Friction-Limited Folding of Disulfide-Reduced Monomeric SOD1

Abstract

The folding reaction of a stable monomeric variant of Cu/Zn superoxide dismutase (mSOD1), an enzyme responsible for the conversion of superoxide free radicals into hydrogen peroxide and oxygen, is known to be among the slowest folding processes that adhere to two-state behavior. The long lifetime, ∼10 s, of the unfolded state presents ample opportunities for the polypeptide chain to transiently sample nonnative structures before the formation of the productive folding transition state. We recently observed the formation of a nonnative structure in a peptide model of the C-terminus of SOD1, a sequence that might serve as a potential source of internal chain friction-limited folding. To test for friction-limited folding, we performed a comprehensive thermodynamic and kinetic analysis of the folding mechanism of mSOD1 in the presence of the viscogens glycerol and glucose. Using a, to our knowledge, novel analysis of the folding reactions, we found the disulfide-reduced form of the protein that exposes the C-terminal sequence, but not its disulfide-oxidized counterpart that protects it, experiences internal chain friction during folding. The sensitivity of the internal friction to the disulfide bond status suggests that one or both of the cross-linked regions play a critical role in driving the friction-limited folding. We speculate that the molecular mechanisms giving rise to the internal friction of disulfide-reduced mSOD1 might play a role in the amyotrophic lateral sclerosis-linked aggregation of SOD1.

Significance

Friction-limited folding has been previously studied; however, the molecular mechanism(s) of internal friction remains a subject of continuing interest. One potential explanation for friction-limited folding might be the formation of transient, nonnative states that impede the productive folding reaction. To test this hypothesis, we monitored the thermodynamic and kinetic folding properties of a monomeric variant of superoxide dismutase (mSOD1) in the presence of small molecule viscogens. We observed friction-limited folding in mSOD1 lacking its internal disulfide bond, but not in its disulfide-containing counterpart. We conclude that the nonnative structure in a peptide model of the C-terminal segment is responsible for the internal friction, and we speculate that this behavior may be involved in the amyotrophic lateral sclerosis-linked aggregation of SOD1.

Introduction

The mechanisms by which a newly synthesized polypeptide chain reaches its properly folded state on a biological timescale remain an area of intense interest (1,2). These mechanisms range from commonly observed and well-understood phenomena such as the formation of partially folded intermediates (3, 4, 5) or the rapid collapse of hydrophobic patches to avoid unfavorable interactions with a solvent (6, 7, 8) to rarely observed and poorly understood events such as internal chain friction retarding the folding process. The hallmark of friction-limited folding is a nonunitary dependence of the folding rate constant upon the bulk solvent viscosity (9,10). Despite the observation of friction-limited folding in several proteins (11, 12, 13, 14, 15, 16, 17), the molecular mechanisms responsible for friction remain a subject of active discussion.

Internal friction in spectrin domains has been attributed to the nonnative docking of its trio of helices (11). By contrast, molecular dynamics simulations of short peptides attributed internal friction to isomerization of local torsional angels (18) and was closely linked to the α-helical propensity of the protein (19,20). This interpretation is consistent with the observation of friction-limited folding in several fast-folding proteins (12, 13, 14, 15), but not others (10,21,22). Schuler et al. have proposed that chain collapse under strongly folding conditions could impede diffusion and be another source of friction (16,17). A yet unexplored explanation for friction-limited folding might be transient nonnative structures in the unfolded ensemble. These putative nonnative intramolecular interactions might be detected by their propensity to slow folding below the diffusion limit (23,24).

If friction-limited folding is a consequence of sampling small nonnative structures in the unfolded ensemble, then a potential candidate might be the Cu/Zn superoxide dismutase 1 (SOD1). SOD1 is a 153-residue homodimeric β-sandwich protein containing eight antiparallel β-strands in each monomer (Fig. 1). Each subunit binds a structural zinc and a catalytic copper ion and contains an intramolecular disulfide bond between C57 in Loop IV after β4 and C146 in β8. Loop IV binds the zinc ion, whereas Loop VII, the electrostatic loop, serves to support Loop IV and guides the anionic substrate to the redox-active copper ion. A recent peptide dissection analysis of SOD1 from our lab demonstrated that the 35-residue C-terminal peptide, containing Loop VII and β8, folds on itself to bring the polar and charged N-terminus and the nonpolar C-terminus into a nonnative juxtaposition (25). This interaction cannot exist in the disulfide-containing form of SOD1, in which C146 in β8 is covalently linked to C57.

