With
that preamble, I'd like to introduce my perspective on neural circuits.
[What are "neural circuits"?]
Behavior is controlled by the nervous system. The nervous system consists
of large numbers of individual cells, which are electrically active and which
communicate with one another by secreting chemicals at sites of contact called
synapses. Somehow, the properties of the cells and their connections
determine behavior. [What?! Yes,
I know that these assertions are simplistic. For a more honest and complete
formulation, click here.]
Let me give you just a couple examples, to illustrate neural organization
at the cellular level. These examples aren't coming from the human brain,
not even from cats or rats or monkeys, but from invertebrates. Now,
when you think of cognition, you may not think first of bugs. But even
with their pin-head brains, insects and spiders and crustaceans are capable
of some pretty fancy feats of perception and motor performance [ref20].
Compared to the best robots we have yet constructed, bugs are rather
impressive pieces of work.
For
my doctoral dissertation [ref1,2], I worked on a small
piece of the nervous system of the spiny lobster. The adjacent diagram [ref3]
shows the front end of the lobster, facing toward the left. The nervous
system is distributed in several cell clusters, with the largest located up
front, and with additional ganglia in each segment of the body. The
big gray lump is the stomach. The lobster's stomach has teeth, mounted
on skeletal levers. The muscles that work these levers and teeth express
a moderately elaborate rhythm -- grab, grind, release, repeat, squeeze, filter,
etc.
The
rhythm of lobster stomach movements is controlled by neural circuitry concentrated
on a tiny ganglion (red arrow) on top of the stomach. In Allen Selverston's
lab at UC San Diego, we studied the cells comprising this ganglion.
Lobsters are especially handy for cellular neurobiology because their nerve
cells are conveniently large and accessible. By placing electrodes into
each cell, it is possible to determine each cell's identity. By carefully
recording and experimentally manipulating the activity of these nerve cells,
it is possible to map the "circuit"
of connections among these cells.
There
is an important point here which may need some emphasis. This ganglion
is not just a collection of several interchangeable, general-purpose nerve
cells. This is an organized set of uniquely identifiable individual
cells and cell sets, each of which can be recognized individually in any particular
specimen.
The cells are interconnected in a very specific way [enlarged
diagram] [ref4], and each cell has specific electrical
parameters upon which the circuit behavior depends. From the interaction
between the properties of the individual cells AND from their effects upon
one another, a rhythmic motor pattern emerges, a pattern which can be switched
or tuned by input from other cells outside the ganglion.
There
are, in fact, dozens of quantitative parameters which are relevant for proper
neural functioning, and these parameters vary in consistent ways among the
various unique individual cells. And this sort of circuit-diagram doesn't
begin to do justice to actual biology. Each cell has elaborately branching
processes [enlarged image].
The
synaptic connections among the cells occur in a complex tangle, and the number
of individual points of contact number on the order of one million [enlarged
image].
Okay so far? Biological systems, even ones as relatively simple as
this 30-cell ganglion, can be impressively complex and intricately organized.
There is no empirical evidence that larger systems, with more cells
and more impressive behavioral capacities, are any less precisely organized.
Just how precise can this organization be? I'll give just one more
example, this time from the famous fruit fly, Drosophila melanogaster
[enlarged image].
Fruit
flies are not convenient for cellular neurobiology. Although a fly's
nervous system contains thousands of cells, the whole animal is is only a
couple millimeters long, not much bigger than the 30-cell ganglion we just
looked at. Nevertheless, its nerve cells can be studied.
Here's
a slice across the nervous system in the thorax [enlarged
image]. This portion of the nervous system contains the circuitry
for establishing the basic coordination of muscles needed for most activities
-- flying, walking, singing, mating. (The brain, in the head, handles
sensory perception -- sight and hearing and smell -- and modulates motor patterns
accordingly.) Notice that this is not just squishy mush. There
is an obvious symmetry, which reflects the fact that particular, identifiable
individual cells have unique and precisely determined features.
