|Home | About | Journals | Submit | Contact Us | Français|
The MM-PBSA MD method was used to estimate the affinity, as represented by log kb, of each of a variety of biologically active molecules to a carbon nanotube in an aqueous environment. These ligand-receptor binding simulations were calibrated by first estimating the log kb values for eight ligands to human serum albumin, HSA, whose log kb values have been observed. A validation linear correlation equation was established [R2 = 0.888 Q2 = 0.603] between the observed and estimated log kb values to HSA. This correlation equation was then used to re-scale all MM-PBSA MD log kb values using a carbon nanotube as the receptor. The log kb of the eight HSA ligands, nine polar and/or rigid ligands and six nonpolar and/or flexible ligands to a carbon nanotube were estimated. The range in re-scaled log kb values across this set of 23 ligands is 0.25 to 7.14, essentially seven orders of magnitude. Some ligands, like PGI2, bind in the log kb = 7 range which corresponds to the lower limits of known drugs. Thus, such significant levels of binding of biologically relevant compounds to carbon nanotubes might lead to alterations in the normal pharmacodynamic profiles of these compounds and be a source of toxicity. Ligand binding potency to a carbon nanotube is largely controlled by the shape, polarity/nonpolarity distribution and flexibility of the ligand. HSA ligands exhibit the most limited binding to a carbon nanotube, and they are relatively rigid and of generally spherical shape. Polar and/or rigid ligands bind less strongly to the carbon nanotube, on average, than nonpolar and/or flexible ligands even though the chosen members of both classes of ligands in this study have chain-like shapes that facilitate binding. The introduction of only a few strategically spaced single bonds in the polar and/or rigid ligands markedly increases their binding to a carbon nanotube.
Research into the toxicity of nanomaterials, generally referred to as “nanotoxicity”1, 2 , is being carried out, but at a far slower pace than research into how nanomaterials can be employed3-5. Significant future efforts in nanotoxicity are going to be required in order to discern the effects that nanoscale materials may have on humans and the environment6, 7. However, it is not clear how to develop an arsenal of meaningful and reliable nanotoxicity screens and assays8. One approach to consider as a member of the arsenal is to perform simulation and modeling experiments which explore the behavior of nanomaterials in relevant biological settings. These simulation and modeling studies could serve as prescreens to identify the more likely sources and mechanisms of nanotoxcity.
Carbon nanotubes are among the currently most popular nanomaterials, and are being evaluated for a wide variety of possible applications. Toxicology studies of the effects of carbon nanotubes have been performed since 20039. However, the results are relatively few, and it is not possible to make full conclusions and accurate human assessments. There have been some recent studies showing the promise in the use of carbon nanotubes for biomedical applications that included exploration of possible toxic side effects10-13. As part of one related investigation of this type, it was found that after injecting mice with single-walled carbon nanotubes, “near-complete clearance [of the nanotubes] from the main organs in 2 months” and “no toxic side effects” were observed14. While these new findings are encouraging, other reports have suggested that nanotubes cause pulmonary damage to the lungs of mice and alter protein expression15, 16. In addition, it has been reported that carbon nanotubes adsorb essential micronutrients17.
One of the applications of carbon nanotubes being explored is as selective membrane channels18, 19. The carbon nanotube is inserted across a cellular membrane in order to selectively transport agents in and out of cells. Recently we have performed simulation and modeling experiments to determine what changes in cellular membrane structure and function can occur due to the insertion of carbon nanotubes across the membrane. The results of our study20 indicate four types of changes in the properties of the membrane bilayer;
A general consideration when exploring the possible toxicity of a drug-candidate is to investigate its binding propensity to sites other than to the receptor it is designed to engage21. Not all such secondary binding is bad, and, in fact, such secondary binding is sometimes useful in delivering the drug candidate to the location of its targeted receptor. But, generally, the more specific and selective is the binding of the drug candidate to only its receptor site, the safer is the drug candidate. In the case of exploring sources of nanotoxicity, the nanoparticle now becomes a possible new receptor site to which drugs and/or other small biologically significant compounds might unexpectedly bind. The binding of a particular drug or biologically significant compound to a nanoparticle may not lead to toxicity, but the possibility necessitates being able to determine if such binding is indeed possible, and to what extent it might occur.
