Computing with DNA

The manipulation of DNA to solve

mathematical problems is redefining
what is meant by “computation”

by Leonard M. Adleman
funny in refrigerators (my view from
Rediscovering Biology undergraduate days in the 1960s at
the University of California at

M y involvement in this story began Berkeley). The field was undergoing

a revolution and was rapidly
in 1993, when I walked into a acquiring the depth and power
molecular biology lab for the first previously associ
time. Although I am a mathematician
and computer scientist, I had done a
bit of AIDS research, which I ated exclusively with the physical sci
believed and still believe to be of ences. Biology was now the study of
importance [see “Bal anced in formation stored in DNA—strings
Immunity,” by John Rennie; Sci of four letters: A, T, G and C for the
entific American, May 1993]. Unfor bases adenine, thymine, guanine
and cyto sine—and of the
transformations that information
tunately, I had been remarkably undergoes in the cell. There was
unsuc cessful in communicating my mathematics here!
ideas to the AIDS research Late one evening, while lying in

C
omputer. The word conjures
up images of keyboards and
monitors. Terms like “ROM,” “RAM,”
“gigabyte” and “megahertz” come to
mind. We have grown accus tomed
to the idea that computation takes
place using electronic compo nents
on a silicon substrate.
But must it be this way? The
comput er that you are using to read
these words bears little resemblance
to a PC. Perhaps our view of
computation is too limited. What if
computers were ubiq uitous and
could be found in many forms?
Could a liquid computer exist in
which interacting molecules perform
computations? The answer is yes.
This is the story of the DNA
computer.
community. So, in an effort to bed reading Watson’s text, I came to
become a more persuasive ad a de scription of DNA polymerase.
vocate, I decided to acquire a This is the king of enzymes—the
deeper understanding of the biology maker of life. Under appropriate
of HIV. Hence, the molecular biology conditions, given a strand of DNA,
lab. There, under the guidance of DNA polymerase pro duces a
Nickolas Chelya pov (now chief second “Watson-Crick” com
scientist in my own lab oratory), I plementary strand, in which every C
began to learn the methods of is replaced by a G, every G by a C,
every A by a T and every T by an A.
modern biology.
I was fascinated. With my own For ex ample, given a molecule with
hands, I was creating DNA that did the se quence CATGTC, DNA
not exist in nature. And I was polymerase will produce a new
introducing it into bacteria, where it molecule with the sequence
GTACAG. The polymerase enables
acted as a blueprint for producing
DNA to reproduce, which in turn
proteins that would change the very
allows cells to reproduce and ulti
nature of the organism.
During this period of intense learn mately allows you to reproduce. For
ing, I began reading the classic text a strict reductionist, the replication of
The Molecular Biology of the Gene, DNA by DNA polymerase is what life
co-au thored by James D. Watson of is all about.
DNA polymerase is an amazing
Watson Crick fame. My concept of
little nanomachine, a single
biology was being dramatically
transformed. Biolo gy was no longer molecule that
the science of things that smelled

54 Scientific American August 1998 Computing with DNA Copyright 1998 Scientific American, Inc.
DNA MOLECULES—with their
sequences of adenine, thymine, guanine
and cytosine (repre sented by the letters
A, T, G and C)—can be used to store
information and to perform com
 putations. The molecule shown here in
color,
GCAGTCGGACTGGGCTATGTCCGA, en
codes the solution to the sample
Hamiltonian Path Problem on the next
page.
1971; and “Randomness in Arithme “Jeez, these things could compute.”
tic,” by Gregory J. Chaitin; Scientific I did not sleep the rest of the night,
American, July 1988]. trying to figure out a way to get DNA
For his study, Turing had invented to solve problems.
a “toy” computer, now referred to as
a Turing machine. This device was
not intended to be real but rather to Copyright 1998 Scientific American, Inc.
be con ceptual, suitable for My initial thinking was to make a
mathematical in vestigation. For this DNA computer in the image of a Tur
purpose, it had to be extremely ing machine, with the finite control re
simple—and Turing suc ceeded placed by an enzyme. Remarkably,
brilliantly. One version of his es sentially the same idea had been
machine consisted of a pair of tapes sug gested almost a decade earlier
by Charles H. Bennet and Rolf
Computing with DNA Landauer of IBM [see “The
and a mechanism called a finite Fundamental Physical Limits of
control, which moved along the Computation”; Scientific Ameri can,
“input” tape reading data while July 1985]. Unfortunately, while an
simultaneously mov ing along the enzyme (DNA polymerase) was
“output” tape reading and writing known that would make
other data. The finite control was Watson-Crick comple ments, it
programmable with very simple in seemed unlikely that any exist ed for
structions, and one could easily other important roles, such as
write a program that would read a factoring numbers.
string of A, T, C and G on the input This brings up an important fact
tape and write the Watson-Crick about biotechnologists: we are a
“hops” onto a strand of DNA and
complementary string on the output com munity of thieves. We steal
slides along it, “reading” each base it
tape. The similarities with DNA from the cell. We are a long way
passes and “writing” its complement
polymerase could hardly have been from being able
onto a new, growing DNA strand.
more obvious.
While lying there admiring this
But there was one important piece
amazing enzyme, I was struck by its
of information that made this
similarity to something described in
similarity truly striking: Turing’s toy
1936 by Alan M. Turing, the famous
computer had turned out to be
British mathematician. Tur ing—and,
universal—simple as it was, it could
independently, Kurt Gödel, Alonzo
be programmed to compute
Church and S. C. Kleene—had
anything that was computable at all. A

