Dissociation Constant (Kd): What It Means and How to Measure Binding Affinity

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-10-02

The dissociation constant, Kd, is the equilibrium constant for the breakdown of a complex into its parts. For a simple bimolecular reaction A + B ⇌ AB, it is written Kd = [A][B]/[AB], where the brackets are molar concentrations at equilibrium [2]. A small Kd means the complex holds together well; a large Kd means it falls apart easily.

You will meet Kd in drug discovery reports, biophysical characterization papers, assay development protocols and docking write-ups. It shows up whenever someone claims that two molecules "bind" or "interact." Knowing what the number actually measures, and what it does not, keeps you from over-reading a figure or repeating an experiment that was never at equilibrium in the first place.

Quick Answer

Kd is the equilibrium dissociation constant for a binding reaction, with units of molarity (M) [2].
For 1:1 binding, Kd = koff/kon, where kon is the association rate constant (M⁻¹ s⁻¹) and koff is the dissociation rate constant (s⁻¹) [2].
Lower Kd means tighter binding. A Kd of 1 nM is a thousand-fold tighter than 1 µM.
Kd equals the free concentration of the titrated partner at which half the binding sites are occupied [2].
Kd is not the same as Ki or IC50. Ki is a thermodynamic constant for an inhibitor complex; IC50 is an assay-dependent concentration that shifts with substrate and enzyme levels [6].
A Kd is only meaningful if you report the temperature, buffer and salt, because affinity changes with conditions [1].

What Kd Actually Describes

Start with the reaction. Two species, A and B, collide and form a complex, AB. The reaction runs both ways. At equilibrium the rates match, and the ratio of concentrations settles at a fixed value:

$$K_d = \frac{[A][B]}{[AB]}$$

The brackets are equilibrium concentrations in mol/L, so Kd carries units of M. It is the reciprocal of the association equilibrium constant Ka, which has units of M⁻¹ [2].

The size of the ratio tells you the physical picture. If Kd is 1 nM, then at equilibrium the product [A][B] is tiny compared with [AB], so almost everything sits in the complex. If Kd is 1 mM, most of A and B stay free. Tight binding is a small number.

Kd also has a kinetic definition. For one-step, non-cooperative 1:1 binding:

$$K_d = \frac{k_{off}}{k_{on}}$$

Here kon is second order with units M⁻¹ s⁻¹, and koff is first order with units s⁻¹. The koff term is the probability per unit time that a given complex falls apart [2]. This is why kinetics and thermodynamics are two views of the same event. A long-lived complex has a small koff, and if kon is near the diffusion limit, that small koff is what produces a small Kd.

Association rate constants for typical proteins cluster in the range of 10⁶ to 10⁷ M⁻¹ s⁻¹, set by collision rates that depend on diffusion and the size of the binding interface [2]. Because kon varies over a narrow window, koff usually decides affinity. An interaction with Kd near 1 µM tends to have koff around 1 s⁻¹, a half-life near 0.7 s. Push Kd to 1 nM and koff drops to roughly 0.001 s⁻¹, with a half-life over 10 minutes [2]. The half-life follows first-order decay:

$$t_{1/2} = \frac{\ln 2}{k_{off}} = \frac{0.693}{k_{off}}$$

That relationship matters when you design a wash step. If your complex has a half-life of seconds, a five-minute wash will erase the signal.

Reading a Binding Curve

The standard way to extract Kd is a titration. Hold one partner at a fixed total concentration, vary the other, and measure how much complex forms. For simple 1:1 binding, the fraction of the fixed component that is bound follows a hyperbola in the free concentration of the titrated partner:

$$\text{fraction bound} = \frac{[L]_{free}}{K_d + [L]_{free}}$$

Half the sites are occupied when [L]free equals Kd [2]. That single sentence is the most useful interpretation of the number. Kd is not "the concentration where binding starts" or "the concentration where binding is maximal." It is the midpoint of the curve.

The hyperbola assumes free ligand is approximately equal to total ligand. That holds when the fixed reactant sits well below Kd. Keeping the fixed partner at least ten-fold below Kd is fine, because then essentially all of the titrated partner stays free [2]. Jarmoskaite and colleagues call this the binding regime, where the concentration giving half-saturation equals Kd. In the intermediate regime, where the labeled species exceeds Kd, you need more titrant to reach half-saturation, and the quadratic binding equation that accounts for bound partner applies [1].

There is a third regime that ruins experiments. When the limiting component sits far above Kd, the curve reports stoichiometry instead of affinity, and no amount of curve fitting recovers a true Kd [1]. Simulations suggest that up to about 10-fold excess of the limiting component over Kd still gives reasonably well-defined values with the quadratic equation, and up to 100-fold only with minimal noise [1]. Isothermal titration calorimetry commonly operates outside the binding regime and fits with the quadratic equation or an equivalent formulation [1].

