Water and the hydrophobic effect

Lesson 1.5 of Biology Primer for Computer Scientists. Download the code as a notebook.

Imagine holding two oily side chains apart in water, then allowing them to come together. Their atoms can make favorable van der Waals contacts. But their meeting also removes some surface from contact with water. The water molecules that occupied that region can move elsewhere, and the remaining water rearranges around the new shape.

If we calculate only the direct interaction between the two side chains, we leave out part of the physical system. The surrounding liquid helps determine whether they stay together. This is the starting point for understanding the hydrophobic effect and the compact cores of many soluble proteins.

Water has neighbors too

A water molecule can donate hydrogen bonds through its two hydrogens and accept through electron pairs on oxygen. In liquid water, these interactions continually change as molecules move. A water molecule does not carry a permanent set of four partners, and the liquid is not a rigid lattice. Nevertheless, the possibilities for hydrogen bonding influence which local arrangements are favorable.

Dissolving something changes those possibilities. A hydroxyl group can exchange hydrogen bonds with water. An ion attracts favorably oriented water molecules around its charge. A hydrocarbon surface has neither an ordinary hydrogen-bond donor nor an acceptor to offer. Water can still attract it through dispersion interactions, but those interactions do not reproduce water's interactions with itself.

The solvent must also make space for any inserted molecule. Imagine removing enough water to create a small cavity, placing the solute inside, then allowing everything to readjust. This is a conceptual way to account for dissolution. The cost of creating space and the favorable interactions made after insertion both contribute to the final result.

Hydrophobic means that exposure to water is unfavorable relative to a suitable nonpolar environment. It does not mean that water and a nonpolar molecule exert no attraction on one another. A small amount of hydrocarbon can dissolve; the unfavorable free-energy balance limits how much.

Count arrangements of the whole system

To compare association with separation, we need free energy. At fixed temperature and pressure,

ΔG=ΔH−TΔS.

Here ΔH measures the enthalpy change, including changes in interactions, and ΔS measures the entropy change. Entropy concerns how many microscopic arrangements are available. For W equally probable arrangements, S=kBln⁡W. Increasing the number of favorable ways the whole system can arrange itself can make a process favorable even when a simple pair-energy calculation misses it.

Constrain two solutes to remain together and they lose some freedom of position and orientation. That is an entropic cost. But bringing their nonpolar surfaces together changes the solvent's available arrangements too. If the solvent benefit and direct contacts outweigh the costs, association becomes favorable under the chosen conditions.

For small nonpolar solutes near room temperature, the reduced entropy of hydration is a useful part of the explanation. Water around a solute has altered positional and orientational possibilities. The familiar drawing of an ordered cage can suggest this restriction, provided we do not imagine a literal shell of frozen water. At larger nonpolar surfaces, creating and reducing an interface becomes important, and the balance of enthalpy and entropy changes. Size and temperature affect the mechanism; “water gains entropy” is not a complete explanation for every hydrophobic process.The distinction between small-solute hydration and larger interfaces is developed in this review of hydrophobic assembly.

For protein reasoning, keep the comparison concrete: how much nonpolar surface touches water before and after the change, what other interactions change, and how much freedom the chain loses. These questions remain useful when a one-sentence account of water's behavior becomes too simple.

A surface-area model we can calculate

A rough solvent model assigns a free-energy cost to exposed nonpolar area:

Gsolvent≈γAnonpolar.

Take γ=0.025 kcal mol⁻¹ Å⁻² as an illustrative coefficient. Burying 100 Ų of nonpolar area then changes this term by about −2.5 kcal/mol, roughly four times the room-temperature thermal scale. This is one modeled contribution, not a prediction of the total folding or binding free energy. The coefficient depends on the model and how area is defined.

There is a geometric reason compact shapes help. For a sphere, A=4πr2 and V=4πr3/3, so A∝V2/3. Combine the volumes of two equal spheres into one sphere: the new radius is 21/3 times the old radius and its area is 22/3 times one original area. Relative to the original total area of two spheres, only 2−1/3≈0.79 remains. About 21% of the area disappears.

Proteins cannot merge their atoms like droplets; their chains and side chains impose different packing constraints. But the surface-to-volume argument explains why a surface penalty favors compact assemblies. It also gives us a model simple enough to manipulate.

Loading the interactive figure…

The spheres interact through an exposed-area penalty and geometric exclusion. Drag one out of a cluster and watch how much accessible area appears. Run the simulation with and without the surface penalty. The model demonstrates how a solvent contribution can favor clustering without an added pairwise attraction between spheres.

The green points in this model mark solvent-accessible surface, the locations a water-sized probe can reach while rolling over the solute. Some points disappear when another sphere blocks the probe. We will compute a similar area for a real protein below.

Collapse is only the beginning of a fold

A soluble protein often buries many of its nonpolar side chains together. Their burial reduces the area exposed to water, while close packing supplies favorable dispersion contacts. Charged and polar side chains often remain exposed, where they can keep interacting with water.

But the chain cannot place each residue independently. Every side chain is attached to a polar backbone. Folding it inward also moves backbone N–H and C=O groups away from water. The structure must accommodate them through hydrogen bonds, internal water, or other favorable interactions. It must also fit the atoms together without severe overlaps and pay for the chain's loss of conformational freedom.