begun a rigorous study of the notion M

(This notion is essentially the con I

of “computability.” This purely H

tent of the well-known “Church’s the S

theoreti cal work preceded the A

sis.”) In other words, one could R

advent of actual computers by about A

program a Turing machine to N

a decade and led to some of the

produce Watson Crick O

major mathematical results of the M

complementary strings, factor O

20th century [see “Unsolved T

numbers, play chess and so on. This

Problems in Arithmetic,” by Howard
re alization caused me to sit up in Scientific American August 1998 55
DeLong; Scientific American, March
bed and remark to my wife, Lori,
Hamiltonian Path Problem versa. The goal is to determine whether a path
exists that will

C onsider a map of cities connected by

commence at the
start city (Atlanta),
finish at the end
certain nonstop city (De
flights (top right). For instance, in the example
shown here,
it is possible to travel directly from Boston to
Detroit but not vice Detroit
 be defined by concatenating the last
name of the city of origin with the first Boston
name of the city of destination (bottom
troit) and pass through each of the right).
remaining cities exactly once. In DNA The complementary DNA city names
computation, each city is assigned a are the Watson-Crick complements of
DNA sequence (ACTTGCAG for the DNA city names in which every C is
Atlanta) that can be thought of as a first replaced Atlanta
name (ACTT) followed by a last name
(GCAG). DNA flight numbers can then
Chicago
by a G, every G by a C, Boston, Chicago and Detroit BOSTON - DETROIT
CITY
every A by a T, and every T in that order. In the BOSTON - ATLANTA
ATLANTA BOSTON
by an A. (To sim plify the computation, this path is CHICAGO - DETROIT
CHICAGO DETROIT
discussion here, details of represented by GCAGTCG
DNA NAME ACTTGCAG
the 3′ versus 5′ ends of the GACTGGGCTATGTCCGA, DNA FLIGHT NUMBER
TCGGACTG GGCTATGT
DNA molecules have been a DNA se quence of length GCAGTCGG
CCGAGCAA
omitted.) For this particular 24. Shown at the left is the GCAGCCGA
problem, only one COMPLEMENT TGAACGTC map with seven cities and ACTGGGCT
Hamiltonian path exists, AGCC TGAC CCGATACA 14 nonstop flights used in ACTGCCGA
GGCTCGTT the actual experiment. ACTGACTT S

—L.M.A. M

L
I

FLIGHT ATGTCCGA
END
and it passes through ATLANTA - BOSTON M
I

Atlanta, ATLANTA - DETROIT L

START S

to create de novo miraculous
molecular machines such as DNA
polymerase. Fortunately, three or
four billion years of evolution have
resulted in cells that are full of
wondrous little machines. It is these
machines, stolen from the cell, that
make modern biotechnology possi
ble. But a molecular machine that
would play chess has apparently
never evolved. So if I were to build a
DNA computer that could do
something interesting, I would have
to do it with the tools that
were at hand. These tools were
essen tially the following:
BOSTON - CHICAGO
in solution meets its Watson-Crick copy in formation from one molecule
complement, then the two strands into an other. For example, DNA
will anneal—that is, twist around polymerase will make a
each other to form the famous Watson-Crick complemen tary DNA
double helix. The strands are not strand from a DNA template. In fact,
cova lently bound but are held DNA polymerase needs a “start
together by
weak forces such as hydrogen
CHICAGO TO DETROIT
bonds. If a molecule of DNA in
solution meets a DNA molecule to
which it is not com plementary (and
has no long stretches of
complementarity), then the two mol
ecules will not anneal.
2. Polymerases. Polymerases

