DNA Computers for Work and Play

information sc i enc e COMPUTERS DNA FOR WORK AND PL AY

TIC-TAC-TOE-PLAYING COMPUTER consisting of DNA strands in solution demonstrates the potential of molecular logic gates.

© 2008 SCIENTIFIC AMERICAN, INC. Logic gates made of DNA could one day operate in your bloodstream, collectively making medical decisions and taking action. For now, they play a mean game of in vitro tic-tac-toe

By Joanne Macdonald, Darko Stefanovic and Milan N. Stojanovic

rom a modern chemist’s perspective, the mentary school in Belgrade, Serbia, we hap- structure of DNA in our genes is rather pened to be having dinner, and, encouraged by Fmundane. The molecule has a well-known some wine, we considered several topics, includ- importance for life, but chemists often see only ing bioinformatics and various existing ways of a uniform double helix with almost no function- using DNA to perform computations. We decid- al behavior on its own. It may come as a surprise, ed to develop a new method to employ molecules then, to learn that this molecule is the basis of a to compute and make decisions on their own. truly rich and strange research area that bridges We planned to borrow an approach from synthetic chemistry, enzymology, structural electrical engineering and create a set of molec- nanotechnology and computer science. ular modules, or primitives, that would perform Using this new science, we have constructed elementary computing operations. In electrical molecular versions of logic gates that can oper- engineering the computing primitives are called ate in water solution. Our goal in building these logic gates, with intuitive names such as AND, DNA-based computing modules is to develop OR and NOT. These gates receive incoming nanoscopic machines that could exist in living electrical signals that represent the 0s and 1s of organisms, sensing conditions and making deci- binary code and perform logic operations to sions based on what they sense, then responding produce outgoing electrical signals. For instance, with actions such as releasing medicine or kill- an AND gate produces an output 1 only if its ing specific cells. two incoming inputs are both 1. Modern-day We have demonstrated some of the abilities of computers have hundreds of millions of such KEY CONCEPTS our DNA gates by building automata that play logic gates connected into very complex circuits, n DNA molecules can act perfect games of tic-tac-toe. The human player like elaborate structures built out of just a few as elementary logic gates adds solutions of DNA strands to signal his or kinds of Lego blocks. Similarly, we hoped that analogous to the silicon- her moves, and the DNA computer responds by our molecular modules could be mixed together based gates of ordinary lighting up the square it has chosen to take next. into increasingly complex computing devices. computers. Short strands Any mistake by the human player will be pun- We did not aim, however, to compete with sil- of DNA serve as the gates’ ished with defeat. Although game playing is a icon-based computers. Instead, because Sto- inputs and outputs. long way from our ultimate goals, it is a good janovic had just finished a brief stint with a phar- n Ultimately, such gates test of how readily the elementary molecular maceutical company, we settled on developing a could serve as dissolved computing modules can be combined in plug- system that could be useful for making “smart” “doctors”—sensing mole- and-play fashion to perform complicated func- therapeutic agents, such as drugs that could sense cules such as markers on tions, just as the silicon-based gates in modern and analyze conditions in a patient and respond cells and jointly choosing computers can be wired up to form the complex appropriately with no human intervention after how to respond. logic circuits that carry out everything that com- being injected. For example, one such smart n Automata built from these puters do for us today. agent might monitor glucose levels in the blood DNA gates demonstrate and decide when to release insulin. Thus, our the system’s computation- Dissolved Doctors molecular logic gates had to be biocompatible. al abilities by playing Near the end of 1997 two of us (Stojanovic and Such molecular modules could have innumer- an unbeatable game of Stefanovic) decided to combine our individual able functions. For instance, in diseases such as tic-tac-toe.

