Could you describe your work in general terms?
|
We constructed a prototype of a medical molecular computer (several trillion such computers can fit in a drop of water) that is able to diagnose disease conditions and release a drug upon positive diagnosis. Made entirely of biological molecules, this computer was successfully programmed to identify – in a test tube – changes in the balance of molecules in the body that indicate the presence of certain cancers, to diagnose the type of cancer, and to react by producing a drug molecule to fight the cancer cells.
|
|
|
| What do you find most surprising or exciting about these results? |
Molecular computer scientists have been searching for a "killer application" that would exploit their advantages over traditional silicon-based technology. Initially, it was hoped that DNA computing might solve computationally-hard problems. However, due to intrinsic limitations, DNA computing is not likely to surpass silicon technology in this field. An alternative paradigm, pursued by our laboratory, proposes to utilize molecular computers for applications in which direct processing of biological information in required. For instance, in situ detection and analysis of molecular signals in living organisms is impossible with electronic computers due to the insurmountable incompatibility of the materials involved, whereas DNA computers are built of like components.
This work represents the first actual proof of concept and the first actual demonstration of a possible real-life application for this kind of computer. The molecular realization is rather intricate and it required a development of a number of molecular-manipulation techniques from scratch. Our system is also quite complex. Tens of DNA strands can each react with any other strand. It turned out that our design was robust enough to cope with this complexity and function reliably despite imperfect function of its component.
|
|
|
What is the "working environment" of the medical computer? Is it able to operate in a living tissue?
|
So far, Our medical computer works in the "sterile" conditions of a solution that contains the salts and optimal pH necessary for enzyme function. Unlike an intracellular or even intercellular environment, this solution does not contain the numerous proteins, nucleic acids, lipids and polysaccharides found in living organisms. Some of these compounds may present a significant danger to the components of our computer in its present form, or, conversely, our computer could inflict damage on these intracellular components.
In other words, our prototype will require significant adjustments before it can be tested in an in vivo environment. It may take decades before a diagnostic molecular computer that is completely safe and can function for long periods without breakdown will be produced and approved for use. |
|
|
Molecular markers for cancers are routinely used for cancer diagnostics.
What is so special about your method?
|
The novelty of our study does not lie in the method of recognizing cancer. In fact, we rely heavily on existing diagnostic methodology in our design. What is different about our approach is its potential for performing diagnostics within tissue itself. Traditional methods require tissue extraction, isolation of the molecular marker in question and its comparison to the norm. Our medical computer might one day be administered as a drug, and be distributed throughout the body by the blood stream to detect disease markers autonomously and independently in every cell. In this way, a single cancer cell could be detected and destroyed before the tumor develops. Even in a late- stage cancer, this kind of treatment could reach every secondary growth, however small, and effectively terminate the disease. Futuristic as this may seem, the unique properties of our computer may one day give them an edge over present “macroscopic” techniques.
|
|
|
| How does your computer perform a diagnosis? |
In our model, “disease” is characterized by the coexistence of several molecular conditions at once. Focusing on several conditions rather than just one greatly improves the precision of the diagnosis, as single markers may act as indicators of more than one disease, or they may be a temporary, nonsignificant phenomenon. DNA chip-based techniques have demonstrated that a given cancer may be precisely characterized by several tens of molecular markers. We estimate that our system could analyze up to 15 markers. While this is less than the full complement of the DNA chip, it may still provide a quite precise identification.
The disease conditions, or markers, that we are able to identify all relate to mRNA transcripts and consist of three types: 1) mutated mRNA sequences that reflect incorrect gene sequence and result in malfunction in the protein transcribed from this mRNA; 2) under-expressed mRNA, which translates into a lower-than-normal amount of a possibly vital transcribed protein, 3) over-expressed mRNA, which leads to an excess of the transcribed protein, and may result in uncontrolled acceleration of intracellular processes and uncontrolled cell division.
|
|
|
How does your computer detect the molecular markers?
|
We developed a mechanism that steers the course of computation by molecular markers. The computer has two states, Yes for a positive diagnosis and No for a negative diagnosis. The computer samples the markers one by one; as long as the markers reflect a disease condition, it remains in the Yes state. Upon sampling a marker in a normal state, the computer switches to No, and this becomes the final output of the computation. If all the markers reflect a disease condition, the final result will be the Yes state, meaning a positive overall diagnosis. This diagnostic output then feeds into a drug-administering process that releases a drug tailored not only to the specific diagnosis, but to the “strength” of the diagnosis (see below).
|
|
|
| What kind of treatment are you able to provide now, and how it might be extended in the future? |
Our current design may release any therapeutic short nucleic acid, either DNA or RNA. Short DNA molecules form the basis for so-called antisense drugs. While approved antisense drugs are presently scarce, there are numerous such drugs in various stages of clinical trials. Short RNA molecules have also been touted as potential drugs, working via the mechanism of RNA interference. A widespread opinion in a pharmaceutical industry states that drugs based on short RNA molecules will become ubiquitous in the near future.
