The Problem With Copying

Jenny Poulton

The system successfully demonstrated in the recent PNAS paper “Litters of self-replicating origami cross tiles” is impressive, but it doesn’t address two key challenges with building fully-functional autonomous Darwinian replicators.

Copying is a process ubiquitous in nature, and indeed the accurate transmission of information from the molecular sequence of one  polymer to another is the core process at the heart of the central dogma of molecular biology. Our best understanding of origin-of-life scenarios is that of simple chain molecules  that are able to produce new copies of themselves that carry the same sequence of constituent units: minimal self-replicators. If these molecules  increase in population exponentially, it allows for Darwinian competition and evolution. Synthesising a successful copying process is a non-trivial task problem which many groups, including that of Braun, Schulman and Otto, have come up with interesting solutions to.

The core issue with creating a copying process can be illustrated as follows. In order for an object to be reasonably defined as a copy of a template, it needs to 1) have a sequence that matches, or correlates with, thetemplate and 2) exist separately from the template. In nature, correlations between copy and template are encouraged by bond specificity, with “matching” bonds being much stronger than unmatched bonds. If we consider a simple system, as in figure 1, this allows the settling on the template at stages 1 and 2. However, these same bonds which encourage accuracy in stages 1 and 2, discourage separation in stage 4. It is very difficult to create a system which successfully manages both of these processes.

Professor Seeman’s new paper follows on from the group’s previous work: “Exponential growth and selection in self-replicating materials from DNA origami rafts”. In the earlier paper, the group design rafts of DNA origami, which are highly programmable, meaning that sticky ends can be designed to encourage matching at stages 1 and 2. DNA origami can also be designed to have bonds which are activated via UV light, and in this system, these are used to create the backbone bonds of the dimers; so that at stage 3, the backbone bonds are created irreversibly. Once this process has been designed, stages 1 and 2 can be achieved through cooling, and stage 4 by heating. The heating at stage 4 does not run the risk of blowing apart the dimers, because of the strong UV activated backbone bonds. This somewhat bypasses the major challenge of replication. This system manages a growth factor of 1.7, with each dimer becoming on average 1.7 dimers at the end of a cycle.

The new paper improves on this system by allowing multiple layers of dimers to be formed in one cycle by increasing the number of sticky ends per monomer to two, figure 2 illustrates this difference. With this setup, the system achieves a growth factor of 4-8.

This achievement is very impressive, but let’s consider which interesting questions about copying that this system answers. This system is profoundly non autonomous, with successive cycles of heating, UV activation and cooling rendering human intervention inevitable. One of the most interesting scenarios of self-replication is the origin of life. By bypassing the issue of creating a system which needs to both have accuracy enhancing copy template interactions, and separation of copy and template, this system ignores the question of  the conditions under which spontaneous self-replication could develop without intervention. Speculation on plausible scenarios can be instructive;  heating/cooling or wetting/drying cycles seem plausible, but Seeman’s cycle seems too complex and contrived to occur naturally.  As an example of someone who uses a plausible cycle, the work of Dieter Braun considers a temperature gradient across a capillary, encouraging thermal cycling. Because there is no external creation of an unbreakable backbone bond, the bond strengths and relative temperatures in the gradient need to be extremely finely tuned. The accumulation of polymers in this system is a genuine breakthrough in our understanding of a plausible origin of life scenario, as it gives us important information on the nature of the chemical building blocks of self replicator. We then come to the next, arguably more complex question: how can we create a system driven purely by chemistry, without any changes in the environment needed at all? Our cells do this by means of enzymes, but are there other ways, and how could they have evolved from the semi-autonomous setting with the cyclic background conditions?

Returning to the Seeman’s paper, to what extent can the system developed be truly considered copying? The mechanism is that there are AA and BB dimers which catalyse the formation of each other. An AA dimer can only create a BB dimer and vice versa. It is difficult to see how the system is generalizable to one in which one arbitrary information is propagated.  In this system, while it is possible for this system to fail to create a dimer during a cycle, the propagation of information is reduced whether an object is created or not. In this system there can be no error propagation because in the event that the object is not created, it cannot then duplicate. The fact that the system is creating dimers only means that you can exactly design the origami rafts so that they are both “end pieces”, with no probability that the system could produce polymers of alternative lengths. If the system was successfully producing trimers, like the work of Otto, then there would be a pool of “centre pieces”. Successfully designing a system which accurately incorporated the correct number of “centre pieces” relative to the template would be a much more sophisticated and generalisable information transfer, with the information being transferred being the length of the polymer.