Equilibration is the other half of the problem. When one partner is in large excess, the approach to equilibrium follows kequil = kon[P] + koff, so equilibration is slowest at the lowest titrant concentration [1]. Set your incubation time from the low end of the concentration range, not the high end. Five half-lives of the equilibration reaction is the recommendation; three half-lives reach 87.5% completion [1]. Assuming a diffusion-limited kon of 10⁸ M⁻¹ s⁻¹, a 1 pM interaction needs about a 10 hour incubation to equilibrate, while a 1 µM interaction needs about 40 ms [1]. Slower kon values stretch those times further.

Worked Example

Take simple 1:1 binding with Kd = 10 nM at 25 °C (T = 298.15 K, R = 8.314462618 J/mol/K, 1 kcal = 4.184 kJ).

Fraction bound. Using f = [L]/(Kd + [L]) with free ligand:

[L] free (nM)	Fraction bound
1	0.0909
3	0.2308
10	0.5000
30	0.7500
100	0.9091
300	0.9677
1000	0.9901

Rearranged, [L] = Kd × f/(1 - f), so 10% bound needs 1.1 nM, 90% needs 90 nM and 99% needs 990 nM. Notice how far above Kd you must go to saturate the sites. This is why a curve that never plateaus cannot define its own amplitude.

Free energy. With a 1 M standard state:

$$\Delta G^\circ_{bind} = RT \ln\left(\frac{K_d}{1\ \text{M}}\right)$$

Plugging in: 8.314462618 × 298.15 × ln(1e-8) = -45,664 J/mol = -45.66 kJ/mol = -10.91 kcal/mol. The negative sign means binding is favorable. The dissociation free energy would be +10.91 kcal/mol. For comparison, Kd = 1 µM, 1 nM and 1 pM give -8.19, -12.28 and -16.37 kcal/mol, and each 10-fold step is 1.364 kcal/mol.

Ligand depletion. The exact quadratic is fraction bound = {(Pt + Lt + Kd) - sqrt[(Pt + Lt + Kd)² - 4 Pt Lt]}/(2 Pt), where Pt is the fixed partner and Lt the total titrant. With Lt = 10 nM:

Pt (nM)	Quadratic	Hyperbola
0.1	0.4988	0.5000
1	0.4875	0.5000
10	0.3820	0.5000
100	0.0901	0.5000

The total titrant needed for half-saturation is Kd + Pt/2, which gives 10.5, 15 and 60 nM for Pt = 1, 10 and 100 nM. A 100 nM fixed partner would make the apparent Kd six times too high.

Kinetics. With kon = 1e6 M⁻¹ s⁻¹, koff = Kd × kon = 0.01 s⁻¹, t1/2 = 0.693/0.01 = 69.3 s, and five half-lives take 5.78 minutes (96.9% complete).

Cheng-Prusoff. For a competitive inhibitor, Ki = IC50/(1 + [S]/Km). An IC50 of 50 nM measured at [S] = Km gives Ki = 50/(1 + 1) = 25 nM. At [S] = 10 Km the same IC50 would mean Ki = 4.5 nM.

How to Measure Kd

Every method is a variation on the same idea: vary one partner, measure the response, fit a binding model. What differs is the readout and the artifacts.

Method	What it measures	Typical affinity range	Main caveat
ITC	Heat from binding; gives affinity, stoichiometry, thermodynamics	nM to sub-mM [8]	Low sensitivity, needs large sample amounts [8]
SPR	Refractive index change near a metal film; real-time on and off rates	sub-nM to low mM [8]	Mass transport, rebinding and immobilization distort results [8]
Fluorescence polarization	Slowed rotation of a labeled partner	Method dependent	Dye can change affinity; run a competition control [2][8]
MST	Directed motion in a temperature gradient	Method dependent	Keep the fluorescent partner at or below expected Kd [8]
BLI	Wavelength shift on a biosensor tip; association and dissociation phases [9]	Method dependent	Range and throughput depend on the instrument; check platform documentation
EMSA	Slower migration of protein-nucleic acid complexes in a native gel	Semiquantitative [8]	Complexes can be lost or gained between equilibration and detection [1]

Two rules apply across all of them. First, vary the incubation time, ideally across at least a 10-fold range, to show that binding reached equilibrium [1]. In a survey of 100 published binding studies, 70 did not report varying incubation time, and most did not document controls for the concentration regime [1]. Second, confirm Kd with a second, independent method, especially when the readout separates bound from free, because complexes can be lost or gained between equilibration and detection [1].

Pull-down and immunoprecipitation assays that wash the pellet disturb the equilibrium before measurement and cannot report a Kd. A valid experiment measures at equilibrium and varies the concentration of one reactant [2]. A binding curve must also reach a plateau to define its amplitude, and Scatchard plots are discouraged because they can hide a failure to reach that plateau [2].

Kd vs Ki vs IC50

These three numbers get mixed up constantly, and the confusion has real consequences for how you compare compounds.

