Small‐Molecule‐Induced and Cooperative Enzyme Assembly on a 14‐3‐3 Scaffold

Abstract Scaffold proteins regulate cell signalling by promoting the proximity of putative interaction partners. Although they are frequently applied in cellular settings, fundamental understanding of them in terms of, amongst other factors, quantitative parameters has been lagging behind. Here we present a scaffold protein platform that is based on the native 14‐3‐3 dimeric protein and is controllable through the action of a small‐molecule compound, thus permitting study in an in vitro setting and mathematical description. Robust small‐molecule regulation of caspase‐9 activity through induced dimerisation on the 14‐3‐3 scaffold was demonstrated. The individual parameters of this system were precisely determined and used to develop a mathematical model of the scaffolding concept. This model was used to elucidate the strong cooperativity of the enzyme activation mediated by the 14‐3‐3 scaffold. This work provides an entry point for the long‐needed quantitative insights into scaffold protein functioning and paves the way for the optimal use of reengineered 14‐3‐3 proteins as chemically inducible scaffolds in synthetic systems.


Protein expression and purification C9-CT52
The C9-CT52 protein and mutants, encoded in a pET28a plasmid, were expressed under sterile conditions in E. coli BL21 (DE3) (Novagen). Two liter cultures of LB medium were used in 5 L baffled conical flask and supplemented with kanamycin (30 µg/mL) inoculated with 25 mL of overnight culture. The culture was incubated at 37 °C and 160 rpm until the optical density OD600 reached 0.6-0.8. Subsequently 0.5 mM isopropyl-β-D-thiogalactopyranoside (IPTG) was added to induce expression and the culture was incubated overnight at 18 °C. Cells were harvested by centrifugation (6,913 g, 10 min, 4 °C) in a Sorvall Evolution Centrifuge with a SLC-300 rotor (Thermo Scientific). The cell pellet was stored at -80° until purification, successively the pellet was resuspended in lysis buffer (10 mL per gram cell pellet, PBS, 370 mM NaCl, 10% glycerol, 20 mM Imidazole, 0.1 mM TCEP, Benzonase® Nuclease (25 U per 10 mL buffer, Novagen), pH 7.4) and cells were lysed using a EmulsiFlexC3 High Pressure homogenizer (Avestin) at 15,000 psi for two rounds. Cell debris was removed by centrifugation (43,206 g, 45 min, 4 °C) in a Sorvall Evolution Centrifuge with a SA300 rotor. The supernatant was applied to a Ni-loaded column (His-Bind® Resin, Novagen) and washed with wash buffer (PBS, 370 mM NaCl, 10% glycerol, 20 mM Imidazole, 0.1 mM TCEP pH 7.4) in presence and subsequently absence of 0.1% Triton-X-100. Protein was eluted from the column by elution buffer (PBS, 370 mM NaCl, 10% glycerol, 250 mM Imidazole, 0.1 mM TCEP pH 7.4). Elution fractions were pooled and cleavage of the His6-SUMO tag was performed in a dialysis bag (MWCO 3.5 kDa, Spectra Laboratories) in the presence of SUMO protease dtUD1 (purified according to reported protocol [2] ) (1:500) and dialyzed against 4L of dialysis buffer(150 mM NaCl, 50 mM Tris, pH 7.8) stirring at 4 °C. The dialysis sample was further purified using a second Ni-column equilibrated with wash buffer and the flowthrough containing the protein was collected. The protein was concentrated using Amicon Ultra Centrifugal Filters (MWCO 10 kDa, Millipore). Concentrated protein was bufferexchanged into assay buffer (20 mM Na2HPO4, 150 mM NaCl, 1 mM EDTA, 2 mM TCEP, pH 7.0) using a PD-10 desalting column (GE Healthcare). Concentration of the protein was determined using a Thermo Scientific ND-1000 spectrophotometer at 280 nm (ε = 25,440 M -1 cm -1 in water under reducing conditions).

S5
Activity assay synthetic substrate Ac-LEHD-AFC Activity assays with synthetic substrate Ac-LEHD-AFC (Enzo Life Sciences, dissolved as 10 mM stock in DMSO) were performed in activity assay buffer (20 mM Na2HPO4, 150 mM NaCl, 1 mM EDTA, 2 mM TCEP at pH 7.0). Assay components 0.02-10 μM 14-3-3 monomer, 0.1 μM C9-CT52 and 0.1 -1 μM FC (dissolved as 14.7 mM stock in EtOH, Enzo Life Sciences) were incubated for 30 minutes at room temperature to allow complex formation and subsequently substrate was added to a final volume of 100 μL. Measurements were done in a 96-well plate (Greiner F-bottom/chimney, black) at 37 °C, using an excitation wavelength of 400 nm (band with 20 nm) and an emission wavelength of 520 nm (band with 25 nm) using a Tecan Infinite F500 plate reader. Activity (U/mg) was calculated using the calibration curve below.

