Biocomputer Research Papers

We now move from system components to the complete biocomputer system, and define the general purpose silicon computer system as a template for biocomputers. Such a template consists of four units: the input and output device (I/O), the arithmetic logic unit, the control unit and the memory (Figure 2) [115].

The first three units collectively build the central processing unit (CPU), typically constructed on a single integrated circuit called a microprocessor. The control unit coordinates the various system components. It decodes the program instructions, and transforms them into control signals, which activate other system parts. This finally results in a change of the system state. Historically the control unit was defined as a distinct part, whereas in modern design this unit is an internal part of the CPU. Busses (often made of groups of wires) interconnect these units. Each unit contains a huge number of small electrical circuits. Switches can turn these circuits on (1) or off (0). A logic gate can perform a logic operation on one or more of such logic inputs and produce a single logic output. Thus, basic elements of any biocomputer unit are switches and logic gates.


As discussed the basic function of a switch is to produce an on or off state. Such switches have been engineered based on transcription regulation, artificial DNA, or RNA.

The DNA based type can either be based on a gene regulatory circuit or on DNA molecule properties. The toggle switch, a synthetic, bistable gene-regulatory network in Escherichia coli, belongs to the first category [116]. This toggle switch is a quite famous one, published in a landmark paper, which helped to kickstart synthetic biology. The toggle is constructed from two repressible promoters, such as that repressor 1 inhibits transcription from promoter 1 and is induced by inducer 1, whereas repressor 2 inhibits transcription from promoter 2 and is induced by inducer 2. The switch can take two stable states, if the inducers are absent: one in which promoter 1 transcribes repressor 2, and one in which promoter 2 transcribe repressor 1. The switch is flipped between these stable states by transient chemical or thermal induction of the currently active repressor. All together, the toggle switch forms an addressable cellular memory unit.

Another type of switch, called I-switch, an artificial DNA nano-device, that has cytosine-rich regions, which act as a sensor for chemical input in the form of protons and functions as a pH sensor based on fluorescence resonance energy transfer (FRET) inside living cells. [117]. The I-switch consists of three oligonucleotides, where two with single stranded overhangs are hybridized onto the adjacent third. At acid conditions these overhangs are protonated, leading to a closed conformation with high FRET. This switch was used to map spatial and temporal pH changes during endosome maturation. These experiments demonstrate the potential of DNA scaffolds responsive to triggers in living cells. These principles might be applied to switches in DNA or RNA scaffolds which assemble proteins [118].

We have already discussed one kind of RNA based switche, the riboswitche, above. Another approach is switches based on an engineered riboregulator, which enable post-transcriptional control of gene expression [119]. This riboregulator is constructed such that a small sequence, complementary to the ribosome binding site (RBS), is inserted downstream from a promoter and upstream from the RBS. After transcription a stem loop is formed at the 5‘ end of the mRNA, which blocks ribosome docking and translation. This mRNA can be targeted by another non-coding RNA and undergo a linear-loop interaction, that expose the obstructed RBS and thus activates expression. Interestingly, this kind of artificial riboregulator have been used to build a genetic switchboard that independently controls the expression of multiple genes in parallel [120].

As mentioned above, it is possible to engineer Boolean logic based on RNAi. A tunable switch has been built based on a synthetic gene network that couples repressor proteins with a design involving shRNA (Figure 3C) [121].

Figure 3

Input/Output (I/O) device: A) In a “digital” biological I/O device input molecules induce due to a set of non-steady state chemical reactions (engineered coherent with a logic scheme) an output molecule. All molecules have a defined concentration...

Although protein based switches, that do not comprise transcription factors, are not uncommon in nature, they have been so far not a major focus [18].

Logic gates

A logic gate is an elementary building block of a digital circuit. These gates can have one or two inputs, but only one output. Inputs and output are of Boolean nature, thus they can be either true (1) or false (0). Different logic operators can be applied on the input. Basic types of logic gates are: AND, OR, NOT (inverter), XOR, NAND, NOR, and XNOR [101] [18]. These operators are the basis for different truth tables (Figure 4). We get a true output from the gate for the following case: AND - both inputs are true; OR - either or both inputs are true; NOT - (has only one input) if the input is false; XOR (either/or) - either input 1 or input 2 is true; NAND - (is an AND gate followed by a NOT gate) both inputs are false, or one is true; NOR (OR followed by NOT) - both inputs false; and XNOR (XOR followed by NOT) - both inputs are true or both are false. All other cases give a false output respectively. Over a period of about two decades DNA, RNA and protein based logic gates have been engineered and classified [122] [101]. A wide range of core machinery and inducers has been developed.

