91av

LIFE is complicated. Over millions of years, nature has evolved the intricate
biochemical reactions which provide the basic machinery of living organisms. In
many ways we seem to know a lot about how these biochemical machines are put
together. We know, for example, that among the key components of these machines
are specialised protein molecules, often with a metal atom at their heart, which
act as enzymes. We know, too, that tiny changes to these enzymes can make a big
difference to what they do: a different reaction begins to take place, or
perhaps the process shuts down completely. Yet when we try to build replicas of
these machines—synthetic molecules that will do the same tricks—they
hardly ever work. Clearly there is still an important gap in our knowledge of
how natural enzymes work.

It is that gap which my own research, and work by several other groups of
researchers around the world, is trying to fill. We don’t work with cells and
organisms. We don’t even handle chemicals and test tubes. All our investigations
take place inside a computer, using a mathematical trick that makes it possible
to run models of the reactions of large molecules on an ordinary desktop
computer. These models are coming up with some spectacular successes in
unlocking secrets of a variety of natural reactions. This could be the
breakthrough we have been looking for to solve problems as diverse as finding a
cure for gout, capturing solar energy, manufacturing fertiliser and designing
more effective anticancer drugs.

To see the contrast between the efficiency of enzymes and the crude “brute
force” of conventional chemistry, consider nitrogen fixing: the conversion of
nitrogen gas in the air into a form that can fertilise the soil. In nature, the
enzyme nitrogenase is responsible for fixing atmospheric nitrogen. At ordinary
outdoor temperatures and pressures, the enzyme rapidly converts nitrogen gas to
ammonia. It’s a far cry from industrial ammonia production—an expensive,
energy-intensive business that requires reactors operating at very high
pressures of more than 50 atmospheres and temperatures of up to 500 °C. The
dream is to do away with all this by building an artificial enzyme that could
turn out cheap synthetic fertilisers to enrich soils around the world.

But before this can be done, every tiny detail of the reaction that is
catalysed by nitrogenase must be defined. The site at which nitrogen reacts with
the enzyme has to be the right shape to allow only nitrogen molecules in. The
channels to this site must “deliver” the nitrogen molecule in the correct
orientation. And the way all the molecules move around and interact—the
dynamics of the reaction—must be carefully controlled.

Chemists have known for some time that the shape of the enzyme molecule is
governed by a long protein chain. At the crucial reaction site sits an atom of
the metal molybdenum. The basic structure and shape of this active site were
discovered by Doug Rees of the California Institute of Technology, Pasadena, in
1993 using X-ray crystallography, and earlier this year he refined the picture
with a similar analysis at much higher resolution. This revealed some surprises.
Inside the protein chains, the molybdenum atom lies in a cluster of iron and
sulphur atoms. This is quite different from what anyone had expected. “Nobody
could have predicted the actual molybdenum site would be so complex,” says
Richard Henderson of the John Innes Centre Nitrogen Fixation Laboratory in
Norwich.

Electron trick

With the full structure of nitrogenase revealed, the scene is set for
computer modellers to take over and highlight the shape and accessibility of the
reactive site—factors that are crucial to its success. Until recently,
attempts to model chemical reactions had to take account of every electron in
every atom in every molecule involved. This used to need supercomputers to
crunch the data for even simple systems with small numbers of atoms (see
“Gulliver among the atoms”, 91av, 3 April 1993, p 34).

But a mathematical technique called density functional theory (DFT) has come
to the rescue. DFT hacks through the jungle of fundamental equations that
represent the electrons in a conventional model, and replaces them with
something much simpler that somehow represents the molecule as a whole. The
trick it uses is to treat the electrons as though they make up a uniform
gas—a free-flowing cloud that under certain circumstances sticks atoms
together like glue. This may not be the way chemists and physicists usually
think about the way electrons form the bonds that hold a molecule together, or
dictate its overall shape, but it works extremely well just the same.