One could envision alternative designs that could result in a release of proteins, long DNA molecules, organic molecules, etc.
|
|
|
How does the computer release the right amounts of the drug at the right time?
|
For each diagnostic computation we employ multiple diagnostic molecules. Therefore, some molecules may terminate in the positive diagnostic state while others terminate in the negative state. The higher the ratio of positive to negative answers, the higher our confidence in the presence of the disease. This raises the dilemma, common to nearly all diagnostic techniques, of the in-between state, in which a diagnosis is not conclusive, or gives a false -positive or –negative result.
A special design feature was created to solve this problem: some of these molecules are designed to release the active drug upon positive diagnosis and do nothing in the case of negative diagnosis. Others release a special "drug suppressor" molecule, which neutralizes exactly one drug molecule, upon negative diagnosis, but do nothing upon positive diagnosis. Thus, the system has a built-in “check-and-balance” function to prevent mishaps in treatment.
The ratio between drug and supressor defines the "drug release function" of a system, i.e. an amount of active drug released for a given Yes:No ratio of the diagnostic process. |
|
|
Compared to a normal computer, how does the molecular computer work?
|
Electronic computers work according to a mathematical model developed by John von Neumann in the 1940s, called the stored program computer. In it, both program and data are stored as words in the computer memory. Each memory word can be accessed by its address. The electronic computer fetches program instructions one by one from its memory and executes them. Typically the instructions are to load data from a certain memory location into a register, to store the content of a register into a memory location, or to perform an operation on registers, e.g. :to add the content of two registers and store the result in a third register.
The stored program computer model is ideal for realization using electronic circuits, and hence it has been the basis of all electronic computers for more than 50 years.
Our molecular computer is based on a very different model, called a finite automaton, which is a simplified version of the Turing machine, a mathematical model of computation developed by the British mathematician Alan Turing in the 1930s. It has data stored as a string of symbols (Turing envisioned a paper tape divided into squares, each holding one symbol) A control unit, which holds the program rules, is in a particular location on the string and can be in one of a finite number of states. The Turing machine starts the computation in an initiate state and on the leftmost symbol in the string. In each cycle, a Turing machine reads a symbol and executes a program rule that specifies what action to take according to the symbol read and the internal state. The actions can be to replace the symbol with a new symbol, move one symbol to the left or to the right, and/or change internal state. A finite automaton is a Turing machine that cannot change the string of symbols, and in each step moves one symbol to the right. Hence the computational result of the finite automaton is the final state upon reaching the last input symbol.
Both the stored program computer and the Turing machine are so-called "universal computers" and as such have equivalent computing power. A finite automaton, however, is much weaker.
Our molecular finite automaton has three components: An input molecule, which stores the data to be processed as well as the fuel needed for the computation, software molecules, which encode program rules, and a hardware molecule which performs the computation, as directed by the software and the input. The molecules float in a watery solution containing some salts needed for the operation of the hardware molecule.
The input molecule is a double-stranded DNA molecule. A mathematical trick is used to store the current state and current symbol to be processed in a "sticky-end" of the DNA molecule, a single-stranded protrusion obtained by having one strand longer than the other. The software molecules also have "sticky-ends", each matching a different state-and-symbol input combination. Through a spontaneous chemical process called DNA hybridization, the input molecule and a matching software molecule temporarily connect to each other. The hardware is a naturally occurring molecule, an enzyme called FokI that attaches to DNA in a specifically coded location and cuts both strands at a fixed distance from that location, creating a sticky end. Using additional chemical and mathematical tricks, the software molecules are designed so that the hardware enzyme attaches to them and, once a software molecule hybridizes with an input molecule, the hardware molecule cuts the input molecule in a programmed location, exposing a new sticky end that encodes the next state and the next symbol to be processed. The final sticky end, obtained at the end of the computation, encodes the computation result.
|
|
|
What other work has been done in this field?
|
The field started in 1994 with the publication of a paper by Len Adleman, describing how one can use DNA to solve combinatorial problems, such as the so-called "traveling salesman problem": finding an optimal route through several cities with given distances between them. The hope was that the parallel nature of DNA manipulation could be used to outperform electronic computers in solving such problems.
This approach involves multiple outside interventions during computation. A solution containing the molecular components is subjected to mixing, separations, PCR amplification, etc several times during computation. Therefore, they are not truly "molecular computers" but rather "lab-assisted molecular computers", where the lab personnel play a crucial role in accomplishing a computation.
