Atoms, electrons, and covalent bonds
Hold a protein at room temperature and its atoms keep moving. Bonds vibrate, side chains turn, and the whole molecule tumbles in solution. Yet a leucine stays a leucine and the chain stays connected. Folding can rearrange the molecule extensively without taking it apart.
To understand how both statements can be true, we need the relative costs of different motions. Stretching a bond, rotating it, and breaking it are different physical operations. Chemistry makes some of them much easier than others, and that difference gives a protein its available degrees of freedom.
What identifies an atom
An atom has a small nucleus containing positively charged protons and uncharged neutrons, surrounded by negatively charged electrons. An element is identified by its proton count, the atomic number. Carbon always has six protons. Nitrogen has seven and oxygen eight. Most of the atoms in a protein are carbon, hydrogen, nitrogen, or oxygen, with sulfur in cysteine and methionine.
Change the neutron count and you change the isotope. Carbon-12 and carbon-13 are both carbon, with six and seven neutrons respectively. Their chemistry is very similar, but some physical properties differ. This lets structural biologists label proteins with carbon-13 and nitrogen-15 for NMR experiments without replacing their chemical building blocks.
Change the electron count and you change charge. Six protons and six electrons give a neutral carbon atom. Sodium, with eleven protons, becomes Na⁺ when it has ten electrons. Chlorine, with seventeen protons, becomes Cl⁻ when it has eighteen. A charged atom or group of atoms is an ion; positive ions are cations and negative ions are anions.
A molecule can therefore contain charged groups, and a whole molecule can itself be an ion. The ammonium group NH₄⁺ contains several covalently connected atoms with one net positive charge. Later we will distinguish that net charge from an uneven distribution of electrons inside an overall neutral molecule.
Shared electrons hold the graph together
Electrons occupy quantum states called orbitals. An orbital describes a spatial probability distribution; the familiar cloud drawing is usually a contour enclosing much of that probability. It is not a little track along which an electron circles the nucleus. Each orbital can hold at most two electrons, and the outer, or valence, electrons are the ones most involved in ordinary bonding.
When atoms form a covalent bond, electron density is shared between them in a lower-energy bonded arrangement. A simple single-bond drawing represents a shared electron pair. A double bond represents two pairs and a triple bond three. More detailed quantum descriptions explain how those electrons are distributed, but the pair-counting picture is enough to read most protein chemistry.
Carbon has four valence electrons. In methane, CH₄, it shares a pair with each of four hydrogens. Counting the shared electrons around carbon gives eight; each hydrogen shares two. This illustrates the octet rule, a useful pattern for common carbon, nitrogen, and oxygen compounds. Hydrogen uses two, and other elements and charged species require their own accounting.
The bond graph records these connections. It needs bond orders as well as adjacency: a carbon with a double bond and two single bonds has three neighbors but four units of bond order. Confusing graph degree with this bond count will give the wrong number of missing hydrogens.
Put a bond on the thermal scale
Chemists often quote energy per mole. A mole contains about particles, so an energy of 80 kcal/mol means 80 kilocalories for that many copies of the event. One kilocalorie is 4.184 kJ.
Thermal energy provides a useful comparison. For one molecule the scale is , with Boltzmann's constant and in kelvin. Per mole it is . At 298 K,
Breaking a typical C–C single bond into separated radicals costs roughly 80–90 kcal/mol. That is well over a hundred times the room-temperature thermal scale. Ordinary thermal jostling therefore does not repeatedly break the protein's backbone. Chemical reactions can change covalent bonds through other pathways; their rates depend on activation barriers and may be altered by catalysts.
By comparison, many rotations about single bonds have barriers of only a few kcal/mol. These motions can occur repeatedly while the covalent graph remains unchanged. A folded protein can be both chemically intact and mechanically restless.
Stiff distances, softer rotations
The four bonds around an ordinary saturated carbon point approximately toward the corners of a tetrahedron, at angles near 109.5°. A carbonyl carbon has three neighbors arranged approximately in a plane, with angles near 120°. These geometries emerge from the electronic structure of the bonded atoms.
Typical bond lengths occupy narrow ranges:
| Bond or group | Approximate length, Å |
|---|---|
| C–C single | 1.53 |
| C=C double | 1.34 |
| Ordinary C–N single | 1.47 |
| Peptide C–N | 1.33 |
| Carbonyl C=O | 1.23 |
| C–O single | 1.43 |
| N–H | 1.01 |
| S–S disulfide | 2.04 |
These are typical values, not exact constants for every molecule. But a C–C single bond of 1.9 Å should immediately look suspicious. Near its preferred length, a bond's energy is often modeled as a spring:
Take an illustrative kcal mol⁻¹ Å⁻². A stretch of 0.1 Å costs 1.5 kcal/mol, already about 2.5 . A stretch of 0.37 Å costs about 20.5 kcal/mol. This explains why a small-looking coordinate error can dominate a calculated energy.
Now fix the length of a single bond and the angles at its two ends, and rotate one end relative to the other. The surrounding atoms move without stretching that bond. Four consecutive atoms define the rotation through a torsion, or dihedral, angle: the angle between the plane of the first three atoms and the plane of the last three.
Turn butane about its middle C–C bond. Its end groups pass through favorable and unfavorable relative positions while the central bond keeps its length. The low-energy arrangements are anti and the two gauche conformations. Switch to the double-bond model to see a much larger rotational barrier.
Rotation is not equally easy about every single bond. Neighboring atoms can collide, rings restrict motion, and electron delocalization can give a nominal single bond some double-bond character. The chain's flexibility comes from the motions still affordable after these restrictions.