So what questions does this system answer? The major claim of this paper is that the incorporation of the ability to increase the number of offsprings per cycle would make systems of this nature more evolutionarily competitive. While this is true, another vital component of a system’s competitiveness is its ability to self-replicate autonomously in a naturally occurring environment. It should be noted that the more offsprings per cycle, the more difficult the tradeoff between accuracy and separation would become. While the UV activated bond is clever, it allows the authors to sidestep this issue, and therefore ignore many subtleties about the issues with copying.

Ribosomes and traffic jams

Blog post by Jordan Juritz

What do rush-hour traffic jams have to do with protein production? As we are all quite familiar with,  congestion arises and the overall flow rate of cars decreases when many vehicles travel on a relatively small number of routes. In “Networks of ribosome flow models for modelling and analyzing intracellular traffic”, Nanikashvili et al. present a model that describes this same phenomenon during protein translation.

The ribosome is the molecular machine responsible for piecing together amino acids to form proteins. mRNA, a long polymer molecule, serves as the working blueprint for the protein. The order in which the amino acids are added is encoded in the sequence of nucleotides on the mRNA. Much like a car on a road, the ribosome binds around the mRNA, and proceeds to synthesise the protein as it steps along the mRNA molecule, reading its instructions from the mRNA as it goes.

A typical yeast cell contains about 60,000 mRNA molecules and about 240,000 ribosomes at any given moment. Space is tight, and the competition for mRNA among ribosomes is very high. As a result, it is expected that there will be multiple ribosomes translating a single mRNA at any one time.

The authors of this paper have employed a network of “ribosome flow models with inputs and outputs (RFMIO)” to describe the precession of the ribosomes through this highly congested system. 

A single RFMIO describes the dynamics of the probability density of ribosomes along a region of mRNA that encodes for a single protein. A single RFMIO describes the progression of a ribosome along a single section of protein-encoding mRNA. Within each RFMIO, the progression of the ribosome is bundled into an arbitrary number of steps. These steps can be fast or slow depending on the depending on the shape of the mRNA (is it straight or bundled up?) or the availability of the required resources, such as tRNAs. Therefore, the authors assign a unique rate to the transition between one step and the next, but only ever permit ribosomes to flow in the forward direction, just as cars can only drive one way down one side of the motorway. Cars travel much more slowly if there are many more cars bunched up ahead of them. Accordingly, in RFMIO models the flow of ribosomes to the next step is high (low) if the next state is free (full).

A single protein is produced once a ribosome has reached the end of an encoding region of mRNA, therefore, the outflow of ribosomes in an RFMIO model is interpreted as the protein production rate. Once a ribosome has reached the end of a single gene on an mRNA, it can stay attached to the mRNA to continue to translate another gene, or it can detach and diffuse back into solution.

The authors proved that after a long time has elapsed the probability distribution of ribosomes within a single RFMIO settles to a steady value. In this steady state, ribosomes produce proteins at a fixed, average rate. However, we know that living systems contain not just one, but many genes. Therefore, the authors considered networks of RFMIOs that enable ribosomes to be shared or recycled by many different genes. The authors proved that for arbitrary interconnection strengths the probability density of ribosomes still converges to a steady state across many different RFMIOs. As a result, many different proteins can be produced at steady rates, the ratios of which are determined by the strength and structure of the network. Crucially, outputs such as the total production rate or ribosome recycling rate are often convex functions of the interconnection parameters. The property of convexity implies that there exists an optimal protein production value that can be found easily (using efficient algorithms) even when the network of RFMIOs becomes very large. It’s like finding the highest point in a country with only one hill.

How do traffic jams arise in these systems? The figure below shows a simple network of two RFMIO connected in series. Both RFMIOs are arbitrarily made up of 3 internal substeps. In the first RFMIO, the rate at which ribosomes flow from internal step 2 to internal step 3 has been chosen to be very slow. As indicated in the corresponding graph, the probability densities after this bottle-neck are much lower than they were before, indicating a ribosome traffic jam has occurred in RFMIO 1! RFMIO 2 takes its input of ribosomes directly from RFMIO 1, therefore the slow transition in the first system has a long-range effect on the probability distribution and production rate of the second system. The implication of this is that, in the case that two genes are present on the same transcript mRNA, a bottle-neck in the ribosome flow on the first gene would reduce the production rate of the second protein. 

