Binding and Kd
Docking, virtual screening, and drug design all chase one experimental number: how tightly a molecule binds its target, the dissociation constant . It spans eleven orders of magnitude, from millimolar fragments to femtomolar biotin–streptavidin, and on the free-energy scale of the last two lessons every factor of ten costs the same 1.36 kcal/mol. Hemoglobin then adds a twist: binding at one site can change binding at the others.
Kd and the binding free energy
The molecule that binds (a drug, an oxygen molecule, a hormone, an ion) is the ligand, and the protein it binds is the receptor or target. Concentrations are in molar units: 1 M is one mole of molecules per liter. Binding happens at far lower concentrations, written with the usual prefixes: mM ( M), µM (), nM (), pM (), fM ().
The strength of the interaction between the ligand and the receptor can be measured experimentally and is often reported as the dissociation constant, , or by the concentration of ligand that inhibits activity by 50%, the IC₅₀. The binding free energy is the thermodynamic quantity that is of interest in computational structure-based design and is defined by
The relationship between the binding free energy and the experimentally determined is
where is the association constant.
is defined by the binding equilibrium as , where square brackets mean concentration. (This is the equilibrium constant you met in general chemistry, written for the complex falling apart.) A logarithm needs a pure number, so the formula above really uses divided by the 1 M standard concentration; that is the "standard state", and the ° in marks it. has units of concentration and a direct meaning: is the free ligand concentration at which half the protein is bound. Rearranged, the fraction of protein with ligand bound (the occupancy ) is
Plotted against this is the logistic , the same curve as the titration curves of lesson 1.3. That's no coincidence, since protonation is binding with H⁺ as the ligand and a pKa is of a dissociation constant. At 298 K each factor of ten in is worth 1.36 kcal/mol:
| (kcal/mol) | Typical of | |
|---|---|---|
| 1 mM | −4.1 | weak fragments (small drug-like pieces), many metabolites (the cell's small molecules) |
| 1 µM | −8.2 | a screening hit (a compound picked out of a large library test), many protein–protein contacts |
| 1 nM | −12.3 | an optimized drug |
| 1 pM | −16.4 | the best antibodies (the immune system's binding proteins) |
| about 10 fM | about −19 | biotin–streptavidin, one of the tightest known |
Fraction bound against free ligand concentration on a log axis. Dragging slides the curve sideways without changing its shape, and the readout converts it to . Raising the Hill coefficient steepens the curve, which the next section explains.
import numpy as np
R, T = 1.987e-3, 298.15
for kd in (1e-3, 1e-6, 1e-9, 1e-12):
print(f"Kd = {kd:7.0e} M dG = {R * T * np.log(kd):6.1f} kcal/mol") # 1 M standard state
# How much does a scoring-function error matter?
for err in (1.0, 2.0, 3.0):
print(f"an error of {err} kcal/mol = a factor of {np.exp(err / (R * T)):4.0f} in Kd")
Kd = 1e-03 M dG = -4.1 kcal/mol Kd = 1e-06 M dG = -8.2 kcal/mol Kd = 1e-09 M dG = -12.3 kcal/mol Kd = 1e-12 M dG = -16.4 kcal/mol an error of 1.0 kcal/mol = a factor of 5 in Kd an error of 2.0 kcal/mol = a factor of 29 in Kd an error of 3.0 kcal/mol = a factor of 158 in Kd
Remember the last lines when you read docking benchmarks. Errors around 2 kcal/mol against measured affinities are common, and 2 kcal/mol is a factor of about 30 in . That's enough to separate a millimolar fragment from a nanomolar drug, but not enough to rank two analogues reliably.
Cooperativity: hemoglobin and myoglobin
For oxygen in blood, the "concentration" is a partial pressure, : the share of the total gas pressure due to oxygen, measured in mm Hg (1 mm Hg = 1 torr). It plays exactly the role plays above.
Hemoglobin is a protein consisting of four polypeptide subunits, each of which surrounds a heme group, an iron-containing ring structure that can reversibly bind a molecule of O₂. Thus each hemoglobin molecule can bind and release up to four O₂ molecules.
