c, Per-residue Cα RMSF of the host (blue), target (pink), and effector (orange) in the ternary complex computed from each trajectory. The clashing region of the target (highlighted in grey) shows considerable flexibility, according with this region being disordered in the crystal structure. d, Comparison of MD simulations to experimental data. Left, crystal conformation of the ternary complex (grey) aligned to representative conformations from each MD trajectory (red, yellow, and light blue). DEER spin-label positions are shown in green. In the crystal structure, the clashing region on the target is disordered. In the MD simulations, although flexible, this region remains mostly ordered, causing additional deformation compared to the crystal structure. Illustrating the differences in strain localization among trajectories shown in b, in the first two trajectories (red and yellow), the switch conformation aligns with the crystal structure and the target deforms more; in the third trajectory (light blue), the target conformation aligns with the crystal structure and the switch deforms instead. Right, experimentally measured DEER distance distribution of the ternary complex (green line representing the best fit to the time domain data and shaded region representing 95% confidence interval from bootstrapping (Methods)) and distance distributions simulated from the crystal structure (grey line) or MD trajectories (dashed lines, colours correspond to the conformations shown at left). The distance distribution simulated from the crystal structure aligns with the left peak in the experimental distance distribution, whereas the distance distributions simulated from the MD trajectories span the experimental distance distribution, suggesting that these trajectories more fully sample the space of ternary complex dynamics.

Extended Data Fig. 4. Nonuniform distribution of strain in the ternary complex.

a, Ratios of fold accelerations in the forward direction to those in the reverse direction, given by the equation: acceleration ratio = (k_off,T:HE/k_off,T:H) ÷ (k_off,TH:E/k_off,H:E). Acceleration ratios corresponding to designs with asymmetric distributions of strain are coloured orange, and to symmetric blue. b, Plot of forward vs reverse acceleration ratios, with linear fits for the symmetric and asymmetric groups. Note that on a log-log plot, the slope of a straight line passing through the origin becomes the y-intercept. c, For each design, design model of host–effector complex in state Y (blue and orange) aligned to the target (pink) to show the designed allosteric clash, and (grey) AF2 prediction of the target position relative to the switch in the ternary complex to show how the clash resolves through global strain. The target deforms downward in designs with an asymmetric distribution of strain, shearing the beta sheet, whereas it deforms outward in designs with a more symmetric distribution of strain, bending the beta sheet. d, Comparison of the AS1 binder–target interface (left) and switch–effector interface (right) (aligned at the host side) in the binary (colours) and ternary (grey) complex conformations, all from crystal structures, showing the binder–target interface deforms substantially more than the switch–effector interface in the ternary complex.

Extended Data Fig. 5. Construction and characterization of the chain reaction.

a, Design model of E2–target, comprising the target LHD101A (with mutations R43V and V69Q) fused to the effector peptide “E2” (cs201B) for hinge cs201. E2 is coloured green and LHD101A is coloured pink. b, Design models of E2–target (green/pink) bound to AS114 (blue) in state X showing no clash (left) and in state Y with the effector peptide (orange) showing a strong clash (right). c, Design models of the reporter hinge “H2” (cs201F with mutation E249L (sticks) which increases E2 on-rate and labelled with Alexa Fluors 555 and 647 at positions indicated by stars) in state X (left) and in state Y with E2–target (right). AS114 and H2 would overlap considerably if simultaneously bound to E2–target, so their binding should be mutually exclusive: AS114 should cage E2 until its release by the effector. d, Kinetics of E2–target and H2 association, measured by a change in FRET efficiency due to the conformational change in H2 upon binding. (Left) FRET time courses (normalized to the initial signal) with varying concentrations of E2–target and 5 nM H2; data (circles) fit with single exponentials (lines). (Right) apparent on-rate constants plotted against E2–target concentration (circles) and a linear fit. The value of the association rate constant (5e + 4 M⁻¹s⁻¹) is higher than the reported value (4.5e + 3 M⁻¹s⁻¹) for the original hinge cs201F with effector cs201B, suggesting that mutation E249L on H2 biases its conformational pre-equilibrium toward state Y to increase the apparent association rate. e, Additional data for the kinetically governed chain reaction shown in Fig. 4c. In the grey control time course, 500 nM AS114 was added to 20 nM H2, then 1 μM effector and 6 μM target was added at time 0, showing that none of these components bind to H2 to cause a change in FRET signal. In the other time courses, pre-incubated 500 nM AS114 and 250 nM E2–target was added to 20 nM H2, then buffer (blue), 1 μM effector (green), 6 μM target (pink), or both (orange) were added at time 0. A baseline drift (obtained from 500 nM AS114 after adding 20 nM H2 then at time 0 adding buffer) was subtracted from each time course, and time courses were normalized to the initial signal. The chain reaction proceeds faster when just excess target is added, probably due to blocking rebinding of E2–target to AS114 after transient dissociation, but this effect is insufficient to achieve full acceleration. The chain reaction also proceeds faster when just effector is added, but probably due to transient rebinding of E2–target to re-form the strained ternary complex, this also does not achieve full acceleration. Adding both effector (to accelerate E2–target dissociation from AS114) and excess target (to prevent E2–target rebinding to AS114) is required to fully accelerate the chain reaction. Note that if a single-chain effector is desired to fully accelerate the chain reaction, the effector and target could be flexibly fused into a single construct. Such a multivalent effector would be reminiscent of CITED2, whose multivalency enables rapid and unidirectional competition against HIF-1α (ref. ).

