Preferred side-chain constellations at antiparallel
coiled-coil interfaces
Erik B. Hadley†, Oliver D. Testa‡, Derek N. Woolfson‡§¶, and Samuel H. Gellman†¶
†Department of Chemistry, University of Wisconsin, Madison, WI 53706; ‡School of Chemistry, University of Bristol, Bristol BS8 1TS, United Kingdom;
and §Department of Biochemistry, University of Bristol, Bristol BS8 1TD, United Kingdom
Edited by Alan R. Fersht, University of Cambridge, Cambridge, United Kingdom, and approved November 21, 2007 (received for review September 24, 2007)
Reliable predictive rules that relate protein sequence to structure
would facilitate postgenome predictive biology and the engineering and de novo design of peptides and proteins. Through a
combination of experiment and analysis of the protein data bank
(PDB), we have deciphered and rationalized new rules for helix–
helix interfaces of a common protein-folding and association motif,
the antiparallel dimeric coiled coil. These interfaces are defined by
a specific pattern of interactions among largely hydrophobic side
chains often referred to as knobs-into-holes (KIH) packing: a knob
from one helix inserts into a hole formed by four residues on the
partner. Previous work has focused on lateral interactions within
the KIH motif, for example, between an a position on one helix and
a dⴕ position on the other in an antiparallel coiled coil. We show
that vertical interactions within the KIH motif, such as aⴕ-a-aⴕ, are
energetically important as well. The experimental and database
analyses concur regarding preferred vertical combinations, which
can be rationalized as leading to favorable side-chain interactions
that we call constellations. The findings presented here highlight
an unanticipated level of complexity in coiled-coil interactions, and
our analysis of a few specific constellations illustrates a general,
multipronged approach to addressing this complexity.
backbone thioester exchange 兩 knobs-into-holes packing 兩 peptides 兩
protein design 兩 protein folding
he ␣-helical coiled coil is a ubiquitous oligomerization
domain found in a wide variety of proteins (1, 2). Coiled-coil
interactions direct and stabilize the multimerization of transcription factors, motor proteins, cytoskeleton components, extracellular matrix components, and membrane-fusion machinery (3,
4). It has been estimated that 3–5% of translated protein
sequences encode coiled-coil domains (5); indeed, we find that,
of the current 46,968 entries in the protein databank (PDB),
2,738 (5.8%) contain coiled-coil interactions of some type
(O.D.T. and D.N.W., unpublished data). Thus, deciphering rules
that link the sequences of coiled coils to their assembly would
have an impact on many aspects of postgenome predictive
biology (6) and aid our ability to engineer or even design
protein–protein interactions (1, 7, 8).
The effects of sequence variation on coiled-coil stability have
been widely studied (9–16), but substantial gaps in our understanding remain. For example, the vast majority of prior efforts
have focused on parallel coiled-coil homo dimers, but other
coiled-coil states are seen in Nature. Oligomers ranging from
trimer through dodecamer have been observed, as have structures with different types (hetero-oligomers) or arrangements
(parallel vs. antiparallel) of helices (3, 13, 17). In particular,
antiparallel coiled-coil dimers have received little attention
despite increasing recognition of their structural and functional
importance both as stand-alone motifs and as components of
globular protein architectures. In the PDB, for example, we find
1,273 coiled-coil structures with ⱕ70% sequence identity; of
these, 800 (63%) are antiparallel dimeric coiled coils (O.D.T.
and D.N.W., unpublished data). Among this subset, 661 (83%)
are intramolecular and the remainder are intermolecular helix
associations (13, 18). Even among the widely studied parallel
T
530 –535 兩 PNAS 兩 January 15, 2008 兩 vol. 105 兩 no. 2
coiled coils, only a limited set of factors has been shown to
influence stability (1–4), and it seems likely that important
determinants of stability and selectivity remain to be identified.
Here, we show that largely overlooked patterns of side-chain
packing at the interhelical interface contribute to antiparallel
coiled-coil dimerization selectivity. We use a unique experimental approach to evaluate the energetic significance of certain
side-chain juxtapositions that have not been examined. We show
how seemingly subtle side-chain variations can lead to surprising
levels of dimerization preference. Furthermore, we demonstrate
that findings from the model system reflect preferences among
coiled-coil proteins of known three-dimensional structure.
As is true for all coiled coils, the formation of antiparallel
coiled-coil dimers is driven by the burial of hydrophobic side
chains arranged along one side of each participating ␣-helix. The
helices typically display a ‘‘heptad repeat’’ of amino acid residues, with the heptad positions designated abcdefg. Hydrophobic
side chains at a and d align to create the dimerization interface.