skills in chemistry and computer science and leukemia, numerous subpopulations of white —The Editors

jean-francois podevinjean-francois work on a project together. As friends from ele- blood cells in the immune system display char-

www.SciAm.com © 2008 SCIENTIFIC AMERICAN, INC. SCIENTIFIC AMERICAN 85 acteristic markers on their cell surfaces, depend- very analogous to the workings of silicon logic ing on the cells’ lineage and their stage of devel- other dna gates. Nevertheless, DNA clearly had a lot of po- opment. Present-day therapies using antibodies computers tential for biocompatible computation, and a eliminate large numbers of these subpopulations couple of other advances gave us the tools to in- Researchers over the years at once, because they target only one of the sur- have devised several ways to vent our own brand of DNA logic gates. face markers. Such indiscriminate attacks can perform computations by First, in 1995 Gerald F. Joyce of the Scripps suppress the patient’s immune system by wiping exploiting DNA’s ability Research Institute in La Jolla, Calif., developed out too many healthy cells, leading to serious to store information in its a method for producing enzymes made out of sin- complications and even death. Molecular mod- sequence of bases. gle strands of DNA that cut other pieces of sin- ules capable of working together to sense and gle-stranded DNA into two segments. These so- 1994: Leonard M. Adleman — analyze multiple markers including perform- of the University of Southern called deoxyribozymes have two short arms that ing logical operations such as “markers A and California solved a puzzle will bind only to another stretch of DNA that has either B or C are present, but D is absent”— known as the Hamiltonian path the correct complementary sequence of bases, so might be able to select the specific subpopula- problem by encoding all the they are very specific about which substrate DNA tions of cells that are diseased and growing out possible solutions (both correct strands they will cleave [see box on page 88]. and incorrect) on a large num- of control and then eliminate only those cells. ber of DNA molecules and Special dye molecules attached to each end of Another application of our modules could be carrying out a series of steps to the substrate strands enable laboratory workers in the analysis of DNA, looking for a large array isolate the molecules with the to monitor the cleaving process. At one end of of possible genetic mutations or identifying one correct solution [see “Comput- the substrate, the dye molecule is a “quencher,” of a wide variety of microbiological pathogens. ing with DNA,” by Leonard M. which prevents the fluorescent marker dye at the Adleman; Scientific Am e r i c a n , Our most advanced tic-tac-toe-playing automa- August 1998]. other end from fluorescing as long as the strand ton combines 32 different short DNA sequences remains intact, keeping the quencher close (oligonucleotides). That many logic gate inputs 1995: Erik Winfree, now at the enough to be effective. After the strand is cut, its could analyze four billion possible combinations California Institute of Technolo- two pieces move apart and the marker dye mol- of oligonucleotides and partition them into thou- gy, proposed that tiles made ecule can fluoresce unhindered. As the work of sands of patterns, each pattern being character- of DNA could be designed to the DNA enzymes progresses, cutting more and perform computations by self- istic of certain pathogens or genotypes. assembling into two-dimension- more strands, the solution gradually lights up al structures [see “Nano with the marker dye’s fluorescent color. Molecular Logic technology and the Double The other key advance came soon after our Researchers reported logic gates based on syn- Helix,” by Nadrian C. Seeman; initial planning, when Ronald R. Breaker of Yale thetic molecules as long ago as the early 1990s. Scientific Am e r i c a n , June 2004]. University reported a way to integrate a deoxyri- In 1993, for instance, A. Prasanna de Silva and bozyme with molecular groups acting as recog- 2004: Ehud Shapiro of the his collaborators at Queen’s University Belfast Weizmann Institute of Science nition modules. These modules work like sensors made AND gates out of small organic molecules in Rehovot, Israel, and Yaakov that either activate or inhibit their attached DNA that would fluoresce only if both hydrogen ions Benenson of Harvard Universi- enzyme when the correct input molecule is bound (from acid) and sodium ions were bound to ty, building on a proposal by to them. Breaker even combined two such mod- them. In 1997 J. Fraser Stoddart, now at North- Paul W. K. Rothemund of ules in a construct that could serve as an AND Caltech, developed a “doctor in western University, and his co-workers made a cell.” Enzymes operating on gate with two small input molecules. Very in- “exclusive OR” (XOR) gates, in which the mol- DNA analyzed whether a com- triguingly, his group has found that such two- ecules fluoresced in the presence of either, but bination of RNA molecules sensor constructs have been used by natural ri- not both, of the inputs (in this case, hydrogen indicative of a disease was boswitches—molecules made of RNA used by ions and molecules called amines). These exam- present in the solution and bacteria to control which of their genes actively responded by releasing another ples, however, were not biocompatible, because molecule as a model for a drug produce proteins [see “The Power of Riboswitch- they required concentrations of acid and other [see “Bringing DNA Computers es,” by Jeffrey E. Barrick and Ronald R. Breaker; compounds that would harm living cells. to Life,” by Ehud Shapiro and Scientific American,January 2007]. In the mid-1990s other researchers exploited Yaakov Benenson; Scientific We saw that we could build our logic gates out DNA’s ability to store information in its se- Am e r i c a n , May 2006]. of DNA enzymes integrated with controlling sen- quence of bases—the molecules conventionally sor modules designed to recognize short DNA abbreviated as A, T, G and C, which pair up to strands having specific base sequences. The DNA form the rungs connecting the two strands of the strands would thus act as inputs to the logic gates famous double-helix structure. Their tech- (an input of 1 if the strand is present; 0 if it is ab- niques, however, were very different from the sent), and the gates’ enzymes would output “1” kind of system we envisaged, namely, one in by cleaving other DNA strands in the solution. which molecular logic gates floating in solution With DNA serving as both inputs and outputs, would process inputs and outputs in a fashion our gates could in principle be chained together