Figure 4

Arithmetic logic unit: Shown are four basic Boolean logic gates (AND, NOT, NOR, and XOR), their symbols and respective truth Table 1 means that the input (a, b) is sensed or the output (out) is released, whereas 0 means not. In the examples system output...

DNA based logic gates: One strategy for engineering a logic gate in vivo is to build a core machinery, based on gene expression regulation [123] [124] [125] [126] [127] [128] [129]. One such system had two inputs such as beta-D-thiogalactopyranoside and anhydrotetracycline (aTC) and a fluorescent protein as output [73]. In order to build such a logic system, a network plasmid was generated composed of a set of three transcription factor encoding genes (LacI, TetR, and lambda cI) and their corresponding promoters. The binding state of LacI and TetR can be changed with the input molecules. Moreover, the system consists of five additional promoters which can be regulated by the three transcription factors. Two of the promoters are repressed by LacI, one is repressed by TetR, and the remaining two are respectively positively or negatively regulated by lambda cI. Altogether, this results in 125 possible networks. Various GFP expressing systems can be formed using a combination of various promoters, input molecules and host strains e.g. E.coli. In such manner functional networks were formed with logic operations such as NOR, NOT, and NAND.

Another option to build a core machinery both in vivo and in vitro is by using DNA aptamers [130] [131]. Yoshida et al built an AND gate by fusing an adenosine-binding DNA aptamer and a thrombin-binding DNA aptamer [130]. Each aptamer binds to partially complementary fluorescence quencher-modified nucleotides, QDNA1 and 2 respectively. When the two inputs adenosine and thrombin are bound both QDNAs are released from the aptamers leading to increased fluorescence intensity. Other input combinations (0 + 0, 0 +1, and 1 +0), lead to the presence of zero or one QDNA and a weaker fluorescence. Similar, an OR gate can be created, if the positions of the fluorophore and QDNA are modified. Another study built an aptamer based nanorobot, which has an open and closed conformation [131]. DNA aptamer–based lock mechanism opens in response to binding of antigen keys. This lock functions as an AND gate, where the aptamer-antigen activation state serves as input, and the nanorobot conformation as output.

Hybridization can serve as another in vitro option for engineering a core machinery feasible for functioning in a logic based network [95] [132]. A two input logic gate of type AND, OR or NOT were constructed by using a branch-migration scheme with a mechanism built on strand recognition and strand replacement. Single stranded nucleic acids are input and output of such a scheme. The gate function is created by sequential base pairing triggered by toehold-toehold binding between single strands and subsequent breaks.

Moreover, in vitro deoxyribozyme-based (DNAzymes) logic gates have been engineered [127] [128] [133]. In order to engineer an AND gate, two different oligonucleotide inputs were hybridized with corresponding controlling elements [127]. This led to the cleavage of the substrate in the presence of both inputs and subsequent conformational change of controlling elements. A NOT and XOR gate was constructed in a similar fashion (Figure 4B).

Recently, a novel in vivo system, called transcriptor, has been used to build permanent amplifying AND, NAND, OR, XOR, NOR, and XNOR gates to control transcription rates (Figure 4A) [134].

RNA based logic gates: The other major class of logic gates is RNA-based. The core machine can be based on RNA aptamer, a riboswitch, ribozymes, hybridization, amber suppressor tRNA, or an orthogonal ribosome [101].

Culler et al demonstrated in vivo that it is possible to engineer an AND gate based on a β-catenin binding RNA aptamer [135]. This aptamer was inserted into the intron position, between a protein-coding exon and an alternatively spliced exon (Ex) containing a stop codon, followed by another intron, the next protein-coding exon and the herpes simplex virus- thymidine kinase (HSV-TK) gene whose product, in turn, is an activator of ganciclovir (GCV). Binding of β-catenin with the RNA aptamer led to mature mRNA which lacked Ex. This led to the expression of HSV-TK. If the alternatively spliced exon was not excluded from the mature mRNA, an early translation termination occurred. This resulted in the synthesis of a nonfunctional peptide. For the induction of apoptosis as output, both expression of HSV-TK and the presence of GCV are required. Another study from the same lab demonstrated the building of AND, NOR, NAND, or OR gates based on RNA aptamer as a core machine (Figure 4C) [88].

We already discussed a simple riboswitch as a structure suitable to build logic. Sudarsan et al. reported a tandem riboswitch core machinery in vivo that facilitate more sophisticated control [136]. They discovered in the 5’ untranslated region of Bacillus clausii metE RNA two naturally occurring riboswitches. Both riboswitches bind independently to two different metabolites, one to S-adenosylmethionine (SAM) and the other to coenzyme B12 (AdoCbl). This binding induced the transcription termination of gene of interest through cis-acting corresponding riboswitches. Only in the absence of both inputs (not SAM and not AdoCbl) we get the full length transcript as output. All together this system functions as a two-input Boolean NOR logic gate.