Figure 1.

Ribbon diagram of mSOD1. The view of mSOD1 along each of the β-sheets is shown, adapted from PDB: 2C9V. The eight β-strands are labeled, and the zinc-binding and electrostatic loops are shown in blue and purple, respectively. The disulfide bond between C57-C146 is shown in yellow.

Supposing the nonnative structure of this peptide might create friction-limited folding in the full-length protein absent of the SS bond, we performed kinetic and equilibrium analyses on the folding of the disulfide-reduced form of a stable monomeric variant of SOD1 (2SH-mSOD1) (26) in the presence of varying amounts of two small molecule viscogens. The SOD1 monomer variant (mSOD1) containing the SS bond (SS-mSOD1) served as a control. We found that 2SH- and SS-mSOD1 closely adhere to a two-state mechanism in glycerol and glucose when examined by both equilibrium and kinetic methods. Using a novel method to fit the kinetic data, we found that the folding and unfolding reactions of only 2SH-mSOD1 were partially limited by the internal chain friction. The simple diffusional processes observed for SS-SOD1 show that the internal friction experienced by 2SH-SOD1 involves the formation of a structure involving one or both of the segments containing the C57 and C146 partners in the disulfide bond. We propose that the nonnative structure formed by the C-terminal region is a source of friction in the folding of SOD1.

Materials and Methods

SOD1 monomer model and expression/purification

mSOD1 has a pair of nonpolar to polar mutations at the dimer interface, F50E and G51E, that prevent dimerization (26). The two cysteines at positions 6 and 111 were replaced with alanine and serine, respectively, to eliminate disulfide scrambling with the C57-C146 disulfide bond in the unfolded state. Recombinant metal-free SS-mSOD1 was expressed and purified according to the procedure previously described (27), and 2SH-mSOD1 was created by incubation with 1 mM tris(2-carboxyethyl)phosphine for several hours at room temperature (28). The purity of the mSOD1 samples were confirmed by mass spectrometry.

Thermodynamic and kinetic circular dichroism experiments

Circular dichroism (CD) spectroscopy was performed on a Jasco-810 spectropolarimeter (Jasco, Easton, MD) equipped with a water-cooled Peltier temperature control system. The equilibrium Gdn-HCl-induced unfolding transitions were monitored from 215 to 240 nm in a 0.2-cm-pathlength synthetic-fused silica cuvette using a scan rate of 20 nm per min⁻¹ and a response time of 8 s. Manual-mixing kinetic folding and unfolding reactions, with a deadtime of 3–5 s, were monitored at 230 nm with a response time of 1–2 s in a 1-cm cuvette. Gdn-HCl concentrations were determined by refractive index measurements on a Leica Mark II refractometer (Reichert Technologies, Buffalo, New York). The protein concentrations were 15 μM for equilibrium experiments and 5–10 μM for kinetic experiments. The buffer conditions were 20 mM HEPES and 1 mM K₂EDTA for SS-mAS-SOD1 experiments and 20 mM HEPES, 1 mM K₂EDTA, and 1 mM tris(2-carboxyethyl)phosphine for 2SH-mAS-SOD1 experiments. The temperature was 20°C for all experiments. The data analyses were performed with an in-house software Savuka available online at https://osmanbilsel.com/.

Solvent viscosity measurements

The solvent viscosity at each polyol concentration was determined using a Gilmont falling ball viscometer (Thermo Fisher Scientific, Waltham, MA). The time of travel was measured 10 times for each sample and was repeated a minimum of two times with fresh sample preparations. All buffers were filtered and degassed before the experiment.

Results

Thermodynamic analysis of SOD1 folding in viscogens

The thermodynamic properties of metal-free SS-mSOD1 and 2SH-mSOD1 were monitored by CD spectroscopy. Both forms exhibited increased stability with increasing concentrations of glycerol, as evidenced by the shift of the midpoint of the equilibrium titration, C_m, to higher concentrations of guanidine hydrochloride (Gdn-HCl) (Fig. 2, left panels). A similar trend was observed for glucose (Fig. 2, right panels), demonstrating that this phenomenon is not specific to the choice of kosmotrope.

Figure 2.

