The computational ability of biological molecules is the basis for tiny machines that will be able to communicate directly with living cells
By: Yaakov Benanson and Ehud Shapira
When the British mathematician Alan Turing conceived the idea of the universal programmable calculating machine, he attributed the word "computer" to a person and not to an object. The year was 1936, and people whose profession was "computer" were engaged in tedious calculations of numbers. The machine that Turing devised to carry out this work - a machine that could calculate any computable problem - laid the foundation for the theoretical study of calculation, and is still the cornerstone of computer science. But Turing never specified what materials it should be built from.
Turing's pure thinking machine had no electrical wires, transistors or logic gates. He himself imagined her as a person equipped with an endless sheet of paper, a pencil and a simple instruction book. His indefatigable computer will read a symbol from the paper tape, change the symbol, then move on to the next symbol, according to the predetermined rules, and it will continue to do so until it is no longer possible to follow any of the rules it has. The electronic calculating machine, built in the 40s from metal and vacuum tubes and later from Zorn components, is probably the only "species" of non-human computer that most people have encountered, but it is not the only form that a computer can take.
Living beings, for example, also perform complex physical processes in the direction of digital information. The biochemical reactions, and ultimately the functioning of the entire animal, are controlled by instructions stored in the animal's genome and encoded in nucleic acid sequences. When comparing the mechanisms of the intracellular biomolecular machines that process the information stored in DNA and RNA, and the mechanism of a Turing machine, striking similarities emerge: both systems process information stored in a string of symbols taken from a fixed alphabet, and both operate through a step-by-step shift -Step along the strings, as they change or add symbols according to a given set of rules.
These similarities inspired the idea that biological molecules could one day serve as raw material for a new breed of computers. Such biological computers will not necessarily have computational power or perform better at traditional computer tasks. Natural molecular machines, such as the ribosome, operate at low speed, only a few hundred operations per second, compared to the billions of operations per second of logic gates in some electronic devices. But biological molecules have a special ability: they speak the language of living cells.
The promise inherent in computers made of biological molecules lies in their ability to operate within a biochemical environment, even within a living being, and to interact with this environment through input and output in the form of other biological molecules. A biomolecular computer may act, for example, as an independent "doctor" inside a cell. It will be able to receive as input signals from the environment that indicate a disease, process them using medical knowledge programmed into it, and produce as output a signal or medicine.
In the last seven years we have worked to realize this vision. We succeeded in creating a biological automaton made of DNA and proteins, which can diagnose in vitro the molecular symptoms of certain types of cancer and "treat" the disease by releasing a drug molecule. The proof of concept was exciting, both because it has strong medical applications, and because it is completely different from what we intended to build when we set out.
Models of molecules
One of us (Shapira) began his research with the understanding that the basic operations of certain biomolecular machines inside living cells - identifying molecular building blocks, cutting and fusing biopolymer molecules, and moving along a polymer - might be used, in principle, to build a universal computer based on the idea of Turing's conceptual machine . The calculation operations of such a Turing machine will be translated into biomolecular terms: one "recognition", two "splits", two "ties", and a shift to the left or right.
Charles Bennett of IBM thought about these things and proposed a hypothetical molecular Turing machine as early as 1982. He was interested in the physics of energy consumption, so he expected that molecules would one day serve as a basis for computing devices with better energy utilization.
The first real demonstration of the computational power of molecules was by Leonard M. Edelman of the University of Southern California, who used DNA to solve a problem that was a heavy task for traditional computer algorithms. The problem, known as the "Hamiltonian route" or the "traveling agent problem", is to find the shortest route between several cities connected to each other by air lines, when each city is visited only once. He created DNA molecules that symbolically represented the cities and flights, mixed trillions of such molecules in a test tube, and received an answer within a few minutes. Unfortunately, it took him much longer to fish out the molecules representing the correct solution from the mixture with the means available to him at the time. Edelman waited for new technologies that would make it possible to create a more practical molecular computer.
