Lineweaver-Burk and Linearized Plots

Even though nonlinear regression software (fitting $v_0$ vs. $[S]$ directly by algorithms such as Levenberg-Marquardt) has replaced hand-drawn linear plots as the gold standard for reporting $K_m$ and $V_{max}$ in the peer-reviewed literature, these classical plots persist for excellent pedagogical and diagnostic reasons. In undergraduate and graduate biochemistry laboratories worldwide, students still plot Lineweaver-Burk graphs by hand or in spreadsheets because the visual pattern of intercept and slope changes under different inhibitor types is immediately legible — a competitive inhibitor increases the slope and x-intercept magnitude while leaving the y-intercept fixed, a pattern that becomes the entry point for the whole subject of Enzyme Inhibition Overview. Pharmaceutical kinetics reports and enzyme assay kits from commercial vendors frequently still include a Lineweaver-Burk-style summary figure alongside the nonlinear fit, precisely because reviewers, regulators, and readers can sanity-check a straight line's slope at a glance in a way that is harder to do with a hyperbola. In drug discovery, high-throughput enzyme inhibition screens generate thousands of dose-response curves; while the underlying $K_i$ values are extracted by nonlinear fitting, quality-control pipelines often flag anomalous data by checking whether a quick double-reciprocal transform of a subset of points looks reasonably linear — a fast visual test for gross pipetting or instrument errors before committing computational resources to full nonlinear analysis. In clinical and diagnostic enzymology (for instance, characterizing mutant forms of enzymes implicated in inherited metabolic diseases, such as phenylalanine hydroxylase variants in phenylketonuria), Eadie-Hofstee plots remain popular in research reports because their more even error distribution across the range of $[S]$ makes deviations from simple hyperbolic (Michaelis-Menten) behavior — such as substrate cooperativity in oligomeric enzymes — visually apparent as curvature in the "line," a diagnostic that is much harder to spot in a double-reciprocal plot where such curvature gets compressed into the crowded, error-inflated region near the axes. The Hanes-Woolf plot, having the mildest error distortion of the three, is often recommended in modern kinetics textbooks as the linearization of choice whenever a linear plot must be used at all — for example in teaching laboratories using inexpensive plate-reader data where nonlinear-fitting software may not be readily accessible, or in quick manual estimates during method development before a full kinetic characterization is undertaken. Beyond classical Michaelis-Menten enzymology, closely related linearization logic recurs throughout biochemistry and pharmacology: Scatchard plots for ligand-receptor binding, Woolf-Augustinsson-Hofstee analyses of transporter kinetics, and even certain enzyme-inhibitor $K_i$ determinations all borrow the same core idea of trading a hyperbola for a line at the cost of redistributing experimental noise — precisely the trade-off this article has made explicit and quantitative.

:::quiz[Which linearization plots v₀ directly against v₀/[S]?]

• Lineweaver-Burk
• Hanes-Woolf
• Eadie-Hofstee ✓
• Scatchard-Michaelis

:::quiz[For the Michaelis-Menten equation with Vmax = 100 μM/min and Km = 1.0 mM, what is v₀ at [S] = 4.0 mM?]

• 50.0 μM/min
• 66.7 μM/min
• 80.0 μM/min ✓
• 90.9 μM/min

:::quiz[On a Hanes-Woolf plot ([S]/v₀ vs [S]), what does the slope represent?]

• Km
• 1/Km
• Km/Vmax
• 1/Vmax ✓

Ask the tutor: How would the Lineweaver-Burk plot's slope and intercepts change if a competitive inhibitor were added at a fixed concentration?

Begin with the object being linearized. For a simple one-substrate enzyme obeying steady-state kinetics,

is a rectangular hyperbola in the $(v_0, [S])$ plane. It passes through the origin, rises steeply at low $[S]$, and approaches the horizontal asymptote $v_0 = V_{max}$ as $[S]\to\infty$. The half-saturation point, $v_0 = V_{max}/2$, occurs exactly at $[S]=K_m$ — this is the operational meaning of $K_m$ established in the study of The Meaning of Km and Vmax. The trouble with reading this curve off graph paper is that the asymptote $V_{max}$ is approached only in the limit, so no finite set of measurements ever touches it; extrapolating "by eye" to find where the curve flattens out is notoriously unreliable, and different observers draw different asymptotes through the same scatter of points. The double-reciprocal transform solves this by mapping the asymptote to a finite, exactly reachable intercept. Taking the reciprocal of both sides of the Michaelis-Menten equation converts the curve into a straight line, so the "point at infinity" for $v_0$ becomes an ordinary, extrapolable y-intercept at $1/[S]=0$. This is the entire logic of every classical linearization: relocate the hard-to-see asymptotic behavior into an easy-to-see finite intercept or slope. Each of the three plots discussed here performs this relocation differently, and it is worth holding all three geometries in mind simultaneously, because they will resurface when diagnosing inhibition patterns.

