ABSTRACT

Conventional analysis of enzyme-catalyzed reactions uses a set of initial rates of product formation or substrate decay at a variety of substrate concentrations. Alternatively to the conventional methods, attempts have been made to use an integrated Michaelis-Menten equation to assess the values of the Michaelis-Menten K_M and turnover k_cat constants directly from a single time course of an enzymatic reaction. However, because of weak convergence, previous fits of the integrated Michaelis-Menten equation to a single trace of the reaction have no proven records of success. Here we propose a reliable method with fast convergence based on an explicit solution of the Michaelis-Menten equation in terms of the Lambert-W function with transformed variables. Tests of the method with stopped-flow measurements of the catalytic reaction of cytochrome c oxidase, as well as with simulated data, demonstrate applicability of the approach to determine K_M and k_cat constants free of any systematic errors. This study indicates that the approach could be an alternative solution for the characterization of enzymatic reactions, saving time, sample and efforts. The single trace method can greatly assist the real time monitoring of enzymatic activity, in particular when a fast control is mandatory. It may be the only alternative when conventional analysis does not apply, e.g. because of limited amount of sample.

1. Introduction

The conventional Michaelis-Menten model is commonly used in biochemistry to assess the values of the catalytic constant k_cat and the Michaelis-Menten constant K_M = (k_cat + k₋₁)/k₁ of irreversible enzyme-substrate reactions [1-5]:

(1)

where k₁ and k₋₁ are the forwards and backwards rate coefficients and E is the enzyme, S the free substrate, ES the enzyme-substrate complex, and P the released product. In the case when k₁, k₋₁ >> k_cat, which is known as the approximation of quasi-equilibrium between E and ES [1], and under the assumption of a quasi-steady state, dES/dt = 0, which holds for S₀ >> E, the MichaelisMenten equation is given by [1,6]:

(2)

The initial rate v₀ = dP(t)/dt at t→0 is a nonlinear function of starting substrate concentrations S₀, where E₀ is the total concentration of enzyme, and P(t = 0) = 0 is the initial condition for the product concentration. If v₀ is plotted as a function of S₀, the parameters k_cat and K_M have to be assessed with Equation (2) at t→0 by nonlinear regression. Another way is to linearize Equation (2) using inverted variables like the 1/v₀ and 1/S₀ and then apply linear regression [7]. The disadvantage of the inverted plots is that they are sensitive to errors for small or large values of v₀ or S₀ [5]. Eadie [8] and Hofstee [5] have proposed a linearization of Equation (2) using non inverted variables v₀ and v₀/S₀ to overcome this limitation. Independent of the particular linearization procedure employed to analyze enzyme kinetics all the methods rely on the determination of the initial rates v₀ in a wide range of starting concentrations S₀. We will refer to these methods as v₀-plots. In cases, when measurements in a wide range of S₀ are not possible or not feasible because of restricted availability of the samples, it would be advantageous to establish a new method for assessment of the enzymatic kinetic parameters for a single value of S₀.

Several attempts have been made to assess the values of K_M and k_cat directly from a time course S(t) of an enzymatic reaction at a single S₀. Different forms of parametrically integrated Michaelis-Menten equation, e.g. (S₀ − S) = E₀ k_cat t − K_M ln(S₀/S), have been fitted to progress reaction curves for enzyme catalyzed reactions to determine k_cat and K_M [9-12]. However, the fit of the implicit parametric solution to the explicit S(t)-curves by linear or nonlinear regression was a demanding numerical problem and often resulted in uncertain values of k_cat and K_M. Only beginning with the work of Schnell and Medoza [13], the explicit solution of the Michaelis-Menten equation for P(t) in terms of Lambert-W function became available, and could provide more accurate results, since no more approximations, than already made in the Michaelis-Menten model, were required. For instance, Goudar and colleagues [14-16] have used the explicit solution of the Michaelis-Menten equation, Equation (3), to analyze single progress curves of enzymatic reactions. The new method promised to be fast and sample saving. Yet surprisingly, the explicit solution of the Michaelis-Menten equation has up to now only occasionally been adopted to estimate k_cat and K_M, even though its theory is well documented [13].