Certain gifted individuals, such as my colleague Mark
Tanouye, can successfully record electrical activity from particular individual
nerve cells, even in this small animal. 
The
circuit which is best known involves a pair of very large nerve fibers which
pass from the head into the thorax and synapse with a thoracic interneuron,
which
in turn synapse with motor neurons which go out to muscles [ref5].
This circuit mediates the rapid escape reflex that enables the fly to
dodge the fly swatter (more on giant axons and escape
reflexes). Within this circuit, particular synapses between particular
parts of identified nerve cells can be reliably located in the same place
in each individual fly [enlarged image].
Once one learns where to look, one can even locate the homologous synaptic
sites in various different species, even in species from different families
of flies [ref6].
What of evolution? Here is a slice across the neck of a fly, showing
the thousands of fibers which connect the head and the thorax [enlarged
image]. 
Note
the giant axons of the escape reflex circuit. Each of the other fibers
also has some particular function. We don't know how many of them are
unique, but it is apparent that several are exceptionally large. Symmetry
is not obvious here, because of incidental distortion. 
Here
is a different species [enlarged image],
in which symmetry is more evident, demonstrating that quite a few individual
fibers can be readily recognized just by size and position. Now, fiber
diameter matters for nerve cell function -- it determines speed, power, and
reliability of signalling (more on nerve cell function).
Fiber diameter is only one of dozens of quantitative parameters which
differentiate nerve cells from one another and which significant for proper
function. Fiber diameter is adjustable by evolution, as evident by differences
among species [ref23] (more on axons in fly necks).
Okay. Now let's think about this a bit. I call your attention
to several points.
First point: Brains have evolved. Biological brains
are the only examples we have of material objects capable of cognition. Even
at the "simple" level of perception, problem solving, decision making,
and motor coordination represented by the behavior of a fly, which you may
or may not wish to denominate as cognition, biological systems far exceed
our best inventions. Biological brains, of course, have been invented
by evolution. If we can discover how evolution did the inventing, that
result might be useful for understanding cognition itself.
Second point: Biological systems display evolvability.
Brains not only have evolved, they have the intrinsic ability to evolve.
This attribute, although often taken for granted, is utterly alien to
the computers and software we often use to model or mimic cognition. Evolvability
is fundamental to all existing cognitive systems.
Third
point: Behavior evolves readily. That is, however the brain
works, the behavior it produces is readily "reprogrammed" by evolution. For
all the attention given to learning, fundamentally the brain is reprogrammed
by changing the hardware. [Yes, this wording is simplistic. Click
for more.] In spite of the learning
capacities of most animals, and the much vaunted plasticity of the human brain,
specific parts of the brain are dedicated -- "hard-wired" in some
squishy sort of way -- to perform specific functions. For that matter,
the basic ability to learn, to adapt to particular and changing conditions
within the span of a single lifetime, is part of the hardware. The ability
that brains have to learn in particular ways at particular times is itself
a built-in, evolved function. [In the words of Bernd Heinrich and Thomas
Bugnyar (Scientific American, April 2007), "By some process that
still remains one of the great unsolved mysteries of biology, exquisitely
precise behaviors can be genetically programmed in animals with brains no
larger than a pinhead."]
Fourth point: We already understand,
at least in basic outline, how evolution works. Darwin's basic theory
has several components. Evolution emerges from the interaction of these
component processes.
- Variation. Organisms vary.
- Heredity. Some variation (in morphology and behavior) is
causally linked to differences in genes.
- Fitness differences. Some hereditary variants work better
than others (at least statistically).
- Reproduction. Organisms reproduce in excess, typically producing
many more offspring than can the environment can support.
- Selection. The conditions of existence determine that those
organisms whose particular variations better suit them for surviving and
reproducing will, on statistical average, be the ones which do survive and
reproduce. To the extent that their advantageous qualities were hereditary,
these qualities will be preserved and perpetuated in subsequent generations.