The relationship between and the amount of O₂ bound to hemoglobin is not linear but S-shaped (sigmoidal). The hemoglobin–oxygen binding curve reflects interactions between the four subunits of the hemoglobin molecule. At low values, only one subunit will bind an O₂ molecule. When it does so, the shape of that subunit changes, altering the quaternary structure of the entire hemoglobin molecule. That structural change makes it easier for the other subunits to bind an O₂ molecule; that is, their O₂ affinity is increased. Therefore a smaller increase in is necessary to get the hemoglobin molecule to bind a second O₂ molecule than was necessary to get it to bind one. This change in affinity is reflected in the increased steepness of the O₂-binding curve. The influence of O₂ binding to one hemoglobin subunit on the O₂ affinity of the other subunits is called positive cooperativity.
In the lungs, where the is about 100 mm Hg, hemoglobin is nearly 100% saturated. The in blood returning to the heart from the body (at rest) is usually about 40 mm Hg, and at this the hemoglobin is still about 75% saturated. When a tissue becomes starved of O₂ and its local falls below 40 mm Hg, the hemoglobin flowing through that tissue is on the steep portion of its binding curve. That means relatively small decreases in below 40 mm Hg will result in the release of lots of O₂ to the tissue.
Muscle cells have their own O₂-binding molecule, myoglobin. Myoglobin consists of just one polypeptide chain associated with an iron-containing ring structure that can bind one O₂ molecule. Myoglobin has a higher affinity for O₂ than hemoglobin does, so it picks up and holds O₂ at values at which hemoglobin is releasing its bound O₂.
The half-saturation pressure plays the role of . Myoglobin follows the simple curve with torr. Hemoglobin is well described near half saturation by the Hill equation, , with torr and Hill coefficient .
Oxygen saturation of hemoglobin and myoglobin against oxygen pressure. Drag the tissue pressure from 40 torr (rest) toward 20 torr (working muscle) and compare how much oxygen each protein would hand over after leaving the lungs at 100 torr.
is not the number of binding sites. Hemoglobin has four, and ; the number of sites is only an upper limit. means positive cooperativity, means independent sites, and means negative cooperativity or a mixture of sites. The change in quaternary structure is a switch between two quaternary states of the tetramer: a low-affinity T state, which deoxyhemoglobin (4HHB) adopts, and a high-affinity R state (lesson 3.6).
Deoxyhemoglobin (4HHB): two α and two β chains, each holding a heme. Each heme's iron is bound to a "proximal" histidine (α His 87, β His 92). When O₂ binds, the iron moves into the plane of the heme and pulls the histidine and its helix with it. That movement is how binding at one site reaches the interfaces between chains.
Open 4HHB in Mol*What a docking score is trying to estimate
The passage below mentions a molecular mechanics force field: an energy function of the atomic coordinates built from bond, angle, torsion, van der Waals, and Coulomb terms. Lesson 4.4 writes it out.
We are interested in the binding free energy of the ligand with the protein, , where , enthalpy, is a result of the changes in van der Waals and Coulombic interactions as water is removed from the binding site and the ligand surface and replaced with van der Waals and Coulombic interactions between the receptor and the ligand. Changes in internal energy of the receptor and the ligand have to be considered also. Entropic changes upon binding are a result of restricting the internal degrees of freedom of the receptor, especially in the binding site, and the ligand, as well as changes in translational and rotational degrees of freedom. The loss of flexibility and other entropic changes are generally unfavorable to complex formation. Another large contributor to the entropy is the solvent, which generally gains entropy as it is released from the protein binding site and the ligand upon complex formation. This so-called "hydrophobic effect" is often the driving force for protein-ligand interactions. The binding free energy is thus a complex term, which can be calculated directly, but at a considerable computational cost.