1. Watson-Crick pairing. As
stated earlier, every strand of DNA
has its Wat son-Crick complement.
As it happens, if a molecule of DNA

G
C
A
WATSON-CRICK
ANNEALING,
G

ATLANTA TO BOSTONBOSTON TO
CHICAGO
C

A
C

C
G
 G
in which Cs pair with Gs G
and As join with Ts, will
result in DNA C A
G A
flight-number
T A
A
C T
C
T T
strands (shown here: At
lanta to Boston, Boston
to Chicago, and Chicago
to Detroit) being held end
C C
G G
T
A
to-end by strands encoding the
complementary DNA city names
(shown here: Boston and Chicago).
56 Scientific American August 1998 Computing with DNA Copyright 1998 Scientific American, Inc.
signal” to tell it where to begin
making the complementary copy.
This signal is provided by a
primer—a (possibly short) piece of
DNA that is annealed to the
template by Watson-Crick
complemen tarity. Wherever such a
primer-template pair is found, DNA
polymerase will be gin adding bases
to the primer to create a
complementary copy of the
template.
3. Ligases. Ligases bind
molecules to gether. For example,
DNA ligase will take two strands of
DNA in proximity and covalently
bond them into a single strand. DNA
ligase is used by the cell to repair
breaks in DNA strands that oc cur,
for instance, after skin cells are ex
posed to ultraviolet light.
4. Nucleases. Nucleases cut
nucleic acids. For example,
restriction endonu cleases will
“search” a strand of DNA for a
predetermined sequence of bases
and, when found, will cut the
molecule into two pieces. EcoRI
(from Escheri chia coli) is a
restriction enzyme that will cut DNA
after the G in the sequence
GAATTC—it will almost never cut a
strand of DNA anywhere else. It has
been suggested that restriction
enzymes evolved to protect bacteria
from viruses (yes, even bacteria
have viruses!). For example, E. coli
has a means (methyla tion) of
protecting its own DNA from EcoRI,
but an invading virus with the deadly
GAATTC sequence will be cut to
pieces. My DNA computer did not
G
G
C
A
T
A
G
G
T
CC
COMPLEMENT OF CHICAGO
COMPLEMENT OF BOSTON
use restriction enzymes, but they necessary—a method of storing
have been used in subsequent information and a few simple
experiments by many other research operations for acting on that informa
groups. tion. The Turing machine stores infor
5. Gel electrophoresis. This is mation as sequences of letters on
not stolen from the cell. A solution oftape and manipulates that
het erogeneous DNA molecules is information with the simple
placed in one end of a slab of gel, instructions in the finite con trol. An
and a current is applied. The electronic computer stores in
negatively charged DNA molecules formation as sequences of zeros
move toward the anode, with and ones in memory and
shorter strands moving more quickly manipulates that information with the
than longer ones. Hence, this operations avail able on the
process separates DNA by length. processor chip. The remark able
With special chemicals thing is that just about any method
and
of storing information and any set of
ultraviolet light, it is pos sible to see
bands in the gel where the DNA operations to act on that information
molecules of various lengths have are good enough.
come to rest. Good enough for what? For univer
6. DNA synthesis. It is now sal computation—computing
possible to write a DNA sequence anything
on a piece of paper, send it to a that can be computed. To get your
commercial synthesis facility and in com puter to make Watson-Crick
a few days receive a test tube comple ments or to play chess, you
containing approximately need only start with the correct input
1018
molecules of DNA, all (or at least information and apply the right
most) of which have the described sequence of opera tions—that is,
sequence. Currently sequences of run a program. DNA is a great way
length approxi mately 100 can be to store information. In fact, the cell
reliably handled in this manner. For has been using this method to store
a sequence of length 20, the cost is the “blueprint for life” for billions of
about $25. The molecules are years. Further, enzymes such as
delivered dry in a small tube and ap poly merases and ligases have
pear as a small, white, amorphous been used to operate on this
lump. information. Was there enough to
build a universal computer?
None of these appeared likely to Because of the lessons of the
help play chess, but there was 1930s, I was sure that the answer
another im portant fact that the great was yes.
logicians of the 1930s taught us:
computation is easy. To build a The Hamiltonian Path Problem
computer, only two things are really
 map (graph). For example, you can
T he next task was choosing a fly nonstop from Boston to Chicago
but not from Chicago to Bos ton. GC
prob Your job (the Hamiltonian Path A
lem to solve. It should not
Problem) is to determine if a
appear to be contrived to fit the G
sequence of connecting flights (a
machine, and it should demonstrate
path) exists that starts in Atlanta T
the potential of this novel method of
(the start vertex) and ends in Detroit C
computation. The problem I chose G
(the end vertex), while passing C
was the Hamiltonian Path Problem.
through each of the remaining cities
William Rowan Hamilton was As
(Boston and Chicago) exactly once. T
tronomer Royal of Ireland in the mid G
Such a path is called a Hamilton ian C
19th century. The problem that has
path. In the example shown on C
come to bear his name is illustrated
in the box on the opposite page. Let
the arrows (directed edges)
represent the nonstop flights C
A
between the cities (ver tices) in the
A T
A C
G
GG

