A consensus now exists that general anesthetics exert their effects through binding and modulating the activities of ion channels and other integral membrane proteins. Unfortunately, despite recent successes 
, it remains difficult to visualize the details of how anesthetics interact with these targets, since integral membrane proteins are challenging systems for high resolution structural studies. Until structure determination for membrane proteins becomes routine, surrogate model systems will remain valuable tools for advancing our understanding of how proteins recognize anesthetics. Apoferritin arguably represents one of the most useful such models now available. Its structure is well characterized, and it recognizes a wide variety of inhaled and intravenous anesthetic agents. Importantly, apoferritin's binding affinities for different anesthetics are comparable to clinically relevant concentrations; in contrast, the prokaryotic ion channel GLIC displays very different patterns of anesthetic recognition than mammalian anesthetic targets 
, despite possessing a pentameric architecture that is similar to those of known mammalian targets such as the acetylcholine and GABAA
Barbiturates can now be added to the list of general anesthetics recognized by apoferritin, providing a second example of recognition of an injected
general anesthetic (the first being propofol). Interestingly, both propofol and the barbiturates act largely via GABA transmission, as both can directly gate and potentiate the inhibitory GABAA
receptor. For propofol and related compounds, affinity for apoferritin is very close to the EC50
for GABA potentiation 
; the same appears true for barbiturates. EC50
values for GABA potentiation by thiopental, pentobarbital, and phenobarbital are close to their Kd
values for apoferritin (), bolstering the argument that the anesthetic binding sites on apoferritin and the GABAA
receptor share structural and physicochemical similarity.
Within the barbiturate class, different compounds potentiate GABA effects and mediate anesthesia with different potencies; these differences are largely mirrored in their affinities for apoferritin. It is therefore interesting to ask which features of the apoferritin-barbiturate interaction are responsible for these differences in affinity. Thiopental and pentobarbital are iso-structural, differing only at a single atom, and adopt essentially identical orientations and conformations within the apoferritin binding site. The slightly larger sulfur atom gives thiopental a slightly larger volume and surface area, but such trivial size differences seem unlikely to explain an approximate five-fold difference in affinity. A more likely explanation is thiopental's more hydrophobic nature, as reflected by its higher partition coefficient (). If indeed the apoferritin binding site preferentially recognizes thiopental on the basis of its lower polarity, then this argues that the form of the anesthetic bound to the protein is the less polar keto tautomer.
Phenobarbital has a substantially lower affinity for apoferritin than either thiopental or pentobarbital. While it is similar to in size to the other two compounds, it is more polar than either, consistent with an apoferritin binding site that prefers less polar ligands. Also, modeling suggests that phenobarbital's rigid phenyl ring could lead to steric clashes between the ligand and the protein; the methylbutyl substituent of thiopental and pentobarbital, while not much smaller than the phenyl ring, is substantially more flexible, and hence more easily accommodated within the binding site.
There are both similarities and differences in how apoferritin recognizes barbiturates versus other, chemically distinct classes of anesthetics (Figures S3
). Similarity can be found in a common binding area shared by haloalkanes, halogenated ethers, propofol analogs, and barbiturates; this area is a small hydrophobic patch created by Tyr-28, Leu-24, and Leu-81 (, S3
, and S4). In the barbiturate complexes this hydrophobic patch interacts with the ethyl substituent, in the propofol complexes this position is occupied by an aromatic ring, and in the halothane and isoflurane complexes it is occupied by two fluorine atoms (Text S1
Diverse general anesthetics utilize a common binding site in apoferritin.
Differences in ligand recognition exist between the apoferritin-barbiturate and apoferritin-propofol complex structures. Even though both propofol and the barbiturates contain aromatic rings, the planes of these rings are rotated nearly 90 degrees between the two classes of drug. The highly hydrophobic propofol analogs bind with their hydroxyl groups (their only potential hydrogen bond donor/acceptor) pointing toward the opening of the protein cavity, away from the cavity walls, and do not use this hydroxyl group to interact with the protein. In contrast, the position of the barbiturates allows them to hydrogen bond with Ser-27 of apoferritin. This hydrogen bond, together with a potential polar interaction between the barbiturate ring and Arg-59, are the most notable differences between how apoferritin recognizes the barbiturates and how it recognizes other general anesthetics. Apoferritin appears to rely principally upon van der Waals interactions when binding other anesthetics, with little or no contribution from polar interactions. van der Waals interactions do contribute to recognition of the barbiturates, particularly between the alkyl substituents of the drugs and hydrophobic side chains of the protein. However, the favorable enthalpies of binding provided by the polar interactions are likely to substantially exceed the contribution from these van der Waals interactions, and dominate the binding energies.