Walsh et al. [17] have pointed out that attempts to use the explicit solution of the Michaelis-Menten Equation (2) to describe reaction time courses have so far met with little success. It is well known, that nonlinear regression, in our case with the Lambert-W function, cannot be started without specifying the initial input parameters K_M,input and k_cat,input, even if the whole progress curve is accurately measured. If K_M,input and k_cat,input are significantly different to the real K_M and k_cat, the nonlinear fit converges very slowly, if at all, and the estimated values of K_M and k_cat are imprecise. To our knowledge, the problem of the strong dependency of the fitted constants K_M and k_cat on the input parameters K_M,input and k_cat,input has not be properly addressed so far. In the present paper we examine the problem and demonstrate, that because of a weak convergence, previous fits of the explicit solution of the Michaelis-Menten equation to a single trace P(t) could not provide a reliable estimation of K_M and k_cat. We propose an alternative fitting method using an explicit solution of the Michaelis-Menten equation in terms of the transformed Lambert-W function, Equation (8), which is much less prone to uncertainties in input parameters than previous single-trace methods. We compare the method in simulated and experimental conditions with conventional v₀-plots to demonstrate that the new approach could be developed to a standard method for a time and sample saving characterization of enzymatic reactions. For that we use simulated data following the ideal irreversible Michaelis-Menten kinetics as well as experimental data on the reaction of cytochrome c oxidase (CcO). At high salt conditions and in the presence of an excess of oxygen, the CcO-reaction meets the single substrate Michaelis-Menten model and no product inhibition takes place. Additionally, the CcO-reaction guarantees full irreversibility of the enzymatic reaction due to formation of water from oxygen during the catalysis [18].

2. Material and Methods

2.1. Preparation of Enzyme Cytochrome c Oxidase

Cytochrome c oxidase (CcO) is the terminal complex of the membrane-bound respiratory chain and spends electrons from ferrous cytochrome c for the reduction of oxygen to water. CcO from Rhodobacter sphaeroides was expressed and purified, using 2l cell culture flasks in a gyratory shaker followed by 12 h solubilisation in detergent [19]. Purified protein was stored in phosphate buffer (50 mM, pH 8 and 0.01% dodecylmaltoside) at −80˚C until use. The enzyme concentration was determined from UV/VIS spectra of air-oxidised and sodium dithionite reduced samples, using the difference of the differential extinction coefficients Δε₆₀₅ − Δε₆₃₀ = 24 mM⁻¹·cm⁻¹ at the wavelengths λ = 605 nm and 630 nm [19]. Ferrous cytochrome c from horse heart (purity > 95% from Fluka) was used without further purification. A solution of cytochrome c was reduced in a fivefold molar excess of sodium dithionite. Reductant was separated via a 5 ml HiTrap desalting column (GE Healthcare) on an ÄktaPurifier FPLC (GE Healthcare) and the protein was stored at −80˚C. The concentration of cytochrome c was determined spectrophotometrically by recording the reducedminus-oxidized difference spectrum of the sodium dithionite reduced and ferricyanide oxidized sample and using the differential extinction coefficient Δε₅₅₀ = 19 mM⁻¹·cm⁻¹ at λ = 550 nm [20].

2.2. Measurements of Enzyme Kinetics

Kinetic experiments of electron transfer from cytochrome c to CcO were performed by the stopped-flow technique [19]. The oxidation of ferrous cytochrome c was monitored at λ = 550 nm by a miniature fibre-optic spectrometer (USB 2000 from Ocean Optics). A 50 nM solution of oxidase was mixed with a 5 to 120 μM solution of cytochrome c, both buffered in a 50 mM phosphate buffer, pH = 6.5, containing 0.05% dodecylmaltoside and 100 mM KCl.

2.3. Simulation of Enzyme Kinetics

The normalized time trace p(t) = P(t)/S₀ of an enzymatic reaction was simulated with the integrated MichaelisMenten equation, Equation (3), for 425 equidistant time points in the range 0 ≤ t ≤ 65 s for the parameters K_M = 220 µM, k_cat = 650 s⁻¹ and E₀ = 50 nM, i.e. for the parameters close to those of the CcO reaction. Pseudorandom noise was generated by the “rnd”-function of the software Mathcad 2001 Professional, MathSoft, Inc., and added to the analytical trace p(t). The amplitude of the noise was ranging from 0 up to 4% of the maximum value p = 1.