"In the future, research in molecular biology may provide improved methods for handling macromolecules," Edelman wrote in 1994 in an initial scientific paper describing the DNA experiment. "It is hoped that the research in chemistry will make it possible to develop synthetic 'designer enzymes'. It is possible to imagine that eventually a multi-purpose computer made of a single macromolecule will be developed, which will be attached to a collection of ribosome-like enzymes that will act on it."
The goal that Shapira set for himself was to develop a concrete logical design for exactly such a device, a device that could function as a basic "operational specification" for a wide group of future molecular computing machines. In 1999 he had a mechanical model made of plastic parts. At this point we joined together to turn the model into actual molecules.
We did not want to immediately attack the difficult challenge of building a molecular Turing machine. We decided to first try to build a simpler, Turing-like machine called a finite automaton. The sole task of this simple machine was to determine whether a string of symbols or letters from a two-letter alphabet, say "a" and "b", contained an even number of b's. This operation can be performed by a finite automaton with only two states and a "program" containing four commands called transition rules. One of us (Benson) came up with the idea of using a double-stranded DNA molecule to represent the input string, and four short double-stranded DNA molecules to represent the transition rules, or "software", of the automaton, and two natural enzymes that act on DNA, FokI And ligase, as "hardware".
The main logical problem we had to solve in designing the automaton was how to represent the changing intermediate states of the calculation, which consist of two factors - the current internal state of the automaton, and a pointer to the symbol in the input string being processed. We managed to do this using a nice trick: at each step of the calculation, the enzymatic hardware "digested" the input molecule, so that the processed symbol was extracted and the next symbol was revealed. Since the symbol can be split into two different sites, each of the resulting versions could represent, in addition to the symbol itself, one of two possible states of the calculation. We then found out that a similar idea had been proposed by Paul Rothmond, a former student of Edelman's, for a molecular Turing machine.
It is worth noting that the computer presented by our team in 2001 operated independently: after the input molecules, the software and the hardware were put into a suitable solution in a test tube, the calculation began and continued in iterations until it was finished without human intervention.
While testing this system, we realized that it not only solves the original problem for which it was created - to determine if a symbol in a string appears an even number of times - but it can do more than that. A two-state, two-symbol automaton has eight possible symbol-state-rule combinations (23), and because our design was modular, we could easily realize all eight possible transition rules using eight different transition molecules. Thus, we could make the automaton perform different tasks by choosing a different "program" - that is, a different mixture of transition molecules.
As we tested our simple molecular automaton with a variety of programs, we also found a way to further refine its performance. One of the tests was the omission experiment, in which the operation of the automaton was tested after removing one component at a time. When the ligase, the enzyme that attaches the software molecule to the input molecule to allow it to be recognized and cleaved by the other enzyme, FokI, was taken out, some calculation seems to have occurred after all. We discovered a previously unknown ability of FokI to recognize and cleave certain DNA sequences even if the two strands of the molecule are not adjacent to each other.
We were delighted with the emerging possibility of removing the ligase from our molecular computer design, as this meant reducing our enzymatic hardware by 50%. More importantly, the ligation was the only operation in the calculation that consumed energy, and eliminating it would have allowed the computer to operate without an external fuel source. Also, without the ligation step, the software molecules will not be consumed during the calculation, but will be recycled.
It took our research group months of painstaking effort and data analysis to bring the ligase-free system to perfection. In the beginning, its efficiency was very low, and it stopped already after one or two calculation steps. But the computational and biochemical challenge continued to spur us on, and in the end Benson, with the help and advice of Rebecca Ader and other colleagues, found the long-awaited solution. Thanks to small but crucial changes we made in the DNA sequences of the automaton, we were able to take advantage of the new feature we discovered in FokI and achieve a significant jump in the computer's performance. In 2003 we already had an independent programmable computer whose input molecules themselves were its only source of fuel. Therefore, the computer can basically process any input molecule, of any length, using a fixed number of hardware and software molecules and without consuming its energy.
However, from a computational point of view, our automaton still looks like a scooter compared to the Rolls Royce of computers we aspired to: a biomolecular Turing machine.