Two or more such interfaces come together through the interdigitation of side chains, which Crick described as ‘‘knobs-intoholes packing’’ (KIH) (19). As illustrated in Fig. 1, each buried
a or d side chain (a ‘‘knob’’) has a local packing environment (a
‘‘hole’’) that is defined by four side chains from the partner. We
divide the four interhelical side-chain–side-chain contacts into
two types (Fig. 1): lateral interactions, for which a line between
the two side chains is approximately perpendicular to the helix
axis; and vertical interactions, for which a line between the two
side chains is approximately parallel to the helix axis.
Most prior analyses of a or d side-chain interactions at the
interface have focused on lateral interactions with a⬘ or d⬘
partners (the prime indicates that the residue resides in a
different helix). Extensive work by Vinson et al., for example, has
shown that the contribution of a given buried a side chain to
parallel coiled-coil dimer stability depends on the identity of the
lateral hydrophobic packing partner (a-a⬘ interaction) (9, 12).
We have recently shown that lateral partners (a-d⬘) influence
antiparallel coiled-coil stability (20). To our knowledge, however, the contribution of vertical side-chain contacts to coiledcoil stability and pairing selectivity has not been analyzed in
detail for either parallel or antiparallel coiled coils (21). Indeed,
given the amount of work necessary for previous studies of
lateral pairing, the prospect of extending such an effort to
examine vertical contacts might seem operationally daunting
(9–16).
Author contributions: E.B.H., O.D.T., D.N.W., and S.H.G. designed research; E.B.H. and
O.D.T. performed research; E.B.H., O.D.T., D.N.W., and S.H.G. analyzed data; and E.B.H.,
D.N.W., and S.H.G. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
¶To
whom correspondence may be addressed. E-mail: d.n.woolfson@bristol.ac.uk or
gellman@chem.wisc.edu.
This article contains supporting information online at www.pnas.org/cgi/content/full/
0709068105/DC1.
© 2008 by The National Academy of Sciences of the USA
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0709068105
is an energetically significant interplay among lateral and vertical
interactions.
Vertical Contacts Affect Antiparallel Coiled-Coil Specificity and Stability. The vertical contact residues for the guest site at a in our
model system are designated a⬘1 and a⬘2; Figs. 2 and 3). In our
original study a⬘1 ⫽ a⬘2 ⫽ Leu (20). To evaluate the importance
of vertical contacts, we changed a⬘1 and a⬘2 to Ile for the 25 a-d⬘
variants considered previously (Leu, Ile, Val, Ala, or Asn at a or
d⬘). As shown in Table 1, switching between a⬘1 ⫽ a⬘2 ⫽ Leu and
a⬘1 ⫽ a⬘2 ⫽ Ile causes a significant change in stability for a number
of lateral a-d⬘ pairings. For example, the most stable lateral
pairing changes from Ile-Leu when a⬘1 ⫽ a⬘2 ⫽ Leu to Leu-Leu
when a⬘1 ⫽ a⬘2 ⫽ Ile. Interpretation of these differences, however,
is complicated by the fact that Leu 3 Ile mutations at a⬘1 and a⬘2
lead to multiple changes in lateral and vertical contacts at the
antiparallel coiled-coil interface.
To isolate the energetic contribution to coiled-coil pairing
preference of vertical contacts within the a⬘1-a-a⬘2 triad, we can
consider hypothetical equilibria such as the one illustrated in Fig.
4. These partial helix net diagrams represent the original NT-C
design and three mutants thereof. In each of these thioesters d⬘ ⫽
Leu, whereas positions a, a⬘1, and a⬘2 vary between Leu and Ile,
with a⬘1 ⫽ a⬘2. The figure explicitly indicates the entire hole into
which each a, a⬘1, or a⬘2 knob inserts on intramolecular coiled-coil
formation. The only difference between the pair of coiled coils
on each side of the equilibrium arrows is the identity of the
vertical contacts within each a⬘1-a-a⬘2 triad (Leu-Leu-Leu and
Ile-Ile-Ile ‘‘homo-triads’’ on the left, Leu-Ile-Leu and Ile-Leu-Ile
‘‘hetero-triads’’ on the right).
Table 1. Thermodynamic data obtained from thioester exchange
of NT-C mutants
d⬘
⌬GCC
a⬘ ⫽ a⬘ ⫽ Leu
a
a⬘ ⫽ a⬘ ⫽ Ile
a
Fig. 2. Thioester model system. (a) Design and sequence of NT-C; Succ ⫽
N-terminal succinyl group. Residues a/d⬘/a⬘ correspond to mutations sites. (b)
Thioester exchange process for NT-C. The thioester-thiol pair on the left
comprises N- (blue) and C-terminal (red) segments, whereas the pair on the
right contains the full-length coiled coil and a small thiol.
Hadley et al.