2.4. Description of the Linear and Nonlinear Regressions

For the linear regression we used the “neigung”-function of the Mathcad 2001 Professional, MathSoft, Inc. The initial rates v_0,exp were determined by linear regression using the first ten data points of the experimental and the computer simulated trace p(t). The same linear regression was applied to Lineweaver-Burk and EadieHofstee plots for estimation of the values of K_M and k_cat. The conventional nonlinear regression with Equation (2), as well as the nonlinear regression with the integrated MichaelisMenten equation was based on the iterative LevenbergMarquardt method implemented in the “minfehl”-function of the Mathcad 2001 Professional.

3. Results and Discussion

3.1. Simulation of Enzymatic Reactions

Our simulations of enzymatic reactions employ an integration of the Michaelis-Menten Equation (2) yielding a closed analytical formula for P(t) in terms of the Lambert-W function W(x) [13], in contrast to frequently applied implicit parametric solutions:

(3)

Note that Equation (3) refers to the fully irreversible enzymatic reaction as constituted by Equation (1). The progress curve P(t) is simulated by Equation (3) at S₀ = 30 µM, K_M = 220 µM, k_cat = 650 s^–1 and E₀ = 50 nM. The result is an ideal product P(t)-curve following the onesubstrate Michaelis-Menten model. Then Equation (3) is fitted with K_M and k_cat to P(t)-curve by nonlinear regression as described in Section 2.4. To study the robustness of the method against deviations of initial input parameters from the true values, we started the nonlinear regression with K_M,input ≠ K_M = 220 µM and k_cat,input ≠ k_cat = 650 s^–1. The extents of the relative variations of the input parameters ΔK_M,input/K_M and Δk_cat,input/k_cat are quantified by

(4)

(5)

The relative deviations of the estimated parameters δK_M,est/K_M and δk_cat,est/k_cat are characterized by

(6)

(7)

The calculation showed, that the relative deviations of the input parameters in the range of –0.5 ≤ ΔK_M,input/K_M ≤ 0.5 and −0.5 ≤ Δk_cat,input/k_cat ≤ 0.5 cause practically equal relative deviations of both estimated parameters: . Therefore, we combined the results for relative deviations of the estimated parameters K_M,est and k_cat,est in a single plot, Figure 1.

The results of a direct fit of Equation (3) to the time course P(t) are summarized in Figure 1(a). If the value of k_cat,input is somewhat different to the exact value, both K_M,est and k_cat,est cannot be determined correctly. On the other hand, K_M,est and k_cat,est are very little sensitive to variations of K_M,input. It means that the fit of Equation (3) to P(t) cannot provide convergence to correct values of K_M and k_cat, if the initial input parameter k_cat,input somenwhat deviates from the true value of 650 s^–1. As far as we know, all previous attempts to determine K_M,est and k_cat,est explicitly using Equation (3) were based on the similar approach as tested above [14-16].

We solve this problem by employing the fact that the apparent rate constant in Equation (3) is given by m = k_cat·E₀/K_M [2]. In terms of m, Equation (3) takes the form:

(8)

where s = S₀ /K_M is the dimensionless parameter. Applying Equation (8) to fit kinetics of an enzymatic reaction by nonlinear regression, we render the procedure robust against errors in initial input values of K_M and k_cat. Using the transformed Equation (8), we obtain very small relative errors for the estimated parameters δK_M,est/K_M ≈ δk_cat,est/k_cat in the range between −7.5 × 10⁻³ and 1.5 × 10⁻², Figure 1(b). For the most values of K_M,input and k_cat,input studied here we calculate even zero deviations. For characterization of the quality of the fit, we use the standard deviation:

(9)

where N = 425 is the number of the equidistant time points t_i at which the function p(t_i) is sampled. Note that when the non-transformed Equation (3) is used, the location of local minima of SD is on a line diagonal to ΔK_M,input/K_M and Δk_cat,input /k_cat axis, see Figure 2(a). This is very unfavorable for nonlinear regression and leads to the weak convergence and mainly to a wrong local minimum. The large deviation from the global minimum causes large uncertainty in K_M,est and k_cat,est. The transformed Equation (8) applied in the new (m, s)-coordinate system exhibits much faster convergence, as Equation (3) in the (K_M, k_cat)-coordinates, Figure 2(b). Contrary to Equation (3), the fit of the apparent rate constant m at a constant parameter s yields the correct m-coordinate of the global minimum of SD. Fitting s at a correct m-coordinate of the minimum greatly facilitates convergence to the well-defined global minimum of SD shown by the arrow in Figure 2(b). The estimated coordinates m_est, s_est of the global minimum can be easily recalculated into the parameters K_M,est and k_cat,est.