- Mutation. As less-fit variation is eliminated by selection,
new variation continually arises by mutation.
- Recursiveness. This process repeats indefinitely. The
output from preceding evolutionary processes is the input for subsequent
evolution.
In modern terms, a gene is a particular DNA sequence which is causally associated
with some aspect of an organism's form and function. Mutations
are changes in DNA sequence. Variations in DNA sequences lead to variations
in form and function. Natural selection on the various forms and functions
leads, within a population, to replacement of one particular sequence by another.
This is standard textbook evolution. Every part of the process is
empirically observable. In many ways, we understand the process quite
well. The piece that is least understood is mutation -- over
a century after Darwin we still don't know much about the nature of the specific
mutations which have led to any particular adaptation -- and most especially
nothing about those leading to behavioral adaptations.
Behavior evolves as one mutation arises, is associated with a small incremental
advantage, and thereby spreads through a population. But we don't know
what kind of mutations can produce adaptive advantages in behavior.
If we want to know
how any particular aspect of cognition emerges, we need to discover the nature
of the evolutionary transitions between pre-cognitive material systems and
those capable of cognition. Within cognitive systems, we can ask the
same question about particular aspects -- face recognition, language acquisition,
even walking upright without falling over. This, I believe, is a key
question for understanding the biological basis for cognition: What
manner of mutations led us from there to here?
You have probably been taught that all the heavy lifting in evolution is
done by selection. That is, although mutation provides raw material,
all the creative shaping is done by selection. In the quote at the start
of this presentation, you might have noticed that Daniel Dennett gave Darwin
credit NOT for the whole package of evolution, but for the specific idea of
natural selection. Happily, we do have a pretty good idea how selection
works in relation to cognition. Basically, smarter is better -- at least
for animals with active life-styles. (I add the "active" qualifier
because there are many life styles, notably those of parasites and of sessile
filter-feeders like clams, for which being smart offers little benefit and
the organisms in question have notably low cognitive capacities.)
Anyway, conventional pedagogy presents natural selection as the principle
designer in evolution. Mutation is usually treated rather dismissively.
Although mutations are necessary as a source of variation, mutations
are nothing but random accidents, the inevitable result of imperfect reproduction.
Does that last sentence sound familiar? Mutations are nothing
but random accidents, the inevitable result of imperfect reproduction.
It should, because it is a commonplace textbook assertion. However,
it is also false -- or, at the very least, a grave oversimplification. Even
Darwin understood the deep significance of variation -- In his Origin of
Species, Darwin wrote,
"I have hitherto sometimes spoken as if the variations . . . had been due
to chance. This, of course, is a wholly incorrect expression, but
it serves to acknowledge plainly our ignorance of the cause of each particular
variation. Some authors believe it to be as much the function of the
reproductive system to produce individual differences, or very slight deviations
of structure, as to make the child like its parents."
Origin
of Species, Chapter 5
And then,
"When the views entertained in this volume on the origin of species, or
when analogous views are generally admitted, we can dimly foresee that there
will be a considerable revolution in natural history. . . A grand and almost
untrodden field of inquiry will be opened, on the causes and laws of
variation." Origin
of Species, Conclusion
Unfortunately, that field remains largely untrodden. We still
do not know much about "the causes and laws of variation" [but see
ref22].
If this talk has one central message, it is this: The mutations
which sustain the evolutionary process are neither random nor accidental.
Mutations do have, in Darwin's words, "causes and laws".
But science has barely begun to understand the signficance of that fact.
Mutations are produced by mechanisms which are themselves the product
of the evolutionary process. Direct Darwinian selection acts on bodily
form and function, to produce adaptation. But selection also operates
indirectly, shaping mechanisms which, recursively, make the evolutionary
process more efficient. Indirect selection favors genetic systems which
increase the probability that some variation will be beneficial (or, equivalently,
which decrease the probability that variation will be deleterious) [ref14].