Ian Dance of the University of New South Wales in Sydney has devised a DFT
model of the workings of nitrogenase, which reveals the various steps in the
reaction between the enzyme, the nitrogen molecules and hydrogen. It can even
show the transient “transition states”—the crucial halfway houses between
reactants and products. As in many biological systems, the active site is inside
a protein pocket shaped to give only the desired reaction. The model shows how
nitrogen (N₂) enters the protein pocket and sidles up to the
molybdenum-iron-sulphur cluster, where one nitrogen atom binds to four iron
atoms. Meanwhile the other nitrogen atom is attacked by three hydrogen atoms to
form a molecule of ammonia. This falls off, leaving the second nitrogen atom to
face attack by more hydrogen. “The molecular structure gives the dance floor and
the set,” says Dance. “Computer modelling gives the choreography.”

This explains why past efforts to mimic nitrogenase, using metal compounds
that bind to nitrogen, have met with so little success. Dance’s model highlights
the chemical structure and shape of the active site which will need to be
incorporated into artificial analogues of the enzyme.

Nitrogen fixing is just one of nature’s complicated reactions that chemists
want to be able to mimic. Photosynthesis in plants, respiration in humans and
other animals and the winding and bonding of strands of DNA are as complex, and
work with the same precision and speed. DFT modelling should be able to help
with all these.

Direct attack

My group is modelling the actions of another reactive site based on
molybdenum, this time with the eventual aim of finding a way to treat gout. Gout
sufferers have painful swollen joints, as King Henry VIII of England would have
testified, caused by an excess of uric acid in the body. Uric acid is produced
by the oxidation of xanthine, a naturally occurring compound related to
caffeine. To treat gout, we need to block the molybdenum-containing enzyme
xanthine oxidase, which catalyses the oxidation reaction—but first we need
to know how it works.

Xanthine oxidase produces uric acid by inserting an oxygen atom into one of
the carbon-hydrogen bonds in the xanthine molecule. Knowing the sequence of
events is crucial to designing molecules which will inhibit uric acid
production. Chemists know that the oxygen atom comes from the active site at the
enzyme’s molybdenum centre. Till now researchers believed that the reaction
occurs when the carbon atom in xanthine is attacked directly by the oxygen
atom.

But our modelling studies show, by contrast, that the carbon atom is attached
to the molybdenum centre before the oxygen is spliced in
(see
Diagram). So to
inhibit this reaction, the carbon atom in xanthine must be prevented from
reaching the molybdenum centre. This knowledge is now being used by chemists to
help them design small drug molecules that will fit into xanthine oxidase’s
“keyhole” and block the action of the enzyme, stopping xanthine reacting and so
preventing gout.

Another reaction being given the DFT treatment is one that plays a central
role in turning light energy into chemical energy in photosynthesis. A key
player in this process is plastocyanin, a complex molecule composed of a protein
chain surrounding a central copper atom. It first gains an electron, to form the
reduced state, and then passes it on and returns to the oxidised state. Each
state has a different structure, and to transport electrons efficiently, the
molecule must flip easily between the two. X-ray crystallography of plastocyanin
showed the molecule to have an unexpected structure, which chemists thought
might be due to the protein forcing the bonds between the copper atom and the
rest of the molecule to take on unusual angles. The protein, they reasoned,
might be holding the copper atom in a strained state somewhere between its
oxidised and reduced forms. This would minimise the reorganisation needed as the
electrons jump on and off.

But DFT calculations published at the end of 1996 by Ulf Ryde and Björn
Roos of the department of theoretical chemistry at the University of Lund,
Sweden, paint a different picture. The Swedish researchers modelled the active
site around the copper centre using only unstrained geometries, and to their
surprise, the model matched plastocyanin almost exactly. “The active site strain
hypothesis is out the window,” says Harry Gray of the California Institute of
Technology in Pasadena. “We need to rethink the role of the protein.” Rather
than acting as scaffolding for a strained copper atom, it could be actively
involved in the switching between states—an added complication for
chemists trying to build molecules that are capable of photosynthesis.