3.2. Advanced Algorithm for Single Trace Regression

The impact of the precision of the input parameters K_M,input and k_cat,input on the errors on the estimated values of K_M,est and k_cat,est becomes stronger with increasing noise level of data. Previously, K_M,input and k_cat,input were determined through a linearization of Equation (3) resulting in uncertain estimates [15,16]. Here we avoid the linearization by using the fact that the simple exponential function

(10)

is an approximation to Equation (8), [2]. First, we estimate the input value of the apparent rate constant via fitting Equation (10) by regression to experimental or simulated p(t) curves. Calculations with simulated p(t) have shown (data are not presented) that for a reliable estimation of it is recommendable to sample p(t) up to the maximum time. The second input parameter for the dimensionless parameter is obtained using the first derivative of Equation (8) for t → 0:

(11)

The value (dp(t)/dt)|t→0 ≈ v_0,exp is estimated by usual linear regression applied to ten or twenty initial points of the p(t) curve. Rearrangement of Equation (11) gives the second input parameter s_input in the form:

(12)

It is recalled that linear regression for v_0,exp and nonlinear regression of the exponential Equation (10) for m_input yield intrinsically wrong estimates of m_input and s_input. However, the estimates are good enough to serve as input parameters for the nonlinear regression with the transformed Equation (8); cf. next Section 3.3. On this way, the determination of m_input and s_input can be easily automated; see the reference in the Appendix 1.6. Finally, the searched kinetic constants are given by K_M,est = S₀/s_est and k_cat,est = m_est·K_M,est/E₀.

3.3. Robustness of Single Trace Regression to Noise

For testing of robustness of single trace regression to error-prone data, noise was added to the error-free function p(t), Equation (8), in form of the pseudo random error:

(13)

where e is the noise amplitude ranging from zero to 0.04, and x is the pseudo-random variable varying in the range of −1 to 1. The function p_err (t) is calculated in N = 425 equidistant points in the time range of 0 to 65 s. The estimates m_est and s_est are determined as averages of five samples of noise of p_err (t). As a measure for the quality of the estimations we use the standard deviation, Equation (9). The resulting SD values are small for all noise levels: SD = 0 at e = 0, SD = 5.8 × 10⁻³ at e = 0.01 and SD = 0.024 at e = 0.04. Symmetric distribution of the relative deviations of the estimated parameters δK_M,est /K_M and δk_cat,est/k_cat around zero suggests no systematic error in the resulting K_M,est and k_cat,est values; see Figure 3, right panels.

As mentioned above, the values of δK_M,est/K_M and δk_cat,est/k_cat are practically equal to each other in the whole range of ΔK_M,input/K_M and Δk_cat,input/k_cat studied here; therefore, they are presented together in the same graph, Figure 3 (right panels). The estimates of K_M,est and k_cat,est, averaged over the results for 25 input pairs of K_M,input and k_cat,input, are summarized in Table 1. Note that the estimates are very close to the actual values K_M = 220 μM and k_cat = 650 s⁻¹, even at e = 0.04.

3.4. Conditions for the Best Convergence

Besides of the noise level of P(t), the quality of analysis depends on the absolute value of the initial substrate concentration S₀. To demonstrate the correlation between noise-level and initial substrate concentration, the relative deviation of estimated K_M and k_cat values as a function of S₀ is depicted in Figure 4(a) for three different levels of noise.