Because many mutations are the product of active biological organization,
the domain in which those mutations are generated can itself be shaped, and
indeed has been shaped, by evolution.
This assertion remains deeply heretical. Nothing like it has been
incorporated into mainstream evolutionary biology. But it is also a
matter of established fact. There is abundant evidence that mutations
arise from complex chemical reactions which are precisely managed by enzymes.
DNA does not sponaneously self-replicate, textbook cartoons notwithstanding.
Furthermore, DNA is not left alone once it forms. It is proof-read,
edited, cut, spliced, rearranged, you name it. There are highly evolved
enzymes for these functions. Mutations don't "just happen", they are
the specific product of enzyme activity, and those enzymes are themselves
the product of evolution.
Here is a paraphrase, in modern terminology, of the quote I just read from
Darwin: "It is as much the function of DNA metabolism to produce individual
differences, or very slight genetic variations, as it is to replicate genes
accurately." Here is how David Thaler put it, in a1994 article
in Science titled The Evolution of Genetic Intelligence: "The
generation of genetic variation is in large part controlled by the genes and
physiology of DNA metabolism... Natural selection acts beyond particular alleles.
It also favors genetic metabolism that generates alleles with a high
probability of passing the tests of environmental selection" [ref7]
Of course, mutations can be accidental. You can
feed mutagenic chemicals to animals, or expose them to x-rays or ultraviolet
light, and damage their DNA. Such unnatural mutations, induced experimentally
in order to break genes and study the consequences, have been very important
for the science of genetics. But such experimental mutagenesis has
warped our perception of how complex and evolved (that is, how biological)
the natural process of mutation really is.
So, what is the significance for cognitive science, of thinking about mutation?
Simply this: In order for complex patterns of organization to
evolve, the appropriate variation must have arisen before it could be selected.
Evolutionary biology has largely ignored this simple requirement. Because
evolution has occurred, variation has been presumed to have been adequate
for the job. If a selective advantage for a step-by-step transformation
can be identified, then we act as if that step-by-step transformation has
been explained. We take for granted the process of mutation, the process
which generates appropriate variation. Because the evolutionary process
has occurred, because we know the result, therefore the necessary and appropriate
variation must have arisen. QED?
But this conventional approach is really missing a big opportunity. Evolvability
is NOT a simple, self-evident attribute, not for bacteria and certainly not
for complex, highly integrated systems like the human nervous system. It
is far from obvious that random accidental mutation should be sufficient for
evolution of cognitive functions, or even for the instinctive behaviors produced
by the nervous system of a fly.
It is not good enough just to presume that random accidental mutations have
explored all possibilities and that selection has done the rest, especially
when we already know of so many elaborate enzymatic mechanisms for tinkering
with DNA. We should be asking, What manner of mutation, and what
manner of encoding between mutation and end product, are most likely to yield
adaptive results? How can it be, that the nervous system can
accommodate changes in its genetic instruction set, such that plausible fiddles
with those instructions can make the system work in better and more elaborate
ways? Can you imagine typographical errors in Shakespeare leading to
a more deeply meaningful play? Or typographical errors in a computer
operating system leading to a more powerful and reliable work station? Remember,
selection cannot work magic. Selection can only choose among available
variations.
Let me elaborate. Another of the talking heads in last week's (Oct.
2001) PBS series on evolution
was Steven Pinker,
author of How the Mind Works, in the Department
of Psychology at Harvard. Pinker emphasized how wonderfully the
parts of our brains have been fine-tuned by evolution. According
to Pinker, higher cognitive functions, particularly language, have evolved
not just by making more brain-stuff, but by wiring it in precise ways. We've
just seen what precise wiring entails, in systems that are simple enough for
us to investigate at the level of specific cell-to-cell connections (above).
And then Pinker noted that our brains contain hundreds of billions
of nerve cells, with hundreds of trillions of connections.