Leu
Ile
Val
Asn
Ala
Leu
Ile
Val
Asn
Ala
⫺1.4
⫺1.7
⫺1.4
0.3
⫺0.7
⫺1.3
⫺1.0
⫺0.8
0.2
⫺0.7
⫺0.9
⫺0.8
⫺0.6
0.4
⫺0.5
0.2
⫺0.1
0.0
0.5
0.4
⫺0.4
⫺0.9
⫺0.9
0.8
0.0
Leu
Ile
Val
Asn
Ala
⫺2.1
⫺1.6
⫺1.6
0.2
⫺0.7
⫺1.9
⫺1.4
⫺1.2
0.1
⫺0.6
⫺1.5
⫺0.7
⫺0.8
0.3
⫺0.4
0.0
⫺0.1
0.2
0.3
0.4
⫺0.7
⫺0.5
⫺0.3
0.3
0.0
Values are reported in kilocalories per mole. Uncertainty ⬇ ⫾0.1 kcal/mol.
PNAS 兩 January 15, 2008 兩 vol. 105 兩 no. 2 兩 531
CHEMISTRY
We have recently developed an experimental design that
facilitates the rapid study of sequence–stability relationships
among antiparallel coiled coils (20, 22–24). Our model system
(NT-C; Fig. 2) consists of two short ␣-helix-prone segments
connected with a flexible linker that allows, but does not enforce,
intramolecular coiled-coil association. The linking segment contains a central thioester bond to allow thiol-thioester exchange.
When NT-C is combined with a small thiol-containing molecule
in aqueous solution at neutral pH, thiol-thioester exchange
occurs rapidly. The equilibrium constant (KTE) provides insight
on the noncovalent attraction between the two ␣-helical segments (Fig. 2b) (20, 22). This system is well suited to exploring
how the interplay between mutations within each helix influences antiparallel coiled-coil stability. As illustrated in Fig. 3, our
original study focused on a lateral (a-d⬘) pair at the interface: an
a position on the N-terminal helix, and a d⬘ position on the
C-terminal helix (20). Synthesis of 10 short peptides (5 alternative residues at each guest site) enabled us to obtain ⌬GTE values
(from KTE) for 25 coiled-coil variants.
Here, we build on this experimental design to show that: (i)
certain combinations of residues in the vertical a⬘-a-a⬘ constellations are favored over others; (ii) vertical a⬘-a-a⬘ constellations
contribute significantly to stability and specificity in antiparallel
coiled-coil dimers; (iii) our findings are robust and not systemdependent; (iv) the preferred combinations are mirrored in
coiled coils of known three-dimensional structure; and (v) there
Fig. 3. Coiled-coil interactions in thioester model system. (a) Helical-wheel
diagram showing the helical regions of NT-C. (b) Partial helical net for NT-C. In
each diagram, the N-terminal segment is shown in blue and the C-terminal
segment is shown in red. Note that the numbering of the mutation sites is
different from that in ref. 20.
BIOPHYSICS
Fig. 1. Knobs-into-hole packing at antiparallel coiled-coil interfaces. (a and
b) Orthogonal views of a knob (light gray) into hole (dark gray) interaction
observed in an experimentally determined protein structure (residues 411 and
58 – 65 of PDB ID code 2ic6). (c) Diagram illustration of b with the heptad
register assignment superimposed. Lateral interactions of the knob residue
are indicated with red arrows, and vertical interactions with blue arrows.
Fig. 5. Oakley model system. (a) Helical-wheel diagram showing the helical
regions of Oakley’s heterodimeric antiparallel coiled coil (25). (b) Partial
helical net for the peptides shown in a. In each diagram, the ‘‘acid’’ peptide is
shown in blue and the ‘‘base’’ peptide is shown in red. The boxes highlight
interactions discussed in the text.
Fig. 4. Partial helical-net diagrams for NT-C and three mutants used to
calculate the discrimination energy (DE). The reported DE value was derived
from the thermodynamic information in Table 1.
⌬GTE values for each of the four coiled coils in Fig. 4 indicate
⌬⌬G ⫽ ⫺0.8 kcal/mol for this equilibrium. Below, we use the
term ‘‘discrimination energy’’ (DE) to describe this type of
thermodynamic comparison, because DE tells us about the
extent to which one pattern of interhelical pairing is favored
relative to another. The result of the equilibrium in Fig. 4 is
designated DELI because we are comparing the impact of the
vertical environments provided by a⬘1 ⫽ a⬘2 ⫽ Leu vs. a⬘1 ⫽ a⬘2 ⫽
Ile. More specifically, we can designate this result as DELI(L/I)
to indicate that the left side has a ⫽ Leu in the Leu-a-Leu vertical
triad, and a ⫽ Ile in the Ile-a-Ile vertical triad. The deduced
DELI(L/I) value indicates that the sum of vertical contacts in the
Leu-Ile-Leu plus Ile-Leu-Ile vertical hetero-triads is 0.8 kcal/mol
more favorable than the sum of vertical contacts in the LeuLeu-Leu plus Ile-Ile-Ile vertical homo-triads.