Doctor DNA
Since the final two-state automaton was too simple to be of real use in solving complex computational problems, we treated it as nothing more than an interesting demonstration of the idea of programmable self-acting biomolecular computers and decided to move on. But when we tried to build more complicated automatons, we soon ran into the same problem that Edelman faced: the "designer enzymes" that he yearned for ten years earlier did not yet exist.
No natural enzyme or enzyme combination is known that can perform the particular recognition, cutting and splicing in a sequential sequence and with the flexibility necessary to realize the design of a Turing machine. It will be necessary to adapt natural enzymes or to engineer entirely new synthetic enzymes. Since science is not yet capable of this, we find ourselves with the design of a biomolecular Turing machine, but have to wait until the parts necessary to build it are invented.
So we went back to our two-state automaton, to see if we could at least find something useful that it could do. We were already thinking about the direction of medical applications, so we wondered if the device would be able to perform some kind of simple diagnosis, such as determining whether a set of conditions representing a particular disease is present in the environment.
Two situations are enough for this task: we called one situation "yes" and the other situation "no". The automaton will start the calculation in yes mode and check condition by condition. If the tested condition is present, the ready state will remain the same, but if the condition is not present, the automaton will move to the not state and will remain so until the end of the calculation process. That is, the calculation will only end in yes if all the conditions of the disease are present. If one of the conditions is not present, the "diagnosis" will be negative.
For this logical design to work, we had to find a way to connect the molecular automaton to its biochemical environment so that it could sense whether certain disease conditions were present. The general idea that the environment can affect the relative concentration of competing transition molecules - and thus affect the calculation - was already proposed in the work plans for the molecular Turing machine. To apply this principle to the sensing of disease symptoms, we had to make the presence or absence of a disease marker the factor that determines the concentration of the software molecules that indicate the symptom.
Many types of cancer are characterized by abnormal levels of any proteins in the cell, the reason for which is the overexpression or underexpression of a protein that acts on certain genes. When a gene expresses the protein it encodes, enzymes in the cell nucleus copy its sequence into an RNA version. This molecular transcript of the gene, called messenger RNA (mRNA), is "read" by the cellular structure called the ribosome, which translates the RNA sequence into the string of amino acids that make up the protein. Therefore, higher than normal or lower than normal levels of certain messenger RNA transcripts can indicate gene activity.
Bennson devised a method in which certain transition molecules preferentially interact with these messenger RNA sequences. The interaction, in turn, affects the transition molecule's ability to participate in the calculation. At a high level of messenger RNA representing a disease state, there will be a dominant presence of the transition molecule Yes? Yes, and this will increase the chance that the computer will find that the symptom is present. In practice, this system can be used for any disease involving abnormal levels of proteins resulting from gene activation, and it can also be adapted to identify harmful mutations in messenger RNA sequences.
Now that we have at our disposal an input mechanism for detecting disease symptoms and the logical structure for performing the diagnosis, we asked this question: what should the computer do when it diagnoses a disease? First we wanted it to produce a visual signal. However, in the molecular world, the distance between producing a signal and the next logical step - administering the drug - is not so great. Benjamin Gil, a student in our team, proposed and built a mechanism that allows the computer to release a drug molecule in response to a positive diagnosis.
But our plan is not yet complete. A central question in computer hardware design is how to build a reliable system from unreliable components. The problem is not unique to biological computers - it is an essential feature of complex systems; The reliability of mechanical devices also decreases as the dimensions are small and the number of components increases. In our case, given the probabilistic nature of the calculation and the imprecise behavior of biomolecular components, it is inevitable that some calculations will give a positive diagnosis even in the absence of some or even all of the symptoms of the disease, and vice versa. Happily, this probabilistic behavior is measurable and predictable, so we could address it with a system of checks and balances.