Today, fewer people (think, believe) molecular “supercomputers” will be built due to inherent limitations of the approach and the rapid advance of electronic computers. Our work does not attempt to compete with electronic computers head-on, but rather to design molecular-scale computing devices that might have applications in areas not accessible to electronic computers, such as "smart components" in biochemical reactions.
Another line of research is "semi-autonomous molecular computation", which is closer to our work. This direction is pursued by Hagiya and Sakamoto in Tokyo University, and by Winfree, Reif and Seeman in the USA. This approach aims at creating systems that evolve in a programmed fashion to perform a computation, while minimizing the external intervention. These researchers were successful in performing very simple computations without much interference outside of regulating temperature during the process (in a PCR manner - multiple cycles of heating/cooling). Still, the experimentally demonstrated computation capabilities of these systems is substantially lower than that of our devices, and they are not completely autonomous.
|
|
|
What exactly is a state machine and a finite automaton?
|
Generally, a state machine consists of 1) a data tape divided into cells, each containing a symbol selected from the tape alphabet, and 2) a computing device driven by transition rules. The device is positioned over one of the cells and is in one of a finite number of internal states. Depending on the symbol detected and on the internal state, a transition rule instructs the device to write a new symbol, change state and move one cell to the left or to the right. The Turing machine is the most general type of state machine, capable of writing on the tape as well as moving in both directions.
A more restricted, yet important, class of state machines is finite automata. A finite automaton is a unidirectional read-only Turing machine. Its input is a finite string of symbols. It is initially positioned on the leftmost input symbol in a default initial state, and in each transition moves one symbol to the right, possibly changing its internal state. Its software consists of transition rules, each specifying a next state based on the current state and current symbol. A computation terminates after the last input symbol is processed, the final state being its "output". Alternatively, it may suspend without completion when no transition rule applies. Some states are deemed “accepting.” An automaton accepts an input if there is a computation with this input that ends in an accepting final state. A finite automaton with two states and an alphabet of two symbols is shown in Fig. 1A. It determines if an input over the alphabet {a, b} contains an even number of a's. Fig. 1B shows the intermediate configurations obtained during a computation of this automaton. A two-state two-symbol automaton can have eight possible transition rules. Programming such an automaton amounts to selecting transition rules and deciding which states are accepting. A selection of such programs is shown in Fig. 1C.
|
 |
|
Fig. 1. Finite automata and inputs for which we show molecular realizations. (A) Automaton A1. Incoming unlabeled arrow marks the initial state, and labeled arrows represent transition rules. A double circle represents an accepting state. (B) A computation of A1 scanning and digesting the input abba. Each row represents an intermediate configuration showing current state and remaining input. The transition rule applied is shown in brackets. (C) The two other automata for which we demonstrate molecular operation.
|
|
|
Is the computer a single unit? The mRNA disease indicators will be present in the cell; does the enzyme that will cleave the computer DNA also occur naturally in the cell?
|
The computer is not a single unit, but rather an assembly of molecules, which interact in a highly selective manner. In some ways the flow of information through the computer resembles the flow of information through signal transduction pathways, where the parts are physically separate but nevertheless provide a coherent function. The enzyme ,FokI, which we employed, is produced naturally in some bacteria, but not in the mammalian cell. Should it be used for in vivo computation, it would need to be either injected into a cell or produced inside it by means of genetic engineering.
|
|
|
Is a part of each computer destroyed during the computation process?
|
| What is destroyed during computation is what we call a "diagnostic molecule", which encodes the medical knowledge and contains an inactive drug precursor. Other parts of the computer, namely the software module, are not consumed. |
|
|
Can you explain how the computer measures the level of indicator mRNA in the cell, particularly how it measures an under-expression of a particular gene?
|
There is a pair of transition molecules for each disease indicator, a positive one which is active when the indicator (which could be high mRNA or low mRNA depending on the context) is present, and a negative one which is active in the normal state.
For under-expressed genes, we designed a system in which the mRNA destroys the positive transition and helps build the negative one. The computer is initially formulated with an active positive transition. When it comes in contact with the mRNA molecules, presence of an indicator (= mRNA low, a disease state for under-expressed gene) leaves an active positive transition and the disease is positively diagnosed in this particular indicator. Once an mRNA is at its normal level (=indicator absent), the positive transition is destroyed and the disease is negatively diagnosed.
|
|
|
| How cheaply can this be manufactured compared to other forms of medication? |
| Since the system is not anywhere near the practical application stage as yet, it is hard to predict its final cost. However, we believe the cost will be of the same order of magnitude as existing drugs. |
|
| How long does it take the computer to break down in the cell if it has a negative response? |
In its present form, the computer is not is not particularly stable inside a cell. Its lifespan would be probably an hour or so. Lifespan, however, is not dependent on the diagnosis.