LIGASES connect A
T T
A A A
C C
G
G
the splinted mole cules. T
G Wherever the A
protein finds two strands of H

C S

DNA in proximity, it will A

covalently bond them into a A

single strand. O

C A
M

CG T
LIGASE
O

G M
I

Computing with DNA Scientific American August 1998 57 Copyright 1998 Scientific American, Inc.
page 56, it is easy to see that a
unique Hamiltonian path exists, and
it passes through the cities in this
order: Atlanta, Boston, Chicago,
Detroit. If the start city were
changed to Detroit and the end city
to Atlanta, then clearly there would
be no Hamiltonian path.
More generally, given a graph with
directed edges and a specified start
ver tex and end vertex, one says
there is a Hamiltonian path if and
only if there is a path that starts at
the start vertex, ends at the end
vertex and passes though each
remaining vertex exactly once. The
Hamiltonian Path Problem is to
decide for any given graph with
specified start and end vertices
whether a Hamiltoni an path exists
or not.
The Hamiltonian Path Problem has
been extensively studied by
computer scientists. No efficient
(that is, fast) al gorithm to solve it
has ever emerged. In fact, it seems
likely that even using the best
currently available algorithms and
computers, there are some graphs
of fewer than 100 vertices for which
de termining whether a Hamiltonian
path exists would require hundreds
of years. In the early 1970s the
Hamiltonian
C vertex. If not, remove
T
Path Problem was shown to be
“NP complete.” Without going into
the the ory of NP-completeness,
suffice it to say that this finding
convinced most theoretical
computer scientists that no
efficient algorithm for the problem
is possible at all (though proving
this re mains the most important
open prob lem in theoretical
computer science, the so-called
NP = P? problem [see “Turing
Machines,” by John E. Hopcroft;
Scien tific American, May 1984]).
This is not to say that no
algorithms exist for the
Hamiltonian Path Problem, just
no efficient ones. For example,
consider the following algorithm:
Given a graph with n vertices,
1. Generate a set of random
paths through the graph.
2. For each path in the set:
a. Check whether that path
starts at the start vertex
and ends with the end
that path from the set.
b. Check if that path passes
through exactly n vertices.
If not, remove that path
 that path passes through
that vertex. If not, remove
that path from the set.
3. If the set is not empty, then
report that there is a Hamil
tonian path. If the set is
empty, report that there
is no Hamiltonian path.

