Myths about disrupting protein-protein interactions with small molecules

The first era of medicinal chemistry was finding small molecules to target proteins. It's still going strong and will continue to do so. The second era that promises a treasure trove at least in principle is finding small molecules for disrupting protein-protein (PP) interactions. So many important processes in our body are regulated by these crucial interactions that finding small molecules to modulate them could promise a bonanza of new therapies.

But disrupting PP interactions is fundamentally different from designing a small molecule to bind to a (mostly) well-defined active site on a protein. PP interaction surfaces are rather flat and expansive and depend on subtle interactions between amino acids that add up to provide substantial binding affinity. Designing a small molecule to disrupt such interactions is somewhat like disrupting the sliding of one door hinge against another by lodging a grain of sand between the two. However, the picture has been simplified somewhat in recent years by the identification of "hotspots"- key amino acid epitopes that provide the bulk of binding interaction. If one could design a small molecule that could provide this major contribution to the free energy of binding, one could have an effective drug that could target PP interactions. After all, a grain of sand can indeed inhibit hinge sliding if it's placed in the right position.

One of the big players in investigating small molecule PP interaction agents has been Jim Wells, formerly at Genentech and Sunesis and now at UCSF. He has a nice Nature review that traces recent successful examples of small molecule PP interaction antagonists. Wells considers six or seven successful stories involving important proteins playing key roles in health and disease. For example, disrupting the inhibition of the pro-apoptotic BAD and BAK by the anti-apoptotic Bcl-2 and Bcl-xl, disrupting the binding of interleukin IL-2 to its receptor, and disrupting the binding of HDM2 to the tumour suppressor p53.

But more importantly, Wells then address some widely held myths about PP interactions that seem to drive pessimism in the field. They are worth taking a look at:

Myth 1: It's very difficult to find a small molecule that would lodge between the rather disordered flat surfaces of two proteins.

While this is true, in practice as is exemplified by almost all the examples, the protein surface does not remain flat when a molecule binds to it. Loops and side chain adapt and form shallow pockets and dents to which the molecule can bind. The fallacy in believing the myth is to assume that small-molecule protein surface binding is a rigid body interaction. It's not, and there's a strong element of induced fit in the process. This is probably the single most important thing to keep in mind, that small molecules will form their own small pockets and bind well to initially flat protein surfaces.

Myth 2: Small molecules that disrupt PP interactions are too large to be drugs

This is part of the partly substantiated myth that large molecules usually don't become drugs, because of many factors including ROF problems. But many molecules disrupting PP interactions cited in the review are about 500-700 Da, perhaps a little large but not intractable as drugs. The authors also calculated the ligand efficiency which is the free energy of binding per non-hydrogen (heavy) atom for the ligands, and found that it was comparable to that of kinase or protease inhibitors. Clearly with sound med chem efforts, it won't be too difficult to have such drugs. Interestingly, since the molecules occupy about half the binding site that the parts of the native protein partner do, their ligand efficiency is almost twice.

Myth 3: Small molecules disrupting PP interactions won't be potent

Just not true. Almost all the molecules found in the cited examples had mid to low nanomolar Ki values, almost as good as the binding constants for the partner proteins.

Myth 4: Screening would not help find novel small molecule PP modulators

Again, not true. Most of the cited molecules were found by HTS. Interestingly, there may be even more wealth in HTS than we have now. As the authors explain, HTS hits are fundamentally going to be limited by chemotypes present in the libarries. After all we can do only as well as chemical space in existing libraries. Existing libraries contain many molecules targeted against kinases, GPCRs and other well-known targets for which common privileged structures have been deduced. But because of the diversity of PP interactions, it is improbable that common scaffolds will exist for disrupting them, and libraries will have to contain novel scaffolds to get better hits. Given this fact, it's impressive and encouraging that the existing libraries could come up with such potent structures for disrupting a few PP interactions. Remarkably, even with such different scaffolds, the ligand efficiency remains more or less constant for the cases studied.

Clearly the field of small-molecule-PP interactions is alive and kicking. In the next few years, hopefully computational, screening and NMR approaches will converge to discover novel agents for these important processes.

Wells, J.A., McClendon, C.L. (2007). Reaching for high-hanging fruit in drug discovery at protein-protein interfaces. Nature, 450(7172), 1001-1009. DOI: 10.1038/nature06526

Elegant switches in an E. Coli ion channel

Molecular dynamics simulations comprise one of the most important tools in the armamentarium of chemists and biologists. Initially a curiosity for theoretical scientists, MD is now an explanatory and predictive tool in chemistry, biology, materials science, engineering and even weather prediction. In the field of biology, some masters such as Martin Karplus have honed this tool to the status of an art. While great leaps have been made in the context of methods, hardware, software and applications in this field, much remains to be still done. One of the reasons is that even now, running microsecond MD is computationally quite expensive. But many important biological events involving biomacromolecules take place on this time scale, thus making the achievement of efficient microsecond MD simulations important.