|
|
|
Do you foresee this technique being used to administer non-therapeutic drugs?
|
| Once the device is made to work in vivo, it may release many kinds of responses, including non-therapeutic drugs or even just signals as diagnostic aids to human physicians. |
|
|
What gave you the idea for the DNA computer?
|
The ideas used in this work were inspired by multiple sources, altough it was mainly concieved in our laboratory. The device consists of the input, computation and output modules. The computation module was invented in our group and developed earlier for different purposes. That work was inspired by the works of C. Bennett, L. Adleman and P. Rothemund. The input module that detects levels of mRNA molecules was invented in our group and it was inspired by the antisense DNA technology and DNA nanoactuators developed by B. Yurke and colleagues. The output module that delivers a drug was invented 'from scratch' in our group.
|
|
|
How exactly does it work? Can you describe it so I can picture it?
|
Any computer applies a program to an input. The result of the computation depends both on the input and the program. If we are able to make either the program or an input depend on some external conditions, then the result of the computation would indirectly provide some information on these conditions. In our case, we took our existing molecular computer and invented a method to change its program as a response to abnormal mRNA levels and/or mutations. In this way, our computation result depends on these levels, which may indicate a disease. Our diagnostic result is translated into a 'therapy' by another method.
|
|
|
How does your method differ from existing molecular computing methods?
|
We have been conducting research in molecular computing for several years, focusing on autonomous, programmable simple molecular computers (automata). Other methods try to solve hard computational problems using additional manual manipulations. Our current result is the first example of a prototype real-life application for molecular computers. Our success directly derives from the features of automonous operation and programmability that are often missing from competing approaches.
|
|
|
What was the technical and/or conceptual breakthrough that allowed this to work?
|
There were several breakthroughs, including the method for sensing abnormal gene levels and mutated genes, the method for analyzing the acquired data and the method for controllable drug release following the diagnosis.
|
|
|
What were the technical challenges you had to address in order to make this work?
|
| In fact, after the systems was specified 'on a paper', it was pretty easy to realize. The main difficulties were conceptual rather than technical. Still, actual implementation required an intensive and concentrated lab effort. |
|
|
What are the pluses and minuses of your method?
|
The method is very flexible since it may potentially detect multiple disease conditions which makes the diagnosis very selective. The drug release method allows the fine tuning of the strength of the treatment even when the diagnosis is uncertain. On the other hand, this is a prototype device that works in the test tube. It will require major modifications to be made compatible with the living systems.
|
|
|
Was there anything surprising about the results? Why?
|
Molecular computer scientists have been searching for a "killer application" that would exploit their advantages over traditional silicon-based technology. Initially, it was hoped that DNA computing might solve computationally-hard problems. However, due to intrinsic limitations, DNA computing is not likely to surpass silicon technology in this field. An alternative paradigm, pursued by our laboratory, proposes to utilize molecular computers for applications in which direct processing of biological information in required. For instance, in situ detection and analysis of molecular signals in living organisms is impossible with electronic computers due to the insurmountable incompatiblity of the materials involved, whereas DNA computers are built of like components.
This work represents the first actual proof of concept and the first actual demonstration of a possible real-life application for this kind of computer. The molecular realization is rather intricate and it required a development of a number of molecular-manipulation techniques from scratch. Our system is also quite complex. Tens of DNA strands can each react with any other strand. It turned out that our design was robust enough to cope with this complexity and function reliably despite imperfect function of its component.
|
|
|
How could this research be applied practically?
|
The published results are a proof of concept. They show that autonomous, molecular scale systems are able to perform such complex tasks as disease diagnosis and treatment. This result will hopefully lead to concerted effort to move from the proof-of-concept stage to the more practical applications or real diagnosis and cure, although these applications lie far ahead.
|
|
|
Even further down the road, what types of uses might it have or eventually lead to?
|
Eventually we envision molecular-scale medical molecular computers that fight a disease with unprecedented efficiency. We also call them 'smart drugs'
|
|
|
| What are the next steps in your research? What are you ultimately aiming for? |
| The next steps may involve an adjustment of our device for in vivo function. |
|
|
| Can you give me a ballpark estimate of when the research could be technically ready to be applied practically? |
We believe that a first working 'smart drug', although more primitive that described by us, may become available within the next 3-4 years. Such drug would probably sense a single disease indicator rather that multiple indicators. More elaborate smart drugs may take about 10 years to emerge, but of course one has to add the duration of the clinical investigations to this estimate.
|
|
|
Who funded the research?
|
| Israel Science foundation, Moross Cancer Institute and the MInerva foundation |