The Eadie-Hofstee and Hanes-Woolf plots achieve the same linearization by different reorganizations of variables, each placing the two parameters into a slope-intercept pair with different sensitivities to experimental scatter — the subject of the derivations below.

The parent relationship, assumed known, is

where:

• $v_0$ is the initial reaction velocity (concentration/time),
• $[S]$ is the substrate concentration (assumed $\gg [E]_{total}$, steady state),
• $K_m$ is the Michaelis constant (same units as $[S]$),
• $V_{max}$ is the limiting velocity at saturating substrate.

The three linear forms are:

where in each case the equation has the form $y = mx + b$, with $(x,y,m,b)$ read off the table below.

Reading: in every case, $V_{max}$ and $K_m$ are recoverable from two independently measurable line features (a slope and an intercept), which over-determines the fit — a useful internal consistency check, since with noise-free data any two of the three quantities (slope, y-intercept, x-intercept) must give the same $K_m$ and $V_{max}$ as the third.

Example 1 — Foundational: constructing a Lineweaver-Burk plot from raw velocity data. An enzyme is assayed at five substrate concentrations, each with a directly measured initial velocity (units: $\mu$M/min for $v_0$, mM for $[S]$), noise-free for this first illustration so the underlying parameters can be checked exactly:

These were generated from $K_m=2.0$ mM and $V_{max}=100\ \mu$M/min; verify with $[S]=2.0$: $v_0=100(2)/(2+2)=50$ — wait, check directly with the formula for each row to confirm consistency. For $[S]=2.0$: $v_0 = 100(2.0)/(2.0+2.0)=200/4=50.00$, not $66.67$ — so these table values instead correspond to $K_m=1.0$ mM, $V_{max}=100$. Confirm: $[S]=2.0 \Rightarrow v_0=100(2)/(1+2)=200/3=66.67$ ✓; $[S]=4.0 \Rightarrow 400/5=80.00$ ✓; $[S]=10 \Rightarrow 1000/11=90.91$ ✓; $[S]=1.0\Rightarrow 100/2=50.00$ ✓; $[S]=0.5\Rightarrow 50/1.5=33.33$ ✓. Good — $K_m=1.0$ mM, $V_{max}=100\ \mu$M/min. Two points on the $1/[S], 1/v_0$ line, say $(2.000, 0.03000)$ and $(0.100, 0.01100)$, give the slope directly:

Since $m=K_m/V_{max}$ and the y-intercept (extrapolating to $1/[S]=0$) is read from the line $1/v_0 = 0.01000(1/[S]) + b$; using the point $(1.000, 0.02000)$:

so $V_{max} = 1/b = 1/0.01000 = 100\ \mu$M/min, and

both recovering the input parameters exactly, as expected for noise-free data — confirming the derivation's algebra without any fitting ambiguity.

Example 2 — Applied: comparing three linearizations on the same noisy data set. A student measures an enzyme's initial velocity at six $[S]$ values, with genuine experimental scatter (units: mM and $\mu$M/min):

Compute the transformed coordinates for each plot. Lineweaver-Burk ($1/[S]$, $1/v_0$):

A linear regression on these six points (computed by standard least-squares on $x=1/[S]$, $y=1/v_0$) gives approximately slope $m\approx 0.01015$ and intercept $b\approx 0.01267$, so

Eadie-Hofstee ($v_0/[S]$, $v_0$):

A least-squares line here (with $v_0$ as the dependent variable) gives slope $\approx -0.83$ and intercept $\approx 76.9$, so $K_m\approx 0.83$ mM and $V_{max}\approx 76.9\ \mu$M/min — close to, but not identical to, the double-reciprocal estimates, illustrating that different linearizations give different fitted parameters from the same noisy data, because each weights the residuals differently. Hanes-Woolf ($[S]$, $[S]/v_0$):

A least-squares fit gives slope $\approx 0.01206$ and intercept $\approx 0.01064$, so

The three methods give $V_{max}$ estimates spanning roughly 77–83 $\mu$M/min and $K_m$ estimates spanning roughly 0.80–0.88 mM from identical raw data — a concrete demonstration that linearization choice matters once real noise is present, because Lineweaver-Burk overweights the low-$[S]$, low-$v_0$ points (where $1/v_0$ blows up), while Hanes-Woolf and Eadie-Hofstee are progressively more forgiving of that region. Nonlinear regression directly on $v_0$ vs. $[S]$, weighting all points equally, is generally the most statistically defensible choice — the linear plots here mainly serve as fast, transparent estimates and diagnostic sanity checks.

