Burial and surfaces
"Buried" and "exposed" have been rules of thumb so far. To compute with them, as a docking program scoring a surface patch or a predictor trained to label residues as buried, you need a definition precise enough to put a number on every atom. That definition is geometric: roll a ball the size of a water molecule over the protein and see what it touches.
Cores and surfaces
The passage below uses two terms unit 3 develops: tertiary structure is the 3D fold of a whole chain, and secondary structure means its regular local shapes, the α-helices and β-strands of lesson 2.2, which pair up into sheets.
A dominant molecular interaction in the tertiary structure of proteins is the hydrophobic effect. Residues with hydrophobic side chains are packed into the core of the protein, away from the solvent, while charged and polar residues form the surface of the protein and are able to interact with polar water molecules and solvated ions. Although residues with hydrophobic side chains fit naturally into the core of the protein, their polar polypeptide backbone does not. For the backbone to participate in the hydrophobic core, its hydrogen bonding groups must be satisfied such that their polarity is, in effect, neutralized. Ordered secondary structural elements (helices and sheets) provide this neutralization through their regular hydrogen bonding patterns. Thus, secondary structural elements are critical to the formation of the hydrophobic core.
Residues with polar side chains can also participate in the hydrophobic core. They are, however, subject to same restrictions as the polypeptide backbone: they must be involved in an interaction that neutralizes their polarity. Buried polar residues can form hydrogen bonds with other polar residues, or with sections of the polypeptide backbone not participating in a regular secondary structure. Alternately, in some proteins small pockets exist in which buried polar residues satisfy their hydrogen bonds with water molecules. These water molecules are completely isolated from the solvent and are integral to the protein structure.
Charged residues can also occur within the hydrophobic core. This arrangement is possible only if the charged residue is paired with another residue of opposite charge such that the net charge is zero. This interaction is known as an ion pair or salt bridge.
Ubiquitin. In the first view its hydrophobic side chains are amber sticks, and most of them point into the space between the β-sheet and the helix. In the second view, the solvent-excluded surface colored by residue class, the outside is mostly polar (aqua) and charged (blue and red), with a few amber patches showing through.
Open 1UBQ in Mol*Rolling a probe
That part of a protein exposed to the surrounding environment of solvent, and other molecules, is generally described as its solvent-accessible surface. It is this surface and its constituent atoms that are important to the protein's function.
Lee and Richards described a protein's solvent-accessible surface by using the "rolling ball" algorithm. The algorithm uses a sphere of a particular radius to roll over the surface atoms of the protein. The radius of the sphere can be altered, but it is usually set to 1.4 Å, the radius of a water molecule. Each atom is represented by its van der Waals radius, and the center of the sphere traces out the solvent-accessible surface. Connolly further developed this algorithm to run on computer graphics systems and describe the related concept of the solvent-excluded surface (otherwise known as the molecular surface). The solvent-excluded surface is where the atom's van der Waals radius comes into contact with the probe and forms the boundary for the excluded volume.
A slice through a made-up protein. The grey discs are atoms at their van der Waals radii, and the green disc is the probe. The orange line is the solvent-accessible surface, the path of the probe's center; it is each atom's radius grown by the probe radius, wherever no other atom's grown radius covers it. The shaded blue region is everything the probe can't reach: the atoms plus crevices and pockets too narrow for it, bounded by the solvent-excluded surface. Move the pointer around the molecule to place the probe, and change its radius: at 0 Å both surfaces collapse onto the atoms, and a larger probe can't follow the groove at the top.
The accessible surface lies a probe radius out from the atoms, so its area is larger, and it splits cleanly into one patch per atom. That makes it the standard way to count how exposed each atom or residue is. The excluded surface is the boundary a water molecule actually feels, and it is what a viewer draws when you ask for "the surface", including Mol*'s molecular surface.
Properties of a molecule such as electrostatic potential (the energy a unit charge would have at each point, set by all the molecule's charges), hydrophobicity, and some noncovalent interactions can be modeled as a surface plane or as a surface that defines a volume in or around the molecule. The surface plane approach maps the property to a curved plane with proximity to the residue or atom comprising that property.