Testing the Ile/Leu Triads in an Independent System. The information
gleaned from our thioester-based model system will not be useful
unless it is broadly applicable to antiparallel coiled-coil dimers.
We turned our attention to a different system to test generality.
Very few heterodimeric antiparallel coiled coils have been
subjected to careful biophysical scrutiny (18); among the bestcharacterized examples is a design generated by Oakley et al. (25)
This system consists of negatively charged (‘‘acid’’) and positively
charged (‘‘base’’) partners. The hydrophobic core comprises
primarily leucine residues, with Asn at a single a position on each
component to control relative helix orientation. The Oakley
system is illustrated in Fig. 5, which identifies the sites we used
for substitutions. The vertical triad on which we focus comprises
the residues designated a*, a⬘1*, and a⬘2* in Fig. 5; in the original
sequence, all three of these residues are Leu. This system should
be sufficiently different from our thioester system to represent
a valid test for generality because the Oakley helices are twice as
long as the helical segments in NT-C, and the residues at the b,
c, and e–g positions vary significantly between the two systems.
A key similarity between the two model systems is a leucine-rich
hydrophobic interface, which enables the comparison of specific
interfacial mutations in these different antiparallel coiled-coil
contexts.
In the Oakley heterodimer, the side chain from residue a*, an
a position on the acid peptide, fits into a hole on the base peptide
532 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0709068105
defined by one lysine and three leucine residues (which include
a⬘1* and a⬘2*). We prepared the a* ⫽ Ile mutant in addition to the
original acid peptide. Two base peptides were prepared, the
original sequence (a⬘1* ⫽ a⬘2* ⫽ Leu) and a double mutant (a⬘1* ⫽
a⬘2* ⫽ Ile), so that we could change the vertical contact residues
for a*. We examined all four acid–base peptide pairings, that is,
the coiled coils that contain all four possible vertical triads
among Leu-Leu-Leu (the original Oakley design), Ile-Ile-Ile,
Leu-Ile-Leu, and Ile-Leu-Ile. Analytical ultracentrifugation
showed that each acid–base pair is dimeric [see supporting
information (SI) Text].
Urea-denaturation experiments revealed significant differences
in the four coiled-coil stabilities (Table 2). There is a reasonable
correlation between ⌬GUrea for the four acid–base peptide pairs
and ⌬GTE for the analogous members of the thioester system (Fig.
6), which is noteworthy in light of the substantial differences
between the two coiled-coil systems. (These differences include
variation in the magnitude of ⌬G for coiled-coil formation: ⫺8 to
⫺11 kcal/mol for the Oakley system vs. ⫺1.4 to ⫺2.1 for the
thioester system.) The four ⌬GUrea values can be used to calculate
DELI(L/I) ⫽ ⫺3.2 kcal/mol for the Oakley system. In a complementary study, we prepared derivatives of the two acid and two base
peptides bearing N- or C-terminal cysteine residues, respectively, to
allow direct determination of the discrimination energy under
native conditions by disulfide exchange (26–28). This approach
indicated DELI(L/I) ⫽ ⫺2.6 kcal/mol, in good agreement with the
value deduced from the unlinked species. Overall, the experiments
based on the Oakley antiparallel coiled coil are consistent with the
thioester experiments in revealing that vertical hetero-triads containing Leu and Ile are preferred relative to the homo-triads. The
quantitative differences between the two systems suggest that the
thioester approach underestimates vertical contact energetics, perhaps because in this case the a⬘2 position is near the ‘‘open’’ end of
the helical hairpin and thus may be subject to fraying (29), which
could reduce the measured free energies.
Table 2. Comparison of thermodynamic data obtained for
Oakley acid/base mutants and from thioester exchange
of the corresponding NT-C mutants
NT-C
a/a⬘1,2
L/L
I/I
I/L
L/I
Oakley heterodimeric coiled-coil
⌬Gfold/TE, kcal/mol
⌬Gfold/Urea, kcal/mol
Tm, °C
⫺1.4
⫺1.6
⫺1.7
⫺2.1
⫺8.0
⫺9.8
⫺10.0
⫺11.0
45
53
55
58
Hadley et al.
Fig. 6. Correlation of ⌬Gfold/Urea for the Oakley acid– base mutations and
⌬Gfold/TE for the corresponding mutations in NT-C. The line corresponds to a
linear regression fit (y ⫽ 3.9321x ⫺ 3.0308, r2 ⫽ 0.855).
which is consistent with the model system results. Furthermore,
by analogy with the theoretical equilibrium shown in Fig. 4, the
data of Table 3 can be used to calculate a theoretical discrimination energy DELI(L/I)⬘ ⫽ ⫺0.81 kcal/mol. The similarity of
this result to the experimental DE value is remarkable, although
both values are subject to errors. Thus, the trend highlighted by
the experimental model studies is precisely mirrored in the
coiled-coil structures of the protein databank: the vertical hetero-triads Leu-Ile-Leu and Ile-Leu-Ile are preferred relative to
the homo-triads Leu-Leu-Leu and Ile-Ile-Ile.