Practical approaches to calculating binding free energies can be divided into four major approaches: first-principles scoring functions, semiempirical methods, empirical scoring functions, and knowledge-based potentials. First-principles methods generally use a molecular mechanics force field. Since the molecular mechanics force field only accounts for enthalpic forces between the protein and the ligand, the estimated binding free energy is only an approximation to the true binding free energy. An empirical scoring function consists of a number of terms known to play a role in protein-ligand complex formation, such as hydrogen bonding and hydrophobic contacts between the protein and the ligand. The weights for each of the terms are derived by fitting the equation using regression methods to a training set of protein-ligand complexes for which both structural information and experimentally determined interaction energies are known. Knowledge-based functions are derived using protein-ligand complexes alone, based on the statistical assumption that interactions that are more frequently observed are more important for complex stability.
The knowledge-based idea is lesson 4.2 run backward. If contacts were Boltzmann-distributed across known structures, an observed frequency would imply an energy up to a constant. As for the entropy term, one careful calculation, for amprenavir, an anti-HIV drug that blocks HIV protease (an enzyme the virus needs to cut its own proteins into working pieces), shows how large it can be:
Conformational entropy reflects the number of conformations that a molecule can assume. In terms of an energy landscape, conformational entropy reflects the number of wells or minima. It is lost on binding when proximity to a bound molecule causes interactions (most often, steric clashes) that prevent previously available conformations from being assumed. Vibrational entropy reflects the motions available to a molecule as a whole (rotation and translation) or to the individual atoms that compose it (torsion, angle bending, and bond stretching). In terms of an energy landscape, vibrational entropy reflects the width of wells or minima.
Gilson and coworkers calculated the entropies for amprenavir, an inhibitory drug molecule that binds to HIV protease. They determined the overall configurational entropy loss on binding to be 26.4 kcal/mol. Of this, at most 4.1 kcal/mol can be accounted for by conformational entropy loss, less than a sixth of the total entropy lost. Therefore, vibrational entropy appears to have an unexpectedly dominant role in binding.
The 26.4 kcal/mol is a penalty larger than the whole measured binding free energy of most drugs. Binding still happens because enthalpy and solvent entropy pay it back. A score computed from a single docked pose contains none of these terms.
Check yourself
A drug binds its target with nM. In the patient's blood its free concentration is 100 nM. What fraction of target molecules carry the drug? What if a resistance mutation weakens binding by 2.7 kcal/mol?
Reasoning
, so 91% of targets are occupied. 2.7 kcal/mol is two factors of ten (2 × 1.36), so becomes about 1 µM and . Occupancy falls from 91% to 9%. A mutation that removes one or two good contacts can turn an effective drug into an ineffective one, which is how many kinase-inhibitor resistance mutations work.
Two docked ligands score −9.5 and −10.0 kcal/mol. A chemist plans to synthesize only the second. Is the difference meaningful?
Reasoning
Probably not. 0.5 kcal/mol is a factor of about 2.3 in , while typical scoring errors are around 2 kcal/mol, a factor of about 30. Within the method's error the two scores are indistinguishable. An experiment, or a more expensive relative free-energy calculation between the two analogues, has to decide.
A mutation at the interface between hemoglobin's α and β chains locks the tetramer in the R state. Predict the mutant's oxygen-binding curve, and whether it would be a better or worse oxygen carrier.
Reasoning
Locked in R, every site has high affinity, as in myoglobin, and the switch between affinities is gone. The curve becomes hyperbolic () with a low . The protein loads oxygen in the lungs but holds on to it at 40 torr and even at 20, delivering much less, so it is a worse carrier. High-affinity hemoglobin variants like this exist in people, and the body compensates by making more red blood cells.
Resting tissue takes only about one in four of hemoglobin's oxygen molecules, which looks wasteful. Using the Hill curve, explain why a carrier with and the same of 26 torr would actually be worse at supplying working muscle.
Reasoning
With , : about 0.79 at 100 torr, 0.61 at 40 torr, and 0.43 at 20 torr. It would leave the lungs only 79% loaded and deliver 36% of its capacity when tissue drops to 20 torr. The real hemoglobin () leaves the lungs about 98% full and drops to about 32% at 20 torr, delivering about 66%. Cooperativity gives both a fuller load in the lungs and a steep release between 40 and 20 torr, where working tissue sits; the oxygen left unused at rest is the reserve that working muscle draws on.