The dissociation constant, usually written as Kd, is a number that describes how tightly a molecule binds to its partner. A small Kd means a tight grip; a large Kd means a weak one. Calculating it sounds straightforward, but in practice the result you get depends heavily on the method you choose, the way you fit your data, and even the conditions inside the tube. For anyone working in drug discovery, protein biochemistry, or related fields, understanding how Kd values are generated and where they can go wrong is just as important as the number itself.
What Kd Represents
When two molecules interact, say a drug and a protein, they are constantly associating and dissociating. At equilibrium, the rate at which new complexes form equals the rate at which existing complexes fall apart. Kd is the concentration of the binding partner (the ligand) at which half of the available binding sites on the target are occupied. If a drug has a Kd of 10 nanomolar, that means you need 10 nanomolar of that drug floating around in solution before half the protein molecules have a drug molecule stuck to them.
Kd can also be understood as a ratio: the dissociation rate constant (koff, how fast things fall apart) divided by the association rate constant (kon, how fast things come together). A molecule that latches on quickly and lets go slowly will have a very small Kd, reflecting strong binding. One that drifts on slowly and falls off quickly will have a large Kd. Surface plasmon resonance (SPR) instruments can capture both of these rate constants directly by tracking how molecules accumulate on and then leave a sensor surface, then use the ratio to derive Kd.
The Major Experimental Methods
No single technique dominates Kd measurement. Different methods suit different situations, and each has quirks that can push the final number around. Here are the most common approaches researchers use.
Isothermal Titration Calorimetry
Isothermal titration calorimetry (ITC) works by measuring the tiny heat changes that occur each time a drop of one molecule is added to a solution of its partner. Because binding events either release or absorb heat, the instrument can track how much binding is happening at each injection. ITC is the only technique that directly measures the enthalpy of binding, and it provides Kd, stoichiometry, and thermodynamic parameters all in a single experiment without needing labels, tags, or competing molecules.1PubMed. Isothermal titration calorimetry in drug discovery For a well-behaved system, the raw data are fitted to a binding model and the Kd drops out of the fit. An example: ITC measurements of cocaine and its metabolites binding to an anti-cocaine antibody produced clean data that fit a simple one-site binding model, with roughly one ligand molecule per antibody binding site.2PubMed Central. Isothermal titration calorimetry determination of thermodynamics of binding of cocaine and its metabolites to humanized h2E2 anti-cocaine mAb
The catch with ITC is that it requires relatively large amounts of purified protein, and the signal can be hard to interpret for very weak or very tight binders. Still, many researchers treat ITC as a gold standard for affinity measurement because it operates label-free in solution.
Surface Plasmon Resonance
SPR instruments like the widely used Biacore platform immobilize one binding partner on a sensor chip and flow the other partner across the surface. As molecules bind, the refractive index near the surface changes, producing a real-time curve called a sensorgram. The shape of that curve encodes the association and dissociation kinetics. By running experiments at a series of concentrations and watching the signal rise toward saturation, you can extract both kinetic rate constants and a steady-state Kd.3PubMed Central. Use of Surface Plasmon Resonance (SPR) to Determine Binding Affinities and Kinetic Parameters Between Components Important in Fusion Machinery
A well-established best practice in SPR analysis is to fit the dissociation phase of the sensorgram first, because the off-rate is relatively unambiguous when the free ligand has been washed away. That dissociation rate constant is then used to constrain the fit of the association phase, which improves the accuracy of the on-rate estimate.4PubMed Central. Determination of rate and equilibrium binding constants for macromolecular interactions using surface plasmon resonance: use of nonlinear least squares analysis methods The ratio of those two rate constants gives Kd.
Fluorescence-Based Methods
Microscale thermophoresis (MST) and fluorescence polarization (FP) are solution-phase techniques that measure binding by tracking changes in how a fluorescently labeled molecule moves or tumbles. MST monitors the directed motion of molecules in a temperature gradient, while FP detects the slowdown in molecular tumbling when a small labeled molecule binds something large. Both can yield Kd values from a titration series.