Preferred Side-Chain Constellations. For each of the homotypic and
heterotypic vertical triads containing Leu and Ile residues, the
structures found in the CC⫹ database were superimposed (Fig.
7). We have used such superimpositions for rationalizing statistically favored interstrand side-chain–side-chain interactions in
␤-sheets (30). The overlaid structures reveal that the statistically
favored heterotypic combinations, Leu-Ile-Leu and Ile-Leu-Ile,
are the ‘‘tightest’’ structurally, as indicated by two types of
analysis. First, we calculated the rmsd values measured across the
side-chain heavy atoms for each of the four sets of superimposed
structures. The rmsd values were smaller for the statistically
favored vertical hetero-triads (0.96 Å for Leu-Ile-Leu and 0.77
Å for Ile-Leu-Ile) relative to the homo-triads (1.58 Å for
Ile-Ile-Ile and 1.52 Å for Leu-Leu-Leu). This trend suggests that
Table 3. Numbers of specific aⴕ-a-aⴕ combinations in natural antiparallel coiled-coil structures
a⬘ ⫽ a⬘ ⫽ Ile
Amino acid at a
Ile
Leu
Sum (all amino acids)
a⬘ ⫽ a⬘ ⫽ Leu
Observed
Expected
Observed/expected
Observed
Expected
Observed/expected
6
19
59
9.4
14.2
0.6
1.3
26
21
103
16.5
24.7
1.6
0.9
Expected numbers were calculated by using the percentage of occurrence of Ile and Leu at all a sites of coiled-coil structures in the CC⫹ database with 70%
sequence identity or lower. These values are 16% and 24%, respectively. Note: the theoretical DELI(L/I)⬘ value is equal to ⫺RT(ln(26/6) ⫺ ln(21/19)); the ratios of
the raw numbers for Ile at both sites or Leu at both sites are self-normalizing, though the same result is obtained by using the normalized values (⫺RT(ln1.6 ⫺
ln 0.9 ⫹ ln1.3 ⫺ ln 0.6)).
Hadley et al.
PNAS 兩 January 15, 2008 兩 vol. 105 兩 no. 2 兩 533
CHEMISTRY
energetic importance of vertical contacts in two antiparallel
coiled-coil model systems led us to ask whether this pattern of
side-chain interactions plays a role in natural coiled coils. We
used the program SOCKET, which searches protein-structure
coordinate files for KIH interactions characteristic of coiled-coil
structures, and the SOCKET-derived database of structurally
verified coiled coils, CC⫹ (13). Recently, we created an interface
to search CC⫹ for various coiled-coil parameters, such as
oligomer state and helix orientation, and also for more specific
features, such as sequence motifs (O.D.T. and D.N.W., unpublished data). By using these resources, we searched the PDB for
examples of antiparallel two-helix coiled-coil structures with
70% sequence identity or less. From this subset we culled
examples of vertical residue triads (a⬘-a-a⬘) in which a was Leu
or Ile and the residues at a⬘ were either both Ile or both Leu;
these vertical triads allow comparison with our experimental
studies. Defining the residues at each site in the a⬘-a-a⬘ vertical
triads limited the number of coiled-coil substructures returned
in our searches: there were 2,072 KIH interactions for all types
of a⬘-a-a⬘ vertical triads in the subset of antiparallel two-helix
coiled-coil structures, but only 162 that matched our constraints.
Nonetheless, the number of structures was sufficient to perform
reliable analyses and draw conclusions.
Table 3 gives the raw counts for each of the residues at the
central a position in the different vertical backgrounds, a⬘ ⫽ a⬘ ⫽
Ile or a⬘ ⫽ a⬘ ⫽ Leu. Expected numbers for each combination
were calculated based on the frequencies of these residues at the
a sites of all KIH interactions in the coiled-coil database with
sequences ⱕ70% identical. The ratio of the observed/expected
numbers is effectively a propensity for a side chain to be in the
specified environment. Gratifyingly, the hetero-triads have favorable propensities (⬎1) and the homo-triads are disfavored,
Fig. 7. Superposition of structures culled from the PDB for vertical triads
containing Ile and Leu in antiparallel two-helix coiled coils. Superpositions
were performed by using the MMTSB toolset (41), and rendered in PYMOL
(42). The PDB and associated chain and residue identifiers for these structures
are given in SI Table 12. For each overlay, a representative backbone is shown
to guide the eye (backbone rmsd ⫽ 0.33– 0.86 Å for the four Ile-Leu combinations at a⬘aa⬘).
