Noncovalent interactions
Bring two carbon atoms from different parts of a protein toward each other. From far away, their interaction is weak. At a suitable separation they attract slightly. Push them another ångström closer and the energy rises sharply. “Closer” has changed from helpful to disastrous.
Now repeat with an N–H group approaching a carbonyl oxygen. A favorable contact can occur at a shorter heavy-atom distance, provided the hydrogen points toward the oxygen. The atoms' identities and the approach direction have changed the answer. A folded protein contains many such local decisions, all constrained by the fact that the participating atoms belong to connected chains.
A contact needs a reference state
A noncovalent interaction leaves the ordinary covalent connectivity unchanged. It can attract or repel, and it can change continuously as groups move. Hydrogen bonds, electrostatic interactions between charged groups, and van der Waals contacts fall into this description.
At room temperature, the thermal scale is about 0.6 kcal/mol. Individual neutral nonbonded contacts are often comparable to this scale or smaller; charged interactions can be considerably larger, depending on their environment. There is no universal energy per “hydrogen bond” or “salt bridge” that can be multiplied by a count to obtain folding stability.
The reason is easiest to see with water. Before two protein groups meet, each may already interact favorably with water. Bringing them together changes those interactions as well as adding the direct protein–protein contact. We must compare the complete before and after states. The energy of a pair in isolation is only one term in that comparison.
Give a hydrogen a direction
A common protein hydrogen bond has the form D–H···A. The donor atom D, usually N or O, is covalently bonded to the hydrogen. The acceptor A has suitable electron density, often a lone pair on N or O, toward which the hydrogen points. The H is partially positive because its donor attracts shared bonding electrons.
The notation draws two different connections. D–H is the covalent bond. H···A is the additional noncovalent interaction. The hydrogen generally remains attached to D in the ordinary hydrogen bond we are considering.
Geometry gives a first test. Donor and acceptor heavy atoms are often roughly 2.7–3.2 Å apart. A 3.5 Å cutoff is a generous candidate search. But the hydrogen must point toward the acceptor; the D–H···A angle is generally more favorable as it approaches 180°. Two suitable heavy atoms can be close while the donor's hydrogen points elsewhere.
The protein backbone repeatedly supplies an N–H donor and a carbonyl oxygen acceptor. An internal proline lacks the usual backbone N–H because its nitrogen is already bonded into its side-chain ring. The amide nitrogen is not an ordinary acceptor: its electronic structure differs from that of a free amine. Identifying donors and acceptors requires the chemical group and protonation state.
| Common group | Donor? | Acceptor? |
|---|---|---|
| Backbone carbonyl oxygen | no | yes |
| Backbone amide N–H | yes | no |
| Serine/threonine hydroxyl | yes | yes |
| Asparagine/glutamine amide | at N–H | at O |
| Deprotonated Asp/Glu carboxylate | no | at O |
| Protonated lysine amino group | yes | no |
| Arginine guanidinium group | at N–H | normally no |
| Tryptophan ring N–H | yes | no |
| Neutral histidine ring | at the N bearing H | at the other N |
Protonating histidine changes this last assignment. Its two neutral forms also differ in which nitrogen bears the hydrogen. Sulfur can participate in some hydrogen bonds too; the N/O rules cover our main examples without exhausting the chemistry.
Why a buried polar group needs attention
Expose a carbonyl oxygen to water and nearby waters can orient their hydrogens toward it. Bury it inside a protein and many of those contacts disappear. If a backbone N–H now points at it, the group has a new interaction that can compensate for part of what was lost.
Helices and sheets repeat this arrangement along the backbone. A buried side chain can also provide a partner, and a water trapped in a cavity may satisfy groups that would otherwise lack one. Calling a polar group “satisfied” describes its interactions; it does not mean that hydrogen bonding removes its partial charges.
An unsatisfied buried donor or acceptor can be costly relative to its water-exposed state. That cost helps explain why introducing or removing one OH group sometimes changes a protein's stability substantially. The same edit on an exposed surface may have a smaller effect because water can rearrange around both versions. Packing changes and protein rearrangements still matter in either case.
Building a helix before seeing a fold
In 1951, Linus Pauling, Robert Corey, and Herman Branson approached protein structure by constructing chains that obeyed known chemistry. They used measured bond lengths and angles, kept the peptide groups planar, and looked for repeating arrangements that allowed favorable hydrogen bonds. Their proposals included the α-helix.Their constraints and proposed geometries appear in the original paper.
