¿ìè¶ÌÊÓƵ

Earlier this year, five research laboratories reported results of related
but independently performed projects that could be the harbingers of an
entirely new sector of the biotechnology industry. Indeed, so great is the
commercial potential of the new approach that, according to some observers,
it could eventually come to dwarf in importance the currently established
recombinant DNA, or genetic engineering, methods for producing biomolecules.
The new approach, known as applied molecular evolution, may be able to generate
an essentially infinate range of entirely novel biomolecules, from vaccines
to industrial catalysts.

The titles of the five papers are about as arcane as anything you could
expect to find in the molecular biology literature, and give no hint of
their real import. Nevertheless, each of the papers shows how the power
of evolution – selection working on random variation – can be exploited
in the laboratory to yield molecules of a desired type.

The recent results, driven as they were by experimental expediency,
focused more on the technology than on the products. But in the long run,
applied molecular evolution will offer a virtually unlimited source of tailor-made
drugs, vaccines, and enzymes. Anything that peptides, RNA and DNA molecules
are theoretically capable of doing, applied molecule evolution will provide
the means of making them do it.

‘There’s an explosion of interest,’ says Larry Gold, of the University
of Colorado, an author of one of the five papers. ‘We thought our work was
unique, but it turned out that we were all doing these things independently.
It’s a glorious moment. We’ve turned half our lab onto this venture.’

It all has to do with shape. More than anything else, shape governs
the interaction of large molecules, whether it concerns antibodies recognising
antigens, enzymes bringing reactants together as a reaction complex, or
hormones slotting into receptors. So in principle, anything that mimics
the shape of a biologically important molecule may be able to replace that
molecule in its function. The trick is to be able to produce the shape mimic
you are interested in.

Suppose, for example, you want to find a peptide that mimics a particular
hormone. If you had the receptor into which the hormone fits in nature to
hand, you might be able to use it as the selector. What you do is expose
the receptor to a huge library of randomly generated peptides. The vast
majority of these peptides will not interact with the receptor at all, but
a few will. These are the ones that just happen to have the required shape
to bind to the receptor by the classic lock-and-key mechanism: the receptor
is the lock, the peptide that fits into it is the key.

Here, then, you would have selected a small number of peptides that
can take the place of the hormone in binding to the receptor: they are shape
mimics of the hormone. But the really interesting issue here is that of
this small selection of shape mimics, some may have an amino acid sequence
similar to that of the hormone in question, while others might be quite
different. This is because there are many ways of producing essentially
the same molecular shape: different combinations of different amino acids
can give the same overall shape. Take this to its next logical step, and
it is clear that peptides may be able to mimic the shape of non-protein
molecules, such as carbohydrates or even certain lipids. This is important,
because not all hormones are proteins.

Even if, in this hypothetical example, the hormone receptor were not
available, shape mimics could still be tracked down. This would require
an extra step in the process, one that exploits the most exquisitely shaped-attuned
system of all: the immune system. Immunise an experimental animal with the
hormone in question, and you produce antibodies whose business end – the
variable region – is complementary in shape to the hormone. In other words,
the antibody’s variable region is the same shape as the receptor into which
the hormone normally fits. The anti-hormone antibody can now be used as
the selector to screen the library of candidate peptides, as before.

In this description, the initial target has been a hormone, and so the
shape mimics that are produced are candidate drugs. If the target had been
some component on the envelope of a virus, the shape mimics would be candidate
vaccines. If the target had been a reaction complex, the shape mimics would
be candidate enzymes. And so on. Whatever is the specific case, the system
gives you the components of applied molecular evolution: the random variation
in the equation is in the initial library of peptides; and the selection
is provided by the lock-and-key demands of the target or its surrogate.
The system is not limited to selection among libraries of peptides, however.
Libraries of RNA and DNA molecules may also be screened for molecules that
bind to specific targets.

Two of the recent crop of five reports were on RNA, three were on peptides,
but all were based on the principle of random variation followed by selection.
In addition, all added a further stage – amplification – that is essential
when the quantity of shape mimics isolated is so small.

