Combinatorial chemistry promises better drugs, new superconductors and an artificial nose, all with a one-thousand-fold increase in productivity. Who could resist?
What made it all possible is combinatorial chemistry. Designing chemicals that change color when other chemicals attach themselves (the basis of the artificial nose) is difficult, if not impossible, to do rationally. The only alternative is brute-force testing of thousands or millions of possible chemicals until you find the right one.
Before you can test this multitude of chemicals you need to make them. Unfortunately, traditional chemistry is slow. The construction of each new chemical is like an adventure in haute cuisine -- tweak the amount of this ingredient, adjust the temperature, add a new spice. Each chemist at a drug company used to be happy to produce a handful of new chemicals each year.
No more. The modern chemist, at Illumina and drug companies worldwide, churns out thousands of potential new nose-sensors or drugs every year, sometimes as many as a million. The key technology is combinatorial chemistry.
Combinatorial chemistry essentially means reacting a set of starting chemicals in all possible combinations. For chemists used to pure, discrete compounds it sounded messy, and initially most of them stayed away. But gradually researchers came up with ways to deal with the complicated mixtures and unmanageable numbers, and chemists forgot their misgivings.
It started with peptides. Everyone knows that peptides -- short stretches of the protein molecules that do all the work in the cell -- are lousy drugs. They get eaten up in the stomach and the survivors rarely get into cells. And yet somehow Mario Geysens strategy for making peptides has become the accepted way of making drugs in the 1990s.
Working at Melbourne University in Australia, Geysen thought he could make lots of peptides in almost the same amount of time that someone else could make one. Each peptide was made on the end of a pin, and for each step in the reaction the pin was dipped into a dish with a new chemical. The trick was to line up hundreds of pins in an array, such that they aligned with hundreds of tiny dishes. Procedures like washing and drying could now be done on all the reactions at once. For some steps -- adding an amino acid as a new link to the peptide chain, for example -- a different chemical had to be added to each dish. But time was still saved, as the chemical reactions could run in parallel. The result was termed a library of peptides.
The next year, in 1985, Richard Houghten of the Scripps Research Institute (La Jolla, Calif.) came up with a similar scheme, but this time each peptide was made attached to bits of plastic enclosed in a mesh bag, dubbed a tea bag. Houghten went on to form the combinatorial chemistry company Houghten Pharmaceuticals, which is now called Trega Biosciences, Inc. (San Diego, Calif.).
The third and most enduring format came in 1991, when Kit Lam of the Arizona Cancer Center (Tucson, Ariz.) made a library of over one million different peptides. "That really showed not only to the peptide chemist but to the medicinal chemist that this could be useful," says Lam.
In the first set of reactions the first link is added to the peptide chain: either A, B, or C. The beads from these three separate reactions are mixed, washed, and re-divided. Each new sample now contains a mixture of A, B and C beads. One sample is reacted with D, another with E, and another with F. The mixing ensures that all possible combinations (AD, AE, AF, BD.....) come out at the end. We end up with nine (3x3) final products after six (3+3) reactions.
"You can instantly see that a parallel method will give you pure compounds," says Czarnik. "But [with split/pool] I can make 10,000 compounds in a four-step reaction sequence with just ten reaction vessels."
Thats 10x10x10x10 compounds with forty (10+10+10+10) reactions. You get down to a total of ten beakers only if you are prepared to wash up in between each of the four steps.
Geysen, Houghten and Lam had started with peptides because Bruce Merrifield of Rockefeller University, New York, had already worked out how to make peptides anchored to a solid material like a plastic bead. Before Merrifields work, which earned him the 1984 Nobel prize in chemistry, there was a painstaking purification of the growing peptide chain after each amino acid was added. Now the chemists could add huge amounts of chemicals, driving reactions to completion in a short time. Excess chemicals can then be washed away, leaving only the beads and therefore the attached peptide product. Others were quick to take advantage of this system to make short stretches of DNA called oligonucleotides.