Example 3 — Exam/competition level: recovering $K_m$ and $V_{max}$ from partial Lineweaver-Burk information, and detecting an inconsistency. A double-reciprocal plot is reported with the x-intercept at $1/[S] = -0.040\ \text{mM}^{-1}$ and the y-intercept at $1/v_0 = 0.0125\ (\mu\text{M/min})^{-1}$. Separately, a single velocity measurement is available: at $[S] = 25$ mM, $v_0 = 68.0\ \mu$M/min. Determine whether this third data point is consistent with the reported intercepts, and if not, identify the most likely experimental source of the discrepancy. Step 1. From the y-intercept, $V_{max} = 1/0.0125 = 80.0\ \mu$M/min. Step 2. From the x-intercept, $-1/K_m = -0.040\ \text{mM}^{-1} \Rightarrow K_m = 1/0.040 = 25.0$ mM. Step 3. Predict $v_0$ at $[S]=25$ mM using Michaelis-Menten with these parameters:

Step 4. Compare to the reported measurement of $68.0\ \mu$M/min. The predicted value (40.0) and measured value (68.0) disagree by nearly a factor of 1.7 — far beyond ordinary assay noise (typically a few percent for a well-controlled spectrophotometric assay). Step 5. Diagnose the likely cause. Because $[S]=25$ mM happens to equal $K_m$ exactly, Michaelis-Menten predicts $v_0=V_{max}/2=40.0$; the measured value of 68.0 is higher than this prediction, i.e., higher than expected given the fitted line. Two structurally different explanations are worth distinguishing: 1. The line's intercepts were extrapolated from data that did not extend to high enough $1/[S]$, so the reciprocal-space fit is dominated by low-$[S]$ points and mis-estimates $V_{max}$ — a hazard flagged explicitly in Example 2. If the true $V_{max}$ is actually higher (say, closer to 90 $\mu$M/min) and the true $K_m$ smaller, the discrepancy could vanish. 2. Alternatively, the $[S]=25$ mM point may itself be an outlier — contamination, pipetting error, or an assay run outside the linear (initial-rate) time regime, artificially inflating $v_0$. The disciplined response is not to force-fit the outlier into the reported line, but to re-plot all raw data (ideally by nonlinear regression on the untransformed $v_0$ vs. $[S]$ data) and examine whether the $[S]=25$ mM point remains anomalous — exactly the diagnostic mindset that later underlies distinguishing genuine inhibition patterns from bad data in Diagnosing Inhibition from Kinetic Plots.

Hans Lineweaver (1907–2009) was an American physical chemist working at the U.S. Department of Agriculture's research laboratories, where much of his career focused on food chemistry and enzymology relevant to agricultural products — including pioneering work on pectin enzymes, protein denaturation, and the physical chemistry of milk and eggs. Dean Burk (1904–1988) was a biochemist who spent decades at the U.S. National Cancer Institute, working broadly across enzymology, cancer metabolism (including early studies of the Warburg effect on tumor respiration), and radiobiology. Their landmark 1934 paper in the Journal of the American Chemical Society, "The Determination of Enzyme Dissociation Constants," introduced the double-reciprocal plot specifically as a practical graphical tool: in an era without computers, fitting a rectangular hyperbola by eye to noisy data was unreliable, but fitting a straight line with a ruler was fast, reproducible, and teachable. The paper's real innovation was not a new chemical insight but a piece of applied numerical analysis — turning intractable curve-fitting into simple linear regression — and it became one of the most cited papers in the biochemical literature of the twentieth century, despite (or perhaps because of) later recognition of its statistical weaknesses. The competing linearizations bear the names of their own originators: George Hanes and Barnet Woolf, working independently in the 1930s in the UK, are credited with the $[S]/v_0$ vs. $[S]$ plot (Hanes' work appeared in a 1932 Biochemical Journal paper on starch-iodine kinetics, with Woolf's contribution often cited from unpublished correspondence relayed by J.B.S. Haldane); B.H.J. Hofstee and later independently George Scatchard-style analyses (adapted from receptor-binding theory) gave the $v_0$ vs. $v_0/[S]$ plot associated with Hofstee's name and, earlier, with work by Woolf and by Augustinsson — a reminder that this family of linearizations emerged from several hands nearly simultaneously as enzymologists worldwide grappled with the same practical problem of hand-fitting hyperbolic data before the era of digital nonlinear regression.