Information-carrying DNA strands can be used to perform simple computations, but have so far been little more than toys. Can molecular computers be more broadly useful — in medicine, for instance?
People have long been fascinated with automata, fashioning them from any available materials to create mechanical creatures or musical devices, or to carry out simple computations. The materials used have even included biological molecules: in 2001, for instance, Yaakov Benenson and colleagues1 built a tiny automaton from DNA strands and enzymes. On page 423 of this issue2, Benenson et al. suggest how such molecular automata might be used to augment so-called antisense technologies, carrying out a diagnosis in vivo (that is, in a living cell) that automatically controls drug delivery.
Antisense drugs are oligonucleotides — short, single-stranded DNA molecules — that offer the promise of treating diseases caused by the expression of a harmful gene, for example a cancer-causing gene3. These drugs are designed to prevent the expression of such a gene: they bind specifically to the messenger RNA (mRNA) strand that is transcribed from the gene, thereby inhibiting the translation of the mRNA into a protein.
Benenson et al.2 were motivated by the idea of achieving the ‘conditional’ release of such an antisense drug: if certain diagnostic conditions are true, such as low expression levels of certain mRNAs and high expression levels of others, then the drug is released. The if–then mechanism is the new element of computation introduced in their scheme.
Benenson and colleagues' diagnosis proceeds in specific steps (transitions), one for each of the diagnostic conditions. After each transition, the state of the diagnosis is either positive (‘yes’), indicating that all conditions tested to date are true, or negative (‘no’), indicating that at least one of the conditions tested so far is false. The diagnosis can thus be viewed as a sequence of transitions of a molecular automaton that can exist in two types of state (Fig. 1). Each transition tests for high or low levels of a particular indicator molecule.
In Benenson and colleagues' experiment, which was done in vitro (that is, in an artificial environment rather than in a cell), the drug is enclosed in the loop of a hairpin-shaped oligonucleotide structure called the diagnostic molecule. Initially, the stem of the diagnostic molecule is composed of four guards, one for each of four diagnostic conditions that the authors test (Fig. 2a). A positive diagnosis for each condition results in removal of the guards, one at a time. The changing diagnostic molecule, which gets shorter with each transition as guards are sequentially removed, is the physical realization of the state of the automaton as the diagnosis progresses.
Physically, each transition is a carefully orchestrated game of shifting allegiances among several oligonucleotide ‘players’. Consider the final ‘yes → yes’ transition, which should diagnose a high level of indicator molecule 4. In addition to the diagnostic and indicator molecules, the cast of players includes a pair that we will call Alice and Eve, and a lone ranger, Bob (Fig. 2b). Eve prefers to pair with indicator 4 rather than with Alice. So if all of the players are added to a solution in which indicator 4 is present, then Eve deserts Alice and pairs with indicator 4. As a result, Alice settles for Bob (Fig. 2c). Furthermore, the Alice–Bob pair can dislodge the final guard, thereby releasing the drug (Fig. 2d). In the absence of indicator 4 in the solution, Alice and Eve would remain paired and the drug would remain captured.
How can the oligonucleotides be designed so that they are likely to shift allegiances according to this plot? Pairing, or formation of a duplex, between two oligonucleotides occurs when a sequence of nucleotides in one oligonucleotide binds to a complementary sequence in the other. Very roughly, the greater the number of nucleotide bonds between the two oligonucleotides, the more stable the duplex. Eve is designed to be complementary to a relatively long unpaired region of indicator 4, whereas Alice is complementary only to a sub-sequence of Eve. So, Eve binds more stably with indicator 4 than with Alice. Also, the Alice and Bob oligonucleotides have (somewhat shorter) complementary sub-sequences, which are likely to form a duplex once Eve has bonded to indicator 4. The game of shifting allegiances plays out somewhat differently when diagnosing low, rather than high, concentrations of an indicator molecule, although the principles are similar.
Also needed is a mechanism for cleaving the guard in the presence of the Alice–Bob pair. This pair can bind to the guard (again, via bonding between complementary single-stranded regions of Alice and the guard, called sticky ends; Fig. 2c). One additional molecule, an enzyme called FokI, recognizes a sequence pattern formed when the Alice–Bob pair is bound to the guard. FokI then cleaves the guard from the hairpin loop, releasing the drug.
Unfortunately, the specific mechanisms proposed by Benenson et al.2 would not work in a living cell: unwanted side effects of the cast of supporting molecules (particularly the FokI enzyme) would be one major problem4. Nevertheless, getting an experiment of this scale to work in vitro is a real achievement. Perhaps more importantly, the work takes a conceptual step forward, by linking the development of molecular automata to antisense therapies. It is plausible that different molecular mechanisms can be found to create diagnostic automata in the cell, building on progress made so far in the use of antisense therapies or cellular computation5,6. Developing such mechanisms would certainly be a good direction for further research.
Benenson, Y. et al. Nature 414, 430–434 (2001).
Benenson, Y., Gil, B., Ben-Dor, U., Adar, R. & Shapiro, E. Nature 429, 423–429 (2004).
Kurreck, J. Eur. J. Biochem. 270, 1628–1644 (2003).
Stein, C. A. J. Clin. Invest. 108, 641–644 (2001).
Weiss, R. & Knight, K. F. Jr in Lecture Notes in Computer Science 2054 (eds Condon, A. & Rozenberg, G.) 1–16 (Springer, New York, 2001).
Weiss, R. et al. Nat. Comput. 2, 47–84 (2003).
About this article
Nature Communications (2017)
Chemical Reviews (2015)
Angewandte Chemie International Edition (2007)
BMC Bioinformatics (2007)
Angewandte Chemie (2007)