For instance, Gold and his colleaque Craig Tuerk were looking for novel
RNA molecules that would bind in a specific way to an enzyme of a virus
that attacks bateria, the bacteriophage T4. The enzyme was the T4 DNA polymerase,
whose activity is essential for the virus’s replication. Under normal circumstances
a short sequence of RNA, eight nucleotides long, is able to bind to the
polymerase, thus controlling its activity. Gold and Tuerk were looking for
novel sequences that would do the same thing.

The first step was to generate a pool of RNA molecules in each of which
a loop of eight nucleotides was randomised for sequence: theoretically,
this gives a total of some 65,536 different molecules (4⁸).
This pool was then exposed to the T4 enzyme, immobilised on nitrocellulose
filters. The RNA molecules that did not bind to the enzymes simply washed
away, leaving behind those whose shape fitted the shape of the control area
of the enzyme. Two of the 65,536 sequences did bind to the enzyme: one of
them was the naturally occurring control sequence, and the other was a novel
one.

The immediate practical concern was to obtain sufficient quantities
of the sequences that bound to the enzyme, and this was done via the magic
of the polymerase chain reaction (PCR). But the immediate theoretical import
of the experiment was that here was an entirely novel RNA sequence. The
effect of the novel sequence is to inhibit the activity of the polymerase,
just as the naturally occurring molecule does in the course of balanced
control of replication. Added in large enough quantities, however, and the
novel sequence could be used to prevent T4 infection.

The same principle can be applied to other viruses: ‘We expect that,
at the least, nucleic acid ligands that inhibit replicative proteins of
epidemiologically important infections can be likewise evolved,’ note Tuerk
and Gold. Here is a potential route for producing highly specific anti-viral
drugs that target the basic molecular machinery of viruses.

While the Colorado laboratory was working on the T4 replicase binding
experiment, Jack Szostak and Andrew Ellington, at Massachusetts General
Hospital, Boston, were also putting RNA through the hoops of applied molecular
evolution. ‘Our interest had been in pre-biotic evolution,’ says Szostak.
‘We had no idea what was going on in Colorado.’ If the chemical evolution
of life first centred on RNA, as many people are coming to believe these
days, then catalytic RNA molecules must originally have come from random
sequences. About five years ago Szostak was wondering what the probability
was of something catalytic coming together at random, and he decided to
try to test it. But he got distracted, and only got down to the job seriously
last year, after Ellington joined him in the lab. ‘Besides, PCR had been
developed by then, and that made it a whole lot easier,’ says Szostak.

The experiment involved exposing random sequence RNA molecules (each
containing 100 nucleotides) to a simple dye, Cibacron Blue. The idea was
the shape of the dye molecule – the ligand – is rather like the transition
state of a reaction complex. Any RNA molecule whose shape would complement
that of the dye would therefore be a candidate as an RNA enzyme, or ribozyme.
In this experiment, the numbers become enormous. For a start, the total
number of possible sequences of a 100-nucleotide RNA strand is 10⁶⁰,
way beyond any practical reach. Ellington and Szostak in their experiment
produced a meagre 10¹³ sequences, an infinitesimal fraction
of the total possible. Nevertheless, by the system of exposure, binding,
and amplification, they were able to pick up as many as 100,000 different
sequences that bound specifically to the dye: that is, one in 10⁸ of the
origianl pool of molecules fitted the shape of the ligand and could therefore
bind specifically. They point out that, of the total ‘universe’ of 100-nucleotide
RNA sequences, a staggering 10⁵² are theoretically capable of
binding to the single ligand represented by the dye molecule.

‘Our results suggest that it may be possible to isolate novel ribozymes
from pools of random-sequence RNAs,’ say Ellington and Szostak. All you
need to have available for experiment are the transition states of useful
reaction complexes to act as selectors – and of course the ability to make
huge pools of random RNA sequences. They point out that, even with a fairly
limited search, it might be possible to recreate the primordial RNA replicase
upon which life on Earth was first built. In any case, this technique opens
up an essentially whole new world of catalysis, one that will at the very
least give a glimpse into the now-vanished pre-biotic world that was based
on RNA enzymes. But as Szostak says, ‘there are likely to be practical applications
too.’ That may prove to be an overly modest prediction.