Logic gates have been engineered with ribozymes as core machinery both in vivo and in vitro [137] [138]. An AND is built when simultaneous hybridization of two oligonucleotide inputs with the ribozyme lead to its activation [137]. Chen et al engineered a YES gate (if input 0 so output 0; if input 1 so output 1) in a system based on a ribozyme, which was inserted into the 3′-UTR of a target transgene [138]. The ribozyme was inactivated in the presence of theophylline, allowing the target transgene to be expressed.

Alternatively a logic gate can be based in vivo on hybridization, with siRNA or miRNA as input [98] [97]. An AND like logic function has been built by using two groups of miRNAs as input and the hBAX protein as output [97]. The miRNAs act as a repressor of activators and repressors in the gate.

Amber suppressor tRNA can be used in vivo as the core machinery for a logic gate [139] [140]. This kind of tRNA identifies the “amber” stop codon (UAG), inserts an amino acid, and do not terminate translation. Anderson et al. utilized an amber suppressor tRNA (SupD) to engineer a two input AND gate [140]. One input is a salicylate responsive promoter, which is linked to the transcription of the amber suppressor tRNA supD. The other input is a arabinose responsive promoter, that regulates the transcription of T7 RNA polymerase. T7 has been mutated to contain two amber stop codons and thus requiring SupD expression for a fully functional T7, which is connected to the expression of green fluorescent protein as an output.

Furthermore, an AND gate has been engineered in vivo using an orthogonal (unnatural) ribosome / mRNA pair [94]. The inputs in this system are two orthogonal rRNAs, which limit the translation of two respective mRNAs. These mRNAs encode two fragments of beta-galactosidase, which's activity is the output of the system.

Protein based logic gates: The third class are protein based logic gates, where a transactivator, an enzyme, chemically inducible dimerization (CID), a T7 RNA polymerase or a zink finger transcription factor can act as core machinery in a logic gate [101] [141].

Transactivator: Various logical gates have been engineered in vivo based on chemically inducible transactivator-based gene circuits [108] [74]. This conception was used by Ausländer et al. to construct several logic gates and combination of them, such as NOT, AND, NAND and N-IMPLY (if a = 0 and b = 1, so output = 1; else output = 0) [74]. Such a N-IMPLY gate was engineered by combining an erythromycin-dependent transactivator and an apple metabolite phloretin-dependent transactivator. The output, fluorescent d2EYFP, was only visible by fluorescent microscopy or FACS analysis in the presence of erythromycin and absence phloretin.

Enzyme based logic gates, such as XOR, N-IMPLY, AND, OR, NOR, NOT, and YES (one input; if input =1, output =1; else output = 0), have been constructed for in vitro systems with a wide variety of inputs, such as glucose, H2O2, NADH, acetaldehyde, starch, phosphate, NAD+, acetylcholine, butyrylcholine, O2 [101] [104] [105] [106] [142]. Baron et al. constructed eg a two input AND gate [104]. Both H2O2 and glucose are in this case necessary input in order to activate the catalytic chain with gluconic acid as output.

Another option for building Boolean logic in vivo is based on a CID system [109] [110]. Bronson et al. utilized a CIT system to engineer a two input AND gate [109]. Dexamethasone– methotrexate input induced the dimerization of an activation domain, B42-glucocorticoid receptor chimera (B42-GR), and a DNA-binding domain, LexA-dihydrofolate reductase chimera (LexA-DHFR). Both B42-GR and LexA-DHFR expression is placed under the control of the GAL1 promoter. Thus galactose is required as second input in the system in order to form the ternary complex. This complex induces, as output of the system, acts as a transcriptional activator and stimulates the transcription of the output, a lacZ reporter gene.

Recently Shis et al. published another interesting option to build an AND gate [141]. A functional T7 RNA polymerase can be built from two fragments, whereas the larger T7 RNA polymerase fragment is encoded by a gene that responds to arabinose and the smaller fragment by a gene that responds to lactose. T7 RNA polymerase will be functional active in the presence of both inputs, arabinose and lactose.

OR, NOR, AND and NAND logic has been based on artificial Cys2– His2 zink finger (ZF) transcription factors as computing elements [75]. Input signals led to expression of corresponding ZF-based transcription factors, which acted on response promoters. An OR gate was constructed, which contained target sites for two different ZF activators [75]. BCR_ABL-1:GCN4 and erbB2:Jun activators were used as ZF-1 and ZF-2, respectively. AmCyan fluorescent protein output, measured by flow cytometry, was observed, when either or both inputs were present.