The objectives of the work reported in this paper were a) to develop a molecular modeling and computational scheme to estimate the binding constants of drugs and small biological compounds to carbon nanotubes, and, b) to compare these binding constants to those observed for drugs binding to a known drug carrier protein, namely human serum albumin, HSA22, 23. The known low to medium affinity range of HSA ligands is the same affinity range one would expect for relatively non- to semi-specific ligand-receptor binding as presented by a carbon nanotube to an arbitrary ligand. Thus, comparative and validation use of the HSA-ligand system constitutes a relevant, reliable and appropriate approach to accomplish the objectives in modeling carbon nanotube- ligand binding.
By being able to estimate the binding constants of drugs and other biologically significant molecules to nanotubes, and then comparing them to known ligand binding constants to HSA, one is in a position to meaningfully postulate which drugs and biological molecules will have their pharmacodynamic profiles24 most influenced by the presence of carbon nanotubes. The alteration of a pharmacodynamic profile of a drug, and/or biologically relevant small molecule, in turn, can be used as a caution flag [pre-screen] for possible nanotoxicity, and thereby signal the need for expanded toxicological testing. Of course, this reasoning assumes that the carbon nanotubes are present in sufficient concentrations to fully realize the pharmacological effects of ligands binding to them.
The biophysical environment assumed in performing the simulation studies reported here is that, at a minimum, a significantly large area of a carbon nanotube surface is exposed to a ligand in an aqueous media to realize the most favorable binding of the ligand to the nanotube. The overall strategy to estimate the binding constants of drugs and small biologically relevant molecules to carbon nanotubes consists of the following steps;
The structure of the carbon nanotube is well-defined owing to its rigidity, and the periodicity in the chemical structure of the carbon nanotube is quite limited and correspondingly permits a full exploration of plausible initial binding modes between a ligand and carbon nanotube. Moreover, validation of the binding estimation methodology on a much more complex system in terms of both 3D-chemical structure and range in atom diversity of the receptor, namely HSA, supplies confidence that the methodology will work using the carbon nanotube as the ‘receptor’. The details on each of the components of the methodology involved in this study are as follows;
The following features, based upon the carbon nanotube, were considered as constraints to introduce diversity into the set of additional ligands used for binding simulations studies to the carbon nanotube;
The set of HSA ligands, and their observed and estimated binding affinities to HSA, expressed as logkb with kb being the binding equilibrium constant, are given in Table 1. An increase of logkb corresponds to an increase in binding affinity. Figure 1 shows propofol bound at its low concentration HSA binding site as determined from the MM-PBSA MD method employing the crystal structures of free and propofol bound HSA as initial MD simulation structures22. Only the residues that form the ‘lining’ of the propofol-HSA binding site are shown in Figure 1.
The linear correlation equation between the observed logkb and the estimated logkb of the ligands to HSA is given by;
An inspection of Part B of Table 1 reveals that the estimated and observed logkb of the ligands generally differ in absolute value by less than one [ a factor of 10 in kb ] with ibuprofen being the largest outlier. A plot of the estimated versus observed logkb values is given in Fig. 2 where its seen that each estimated logkb value is larger than the observed value for each ligand. An R2 = 0.89 in eq. (2) indicates that the estimated logkb captures about 90% of the variance in the observed logkb over a range in binding of nearly four orders of magnitude. Thus, it is reasonable to postulate that the application of the methodology employed for estimating ligand-HSA binding, including the MM-PBSA approach, can be applied to at least differentiate strong and weak binding ligands to a carbon nanotube. In addition, the estimated magnitude of binding of a ligand to a carbon nanotube can be meaningfully compared to the range in estimated binding constants of known ligands to HSA.
Table 2 reports the chemical structures along with the use and function of a set of selected polar and/or rigid ligands whose chemical structures were selected to introduce diversity with respect to the three constraints presented in Section E of Methods. These ligands would not be expected to bind particularly well to the carbon nanotube based upon constraints a) and b) and chemical intuition.
A set of nonpolar and/or flexible ligands, whose chemical structures are shown in Table 3, have also been chosen as binding candidates to a carbon nanotube. These ligands have chemical structures which are conducive to binding to the carbon nanotube based on constraints a) and b) given in Section E of Methods.