In a study comparing the two approaches for peptide-protein interactions, MST determined Kd values ranging from about 7 to 17 micromolar for several peptides binding to the enzyme calcineurin, and FP served as a complementary check for a related peptide.5PubMed. Microscale thermophoresis and fluorescence polarization assays of calcineurin-peptide interactions These methods are popular because they use small sample volumes and work well in high-throughput settings. The requirement for fluorescent labeling is the main limitation, since the label itself can sometimes alter binding behavior.
Thermal Shift Assays
Thermal shift assays (TSA), sometimes called differential scanning fluorimetry, measure how much a protein’s melting temperature increases when a ligand binds and stabilizes it. Historically, researchers have estimated Kd from these thermal shifts using a classical formula, but recent work has shown that this classical approach produces Kd values that deviate substantially from ITC results and shift unpredictably depending on the ligand concentration used. Newer fitting methods have been developed that show much better agreement with ITC across a range of conditions.6PubMed Central. Determination of Protein–Ligand Binding Affinities by Thermal Shift Assay This is worth knowing because TSA is cheap and fast, making it attractive for early-stage screening, but the accuracy of the Kd you extract depends on which math you apply to the melting curves.
Why the Same Interaction Can Give Different Kd Values
A recurring frustration in binding studies is that different methods often return different numbers for what should be the same interaction. This is not always a subtle discrepancy. A comparative study of a small-molecule inhibitor binding to a protein target found Kd values of roughly 7 nanomolar by SPR, about 69 nanomolar by ITC, and around 336 nanomolar by MST.7bioRxiv. Comparative Analysis of the Techniques for the Determination of Binding Affinity between a Small Molecule Inhibitor and a Protein Target That is a roughly 50-fold spread across three common platforms.
Several factors drive these discrepancies. SPR immobilizes one partner on a surface, which can alter its behavior compared to a free-floating protein in solution. ITC measures a thermodynamic quantity in solution but requires high concentrations and can struggle with compounds that have limited solubility. MST requires labeling, which introduces its own perturbation. Temperature, buffer composition, and the mathematical model chosen for curve fitting all add variability. The practical lesson: whenever you see a Kd reported in a paper, the method matters. A 10-nanomolar Kd from SPR and a 10-nanomolar Kd from ITC carry different kinds of confidence, and comparing Kd values across methods without acknowledging these differences can be misleading.
Getting from IC50 to Kd
In drug screening, the number you measure first is usually not Kd but IC50, the concentration of a compound needed to inhibit binding or activity by half. IC50 is convenient to measure but depends on assay conditions, particularly the concentration of the competing ligand. To convert an IC50 into something closer to an intrinsic affinity value, researchers commonly use the Cheng-Prusoff equation, which adjusts for the concentration and affinity of the competing ligand to yield Ki, the inhibition constant.8PubMed Central. Binding Curve Viewer: Visualizing the Equilibrium and Kinetics of Protein–Ligand Binding and Competitive Binding – Section: Comparison of the Apparent Ki and Ki Calculated by Using Different Equations
A related approach flips this around: by measuring IC50 at several different concentrations of a labeled ligand and then plotting those IC50 values against ligand concentration, you get a straight line whose slope and intercept reveal both Ki and Kd.9PubMed. A novel method for determination of the affinity of protein: protein interactions in homogeneous assays This is useful because it extracts the intrinsic binding constant from a competition experiment, which is often easier to set up than a direct binding assay. The key assumption behind the Cheng-Prusoff equation is that the ligand and receptor concentrations are arranged so that depletion effects are negligible. When the inhibitor binds so tightly that a significant fraction of it gets soaked up by the target, the simple equation breaks down and tighter-binding correction formulas are needed.
Data Fitting Matters More Than You Might Think
Once raw binding data are in hand, the Kd value lives or dies by the fitting procedure. Historically, binding data from radioligand experiments were linearized using Scatchard plots, where bound-over-free ligand was graphed against bound ligand to produce a straight line. The slope of that line gave the negative reciprocal of Kd. This approach had serious problems. A Monte Carlo simulation study found that linear regression on Scatchard-transformed data produced less accurate Kd estimates than nonlinear regression on the untransformed binding curve, and the gap grew worse when data were noisy or the concentration range was limited. Under those less-than-ideal conditions, Scatchard analysis sometimes returned physically impossible negative values for Kd, while nonlinear fitting of the same data gave reasonable results.10PubMed. Model testing in radioligand/receptor interaction by Monte Carlo simulation
Modern software uses nonlinear least-squares fitting as the default, which is a major improvement. But the choice of binding model still matters. A simple one-site model assumes one type of binding site; if the system actually has two classes of sites with different affinities, forcing a one-site model will spit out an average Kd that does not accurately describe either site. Residual plots and statistical measures of goodness-of-fit help identify when a more complex model is needed, though adding parameters to a model always risks overfitting.
