Thermodynamics of Ligand Binding
While all ligands in the CGS 27023 series have been previously described,8
the thermodynamics of binding were unknown. The free energy, enthalpy, entropy, and heat capacity of binding were determined by isothermal titration calorimetry. Heat capacity experiments were conducted at 15°C, 25°C, and 37°C. Ligand binding was characterized in three buffers - MOPS, HEPES, and TRIS – in order to account for proton transfer events.32
Observed enthalpies of binding were plotted against enthalpies of ionization. Linear regression was then used to determine the number of protons, n
, transferred upon binding. Because stromelysin-1 undergoes protonation of H224 upon binding,5
it is critical to ensure that analyses are conducted with the actual thermodynamics of binding. In all cases, the calculated n
was between 0.3 and 0.35 protons mol−1
, which is within error of measurement. The derived thermodynamic parameters for binding are shown in .
Thermodynamics of binding for compounds 1–4. ΔG, ΔH, and TΔS are described in kcal mol−1. ΔCP is reported as cal mol−1 K−1
Previously, enthalpy-entropy compensation has been rationalized by changes in protein or host hydration upon binding.33,34
Fortunately, hydrophobic hydration affects ΔCP
in a predictable fashion. While changes in electrostatics or vibrational states of a protein during binding could, in principle impact ΔCP,35
is the best available measurement of changes in solvent-exposed hydrophobic suface area and is therefore considered a hallmark of the hydrophobic effect.36–38
The values of ΔCP
observed here, ranging from 60–72 cal mol−1
, show no evidence of changes in protein solvation upon binding. Additionally, there is no evidence that the presence of polar ligand functional groups accounts for the observed enthalpy-entropy compensation. Clearly, transfer of water from the protein to the ligand upon binding would contribute to the observed ΔCP
. However, if the transfer of water molecules from the protein surface to polar ligand functionalities contributed significantly to enthalpy-entropy compensation, such trends should correlate the presence or absence of a polar moiety. Compounds 1
contain a pyridyl group absent in the compounds 2
. Nonetheless, compounds 1/2
have similar heat capacities and thermodynamic parameters. We therefore reject changes in protein solvation upon binding as an explanation for the observed thermodynamic parameters and will investigate a possible structural basis for enthalpy-entropy compensation in this system.
Determination of Protein-Ligand Crystal Structures
Both SCD complex structures for compounds 3
crystallized in the same space group with identical cell dimensions (see Supplemental Materials
). Unsurprisingly, both structures of SCD share the same fold as other previously reported SCD.5,23,39–46
Superposition of the SCD structures in complex with inhibitors 3
gives an rmsd of 0.130 Å for 129 C(α) positions (). Difference electron density for the inhibitors was visible in the first unbiased maps. The asymmetric unit of both structures is composed of one monomer, both of which are essentially complete apart from portions of the “long, flexible loop” (amino acids 210–234).5
In the SCD structure complexed with inhibitor 3
, the electron density for amino acids 126, 213, 214, 218, 224, 231, 232, 237, 241 and 248 is weak; amino acids 215, 216 and 225–230 are absent. In the structure bound with compound 4
, the electron density for amino acids 213, 224, 231, 233, 236, 243 and 246–248 is poorly defined, while residues 215, 216 and 225–231 are completely disordered. Disordered amino acids within this flexible loop have been previously reported.5,23,42
The positions and interactions of the catalytic and structural Zn2+
and three Ca2+
ions are essentially identical in both complexes and consistent with previous reports.5,23,39–46
One sulfate ion is also present in the SCD structures complexed with compounds 3
at the same position observed for SCD complexed with other nonpeptide inhibitors.23
Figure 5 Average variation of C(α) upon ligand binding. Based on protein superimposition of PDB files 1BM6, 2JT5, 1BQO, 2JNP, 1B3D, 2USN, 1B8Y, and the crystal structures described herein. Solid line depicts ligands binding only the S1’ hydrophobic (more ...)