Cell to cell communication based logic gates: Logic systems built on gene expression regulation can be expanded to multicellular engineered networks [112] [143]. Different logic gates were carried in one study by different strains of E.coli, which communicate by quorum sensing (see above) (Figure 4D) [112]. Input was aTC or arabinose. Colonies containing different gates were wired together via quorum molecules. Different combinations of colonies containing specific simple logic gates resulted in the construction of 16 two-input Boolean logic gates. Different combinations of 2 input molecules such as NaCL, galactose, 17 beta-estradiol, doxycycline, galactose, or glucose were used in another study which builds a multicellular network based on gene expression regulation [143].

Finally, one might ask how many gates can be interconnected with the present technology in a circuit. A study by Privman et al. tried to determine this maximum number under optimal noise reduced conditions [144]. They concluded that under such conditions, logic gates can be concatenated for up to order 10 processing steps. Beyond that, it will be necessary to engineer novel systems for avoiding noise buildup.

Input and output (I/O) device

A biomolecular I/O device is basically an engineered set of chemical reactions with input and output molecules with distinct concentrations, formalized as e.g for the case of a two input device: [input molecule 1] + [input molecule 2] <-> [output molecule] (Figure 3A).

In order to act in a digital manner, the concentrations need to be defined as distinguishable high or low, which can be translated to Boolean logic (low as 0 or of, and high as 1 or on) (Figure 3B). As we already discussed above, a variety of interesting devices have been constructed (Figure 3C) [79] [92] [93] [116] [117] [119] [120] [121] [137]. However, reaction kinetics and dynamics are often difficult to predict as values in a living cell are often continuous, can variate to a certain degree, are away from a steady state, and can be difficult to quantitate [18]. Thus, to facilitate Boolean logic, thresholds of inputs and outputs must be well defined, which can be difficult to achieve in biological systems [18]. Depending on the kind of system this can this be concentrations, localization of biomolecules or enzyme activity [101]. A linear system can contribute to minimize retroactive effects; as such a system allows applying well defined control theory. Oishi et al tried to address these kinds of problems and tried to identify design principles for an ideal linear I/O system [145]. Their implementation of such an I/O systems was based on idealized chemical reactions, and on enzyme-free, entropy-driven DNA reactions.

Arithmetic logic unit

The arithmetic logic unit performs two classes of operations: arithmetic and logic. Both have been engineered in biological systems.

Biological computers have shown t no be able to execute simple arithmetic such as addition and subtraction, as Ausländer et al have demonstrated by a combinatorial assembly of chemically inducible transactivator-based logic gates [74]. Moreover, it has been shown that more complex arithmetic is achievable. A combination of several DNA hybridization based logic gates make it e.g. possible to calculate the integer part of a square root of a four-bit binary number [132] [146].

Logic can be built, as discussed by means of logic gates (Figure 4).

The ability of biocomputers to solve logic problems beyond the Hamiltonian path problem have been demonstrated by the implementation of several logic requiring games [35] [147] [133] [47] [86]. A molecular automaton was engineered, which was able to play a game which covers all possible responses to two consecutive sets of four inputs [147]. Moreover, a deoxyribozyme-based automation is able to play a complete game of tic-tac-toe [133]. A device based on DNA recombination was able to solve a sorting problem, where a stack of distinct objects needed to be placed into a certain order and orientation using a minimal number of manipulations [47]. A molecular algorithm based on ribonuclease digestion to manipulate strands of a 10-bit binary RNA library has been used to address the so called “Knight problem” which asks what configurations of knights in a chess game can one place on a 3 x 3 chess board such that no knight is attacking any other knight on the board [86]. Moreover, DNA hybridization based logic has been used to implement simple logic programs [148]. This logic system consisted of molecular representations of facts such as Man(Socrates) and rules such as Mortal(X) <-- Man(X) (Every Man is Mortal). The system was able to answer molecular queries such as Mortal(Socrates)? (Is Socrates Mortal?) and Mortal(X)? (Who is Mortal?).

Control unit / Central processing unit

A state machine is a theoretic mathematical model, which helps to understand what is going on in the central processing unit of a computer, and which can be experimentally implemented (Figure 5A) [18] [149] [150].

Figure 5

Control unit; central processing unit: A) A final state machine, as shown here, is a theoretical model which can help to understand what is going on in the central processing unit. Simplified: Symbols a and b are written on a tape, which is read by the...

