THE spiders in Fritz Vollrath鈥檚 laboratories have been confined to
barracks鈥攔anks of square plastic frames, stacked neatly like books on a
shelf. Each contains an arachnid and its ephemeral story written in silk. Until
a few weeks ago they had the run of the place, suspending their webs across the
ceiling and down the walls. Spider excrement wasn鈥檛 a problem, says Vollrath
dryly, so long as you covered the computer keyboards with paper every night. The
cleaners were not as tolerant. So, now, the spiders are locked up.
But there is no suppressing the enthusiasm that Vollrath, professor of
zoology at the University of Aarhus in Denmark, and his team have for their
eight-legged research subjects. Especially since they feel close to cracking the
problem of how spiders know what shape and form their webs should take. With a
combination of field studies, laboratory experiments and computer models,
Vollrath and his PhD student Thiemo Krink have created cyber-spiders that can
produce webs as skilfully as their living counterparts.
Cyber-spiders have no substance, they are disembodied computer programs that
spin virtual webs. Mimicking the behaviour of real spiders, they employ
mathematical rules to make decisions about where to attach a new thread in the
stepwise process of web-building. What鈥檚 more, like real spiders, cyber-spiders
have honed their decision-making skills by a process of evolution. A
computer version of natural selection sees to it that only cyber-spiders
producing the best virtual webs survive.
Advertisement
Now, Vollrath and Krink are pushing their cyber-spiders to the
edge鈥攕eeing whether they react to environmental change in the same way as
real spiders when they are forced to build their webs in wind tunnels,
centrifuges and under other extreme conditions. And so far so good. 鈥淲e must be
very close to the real rules,鈥� say Vollrath, and the surprise is just how simple
they seem to be. Spiders, it turns out, can build a huge variety of webs with a
very limited range of abilities. They can measure distances and angles. They can
assess the orientation of their bodies in relation to gravity. But
astonishingly, given the complexities of webs, spiders have no overall plan to
guide them. Their memories last no longer than a couple of web moves.
Hot spot
Globally speaking, understanding how spiders make decisions about their webs
might seem like a clever, but narrow, achievement. In fact, it immediately plugs
you into one of the hottest areas in zoology鈥攖he quest to understand how
environment influences the genetic component of animal behaviour.
Each web is a relic of a fleeting foraging behaviour. And many spiders,
including the common garden cross spider, Araneus diadematus, the focus
of most web research, build a new web each day. That鈥檚 around 200 webs in a
lifetime, each varying slightly depending on environmental conditions. Subtle as
they are, these variations let biologists monitor with unrivalled precision how
spiders shift their behaviour to suit conditions. And measuring the success (or
otherwise) of each behavioural shift is simplicity itself: you just count the
number of prey trapped in the web.
Vollrath has been studying spiders for over two decades. He got into the
field quite by chance. As a young neurobiologist he grew tired of the sedentary,
lab-bound lifestyle. A contact offered him a field trip to Panama to study
spider 鈥渒leptoparasites鈥�, little spiders that live on the webs of bigger spiders
and steal their prey. Vollrath was hooked. He now heads one of the biggest
spider research groups in the world, with collaborators both in Aarhus and at
the University of Oxford. They are a select little band, which is part of the
fascination. 鈥淚f I have an idea, I have to go and test it myself鈥攖here
just isn鈥檛 the mass of scientific literature you get in other fields.鈥�
Web research begins with some simple observations. Watch enough spiders build
new webs in the wild and it becomes apparent that the behaviour can be broken
down into various stages. First, Araneus explores the web site,
clambering around trailing silken thread to get the lie of the land. Then it
begins to build. The starting point is a Y-shaped structure produced by
stringing a thread between two high points and abseiling off the middle to
attach a second thread to a lower site. The junction of the Y is the web鈥檚 hub,
the arms and stem are the first 鈥渟pokes鈥�. Next, the spider moves round the hub,
extending it and attaching more spokes as it goes. After that, it spirals out
from the hub to lay a further thread, making between four and eight turns of the
web.
During the final stages of web construction, the spider will inch back along
this 鈥渢emporary spiral鈥�, destroying it as it goes. But in the meantime, it has
other priorities. Half-completed, the web is made entirely of non-sticky silk,
allowing the spider to move around with ease. If this were the finished web,
prey would bounce off it like a trampoline. So the spider must produce another
specially coated silk that absorbs water from the atmosphere, making it both
sticky and stretchable.
Armed with such silk the spider finishes its handiwork by creating the
鈥渃apture spiral鈥�. From the outer reaches of the web (normally near the bottom)
it first zigzags its way back, using the temporary spiral as a handrail and
laying threads as it goes. Nearer the hub the zigzags develop into full circles,
the spider completing up to 50 of these to finish the capture spiral. The final
stage is to fine-tune tension in the web as a whole, by making adjustments at
the hub. Then the spider sits and waits.