The absolute values of the relative deviations of the estimated parameters K_M and k_cat became larger with decreasing S₀ and increasing noise amplitude e. The dependence of |δK_M,est|/K_M and |δk_cat,est|/k_cat on S₀ can be referred to large rounding errors of the nonlinear regression at small values of initial substrate concentrations, especially when S₀ << K_M (here at about S₀ = 10 µM), and at larger level of noise. Since the parameter s in Equation (8) interrelates S₀ to K_M, we estimated for a given parameter s the ranges of the apparent rate constant m, which condition a correct determination of K_M and k_cat. Reliable results of the fit we obtain for s = 0.01 and m between 0.02 and 1.2 s⁻¹, for s = 0.1 and m between 0.024 and 1.9 s⁻¹ as well as for s = 1.0 and m between 0.07 and 1.2 s⁻¹. In that range of the parameters the

computation algorithm for Lambert-W function and nonlinear regression implemented in Mathcad 2001 Professional yield exact values of K_M,est and k_cat,est (data are not shown). The precision of the value of S₀ appears to be very important for a correct analysis of progress curves, because of the direct dependence of s on S₀. Already the relative deviation ΔS₀/S₀ = (S_0,meas − S₀)/S₀ of the measured value S_0,meas from the exact value S₀ in the range ± 0.5% causes a hundredfold larger relative error in the estimated values of K_M,est and k_cat,est, Figure 4(b). In order to keep the relative error in K_M,est and k_cat,est within ± 20% error margin, the relative deviation ΔS₀/S₀ should be in the range between −0.18 and +0.21%. Therefore, the single trace regression requires very precise values of the starting substrate concentrations. Generally, the demanded precision of S₀ should be better than about ±0.1%, which is nowadays available in most laboratories.

3.5. Testing v₀-Plots and Single Trace Regression with Simulated Data

We tested the limits and quality of the new single trace regression as well as commonly used v₀-plots using computer generated data with pseudo-random noise. One source of errors in v₀-plots is an inaccurate experimental determination of v₀. In our simulations, we determine the realistic error in v₀ using a computer simulated trace p(t) with a known level of pseudo-random noise, as opposed to previous computer tests, in which a fixed error was added to v₀ [5]. Three classical v₀-plots (LineweaverBurk, Eadie-Hofstee and nonlinear) as well as the single trace regression method were applied to the same data set. The results are presented in Table 2 (see also Tables 4 and 5 in Appendix).

Table 2 clearly shows that the single trace regression with Equation (8) yields the most precise results for K_M and k_cat when error prone simulated data are analyzed.

The nonlinear v₀-plot yields good estimates coinciding with actual values within their standard deviations. The Eadie-Hofstee plot yields less accurate results, the estimate does not compare to the true value k_cat, even if within the error margin. The Lineweaver-Burk plot yields unrealistic negative values of K_M and k_cat. Presumably, an uneven distribution of error over S₀-values leads to the unrealistic results. This could be to a certain degree corrected by disregarding data points for smaller S₀-values, a strategy which is frequently employed by experimentalists exploiting the Lineweaver-Burk plot. Our analysis adheres strictly to all values of the plots to ensure the full comparability of all the methods. We also did not perform a weighting of the data points corresponding to their precision, which additionally may have led to improved results. Another reason for the failure of the Lineweaver-Burk plot is the narrow width of the analyzed S₀ range between 3 and 100 μM, compared to the K_M-value of 220 μM. Generally, a wider range of S₀- values used for analysis increases precision of results of all v₀-plots considerably [5]. Regarding this fact, our new single trace regression method could be of great benefit, especially in cases, where a wide range of S₀ values cannot be addressed experimentally. As expected, results for a wider range of staring concentrations S₀ are more precise, than those for the narrow range of S₀-values (see Figure 7 and Table 4 of Appendix). The reason for the notably more precise results of the single trace regression compared to the three v₀-plots in Table 2 could be the fact that the initial rates v₀ are usually underestimated in their experimental determination (cf. Appendix Figure 6). The effect of the underestimation of v₀ on the results of the v₀-plots is actually hardly predictable. Dependent on the actual noise level, the estimated values of K_M and k_cat can be systematically shifted either to lower or to higher values. In that case, the small error margins of K_M and k_cat are misleading, as demonstrated for the k_cat value determined by the Eadie-Hofstee plot in Table 2. Differently to the v₀-plots, the results of the new single trace method do not show any systematic deviation from the actual K_M and k_cat values at all three levels of noise studied here, see Figure 3. In principle, the assessment of K_M and k_cat can be improved by repeating the single trace experiments and averaging, as opposed to the v₀-plots using systematically underestimated values of v_0,exp. Therefore, we may conclude that the new method has a potential to yield more precise results than the conventional v₀-plots.