The three recently reported experiments on peptide shape mimics follow
this same pattern. For instance, Steven Cwirla and his colleagues at Affymax
Research Institute in Palo Alto, California, screened a large library of
small peptides (each containing six amino acids) for specific binding to
a part of the natural opiate, beta-endorphin. They produced a pool of 3
x 10⁸ peptides, of which 51 turned out to bind to the target.
This is about a one in a 10⁷ hit. Using a pool of 15-residue
peptides, James Devlin and his colleagues, of Cetus Corporation, Emeryville,
California, found a one in two-million hit for binding to the protein strepavidin.
Last, Jamie Scott and George Smith, of the University of Missouri, screened
about 40 million six-residue peptides for thier ability to bind with an
antibody. Nineteen peptides did so giving a one in two-million hit rate.

Working with small peptides such as these is technically more demanding
than working with RNA or DNA. One reason is that in order to get the random
peptide pool, a random DNA library must first be generated, and its products
must then be cloned into some kind of vector, such as a phage. The pool
of vectors can then be made to transcribe the inserted DNA into peptides,
which may be displayed on the surface of the phage. It is this pool of vectors,
bearing the peptides on their surfaces, that is then exposed to the target
molecule. Once the shape mimics have been selected out in the usual way,
their amplification is much more cumbersome than simply going to the PCR
bench. It is also more difficult to generate as large a pool of candidate
peptides by this method than it is to make a large number of RNA or DNA
molecules. ‘There is tremendous scope for improving the technology here,’
observes Szostak.

Technology aside, it becomes clear that applied molecular evolution
is actually a numbers game, and one with surprising dimensions. Specifically,
how big does the pool of random molecules have to be in order to have a
reasonable chance that it contains at least one with the right shape? ‘You
might think you’d have to deal with huge numbers when you are looking for
particular shapes,’ says Stuart Kauffman of the University of Pennsylvania
and the Santa Fe Institute. ‘But it turns out not to be so. It turns out
to be very manageable.’ Kauffman is one of applied molecular evolution’s
great enthusiasts and promoters. With his colleague Marc Ballivet, Kauffman
developed methods of generating large peptide and nucleic acid pools, and
filed very broad patents in March 1985. Since then patent applications have
been flying thick and fast from various laboratories, a sure sign of a field
on the move.

As Kauffman says, if the numbers game wasn’t favourable, the field wouldn’t
be going anywhere – or at least, would be advancing much more slowly. Like
Szostak and a number of other workers in this field, Kauffman had come into
it via an interest in pre-biotic evolution. ‘It is the question of what
is the probability that an arbitrary protein will catalyse an arbitray reaction,’
explains Kauffman. ‘It dawned on me that the way to approach this problem
was to generate random DNA sequences, random RNAs, and therefore random
proteins, and just go and find out.’ This was to be Szostak’s line of thinking
too.

As we saw earlier, even with a protein of fairly modest size – 100 amino
acids – there is a gargantuan number of possible amino acid sequences:
10¹³⁰, which vastly exceeds the number of particles in the Universe.
If every single sequence represented a single possibility in shape recognition,
practitioners in applied molecular evolution would need a great deal of
patience, and a big budget, to generate all possible sequences. ‘Many years
ago I was thinking about this, and I felt that there ought to be a lot fewer
shapes than there were polymer sequences,’ says Kauffman. ‘But I didn’t
know how to think about it.’

Then along came a paper by George Oster and Alan Perelson, of the University
of California at Berkeley. In it they extended an idea proposed by John
Maynard Smith, of the University of Sussex, some years earlier: the idea
of a ‘shape space’ that represent all the three-dimensional possibilities
of polymers. (They included other parameters too, such as charge, dipole
moment, hydrophobicity and so on, but shape was the key feature.) A point
in the space represents a very particular shape, and the question was, how
does the real world of polymer molecules match up to the essentially infinite
number of points in total shape space? Oster and Perelson approached this
by thinking about the immune system, whose business is recognising shapes.