86 SCIENTIFIC AMERICAN © 2008 SCIENTIFIC AMERICAN, INC. November 2008 [applications] jobs for intelligent DNA DNA logic gates could have many applications, ranging from medical treatments to counterterrorism.

injectable pancreas targeted treatment counterterrorism Logic gates operating in the bloodstream of Gates that sense different markers on white blood DNA-based chemical sensors, along with DNA logic a diabetic patient could monitor glucose levels cells and combine their data could target leukemia gates, could sniff out previously unknown nerve and release insulin when appropriate. cells for destruction while sparing healthy cells that agents such as Soviet-made “novichok” chemicals may have some but not all the same markers. as well as more familiar ones such as sarin.

to form complex circuits. Like wires in electrical stem will block the enzyme’s activity. We call circuits, the base sequences of the sensors and the this structure a sensor or a YES gate because enzymes would control which gates’ outputs adding the input strand (say, “input X”) for the [THE AUTHORs] “connected” to which inputs, even as all the gates stem-loop controller opens the stem, exposing Joanne Macdonald, Darko Ste- sloshed around independently in a test tube. the enzyme’s substrate-matching region and al- fanovic and Milan N. Stojanovic After some less than successful attempts us- lowing it to function. The enzyme’s output (spe- bring very different backgrounds ing other designs, we settled on DNA structures cific cleaved strands of DNA) in essence says, to the task of programming DNA known as stem-loops for our recognition mod- “YES, input X is present.” to compute. Macdonald is an as ules. Sanjay Tyagi and Fred Kramer, both at the Adding a second stem-loop with a different sociate research scientist at Public Health Research Institute in Newark, loop sequence (Y) on the other of the enzyme’s Columbia University. She conducts biology-related research within N.J., had reported that stem-loops switch be- two arms yields an AND gate. Only if input X the division of clinical pharmacolo- tween two shapes, or conformations. In the AND input Y bind to it can the enzyme function gy and experimental therapeutics, closed conformation the DNA strand making up and cleave DNA [see box on page 89]. and pursues practical applications the stem-loop folds onto itself, and the two ends We make an inhibitory controller—one that of DNA computing for viral detec- zip together, forming a stem along with a loop of will deactivate the enzyme when the correct in- tion. Stefanovic is an associate professor of computer science at unzipped DNA, like the outline of a lollipop. An put binds to the loop—by plugging a stem-loop the University of New Mexico input DNA strand consisting of the sequence of sequence into the “back” of the enzyme. Now working on algorithms for memory bases complementary to the loop will bind to it, when the stem is closed, the enzyme is intact and management in computers. He is but in forming a stretch of the familiar double produces output. The relevant input strand will the recipient of a U.S. National Sci- helix it pries the stem apart—the double-helical open the stem-loop and deform the enzyme ence Foundation (NSF) CAREER award. Stojanovic is associate DNA cannot form a tight enough curve to main- enough to inactivate it. Of course, this inactiva- director of the division of clinical tain the closed loop. tion will not remove output strands already pro- pharmacology and experimental Depending on how we attach a stem-loop to duced by the gate, so in isolation this NOT gate therapeutics at Columbia and direc- a DNA enzyme, opening the loop may either ac- does not function as conveniently as an electron- tor of the NSF Center for Molecular tivate or inhibit the enzyme’s activity. If one of ic NOT gate. But the NOT unit comes into its Cybernetics. He is a Leukemia & Lymphoma Society Fellow. MAYA is the enzyme’s two substrate-matching arms own when combined with the AND gate struc-