BIOPHYSICS
Analysis of Antiparallel Coiled-Coil Structures from the PDB. The
Fig. 8. Significant discrimination energies DELI(X/Y) for d⬘ ⫽ Leu and d⬘ ⫽ Ala (see Fig. 4). Values were determined to be significant if DE ⱖ0.4 kcal/mol (two
times the expected error). Positive values (blue) indicate a preference for X paired with leucine vertical contacts, whereas negative values (red) indicate a
preference for X paired with isoleucine vertical contacts. Note that only one side of the diagonal (dotted line) is shown because DELI(X/Y) ⫽ ⫺DELI(Y/X). These
graphs are intended to show qualitatively that the residue at d⬘ can strongly influence vertical pairing preferences of the residue at a⬘. DE data are available in
SI Tables 7–11.
the favored vertical triads lead to preferred conformations of
side chains, or what we term preferred side-chain constellations.
In a second mode of analysis, we examined the side-chain
dihedral angles for all four vertical triads culled from the CC⫹
database (see SI Table 12). The results indicate that the heterotriads sample fewer of the possible side-chain conformations
than do the homo-triads. This trend supports the notion that the
conformational space sampled is tighter for the favored heterotriads than for the homo-triads. Indeed, for the hetero-triads,
preferred side-chain conformations were apparent. For instance,
the dominant constellation for Leu-Ile-Leu had Ile in the g ⫺t
conformation and both Leu in t g⫹; for Ile-Leu-Ile, the dominant
constellation had the side-chain conformation pattern g ⫺t, t g⫹,
g ⫺t. Each side-chain conformation in these dominant constellations is fully compatible with the ␣-helix. All Ile display the g
⫺t conformation, which is the favored rotamer (81%) for Ile in
␣-helices (31), and all Leu display t g⫹, the second-most
populated conformer (30%) for Leu in ␣-helices. Further inspection of the structures revealed that in the homo-triads only
‘‘glancing’’ contacts are made between individual methyl groups
from different side chains. In contrast, in the favored heterotriads more intimate contacts occur between side chains, with
the terminal methyl groups of the knob making multiple contacts
with the vertical partners to give ‘‘nested-packing’’ interactions
(32).
Interplay Among Vertical and Lateral Interactions. The analyses
described above show that the data obtained by using our
thioester-based antiparallel coiled-coil model system provide
insight that is relevant to side-chain–side-chain packing preferences in natural coiled coils. This correlation encouraged further
analysis of the data in Table 1. These ⌬GTE data allow the
calculation of 50 independent DELI(X/Y) values, for X ⫽ Leu,
Ile, Val, Asn, or Ala and Y ⫽ Leu, Ile, Val, Asn, or Ala (see SI
Tables 7–11). Each of these DELI(X/Y) values reflects an
energetic comparison of two different antiparallel coiled-coil
pairings: Leu-X-Leu plus Ile-Y-Ile vs. Leu-Y-Leu plus Ile-X-Ile.
Because DELI(X/Y) is meaningless if X ⫽ Y, and because
DELI(X/Y) ⫽ ⫺DELI(Y/X), 10 independent DELI(X/Y) values
result when X and Y each vary among five residues. The number
of DELI(X/Y) values that can be calculated from our ⌬GTE data
expands by a factor of five because we have examined five
different residues (Leu, Ile, Val, Asn, or Ala) as the lateral
partner (d⬘) for the central residue in the a⬘-a-a⬘ vertical triads
under scrutiny. Comparing these 5 sets of 10 DELI(X/Y) values
should indicate whether there is an energetically significant
interplay between the lateral a-d⬘ and the vertical a⬘-a-a⬘ inter534 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0709068105
actions experienced by a given a residue when it occupies its hole
on the partner helix.
As a prelude to exploring the interplay between lateral and
vertical interactions, we used an absolute DELI value of ⱖ0.4
kcal/mol (twice the estimated experimental uncertainty) as the
criterion to identify those cases in which there is a significant
preference for one pair of dimers relative to the alternative pair.
By this criterion, 21 of the DELI values generated from our ⌬GTE
data (42%) are significant. This proportion indicates that the
impact of vertical interactions within antiparallel coiled coils is
not limited to the Leu-Ile combinations examined above, but
rather that many vertical triads can exert a substantial influence
on antiparallel coiled-coil dimerization selectivity.
Comparison among the five sets of 10 DELI(X/Y) values
indicates that there is an energetically significant interplay
between the lateral partner d⬘ and the vertical partners a⬘ in their
interactions with an a residue from the partner helix (a knob
fitting into an a⬘, d⬘, a⬘, e⬘ hole). We illustrate this point in Fig.
