Noncovalent Forces That Stabilize Folded Proteins

Rational protein engineering for industrial and therapeutic stability routinely exploits every force described here: engineered disulfide bonds are added to industrial enzymes (proteases, lipases, cellulases used in detergents and biofuel production) to extend their working lifetime at high temperature; surface salt-bridge networks are redesigned to raise the melting temperature of antibody fragments and vaccine antigens for room-temperature storage in the field. Drug design against protein–protein interfaces depends critically on getting van der Waals packing and hydrogen-bond geometry right — a single mismatched methyl group at a binding interface can cost several kJ/mol of binding affinity, the same order of magnitude as the entire folding free energy of a small domain, which is why structure-based drug design tools score candidate molecules using explicit Lennard-Jones and electrostatic terms nearly identical in form to those derived here.

Insulin is a classic case where disulfide chemistry is directly therapeutic: its two chains are held together by two interchain disulfide bonds plus one intrachain bond, and engineered insulin analogs (insulin lispro, insulin glargine) modify surface charge and packing to tune subcutaneous absorption kinetics without disturbing this disulfide scaffold. Antibody engineering exploits interchain disulfides to link heavy and light chains, and next-generation bispecific antibody formats are stabilized by additional engineered disulfide bonds at the domain interfaces. In structural biology and biophysics, hydrogen-bond and van der Waals terms form the core of every molecular dynamics force field (AMBER, CHARMM, OPLS) used to simulate folding pathways, predict mutational effects, and model drug binding; the Lennard-Jones potential derived here is, essentially unmodified, the workhorse nonbonded term in these simulations. Finally, in synthetic biology, de novo protein designers (e.g., Rosetta and deep-learning-based design pipelines) explicitly optimize hydrogen-bond networks, buried salt bridges, and core packing density as primary design objectives, using energy functions built directly from the physical principles worked out in this article — a direct, modern descendant of Pauling's original stereochemical reasoning.

Ask the tutor: How do these individual noncovalent forces combine quantitatively with the hydrophobic effect to set the overall thermodynamic stability measured in a real denaturation experiment?

A folded protein is held together the way a house of cards is held together — not by any single strong joint, but by many weak contacts arranged so that removing one destabilizes the rest. The four forces surveyed here differ enormously in physical origin and in how they scale with distance, and a mechanistic understanding requires treating each on its own terms before assembling the composite picture.

Hydrogen bonds form when a polarized N–H, O–H, or S–H group (the donor) shares its hydrogen with a lone pair on an electronegative acceptor atom, typically the backbone carbonyl oxygen or a side-chain oxygen/nitrogen. The interaction has significant electrostatic character (donor δ+···acceptor δ−) plus a smaller covalent/charge-transfer contribution from partial overlap of the O–H σ orbital with the acceptor lone pair. This mixed character explains why hydrogen bonds are strongly directional: the bond is strongest when the D–H···A angle is close to 180° and the donor–acceptor distance is 2.7–3.0 Å for N–H···O=C. In an α-helix, every backbone C=O (residue i*) hydrogen-bonds to the N–H of residue i+4, tiling a continuous helical ladder of ~3.5–3.8 hydrogen bonds per turn. In β-sheets, the H-bonds run between adjacent strands, either parallel or antiparallel, producing the characteristic pleated geometry.

Van der Waals interactions arise from instantaneous dipole–induced dipole attraction (the London dispersion force) between any two atoms at close range, balanced against Pauli exclusion repulsion when their electron clouds begin to overlap. The attractive term falls off as $1/r^6$ (fast, short-ranged), and the repulsive term rises as $1/r^{12}$, giving the Lennard-Jones potential its signature well-defined minimum — the van der Waals radius sum. Individually these interactions are tiny (0.4–4 kJ/mol per pair), but a well-packed hydrophobic core brings thousands of atom pairs into simultaneous contact, and the cumulative attraction becomes a major stabilizing term, while the repulsive wall enforces the exquisite steric complementarity ("knobs into holes") that gives a protein fold its uniqueness.