Extended Data Fig. 6. Construction and characterization of rapid sensors.

a, Structural model of the best SARS-CoV-2 sensor construct, comprising AScov (blue), the SmBiTgraft peptide with the effector (orange) and grafted SmBiT (green), LgBiT (grey or cyan), and flexible linkers (black). In this design, SmBiT is caged in a helical conformation when SmBiTgraft is bound to the switch and is free to reconstitute the luciferase when SmBiTgraft is released. To form SmBiTgraft, SmBiT was grafted onto the effector peptide such that most of its hydrophobic residues are buried within the switch–SmBiTgraft interface when in the bound helical conformation. Because the original effector peptide binds so strongly to the switch, it could accommodate replacing some of its interface residues with residues from SmBiT without reducing its affinity so much that it no longer effectively cages the SmBiT. b, SPR data showing sfGFP-SmBiTgraft binding to AS0 (blue and orange, association phase), slow subsequent dissociation in the absence of target (blue), and rapid subsequent dissociation upon addition of 10 μM target (orange) caused by rapid target binding to form a transient ternary complex, causing the spike at the beginning of the dissociation phase. c–f, Rapidly sensing the target through a facilitated dissociation mechanism (top), slowly sensing the effector limited by the slow base exchange rate of SmBiTgraft between binding AS0 and LgBiT (middle), and rapidly sensing the SARS-CoV-2 RBD with facilitated dissociation (bottom). c, Schematics showing the mechanism of sensing. d, Luminescence time courses (normalized to the initial signal) of 10 pM sensor construct then at time 0 adding varying concentrations of analyte (target, effector, or SARS-CoV-2 RBD); data (colours) fit (black) with single exponentials up to the maximum signal for time courses which showed appreciable signal increase. In some time courses, signal slowly decreases due to depletion of luciferase substrate. e, Luminescence signal fold change plotted against analyte concentration. f, Sensor response rate constant plotted against analyte concentration for time courses that showed appreciable signal increase.

Extended Data Fig. 7. Detailed functional characterization of ASNeo2.