This is not a perfect algorithm; never

thele
ss, if
the
gene
GG ratio
n of
paths
is ran
dom
enou Seven Days in a Lab
gh
and
the
F o

result r my
ing
expe
set
large rimen
enou
gh, t, I
then
there soug
is a
ht a
high
prob Ham
abilit
y that iltoni
it will an
give Path
the Probl
corre em
ct small
answ enou
er. It gh to
is be
this solve
algori d
thm readil
that I y in
imple the
ment lab,
ed in yet
the large
first enou
DNA gh to
comp provi
utatio de a
n. clear
“proo
f of
princi
ple”
for
AA
DNA
from the set. comp
c. For each vertex, check if utatio
 n. I
chos T
e the T
seve
n-city C
, A
14-fli
ght G
map
G
show
G
n in
the
inset
on
page
56. A
nons
cienti
fic
study
has
show
n
that it
takes
abou
t 54

DNA POLYMERASE

A
G
C G
C C
C
T
example, a primer—GGCT, representing the complement
of the first name of Detroit—is annealed to the right end
of a DNA strand. The primer signals the polymerase to
begin making a Watson-Crick complement of the strand.
A
After the polymerase is done, the double helix is split
into two strands so that Watson-Crick comple ments of
each half can be made. This process is repeated to
GC obtain a large number of copies of mole cules that have
POLYMERASE CHAIN REACTION, or PCR, is used to the correct start and end cities. Gel electrophoresis is
replicate DNA molecules that begin with the start city then used to isolate those molecules that have the right
(Atlanta) and terminate with the end city (Detroit). In this sequence length of 24.

58 Scientific American August 1998 Computing with DNA Copyright 1998 Scientific American, Inc.

G
seconds on average to find the
unique Hamiltonian path in this
graph. (You may begin now....)
To simplify the discussion here,
con sider the map on page 56,
which con tains just four
cities—Atlanta, Boston,
Chicago and Detroit—linked by
six flights. The problem is to
determine the existence of a
Hamiltonian path start ing in
Atlanta and ending in Detroit.
I began by assigning a random
DNA sequence to each city. In
our example, Atlanta becomes
ACTTGCAG, Boston
TCGGACTG and so on. It was
conve nient to think of the first
half of the DNA sequence as
the first name of the city and
the second half as the last
name. So Atlanta’s last name is
GCAG, whereas Boston’s first
name is TCGG. Next, I gave
each nonstop flight a DNA
“flight number,” obtained by
concate nating the last name of
the city of origin with the first
name of the city of desti nation. G
In the example on page 56, the
Atlanta-to-Boston flight number
be comes GCAGTCGG.
Recall that each strand of DNA
has its Watson-Crick
complement. Thus, each city
has its complementary DNA
C
G
C
G
PRIMER
C
name. Atlanta’s complementary
name becomes, for instance,
TGAACGTC. After working out
these encodings, I had the
complementary DNA city
names and the DNA flight
numbers synthesized. (As it
turned out, the DNA city names
themselves were largely
unnecessary.) I took a pinch
T (about 1014 molecules) of each
T of the different sequences and
put them into a common test (TGAC) is complementary to and it goes through Atlanta,
tube. To be gin the the beginning of the latter Boston, Chicago and Detroit, in
computation, I simply added (ACTG). In this manner, that order. Thus, the mole cule
water—plus ligase, salt and a complexes will grow in length, encoding the solution will have
few other ingredients to with DNA flight numbers splint the sequence
approximate the condi tions ed together by complementary GCAGTCGGACTGGGCT
inside a cell. Altogether only DNA city names. The ligase in ATGTCCGA.
about one fiftieth of a teaspoon the mixture will then Unfortunately, although I held
of solution was used. Within permanently concatenate the the so lution in my hand, I also
about one second, I held the chains of DNA flight numbers. held about 100 trillion
answer to the Hamiltonian Path Hence, the test tube contains molecules that encoded paths
Problem in my hand. molecules that encode random that were not Hamiltonian.
To see how, consider what paths through the different These had to be eliminated. To
transpires in the tube. For cities (as required in the first weed out molecules that did not
example, the Atlanta-to Boston step of the algorithm). both begin with the start city
flight number (GCAGTCGG) Because I began with such a and terminate with the end city,
and the complementary name large number of DNA I relied on the polymerase
of Boston (AGCCTGAC) might molecules and the chain reaction (PCR). This
meet by chance. By design, the problem contained just a important technique re quires
former sequence ends handful of cities, there was a many copies of two short
virtual certainty that at least pieces of DNA as primers to
one of the molecules formed signal the DNA polymerase to
would encode the Hamiltonian start its Watson-Crick
path. It was amazing to think replication. The primers used
that the solution to a were the last name of the start
mathematical problem could be city (GCAG for
stored in a single molecule!
Notice also that all the paths
were created at once by the
simultaneous in teractions of
literally hundreds of tril lions of
molecules. This biochemical re
action represents enormous
parallel processing.
A