Polyol dependence of equilibrium titrations. The equilibrium titration of SS-mSOD1 (top) and 2SH-mSOD1 (bottom) at increasing concentrations of glycerol (circles) and glucose (squares), monitored by CD spectroscopy at 230 nm, are shown with a two-state fit to the data (solid lines). SS-mSOD1 with glycerol percentages by volume (top left) are as follows: 0% (white), 10% (red), 16% (orange), 24% (yellow), 28% (green), 30% (blue), and 32% (purple). SS-mSOD1 with glucose percentages by volume (top right) are as follows: 0% (white), 11% (dark red), 17% (dark orange), 20% (dark green), 24% (dark blue), and 26% (dark purple). 2SH-mSOD1 with glycerol percentages by volume (bottom left) and with glucose percentages by volume (bottom right) follow the same color scheme.

The data were fitted to a two-state model assuming a linear dependence of the free energy of folding on the denaturant concentration (29), and quantitative estimates of the free energy of folding, ΔG⁰_eq, are provided in Table S1. Over the range of cosolvents explored, 0–32% by volume, the unfolding transitions for both forms of mSOD1 were well described by two-state models, and the stabilities were found to increase by ∼2-fold (Tables S1 and S2). Interestingly, the trend is reversed for the denaturant dependence of the free energy of folding, the m_eq-values, which decrease by ∼25% over this range in glucose or glycerol concentrations for both SS-mSOD1 and 2SH-mSOD1 (Tables S1 and S2). Although the decrease in the m_eq-value could reflect the presence of partially folded states, whose population is increased by the kosmotropes, companion kinetic studies of the folding reactions offer an alternative explanation (see below).

Kinetic analysis of SOD1 folding in viscogens

To test for the presence of internal friction-limited folding, we performed kinetic folding experiments on SS-mSOD1 and 2SH-mSOD1. The unfolding and refolding reactions for SS-mSOD1 and 2SH-mSOD1 were monitored by CD spectroscopy. The resulting kinetic traces were well fitted by single-exponential functions that provided the apparent rate constant, k, and its reciprocal, the relaxation time, τ. The excellent agreement between the ellipticities at the beginning of both unfolding and refolding reactions and the estimated ellipticities of the native and unfolded states under the same conditions eliminates the possibility of partially folded species with significant amounts of secondary structure forming within the deadtime of the experiment (Fig. S1). The chevron plots, which show the observed relaxation times as a function of final denaturant concentration, are given in Fig. 3. These data demonstrate that increasing concentrations of glycerol and glucose increase the free energy of folding between the native and unfolded states by decreasing the refolding times and increasing the unfolding relaxation times. The net effect of these changes moves the maximal relaxation time, at which k_u = k_f, to progressively higher concentrations of denaturant, concordant with the C_m-values from the equilibrium studies.

Figure 3.

Polyol dependence of the folding kinetics. Chevron plots at increasing concentrations of glycerol (circles) and glucose (squares) of SS-mSOD1 and 2SH-mSOD1, obtained by CD spectroscopy at 230 nm, are shown with an independent fit of each chevron to a two-state model (solid lines). The color and symbol schemes are described in the caption of Fig. 2.

The chevron data were well described by a simple two-state kinetic model linking the native and unfolded states. The predicted free energy differences in the absence of denaturant between the folded and unfolded states were calculated from ΔG⁰_eq = −RT ln (k_u/k_f), where the rate constants in the absence of denaturant were obtained by linear extrapolations of the unfolding and refolding legs of the chevron plot to determine τ_f/τ_u, the reciprocal of k_f/k_u. The m_eq-value can also be determined by summing the absolute values of the m_u^‡- and m_f^‡-values, the slopes of the two legs of the chevron plot (29). The resulting values are listed in Tables S3 and S4. The excellent agreement between the kinetic and equilibrium estimates of stability, the m_eq-values, and the C_m-values for both SS-mSOD1 and 2SH-mSOD1 over the range of kosmotrope concentrations examined (Fig. S2), along with the absence of burst-phases for both unfolding and refolding reactions (Fig. S1), rules out partially folded states as an explanation for the reduction of m_eq-values at increasing kosmotrope concentrations.