State is defined as all the stored information, at a given point in time, to which the circuit or program has access. The output of a circuit or program is determined by its input and states. The simplest form of such a state machine is called finite state machine (or finite state automata). In simple terms, this machine contains a tape with symbols a and b. The tape can move in one direction and the machine can read the symbol on the tape. The machine changes its state due to the letter it reads. A string transducer is a state machine that also can write symbols and a Turing machine can in addition move from left to right [151].

State machines have been engineered with biomolecules [152] [153] [154] [40] [155] [156] [157] [158]. Hagiya et al. built in 1997 the first state to state transition system by guiding DNA polymerase based DNA extension by a template strand with a transition rule sequence [152] [153]. The present state is encoded by the 3’-end sequence of a single-stranded DNA molecule. The template strand (rule) enclosed a binding site for the 3-end of the DNA molecule and the extension template. State transition occurred, if the current state is annealed onto an appropriate portion of DNA encoding the transition rules and the next state was copied to the 3’-end by extension with polymerase. The extension template represents the new state.

The first experimental implementation of a finite state machine, comprising DNA and DNA-manipulating enzymes, was published by Benenson et al. in 2001 (Figure 5B) [154] [18]. Similar to the concept developed by Benenson et al. several finite state machine were later developed in which the transitions were executed by autonomous biochemical steps based on DNA sticky end recognition, ligation and digestion [40] [156]. This system was expanded by Adar et al. to allow stochastic computing. The core of this form of computing is the choice between alternative computing paths (biochemical pathways), each with a prescribed probability, which were programmed by the relative molar concentrations of the software molecules coding for the alternatives [155]. Another finite state machine based on DNA aptamer generated different configurations (outputs) in response to a set of two different groups’ chemical inputs [157]. Moreover, by using molecular finite state machines simultaneously with fluorochrome labeled DNA it was possible to distinguish between two distinct images encrypted onto a DNA chip [158].


DNA's biological role is to encode huge amounts of data, theoretically up to two bits per nucleotide or 455 exabytes per gram of ssDNA [43]. It has been recently shown, that it is possible to encode arbitrary digital information in DNA, e.g. an html-coded draft of a book that included 53,426 words, 11 JPG images and 1 JavaScript program into a 5.27 megabit bitstream [43]. The oligonucleotides library was engineered by utilizing next-generation DNA synthesis techniques. In order to read the encoded book, the library was amplified by PCR and subsequent sequenced. A similar study encoded computer files totalling 739 kilobytes into a DNA code [44].

Another important feature of DNA is the relatively permanence of the storage. Even after the cells die, one might be able to recover information from the DNA. These storage abilities make DNA suitable as core machinery for engineered memory devices.

Several biological storage devices have been engineered [41] [42] [45] [159]. Some feedback motifs in natural systems exhibit memory such as mutual inhibition and auto regulatory positive feedback [41

Today’s silicon-based microprocessors are manufactured under the strictest of conditions. Massive filters clean the air of dust and moisture, workers don spacesuit-like gear and the resulting systems are micro-tested for the smallest imperfection. But at a handful of labs across the country, researchers are building what they hope will be some of tomorrow’s computers in environments that are far from sterile-beakers, test tubes and petri dishes full of bacteria. Simply put, these scientists seek to create cells that can compute, endowed with “intelligent” genes that can add numbers, store the results in some kind of memory bank, keep time and perhaps one day even execute simple programs.

All of these operations sound like what today’s computers do. Yet these biological systems could open up a whole different realm of computing. “It is a mistake to envision the kind of computation that we are envisioning for living cells as being a replacement for the kinds of computers that we have now,” says Tom Knight, a researcher at the MIT Artificial Intelligence Laboratory and one of the leaders in the biocomputing movement. Knight says these new computers “will be a way of bridging the gap to the chemical world. Think of it more as a process-control computer. The computer that is running a chemical factory. The computer that makes your beer for you.”

As a bridge to the chemical world, biocomputing is a natural. First of all, it’s extremely cost-effective. Once you’ve programmed a single cell, you can grow billions more for the cost of simple nutrient solutions and a lab technician’s time. In the second place, biocomputers might ultimately be far more reliable than computers built from wires and silicon, for the same reason that our brains can survive the death of millions of cells and still function, whereas your Pentium-powered PC will seize up if you cut one wire. But the clincher is that every cell has a miniature chemical factory at its command: Once the organism was programmed, virtually any biological chemical could be synthesized at will. That’s why Knight envisions biocomputers running all kinds of biochemical systems and acting to link information technology and biotechnology.

All of these operations sound like what today’s computers do. Yet these biological systems could open up a whole different realm of computing. “It is a mistake to envision the kind of computation that we are envisioning for living cells as being a replacement for the kinds of computers that we have now,” says Tom Knight, a researcher at the MIT Artificial Intelligence Laboratory and one of the leaders in the biocomputing movement. Knight says these new computers “will be a way of bridging the gap to the chemical world. Think of it more as a process-control computer. The computer that is running a chemical factory. The computer that makes your beer for you.”