Buoyed by the confidence gained by modelling such a complex natural
structure, modellers are now joining in the fight against cancer, and tackling
the tricky problem of how the metal atoms that are an integral part of certain
anticancer drugs bind to the DNA of tumour cells.

The drug cisplatin, which is widely used to treat testicular and ovarian
cancers, has a platinum atom at its centre that binds directly to nitrogen atoms
on two guanine bases, preventing the two DNA strands from replicating by either
zipping them together or introducing kinks in the chain. But some tumours are
not affected by these kinks, so new compounds are desperately needed. Designing
improved drugs will need a much better understanding of platinum-DNA
binding.

Computer modelling can help here by looking at the transition states which
cisplatin goes through in binding to DNA—the choreography involved.
Building a frame-by-frame “film” of the drug approaching a strand of DNA and
grabbing hold of the binding sites should allow chemists to design a new
generation of drugs based on cisplatin that will attack different parts of the
DNA chain.

My own research has recently been successful in a similar area: modelling the
way a cancer-imaging compound binds to tumour DNA. In conjunction with the
British health science group Amersham International, I have been analysing
compounds in which the metal technetium is surrounded by a variety of small
organic molecules. Despite them all having very similar structures, of the 12
compounds I analysed only one is highly effective at binding to tumours, a
compound called Tc-HL91. Our job was to find out why it is so
specific.

Precise shape

Compounds based on the radioactive isotope of technetium (Tc–99) are
useful for medical imaging because they emit gamma rays which pass harmlessly
through body tissue to a detector, giving a kind of “inside-out” X-ray of the
patient. If the compound binds selectively to tumour cells, then we have a
quick, noninvasive method for diagnosing and pinpointing cancer. Tc-HL91 is
particularly good at binding to and highlighting oxygen-deficient, or “hypoxic”,
tumours in this way. Such tumours tend to be resistant to radiotherapy, so a
simple test to show whether a tumour is hypoxic or not would help doctors decide
whether radiation treatment is worth pursuing. Tc-HL91 differs only slightly
from its less successful relatives—an extra carbon atom here, a smaller
carbon chain there—yet it binds to tumour cells with remarkable
selectivity. But before I started my computer modelling work nobody knew the
precise shape taken up by the small molecules attached to the technetium atom in
Tc-HL91.

We began by using DFT to calculate the shapes of other technetium compounds
which had already been examined by X-ray crystallography. The structures
predicted by our models matched the real structures, which gave us confidence to
try and predict the structure of Tc-HL91. Using the same DFT calculations, we
produced a model structure for Tc-HL91 which highlighted some important
differences compared to its relatives: the central active binding site around
the technetium atom in Tc-HL91 has a unique shape and chemistry. Our modelling
was confirmed in July last year by chemists at Amersham International, who
successfully crystallised Tc-HL91 and examined it with X-ray
crystallography.

We are now using DFT modelling to investigate the compound’s behaviour in
water—a more accurate picture of what it will look like when circulating
through the body. Unlike its related compounds, Tc-HL91 undergoes a chemical
change in water, which is revealed by a colour change: while dissolved in
organic solvents it is green, but in water it turns colourless, perhaps because
a hydrogen atom from a water molecule jumps onto one of the oxygen atoms bound
to the technetium reactive site. Our study sheds light on why this reaction
takes place in Tc-HL91 but not in any of the related technetium compounds. If
this change in structure helps explain why Tc-HL91 targets tumours so
effectively—and we think it will—we will have a handle on the design
of a variety of selective tumour-binding compounds based on other metals.

With increasingly powerful machines, this is an exciting time to be doing
computer modelling. Like other DFT modelling groups around the world, we expect
to uncover more and more of the secrets of nature’s success. Our results should
provide invaluable pointers for chemists aiming to design powerful new drugs and
molecules that can, at last, faithfully mimic the actions of their natural
counterparts.