It is noteworthy that the denaturant dependence of the unfolding and refolding relaxation times (Tables S3 and S4), the m_u^‡- and m_f^‡-values, for SS-mSOD1 and 2SH-mSOD1 vary with glycerol and glucose concentration in a fashion that leaves their Tanford β (β_T) values unchanged across the range of glycerol and glucose concentrations explored (Fig. S3). β_T reflects the fraction of the buried surface area in the transition state ensemble (TSE) relative to the buried surface area in the native state and is calculated as β_T = m_f^‡/(m_f^‡ − m_u^‡) (30). The β_T-values for SS-mSOD1 and 2SH-mSOD1 are independent of kosmotrope concentration, and the average values are ∼0.68 ± 0.02 and ∼0.75 ± 0.01 kcal mol⁻¹ M⁻¹, respectively. The robust positions of the TSEs in terms of the buried surface area for both oxidized and reduced mSOD1 demonstrate that these additives have a common but proportional effect on the structures and the free energies of the TSE and the unfolded state. The reduction in the m_eq-value in the presence of kosmotropes (Tables S1–S4) is, therefore, a consequence of the solvent composition and does not reflect the presence of an additional thermodynamic state. This conclusion is consistent with a previous study that observed the compaction of the unfolded state of SNase in response to viscogen (31). The consistently higher m_eq-value (Tables S1–S4) and the greater β_T-value for 2SH-mSOD1 (Fig. S3) likely reflect the increased solvent exposure of side chains and backbone in an unfolded ensemble that is not constrained by the disulfide cross-link.

Viscosity dependence of folding

Glycerol and glucose also act to substantially increase the solvent viscosity, providing a test for intramolecular friction in limiting diffusive folding reactions. The rate constants of diffusion-limited reactions are expected to vary linearly with solvent viscosity, according to the Kramer’s formalism (9,22,32, 33, 34, 35, 36)

$$k=\frac{1}{\tau }=\frac{C}{\eta +\sigma }\text{exp}\left(\frac{-\Delta G^{0‡}}{RT}\right),$$ (1)

where k is the rate constant, τ is the corresponding relaxation time, C is an adjustable parameter, η is the viscosity of the solution, σ is the internal friction, ΔG^0‡ is the intrinsic barrier height under standard conditions, R is the gas constant, and T is the absolute temperature. Thus, in the absence of significant internal friction, the folding rate constant should vary inversely with the solvent viscosity, k ∝ 1/η (37,38). A less-than-unitary dependence of the observed rate constant on the solvent viscosity, i.e., a nonzero value for σ, has been taken as evidence for chain friction in limiting the folding of three helix-bundle spectrin domains (11,16,39,40), cytochrome c (14), an ultrafast folding protein (12), and unfolded and intrinsically disordered proteins (17,41,42). Because viscogens significantly alter the thermodynamic properties, an isostability approach, in which the refolding and unfolding rate constants were interpolated under strongly refolding and unfolding conditions to a constant thermodynamic driving force, is typically employed (21,32,43). This approach, however, does not account for the observed change of the m_f^‡- and m_u^‡-values for SOD1 in response to the presence of viscogen (Tables S1–S4).

To address this issue, we developed a method of globally fitting the chevron data by introducing parameters that describe the viscogen dependence of the thermodynamic and kinetic properties of the folding reaction. It has been previously shown that the effects of Gdn-HCl and viscogens on the viscosity of the solution are independent and additive in the range of concentrations examined in this study (44). When plotted as a function of viscogen concentration, we observed that ΔG⁰_eq, m_eq, m_f^‡, and m_u^‡ all vary linearly with increasing viscogen concentration (Figs. 4 and 5). This behavior is most simply explained by the independent and additive effects of denaturant and viscogen on the folding and unfolding rate constants.

Figure 4.

Viscogen dependence of ΔG⁰_eq for SS-mSOD1 and 2SH-mSOD1. The ΔG⁰_eq of SS-mSOD1 (blue) and 2SH-mSOD1 (red) as a function of either glycerol concentration (light circles) or glucose concentration (dark squares) is shown along with linear fits of the data (solid and dashed for glycerol and glucose, respectively). The ΔG⁰_eq in the absence of viscogen is shown as a diamond for both SS-mSOD1 and 2SH-mSOD1.

Figure 5.

Viscogen dependence of m-values for SS-mSOD1 and 2SH-mSOD1. The m_eq (top), m_f^‡ (middle), and m_u^‡ (bottom) are shown for SS-mSOD1 (blue) and 2SH-mSOD1 (red) as a function of either glycerol concentration (light circles) or glucose concentration (dark squares) along with linear fits to the data (solid and dashed for glycerol and glucose, respectively). The units of m_eq, m_f^‡, and m_u^‡ are in kcal mol⁻¹ [D] ⁻¹. The m-values in the absence of viscogen are shown for SS-mSOD1 and 2SH-mSOD1 as blue and orange diamonds, respectively. m_eq was calculated as m_f^‡− m_u^‡.