This explains why “put hydrophobic residues inside” cannot specify a unique native structure. Many compact arrangements bury nonpolar area. Most will pack badly, leave polar groups without partners, or require unfavorable backbone geometry. The detailed chemical constraints select among the arrangements that a broad tendency toward burial makes worth considering.

Look through ubiquitin's nonpolar side chains in the first view, then switch to the surface. In the third view, find Leu 8, Ile 44, and Val 70. These nonpolar groups remain exposed and form a recognition patch used by many ubiquitin-binding partners.

Open 1UBQ in Mol*

An exposed patch is not necessarily a mistake in the fold. It can be part of the protein's job. A partner can cover that patch when it binds, creating additional nonpolar burial at the interface. A patch therefore suggests a place to investigate, although geometry and other chemical contacts still decide which partner can bind there.

Measure how much water can reach

The Shrake–Rupley algorithm estimates solvent-accessible area by placing sample points around each atom, enlarged by a water-sized probe. A point is accessible if no other atom blocks the probe at that location. Summing exposed points gives an approximate area.

For each residue, divide that area by a reference maximum for the residue type. The ratio lets us compare small alanine with much larger tryptophan. The code classifies below 20% as buried; this is a chosen threshold for the calculation, not a chemical phase transition.

import numpy as np
from Bio.PDB.SASA import ShrakeRupley
from primer import structure, THREE_TO_ONE

chain = structure("1UBQ")[0]["A"]
for water in [r for r in chain if r.id[0] != " "]:
    chain.detach_child(water.id)
ShrakeRupley().compute(chain, level="R")

# Largest accessible area of each residue type in an extended chain (Ų, Tien et al. 2013)
max_area = {"A": 129, "R": 274, "N": 195, "D": 193, "C": 167, "Q": 225, "E": 223, "G": 104, "H": 224, "I": 197,
            "L": 201, "K": 236, "M": 224, "F": 240, "P": 159, "S": 155, "T": 172, "W": 285, "Y": 263, "V": 174}
exposure = {r: r.sasa / max_area[THREE_TO_ONE[r.get_resname()]] for r in chain}
classes = {"hydrophobic (AVLIMFW)": "AVLIMFW", "charged (DEKR)": "DEKR", "polar and other": "CSTNQHYGP"}
for name, letters in classes.items():
    values = [x for r, x in exposure.items() if THREE_TO_ONE[r.get_resname()] in letters]
    buried = sum(x < 0.2 for x in values)
    print(f"{name:22} {len(values):2} residues, mean exposure {np.mean(values):.0%}, {buried} buried (< 20%)")
hydrophobic (AVLIMFW)  25 residues, mean exposure 16%, 17 buried (< 20%)
charged (DEKR)         22 residues, mean exposure 44%, 2 buried (< 20%)
polar and other        29 residues, mean exposure 42%, 5 buried (< 20%)

Compare the hydrophobic and charged groups. In ubiquitin, the hydrophobic group has lower average exposure and more buried residues. The result is a population pattern, so an exposed leucine or a partly buried charged residue does not invalidate it. Inspecting the exceptions often leads us to a binding surface or a specific internal interaction.

Now change the environment in your mental picture. Put a protein surface against the nonpolar tails of a membrane, and hydrophobic residues can be favorable on that exterior. Put a polar-lined channel through the membrane, and water and suitable ions can occupy its interior. The same chemical groups follow the same interaction rules; what changes is what touches them.

Check yourself

A calculation says that a mutation buries an extra 80 Ų of nonpolar area. With γ=0.025 kcal mol⁻¹ Å⁻², what change does the surface term predict? Why should you hesitate to call this the mutation's stability change?

Reasoning

The term changes by −80(0.025)=−2.0 kcal/mol. The mutation can also change packing, hydrogen bonds, charge states, and the conformations available to folded and unfolded chains. Those changes belong in the stability comparison as well.

Replace a buried leucine with an aspartate. Predict a source of destabilization, then give two responses that could reduce the cost.

Reasoning

A usually charged carboxylate has lost favorable hydration in a site formerly occupied by a nonpolar side chain; the replacement may also leave a packing defect. The protein could admit water or rearrange the side chain toward solvent. Its pKa could also rise enough to favor neutral COOH. Which response occurs depends on the surrounding structure and pH.

A model rewards only low exposed nonpolar area. Could it distinguish a correctly folded protein from a compact arrangement with badly directed backbone N–H groups?

Reasoning

Not reliably. Both can bury similar areas. The model has omitted the geometry and chemistry of backbone hydrogen bonds, so it cannot penalize an unsatisfied buried donor merely because it points the wrong way. A good collapse signal is insufficient for an atomic model of a fold.

If nonpolar burial often favors folding, why might a soluble protein retain a nonpolar surface patch? How would you turn that idea into an experiment?

Reasoning

The patch may bind another molecule, which covers it upon association. One could alter a patch residue and measure binding while also checking that the protein remains folded. Otherwise, reduced binding might simply result from destabilizing the whole protein rather than changing the binding surface.