The exploitation of polar interactions for barbiturate recognition has interesting implications for anesthetic binding. First, it means that the apoferritin site (and probably any site that responds to multiple classes of general anesthetics) must be versatile, and able to rely to differing degrees upon hydrophobic versus polar interactions, depending upon the ligand. Such an amphipathic site should be able to recognize different targets in a versatile manner 
. For example, apoferritin contains polar side chains within its hydrophobic cavity, but these side chains are capable of interacting with partners other than the ligand. In the absence of ligand, the hydroxyl group of Ser-27 can hydrogen bond with a backbone carbonyl oxygen of the protein; in the presence of barbiturates, this same side chain interacts with the drug. In this way, anesthetics can competitively alter hydrogen-bonding patterns within their binding site, and potentially affect protein stability or dynamics 
Another way in which proteins can flexibly recognize different ligands is by stretching or shrinking their binding pockets. However, proteins are limited in the degree to which they can expand or contract a cavity before their structural integrity is compromised. Thus, the volume of the anesthetic-binding cavity in apoferritin varies to some degree when binding different ligands, but affinity is reduced for ligands that are too large to be comfortably accommodated within the acceptable range of cavity volumes. Interestingly, thiopental and pentobarbital are larger than other apoferritin anesthetic ligands examined to date, and yet they are still bound with reasonably high affinity. We suggest that an explanation lies with the polar interactions that the barbiturates make within the apoferritin binding pocket.
For ligands that rely solely on van der Waals interactions for recognition, larger ligands have larger surface areas, and can therefore develop larger favorable enthalpies of binding; however, they will also be bound more snugly in the binding pocket, imposing an entropic penalty. At some point, the incremental enthalpic gains associated with increased surface area will be outweighed by the increasing entropic penalty. For nonpolar ligands, the competing enthalpic and entropic effects lead to maximum affinity when the ligand's volume corresponds to approximately 50–60% of the available cavity volume 
. In contrast, strong polar interactions between a ligand and its target contribute substantially larger favorable enthalpies of binding than do van der Waals interactions, and so when polar interactions are present, greater losses in conformational entropy (and hence higher packing densities) can be tolerated. This appears to be the case with anesthetic binding to apoferritin (). For propofol-like compounds that make no polar interactions with the protein, affinity decreases when ligand volume is increased beyond 50% of the cavity volume 
. However, for compounds like thiopental and pentobarbital that make polar contacts with apoferritin, ligands can fill well over 50% of the available cavity volume, yet still bind with high affinity. Unfortunately, it is not possible to extract the loss in ligand conformational entropy from our calorimetric data, since calorimetric estimates of ΔS include contributions from the displacement of water.
The packing density/affinity relationship for barbiturates is offset from that of the propofol-like compounds, interpreted here as the addition of an entropic penalty offset by new enthalpic gains.
We caution that understanding why a particular anesthetic binds with high affinity to a given target will not necessarily provide insights into general anesthesia, since anesthetic side effects are presumably also mediated by interactions with protein targets, some of which may occur with high affinity. However, the side-effect profiles of currently-used anesthetics vary considerably, suggesting that, unlike anesthetics' clinically useful effects, side effects tend to derive from chemotype-specific interactions. Therefore, targets that recognize many different anesthetic chemotypes are likely to reveal binding properties that are relevant to the production of the anesthetized state.
In summary, we present the first structures of barbiturates bound to a specific protein target. They add support to the notion that direct protein interactions subserve anesthetic effects. Further, these data emphasize that apoferritin is a unitary anesthetic binder, capable of accommodating different general anesthetic chemotypes. Its binding site contains a three-dimensional distribution of amphipathic features that allow recognition of a diverse collection of anesthetic ligands with remarkably high, and clinically relevant, affinity. Thus, it would seem that the apoferritin site can be considered a prototypical anesthetic binding site, and used for purposes as diverse as database mining or screening of novel anesthetic candidates.