We created two types of calculation molecules: one is supposed to release a drug when the calculation ends with a yes, and the other releases a substance that suppresses the same drug when the calculation ends with a no. By changing the relative concentrations of the two types of molecules, we could control the certainty threshold of the diagnosis. In the future, if our molecular automaton is launched on a medical mission, it will be possible to program it to exercise a similar judgment.
An explosive new strain
It turned out that our simple scooter carried us further than we thought it could, and in a slightly different direction than we intended. Our biomolecular computer has so far only been demonstrated in vitro. The image of its biological environment is made by adding different concentrations of RNA and DNA molecules and putting all the components of the automaton into the same test tube. Now we have set as our goals to make it work inside a living cell, to see it computer inside the cell and to get it to communicate with the environment.
Inserting the automaton into the cell is a challenge in itself, because most insertion systems are designed for either DNA or protein, while our computer contains both. We must therefore find a way to infuse both together. Another hurdle is finding a means to watch all aspects of the calculation as they occur inside the cell, to make sure the automaton can work without the cell's activities interfering with the calculation steps, or the computer components affecting the cell's behavior in unwanted ways. And finally, we will look for alternative means to create a connection between the automaton and its environment. Very recent work in cancer research suggests that microRNAs, which are small molecules with important control functions in cells, are better markers of the disease, and so we began to change the design of our computer so that it would "talk" to microRNAs instead of Messenger RNA.
Although we are still far from operating our device inside living cells, and certainly on living creatures, we have a proof of concept. Our in vitro demonstration, which linked biochemical disease symptoms directly to the basic computational steps of a molecular computer, showed that an autonomously operating molecular computer can communicate with biological systems and perform meaningful biological evaluations. Its input mechanism can sense the environment in which it operates, its calculation mechanism can analyze the environment, and its output mechanism can affect the environment in an intelligent way based on the results of the analysis.
Our automaton thus fulfilled the promise that biomolecular computers would enable direct interaction with the biochemical world. It also brings computational science back to Turing's original vision. The first calculating machines had to deviate from the idea to accommodate the properties of electronic components. Only a few decades later, when molecular biologists began to reveal the working mechanisms of tiny machines inside living cells, computer scientists recognized in them systems similar to Turing's abstract idea of computation.
We do not intend to claim that molecules will replace electronic machines in all computational tasks. Each of the two varieties of computers has its own advantages, and they can coexist. Because biological molecules can directly access the information encoded in other biological molecules, they have a natural fit for living systems that electronic computers could never have. Therefore, we believe that our experiments demonstrate the essential importance of this new breed of computer in a wide variety of uses. The biomolecular computer detects life signals.
Computing machines: conceptual and natural
Mathematician Alan Turing envisioned the properties of a mechanical computer in 1936, long before intracellular molecular machines were seen and studied. When the working mechanisms of nature's tiny automatons were later identified, surprising similarities to Turing's ideas were discovered: both systems store information in strings of symbols, both process the strings step by step, and both change or add symbols according to fixed rules.
Turing machine
This hypothetical device operates on an information encoded tape bearing symbols such as . a and b A control unit with read/write capability processes the tape, symbol by symbol, according to instructions provided by transition rules, which refer to the internal state of the unit. The transition rule in this example therefore dictates that if the state of the control unit is 0 (SO ), and the called symbol is a, the unit should change its state to 1 (S1), change the symbol to b, and move one place to the left (L).
biological machine
The ribosome, an organelle in the cell, reads information encoded in gene transcripts called messenger RNA (mRNA) and translates it into amino acid sequences to create proteins. The symbolic alphabet of mRNA is made up of triplets of nucleotides called codons, each of which corresponds to a specific amino acid. While the ribosome processes the mRNA strand, codon by codon, helper molecules called adapter RNA (tRNA) bring in the correct amino acids. The tRNA confirms the codon match and then releases the amino acid that will join the elongating chain.
Overview/ Living Computers
– Natural molecular machines process information in a similar way to the Turing machine, which was an early conceptual computer.
- A Turing-like automaton built from DNA and enzymes can perform calculations, receive input from other biological molecules and output tangible results, such as a signal or medicine.