jean-francois podevinjean-francois named after his daughter. serves as one side of the stem, then the closed ture. The resulting gate, which we call AND-

www.SciAm.com © 2008 SCIENTIFIC AMERICAN, INC. SCIENTIFIC AMERICAN 87 [WORKING PARTS] AND-NOT, produces output only if inputs X Molecular Modules AND Y AND NOT Z are present. That function, also known as an INHIBIT gate, turned out To perform as logic gates analogous to silicon ones, a technology must produce specific to be very useful for our tic-tac-toe automata. outputs in response to a variety of inputs. DNA enzymes and recognition modules The most important aspect of our system is provide these output and input functions for a system based on DNA in solution. that it is highly modular. We can use hundreds A and theoretically millions of different base se- A G quences for the inputs, and we can also change Output ENZYME G deoxyribozyme C A DNA enzyme called a de- C the sequences of the output strands. We could G oxyribozyme (top) consists of A C C even switch the underlying enzyme to be a ligase, single-stranded DNA folded C G into a “core” structure with C T one that joins together short strands to produce C A arms at each end that can Enzyme longer ones. Indeed, Andrew D. Ellington’s bind to a substrate DNA A core G strand that has the comple- Substrate T C group at the University of Texas at Austin has mentary sequences of bases binding G region G studied ligase-based switches extensively. separated by a specific se- T A quence of three other bases The functioning of the gates is also autono- (dark gray). The enzyme ACG T G A T C T T C T mous. That is, once we trigger a computation by cleaves the strand into two adding the input to the solution, no more human pieces (bottom). The process T C A C T A T rA G G A A G A G can be monitored by attach- intervention is required. In essence, DNA mol- Fluorescent Substrate Quencher ing a fluorescent molecule group group ecules make the decisions on their own, based at one end of the substrate strands and a quencher mol- on whatever inputs they receive. ecule at the other end. The Our gates do have some significant differenc- quencher molecule blocks T C A C T A T rA es, however, from the silicon-based logic in elec- fluorescence until the cleav- trical circuits. First, we cannot reset our gates. ing of the strand takes it G G A A G A G out of range. Once an input strand is bound to a stem-loop Cleaved substrate controller, it tends to remain there for the rest of the computation. Nor can the cleaved oligonu- cleotide output strands be reassembled. Our ul- Input SenSOR In a DNA structure called a stem-loop, the DNA folds onto itself and zips together to form timate biomedical goals do not require a gate-re- a double-stranded stem with a single-stranded loop (left). When a matching input strand set function, but it would be useful for potential binds to the loop, it pries the stem apart (right). molecular robotics applications (involving mov- CLOSED Stem-loop OPEN Stem-loop ing parts). We are exploring the use of ligase enzymes to reassemble output strands. G T C A C G G G C Second, electronic gates have a threshold T C G T G G A G T C Input G AGAAGT C A T C voltage at which their switching happens, and C binding G G G A C their outputs are tied to specific voltages so that G A region C T CTTCA T T G AGAAGT C G G C they cannot linger at an intermediate voltage. T C G T A C G Input Thus, the 0s and 1s are well defined, and the log- A T C T A A ic is truly digital. Solutions of our gates, in con- C G C A C A trast, change in continuous fashion between the A C T C G T C T inactive and the fully active forms depending on T C how many inputs we add to the fluid. This be- A havior would be important if we were attempt- sensor gate ing to build the molecular equivalent of a per- A stem-loop attached to the arm of an enzyme blocks the enzyme’s function (1) until an input DNA sonal computer, but it does not matter for many strand opens the controller and exposes the arm (2), enabling the enzyme to bind and cleave substrates (3). This structure is also called a YES gate because it signals, “Yes, the input is present.” biomedical applications. DNA Plays Tic-Tac-Toe ●1 ●2 ●3 With a general approach to constructing molecular logic gates in our hands, we looked for an objective test of their ability to compute. We wanted to apply our logic gates in a situation in which everyone would immediately see that the molecules were making decisions. A traditional test for a new computer system is to make it play