Imagine trying the construction yourself. Twist the chain so that one N–H reaches an oxygen. The next peptide group comes with it because the atoms are connected. A usable repeating shape must let the next donor reach a suitable acceptor too, without stretching bonds or driving atoms through one another. Satisfying one contact is easy; repeating the arrangement along a chemically realistic chain is a much narrower problem.
That reasoning is still useful when debugging a predicted structure. If a helix looks plausible as a ribbon, inspect the repeated N–H···O geometry. The ribbon can conceal the failure of the very contacts that make the local shape favorable.
Opposite charges in different surroundings
A lysine amino group and an aspartate carboxylate can make an ion pair, often called a salt bridge. A common structural search finds a positive nitrogen within about 4 Å of a negative oxygen. This is a geometric criterion for a candidate interaction; the protonation assignments must also be credible. A salt bridge can include hydrogen bonds, so these categories are not mutually exclusive.
For two point charges, a simple continuum electrostatic model gives
with charges in units of , distance in Å, and relative dielectric constant . Opposite charges make the product negative, giving attraction. Like charges give repulsion.
In vacuum . In liquid water near room temperature it is about 80. Water molecules reorient and polarize in response to a charge, reducing the interaction between charges compared with vacuum. Replacing all that molecular behavior with one dielectric constant is an approximation, especially at distances of only a few ångströms.
For +1 and −1 charges 4 Å apart, the formula gives −83 kcal/mol in vacuum and about −1.04 kcal/mol using water's dielectric. Using an illustrative protein-interior value of 4 gives −20.75 kcal/mol. That last number looks like a strong argument for burying opposite charges together. What did the calculation leave out?
It left out the water that stabilized each charge before burial. Solvation is the interaction with the surrounding solvent; desolvation is the loss or replacement of those interactions. A buried ion pair can gain a stronger direct attraction while paying a large desolvation penalty for both groups. The net effect depends on geometry, access to water, neighboring groups, and protonation. A protein's interior also does not have one universally correct dielectric constant.
This is why we usually find charged side chains at the surface, and why a buried charged group makes us look for stabilizing partners or internal waters. It is a preference, not a rule that every buried charge must pair one-to-one with an opposite charge.
Attraction has a short-range wall
Even electrically neutral atoms attract through correlated fluctuations of their electron clouds. These dispersion interactions occur between polar and nonpolar groups alike. Individually they are weak, but well-packed surfaces can make many favorable contacts.
At shorter separations, overlap of occupied electron clouds produces strong repulsion. Van der Waals radii summarize typical nonbonded sizes: roughly 1.7 Å for carbon, 1.55 Å for nitrogen, 1.52 Å for oxygen, and 1.8 Å for sulfur. They are useful comparison scales, not hard boundaries painted on atoms.
For ordinary nonbonded carbon pairs, a separation around the sum of the radii is already quite close. Two carbons 2.6 Å apart overlap that sum by about 0.8 Å and should raise a steric-clash warning. Covalently bonded atoms are a different case; a C–C bond near 1.53 Å is entirely normal.
A common model combines a shallow attraction and a steep repulsion in the Lennard-Jones potential:
Here is the well depth, unrelated to the dielectric constant above. At the energy is zero. The minimum lies at and has energy . At larger separations the attraction decays as ; at shorter ones the repulsive term rapidly dominates. The power twelve is a convenient approximation to the repulsion, not a fundamental law of electron clouds.
Compare the nonpolar and charged pairs. First find the minimum, then move a little closer and compare the energy with the shaded thermal scale. For the charged pair, changing the dielectric from 80 to 4 makes the Coulomb contribution twenty times larger at the same distance. The plot omits the cost of moving the charges out of water.
A particular carbon-pair potential may have a well only 0.1–0.2 kcal/mol deep. The location of that minimum depends on the chosen parameters and is not identical to every tabulated radius sum. A physically useful model has to describe favorable packing as well as penalize overlap.