Just watching this behaviour can give clues about the decision processes that
govern web-building. Site choice, it transpires, is almost certainly random. But
the lack of symmetry of most webs is quite deliberate. The capture spiral is
invariably longer than it is wide, with the hub placed high up, off-centre. It鈥檚
as if the spider鈥檚 web-building program contains a rule that says 鈥済ravity means
you can run down faster than up, so prey caught in the bottom of the web is most
likely to provide dinner鈥�.
It鈥檚 clear too that spiders build webs with widely varying numbers of spokes.
It鈥檚 as if they decide in advance what maximum geometric angle they are prepared
to accept between the spokes and continue to add new ones until all the gaps are
filled. Spiders also seem to decide beforehand the spacing between the
capture-spiral threads linking each spoke. And intriguingly, even though
Araneus has poor sight, it manages to keep these spacings regular.
Vollrath has discovered precisely how. By observing the behaviour of spiders
with half-length legs鈥擜raneus sheds its legs if they become
damaged and regenerates new but shorter ones鈥攈e found that they measure
the distance between consecutive threads with their front two legs.
In fact, web-building is even more subtle than this picture suggests. In one
experiment, Vollrath tried to trick spiders into creating distorted webs by
cutting a single thread of the capture spiral soon after its construction. When
the spider spirals back round to the damaged area it finds that it must reach
twice as far as expected. Surprisingly, however, any distortion in spacing is
compensated for by the spider within a few full turns of the spiral. If the
spider based its decisions on distances alone, says Vollrath, the blip would
stay in the web all the way up. But spiders can also measure the angle at which
a thread is attached to a spoke. What鈥檚 more, they have an expectation of what
the correct angle should be, based on the memory of what they have just done and
how they are oriented on the web. 鈥淚f the angle is too far out then they find an
intermediary between the expected and the found,鈥� says Vollrath.
And it turns out that all these angle and distance measurements are
influenced by gravity. During a spacelab mission in 1963, NASA discovered what
happens when Araneus builds webs in the low-gravity environment of
space. The photos sent back to Earth were poor, but the space webs were clearly
more symmetrical than earthbound webs. Vollrath got a similar result simply by
forcing his spiders to build webs in horizontal planes, thereby removing any
advantage of having a skewed web. But Vollrath wanted to go further and discover
precisely how gravity influences each of the spider鈥檚 web-making decisions. So
he built a contraption akin to a fairground wall of death and started spinning
his spiders in it.
Downhill fast
鈥淭he faster you spin a spider, the more problems it has in knowing where it
is,鈥� he says. The disoriented spider stops producing a nice regular spiral and
begins moving backward and forward, laying down threads at random. But spin the
spider at 120 revs per minute鈥斺�漷hat鈥檚 like cycling downhill really fast
with the spider building its web within your wheel鈥攁nd it will start
producing regularly shaped webs again. This is because the effects of gravity
are cancelled by the centripetal force.
Armed with results from these spinning experiments, Vollrath could model the
influence of gravity on the web-building rules. He discovered that spiders
continuously reassess gravity when making their webs, for example, and that
gravity shapes the web more directly than factors such as prey size. Other
experiments show that spider body weight鈥攕omething that would obviously be
affected by gravity鈥攊s a factor too.
Spiders spin their webs according to their means. In other words, they know
in advance how much silk they have stored in their abdomen and adjust the mesh
size of their capture web accordingly. Vollrath found that if he destroyed a
spider鈥檚 web soon after it had been built the spider would build another using
slightly less silk. Repeat the procedure and the webs become progressively
smaller. 鈥淭here is a very good correlation between the amount of silk in the web
and the time between webs,鈥� he says.
But that is not all. For Vollrath has also discovered how spiders judge the
amount of silk they are carrying. They sense their own body weight in some as
yet unknown way. You can make spiders artificially heavy by spinning them in a
centrifuge. 鈥淎ll the way up to 15 g [which increases perceived weight
to 15 times as much as usual] they still build webs,鈥� says Vollrath. 鈥淭he web is
no longer very good, but it鈥檚 good enough to catch prey.鈥� In addition,
because the spider mistakenly thinks it has lots of silk, the threads get
progressively thicker and the web smaller as the deceived spider runs out of
building materials. Vollrath found a similar effect when spiders constructed
their webs in wind tunnels. 鈥淭he spider can adjust its engineering to local
conditions,鈥� he comments. In the case of wind, this means deciding in advance to
produce stiffer silk.
Still, working out that web-building decisions depend on gravity, distances,
angles, wind speed and so on is only half the battle. The crunch comes, says
Vollrath, when you try to discover how the rules work together and their
relative importance in producing the finished web. This is where computer
modelling plays a crucial role.