The additivity is captured in the exponential terms in Eq. 2.

$$k_{obs}=\frac{k_f}{\eta +{\sigma }_f}\text{exp}\left(\frac{\left(m_f^{‡}+{\alpha }_f^{‡}\left[Visc\right]\right)\left[Den\right]-m_{vf}^{‡}\left[Visc\right]}{RT}\right)+\frac{k_u}{\eta +{\sigma }_u}\text{exp}\left(\frac{\left(m_u^{‡}+{\alpha }_u^{‡}\left[Visc\right]\right)\left[Den\right]+m_{vu}^{‡}\left[Visc\right]}{RT}\right),$$ (2)

where [Visc] and [Den] are molar concentrations of viscogen and denaturant, k_obs is the observed rate constant at that viscogen and denaturant concentration, k_f and k_u are the rate constants in the absence of viscogen, m_f^‡ and m_u^‡ represent the denaturant dependence of the rate constants, α_f^‡ and α_u^‡ are the viscogen dependence of m_f^‡ and m_u^‡, m_vf^‡ and m_vu^‡ represent the viscogen dependence of the transition state free energy, η is the solution viscosity, and σ_f and σ_u are the chain friction terms for folding and unfolding. Note that the viscogen dependence terms (α_f^‡, α_u^‡, m_vf^‡, and m_vu^‡) are unique for a specific viscogen, whereas the other terms (k_f, m_f^‡, σ_f, k_u, m_u^‡, and σ_u) are universal parameters for a protein. In this method, only the parameters α_f^‡, α_u^‡, m_vf^‡, m_vu^‡, σ_f, and σ_u are altered to obtain the fit; the values for the parameters k_f, m_f^‡, k_u, and m_u^‡ are fixed values obtained from the chevron fits in the absence of any viscogen (Tables S3 and S4), and the parameters[Visc], [Den], η, and are known constants.

Global fits of the chevron data for both SS-mSOD1 and 2SH-mSOD1 to Eq. 2 provided an excellent description of the results (Fig. 6). Supporting the validity of the analysis is the excellent agreement between 1) the results of the individual linear regression values for terms α_f^‡, α_u^‡, m_vf^‡, and m_vu^‡ and those obtained from the global fit (Table S5) and 2) the values of the friction coefficients for the unfolding and refolding reactions in the presence of glucose and glycerol (Table S6). A global fit of the combined glucose and glycerol data sets, subjected to a rigorous error analysis (Fig. S4), finds the folding and unfolding friction coefficients for SS-mSOD1 are σ_f = 0.07 ± 0.09 and σ_u = 0.00 ± 0.01. SS-mSOD1 experiences neither friction-limited folding nor unfolding. By distinct contrast, the friction coefficients for 2SH-mSOD1 are σ_f = 1.02 ± 0.24 and σ_u = 0.87 ± 0.13. We conclude that 2SH-mSOD1 experiences internal chain friction during both folding and unfolding reactions, slowing access to the productive TSE from both the native and the unfolded states of 2SH-mSOD1.

Figure 6.

Global fit of folding kinetics. Global fits to the sets of chevrons in each panel, obtained using Eq. 2, are shown for SS-mSOD1 (top) and 2SH-mSOD1 (bottom). The colors and symbols were carried forward from Fig. 2.

Discussion

Internal friction in proteins

Friction in condensed phase reactions arises from the slowing of a reaction by the exchange of energy and momentum of a molecule with its surroundings, typically the solvent. For proteins, the chain may play a role analogous to the solvent via intramolecular interactions, e.g., hydrogen bonds, van der Waals interactions (45), and electrostatic interactions (46). A common theme in friction is the rate at which the solvent, or the rest of the chain in the case of proteins, equilibrates relative to the molecular motion of a reaction. For proteins, the equilibration time is likely to be affected by nonnative interactions with residence times greater than the chain reconfiguration time. The nonnative docking of helices in spectrin (11) and the nonnative conformation of the C-terminal peptide of SOD1 (25) are candidate states that may contribute to friction in this manner.