a game of strategy. The rules of a game provide moorman tommy

88 SCIENTIFIC AMERICAN © 2008 SCIENTIFIC AMERICAN, INC. November 2008 [LOGIC GATES] a challenge with a straightforward measure of success: the system will either be able to play the HOW DNA COMPUTES game or not. Game-playing ability is intimately connected with general computational ability. Combining DNA enzymes with stem-loop controllers yields a variety of fundamental We chose the classic children’s game of tic- logic gates that use short strands of DNA as both inputs and outputs. The cleaving tac-toe for our demonstration. In this game, action of the enzyme produces the strands that serve as the gate’s output of 1. No played on a 3  3 grid, two players try to put cleaving is an output of 0. three marks in a row while blocking the opponent from doing the same. Tic-tac-toe is one of AND Gate the simplest two-player games of perfect infor- A logical AND gate has two inputs and produces an output of 1 only if both inputs are 1. A deoxy- mation, meaning that a player knows everything ribozyme with a stem-loop on each of its arms acts as an AND gate. The closed stems disable the enzyme (left), and only when both loops’ matching input strands are added can the enzyme that there is to know about the state of the game cleave substrates (middle). Truth table (right) summarizes the gate’s function. at each move (unlike, for instance, most card games, in which rivals’ cards are unknown). Tic- tac-toe will always end in a draw if both parties 0 0 0 play well, but our device will exploit any mistake the opponent makes. 1 0 The game is simple enough that we can en- 0 code all decision making into logic operations that examine only the opponent’s moves. That 0 1 is, when you are using a fixed strategy, even if 0 you remember only what your opponent’s moves have been, you can work out what your own past 1 1 1 moves must have been and therefore what the current board position is and what your strategy dictates as your next move. We condensed that chain of reasoning down to a network of logic gates that takes the opponent’s moves as inputs and produces your next move as the output. In AND-AND-NOT Gate A stem-loop controller on the “back” of a deoxyribozyme acts as a NOT 2002 we set out to build just such a network out 0 of DNA logic gates, a tic-tac-toe-playing autom- input that inhibits the enzyme when the matching input strand is present. If the stem-loop’s input strand is not present (0), the stem remains 0 0 0 aton that we christened MAYA (molecular array closed and the enzyme cleaves substrates to produce output strands, of YES and AND-AND-NOT gates). provided that the enzyme’s arms are free (left). When the input strand binds to the controller, the stem opens, deforming the enzyme core and 0 MAYA consists of nine wells corresponding rendering it inactive (middle). A deoxyribozyme with controllers on 1 0 0 to the squares of the tic-tac-toe grid. Each well both arms and its back thus behaves as an AND-AND-NOT gate. The contains its own precisely defined set of DNA enzyme is active, cleaving substrates and thus producing the 1 output, only if inputs X (blue) AND Y (purple) AND NOT Z (yellow) are present. 0 logic gates in solution [see box on next page]. 0 1 0 The enzymes of these gates are all designed to cleave the same substrate DNA strand, which is also in all the wells, but they require magnesium 0 1 1 ions to function. Thus, adding magnesium ions 1 stirs MAYA into action. Because the enzymes in the central well have no stem-loop controllers on 1 them, they start cleaving the substrate immedi- 0 0 0 ately. The fluorescence from the central well in- creases, signaling that M AYA has taken the cen- 1 tral square as the opening move. 1 0 0 The human (let’s call him Harry) has eight input strands (one for each of the eight remaining 1 squares) for inputting his moves. The base se- 0 1 0 quences of these strands are complementary to the sequences on the stem-loops that control 1 MAYA’s DNA gates. To move in square 4, for 1 1 0 instance, Harry adds input 4 to all nine of