- This working computer built from the molecules of the code of life demonstrates the feasibility of its operation and may prove to be a valuable medical tool.
A model of a molecular Turing machine
A Turing machine made of biological molecules will use their natural ability to recognize symbols and connect molecular subunits together or break the connections between them. A plastic model built by one of the authors (right) is used as working plans for such a system. The symbols are written on yellow "molecule" stones. Blue software molecules indicate the state of the machine and define transition rules. Bumps on the stones differentiate them physically.
איך זה עובד?
The machine works on a string of symbol molecules. In the position of the control unit in the middle, the symbol and the current state of the machine are defined.
Construction of a molecular automaton
Since living things process information, their materials and mechanisms are suitable for computational operations. The DNA molecule exists to store information in an alphabet of nucleotides. Cellular mechanisms read and change the information using enzymes and other molecules. This operating system relies on the chemical affinities between the molecules, through which they recognize and bind each other. Building a Turing machine-like device from molecules therefore means translating the concepts into the language of molecules.
A simple conceptual computer, called a finite automaton, can move in only one direction and can read a series of symbols and change its internal state according to transition rules. A two-state automaton can therefore answer yes-no questions by switching between two states labeled 1 and 0. The state of the automaton at the end of the calculation represents the result.
The raw materials for a molecular automaton include DNA strands in various arrangements, which are used both as input and as software, and the DNA splitting enzyme, FokI, which is used as hardware. Nucleotides, whose abbreviated names are A, C, G and T, encode here both the symbols and the internal state of the machine.
A calculation that works independently
A hardware-software coupling identifies the state conjunction and its complementary symbol on the input molecule. The molecules connect to form a hardware-software-input conjugate, then FokI cleaves the input molecule and reveals the next symbol.
A new hardware-software coupling identifies the next state and symbol on what is left of the input molecule.
Responses continue until no rule is matched or until the end symbol is revealed.
In this example, computational splits leading to the final output (far right) produced a four-nucleotide termination symbol indicating machine state 0, which is the result of the computation.
About the authors
Ehud Shapira and Yaakov Benanson started collaborating to build a molecular automaton in 1999. Shapira is a professor in the Departments of Computer Science and Biological Chemistry at the Weizmann Institute of Science, and holds the Harry Weinerb Professorial Chair. Shapira was already an accomplished computer scientist and software pioneer with a growing interest in biology in 1998, when he first designed a model of a molecular Turing machine. Benanson, who completed his studies for a master's degree in biochemistry at the Technion, was a doctoral student with Shapira the following year. Benanson, currently a fellow at the Bauer Center for Genome Research at Harvard University in the USA, is working on new molecular tools to influence living cells.
Doctor DNA
After showing that an automaton made of DNA and enzymes can perform abstract yes-or-no calculations, the authors wanted to give the device a practical query it might encounter inside a living cell: Are there signs of disease present in the environment? If the answer is yes, the automaton can produce as an output active drug therapy. The basic computational idea has not changed from the previous design: couplings of transition molecule "software" and enzyme "hardware" process symbols in a diagnostic molecule, splitting it again and again and revealing the next symbols. Also, the new task requires a means that will allow the disease markers to create input for the calculation and mechanisms that will verify the correctness of the diagnosis and produce treatment.
And more on the subject
A Mechanical Turing Machine: Blueprint for a Biomolecular Computer. Presented by Ehud Shapiro at the 5th International Meeting on DNA Based Computers, Massachusetts Institute of Technology, June 14–15, 1999. www.weizmann.ac.il
Programmable and Autonomous Computing Machine Made of Biomolecules. Y. Benenson, T. Paz-Elizur, R. Adar, E. Keinan, Z. Livneh and E. Shapiro in Nature, Vol. 414, pages 430–434; November 22, 2001.
An Autonomous Molecular Computer for Logical Control of Gene Expression. Y. Benenson, B. Gil, U. Ben-Dor, R. Adar and E. Shapiro in Nature, Vol. 429, pages 423–429; May