Computational Prediction of Kd
Measuring Kd experimentally for every possible drug candidate is expensive and slow. Computational chemistry tries to predict binding affinity from molecular structure, using the three-dimensional shapes and energetics of a protein-ligand pair to estimate how tightly they will bind. Among computational approaches, free energy perturbation (FEP) methods are considered particularly accurate, typically achieving agreement with experimental data to within about 1 kilocalorie per mole of binding free energy. In a test across eight targets and eight sets of related compounds, one implementation of absolute FEP produced a moderate correlation with experimental binding energies.11PubMed Central. Understanding the impact of binding free energy and kinetics calculations in modern drug discovery – Section: 2.2 Alchemical free energy perturbation
To put “within 1 kcal/mol” in perspective: because binding free energy and Kd are linked through a logarithmic relationship, an error of 1 kcal/mol in free energy translates to roughly a five-fold error in the predicted Kd. That sounds rough, but for an early-stage drug campaign trying to rank thousands of candidates, being consistently within five-fold of the true answer is actually useful. The challenge is that real-world accuracy varies substantially depending on the target, the chemical series, and the quality of the structural data fed into the calculation. Computational Kd predictions are a complement to experiments, not a replacement.
When a Single Kd Is Not Enough
The standard Kd calculation assumes a simple 1:1 interaction in which one molecule of ligand binds one site on the target. Several biologically important situations violate that assumption and require a different framework.
Avidity and Multivalent Binding
Antibodies, for example, have two identical binding arms. When both arms engage targets on the same surface simultaneously, the observed “functional affinity” can be far tighter than the affinity of either arm measured individually. This effect, called avidity, arises because once one arm binds, the second arm is held in close proximity to its target, dramatically increasing the chance it will bind and rebind even after momentary dissociation.12PubMed Central. Exploring avidity: understanding the potential gains in functional affinity and target residence time of bivalent and heterobivalent ligands The result is a much slower apparent off-rate and a deceptively small measured Kd that does not reflect the intrinsic strength of either individual binding event.
Distinguishing avidity from true affinity matters in antibody engineering and other multivalent systems. A study of the interaction between IgG antibodies and the neonatal Fc receptor (FcRn) used a biosensor technology that could tease apart the single-arm affinity from the two-arm avidity effect within a single measurement, something that classical SPR struggled to do cleanly.13PubMed Central. Insight into the avidity-affinity relationship of the bivalent, pH-dependent interaction between IgG and FcRn If you report a Kd for an antibody without specifying whether it reflects monovalent affinity or bivalent avidity, the number is ambiguous at best.
Cooperativity
Some proteins have multiple binding sites that influence each other. When binding at one site makes binding at the next site easier, the system shows positive cooperativity and the binding curve steepens relative to what a simple 1:1 model would predict. The Hill coefficient, often extracted from dose-response data, quantifies this steepness. A Hill coefficient greater than one signals that multiple binding events are linked, though it cannot by itself tell you the exact mechanism (whether the sites communicate through shape changes, competition, or some other route). A Hill coefficient below one can look like the sites are interfering with each other, but in enzymatic inhibition experiments this can instead mean that some enzyme-inhibitor complexes retain partial activity, not that binding is genuinely anti-cooperative.14PubMed Central. Hill coefficients, dose-response curves and allosteric mechanisms Fitting cooperative systems with a standard Kd equation that assumes independent sites will give misleading results.