Kd is the general equilibrium dissociation constant for any complex. Ki is the dissociation constant of an enzyme-inhibitor or target-inhibitor complex, and it is an intrinsic thermodynamic constant that does not depend on substrate concentration [6]. IC50 is the total inhibitor concentration that reduces an activity by 50%, and it does depend on the concentrations of enzyme or target, substrate or ligand, and other assay conditions [6].

For classic Michaelis-Menten enzymes, the Cheng-Prusoff relations convert between them:

Inhibition type	Relation
Competitive	Ki = IC50/(1 + [S]/Km)
Uncompetitive	Ki = IC50/(1 + Km/[S])
Noncompetitive	Ki = IC50

For tightly bound inhibitors, subtract the enzyme concentration: competitive Ki = (IC50 - [E]/2)/(1 + [S]/Km) [7]. These equations assume classic Michaelis-Menten kinetics, a known inhibition mechanism and free inhibitor approximately equal to total, and they fail for tight-binding inhibitors [7]. When a ligand competes with an inhibitor, the target-ligand Kd enters the IC50-to-Ki conversion [6].

The practical point: IC50 values from two labs are not comparable unless the assay conditions match. Ki and Kd are more comparable, provided temperature, buffer and salt are reported.

Common Mistakes

Reporting a Kd from a washed pull-down. Washing removes the equilibrium before you measure it, so the number reflects wash kinetics, not affinity. Measure at equilibrium and vary one reactant [2].
Fitting a curve that never plateaus. Without saturation, the amplitude is undefined and Kd is unconstrained. Scatchard plots hide this failure [2].
Ignoring the concentration regime. If the fixed partner sits far above Kd, the curve reports stoichiometry, not affinity. Keep it at least ten-fold below Kd where possible [1][2].
Assuming equilibrium without testing it. Incubating for a fixed time and hoping is not a control. Vary incubation time across at least a 10-fold range [1].
Treating Kd as a single universal number. Puf4 RNA binding differed 100-fold in affinity between 0 °C and 25 °C, so temperature, buffer and salt belong with every reported Kd [1].
Confusing Kd with IC50. IC50 shifts with substrate and enzyme concentration; Ki does not [6].
Dropping the standard state when computing ΔG. Kd must be divided by 1 M to make it dimensionless before taking a logarithm [4].
Flipping the sign of ΔG. Papers differ on whether they report a binding or dissociation free energy. State which direction you mean.

Limitations

The hyperbola and Kd = koff/kon hold only for one-step, single-site 1:1 binding. Cooperative, multi-site or conformationally gated binding needs other models, and those should be considered only after controls for equilibration and titration have been performed [1].

The sign of ΔG for binding differs between papers, and the value depends on the 1 M standard state. For the binding direction A + B → AB, ΔG°bind = -RT ln(Ka × c°) = +RT ln(Kd/c°), so a Kd below 1 M gives a negative, favorable value. Papers that report a free energy of dissociation use the opposite sign, ΔG°diss = -RT ln(Kd/c°). Read the definition before comparing numbers across sources.

Affinity ranges quoted for ITC and SPR come from one methods paper written by the developers of MST, so treat them as approximate [8]. No single verified source covers BLI affinity range or throughput, so check the current instrument documentation for your platform. The kequil = kon[P] + koff expression is the standard pseudo-first-order form; it predicts that equilibration is slowest at low titrant concentration [1].

Classical biochemical methods such as EMSA and ELISA are typically limited to semiquantitative interaction analysis [8]. They can rank interactions, but they are weak tools for reporting a precise Kd.

Frequently Asked Questions

What is Kd in simple terms?

Kd is the free concentration of the titrated partner at which half the binding sites are occupied. A small Kd means the complex is stable at low concentrations; a large Kd means you need a lot of partner to fill the sites. It has units of molarity [2].

How do I measure Kd?

Pick a method whose readout matches your system, hold one partner at a fixed concentration below Kd, titrate the other, and fit the resulting curve. Confirm that binding reached equilibrium by varying incubation time, and verify the value with a second method [1]. ITC, SPR, fluorescence polarization, MST, BLI and EMSA are all used, each with its own artifacts [8][9][10].

What is the difference between Kd and Ki?

Ki is the dissociation constant of an enzyme-inhibitor or target-inhibitor complex, and it does not depend on substrate concentration. Kd is the general term for any complex at equilibrium [6]. In practice they are the same kind of quantity applied to different complexes.

Why is my IC50 different from my Kd?

IC50 depends on the enzyme or target concentration, the substrate or ligand concentration and other assay conditions, while Kd and Ki are intrinsic constants [6]. Use the Cheng-Prusoff relations to convert, and check whether the inhibitor is tight-binding before trusting the result [7].

What do koff and kon tell me that Kd does not?

Kd tells you how much complex exists at equilibrium. kon and koff tell you how fast the system gets there and how long the complex survives. Two interactions with the same Kd can have very different koff values, which changes how they behave in a wash step or a time-resolved assay [2].