3.6. Testing Single Trace Regression with Experimental Data

To gauge the conventional methods to the single trace regression analysis under experimental conditions of enzymatic catalysis we have chosen the reaction of CcO with ferrous cytochrome c, because at high salt conditions and surplus of oxygen it meets the single substrate Michaelis-Menten model and guarantees full irreversibility of the enzymatic reaction due to formation of water from oxygen during the catalysis as well as the exclusion of product inhibition [18]. The high ionic strengths conditions lead further to slower kinetics of this enzymatic reaction compared to the optimum rate at low ionic strength; see, e.g., review by Cooper [18].

Initial reaction rates v_0,exp are measured spectrophotometrically at different starting concentrations S₀ of the substrate, ferrous cytochrome c. The corresponding Lineweaver-Burk, Eadie-Hofstee and nonlinear v₀-plots are presented in Figures 5(a)-(c).

The resulting kinetic parameters K_M and k_cat of the plots are summarized in Table 3.

A quantitative comparison of the simulated product curves p(t) at different levels of noise with the experimental p(t) by Equation (9) shows that the experimental level of noise (compared to Table 1) allows for a definite determination of K_M and k_cat by all three v₀-plots studied here. A representative result of a fit of Equation (8) to a single trace of an enzymatic reaction is shown in Figure 5(d). The sevenfold repetition of the single trace regression at seven different S₀ values (see Appendix, Table 5) enables computation of standard deviations of K_M and k_cat in Table 3. It is apparent from the Table 3 that the Lineweaver-Burk plot yields the most imprecise results, what is in line with the conclusion of Hofstee [5]. The nonlinear v₀-plot provides the smallest standard deviations of k_cat compared with single trace regression or Lineweaver-Burk and Eadie-Hofstee plots. Within the relative large error margin of K_M and k_cat, the results of the single trace regression are at least consistent with the results of the v₀-plots. The variations of the kinetic parameters K_M and k_cat determined by single trace regression at different substrate concentrations S₀ reflect the limits of the method caused by experimental errors; see Table 5 in Appendix. However, the experimental result itself does not suggest any systematic deviations of K_M and k_cat with S₀.

In our experiment it is not clearly evident whether the single trace regression or the v₀-plots are generally more precise in determination of k_cat and K_M. At least we demonstrate here that single trace regression leads to realistic estimates, which are in good agreement with the values of the v₀-plots.

4. Conclusions

The analytical integration of the Michaelis-Menten Equation (2) in terms of the transformed Lambert-W function, Equation (8), provides a reliable tool to determine the Michaelis-Menten K_M and turnover k_cat constants from the analysis of a single reaction trace. Supported by our advanced nonlinear regression algorithm, this new method allows determining K_M and k_cat much quicker than by conventional linearization or by nonlinear plots using a set of (v_0,exp, S₀) pairs.

The method can be easily implemented as a time and sample saving tool in the online characterization of enzymatic activity. The kinetic parameters K_M = 219 ± 4

μM and k_cat = 790 ± 120 s^–1 determined with Equation (8) for the CcO reaction are quite conform to results of conventional v₀-plots. The fact that no systematic errors occur in K_M and k_cat in the large range of S₀ tested here suggests that application of the single trace method at S₀ < K_M can complement, if not replace, the conventional methods requiring measurements in the larger range of S₀ (0 < S₀ < 2 K_M, Hofstee (1959), [5]). Note that for some enzymatic reactions, like oxidation of cytochrome c by CcO studied here, the region S₀ > 0.5 K_M is experimenttally not accessible because the absorption of cytochrome c becomes too strong for correct spectrophotometric determination.

We can therefore expect that the new single trace nonlinear regression will apply to all single substrate irreversible enzymatic reactions whenever a reliable, precise and fast assessment of the kinetic parameters K_M and k_cat is demanded. The single trace regression with the transformed Lambert-W function can be adopted to analyze reversible enzymatic reactions [21,22], as well as to enzyme kinetics of multiple alternative substrates [23]. In any case, the method will provide a new tool for biotechnology that saves sample and time by reducing the number of experiments at various S₀ mandatory for the determination of K_M and k_cat by v₀-plots.