tommy moorman tommy MAYA’s wells. MAYA signals its move in re-

The first-generation automaton, MAYA-I, proves the potential of DNA logic gates by playing a perfect game of tic-tac-toe, albeit with some restrictions to simplify its programming. MAYA plays first, selecting the central square (5), and the human player’s first move must be in either the upper left corner (square 1) or the left side (square 4).

12 3 2 1 MAYA-I’S STRUCTURE The computer’s 3  3 array of wells contains a variety 6779 of molecular gates in solution, along with substrate 4 1649 strands (not shown). In wells where any enzymes 1 become active, the cleaved substrates fluoresce red. 89 The “gate” in the central well is a DNA enzyme with no stem-loop controllers. 4 5 6 6 6 1213 6 6 17 EXAMPLE GAME 1 18 The human, “Harry,” adds magnesium 6 ions to all nine wells to switch MAYA on. 19 The enzymes in well 5 cleave substrate strands, and the well lights up, signaling MAYA’s opening move (X). 789 9 9

42 43 7 1 5 9 9 2629 4 Mg+ 46 47 7 7 79 9 4 6869 48 78

For his first move, Harry takes square 4. To block MAYA from taking the diagonal, Harry desperately tries blocking MAYA He tells MAYA by adding input strand 4 to Harry takes square 9 by inputting strand 9 by taking square 7. all the wells. to all the wells.

13 2 1 4 9 7 4 49 79

Input strand 4 activates the YES-4 gates in His two inputs activate the 4-AND-9 gates Unfortunately for Harry, his inputs now well 1, which lights up; MAYA has taken in well 3; MAYA takes that square. activate the 7-AND-9-AND-NOT-1 gate in square 1 for its second move. well 2 (he has not added strand 1), and MAYA takes that square to win.

sponse by turning on the fluorescence in another stricted Harry’s first move to be either the upper of the wells. left corner (square 1) or the left side (square 4). As the game progresses, each well contains in- Those two moves are representative of all the put strands representing all Harry’s moves, and moves that Harry might make in response to the combination of gates in each well processes MAYA’s opening move in the center because the those inputs. After every move, one of the wells board is symmetric. If he moved somewhere else, contains a gate that the last input triggers in com- the board could be rotated to make it a move in bination with the previous inputs. That well either square 1 or 4. With that restriction, the lights up to indicate a move by MAYA. strategy we chose for MAYA allows 19 different

To simplify MAYA’s programming, we re- possible games to be played. In one of the games, moorman tommy