M
I

with TCGG, and the latter starts

A

with AGCC. Because these

A

sequences are complementary, O

they will stick togeth er. If the

M

resulting complex now en

T

counters the Boston-to-Chicago

flight number (ACTGGGCT), it,
too, will join the complex For the map on page 56, there
because the end of the former is only one Hamiltonian path,

Computing with DNA Scientific American August 1998 59 Copyright 1998 Scientific American, Inc.

IRON BALL

G
T
C

C
G

T G C
G
C
G G
T
A

G
C
 A
A
T
A
A

C
G

C
PROBE
C MOLECULES
A
A

M
I

C
H

T
C

T
C
C
GG
G
T

A
in liquid. Because of Watson-Crick affinity, the probes
capture DNA strands that contain Boston’s name
(TCGGACTG). Strands missing Boston’s name are
then discarded. The process is repeated with probe
molecules encoding the complementary DNA name of
Chicago. When all the computational steps are
completed, the strands left will be those that encode
the solution GCAGTCGGACTGGGCTATGTCCGA.

PROBE MOLECULES are used to locate DNA strands

encoding paths that pass through the intermediate right length (in the exam ple on page 56, a length of
cities (Boston and Chicago). Probe molecules 24). All other molecules were discarded. This
containing the complementary DNA name of Boston completed step 2b of the algorithm.
(AGCCTGAC) are attached to an iron ball sus pended To check the remaining se
Atlanta) and the Watson-Crick The result was that molecules with the
comple ment of the first name of the both the right start and end cities quences for whether their paths
end city (GGCT for Detroit). These were reproduced at an exponential passed through all the intermediary
two prim ers worked in concert: the rate. In contrast, molecules that cities, I took advantage of Watson
first alerted DNA polymerase to encoded the right start city but an Crick annealing in a procedure
copy complements of sequences incorrect end city, or vice versa, called affinity separation. This
that had the right start city, and the were duplicated in a much slower, process uses multiple copies of a
second initiated the duplication of linear fashion. DNA sequences that DNA “probe” mol ecule that encodes
molecules that encoded the correct had neither the right start nor end the complementary name of a
end city. were not duplicated at all. Thus, by particular city (for example, Boston).
PCR proceeds through tak ing a small amount of the These probes are attached to
thermocycling, repeatedly raising mixture after the PCR was microscopic iron balls, each approxi
and lowering the tem perature of the completed, I obtained a solution mately one micron in diameter.
mixture in the test tube. Warm containing many copies of the I suspended the balls in the tube
conditions encourage the DNA molecules that had both the right con taining the remaining molecules
polymerase to begin duplicating; a start and end cities, but few if any under conditions that encouraged
hot ter environment causes the molecules that did not meet this Watson Crick pairing. Only those
resulting an nealed strands to split criterion. Thus, step 2a of the molecules that contained the
from their dou ble-helix structure, algorithm was complete. desired city’s name (Bos ton) would
enabling subsequent replication of Next, I used gel electrophoresis to anneal to the probes. Then I placed
the individual pieces. identify those molecules that had a magnet against the wall of the