Table 4 contains the composite set of 23 ligands described in Tables Tables11 – 3 whose binding potencies to a carbon nanotube, as measured by logkb, have been determined by applying the MM-PBSA MD method and eq. (2). Part A of Table 4 includes the ligands of Table 1 which are also observed to bind to HSA. Part B of Table 4 reports the calculated logkb values of the set of selected polar and/or rigid ligands given in Table 2, and Part C of Table 4 reports the calculated logkb to a carbon nanotube of the set of nonpolar and/or flexible ligands given in Table 3.
Each of the HSA ligands, except for ibuprofen, is calculated to bind less strongly to a carbon nanotube than is correspondingly both observed and estimated to HSA, see Table 2. Ibuprofen is estimated, using eq.2, to bind more strongly to a carbon nanotube than is observed for its binding to HSA, and about the same as its estimated logkb to HSA based upon eq.2. However, ibuprofen is the major outlier in the set of ligands whose estimated binding to HSA is known, see Table 1, Part B. Moreover, ibuprofen is estimated to bind much more strongly to HSA than observed. Thus, caution is warranted in accepting the absolute estimated high affinity of ibuprofen to the carbon nanotube.
The average observed binding of the eight HSA ligands to HSA is 4.47 with a range of [1.98 to 5.88]. Correspondingly, the average estimated, using eq.2, binding of the eight HSA ligands to a carbon nanotube is 2.99 with a range of [0.25 to 5.90]. Keeping in mind that the sample set of eight ligands is small, these averages and ranges in binding would suggest;
The preferred binding mode of propofol to a carbon nanotube, based upon the docking procedure employed and the MM-PBSA approach, is shown in Fig. 4b. Figure 4a shows the initial MDS propofol-nanotube docking structure, and in both parts of Fig. 4 the water molecules have been suppressed in the visualization. propofol exhibits little geometric specificity in preferred binding modes to the nanotube. That is, propofol is equally bound in any of a large number of different binding geometries. This situation arises seemingly because both propofol and the carbon nanotube are rigid and there is only one polar hydroxyl group on the otherwise nonpolar and ellipsoidal propofol molecule. Hence, there is little geometric specificity regarding the binding of propofol to the carbon nanotube. The relatively high binding affinity [log kb = 4.57] of propofol to the nanotube among the eight HSA ligands arises from the rigidity of both the nanotube and propofol. This rigidity results in a small loss in binding free energy due to an entropy change upon binding. The entropy loss, ΔS, including desolvation entropy, of propofol binding to the carbon nanotube is given in Table 5. It is seen that this is the smallest entropy [and, therefore, free energy] change of the three exemplary compounds given in Table 5.
The average estimated binding of the nine rigid and/or polar ligands, using eq.2, to a carbon nanotube is 4.23 with a range in binding affinity of [ 1.88 to 6.95]. This is a much higher average binding affinity to the nanotube than determined for the eight HSA ligands to the nanotube [2.99], and not too much different from the average binding of the HSA ligands to HSA [4.47]. The range in estimated log kb of over 5 orders of magnitude suggests a marked specificity in binding. This is intuitively surprising as one would have expected all of these compounds to be relatively poor binders based upon their rigidity and/or polarity. Dabigatran is actually predicted to bind at the low-end of the range characteristic of drugs [10-7]35.
Some of these polar and/or rigid molecules, including dabigatran and montelukast, the two compounds predicted to best bind to the carbon nanotube, have sufficient flexibility to snugly fit their nonpolar ‘sides’ onto the nanotube surface. This conformational flexibility is shown by argatroban, an average binder in this class of ligands to the carbon nanotube, in Fig. 5. Part A of Fig. 5 shows argatroban in its preferred solution conformation prior to performing an MDS docking simulation to the carbon nanotube. Part B of Fig. 5 shows argatroban in its preferred bound form to the carbon nanotube. Clearly argatroban has undergone a major conformational change upon binding to the carbon nanotube. In solution argatroban adopts a balled-up, oil-drop conformation to minimize exposure of nonpolar groups to solvent, and maximize polar group interactions with the aqueous media. In the bound state argatroban adopts an open extended conformation in which one ‘side’ of its surface is largely nonpolar and binds to the carbon nanortube, and the other ‘side’ is more polar and exposed to the aqueous media. This conformational change corresponds to only a moderately larger change in ligand binding entropy as compared to propofol, see Table 5, even though the unbound and bound geometries are very different as is evident in Fig. 5.