8 by comparing the significant DELI values for d⬘ ⫽ Leu and for
d⬘ ⫽ Ala. (Analogous results for d⬘ ⫽ Ile, Val, and Asn may be
found in SI Fig. 18.) This comparison shows that the identity of
d⬘ can determine whether DELI is significant for an X/Y pair as
the knob residues (a). To our knowledge, all published efforts to
account for or predict sequence–stability or sequence–
selectivity relationships among coiled coils focus on pairwise
interactions between side chains that are brought into contact on
coiled-coil formation. The distinctions among our d⬘ ⫽ Leu and
d⬘ ⫽ Ala datasets, however, show that the energetic impact of
having both vertical neighbors (a⬘) Ile or both Leu on the knob
residue at a is influenced by the identity of the d⬘ residue. Thus,
treating vertical contacts in a simple pairwise or even triad
fashion is not adequate for complete rationalization of coiledcoil stability or pairing preference. Similarly, treating lateral
interactions in a simple pairwise fashion is not adequate, because
of interplay involving vertical contacts (see SI Fig. 19).
Conclusions. We have uncovered a substantial thermodynamic
influence exerted by previously unexplored patterns of intermolecular side-chain contacts at antiparallel coiled-coil interfaces.
Our results provide the first quantitative analysis of vertical
contact effects on coiled-coil pairing preference. Data derived
from model systems indicate that vertical packing interactions
involving buried side chains at a positions of the coiled-coil
heptad repeat can have a significant effect on antiparallel
dimerization selectivity. This trend is found in natural antiparallel coiled-coil dimers of known structure, with certain a⬘-a-a⬘
hetero-triads, Ile-Leu-Ile and Leu-Ile-Leu, favored relative to
the corresponding homo-triads. Analyses of the favored strucHadley et al.
tural motifs reveal preferred constellations in which the side
chains adopt favorable dihedral angles and form intimate
contacts.
Based on these results, it seems likely that vertical interactions
represent an important but largely unrecognized component of
the ‘‘code’’ that determines coiled-coil interaction preferences in
complex biological environments. Past studies of coiled-coil
pairing preferences, mostly involving parallel dimers, have identified relatively few mechanisms by which selectivity can be
achieved. This limited set of rules has allowed considerable
progress in the areas of coiled-coil prediction and design (1, 3),
but our findings suggest that additional advances will be possible
based on a more sophisticated treatment. Previously elucidated
factors include lateral pairing of nonpolar side chains (9–16),
which is driven by stereochemical complementarity, and lateral
pairing of polar side chains (9–16, 33), which is driven by
hydrogen bonding (at least for Asn-Asn pairs). In addition,
pairing preferences are influenced by the electrostatic interactions (attractive or repulsive) between ionized side chains (34–
39). Our findings indicate that vertical interactions must be
considered as well if we are to understand the origins of
dimerization selectivity among natural coiled-coil sequences and
expand current abilities to predict and design such interactions.
Here, we have explicitly addressed interactions associated with
a⬘-a-a⬘ vertical triads in antiparallel dimers, but the approaches
we have used can readily be extended to d⬘-d-d⬘-type vertical
interactions at antiparallel coiled-coil interfaces, to d⬘-a-d⬘ and
a⬘-d-a⬘ vertical interactions at parallel coiled-coil interfaces, and
to other more-complex combinations of side chains.
In addition to highlighting the role of vertical contacts at
coiled-coil interfaces, our results show that simple pairwise
analysis of side-chain–side-chain interactions at such interfaces
does not fully capture the factors that control stability or partner
preference. High-fidelity interactions are common and necessary in biology, and proteins must find their partners in the
cellular milieu efficiently and without engaging in promiscuous
associations. Coiled-coil motifs represent a natural mechanism
for guiding and cementing protein–protein interactions. Because
3–5% of the proteome is estimated to encode such motifs (5),
however, selectivity in coiled-coil pairing constitutes an extraordinary feat of molecular recognition and discrimination. Our
results provide an expanded framework for exploring partner
selectivity among natural coiled-coil sequences and for designing
new sequences that display high-fidelity pairing with a specified
partner.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
23. Woll MG, Hadley EB, Mecozzi S, Gellman SH (2006) J Am Chem Soc 128:15932–15933.
24. Hadley EB, Witek AM, Freire F, Peoples AJ, Gellman SH (2007) Angew Chem Int Ed
46:7056 –7059.