Conformational Selection and Induced Fit
The textbook picture of binding has a rigid lock receiving a rigid key. In reality, proteins are flexible, and binding often involves changes in the protein’s shape. Two competing models describe this: conformational selection, where the ligand binds a pre-existing shape among the protein’s natural fluctuations, and induced fit, where the ligand first contacts the protein and then coaxes it into a new shape. Both models produce measurable binding kinetics, but they predict different behaviors when you plot the observed rate of binding against ligand concentration. Induced fit always produces rates that increase with ligand concentration, while conformational selection can produce rates that increase, decrease, or stay flat, depending on the underlying rate constants.15PubMed Central. Conformational selection or induced-fit? A critical appraisal of the kinetic mechanism If you are extracting a Kd from kinetic data and the wrong mechanism is assumed in the fitting model, the resulting affinity estimate can be off.
Kd in Drug Discovery and Why Residence Time Entered the Conversation
For decades, drug discovery programs pursued the tightest possible Kd as the primary goal for lead optimization. A tighter binder seemed like it should be a better drug. But a Kd measured at equilibrium in a test tube does not capture what happens inside the body, where drug concentrations fluctuate constantly as a dose is absorbed, distributed, and cleared. The concept of residence time, defined as how long the drug stays physically bound to its target once it has attached, addresses this gap. A drug with a long residence time can remain effective even as its free concentration in the blood drops below the Kd, because the molecules already bound to the target are slow to let go.16PubMed Central. Drug-target residence time: critical information for lead optimization
Residence time is governed primarily by koff, the dissociation rate constant. Two drugs can share the same Kd while having very different kon and koff values, and this difference can translate into different durations of action in patients. Optimizing solely for Kd might lead a team to improve kon (making the drug find its target faster) without changing koff, which would tighten the measured affinity in a test tube without actually extending the drug’s effect in the body. Recognizing this, many modern drug discovery programs now track koff alongside Kd.
Environmental Conditions That Shift the Number
Kd is not a universal constant for a given pair of molecules. It changes with temperature, pH, ionic strength, and the presence of other molecules in solution. This is especially relevant when trying to relate a laboratory measurement to what happens inside a cell. The interior of a cell is crowded with macromolecules, proteins, nucleic acids, polysaccharides, and other large structures that collectively occupy a significant fraction of the available volume. This molecular crowding can push binding partners closer together and favor the bound state over the unbound state, effectively tightening affinity. However, the magnitude of this stabilization is often more modest than you might expect, on the order of about 1 kcal/mol of additional stabilization for protein subunit association, which translates to only about a five-fold tightening in Kd.17PubMed Central. Effect of macromolecular crowding on protein binding stability: modest stabilization and significant biological consequences
pH matters for any interaction in which ionizable groups on the protein or ligand participate in binding. The antibody-FcRn interaction mentioned earlier is a dramatic example: the entire biological function of FcRn depends on the fact that it grips IgG tightly at low pH inside endosomes and releases it at the higher pH of the bloodstream.18PubMed Central. Insight into the avidity-affinity relationship of the bivalent, pH-dependent interaction between IgG and FcRn Engineers designing antibody variants with longer half-lives exploit this by tuning affinity specifically at endosomal pH while trying to keep binding weak at neutral pH. When you see a Kd quoted without the pH and temperature at which it was measured, treat it with some skepticism, because the same pair of molecules at a different pH could easily show a Kd that differs by an order of magnitude or more.
Practical Checklist for Evaluating a Reported Kd
If you are reading a paper or evaluating data from a collaborator, a few questions can help you judge whether a Kd value is reliable and comparable to others in the literature:
- Method: Was the Kd measured by a direct binding technique (ITC, SPR) or derived indirectly from competition data? Indirect derivations carry additional assumptions.
- Fitting model: Was a 1:1 binding model used, and if so, is there evidence that the system actually behaves as 1:1? Residual plots or chi-squared values can indicate poor fits.
- Conditions: Were temperature, pH, buffer, and ionic strength reported? A Kd without these parameters is hard to reproduce or compare.
- Valency: For antibodies or other multivalent molecules, does the reported Kd reflect monovalent affinity or avidity? This can shift the apparent value by orders of magnitude.
- Cross-method agreement: Was the Kd confirmed by at least one independent method? Given that different platforms can disagree by 50-fold or more for the same interaction, single-method values deserve less confidence.
Kd calculation is not a black box that takes in data and spits out truth. Every step, from the experimental technique to the curve fit to the binding model, introduces assumptions. Knowing what those assumptions are, and when they might fail, is what separates a Kd value you can build on from one that might send your project in the wrong direction.