The same atoms, arranged differently
Ethanol and dimethyl ether have the same molecular formula and different bond graphs. They are structural isomers. Butane's anti and gauche arrangements have the same graph and differ by a rotation that is readily accessible; they are conformations of the same molecule.
A C=C double bond gives a different situation. Part of its bonding depends on aligned electron orbitals. Twisting one end by 90° disrupts that alignment and carries a large energy cost. Suitable substituents can therefore have distinct cis and trans arrangements that do not readily interconvert at room temperature.
In some chemical groups, assigning every bond as purely single or double also fails. Draw a carboxylate, –COO⁻. We can put the C=O on either oxygen, but the real group has delocalized electrons and two similar C–O bonds, both about 1.25 Å. These alternative drawings are resonance forms. The molecule is not rapidly switching between the two drawings; neither drawing alone describes it.
The peptide group also has delocalized electrons. Its C–N bond is shorter than an ordinary C–N single bond and resists rotation, keeping the group approximately planar. Cis–trans isomerization is possible but much slower than many ordinary single-bond rotations, and can limit folding rates. Much of the backbone's remaining flexibility lies in two torsions per residue, φ and ψ, which we will study in chapter 3.
Recover a graph from a structure
The regularity of bond lengths lets us try a small inference problem. Given ubiquitin's heavy atoms, connect two atoms when their separation is less than the sum of their covalent radii plus a tolerance. A covalent radius is an atom's approximate contribution to a bond length.
import numpy as np
from collections import Counter
from primer import structure
ubq = structure("1UBQ")
atoms = [a for a in ubq[0].get_atoms() if a.get_parent().id[0] == " "] # protein atoms only, no water
print(len(atoms), "atoms:", dict(Counter(a.element for a in atoms)))
xyz = np.array([a.coord for a in atoms])
covalent_radius = {"C": 0.76, "N": 0.71, "O": 0.66, "S": 1.05} # Å
r = np.array([covalent_radius[a.element] for a in atoms])
dist = np.linalg.norm(xyz[:, None, :] - xyz[None, :, :], axis=-1)
bonded = (dist < r[:, None] + r[None, :] + 0.4) & (dist > 0)
i, j = np.nonzero(np.triu(bonded))
print(len(i), "bonds inferred")
602 atoms: {'N': 105, 'C': 378, 'O': 118, 'S': 1}
608 bonds inferredThis rule produces a candidate bond graph from positions. It works well for this simple structure, but it does not infer bond orders, identify protonation states, or handle every metal site and unusual ligand. Production tools also use chemical templates and explicit connectivity information where available. A legacy protein PDB file often relies on that known residue chemistry rather than listing every standard bond.
Now count the neighbors of each atom in the inferred graph:
degree = bonded.sum(axis=1)
for element in "CNOS":
counts = Counter(int(d) for a, d in zip(atoms, degree) if a.element == element)
print(element, dict(sorted(counts.items())))
C {1: 50, 2: 127, 3: 201}
N {1: 24, 2: 78, 3: 3}
O {1: 118}
S {2: 1}The file omits hydrogens, so these are heavy-atom degrees. A methyl carbon has one heavy neighbor; its other three bonds lead to hydrogens. A carbonyl carbon has three heavy neighbors and a double bond to one of them. Recovering the complete chemistry requires more than insisting that every carbon have four neighbors.
The distances contain another clue. Group the inferred bonds by their elements:
kind = np.array(["".join(sorted(atoms[a].element + atoms[b].element)) for a, b in zip(i, j)])
lengths = dist[i, j]
import matplotlib.pyplot as plt
fig, ax = plt.subplots(figsize=(6.4, 3.2))
for k, c in [("CC", "#2a78d6"), ("CN", "#eb6834"), ("CO", "#1baf7a")]:
ax.hist(lengths[kind == k], bins=np.arange(1.15, 1.62, 0.01), histtype="step", lw=2, color=c, label=k[0] + "–" + k[1])
ax.set_xlabel("bond length (Å)")
ax.set_ylabel("bonds")
ax.legend()
plt.show()
The C–N lengths separate into populations near 1.33 and 1.47 Å. The shorter population includes peptide bonds. The C–O distribution contains short carbonyl bonds and longer single bonds. We can see the consequences of electron sharing in an array of measured coordinates.
Remember that crystal models are refined using expected chemical geometry as well as experimental data. This agreement is a useful check of the model, but the histogram is not an independent discovery of bond lengths from entirely unconstrained observations.
Check yourself
A carbonyl carbon has three heavy-atom neighbors and no hydrogens. Is that incompatible with carbon's usual four-bond count?
Reasoning
No. One neighbor is connected by a double bond and the other two by single bonds. The graph degree is three, while the total bond order is four. An adjacency matrix without bond orders loses this distinction.
Using kcal mol⁻¹ Å⁻² and Å, estimate the stretching penalty for a bond predicted at 1.9 Å. Compare it with kcal/mol.
Reasoning
The stretch is 0.37 Å, giving kcal/mol, or about 34 . The spring model is approximate this far from equilibrium, but the scale already identifies a severe geometric error. It is much larger than an ordinary thermal fluctuation.
An atom has eight protons and ten electrons. A second has eight protons, eight electrons, and one additional neutron. Which identity or property changes in each case?
Reasoning
Both are oxygen because they have eight protons. The first has charge −2. The second is neutral oxygen with a different isotope mass. Changing electron or neutron counts does not change the element.
A simulation unfolds a protein while leaving all its covalent bonds intact. What can change, and which graph from lesson 0.3 will change most?
Reasoning
Many torsion angles can change, with smaller bond-length and bond-angle fluctuations. Distant sections of the chain separate and meet different neighbors, so the contact graph changes extensively. The covalent bond graph remains the same unless a reaction such as disulfide reduction is also modeled.