Electrostatic interactions include salt bridges (ionized Asp/Glu carboxylates paired with ionized Lys/Arg/His side chains) and longer-range charge–dipole and dipole–dipole effects, notably the interaction of a charged group with the macroscopic dipole moment of an α-helix (created by the net alignment of ~3.5 D peptide-bond dipoles per residue along the helix axis, giving the N-terminus a partial positive character and the C-terminus a partial negative character). Coulomb's law in vacuum predicts huge energies for close ion pairs, but water's high dielectric constant (ε ≈ 80) and competing ion solvation screen most of this, so in situ salt bridges typically contribute only a few kJ/mol to stability — sometimes even destabilizing the fold if the desolvation penalty of burying the charges exceeds the electrostatic gain.

Disulfide bonds are the outlier: a genuine covalent S–S bond (bond energy ≈ 250 kJ/mol) formed by oxidative coupling of two cysteine thiols, typically in the oxidizing environment of the endoplasmic reticulum or periplasm (rarely in the reducing cytoplasm). A disulfide does not usually add much enthalpic stabilization directly (breaking one costs less than its raw bond energy once you account for the fact that the reduced dithiol is also stable), but it dramatically stabilizes the folded state entropically, because a covalent cross-link drastically reduces the number of accessible conformations — and hence the conformational entropy — of the unfolded chain.

The hydrogen bond energy is often modeled (for an electrostatic-dominated picture) using a dipole–dipole or Coulombic approximation, but the essential quantitative content for this chapter is captured by three relationships: the Lennard-Jones potential for van der Waals contacts, Coulomb's law with dielectric screening for electrostatics, and the free-energy bookkeeping that assembles all terms (plus the hydrophobic/conformational-entropy terms treated elsewhere) into the net stability of the fold.

where: $U_{\text{LJ}}(r)$ is the pair interaction energy at center-to-center distance $r$; $\varepsilon$ is the well depth (typically 0.4–4 kJ/mol for C···C, O···O, etc.); $r_{\min}$ is the distance at which $U_{\text{LJ}}$ is minimized (sum of van der Waals radii, e.g., ~3.4 Å for two carbons).

Reading: at $r = r_{\min}$ the potential equals $-\varepsilon$ (maximal attraction); for $r < r_{\min}$ the $r^{-12}$ term dominates and the pair repels steeply; for $r \gg r_{\min}$ both terms vanish and atoms simply stop interacting. This is why packing in a protein core must be nearly perfect — being 0.5 Å too close costs far more energy than being 0.5 Å too far costs in lost attraction.

where: $q_1, q_2$ are the point charges (in Coulombs, e.g. $\pm e = \pm 1.602\times10^{-19}\,\text{C}$ for a unit ionic charge); $r$ is the separation; $\varepsilon_0 = 8.854\times10^{-12}\,\text{F/m}$ is the vacuum permittivity; $\varepsilon_r$ is the relative (effective) dielectric constant of the medium between the charges, ranging from ~2–4 (buried, protein interior) to ~80 (bulk water).

Reading: the $1/\varepsilon_r$ factor is the single most important correction separating textbook electrostatics from real protein behavior — the same salt bridge that would be worth hundreds of kJ/mol in vacuum is worth only a few kJ/mol once immersed in a polarizable, ion-screened aqueous environment.

where: each $\Delta$ term is the (folded − unfolded) contribution of that interaction category; $\Delta S_{\text{conf}}$ is the conformational entropy lost by restricting backbone and side-chain rotamers to one folded state (a large unfavorable term, since $\Delta S_{\text{conf}}<0$ on folding, so $-T\Delta S_{\text{conf}} > 0$).

Reading: $\Delta G_{\text{fold}}$ (typically −20 to −65 kJ/mol) is a small residual of large, partially canceling contributions — the individual terms can each be hundreds of kJ/mol, but they overwhelmingly cancel between the folded and unfolded states because the unfolded chain is also hydrogen-bonded to water and also has van der Waals contacts, just with solvent instead of itself.