25. McClain DL, Woods HL, Oakley MG (2001) J Am Chem Soc 123:3151–3152.
26. Lumb KJ, Kim PS (1995) Biochemistry 34:8642– 8648.
27. Harbury PB, Zhang T, Kim PS, Alber T (1993) Science 262:1401–1407.
28. Bilgicer B, Xing X, Kumar K (2001) J Am Chem Soc 123:11815–11816.
29. Rohl CA, Baldwin RL (1994) Biochemistry 33:7760 –7767.
30. Hutchinson EG, Sessions RB, Thornton JM, Woolfson DN (1998) Protein Sci 7:2287–2300.
31. Lovell SC, Word JM, Richardson JS, Richardson DC (2000) Proteins 40:389 – 408.
32. Wouters MA, Curmi PMG (1995) Proteins 22:119 –131.
33. Gonzalez L, Woolfson DN, Alber T (1996) Nat Struct Biol 3:1011–1018.
34. Krylov D, Mikhailenko I, Vinson C (1994) EMBO J 13:2849 –2861.
35. Grigoryan G, et al. (2006) PloS Comp Biol 2:551–563.
36. Campbell KM, Lumb KJ (2002) Biochemistry 41:7169 –7175.
37. Burkhard P, Ivaninskii S, Lustig A (2002) J Mol Biol 318:901–910.
38. McClain DL, Gurnon DG, Oakley MG (2002) J Mol Biol 324:257–270.
39. Ryan SJ, Kennan AJ (2007) J Am Chem Soc 129:10255–10260.
40. Bernstein FC, et al. (1977) J Mol Biol 112:535–542.
41. Feig et al. (2001) MMTSB Tool Set (MMTSB NIH Research Resource, The Scripps Research
Institute, La Jolla, CA).
42. DeLano WL (2002) The PyMOL Molecular Graphics System (DeLano Scientific, Palo Alto,
CA). Available at: www.pymol.org. Accessed November 18, 2007.
15.
16.
17.
18.
19.
20.
21.
22.
Hadley et al.
Backbone Thioester Exchange Assays. Assays were typically initiated by mixing
⬇0.1 mM each peptide in 50 mM buffer (phosphate, pH 7) and by allowing
equilibration for 90 min. Details are provided in SI Text.
ACKNOWLEDGMENTS. We thank Josh Price for assistance with the CD data
workup and Beth Bromley for valuable discussions on mutant cycles. T his work
was supported National Institutes of Health Grant GM-61238, and by a Biotechnology and Biological Sciences Research Council and Engineering and
Physical Sciences Research Council (United Kingdom) studentship (to O.D.T.).
PNAS 兩 January 15, 2008 兩 vol. 105 兩 no. 2 兩 535
CHEMISTRY
Analysis of PDB Structures. The November 1, 2007, release of the PDB (40) was
searched for helix– helix interactions with knobs-into-holes packing by using
the program SOCKET (14) with a packing cutoff of 7.0 Å. More sophisticated
searches were used to create a subset comprising two-helix, antiparallel
coiled-coil motifs with 70% sequence identity at the most, and with ␣-helices
longer than 11 aa. Full details of CC⫹ and our analysis interface will be
presented elsewhere. Amino acid frequencies from this subset of interactions
were normalized as described in the text.
BIOPHYSICS
11.
12.
13.
14.
Woolfson DN (2005) Adv Protein Chem 70:79 –116.
Mason JM, Arndt KM (2004) ChemBioChem 5:170 –176.
Lupas A (1996) Trends Biochem Sci 21:375–382.
Gruber M, Lupas AN (2003) Trends Biochem Sci 28:679 – 685.
Wolf E, Kim PS, Berger B (1997) Prot Sci 6:1179 –1189.
Fong JH, Keating AE, Singh M (2004) Genome Biol 5:R11.
Hill RB, Raleigh DP, Lombardi A, DeGrado WF (2000) Acc Chem Res 33:745–754.
Petka WA, Harden JL, McGrath KP, Wirtz D, Tirrell DA (1998) Science 281:389 –392.
Acharya A, Rishi V, Vinson C (2006) Biochemistry 45:11324 –11332.
Mason JM, Schmitz MA, Muller K, Arndt KM (2006) Proc Natl Acad Sci USA 103:8989 –
8994.
Havranek JJ, Harbury PB (2003) Nat Struct Biol 10:45–52.
Acharya A, Ruvinov SB, Gal J, Moll JR, Vinson C (2002) Biochemistry 41:14122–14131.
Walshaw J, Woolfson DN (2001) J Mol Biol 307:1427–1450.
Keating AE, Malashkevich VN, Tidor B, Kim PS (2001) Proc Natl Acad Sci USA 98:14825–
14830.
Tripet B, Wagschal K, Lavigne P, Mant CT, Hodges RS (2000) J Mol Biol 300:377– 402.
Wagschal K, Tripet B, Lavigne P, Mant C, Hodges RS (1999) Protein Sci 8:2312–2329.
Walshaw J, Woolfson DN (2003) J Struct Biol 144:349 –361.
Oakley MG, Hollenbeck JJ (2001) Curr Opin Struct Biol 11:450 – 457.
Crick FHS (1953) Acta Crystallogr 6:689 – 697.
Hadley EB, Gellman SH (2006) J Am Chem Soc 128:16444 –16445.
Kwok SC, Hodges RS (2004) J Biol Chem 279:21576 –21588.
Woll MG, Gellman SH (2004) J Am Chem Soc 126:11172–11174.
Methods