Vollrath has been using computers to refine his ideas about spider behaviour
for years. The early work was done at Oxford with Nick Gotts, a psychologist
from the artificial intelligence laboratory at the University of Sussex. The
result was a computer model called Theseus which was developed further by Krink
and Peter Fuchs.
The latest generation of simulations, created by Krink, are something
special. Instead of attempting to simulate perfect web-building in one
megaprogram, Krink has taken the various aspects of the behaviour and assigned
them to between 50 and 60 individual cyber-genes. Each encodes a decision
rule鈥攕uch as mesh size and angle of thread attachment鈥攃ontrolled by
a set of parameters which modify its calculations,
and a weighting which determines how much the rule influences the cyber-spider鈥檚
behaviour [see
Diagram]. All cyber-spiders carry a
full complement of web-building genes but the parameters and weighting
controlling them are left to chance. The result is a population of spiders that
can build webs, after a fashion. 鈥淭he starting webs are always random,鈥� says
Krink. 鈥淭hen we let them evolve.鈥�
Sexy silk
They can do this because Krink鈥檚 spiders can reproduce sexually. When two
individuals get together their genes are shuffled in just the same way as they
are when real spiders mate. The genes can also mutate. The result is a mixed bag
of offspring. Each is allowed to build a web and then natural selection gets to
work on them. 鈥淭he system selects for those that are most efficient at using
their silk to catch prey,鈥� says Krink. The webs are peppered with prey and the
most efficient鈥攖hose that best balance costs such as silk production and
construction time against the benefits of catching prey鈥攕urvive the
selection process. Spiders with these genes get to mate and produce another
generation.FIG-mg20424002.GIF
It generally takes no more than 50 generations for the population to produce
webs that are as efficient as possible given the genes available in the first
place. Spiders with a different gene pool, however, might solve the problem even
more efficiently. That鈥檚 why Krink has now developed a computer model that
allows more successful populations to influence less efficient web-builders,
simulating what happens in the real world.
The first generation of spiders is divided into subpopulations, each of which
tries to solve the foraging problem as best it can with the genes it has been
dealt. But all is not lost for those dealt poor genes. In Krink鈥檚 program,
cyber-spiders are allowed to move to new subpopulations, taking their genes with
them. And in this interbreeding, it鈥檚 always the most successful subpopulations
that send out most migrants. To the low-achievers the incoming genes are like an
infusion of new ideas. Run the program long enough, and eventually all the
spiders will end up foraging as well as possible given all the genes available.
And crucially, because each rule is encoded on a separate cyber-gene, the
researchers can go back and analyse how it fared throughout the selection
process. This enables them to work out its importance relative to the other
genes.
But how close are these rules to simulating the decision processes of real
spiders? The test comes when Vollrath and Krink compare the performance of their
fully evolved cyber-spiders with the best that nature can offer.
This work has only just begun. But cyber-spiders have already achieved some
success as understudies for real ones. One test uses a real web, digitised and
then stripped of its capture spiral. A cyber-spider is placed on it at the point
where the temporary spiral was finished. 鈥淚t has no idea about what has happened
before. But it is able to complete the complex structure of the web with just
local information,鈥� says Krink, who has devised a statistical method to show
that the impostor鈥檚 efforts pass muster.
Sticky snare
In another experiment, cyber-spiders forced to build webs in long, thin
spaces positioned the hubs way off-centre and constructed highly distorted
capture spirals to fill the available space. Such webs occur naturally in two
species of spiders that eat moths. Moths are particularly slippery prey because
they shed scales from their wings when caught to evade the spider鈥檚 sticky
snare. 鈥淟adder-web鈥� spiders have evolved two strategies. Either they concentrate
their web-building efforts in the area directly below the hub, so as to outrun
their moth prey, or they build upward from the hub hoping that their prey will
tumble down into their waiting jaws.
Vollrath has a library containing thousands of photos of real spiders鈥� webs
created under all sorts of experimental conditions. But the puzzle he most wants
to solve is the peculiarity known as cribellate spiders.
These make the same sort of two-dimensional orb webs as Araneus,
but they cannot produce sticky silk. Instead they build their capture spiral
using a tangle of extremely fine silk fibres that have been 鈥渂ack-combed鈥� to
create a fuzzy wool-like texture. The question is, did cribellate spiders and
those that produce sticky silk have a common ancestor that built orb webs, or
did orb-building evolve twice, once for the cribellates and once for
Araneus-type web-builders? 鈥淒o cribellates use the same set of rules or
a subset of rules that can be evolved into the other rules without putting
anything new in?鈥� asks Vollrath. If the answer is yes, then the two groups do
share a single orb web-building ancestor. 鈥淲e鈥檙e working on it,鈥� he adds.
If Vollrath and Krink can solve such mysteries their research will have come
full circle. The principles of evolution have helped them refine their ideas
about web-building. Now, armed only with an understanding of the rules that
govern that behaviour, they can start to untangle some of the threads of spider
evolution itself.