The variable region of an antibody recognises not just a point in shape
space, they said, but rather a cloud of points, or ball, of similar shapes.
They then went on to say that you would expect even the most primitive immune
system to be able to cover a fair fraction – say a third – of total shape
space in the spectrum of antibodies it is capable of generating. The newt’s
immune system is one of the simplest, and it produces 10,000 antibodies.
From this Oster and Perelson calculated that, given overlap, some 10⁸
antibodies effectively cover all shape space. ‘That was amazing,’ says Kauffman.
‘It makes the world finite and manageable.’ This, remember, is molecular
shape, and it is not confined just to the space encompassed within individual
antibody variable regions: logically it must include molecular shape in
all its forms, whether nucleic acid, protein, carbohydrate, or lipid.

Kauffman then extrapolated the idea to catalysts. ‘One day I was thinking
about this, and realised we need catalytic task space,’ he recalls. The
question is, how do you define catalytic task in this context? ‘To a first
approximation, I don’t know. To a second approximation, it is the binding
of a transition state in a reaction complex. Shape. The same kind of thing.’
Was there any evidence for this? ‘Just as you can ask whether a carbohydrate
and a peptide have the same shape, you can ask whether there are enzymes
that catalyse very different reactions by the same site.’ There are indeed
such enzymes, and so by analogy with the antibody argument you can begin
to calculate the extent of catalytic task space and the number of enzymes
required to cover it all. ‘The answer comes to be about 10⁸,’
concluded Kauffman. ‘Isn’t that amazing, that the two numbers – for antibodies
and enzymes – should come out to be rougly the same?’

The notion that the world of biopolymers is effectively represented
by a mere 100 million shapes, give or take an order of magnitude, is mind-boggling.
But the kinds of numbers reported in the five recently published papers,
on very different systems, do in fact fall into this order of magnitude.
‘It means that with 100 million molecules, maybe a billion, you can do any
reaction you want,’ says Kauffman. ‘It’s a universal constructor set. I’m
pretty sure ‘I’m right. I’d bet a lot on it.’

The search for the right shape

The idea of applied molecular evolution is therefore the search for
the right molecular shape. Even if there are only 100 million shapes out
there to be screened, it might not always be possible to hit on the optimum
shape quickly. In which case, another technique of modern molecular biology
can be brought in: directed mutation. If you have a shape that is good,
but not the optimum, it might be possible to improve it. By analogy with
fitness in population genetics, you can think of the fitness of shape (for
a particular task) as being represented by a rugged langscape: the peaks
represent fitness optima, the troughs reduced fitness. The task is to get
your sub-optimal shape to be the highest peak possible.

Again, it is a numbers game. If you move in steps of single mutations,
the number of possibilities you have in a protein of 100 amino acids is
small – just 1900. With variants moving by two mutations at a time, you
suddenly jump to millions of possibilities. And with three-mutant variants,
you are quickly into the trillions, and rising. Although it would be technically
easier to work with one-mutant variants, this quickly leaves you stuck on
sub-optimal peaks. In order to make leaps from these peaks to more desirable,
higher peaks you have to take bigger jumps: two- and three-mutant variants.
This mutation across fitness landscapes takes you into novel molecular shapes,
territory never previously inhabited by living organisms.

This technique of selection, mutation, and further selection, ‘heralds
a new era in novel molecular design unrestrained by the rules that govern
organismic survival and replication,’ say Tuerk and Gold. ‘It produces unpredictable
and unimaginable molecular configurations of nucleic acids and proteins
with any number of targeted functions. Tuerk and Gold name their procedure
for doing this SELEX, and have patent applications relating to aspects of
it. Kauffman and Ballivet’s 1985 patent also covers the generation of large
numbers of mutant variants.

‘Everywhere I look I see the following,’ says Kauffman. ‘Biotechnology
is going a whole new way, cloning known genes and so on. And there’s a whole
new way, that of evolving novel biopolymers. In this new technology we are
going to have to learn how to handle millions of different molecules; we
are going to have to learn how to evolve across sequence space through mutation.’
If applied molecular evolution is to be more than fantasy, whole new technologies
will be necessary, particularly in instrumentation and software. But as
Kauffman puts it: ‘Because the task is finite, it feels inevitable.’