This should raise an intriguing question. How do you accomplish
the "precise wiring" of hundreds of trillions of connections, by a process
in which one mutant gene gradually replaces another over many generations?
How many genes must be involved? How many gene replacements would
it take? According to the latest results from the Human Genome Project,
the human genome contains only a few tens of thousands of genes. And
presumably we need at least a few thousand of those genes for making and running
all the other parts of the body. Am I the only one who sees a problem
here?
It's not too big a trick to use a small set of instructions to construct
a system consisting of many individual parts. Consider a checkerboard.
64 squares, each with a particular position and color. That checkerboard
can be specified with about four instructions: make an array of squares,
8 this way, 8 that way, with alternating colors. But what if you wanted
to make one particular square green? And another particular square orange?
It's one thing to make a complex pattern. It's quite another thing
to make it in such a way that particular elements can be precisely tuned for
particular functions, and do so with a seemingly restricted instruction set.
The genome is not a limitless reservoir of information. It is
quite finite, and and it appears to be relatively small compared to the system
which it specifies. (And, given the slowness of the evolutionary process,
the amount of time available for the genome to evolve is also quite finite.)
One
might intuitively expect the evolutionary process to slow down asymptotically
over time. The more elaborate and functionally sophisticated a nervous
system becomes, the harder it should be to make further improvements by random
trial and error. In fact, the evolution of instinctive behavior, which
must involve genetically-based reorganization of complex neural connections,
apparently occurs faster and with greater ease than the evolution of morphology.
Why should there be a gene -- how could there be a gene --such that
by mutating it we make a more appealing courtship song, or a more effective
spider web, or a more versatile innate grammar? What kind of
prior organization, whether of DNA sequences or nerve cells, will permit such
a process to work? These are the sort of questions that should be asked,
but very seldom are.
While thinking about such questions (back in the early 80's, when I first
became impressed with the problem of evolvability), I tried to imagine some
evolvable patterns of organization which would in turn facilitate the evolutionary
process. The metaphor of "tuning knobs" presented itself.
Evolution might work more efficiently if adaptively (cognitively) interesting
features were adjustable.
At the time, I was enjoying books by Douglas
Hofstadter (Cognitive Science
at Indiana University). In Metamagical Themas,
Hofstadter writes, "Variations on a theme is the crux of creativity." And
he then introduced the metaphor of "twiddling a knob on a concept" [ref8].
Variations on a theme are obvious in biology. How, I wondered,
might the genome implement the idea of twiddling a knob on a concept, given
that protein-coding genes (the standard "genes" of textbook definition)
are rather brittle and conservative?
The genome includes a lot of stuff which is called "junk DNA." The
idea of junk DNA is sometimes understood, metaphorically, as a resource for
tinkering. But rather more commonly, it is understood as useless trash.
Among the various categories of junk DNA are repetitive sequences --
stretches of DNA in which a pattern of bases repeats itself over and over
dozens or hundreds of times. Not only is the information content apparently
minimal, but mutations in such sequences are extremely frequent. Repetitive
sequences are notoriously unstable, most notably by increasing or decreasing
the number of repetitions. Surely, these sequences are just useless
trash. Surely such uninformative, unstable sequences could not play
a role in precise adaptation?
However, what if information was represented in the length -- the number
of repetitions -- of such DNA sequences, such that changes in length would
exert a quantitative effect on the expression of some hereditary trait? Then
frequent mutations in length would amount to twiddling a knob on the trait,
allowing for rapid and precise evolutionary tuning.
I tried to publish this idea in 1985, but didn't succeed. Then in
the early 90's, things became quite exciting. 
In
the journal Science, under the headline, "The Puzzle of the Triplet
Repeats", came news of some peculiar human neurological diseases associated
with repetitive DNA triplets [ref9]. The severity
of the disease was associated with the number of repetitions, and mutations
affecting repeat number were quite frequent. This was exciting. The
pull-quote, "No one expected that DNA sequences could be so unstable
or behave as these do," was especially provocative. 