Measuring exposure
The standard algorithm, Shrake and Rupley's (1973), samples. Put about a hundred evenly spaced points on each atom's expanded sphere (van der Waals radius plus 1.4 Å), throw away the points that fall inside any other atom's expanded sphere, and the surviving fraction times the sphere's area is that atom's accessible area. It fits in a dozen lines of NumPy:
import numpy as np
from primer import structure
RADII = {"C": 1.70, "N": 1.55, "O": 1.52, "S": 1.80} # van der Waals radii, Å
def sphere_points(n=100):
"""n roughly even points on a unit sphere (a Fibonacci lattice)."""
i = np.arange(n) + 0.5
polar = np.arccos(1 - 2 * i / n)
turn = np.pi * (1 + 5 ** 0.5) * i
return np.c_[np.cos(turn) * np.sin(polar), np.sin(turn) * np.sin(polar), np.cos(polar)]
def shrake_rupley(coords, radii, probe=1.4, n=100):
R = radii + probe # each atom's sphere, grown by the probe radius
unit = sphere_points(n)
area = np.zeros(len(coords))
for i, (center, r) in enumerate(zip(coords, R)):
others = np.flatnonzero(np.linalg.norm(coords - center, axis=1) < r + R.max())
others = others[others != i]
points = center + r * unit # n points on this atom's sphere
dist = np.linalg.norm(points[:, None] - coords[others][None], axis=2)
exposed = (dist >= R[others][None]).all(axis=1) # not inside any neighbor's sphere
area[i] = 4 * np.pi * r**2 * exposed.mean()
return area
chain = structure("1UBQ")[0]["A"]
atoms = [a for a in chain.get_atoms() if a.get_parent().id[0] == " "] # protein only, no waters
coords = np.array([a.coord for a in atoms])
area = shrake_rupley(coords, np.array([RADII[a.element] for a in atoms]))
print(f"{len(atoms)} atoms, accessible surface {area.sum():.0f} Ų")
print(f"atoms with no accessible surface at all: {(area == 0).mean():.0%}")
602 atoms, accessible surface 4859 Ų atoms with no accessible surface at all: 46%
Biopython's ShrakeRupley does the same with a slightly different radius table, and its total for ubiquitin
comes out within 1% of ours. To compare residues of different sizes, divide each residue's area by the largest it could have, measured on the same residue
flanked by glycines in an extended chain (Tien et al. 2013). The result is the relative solvent
accessibility, from 0 (fully buried) to about 1 (as exposed as it gets). Below about 0.2 a residue is
usually called buried. The cutoff is a convention; 0.16, 0.20, and 0.25 are all in use.
from Bio.PDB.SASA import ShrakeRupley
from Bio.PDB.DSSP import residue_max_acc
model = structure("1UBQ")[0]
chain = model["A"]
for res in [r for r in chain if r.id[0] != " "]:
chain.detach_child(res.id) # drop waters: we want exposure to bulk solvent
ShrakeRupley().compute(model, level="R")
max_asa = residue_max_acc["Wilke"] # Tien et al. 2013 maxima, Ų
rsa = {res.id[1]: res.sasa / max_asa[res.get_resname()] for res in chain}
buried = [res for res in chain if rsa[res.id[1]] < 0.2]
print(f"Biopython total: {sum(r.sasa for r in chain):.0f} Ų")
print(f"{len(buried)} of {len(chain)} residues buried:", " ".join(f"{r.get_resname()}{r.id[1]}" for r in buried))
Biopython total: 4893 Ų 24 of 76 residues buried: ILE3 VAL5 THR7 ILE13 LEU15 VAL17 ASP21 ILE23 VAL26 LYS27 ILE30 ILE36 GLN41 LEU43 ILE44 PHE45 LEU50 THR55 LEU56 TYR59 ILE61 SER65 LEU67 LEU69
Most of the buried residues are I, L, V, and F. The rest are the exceptions.