Ligand Binding Changes the Local Environment of Tryptophan
Tryptophan is one amino acid that is particularly conducive to study by Raman spectroscopy as specific bands exist that characterize hydrogen bonding, side chain orientation, and hydrophobic environment.19
In resonance Raman spectroscopy, UV excitation amplifies the tryptophan spectrum making them even more apparent. The ratio of the W7 fermi doublet of tryptophan (1340/1360 cm−1
) has been repeatedly validated as a marker of hydrophobicity in the tryptophan local environment.48–50
Unfortunately, attempts at solution phase Raman spectroscopy were limited by sample fluorescence and the Raman microscope available for DCDR did not have the option of UV excitation. Thus, the studies described here were limited to non-resonance Raman spectroscopy. In their initial report on the W7 band, Miuri and coworkers note that the ratio of the W7 doublet cannot be used with visible light excitation as CH aliphatic stretches move overlap the 1340 cm−1
Additionally, the 1360 cm−1
is not always discernible in non-resonance spectra,51
even if readily apparent in the resonance Raman spectra.48,49
One structurally-sensitive vibrational band in the non-resonance Raman spectroscopy of proteins is the W18 stretch (760 cm−1
), corresponding to the indole ring of tryptophan. Because the W18 stretch is among the most intense in the Raman spectrum, changes in intensity are readily detected, and the intensity of this band is related has been previously correlated to the number of hydrophobic contacts made by the indole ring. This deduction is based on work using the model compound, 3-methylindole. By measuring the W18 band intensity of 3-methylindole in solvents including H2
O, 1:1 H2
O:EtOH, EtOH, and pentane, Miura et al.
demonstrated that the band intensity varied inversely with solvent hydrophobicity.52
Because band intensity was comperable in vapor and aqueous spectra, Miura et al. concluded that an increased number of hydrophobic contacts on the indole ring resulted in a relatively decreased band intensity.52
When considering the W18 band in difference Raman spectra where the spectra of unbound protein is subtracted from that of bound protein, increased hydrophobic contacts on tryptophan in the bound state leads to the W18 band appearing as a negative feature. shows negative features in the difference spectra at ~ 760 cm−1
corresponding to the W18 stretch, indicating the presence of an increased number of hydrophobic contacts in the bound state.
There is a significant temptation attempt to analyze the relative intensities of the various W18 ligand bands through the generation of a double difference spectrum. One must, however, exercise caution as any error in the initial Raman spectra has been significantly propogated as two difference spectra were generated. Assuming a 5% error in the intial measurements (inclusive of buffer subtraction), propogation of errors indicates that the error in the difference spectra is 7%. Generation of a double difference spectra would further increase the error to 10%. To determine if any reliable trends in relative W18 band intensities were present, a series of double displacement spectra were generated (supplemental materials Figures S6–S7
). Unfortunately, these spectra yielded unreliable, and at times, contradictory results - likely due to the small variations in peak intensity and the substantial error associated with generating double difference spectra. To ascertain trends amongst ligands, nondifference Raman spectroscopy will have to be combined with other high resolution techniques, such as isothermal titration calorimtery x-ray crystallography, or NMR protein structure determinations.
Despite this limitation, nondifference Raman spectroscopy does provide clear evidence that the number of hydrophobic contacts within the protein interior of stromelysin-1 increases upon ligand binding. Additionally, this technique is reasonably accessible and does not require protein crystals which can be difficult to obtain. There are two ways in which interior tryptophans could experience increased hydrophobic contacts – residue translocation or protein contraction with increased packing of the protein interior.
We first consider the possibility of tryptophan translocation. X-ray crystallographic data show no major change of tryptophan conformation upon ligand binding (). For tryptophan translocation to account for the observed W18 band, the translocation must differ significantly between compounds 1 and 4 since the W18 band is much stronger for both compounds 3 and 4 than for compound 1. shows the change in position of C(α) and C1 and the χ 1,2 from apo protein. While the χ1,2 angle for W92 does change significantly upon binding, the change is consistent across the series. Also, compound 1 exhibits the largest change in W92 χ1,2 dihedral angle but the smallest change in W18 intensity upon ligand binding. These small changes in χ1,2 dihedral are also evidenced by the W3 tryptophan stretch in all difference spectra. Thus, the structural data show no evidence for translocation.
Conformation of the three internal tryptophans of unbound SCD (pink), SCD bound to 1 (blue), and SCD bound to 4 (purple)
Conformations of stromelysin-1 catalytic domains.