REFERENCES

Appendix

1. Comparison of the Approaches to Assess K_M and k_cat

1.1. Linearizations of the Michaelis-Menten Rate Equation

Enzyme kinetics is most commonly analyzed by measuring the initial reaction rate v₀ at different substrate concentrations S₀. A number of linearization procedures are available for the evaluation of such datasets [3]. Hofstee [5] has pointed out that there is no real basis for the continued use of inverted linearization, like the 1/v₀ versus 1/S₀, the so-called Lineweaver-Burk plot [7]:

(14)

because the plot is sensitive to errors for low and high values of v₀ or S₀. Eadie [8] and Hofstee [5], proposed to use the v₀ versus v₀/S₀ plot, now commonly known as Eadie-Hofstee plot:

(15)

The maximum rate of the product formation is conventionally defined by

(16)

1.2. Non-Linearized Michaelis-Menten Equation

Beginning with the wide availability of powerful computers and nonlinear regression software, the Eadie-Hofstee and other linear plots have been partly superseded by nonlinear regression method that should be significantly more accurate and no longer computationally inaccessible. The nonlinear regression method uses the Michaelis-Menten Equation (2) at t→ 0 in the nonlinear form

(17)

1.3. Error in the Initial Reaction Rate v₀

Practical determination of v₀ always leads to underestimated values of the initial reaction rate v_0,exp because of the essential non-linearity of the product function P(t) (Figure 6). For instance, if the idealistic curve P(t) is simulated with Equation (3) in N = 425 equidistant points t_i from the time region 0 £ t/s £ 65, the sampling time interval is Δt = t_i+1 – t_i = 0.153 s. If the two first points are used, the “experimental” value v_0,exp = (P(t₁) − P(t₀)) / Δt = 3.866 µM/s is by −0.9% smaller than the correct value v₀ = 3.90 µM/s calculated by Equation (17). It is clear, the shorter the interval Δt = t₁ − t₀, the smaller the difference between v_0,exp and v₀. Since in experiment the first time point t₁ cannot be set arbitrarily close to t₀ = 0, the experimental value v_0,exp is necessarily smaller than v₀. In a more realistic case, when ten initial points are used to calculate v_0,exp by linear regression, the underestimation of v_0,exp is by −8.2%. It is important to realize, that the initial part of P(t) is an essentially nonlinear function, see insert in Figure 6. Therefore, linear regression does not apply to determine v₀. Alternatively, v_0,exp can be obtained by fitting a second degree polynomial equation P(t) = at² + bt + c to the data by nonlinear regression [4]. Extrapolation of the first derivative of the equation to t→ 0 yields v_0,exp = b. The nonlinear regression can improve precision of the assessment of v_0,exp considerably. Still, the results on v_0,exp are very prone to errors in initial estimates of the coefficients a, b and c, as well as to the number of data points and time range taken for the fit. For instance, when ten initial points are used to calculate v_0,exp, Figure 6, at the initial parameters a = –0.2 s⁻², b = 3 s⁻¹ and c = 0, the underestimation of v_0,exp is only −0.14%, at a = −0.5 s⁻², b = 1 s⁻¹ and c = 0 the underestimation is −0.16%. When twenty initial points are used to calculate v_0,exp, the underestimations are −0.56% and −0.58%, respectively.

1.4. Testing v₀-Plots and Single Trace Regression with Simulated Data

For a computational testing of commonly used v₀-plots and the single-trace regression we used narrow and wide ranges of starting substrate concentrations S₀ (c.f. 2.3 and 2.4). A broader S₀-range, up to ~2 K_M, is generally favorable for an analysis based on v₀-plots compared to the analysis in a narrow range of S₀ used for the results presented in Table 2 [5]. In Figure 7 the Eadie-Hofstee (a,d), Lineweaver-Burk (b,e) and nonlinear regression

v₀-plots (c,f) are shown for the S₀-range from 0 to 100 µM (left side) and from 0 to 500 µM (right side), respectively. Notably, the K_M-value for the simulations was set to 220 µM. The diagrams are based on data generated by the simulation of the enzymatic product curve P(t) with Equation (3) and a noise magnitude of 2 µM.