Writing in Science, scientists from D. E. Shaw company, Columbia and the Hebrew University of Jerusalem have a lovely paper documenting the application of the new and innovative MD program Desmond to the dynamics of a bacterial ion channel that transports Na+ ions using the electromotive force generated by proton transport. Desmond is supposed to enable efficient microsecond MD. It is going to be interfaced with the Maestro interface developed by Schrodinger and is due to be released this year I believe. Currently the fastest MD program on a single processor is GROMACS. While head to head comparisons of GROMACS and Desmond have not been reported to my knowledge, Desmond is supposed to be very fast on multiple processors, a facility that many can now afford to have.

In the Science paper, the researchers apply Desmond to understand the transport mechanism of the Na+/H+ antiport ion channel in E. Coli. This protein is crucial for E. Coli to survive harsh conditions of pH, alkalinity, and ionic lithium environments. The authors basically focus on the protonation state of certain key aspartates and find something pretty interesting- two crucial aspartates essentially act as switches that decide whether Na+ ions would be transported to the cytoplasm or to the periplasm. Using many long MD simulations involving different protonation states, the authors discovered that one of the carboxylates always has to be protonated. This acts like a "master aspartate" switch. Once this switch's state is set, it's the state of the other switch that decides the direction of transport- protonated leads to expulsion of the Na+ into the periplasm, while deprotonated leads to expulsion into the cytoplasm.

The observation reminded me of a high-school "staircase lighting" electricity experiment. A master switch had to be always on for the assembly to work. The On/Off state of another switch would then govern whether current flowed or not.

Using this discovery as the basis for exploring further conformational changes related to it, the authors come up with an elegant stepwise mechanism for the transport of Na+ and H+ ions that accounts for the observed stoichiometry of one Na+ ion for every two H+ transported. Using free energy perturbation binding affinity calculations, the authors also rationalize the channel's observed selectivity for Na+ over K+, and slightly for Li+ over Na+.

There is also a a very intriguing explanation for the pH sensitivity of this ion channel. The crystal structure of the channel is solved at pH 4, and it is inactive at this pH. How does the channel get activated at higher pH? To explore this, the authors do something simple but quite clever. They first identify all the key aspartates lining the channel and determine their pKa values. Perhaps not surprisingly, the pKa values of these are abnormally high- not an uncommon observation for amino acids in the unusual environments in protein interiors. They then selectively deprotonate one aspartate keeping all others protonated and do MD on the resulting structures. If there is a key "pH sensing" aspartate, its protonation state will likely govern a conformational change from inactive-active. Indeed, one aspartate, D133, turns out to modulate a conformational change involving two helices when it is deprotonated. Crucially, this results in the "master aspartate" noted above to move away from the Na+ entry/exit pathways. Thus it can no longer bind the ion, resulting in an inactive channel. Mutagenesis studies also support the observations.

A neat conclusion from a beautiful set of experiments. A fascinating example of how nature essentially and surprisingly uses high-school chemistry to modulate movements in complex proteins. And a highly successful and inspiring example of how efficient, long MD simulations can shed light on these crucial processes.

Arkin, I.T., Xu, H., Jensen, M.O., Arbely, E., Bennett, E.R., Bowers, K.J., Chow, E., Dror, R.O., Eastwood, M.P., Flitman-Tene, R., Gregersen, B.A., Klepeis, J.L., Kolossvary, I., Shan, Y., Shaw, D.E. (2007). Mechanism of Na+/H+ Antiporting. Science, 317(5839), 799-803. DOI: 10.1126/science.1142824

Computational modeling of GPCRs: not too bad

GPCRs constitute one of the most important family of proteins in our body, both for their innate importance in signal transduction and neurotransmission, and as important targets for drugs. Many of the important drugs on the market today target GPCRs. And yet there is an unusual gap between knowledge and application when it comes to this important family. That's because only two crystal structures of GPCRs are known. And one of them was derived last year, so there's been a real dearth of structural information about GPCRs for a long time.

We do know something about many GPCRs, however. We know that they are 7-TM receptor-spanning proteins. And the two structures we do know about shed valuable insight on GPCR function. One is rhodopsin which has been around for a while. Then there was big news last year about the second important GPCR whose structure was determined- the ß-2 adrenergic receptor.