As a bridge to the chemical world, biocomputing is a natural. First of all, it’s extremely cost-effective. Once you’ve programmed a single cell, you can grow billions more for the cost of simple nutrient solutions and a lab technician’s time. In the second place, biocomputers might ultimately be far more reliable than computers built from wires and silicon, for the same reason that our brains can survive the death of millions of cells and still function, whereas your Pentium-powered PC will seize up if you cut one wire. But the clincher is that every cell has a miniature chemical factory at its command: Once the organism was programmed, virtually any biological chemical could be synthesized at will. That’s why Knight envisions biocomputers running all kinds of biochemical systems and acting to link information technology and biotechnology.

Realizing this vision, though, is going to take a while. Today a typical desktop computer can store 50 billion bits of information. As a point of comparison, Tim Gardner, a graduate student at Boston University, recently made a genetic system that can store a single bit of information-either a 1 or a 0. On an innovation timeline, today’s microbial programmers are roughly where the pioneers of computer science were in the 1920s, when they built the first digital computers.

Indeed, it’s tempting to dismiss this research as an academic curiosity, something like building a computer out of Tinker Toys. But if the project is successful the results could be staggering. Instead of painstakingly isolating proteins, mapping genes and trying to decode the secrets of nature, bioengineers could simply program cells to do whatever was desired-say, injecting insulin as needed into a diabetic’s bloodstream-much the way that a programmer can manipulate the functions of a PC. Biological machines could usher in a whole new world of chemical control.

In the long run, Knight and others say, biocomputing could create active Band-Aids capable of analyzing an injury and healing the damage. The technology could be used to program bacterial spores that would remain dormant in the soil until a chemical spill occurred, at which point the bacteria would wake up, multiply, eat the chemicals and return to dormancy.

In the near term-perhaps within five years-“a soldier might be carrying a biochip device that could detect when some toxin or agent is released,” says Boston University professor of biomedical engineering James Collins, another key player in the biocomputing field.

The New Biology

Biocomputing research is one of those new disciplines that cuts across well-established fields-in this case computer science and biology-but doesn’t fit comfortably into either culture. “Biologists are trained for discoveries,” says Collins. “I don’t push any of my students towards discovery of a new component in a biological system.” Rockefeller University postdoctoral fellow Michael Elowitz explains this difference in engineering terms: “Typically in biology, one tries to reverse-engineer circuits that have already been designed and built by evolution.” What Collins, Elowitz and others want to do instead is forward-engineer biological circuits, or build novel ones from scratch.

But while biocomputing researchers’ goals are quite different from those of cellular and molecular biologists, many of the tools they rely on are the same. And working at a bench in a biologically oriented “wet lab” doesn’t come easy for computer scientists and engineers-many of whom are used to machines that faithfully execute the commands that they type. But in the wet lab, as the saying goes, “the organism will do whatever it damn well pleases.”

After nearly 30 years as a computer science researcher, MIT’s Knight began to set up his biological lab three years ago, and nothing worked properly. Textbook reactions were failing. So after five months of frustratingly slow progress, he hired a biologist from the University of California, Berkeley, to come in and figure out what was wrong. She flew cross-country bearing flasks of reagents, biological samples-even her own water. Indeed, it turned out that the water in Knight’s lab was the culprit: It wasn’t pure enough for gene splicing. A few days after that diagnosis, the lab was up and running.

Boston University’s Gardner, a physicist turned computer scientist, got around some of the challenges of setting up a lab by borrowing space from B.U. biologist Charles Cantor, who has been a leading figure in the Human Genome Project. But before Gardner turned to the flasks, vials and culture dishes, he spent the better part of a year working with Collins to build a mathematical model for their genetic one-bit switch, or “flip-flop.” Gardner then set about the arduous task of realizing that model in the lab.

The flip-flop, explains Collins, is built from two genes that are mutually antagonistic: When one is active, or “expressed,” it turns the second off, and vice versa. “The idea is that you can flip between these two states with some external influence,” says Collins. “It might be a blast of a chemical or a change in temperature.” Since one of the two genes produces a protein that fluoresces under laser light, the researchers can use a laser-based detector to see when a cell toggles between states.

