Biochemical oscillators are fascinating systems. Watching a “soup” of reagents periodically blink or change its state is not only oddly satisfying, but it can give us a glimpse into how biochemical matter might have evolved into sustained life thanks to elementary “machines” performing autonomous, repeated tasks. For a molecular oscillator to be a useful machine in this context, it should be able to direct not only other molecular processes, but also mechanical ones that could exert forces or organize other molecules in space and time. Cytoskeletal scaffolds, for example, periodically reorganize and partition a multitude of cellular components. Could we build a minimal system in which a molecular clock directs periodic shape changes in a molecular scaffold?
This vision was at the center of many Skype meetings among young scientists in our international team, who had the bold ambition of rationally building artificial, life-like behaviors using nucleic acids as building blocks. We had just demonstrated that an in vitro transcriptional oscillator could be used to impart periodic conformation changes in an elementary DNA machine [1,2]. We were excited at the prospect of scaling up this demonstration to much larger DNA objects, gaining the ability to control self-assembly at the micron-scale using a programmable molecular clock. We identified DNA nanotubes as an excellent model system, because of their predictable assembly patterns from nanometer-sized tiles , and their structural similarities with cytoskeletal filaments. We then systematically engineered and interconnected these components, as reported in our Nature Chemistry article.
Designing a system in which a dynamic molecular circuit can activate or deactivate self-assembling elements appeared to be a challenging, but not daunting task: all the parts (transcriptional oscillator and DNA nanotubes) had been previously characterized in isolation [1,2,3], and any nucleic acid components can be in principle easily interconnected by designing nucleic acid molecules carrying out the desired reactions . Due to the predictability of Watson-Crick base pairing, there are plenty of rational approaches to design and optimize nucleic acid sequences that carry out desired assembly or kinetic tasks.
Figure 1: Architecture of our molecular oscillator directing DNA nanotube self-assembly
First, we set off to develop a method to reversibly control assembly and disassembly of DNA nanotubes: we engineered the assembling units of our DNA nanotubes, known as DX tiles , to include an overhang (toehold) at one of the sticky ends that control tile-tile binding. Toeholds make it possible for strand “invaders” to destabilize the tile-tile bonds, promoting disassembly. In turn, “anti-invader” strands displace invaders bound to the tiles, thereby reactivating nanotube polymerization. By supplying invaders and anti-invaders in repeated cycles we demonstrated reversible assembly and disassembly of the nanotubes, quantitatively tracking their mean length over time, and capturing these dynamics with a coarse-grained mathematical model .
Encouraged by the success of our strategy to control nanotube assembly reversibly, we set off to adapt our transcriptional oscillator [1, 2] to release nucleic acid strands controlling assembly and disassembly. Despite the simplicity of the oscillator, which includes only 7 oligonucleotide species (two synthetic genes, or genelets, interconnected in a negative feedback loop) and 2 enzymes (RNA polymerase and RNase H), we were hit by a multitude of system-level challenges we had not anticipated.
At the most basic level, we had to ensure that DNA nanotubes could assemble under transcriptional conditions in the presence of enzymes fueling the oscillator dynamics. A careful redesign of the DNA tiles was necessary, in particular, to minimize the effects of transcripts produced by RNA polymerase binding non-specifically to nanotubes, an unexpected phenomenon that caused many of our nanotube variants to melt in the presence of RNA polymerase .
At the level of reaction design, we had to couple the oscillator and the nanotubes with a mechanism producing enough molecules for invasion of tiles, while minimizing perturbations on the oscillator. Based on prior experience, we used an “insulator” circuit to transcribe an RNA invader to disassemble our nanotubes ; with an RNA invader, re-assembly of nanotubes can be obtained via RNase H-mediated degradation of invader bound to tiles, instead of via release of an anti-invader (Figure 1). Yet, sharing RNase H between the oscillator and the invader-bound tiles causes an indirect coupling between the two, and requires careful tuning of the oscillator components for correct operation. Further, oscillations become dampened as enzymes progressively lose activity, NTPs are depleted, and waste products accumulate.
Upon careful design and calibration of each component, we demonstrated that the outputs of an autonomous transcriptional oscillator can be used to direct temporal fluctuations of the mean length of a population of DNA nanotubes. Our models are able to capture the behaviors we observed, supporting our strategy and interpretation of phenomena. Yet, we feel that this is a partial success as we achieved at most two oscillations in nanotube length, falling short of our ambition of demonstrating multiple, dramatic cycles.
A very valuable outcome of this project, however, is that it points to a range of challenges that must be solved before we can claim that it is possible to rationally build active, dynamic molecular systems and materials. Three are, in our view, the most significant, and are related to the need of interconnecting multiple components to achieve the target dynamic behaviors. First, in the absence of compartments, distinct parts can become coupled in undesired, indirect ways that may hamper the intended dynamics of the interconnected system; careful reaction design, compartmentalization, and vascularization have the potential to mitigate this challenge. Second, components expected to operate out of equilibrium (such as oscillators) are often fragile, thus it is desirable to develop molecular “signal generators” that are insensitive to perturbations. Third, sustained dynamic behaviors require in-flow of fuel, and out-flow of waste, making it essential to design ad hoc reactions or compartments for fuel and waste management. Methods to mitigate these challenges may take inspiration from living cells as well as from traditional engineering approaches. We hope that our project will serve as a stepping stone in this direction, and promote the development of new biomaterials with integrated dynamic behaviors, and also, many stimulating conversations among ambitious and curious researchers.
 Franco, E. et al. Timing molecular motion and production with a synthetic transcriptional clock. Proc. Natl Acad. Sci. USA 108, E784–E793 (2011).
 Kim, J. & Winfree, E. Synthetic in vitro transcriptional oscillators. Mol. Syst. Biol. 7, 465 (2011).
 Rothemund, P. W. K. et al. Design and characterization of programmable DNA nanotubes. J. Am. Chem. Soc. 126, 16344–16352 (2004).
 Soloveichik, D., Seelig, G. & Winfree, E. DNA as a universal substrate for chemical kinetics. Proc. Natl Acad. Sci. USA 107, 5393–5398 (2010).
 Mardanlou, V. et al. A coarse-grained model captures the temporal evolution of DNA nanotube length distributions. Nat. Comput. 17, 183–199 (2018).
 Schaffter, S. et al. T7 RNA polymerase non-specifically transcribes and induces disassembly of DNA nanostructures. Nucleic Acids Res. 46, 5332–5343 (2018).