Example 1 — Foundational: energy of a single backbone α-helix hydrogen bond. Model the N–H···O=C hydrogen bond as a simple point-charge Coulombic interaction between partial charges $\delta_H = +0.2e$ on the amide hydrogen and $\delta_O = -0.4e$ on the carbonyl oxygen (typical partial charges from protein force fields), separated by the H···O distance of 1.95 Å (i.e., $1.95\times10^{-10}$ m), in an effective local dielectric of $\varepsilon_r = 3.5$ (buried helix interior, not bulk water):

Numerator: $(0.2)(0.4)(1.602\times10^{-19})^2 = 0.08\times2.566\times10^{-38} = 2.053\times10^{-39}\,\text{C}^2$. Denominator: $4\pi(8.854\times10^{-12})(3.5)(1.95\times10^{-10}) = (1.113\times10^{-10})(3.5)(1.95\times10^{-10})$. Compute stepwise: $1.113\times10^{-10}\times3.5 = 3.895\times10^{-10}$; then $\times1.95\times10^{-10} = 7.596\times10^{-20}$.

Convert to molar energy:

This falls squarely within the accepted experimental range for a backbone hydrogen bond (8–30 kJ/mol), confirming the partial-charge Coulomb model captures the right order of magnitude even though the true interaction has additional short-range quantum mechanical (charge-transfer) character not included here. As a limiting check: if $\varepsilon_r$ were instead the bulk-water value of 80 (physically appropriate for a solvent-exposed loop hydrogen bond rather than a buried helical one), the same calculation gives $U \approx -0.71$ kJ/mol — vanishingly weak, which correctly predicts that hydrogen bonds exposed to bulk solvent contribute almost nothing net to folding stability, because the donor and acceptor could instead hydrogen-bond to water with comparable energy. This is precisely why burial of backbone hydrogen bonds away from water, achieved by forming regular secondary structure, is what makes them net stabilizing.

Example 2 — Applied: net stability change from removing a buried salt bridge by mutagenesis (Arg → Ala). A protein engineer mutates a buried Arg residue (forming a salt bridge with a nearby Asp, at $r = 2.8$ Å, in an effective local dielectric $\varepsilon_r = 10$, appropriate for a mostly buried but not fully desolvated interior pocket) to Ala, removing the charge entirely. Estimate the expected change in folding free energy. First compute the vacuum-scale Coulomb term at $r=2.8\times10^{-10}$ m:

Denominator: $4\pi\varepsilon_0 = 1.113\times10^{-10}$; times $2.8\times10^{-10}$ gives $3.116\times10^{-20}$.

Molar: $8.235\times10^{-19}\times6.022\times10^{23} = 4.96\times10^{5}\ \text{J/mol} = 496\ \text{kJ/mol}$ (magnitude; attractive so negative sign):

Screen by $\varepsilon_r=10$:

Now subtract a desolvation penalty. Because this pair is substantially buried (only partially accessible), assume a desolvation cost of 35 kJ/mol (a reasonable mid-range estimate for a mostly buried monovalent pair, smaller than a fully buried pair's 60–80 kJ/mol but larger than a fully exposed pair's ~3 kJ/mol):

So the intact salt bridge is predicted to stabilize the fold by roughly 14.6 kJ/mol relative to having no interaction at all. Mutating Arg→Ala removes both the favorable electrostatic term and the desolvation penalty (since Ala has no charge to desolvate), so the net expected destabilization upon mutation is approximately the full 14.6 kJ/mol — i.e., the mutant is predicted to have $\Delta G_{\text{fold}}$ less negative (less stable) by about 14–15 kJ/mol, a large effect for a single point mutation, consistent with experimentally observed values for disruption of strongly buried salt bridges (often 8–20 kJ/mol) as measured by urea or thermal denaturation assays. As a check on scope: for a fully solvent-exposed salt bridge ($\varepsilon_r\to80$, desolvation penalty →0), the same style of calculation collapses toward the few-kJ/mol values found in Step 3 of the first derivation, correctly reproducing the well-known experimental observation that surface salt bridges contribute far less to stability than buried ones.