Protein contraction would also increase the number of hydrophobic contacts on tryptophan. Because proteins contain intraprotein voids53–57
and have average interior surface complementarities of only 60–80%,58,59
contraction is feasible. An analysis of 50 protein structures demonstrated that most buried tryptophans only have ~60% surface complementarity (surface area contacted by another amino acid residue); 40% of the tryptophan surface lacks any hydrophobic contacts.58,59
Lysozyme is a protein for which ligand binding and protein contraction were convincingly linked using both volume and intrinsic compressibility measurements.60
Ligand binding by lysozyme also decreases the W18 mode intensity and led to negative features at ~ 760 cm−1
which could be contributed to by protein contraction. While direct correlation of the W18 band with protein contraction has not, to our knowledge, been reported in the literature, supporting data has, in fact, been published by several groups independently. In his 2004 review of protein contraction/expansion as a driving force for binding and enthlapy-entropy compensation, Dudley Williams and coworkers described two model systems - avidin-biotin and hemoglobin-oxygen.61
Avidin-biotin binding is a representative system of protein contraction. Using H/D exchange of amide backbone proteins characterized by MALDI, binding of biotin by avadin led to a marked decrease of H/D exchange of the protein backbone, which was interpreted as increased protein packing and therefore decreased access of D2
O to the amide protons.62
Thermodynamically, binding of biotin and avadin demonstrates a marked enthalpic benefit and entropic penalty of binding.63,64
This system has also been studied by difference resonance Raman spectroscopy to analyze changes in the local tryptophan environments upon binding. Both the W7 and W18 bands experienced an increase in intensity indicative of an increase in local hydrophobicity about the indole ring.65
In conglomeration, these results support the use of the W18 band as a marker for protein contraction.
Similarly, when investigating protein expansion as demonstrated by hemoglobin binding to oxygen, the converse finding are obtained.61
Hb exists in two forms in the blood - a “tense” rigid state having a low affinity for O2
and a high affinity “relaxed” state that avidly binds both O2
Using H/D exchange and ESI experiments, Williams and coworkers demonstrated that transitioning from a tense to relaxed state upon binding O2
resulted in increased H/D exchange and protein expansion.62
Subsequent binding events lead to relaxation of additional subunits until the protein is fully relaxed after binding of the fourth O2
molecule. Thermodynamically, this phenomena was entropically driven and entropically opposed, with each successive binding event resulting in a smaller entropic benefit and enthalpic penalty of binding.67,68
The concept of protein expansion is also supported spectroscopically by analyzing work by Nagatomo et al,69
which showed decreased hydrophobicity of tryptophans using the W18 band upon carbon monoxide binding. Because this was a non-resonance study, the tryptophan spectra is not enhanced and the 1360 cm−1
peak is not readily visible for analysis. Carbon monoxide is more facile to use for spectroscopic experiments as its higher affinity results in a more stable complex. Nonetheless, the binding site and binding mode are conserved and protein dynamics should be conserved in both cases.
Ideally, our Raman spectroscopy work would be supplemented by physical measurements of intrinsic compressibility, but these require crystals that we were unable to acquire. Nonetheless, either tryptophan translocation or protein contraction must occur to produce the W18 observations in these studies, and data supporting translocation was lacking. This lack of evidence for translocation combined with the previous correlation of decreased W18 band and protein volume for lysozyme leads us to the conclusion that protein contraction is responsible for the increased hydrophobic contacts on tryptophan.
Consideration of the Interfacial Mobility Model
The interfacial mobility model postulates that high affinity ligand binding results from tightening of the protein-ligand interface, which in the case of stromelysin-1, we propose arises from protein contraction about the ligand. Consequences of protein contraction would include increased internal protein packing, rigidification, and decreased residual movement. Thermodynamically, the interfacial mobility model manifests with an enthalpic benefit and entropic penalty of binding that is inversely proportional to ligand complexity. While it is currently impossible to deconvolute accurately and reliably the contribution of van der Waals interactions and hydrogen bonding to binding enthalpy, there has been success in correlating entropies of binding to either residual protein motion or residual entropy.70,71
As both decreased residual protein movement and increased internal packing are consequences of protein contraction, we correlate our Raman observations of internal packing (), and therefore contraction and decreased residual protein motion, with the experimental entropies of binding ().
In the case of compounds 1 and 2, both enthalpy and entropy contribute to ligand binding, and the magnitude of entropic contribution is comparable (2.1 and 2.7 kcal mol−1 respectively). Entropy does not contribute significantly for binding of compound 3 (–0.1 kcal mol−1), but a significant entropic penalty of binding occurs for compound 4 (–4.7 kcal mol−1). Based on these thermodynamic observations, the interfacial mobility model predicts that compounds 1 and 2 would experience relatively small and comparable contractions as they have comparable entropies of binding. Binding of compound 3 should cause a protein contraction greater than that for compounds 1 and 2 but less than for compound 4.