But oligonucleotides and peptides are equally lousy as drugs, and for similar reasons. What was needed was a way to make similarly huge numbers of chemicals, but chemicals that looked like drugs.
The company researchers who tackled this -- at Chiron Corporation, Emeryville, Calif., and Affymax Research Institute, Santa Clara, Calif. -- initially stayed with the links-in-a-chain approach. Chirons peptoids, made in 1992, looked like peptides but the linkages were altered to make them more stable.
The next evolutionary step came the following year. "People realized that one could think more broadly and make unnatural structures; that one didnt have to make polymeric structures at all," says Mark Gallop, director of combinatorial chemistry at Affymax. "You could make chemicals like the organic heterocycles that pharmaceutical companies had historically turned into drugs." Jonathan Ellman and colleagues at the University of California, Berkeley, and Sheila DeWitt and the bioorganic group at Parke Davis made the first libraries of these circular chemicals -- molecules in the same family as the sedative valium.
They were not alone. "They just happened to be the first ones to publish, but there were lots of people doing similar things at the same time," says Lam. "The field just exploded around that time."
A circular chemical is a perfect starting point for a library. Hundreds of different side-branches can be added to the various chemical groups poking out of the circle. This is now true combinatorial chemistry, where one of a large number of possible building blocks is added to each of several variable sites on a molecule.
If you make your chemicals in mixtures you save time, but you are left with a problem. What chemical is on what bead?
The simplest way to find out is to look, directly, with tools that detect the exact molecular weight (using a mass spectrometer) or the electronic environment of particular atoms (using a nuclear magnetic resonance (NMR) machine). There is just enough of a chemical on a standard bead for identification using these methods. Some methods work with the chemical attached to the bead; for others the linker that holds the chemical to the bead must be sheared with light or acid or base.
If a library has been made for one specific purpose this approach can work. "The beauty of combinatorial chemistry is that you dont really care about identifying everything, you just have to identify the active chemicals," says Ron Zuckermann, director of bioorganic chemistry at Chiron. Having determined the make-up of the few chemicals that turn off his protein of interest, Zuckermann will happily re-make these chemicals in bulk for further study.
But if you have a large library that will be used in many different projects, you may want to know what the chemical on every single bead looks like, and doing 100,000 NMR runs will take far too much time and money. This is where tags come in.
The first tags were peptides and oligonucleotides. Each time you added a certain nucleotide building block, say adenine, you could add a particular amino acid, say leucine. To identify the final oligonucleotide, the peptide is chewn off, one amino acid at a time. The order of amino acids gives the order of nucleotides, leaving the oligonucleotide intact for testing as a drug.
Once the chemistries got more complicated, the tags got more sophisticated. Affymax and Pharmacopeia, Inc. (Princeton, New Jersey) made tags that were more resistant to degradation by harsh chemicals, and more easily detected. But the complexity award goes to IRORI (La Jolla, Calif.), who embedded tiny radio-transponders in containers reminiscent of Houghtens tea bags. A computer assigns each transponder a code and a corresponding chemical. After each chemical step, the IRORI machine uses radiofrequency communication to read the code (at the rate of ~1000/hour). This information is used to redirect the container to the next appropriate reaction.
"I remember when I first saw it I thought it was cute, but that it looked like a two-dollar solution to a two-cent problem," says Czarnik, who until recently was the vice-president for chemistry at IRORI. "But it is actually a very elegant solution to a problem that there arent obvious alternatives to."
IRORIs next-generation tag is a 2D bar code: a patchwork of light and dark squares which is laser-etched on a ceramic surface, and then read by a camera. And a group at Pfizer, Inc. (Groton, Conn.) led by Eric Roskamp has developed a simple numbering system, which uses a heated printing press, graphite ink and optical character recognition.For chemists who spurn beads and stick to the more traditional ways of solution chemistry, making and decoding mixtures is still a possibility. These decoding methods are collectively called deconvolution. One possibility is shown to the left, for a chemical made of two parts. In each mixture, one of the two positions is held constant, while the other is varied. The results from the first set of mixtures indicates the best chemical for part A, and the second set of mixtures gives the best part B. The chemists would then make the single chemical AB and test its properties.