a, Schematic depiction of the labelling strategy for single-molecule tracking experiments. b, Single-molecule trajectories of IL-2Rβ (red), γ_c (blue) and ASNeo2-induced heterodimers (magenta). c, Data from Fig. 5e as box plots to display datapoint variation including Neo2 (+/- effector) (green and orange, left side) and single-molecule tracking experiments with labelled ASNeo2 and γ_c (right side). Sample sizes and independent repeats are: unstimulated: 37 and 3; ASNeo2: 32 and 3; ASNeo2 + Effector: 33 and 3; Neo2: 44 and 4; Neo2 + Effector: 26 and 2; labelled ASNeo2: 21 and 2; labelled ASNeo2 + Effector: 18 and 2. d, Relative dimerization in relation to receptor cell surface density ratio indicates that high dimerization data variance is caused by differing IL-2Rβ to γ_c ratios at the plasma membrane. Even at high γ_c excess, effector-bound ASNeo2 shows no residual affinity for γ_c. e, Diffusion properties of IL-2Rβ and γ_c are reverted to the ground state after addition of effector. f, Immobile particles are increased upon stimulation, but not decreased after effector addition, potentially indicating receptors internalizing in membrane proximal endosomes. For e,f, the left box always corresponds to IL-2Rβ and the right one to γ_c. Sample sizes for d–f are as in c. g, Normalized localization density over time confirms minimal single-molecule bleaching in long-term single-molecule tracking experiments. Sample sizes and independent repeats are: without Effector: 5 and 5; with Effector: 3 and 3. h,i, Dissociation of ASNeo2-induced IL-2Rβ/γ_c dimers at the cell surface upon addition of 10 µM effector as detected by time-lapse single-molecule co-tracking (h) with colour-coded corresponding co-trajectories (i). j,k, Conformational change of ASNeo2 bound to the cell-surface receptor probed by smFRET. j, FRET efficiency histograms for ASNeo2 E4C/K211C labelled with Cy3B and ATTO643 bound to cells expressing IL-2Rβ and γ_c in the absence (blue) and presence (yellow) of 10 µM effector. Sample sizes are: without Effector: 7; with Effector: 5. k, smFRET co-localizations of one individual cell before the effector was added (left) and after it was added (right) colour-coded for FRET efficiency, highlighting the observation of individual molecules. Statistics for c,e,f were performed using two-sided two-sample Kolmogorov–Smirnov tests (ns, not significant, P values noted). Box plots show the distribution of the dataset, highlighting the median, quartiles, and outliers, with whiskers extending to the range limits. Scale bars in b,i: 5 µm.

Extended Data Fig. 8. Characterization of cyclic permutations of ASNeo2.

a, Design models of ASNeo2 and selected cyclic permutations in state X, rainbow-coloured from N terminus (blue) to C terminus (red) to illustrate the protein topology. In ASNeo2, the switch is at the N terminus and Neo2 is at the C terminus. In the cyclic permutations, although the relative position of the switch and Neo2 changes minimally, the switch is in the middle of the protein, part of Neo2 is at the N terminus, and the other part is at the C terminus. This way, the regulatory switch cannot degrade without also breaking Neo2. b, SEC purifications performed on a Superdex 200 Increase 10/300 GL column. The cyclic permutations are prone to aggregation during expression, but distinct monomer peaks can be picked out. c, Fast effector-concentration-dependent dissociation of γ_c from the ASNeo2–IL-2Rβ–γ_c complex upon addition of peptide effector. Data (grey) fit (colours) as described in methods (neglecting the accumulation modelling because accumulation on the SPR surface was negligible with these proteins). d, Rate constants of facilitated γ_c dissociation computed from the model fit by ln(2)÷{half-time of γ_c–host interaction} plotted against effector concentration (circles) and fit with hyperbolic equations (black lines).

Extended Data Fig. 9. Additional characterization of differential effects of transient ASNeo2 stimulation on T cell behaviour.

Human T cells were stimulated with 1 nM (a–g) or 5 nM (i) ASNeo2 for 5 min (a–h), 30 min (h), or 25 min (i) or left untreated as a control. Signalling was either sustained by continued ASNeo2 treatment or terminated by the addition of 10 μM effector. Cells were collected for counting and phenotypic analysis by flow cytometry after 72 h (b,c; n = 4) or 48 h (a,e–g; n = 4), for pSTAT signalling analysis by flow cytometry after 20 min (h, n = 3), or for RNA-seq analysis after 6 h (i, n = 3). a, Changes in viable T cell counts from 0 h to 48 h across each group. b,c, Representative flow-cytometry histogram of CFSE (b) and quantitative analysis of divided cells (c). d, Frequencies of live T cells. e, Mean fluorescence intensity (MFI) of Ki-67, CD25, GATA3, CD69, and BCL2. f,g, Representative flow-cytometry plots (f) and quantitative analysis of caspase-3+ cells (g; n = 4). h, Dose-dependent pSTAT5 curves. i, Transcripts per million (TPM) of CDK4 and POLD2 in the MYC_TARGETS_V1 gene set (left); TPM of DOCK2 and KIF1B in the MITOTIC_SPINDLE gene set (right). a,c–e,g, Statistics were obtained from ANOVA with two-sided Tukey’s post-test (ns, not significant (P > 0.05), *P = 0.04, **P = 0.001, ***P = 0.0002, ****P < 0.0001). Lines or bars represent means; error bars represent s.e.m. (a,h,i) or SD (c–e,g). n refers to biologically independent samples.