See the contact distances in ubiquitin
Let us measure every pair of heavy atoms that is more than three covalent bonds apart. The exclusion removes short distances created by local covalent geometry. Then compare carbon–carbon pairs with pairs involving two nitrogens or oxygens.
import numpy as np
import matplotlib.pyplot as plt
from primer import structure
atoms = [a for a in structure("1UBQ")[0].get_atoms() if a.get_parent().id[0] == " "]
xyz = np.array([a.coord for a in atoms])
element = np.array([a.element for a in atoms])
dist = np.linalg.norm(xyz[:, None] - xyz[None], axis=-1)
# Bond graph from distances (lesson 1.1), then drop pairs within three bonds of each other.
radius = np.array([{"C": 0.76, "N": 0.71, "O": 0.66, "S": 1.05}[e] for e in element])
bond = ((dist < radius[:, None] + radius[None] + 0.4) & (dist > 0)).astype(int)
near = bond.copy()
for _ in range(2):
near = ((near + near @ bond) > 0).astype(int)
np.fill_diagonal(near, 1)
i, j = np.nonzero(np.triu(near == 0))
d = dist[i, j]
polar = np.isin(element[i], ["N", "O"]) & np.isin(element[j], ["N", "O"])
carbon = (element[i] == "C") & (element[j] == "C")
fig, ax = plt.subplots(figsize=(6.4, 3.2))
bins = np.arange(2.4, 5.0, 0.05)
ax.hist(d[carbon], bins=bins, histtype="step", lw=2, color="#eda100", label="C···C")
ax.hist(d[polar], bins=bins, histtype="step", lw=2, color="#2a78d6", label="N/O···N/O")
ax.set_xlabel("distance between nonbonded atoms (Å)")
ax.set_ylabel("pairs")
ax.legend()
plt.show()
print(f"closest C···C: {d[carbon].min():.2f} Å; N/O pairs closer than 3.2 Å: {(d[polar] < 3.2).sum()}")
closest C···C: 3.41 Å; N/O pairs closer than 3.2 Å: 78
Carbon pairs begin appearing near the expected packing scale. N/O pairs can occur at shorter distances, consistent with their smaller sizes and, for chemically suitable pairs, favorable hydrogen bonds or ionic contacts. The histogram cannot assign an interaction to an individual pair: it has no donor direction or protonation information. The lab will add some of that information explicitly.
Follow the backbone N···O contacts in the first view. Some join distant stretches of the chain; others repeat within a helix. Then inspect three close positive–negative side-chain pairs in the second view. The dashed measurement lines mark distances, not additional covalent bonds.
Open 1UBQ in Mol*What the energy function must keep track of
Suppose we move one side chain to improve a salt bridge. That same movement may create a steric clash, remove a hydrogen bond, or expose nonpolar surface to water. The terms cannot be optimized independently because they depend on shared coordinates.
Their nonlinear distance dependence makes this harder. Halving a distance multiplies by 4,096. A small placement error on the repulsive wall can overwhelm many favorable contacts elsewhere. Representing each residue by one sphere avoids much of the atomic work, but loses hydrogen directions, side-chain shapes, and the locations of particular charges. A coarse model needs interaction rules appropriate to the information it actually keeps.
Check yourself
An asparagine nitrogen and a backbone oxygen are 2.9 Å apart. Is that sufficient to identify a hydrogen bond? What additional geometric observation would help?
Reasoning
They are a plausible donor–acceptor pair at a favorable distance. We still need the donor hydrogen to point toward the oxygen. Placing or observing the hydrogen lets us check H···O distance and the N–H···O angle. Proximity alone does not establish the directional interaction.
Two opposite unit charges remain 4 Å apart while a model changes their dielectric from 80 to 4. How much does the direct attraction change, and why is this not the folding free-energy change?
Reasoning
It becomes twenty times stronger, from about −1.04 to −20.75 kcal/mol in the point-charge model. A folding comparison must also include lost interactions with water, rearrangements, changes in protonation, and the other parts of the protein. The pair term omits those contributions.
Removing a buried serine OH leaves a backbone carbonyl without its only hydrogen-bond partner. Give a reason to expect destabilization and two ways the protein could partly compensate.
Reasoning
The carbonyl has lost a favorable contact while remaining inaccessible to bulk water. A local rearrangement could supply another donor, or a water molecule could enter the newly created space. Without compensation, destabilization is plausible; its magnitude cannot be obtained by counting one lost hydrogen bond alone.
A coarse model puts one point at each Cα. It predicts the correct contact map but gives no atom-level coordinates. Which checks can you already perform, and which must wait?
Reasoning
We can inspect overall compactness, residue neighborhoods, and contacts between distant sequence positions. We cannot inspect side-chain overlaps, individual salt-bridge geometries, or the directions of backbone hydrogen bonds. Those require additional coordinates and chemical assignments.