90 SCIENTIFIC AMERICAN © 2008 SCIENTIFIC AMERICAN, INC. November 2008 © 2008 SCIENTIFIC AMERICAN, INC. Harry plays perfectly and the game ends in a ently within a mixture than they did on their draw. In the remaining 18 games, MAYA ex- own, necessitating other redesigns. Finally, after MAYA-II ploits his mistakes and wins. three consecutive summers and many Saturdays, The second generation of To work out all the required gates for the au- through some changes of inputs and many small the authors’ tic-tac-toe-play- tomaton, we considered every move in all 19 adjustments of gate sequences and concentra- ing DNA computer, MAYA-II, games and determined which gates would pro- tions, our team had a system in which we could goes beyond MAYA-I in duce the desired move. The hardest part was clearly distinguish active and inactive gates in all several respects. matching the strategy requirements with our wells, for all the games, reproducibly. The human player may make logic-gate technology. Although our gates are any legal move in response to designed to output DNA strands that could in Implications MAYA-II’s opening move, principle serve as inputs to other gates, for Integrating more than 100 molecular logic com- increasing the number of possi- MAYA we chose to avoid relying on that feature ponents in a single system represented a substan- ble games to 76. MAYA-II wins and the extra complications it might engender. tial milestone. In the jargon of electronics, 72 of them and draws 4. Altogether we took less than three months to de- MAYA-II is the first “medium-scale integrated 32 logic gates cleave green- sign and develop MAYA and fully test all 19 molecular circuit.” Our work on a device of such fluorescing substrates to high- games in the laboratory. complexity let us refine our deoxyribozyme log- light the human’s squares. ic gates as plug-and-play computing primitives. MAYA-II New efforts in our laboratories now proceed 96 logic gates compute Not content with MAYA’s limitations, we built more smoothly with existing components, and MAYA-II’s moves and indicate an unrestricted version, MAYA-II. We also we can design gates that usually work immedi- them with red fluorescence. A computer program designed made MAYA-II more user-friendly, displaying ately without needing any fine-tuning. the arrangement of gates. both players’ moves in two different fluorescent We could integrate our method with other colors. The automaton still goes first and claims molecular computing approaches developed re- the middle square, but Harry the human can cently. For example, Erik Winfree’s group at the then take any of the remaining eight squares. California Institute of Technology came up with MAYA-II plays four times as many possible impressive “strand displacement cascades,” games as MAYA, winning 72 of them and draw- which could be used to analyze mixtures of oli- ing four. gonucleotides in a similar fashion. In this We wrote a computer program (for a stan- scheme, strands of DNA combine, joining and dard silicon-based computer) to determine an displacing one another mostly without the appropriate arrangement of logic gates. The re- need for any catalysts analogous to the DNA sulting design calls for 128 different logic gates, enzymes of our gates. Winfree’s system has 96 for deciding and signaling the automaton’s been demonstrated with a cascade of five moves using red fluorescence and 32 to highlight units. In comparison, our present system suffers Harry’s moves in green fluorescence. from becoming prohibitively slow if three layers The sheer size of this automaton made build- of gates are combined. MAYA-II, for all its com- ing and testing MAYA-II an enormous chal- plexity, functions as a single layer of gates and lenge. One of us (Macdonald) led the project and takes around 15 minutes to carry out a move. ➥ more to trained several high school students to test au- For our decision-making molecules, we are explore tomata, mostly during summers and on Satur- now very confident about putting many gates to- days. The students checked all 76 games multi- gether, and tasks representing fresh challenges A Deoxyribozyme-Based Molecu- ple times. They had to make changes in MAYA- beckon. We hope one day to report a mixture of lar Automaton. Milan N. Stojanovic and Darko Stefanovic in Nature Bio- II’s design to deal with several problems (and molecules that can be taught a strategy by play- technology, Vol. 21, No. 9, pages then recheck all the games after each tweak). ing example games with them or by introducing 1069–1075; September 2003. Our chief concern going into the project was some selection to eliminate the gates that encode that some sequences might bind in unintended losing strategies. We might then develop autom- Medium Scale Integration of places. Our computer-modeling tools were not ata that we can train to recognize cancer cells. Molecular Logic Gates in an Automaton. Joanne Macdonald et al. advanced enough to be able to predict such dif- But perhaps the most important next step of in Nano Letters, Vol. 6, No. 11, pages ficulties. In fact, spurious binding was relatively our program is to incorporate new primitives to 2598–2603; November 2006. rare. Instead the more serious problem turned carry out more functions, such as sensing and out to be individual gates cleaving their sub- moving (or “actuating”). These are automata MAYA II, a Second-Generation Tic- strates at different rates. We (or, rather, our stu- that would take action based on the presence of Tac-Toe Playing Automaton. Online at http://tinyurl.com/4mvbnm dents) had to adjust concentrations and struc- a given input. Our plug-and-play system would tures to correct for this variability. We also then be moving well beyond “play” and would Eric Winfree’s home page: n tommy moorman tommy quickly discovered that some gates acted differ- be ready for some real work. www.dna.caltech.edu/~winfree