After
all, this is exactly the behavior I had speculated, a decade earlier. Here
we had repetitive DNA "twiddling a knob" on a disease. I
wrote to Science. My letter was published on February 4, 1994
[ref10]. The very next week, Science published
an article with the title "Transcriptional Activation Modulated by Homopolymeric
Glutamine and Proline Stretches" [ref11]. This
is dense jargon; I almost missed the article altogether. "Transcriptional
activation" means control of gene activity. "Homopolymeric"
means "a polymer of like units"; glutamine and proline are amino
acids, each encoded by a triplet of DNA bases. So, homopolymeric glutamine
and proline stretches translates to DNA triplet repeats. What the title
says, then, is that genes can be regulated by triplet repeats. And
the data showed not only that the number of repetitions had a quantitative
effect on gene expression but that such sequences turned up in all sorts of
regulatory genes.
I have since learned that this same idea emerged independently in at least
four locations. Here at SIU with me, at the University of Zürich
with the researchers cited above, at the Weizmann Institute of Science with
a theoretician named Ed Trifonov (a Russian emigre to Israel), and at the
Hebrew University in Jerusalem with quantitative geneticists Morris Soller
and Yechezkel Kashi.
I have collaborated with the latter group, reviewing the extensive evidence
that a mechanism of genetic regulation based on repetitive DNA is pervasive
and interpreting this mechanism as evidence of an evolutionary-tuning-knob
function [ref12,13,14]. Although the idea has still
not caught on in a big way, the more general concept of evolving strategies
for mutation is receiving some serious attention. A meeting at the Rockefeller
Institute in 1998, convened by Nobel prize winner Werner Arber to discuss
"Molecular Strategies in Biological Evolution", received news coverage in
Science [ref15] (proceedings later published by
NYAS [ref19]). Participants at this meeting expressed
delight in discovering colleagues, working in various different disciplines,
who shared similar vision. The "tuning knob" metaphor was introduced
by Trifonov (the Russian theoretician), who was astonished to learn that I
had published the same metaphor the preceding year. Then, in January
1999, the evolutionary tuning knob concept was picked up in an article in
Scientific American, "DNA Microsatellites: Agents of Evolution?", by
Richard Moxon and Chris Wills [ref16].
The basic idea of molecular strategies for mutation has been expanded to book
length by Lynn Helena Caporale (one of the organizers of the Rockefeller meeting
mentioned above) [ref19].
Are "tuning knob genes", based on repetitive DNA, important for
cognitive science? Nobody knows. But consider the following. The
period gene in Drosophila contains a hexanucleotide repeat in which
variation seems to affect the temperature sensitivity of the fly's circadian
rhythm [ref17]. The human D4 gene, which encodes
a dopamine (neurotransmitter) receptor, contains a 48-nucleotide repeat in
which variation has been linked to personality differences in novelty-seeking
[ref18]. Most of the genetic diseases associated
with triplet repeat expansion are neurological diseases. The
inheritance pattern of bipolar disorder (manic-depression) suggests genetic
anticipation, the hallmark of the triplet-repeat diseases. It is perhaps
not too far-fetched to speculate that mutational adjustment of repeat-based
genetic "tuning knobs" has been involved in rapid evolution of the
cognitive abilities which characterize our species -- most notably language
facility, tool use, and sophisticated social structures based on substantial
differences in talent and personality. [Evidence continues to accumulate,
that tandem repeats are especially important in nervous system genetics (ref21).]
The concept of tuning knobs is quite simple. I hope that the genetic
tool kit of mechanisms underlying evolvability will grow more interesting,
once folks catch on that (1) evolvability may be a most significant feature
of cognitive systems, and that (2) the mutations which are needed to enable
the evolvability of cognitive systems are not random accidents [ref22, ref23].