Given the paucity of structural information and the availability of two structures, a logical question is whether computational modeling can teach us something new about GPCRs whose structure is unknown. To this end, Stefano Costanzi at the NIH did a nice set of experiments which he published in J. Med. Chem. He attempted to build a homology model of the adrenergic receptor based on the sequence and structure of rhodopsin. Since we now have a crystal structure of the adrenergic GPCR, we have something concrete to compare modeled structures and ligand orientations to.

Costanzi was particularly interested in knowing how a small molecule-carazolol- binds to the modeled GPCR. This is important both from a structural and functional drug-discovery point of view. His results indicate that we can do pretty well. In essence, he built two models of the receptor, one of them de novo. While the models were similar to rhodopsin in the conserved regions, the important differences were with respect to a loop that flaps on top of the protein. In one model the loop was buried inside the binding pocket, and in the other one it was open. Docking of carazolol into the buriled-loop model using the Glide program from Schrodinger gave a binding pose in which the ligand was, not surprisingly, buried deeper into the cavity compared to the crystal structure. This was naturally the effect of the loop blocking part of the pocket. The other model in which the loop was not buried gave much better results. Curiously, the ligand was buried a little deep in the pocket even in this model, even though it was much less buried compared to the previous one. It still misaligned considerably with the experimental pose. Inspection revealed that there was a Phe in the pocket which was anti in the model but +gauche in the crystal structure. Since the corresponding residue in rhodopsin was Ala, there was no way this unusual conformation could have been predicted ab initio. Fixing the conformation of this residue to +gauche suddenly gave excellent alignment with the ligand orientation in the crystal structure.

An instructive piece of work that shows that homology modeling and docking of ligands into GPCRs of unknown structure can be fruitful. However, it also indicates caveats like the Phe conformation which are hard to account for de novo. However, since structures of members in this important family of proteins are unavailable anyway, even some predictive ability might be welcome in this area.

Costanzi, S. (2008). On the Applicability of GPCR Homology Models to Computer-Aided Drug Discovery: A Comparison between In Silico and Crystal Structures of the ß2-Adrenergic Receptor. Journal of Medicinal Chemistry DOI: 10.1021/jm800044k

Hexacyclinol as a test case: ab initio C13 chemical shift prediction

Anybody heard of this natural product called hexacyclinol and how doubts were raised about its synthesis and structure? Kidding obviously. I am going to assume that any organic or related chemist who has not heard of hexacyclinol has not heard of Robert Burns Woodward by default.

Well, in any case, recall that the high point of that deb(acle)ate was Scott Rychnovsky's demonstration by using quantum chemical prediction of C13 chemical shifts that a structure quite different from hexacyclinol fit the C13 NMR data much better compared to JJLC's structure. To do this Rychnovsky used DFT methods and the mpw1pw91 functional which was tried, tested and proven to be a reliable tool for C13 chemical shift prediction by Bifulco and others. (excellent general review here which deals with calculation of both shifts and 2 and 3 bond homo and heteronuclear coupling constants)

The point of value for the organic chemist from the whole exercise was the fact that C13 chemical shift prediction could not just be used to distinguish regioisomers whose identity might be ambiguous but, based on Rychnovsky's analysis, also can be used to correctly assign misassigned C13 peaks. To me this is the greatest benefit of the analysis for the practicing organic chemist.

Henry Rzepa and Christopher Braddock at Imperial College in London have now demonstrated the application of this increasingly valuable method to the correct assignment of some interesting halogenated natural products called obtusallenes. In this case there was ambiguity about the positions of a chlorine and a bromine. The proton chemical shifts were very similar and could not be used to assign the positions. Rzepa and Braddock used the mpw1pw91 functional not just for the chemical shift calculation but also for the optimization. Fast computational power has made this possible now. The bottom line is that average C13 shift deviations are much more for the incorrect regioisomer. Using the method, the authors also re-assigned two ambiguous peaks. In addition, they determine that the 6-31G (d,p) basis set gives some errors for certain functional groups while using the aug-cc-pVDZ basis set (all that's left to say is "warp speed" now) basis set eliminates these errors.

A short, neat demonstration of the increasing value of quantum chemical NMR prediction methods for the practical organic chemist.

Braddock, D.C., Rzepa, H.S. (2008). Structural Reassignment of Obtusallenes V, VI, and VII by GIAO-Based Density Functional Prediction. Journal of Natural Products, 71(4), 728-730. DOI: 10.1021/np0705918