It is evident in Figure 7 that the data points obtained in the broad range of S₀ (d,e,f) deviate less from the regression curves compared to the respective plots obtained for the narrower S₀-range (a,b,c). This observation is consistent with results of the corresponding plots presented in Table 4. Standard deviations of the K_M and k_cat values in Table 4 are significantly smaller compared to the corresponding deviations in Table 2. Additionally to Table 2, in Table 4 the results of simulations at the noise-magnitudes of 0 and 2 µM are presented too. However, the general features of the impact of the S₀-range on quality of the analysis are throughout the same as in Table 2. The Lineweaver-Burk plot does not yield unphysical, negative results, yet provides the most imprecise results compared to other v₀-plots. The Eadie-Hofstee and the nonlinear regression plot yield considerably more precise results than the Lineweaver-Burk plot. However, all v₀-plots are not able to correctly determine the true values of K_M and k_cat in the range of their standard deviations. At low noise-magnitudes, the results of the v₀-plots are generally overestimated. With increasing noise-level, this effect may be (over-) compensated by imprecision of results.

As in the case of the narrow S₀-range, the single trace regression method yields reasonable results for all investigated noise-magnitudes and shows only nonsystematic errors of the determined K_M and k_cat values (see Table 4).

1.5. Testing v₀-Plots and Single Trace Regression with Experimental Data on CcO

The single trace regression method is able to determine K_M and k_cat for every single time course of an enzymatic reaction at a single S₀-value, contrary to commonly used v₀-plots. In our case, the K_M and k_cat parameters are determined for time traces of the enzymatic reaction of CcO with ferrous cytochrome c at seven different S₀- values ranging from 2.68 to 59.35 µM. The results of the new regression method are presented in Table 5; the corresponding mean values and standard deviations of K_M and k_cat are shown in Table 3. For the experimental curves no systematic deviations of the resulting K_M and k_cat are observed, as for the simulated data.

1.6. Mathcad Code for Michaelis-Menten-Lambert Analysis of a Single Progress Curve of Enzyme Kinetics

The complete Mathcad code for analysis of single progress curves of enzymatic reactions with the MichaelisMenten-Lambert equation can be found on the homepage of the corresponding author:

http://www.uni-bielefeld.de/chemie/arbeitsbereiche/pc3-hellweg/work/sergej/MathcadCode.pdf The Mathcad code begins with a generation of the Lambert-W function W(x). For this purpose we use the standard Mathcad function “wurzel” to solve the equation ye^y – x = 0 with an initial value of y = 0.1 and tolerance = 10⁻¹⁵.

a) The second part of the code starts with an input of the total substrate concentration S₀ as well as the total enzyme concentration E₀, and the upload of an experimental product progress curve P(t).

b) An initial value of the apparent rate constant m is estimated from the experimental normalized product curve p(t) = P(t)/S₀. Here we use the exponential approximation and the 2/3-amplitude criterion to estimate the characteristic relaxation time t = 1/m of p(t). To obtain a more precise estimate of m we use the initial estimate of m further as an input parameter for a nonlinear fit of p_exp(t) to the experimental product curve p(t). Therefore, the standard function “Minfehl” implemented in Mathcad is applied.

c) An initial reaction rate v₀ of the product curve p(t) is determined by the standard function “neigung” using the first 11 points of p(t).

d) An estimate of the dimensionless input parameter s is calculated by s = |m/v₀ − 1|.

e) Exact values of the Michaelis-Menten parameters K_M and k_cat are determined by the “Minfehl” function using the experimental p(t)-curve and the estimates of m and s as input parameters.

In the third part, the calculated Michaelis-Menten parameters K_M, k_cat and V_max are summarized together with results of the fit of the integrated Michaelis-Menten Equation (8) to the experimental p(t)-curve.

Abbreviations

CcO: cytochrome c oxidase;

E₀: Total enzyme concentration;

S₀: Starting substrate concentration;

k_cat: Catalytic constant, a key parameter in enzymatic analysis;

K_M: Michaelis-Menten constant, a key parameter in enzymatic analysis;

K_M,input, k_cat,input: Input parameters for K_M and k_cat for their estimation by nonlinear regression;

K_M,est, k_cat,est: Estimated values of K_M and k_cat by nonlinear regression;

m: The apparent rate constant, defined as m = k_cat·E₀/K_M;

s: The dimensionless parameter, defined as s = S₀/K_M;

P(t): Progress curve of the product concentration of an enzymatic reaction;

p(t): Normalized product concentration, defined as p(t) = P(t)/S₀, ranging from 0 to 1;

v₀: Initial reaction rate of the progress curve;

v₀-plots: Methods for analyzing enzyme kinetics based on a variety of initial rates v₀ at different S₀.