Example 3 — Exam/competition level: comparing the entropic stabilization from an engineered disulfide bond. A protein engineer introduces an artificial disulfide bond (two Cys residues, forming an S–S cross-link) between two loop positions separated by $n=40$ residues in the unfolded chain, hoping to stabilize the fold. Estimate the entropic contribution to stabilization using the simple polymer loop-closure model, in which cross-linking reduces the configurational entropy of the unfolded state, thereby stabilizing the folded state relative to unfolded — since the loss of entropy on cross-linking has already happened in the unfolded ensemble sense, but the folded state entropy is barely changed (it's already compact). The entropy cost of forming a loop of $n$ links in a random-coil polymer (Flory-type treatment, simplified Jacobson–Stockmayer form) scales as:

where $C$ is a constant absorbing local chain stiffness and the entropy of the free chain ends (commonly $C\approx 1$ to $3$ for polypeptides depending on parameterization; take $C=1.5$ here as a representative mid-range value), $R = 8.314\ \text{J/(mol}\,\cdot\,\text{K)}$, and $n=40$. Compute $\ln(40) = 3.689$.

This is the entropy lost by the unfolded chain upon cross-linking (forcing the two Cys residues together removes a huge number of the conformations the free chain could otherwise sample). At physiological temperature $T=310\ \text{K}$, the corresponding stabilization of the folded state relative to the now entropically-restricted unfolded state is:

(Here the sign convention is: because the unfolded state loses entropy, its free energy $G_{\text{unfolded}}$ rises by $-T\Delta S_{\text{loop}} = +18.1$ kJ/mol, which is equivalent to the folded state becoming more stable relative to unfolded by 18.1 kJ/mol.) This value is comparable in magnitude to the entire native folding free energy of many small proteins (20–65 kJ/mol), which is precisely why engineered disulfide bonds are a standard and often dramatically effective strategy in protein engineering for enhancing thermostability, particularly for industrial enzymes and antibody fragments. As a limiting-case check: if the two cysteines were adjacent in sequence ($n\to$ very small, e.g. $n=2$, an unrealistic but illustrative limit), $\ln(2)=0.693$ and $\Delta S_{\text{loop}} = -8.314[1.5(0.693)+1.5] = -8.314[1.04+1.5]=-8.314(2.54)=-21.1\ \text{J/(mol}\,\cdot\,\text{K)}$ — a smaller magnitude entropy loss (and hence smaller stabilization) than the $n=40$ case, which correctly reflects the physical intuition that cross-linking two residues that are already close in sequence removes relatively little conformational freedom, whereas stapling together two distant loop regions (large $n$) removes a great deal of entropy from the unfolded ensemble and thus provides much greater stabilization — up to a point; at very large $n$ the $\ln n$ growth is slow, so returns diminish and other factors (loop strain, unfavorable folded-state geometry) begin to dominate the real design outcome.

Linus Pauling's structural chemistry laid the conceptual foundation for nearly everything in this article. Working at Caltech in the 1930s–1950s, Pauling combined X-ray crystallography, quantum mechanical bonding theory, and painstaking model-building to establish the geometric rules of the hydrogen bond — its near-linear preferred geometry, its characteristic donor–acceptor distances, and its partly covalent, partly electrostatic character. His 1951 papers with Robert Corey and Herman Branson, built directly from these hydrogen-bonding rules, predicted the α-helix and β-sheet purely from stereochemical principles — before a single protein crystal structure had confirmed them. Pauling's insistence that hydrogen bonds have a well-defined, calculable geometry (rather than being a vague "sticky" force) transformed protein structure from qualitative description into a predictive, mechanistic science, directly enabling the framework used today to rationalize how noncovalent forces encode a specific three-dimensional fold.