Kinetic model of scaffold-mediated enzyme activity under smallmolecule control
Derivation and numerical solution Monovalent caspase-9 fusion proteins are recruited onto bivalent 14-3-3 scaffolds in the presence of the small molecule FC. In the kinetic model, the ternary complex consisting of two caspase-9 monomers bound to the bivalent 14-3-3/FC scaffold is assumed to be enzymatically active. In addition, caspase-9 monomers can homo-dimerize in a scaffoldindependent fashion resulting in catalytically active dimers. Figure S7 shows a schematic representation of all the scaffold-mediated and scaffold independent equilibria involved as well as the catalytically active species. Our kinetic model is based on a two-step procedure. The main assumption of the model is that equilibration of the monovalent C9-CT52, FC and 14-3-3 species is much faster than the catalytic conversion of the substrate. Therefore, in the first step the equilibrium concentration of all species is found by solving algebraic expressions for the equilibrium concentrations of each of the different species as a function of the parameters (the equilibrium constants and the total concentrations). In order to verify these assumptions, we have modeled the scaffold equilibration using realistic kinetic parameters which reveal that this assumption only leads to small errors (typically between 5-10%). We have incorporated the possibility for cooperative binding of the second monovalent caspase-9 fusion protein as many scaffoldmediated protein complexes display some level of cooperativity. [3] In contrast, binding of the small molecule FC to two binding sites on the 14-3-3 scaffold occurs in an independent fashion. In the second step, we model the catalytic conversion of the substrate by recruited caspase dimers on the 14-3-3 scaffold as well as free caspase-9 homo-dimers by assuming that the rate of product formation can be described by a Michaelis-Menten approximation. Let us define the following quantities: Eq. 1 Next, we write down the mass-balances of the total concentration of FC and C9-CT52 species partitioned between their respective free forms and various complexes (Eq. 2-3).
Eq. 2 Eq. 3 Now, we write down the equilibrium equations of the dissociation constant Kd,FC and its related species, while taking into account the appropriate statistical factors. There are 3 possible reactions: the binding of FC to a free scaffold molecule, the binding of a second FC to a BF complex and the binding of FC to BFC complex (Eq. 4-6).
Eq. 4 Eq. 5 Eq. 6 Next, we derive expressions relating the C9-CT52 containing species to the equilibrium constant Kd and the cooperativity parameter σ, while taking into account the appropriate statistical factors (Eq. 7-9). The homo-dimerization of C9-CT52, described by dissociation constant Kd,C9, is also considered (Eq. 10). Eq. 7 Eq. 8 Eq. 9 Eq. 10 These equilibrium equations can be rewritten to obtain expressions for all complexes BF, BFF, BFFC, BFFCC, BFC and CC as a function of the dissociation constants relevant parameters and free concentrations of all species (Eq. 11-16.).
Eq. 11 Eq. 12 Eq. 13 Eq. 14 Eq. 15 Eq. 16 Finally, equations 11-16 are substituted in the mass-balance equations (eq. 1-3) to arrive at the following expressions: Eq. 17 Eq. 18 Eq. 19 Custom-written Matlab scripts are used to solve the coupled non-linear equations 17-19 for the free concentrations B, F and C. These can subsequently be used to calculate the steady-state concentrations of all other species (eq11-16), most importantly the enzymatically active complexes BFFCC and CC. The total amount of enzymatically active compounds in steady-state, E and initial substrate concentration S0 are used to find the initial reaction rate v0 as given by the Michaelis-Menten equation (Eq. 20). Furthermore, a composite parameter was introduced, which represents all parameters that influence only the height of the activity curve (KM, kcat and S0).

Eq. 20
The initial reaction rate is converted into enzymatic activity measured in U/mg. For this, a conversion factor is used that contains the total C9-CT52 concentration Ctot, the volume V of the experimental reaction mixture and the molecular weight Mw of C9-CT52 (Eq. 21).

Eq. 21
Data analysis and parameter estimation We performed non-linear least square analysis on the activity assay data as reported in Figure 3.A by comparing the activity data to the computed activity. Previous work has reported accurate values for Kd,FC (66 μM [4] ) and as such this constant was assumed constant during the non-linear least square optimization. To increase the accuracy of the estimated parameters, we used three datasets in the optimization obtained by scaffold titrations using different concentrations of FC (Figure 3.A and Figure S.8). During the fitting, a composite parameter was introduced, which incorporates all parameters that only influence the height of the activity curve (KM, kcat and S0). S0 is set at 200 μM according to the experimental parameters and the determined KM (650 μM) and kcat (0.2 s -1 ) are both in accordance with literature. [5][6][7] To compensate for length differences between the datasets, residuals were multiplied by a relative weight factor. Nonlinear least square minimization of the data was performed using the Matlab function lsqnonlin, a subspace trust region method based on the interior-reflective Newton method. In order to prevent entrapment in a local minimum, 30 different starting values of Kd, , Kd,C9 and were defined, and the best fit (defined as the fit with the lowest squared 2-norm of the residuals) is taken as the final solution for the optimized values. The different initial parameter sets are defined using a latin hypercube sampling method (Matlab function lhsdesign). The standard deviation of the parameters were S11 calculated using the Fischer information matrix. [8] The optimized values of Kd, , Kd,C9 and and their standard errors are reported in