Of course, if you do parallel synthesis you have none of these worries. As long as you keep track of each compound during all the chemical transformations you can deduce its final structure.
If making libraries is so easy, perhaps this is a case of the bigger the better. The first hitch in that argument comes with planning how to make the library, which involves optimizing a prototype of each class of chemical reaction. With the IRORI system Czarnik says: "Once the route to make a library exists, we anticipate that 10,000 compounds will take about two months with one to two people. But developing the route can take six months."
After chemicals are made they must be tested, to see if they jam the protein they are meant to jam. Testing also takes time and money.
"I think a lot of people are sitting back and wondering if making millions and millions of compounds is really worth it," says Vince Anido, CEO of CombiChem, Inc. (San Diego, Calif.). "The cost is enormous."
"There is as much range of opinion [on library size] as there are people," says Czarnik. "There are devout believers in spending as much time thinking about what you should make as possible, and then never making more than a few hundred compounds. Then there are those who say I would like to know exactly what to make, but I would also like to know the face of God, so perhaps we should make lots of compounds."
One way of whittling down library size is to bring in the practitioners of rational drug design -- scientists who use the shape (or structure) of the target to deduce which one chemical key will fit the protein lock. The reality is often not so simple.
"The idea of rational drug design is that based on a structure you can predict one perfect drug," says Zuckermann. "That isnt possible now, and that day may never come. But [rational drug design] can certainly spit out a family of 10,000 compounds that it would be good to make." The computer modeling indicates the rough shape for the key, and custom programs generate a list of chemicals that might fit that description.
Smaller libraries are also more suited to the later stages of drug discovery. The first chemical that shows promise in drug testing rarely ends up being the final drug. Instead there is a period of optimization, in which hundreds of variants of the original lead chemical are made and tested. Combinatorial chemistry can be used to make more variants more rapidly. Usually one or more of the variants is an improvement on the original lead chemical.
For combinatorial chemists who still yearn for beefy libraries, the holy grail is a library that contains a lead for every possible protein target. The first step in reaching that holy grail is to minimize duplication, by making the library as structurally diverse as possible. "You can make sure you get diversity of shape, size, and hydrophobicity," says Zuckermann. "Software for that is advanced pretty far, and we have a group at Chiron that does that. Its useful for avoiding structural similarity. But trying to find parameters that are really meaningful experimentally is very difficult." In the vast universe of chemical shapes, how can you ever define similarity?
CombiChems Anido has responded to this dilemma by making everything, at least in a computer. CombiChem researchers have programmed their computers with every chemical building block they can come up with, and the known laws of chemical reactions, and let the computer come up with its own trillion-member chemical library. "Its like taking the alphabet and software that encodes the rules of English," says Anido. "Using those rules, you make all possible words."
Anido initially does a standard screen for lead chemicals, however poor, in a diverse 10,000-member library. Computers tell him what common hole, protrusion, or corner in all these chemicals might have made them all stick to the same target protein. The computer then searches the trillion-member library for other chemicals with that common shape. Anidos team makes these new chemicals, and the whole process is repeated. Four to six repetitions, each taking approximately three months, usually yields a chemical that is good enough to hand over to a large pharmaceutical company for animal and human testing.
And that is where things slow down. "Combinatorial chemistry has sped up lead discovery dramatically -- finding an active molecule in an assay," says Zuckermann. "But to get a drug on the market you need to do toxicology, animal and human testing, and these take just as long."
It is a long way from peptides to artificial noses for landmines, but thanks to the speed and numbers of combinatorial chemistry it has taken just over a decade. And according to Kim Janda, a combinatorial chemist at Scripps Research Clinic, La Jolla, Calif., the number of applications will keep growing. "Just about anywhere you need better things faster," he says, "you can use combinatorial chemistry."