60 Scientific American August 1998 Computing with DNA Copyright 1998 Scientific American, Inc.
 molecules to break free Molecular computers Re bek, Jr., of the
from the probes and G also have the potential Massachusetts Institute
redissolve in the liquid. CC for extraordinary energy of Technology creates
Next, I reapplied the cules left in the tube effi ciency. In principle, molecules that can
magnet to attract the should be precisely one joule is sufficient for reproduce—informing us
balls again to the side of those encoding approximately 2 × 1019 about how life on the
the test tube, but this Hamiltonian paths. ligation op erations. This earth may have arisen
time without any Hence, if the tube is remarkable consider [see “Synthetic
molecules attached. The contained any DNA at ing that the second law Self-Replicating
liquid, which now all, I could conclude that of thermody namics Molecules,” by Julius
contained the desired a Hamilto nian path dictates a theoretical Rebek, Jr.; Scientific
DNA strands (in the existed in the graph. No maximum of 34 × 1019 Ameri can, July 1994].
example, encoding paths DNA would indicate that Stimulated by research
(irreversible) operations
that went through no such path existed. on DNA computation,
per joule (at room
Boston), could then be Fortunately, to make this Erik Winfree of the
temperature). Existing
poured into a new tube deter mination I could California Institute of
supercomputers are far
for further screening. Theuse an additional PCR Technology synthesizes
less efficient, ex ecuting
process was repeated forstep, followed by another 9 “intelligent” molecular
the remaining at most 10 operations com plexes that can be
gel-elec trophoresis
intermediary cities (Chi operation. To my delight, per joule. “programmed” to
cago, in this case). This the final analysis Experimental and assemble themselves
iterative proce dure, theoretical scien tists into predetermined
revealed that the mol
which took me an entire ecules that remained did around the world are structures of arbitrary
day to complete in the indeed encode the working to exploit these complexity. There are
lab, was the most te properties. Will they suc many other examples. It
desired Hamiltonian
dious part of the that can compete with is the enor mous
path. After seven days in
T experiment. electronic com puters? potential of this new area
the lab, the first DNA
At the conclusion of the computation was That remains to be seen. that we should focus on
affinity sepa rations, step complete. Huge financial and and nurture.
2c of the algorithm was intellectual investments For me, it is enough just
over, and I knew that the A New Field Emerges over half a century have to know that computation
DNA mole made electron ic with DNA is possible. In
computers the marvels ofthe past half-century,
T
C W hat about the our age— they will be
hard to beat.
biology and com puter
science have blossomed,
future? It is clear But it would be and there can be little
shortsighted to view this doubt that they will be
that molecular computers
research only in such central to our scientific
have many attractive
practical terms. My and econom ic progress
properties. They pro vide
experiment can be in the new millennium.
extremely dense
viewed as a But biology and
information stor age. For
manifestation of an computer science—life
example, one gram of
emerging new area of and computation—are
DNA, which when dry
science made possible related. I am confi dent
would occupy a volume
by our rapidly developing that at their interface
of approximately one
ability to control the great discov
cubic centimeter, can
molecular world.
store as much
Evidence of this new
information as ap
“molecular science” can
proximately one trillion
be found in many places.
CDs. They pro vide
For example, Gerald F.
G enormous parallelism.
Joyce of Scripps
test tube to attract and Even in the tiny
Research Institute in La
hold the metal balls to experiment carried out in
Jolla, Calif., “breeds”
the side while I poured one fifti eth of a teaspoon
trillions of RNA
out the liquid phase of solution, approxi
14 molecules, generation
containing molecules mately 10 DNA flight after gener ation, until
that did not have the numbers were “champion” molecules
desired city’s name. simultaneously evolve that have the
I then added new solvent concatenated in about catalytic properties he
and removed the magnet one second. It is not seeks [see “Directed
in order to resuspend the clear whether the fastest Molecular Evo lution,” by
balls. Raising the supercomputer available Gerald F. Joyce;
temperature of the today could accomplish Scientific American,
mixture caused the such a task so quickly. December 1992]. Julius
A
 ceed in creating molecular computers
The Author
LEONARD M. ADLEMAN received a Ph.D. in computer sci
ence in 1976 from the University of California, Berkeley. In
1977 he joined the faculty in the mathematics department at
the Mas sachusetts Institute of Technology, where he
specialized in algorith mic number theory and was one of the
inventors of the RSA public key cryptosystem. (The “A” in RSA
stands for “Adleman.”) Soon after joining the computer
science faculty at the University of South ern California, he
was “implicated” in the emergence of computer viruses. He is
a member of the National Academy of Engineering.
Computing with DNA Scientific American August 1998 61 Copyright 1998 Scientific American, Inc.
eries await those who seek them. SA
Further Reading
Molecular Computation of Solutions to Combinatorial Problems.
Leonard M. Adleman in Science, Vol. 266, pages 1021–1024;
November 11, 1994.
On the Path to Computation with DNA. David K. Gifford in
Science, Vol. 266, pages 993–994; November 11, 1994. DNA
Solution of Hard Computational Problems. Richard J. Lipton in
Science, Vol. 268, pages 542–545; April 28, 1995. Additional
information on DNA computing can be found at http://
users.aol.com/ibrandt/dna_computer.html on the World Wide
Web.