Internal friction in SOD1

The introduction of glucose and glycerol to the solution and the accompanying increase in solvent viscosity revealed friction-limited refolding and unfolding for 2SH-mSOD1. The contrasting behavior of SS-mSOD1 focuses attention on β8, which is prevented from forming a nonnative interaction with the N-terminus of Loop VII when linked to Loop IV by an SS bond. The conclusion that the nonnative interaction between Loop VII and β8 is responsible for chain friction in both unfolding and refolding is consistent with a mutational analysis that mapped the structure of the TSE onto β2, β3, β4, and β7, the internal strands on both faces of the β-sandwich (Fig. 1; (47)). These results predict that the Loop VII-β8 sequence would be released before the TSE in unfolding and only engage after the formation of the TSE in refolding. The C-terminus of SOD1 would have ample opportunity to collapse on itself and introduce chain friction for both reactions. Although the marginal stability of the Loop VII-β8 peptide to denaturant appears to challenge this explanation for friction-limited unfolding, the presence of the TSE in the full-length protein and high concentrations of viscogen make comparisons ambiguous and highlight the limitations of peptide models for unfolded proteins (25).

mSOD1 may be uniquely sensitive to local nonnative interactions in the unfolded ensemble because its long-range contact order would require diffusion-mediated contacts with residues distal in sequence (47,48). Consistent with this hypothesis, mSOD1 is the slowest protein known that folds by a two-state mechanism, with a folding time of ∼10 s (29,49). High-energy states sampled by the unfolded ensemble, before productive folding, could potentially contribute to chain friction. Such states were observed in several NMR studies of SOD1 that found evidence of nonnative dimer formations (50, 51, 52). Interestingly, all of these states were found to involve the regions involved in or directly adjacent to the cysteines forming the disulfide bond. It is possible that the observation of internal friction in the monomeric construct in our study is related to the observation of similar structural perturbations in the SOD1 dimer.

A potential link between internal friction, aggregation, and amyotrophic lateral sclerosis in SOD1

Intramolecular interactions that give rise to chain friction have the potential to become intermolecular interactions and thereby drive intermolecular aggregation. This possibility is relevant to this study because aggregates of SOD1 have been linked to the invariably fatal neurodegenerative disease amyotrophic lateral sclerosis (ALS) (53). Two lines of evidence support a role for the C-terminus of SOD1 in aggregation and toxicity in ALS. Work by Ivanova et al. (54) showed that only a hexapeptide containing residues 147–153 from β8 accelerated fibril formation of both wild-type metal-free SOD1 and ALS-linked mutants of SOD1. The relevance of a role for β8 in the aggregation of SOD1 is also evident in the observations of Furukawa et al. (55), who found that the segment containing β8 and portions of the preceding Loop VII was the only region resistant to proteolysis in aggregates of ALS variants of SOD1 isolated from mouse motor neurons.

The nascent SOD1 chain before disulfide bond formation, dimerization, and full metalation has several properties that distinguish it from many proteins and would increase its propensity to aggregate. First, 2SH-mSOD1 is only marginally stable, resulting in ∼1% of the population occupying the unfolded state at equilibrium (Table S2; (56)). ALS mutations in 2SH-mSOD1 dramatically destabilize the native state, resulting in a preponderance of the unfolded state at a neutral pH and physiological temperatures (57). The shift to favor the unfolded state of 2SH-mSOD1 is compounded by its exceedingly slow folding reaction, τ ∼10 s, in the absence of a denaturant (29). The long lifetime and preponderance of the unfolded nascent chain in SOD1 ALS variants would enhance the probability for the same sequences responsible for intramolecular friction to form intermolecular interactions and thereby nucleate aggregation (54,55). A nascent SOD1 polypeptide chain, with nonnative collapsed structure at the C-terminus, could be a nearly universal target for therapeutics designed to ameliorate the deterioration of motor neurons in ALS caused by mutant forms of SOD1.

Author Contributions

N.R.C. collected the data, conceived and designed the analysis, performed the analysis, interpreted the results, and wrote the manuscript. C.K. designed the experiments, collected the data, and edited the manuscript. O.B. conceived and designed the analysis, contributed data analytical tools, and edited the manuscript. J.A.Z. designed the experiments and edited the manuscript. C.R.M. designed the experiments, conceived and designed the analysis, interpreted the results, and edited the manuscript.

Editor: Doug Barrick.