The exceptions, in one small protein
Every buried polar or charged side chain should have a partner to neutralize it. Check that for ubiquitin by looking for nitrogen or oxygen atoms within 3.5 Å of each buried polar side chain's N and O atoms:
from Bio.PDB import NeighborSearch
POLAR = {"ASP": ["OD1", "OD2"], "GLU": ["OE1", "OE2"], "LYS": ["NZ"], "ARG": ["NE", "NH1", "NH2"],
"SER": ["OG"], "THR": ["OG1"], "ASN": ["OD1", "ND2"], "GLN": ["OE1", "NE2"], "TYR": ["OH"],
"HIS": ["ND1", "NE2"]}
search = NeighborSearch([a for a in chain.get_atoms() if a.element in "NO"])
for res in buried:
name = res.get_resname()
if name not in POLAR:
continue
partners = sorted({f"{b.get_parent().get_resname()}{b.get_parent().id[1]}.{b.get_id()} "
f"({b - res[atom]:.1f} Å)"
for atom in POLAR[name] for b in search.search(res[atom].coord, 3.5)
if b.get_parent() is not res})
print(f"{name}{res.id[1]:<3} rsa {rsa[res.id[1]]:.2f} partners: {', '.join(partners)}")
THR7 rsa 0.16 partners: LEU8.N (3.2 Å), LYS11.O (3.1 Å), THR9.N (3.3 Å), THR9.OG1 (3.2 Å) ASP21 rsa 0.17 partners: ASN25.ND2 (3.4 Å), GLU18.N (3.1 Å), GLU18.O (3.5 Å) LYS27 rsa 0.06 partners: ASP52.OD2 (2.9 Å), GLN41.O (3.2 Å) GLN41 rsa 0.00 partners: ILE36.O (3.0 Å), LYS27.O (3.0 Å) THR55 rsa 0.18 partners: ASP58.N (3.3 Å), ASP58.OD1 (3.4 Å), ASP58.OD2 (3.5 Å), SER57.OG (3.0 Å) TYR59 rsa 0.13 partners: GLU51.N (3.0 Å) SER65 rsa 0.12 partners: GLN62.N (3.2 Å), GLN62.O (2.5 Å)
Every one has a partner. Gln 41 is completely buried and hydrogen-bonds to two backbone carbonyl oxygens. Lys 27 is the buried charge, and its amine sits 2.9 Å from an oxygen of Asp 52: a salt bridge. The threonines and serine reach backbone N–H or C=O groups next to them, Tyr 59's hydroxyl hydrogen-bonds to a backbone nitrogen, and the one buried aspartate, Asp 21, accepts hydrogen bonds from the backbone N–H of Glu 18 and the side-chain amide of Asn 25.
The third kind of exception goes the other way: hydrophobic residues left on the surface on purpose. Ubiquitin's best-known feature is a hydrophobic patch formed by Leu 8, Ile 44, and Val 70. It is where almost every protein that recognizes ubiquitin binds. Our numbers give Ile 44 a relative accessibility of 0.15, just under the cutoff, while Leu 8 (0.69) and Val 70 (0.22) are exposed. An exposed hydrophobic patch costs a little stability and buys a sticky spot for a partner, so finding one on a structure is a good hint of a binding site.
First view: the buried Lys 27–Asp 52 salt bridge, with the distance between the charged nitrogen and oxygen, and the buried Gln 41 nearby. Second view: Leu 8, Ile 44, and Val 70 (amber) poking through the surface on one face of the protein, the patch that ubiquitin-binding proteins grab.
Open 1UBQ in Mol*Membrane proteins turn the pattern inside out, since the outside of a membrane-spanning segment faces lipid instead of water (lesson 3.5).
Check yourself
Hemoglobin is four chains, two α and two β, packed together (an α₂β₂ tetramer). Would a residue's relative accessibility computed from a single chain match the value computed from the whole tetramer? Which would you use to find the residues at the interface between chains?
Reasoning
Not for residues at a chain–chain contact. In the isolated chain the partner chains are gone, so residues that touch them look exposed; in the tetramer they are buried. That difference is the standard way to define an interface: compute each residue's accessible area in the single chain and in the complex, and the residues whose area drops are the interface.
You double the probe radius from 1.4 to 2.8 Å. What happens to the total solvent-accessible area of a protein, and to the solvent-excluded volume?
Reasoning
The excluded volume grows, because a bigger probe fits into fewer crevices, so more of the small grooves and pockets count as inside. The accessible area changes in two opposing ways. Every atom's sphere grows by 1.4 Å, which inflates the area, but more of each sphere is now covered by its neighbors' larger spheres and more crevices close off. Past about 1.4 Å the inflation wins: Biopython gives ubiquitin 4,890 Ų with a 1.4 Å probe and 5,140 Ų with 2.8 Å. Below 1.4 Å the crevices win, and a vanishingly small probe sees about 7,960 Ų, the bare van der Waals surface with every crevice open.
In a new structure you find a buried aspartate with no positive residue nearby and no hydrogen-bond partner within 3.5 Å. List what could explain it before you call it an error.
Reasoning
It may be protonated and neutral: burial raises an aspartate's pKa (lesson 1.3), and a neutral carboxylic acid costs much less to bury. It may bind a metal ion or an ordered water molecule that the model doesn't include, or that sits outside your search because waters were stripped. It may have a partner just beyond 3.5 Å in a lower-resolution structure. It may be functional, for example a catalytic residue kept in a strained state on purpose. Only after ruling those out is a modeling error the best explanation.
A drug-design group wants to block ubiquitin from binding its partners. Using only this lesson's numbers, where would you suggest a small molecule should bind, and why there?
Reasoning
At the Leu 8, Ile 44, Val 70 patch. It is the exposed hydrophobic surface that partners already bind, and a small hydrophobic molecule pays the same desolvation price there that a partner does, getting nonpolar surface out of water. Buried residues are out of reach without unfolding the protein, and the charged surface is solvated by water and mostly bound weakly.