In January, in the journal Nature, Gardner, Collins and Cantor described five such flip-flops that Gardner had built and inserted into E. coli. Gardner says that the flip-flop is the first of a series of so-called “genetic applets” he hopes to create. The term “applet” is borrowed from contemporary computer science: It refers to a small program, usually written in the Java programming language, which is put on a Web page and performs a specific function. Just as applets can theoretically be combined into a full-fledged program, Gardner believes he can build an array of combinable genetic parts and use them to program cells to perform new functions. In the insulin-delivery example, a genetic applet that sensed the amount of glucose in a diabetic’s bloodstream could be connected to a second applet that controlled the synthesis of insulin. A third applet might enable the system to respond to external events, allowing, for example, a physician to trigger insulin production manually.

GeneTic Tock

As a graduate student at Princeton University, Rockefeller’s Michael Elowitz constructed a genetic applet of his own-a clock.

In the world of digital computers, the clock is one of the most fundamental components. Clocks don’t tell time-instead, they send out a train of pulses that are used to synchronize all the events taking place inside the machine. The first IBM PC had a clock that ticked 4.77 million times each second; today’s top-of-the-line Pentium III computers have clocks that tick 800 million times a second. Elowitz’s clock, by contrast, cycles once every 150 minutes or so.

The biological clock consists of four genes engineered into a bacterium. Three of them work together to turn the fourth, which encodes for a fluorescent protein, on and off-Elowitz calls this a “genetic circuit.”

Although Elowitz’s clock is a remarkable achievement, it doesn’t keep great time-the span between tick and tock ranges anywhere from 120 minutes to 200 minutes. And with each clock running separately in each of many bacteria, coordination is a problem: Watch one bacterium under a microscope and you’ll see regular intervals of glowing and dimness as the gene for the fluorescent protein is turned on and off, but put a mass of the bacteria together and they will all be out of sync.

lowitz hopes to learn from this tumult. “This was our first attempt,” he says. “What we found is that the clock we built is very noisy-there is a lot of variability. A big question is what the origin of that noise is and how one could circumvent it. And how, in fact, real circuits that are produced by evolution are able to circumvent that noise.”

While Elowitz works to improve his timing, B.U.’s Collins and Gardner are aiming to beat the corporate clock. They’ve filed for patents on the genetic flip-flop, and Collins is speaking with potential investors, working to form what would be the first biocomputing company. He hopes to have funding in place and the venture launched within a few months.

The prospective firm’s early products might include a device that could detect food contamination or toxins used in chemical or biological warfare. This would be possible, Collins says, “if we could couple cells with chips and use them-external to the body-as sensing elements.” By keeping the modified cells outside of the human body, the startup would skirt many Food and Drug Administration regulatory issues and possibly have a product on the market within a few years. But Collins’ eventual goal is gene therapy-placing networks of genetic applets into a human host to treat such diseases as hemophilia or anemia.

Another possibility would be to use genetic switches to control biological reactors-which is where Knight’s vision of a bridge to the chemical world comes in. “Larger chemical companies like DuPont are moving towards technologies where they can use cells as chemical factories to produce proteins,” says Collins. “What you can do with these control circuits is to regulate the expression of different genes to produce your proteins of interest.” Bacteria in a large bioreactor could be programmed to make different kinds of drugs, nutrients, vitamins-or even pesticides. Essentially, this would allow an entire factory to be retooled by throwing a single genetic switch.

Amorphous Computing

Two-gene switches aren’t exactly new to biology, says Roger Brent, associate director of research at the Molecular Sciences Institute in Berkeley, Calif., a nonprofit research firm. Brent-who evaluated biocomputing research for the Defense Advanced Research Projects Agency-says that genetic engineers “have made and used such switches of increasing sophistication since the 1970s. We biologists have tons and tons of cells that exist in two states” and change depending on external inputs.

For Brent, what’s most intriguing about the B.U. researchers’ genetic switch is that it could be just the beginning. “We have two-state cells. What about four-state cells? Is there some good there?” he asks. “Let’s say that you could get a cell that existed in a large number of independent states and there were things happening inside the cell…which caused the cell to go from one state to another in response to different influences,” Brent continues. “Can you perform any meaningful computation? If you had 16 states in a cell and the ability to have the cell communicate with its neighbors, could you do anything with that?”

By itself, a single cell with 16 states couldn’t do much. But combine a billion of these cells and you suddenly have a system with 2 gigabytes of storage. A teaspoon of programmable bacteria could potentially have a million times more memory than today’s largest computers-and potentially billions upon billions of processors. But how would you possibly program such a machine?

Programming is the question that the Amorphous Computing project at MIT is trying to answer. The project’s goal is to develop techniques for building self-assembling systems. Such techniques could allow bacteria in a teaspoon to find their neighbors, organize into a massive parallel-processing computer and set about solving a computationally intensive problem-like cracking an encryption key, factoring a large number or perhaps even predicting weather.