The average estimated binding of the six flexible and/or nonpolar ligands, using eq.2, to a carbon nanotube is 5.82 with a range of [4.27 to 7.14], see Table 4, Part C. This is a much higher average binding affinity to the nanotube than is observed for the eight HSA ligands to HSA [4.47]. The range in estimated log kb is a bit less than 3 orders of magnitude. Thus, as is intuitively suggested, flexible and/or nonpolar ligands readily bind with considerable strength to a carbon nanotube. Both DMPG and PGI2 are predicted to bind at the low-end of the range characteristic of drugs [10-7]35.
Figure 6 illustrates the lowest free-energy binding mode of PGI2 to a carbon nanotube with, in this case, the waters of the MDS shown. The suggestion that these types of ligands might bind strongly to a carbon nanotube, owing to their flexibe nonpolar chains wrapping around the nanotube, is born out by inspection of Fig.6. It is readily seen, especially in the top view shown in Part B of Fig. 6, how the PGI2 chain partially coils about the surface of the nanotube in what is marginally the lowest free energy state among many about equally preferred, and near geometrically equivalent, conformations and alignments. And it is also clear, particularly from the side view in Part A of Fig.6, that the polar groups of PGI2 are exposed to water in the bound state. PGI2 loses a considerable amount of conformational entropy as compared to both propofol and argatroban, see Table 5, upon binding to the carbon nanotube, but the corresponding loss in free energy is more than compensated by the gain in binding enthalpy.
An inspection of Fig. 6 reveals that there are few waters in direct contact with the carbon nanotube, whereas the aliphatic chains of PGI2 are directly interfacing with the surface of the nanotube. The rigid nanotube acts as an entropy reservoir to the water molecules leading to a cage-like structure of waters about the nanotube. Water molecules can also migrate into the interior of the nanotube as can be seen in Fig.6, Part B.
The principal finding from this study is that several of the drugs and biologically active compounds investigated have estimated binding affinities to carbon nanotubes that fall into the upper range, and beyond, observed for the binding of drugs to HSA. Moreover, a few of the ligands considered in this study are predicted to bind to the carbon nanotube in the 10 -7 range which is often considered a lower limit of affinity when developing drugs35. Thus, the presence of nanotubes in the body could alter the normal pharmacodynamic profiles of drugs and biologically active compounds.
There are many drugs and small biological molecules that can and should be considered in a study of this type. For example, steroids and related endogenous compounds would be good candidates because of their essential biochemical roles. However, such types of structures were not considered in this initial study because the planar geometry and rigidity of the steroid core would not be expected to interact very significantly with the rigid, cylindrical structure of a carbon nanotube. However, they are many prime candidates for future nanomaterial computational-based binding studies. Also, there is the consideration that the entire bound drug-carbon nanotube complex could bind selectively to a receptor site. However, the large size, high rigidity and geometric shape of a carbon nanotube make it highly unlikely that a bound drug-carbon nanotube complex could perform as a novel ligand to a receptor with any reasonable affinity potency.
It was not possible to identify any experimental data for the binding of any molecule to a carbon nanotube. Discussions with some groups involved in making such binding measurements indicated that attempts to make such affinity measurements had proven very difficult and unreliable owing to the high hydrophobicity of the carbon nanotube. Moreover, it needs to be remembered that a primary reason for doing computational chemistry is to make computational estimations when measurements cannot be made. Thus, this work is putting forth binding predictions in the absence of being able to supply confirmatory experimental measurements.
Unsuccessful attempts were made to develop classic QSAR models for the binding of the ligands of Table 1 — Part A to HSA. Classic QSAR descriptors and modeling cannot seemingly capture the subtle 3D-geometric features, especially for low to medium level binding affinity events, needed to build a meaningful QSAR model. Moreover, it is our position that the simplicity of the 3D-structure of the carbon nanotube actually requires an even more detailed 3D modeling approach, like MM-PBSA, than is needed for a receptor like HSA , in order to capture information on the sub-angstrom level to permit accurate differentiation in the ligand-receptor binding interactions involving a carbon nanotube.