While internal packing, as demonstrated by the W18 tryptophan band, is one proxy for protein contraction, a second proxy marker for this phenomenon is decreased residual movement. Additionally, x-ray crystallography is high resolution and likely to yield insights into minor differences between ligands. To this end, an analysis of crystal structures was undertaken.5,6,23,39,45,72
Ten solved bound structures, including those for compounds 1
, and 4
were superimposed with apo SCD using SuperPose29
and the difference between C(α) positions for apo- and holoprotein was measured for all residues. These results are plotted graphically in . The average difference in C(α) position across all residues and structures is approximately 1 Å. There are, however, certain protein regions where the difference C(α) is much greater, and these regions correspond to random coil secondary structure. All structures demonstrate increased variability of C(α) position for the loop encompassing residues 221–233. Interestingly, two loops (150–160 and 170–175) demonstrate an increased difference of C(α) position variability for large ligands, as defined by binding more than just the S1’ and Zn2+
subpockets, versus small ligands that bind only the S1’ and Zn2+
subpockets. The average difference in C(α) for large compared with small ligands is 0.9 ± 0.2 versus 2.0 ± 0.5 Å for residues 148–161, respectively. Similarly, the average difference in C(α) for residues 169–175 is 1.1 ± 0.2 versus 1.8 ± 0.3 Å, respectively, for large and small ligands. It is intriguing that these regions in relatively close proximity to the active site () demonstrate increased variability in C(α) large relative to small ligands as this could reflect decreased residual movement of these residues. If C(α) variability between superimposed protein structures truly reflects residual protein motion, this would further support the interfacial mobility model.
Regions of variable mobility for SCD upon ligand binding. Regions depicted in red show a clear increase in mobility when SCD is bound to large ligands. Regions depicted in orange show a trend towards increased mobility when bound to large ligands.
We have presented data that protein contraction, as judged by the W18 Raman band, varies inversely with ligand complexity. Additionally, crystallographic analysis reveals reduced motions in two loops and possibly an α-helix near the active site upon binding of high affinity small ligands. The W18 Raman band indicates that for compounds 1 and 2, the protein contracts identically even though 2 lacks the pyridyl substituent. Similarly, compounds 3 and 4 behave similarly despite the difference of pyridyl moiety. shows the subpockets of the SCD active site. The majority of subpockets are composed of multiple secondary features. The S1’ hydrophobic pocket is made up of a loop (residues 218–222) and an alpha helix (residues 195–201). The S1 groove is formed by a loop (residue 163) and beta sheet (residue 165–166); that same loop also forms the P1 groove (residues 162, 164). Thus, while lacking the pyridyl moiety ring, compound 2 still contacts the P1 loop through binding of its isopropyl group. Although compound 3 contains the pyridine moiety, the structure reveals that this substituent is rotated away from the P1 groove. Thus, compounds 1 and 2 contact the same secondary features, the same observation is true for compounds 3 and 4. The p-methoxyphenyl group for compounds 3 and 4 interacts more extensively with the protein than seen in compound 1. This could result in tightening of the S1’-ligand interface around the smaller ligands.
Meshwork structures of SCD subpockets. A) Empty meshwork of SCD. S1’ pocket is in blue, P1 groove in red, S1 groove in purple. (B) Meshwork of bound compound 1. (C) Meshwork of bound compound 3. (D) Meshwork of bound 4.
When Krishnamurthy et al.
first described the interfacial mobility model, they defined ligand size as the number of “distal residues” contacted.1
In the context of human carbonic anhydrase, the distal residues were amino acids at an increased distance from the catalytic zinc in the conical active site. Because SCD has an active site groove, we originally thought to define ligand size as a function of subpockets occupied. This assumption is flawed because small changes in ligand structure can
dramatically alter the binding mode, i.e. absence of the isopropyl moiety allowed reorientation of the pyridyl moiety in the active site. While a priori
we had expected to see a linear trend in protein contraction, this ligand rearrangement in the active site caused compound pairs 1
to give similar results despite possessing moieties that could have bound additional subpockets. The small ligands 3
bind loop 218–222 and α-helix 195–201 in addition to the catalytic zinc. The large ligands 1
interact with these residues as well as loop 162–164 and β-sheet 165–166. Once “ligand size” was defined by the secondary structural elements involved in ligand binding, which correlates closely with residues contacted, the degree of protein contraction correlates well with “ligand size.” Thus, trends in protein contraction and thermodynamics are in good accord with those predicted by the interfacial mobility model.1