Researchers at MIT have long been interested in methods of computing that employ many small computers, rather than one super-fast one. Such an approach is appealing because it could give computing a boost over the wall that many believe the silicon microprocessor evolution will soon hit. When processors can be shrunk no further, these researchers insist, the only way to achieve faster computation will be by using multiple computers in concert. Many artificial intelligence researchers also believe that it will only be possible to achieve true machine intelligence by using millions of small, connected processors-essentially modeling the connections of neurons in the human brain.

On a wall outside of MIT computer science and engineering professor Harold Abelson’s fourth-floor office is one of the first tangible results of the Amorphous Computing effort. Called “Gunk,” it is a tangle of wires, a colony of single-board computers, each one randomly connected with three other machines in the colony. Each computer has a flashing red light; the goal of the colony is to synchronize the lights so that they flash in unison. The colony is robust in a way traditional computers are not: You can turn off any single computer or rewire its connection without changing the behavior of the overall system. But though mesmerizing to watch, the colony doesn’t engage in any fundamentally important computations.

Five floors above Abelson’s office, in Knight’s biology lab, researchers are launching a more extensive foray into the world of amorphous computation: Knight’s students are developing techniques for exchanging data between cells, and between cells and larger-scale computers, since communication between components is a fundamental requirement of an amorphous system. While Collins’ group at B.U. is using heat and chemicals to send instructions to their switches, the Knight lab is working on a communications system based on bioluminescence-light produced by living cells.

To date, work has been slow. The lab is new and, as the water-purity experience showed, the team is inexperienced in matters of biology. But some of the slowness is also intentional: The researchers want to become as familiar as possible with the biological tools they’re using in order to maximize their command of any system they eventually develop. “If you are actually going to build something that you want to control-if we have this digital circuit that we expect to have somewhat reliable behavior-then you need to understand the components,” says graduate student Ron Weiss. And biology is fraught with fluctuation, Weiss points out. The precise amount of a particular protein a bacterial cell produces depends not only on the bacterial strain and the DNA sequence engineered into the cell, but also on environmental conditions such as nutrition and timing. Remarks Weiss: “The number of variables that exist is tremendous.”

To get a handle on all those variables, the Knight team is starting with in-depth characterizations of a few different genes for luciferase, an enzyme that allows fireflies and other luminescent organisms to produce light. Understanding the light-generation end of things is an obvious first step toward a reliable means of cell-to-cell communication. “There are cells out there that can detect light,” says Knight. “This might be a way for cells to signal to one another.” What’s more, he says, “if these cells knew where they were, and were running as an organized ensemble, you could use this as a way of displaying a pattern.” Ultimately, Knight’s team hopes that vast ensembles of communicating cells could both perform meaningful computations and have the resiliency of Abelson’s Gunk-or the human brain.

Full Speed Ahead

Even as his lab-and his field-takes its first steps, Knight is looking to the future. He says he isn’t concerned about the ridiculously slow speed of today’s genetic approaches to biocomputing. He and other researchers started with DNA-based systems, Knight says, because genetic engineering is relatively well understood. “You start with the easy systems and move to the hard systems.”

And there are plenty of biological systems-including systems based on nerve cells, such as our own brains-that operate faster than it’s possible to turn genes on and off, Knight says. A neuron can respond to an external stimulus, for example, in a matter of milliseconds. The downside, says Knight, is that some of the faster biological mechanisms aren’t currently understood as well as genetic functions are, and so “are substantially more difficult to manipulate and mix and match.”

ill, the Molecular Sciences Institute’s Brent believes that today’s DNA-based biocomputer prototypes are steppingstones to computers based on neurochemistry. “Thirty years from now we will be using our knowledge of developmental neurobiology to grow appropriate circuits that will be made out of nerve cells and will process information like crazy,” Brent predicts. Meanwhile, pioneers like Knight, Collins, Gardner and Elowitz will continue to produce new devices unlike anything that ever came out of a microprocessor factory, and to lay the foundations for a new era of computing.

Who’s Who in BiocomputingOrganizationKey ResearcherFocus Lawrence Berkeley National Laboratory Adam Arkin Genetic circuits and circuit addressing Boston University James J. Collins Genetic applets Rockefeller University Michael Elowitz Genetic circuits MIT Thomas F. Knight Amorphous computing

Simson Garfinkel

Simson L. Garfinkel is a computer security research scientist whose interests include digital forensics, security, personal information management, privacy, and terrorism. He holds six U.S. patents for his computer-related research, has published… More dozens of journal and conference papers in security and computer forensics, is the author or co-author of 14 books, and has started five companies.

One thought on “Biocomputer Research Papers

Leave a Reply

Your email address will not be published. Required fields are marked *