However, once a validated high-level method like MM-PBSA has been used to predict the logkb values for a set of [preferably analog] ligands to a carbon nanotube, the resulting data set can be considered a training set for use in alternate modeling approaches including QSAR analyses. In principle, classic QSAR models might be constructed which could serve as rapid virtual high-throughput screens for evaluating the affinity propensity of members of libraries of compounds to carbon nanotubes.
It is dangerous to speculate why each estimated logkb value for each HSA ligand is larger than its corresponding measured value, see Table 1 and Fig. 2. However, it is possible that the HSA-ligand binding simulations are estimating greater changes in solvation free energies in the binding processes than actually occur. This, in turn, may be attributed to too short a simulation time and/or not considering long-range solvation reorganization behavior. Along these lines of reasoning, there is no apparent reason why ibuprofen binding to HSA should be an outlier more so than any of the other ligands studied. It is possible that overestimation of the changes in desolvation free energy in the simulations may be most attenuated for the HSA-ibuprofen system.
PGI2 is a metabolite of arachidonic acid, and, as such, is a naturally occurring prostaglandin that has potent vasodilatory activity and inhibitory activity of platelet aggregation31. PGI2 has been developed as a drug known as Flolan36.
The in vitro half-life of PGI2 in human blood at 37°C and pH 7.4 is approximately 6 minutes36. Hence, the in vivo half-life of PGI2 in humans is expected to be no greater than 6 minutes. The results of this current study suggest that PGI2 will strongly bind [in the10-7 range] to an exposed carbon nanotube. Thus, it is reasonable to suggest that nanotubes could serve as reservoirs for PGI2 molecules, and thereby effectively increase the normal half-life of this potent vasodilator and platelet aggregation inhibitor. However, it must be kept in mind that there is little definitive information as to how carbon nanotubes distribute in living systems, including humans.37 One study using mice indicates a bioaccumulation of carbon nanotubes in the body. Another study suggests that uptake of carbon nanotubes into the lung may pose an asbestos-like threat to humans.38 Very recently, strong evidence has been reported that carbon nanotubes can enter into, and migrate about, the interior of human cells.39
Overall, the findings from this study indicate that ligand binding potency to a carbon nanotube is largely controlled by the shape, distribution of polar/nonpolar groups and flexibility of the ligand. Known ligands that bind to HSA exhibit limited binding to a carbon nanotube. These HSA ligands tend to be spherical/ellipsoidal in shape, relatively rigid, small to moderate in size and quite variable in their distributions of polar/nonpolar character. The general rigidity and spherical shape of the HSA ligands seemingly most restrict their binding to a carbon nanotube. The polar and/or rigid ligands bind less strongly to the carbon nanotube, on average, than do the nonpolar and/or flexible ligands even though both classes of ligands have chain-like shapes that facilitate binding. The introduction of only a few strategically spaced single bonds in the polar and/or rigid ligands markedly increases their binding to the carbon nanotube. The ability of a ligand to adopt a low-energy conformation characterized by a nonpolar ‘side’ and a ‘polar side’ also facilitates binding to the a carbon nanotube.
Still, the results of this study cannot be meaningfully ‘transferred’ to other nanomaterials for most of the same reasons that findings from one biological ligand-receptor system are not usually representative of another biological ligand-receptor system. The geometries and chemistries involved are sufficiently different between different classes of nanomaterials so as to negate any type of significant extrapolation. Indeed, preliminary results of work we have in progress suggest that simply putting side chains on a carbon nanotube imparts highly distinct properties reflective of the type, location and distribution of the side chains relative to the core carbon nanotube structure.
This work was funded by the National Institutes of Health through the NIH Roadmap for Medical Research, Grant 1 R21 GM075775. Information on Novel Preclinical Tools for Predictive ADME-Toxicology can be found at http://grants.nih.gov/grants/guide/rfa-files/RFA-RM-04-023.html. Links to nine initiatives are found here http://nihroadmap. nih.gov/initiatives.asp. This work was also supported, in part, by The Procter & Gamble Company. Resources of the Laboratory of Molecular Modeling and Design at UNM, and at The Chem21 Group, Inc., were used in performing these studies.