Figure 1: A series of RMIOs. A bottleneck in the first RFMIO causes a low turnover of proteins in the second RFMIO.

This model can be used to do more than just simulate biological traffic jams. Consider the network of parallel RFMIOs shown in Figure 2. These two RFMIOs represent two genes that compete for a common pool of ribosomes, with a proportion “v” going into RFMIO 1 and proportion “1-v” going into RFMIO 2. “v” represents the binding strength of ribosomes with the starting point of the gene represented by RFMIO 1. This model can be solved to find the binding strengths that optimise the combined protein production rate, “y”. This class of models can be used to aid the design of synthetic biological circuits to tune the production rates of the system.

Figure 2: Two RFMIOs in parallel. The pair of systems compete for a share of the total ribosome pool. The protein production rate is maximised for the most efficient resource allocation.

An understanding of the properties of flowing traffic helps engineers to design more efficient transport systems, and the same is true for engineering projects within the cell. Bioengineers employ modelling techniques, such as these networks of RFMIOs, to aid in the design of smarter, more efficient nano-machines and biological systems. Bioengineering can give rise to novel medicines and technologies, and it helps us to understand just a little bit more about how nature solved these tricky problems in the first place.

Figure 3: A complex genetic regulatory pathway, introducing metabolites and enzyme concentrations. This system can be described by RFMIOs.

References:

Nanikashvili, I., Zarai, Y., Ovseevich, A., Tuller, T., & Margaliot, M. (2019). Networks of ribosome flow models for modeling and analyzing intracellular traffic. Scientific Reports, 9(1), 1703. https://doi.org/10.1038/s41598-018-37864-1

Towards Quantitative DNA Reaction-Diffusion Chemical Networks

By Javier Cabello Garcia

Have you ever wondered how zebras got their stripes? Or why you (probably) were blessed with five toes instead of webbed feet? These and other pattern-formation phenomena in Nature are produced by reaction-diffusion systems. These systems are defined as those in which the concentration of one or more species of chemical compounds change in time and space. The change of concentration with time is not exotic, since every chemical reaction leads the transformation of one species into another during time. However, space dependency due to a diffusion factor is introduced when the system is not well stirred - i.e., concentrations are not the same in all regions of space.

Diffusion is the spontaneous spread of particles as a result of random motion in a solution. This motion produces the transition of particles from regions where they are present in high concentration to others where their concentration is lower, following concentration gradients. Concentration imbalances not only cause diffusion, but lead to different chemical reaction rates in different regions. This heterogeneity prompts the formation of patterns, which can show interesting behaviours like wave fronts or oscillations.

The pattern formation of reaction-diffusion systems is common in many biochemical processes; several lines of research have therefore addressed the design and quantitative study of these systems. DNA has been proposed several times as the ideal substrate for the production of quantitative chemical networks. The suitability of DNA comes from the predictability of the DNA interactions and the thorough characterisation of its reaction rates in well-stirred conditions. However, reaction-diffusion networks strongly depend on the diffusion speeds of each species as well. This means that for DNA networks the diffusion speed of each DNA species has to be known and tunable.

Alas, the diffusion speed of individual DNA strands is hard to control. It is largely determined by size and shape, and varies relatively little for the sort of short DNA systems typically used. Previous attempts to modify the diffusion of strands relied on making them transiently stick to a solid matrix, effectively immobilising the strands temporarily. In a recent paper, Rodjanapanyakui et al. present a matrix-free approach, were they modulate the diffusion speeds of particular DNA species in solution by specific non-covalent binding. The “modulated” strand is complementary to an “anchor” strand, which is attached to a large polymer that is free in the solution. Binding to the anchor reduces the diffusion of the modulated strand a great deal. A “competitor” strand, which can compete with the modulated strand for binding to the anchor (by a toehold exchange reaction), allows temporary release of the modulated strand from the anchor. By varying the concentration of competitors, different average diffusion rates can be obtained.

Fig. 1 Diffusion modulation system. a) Strands of the system. b) In the presence of a large excess of the competitor strand (C) the modulated strand (T) can diffuse freely. When no C is present, T binds to the anchor (A) complexes, which diffuse at a significantly slower speed. c) C release T by a toehold exchange reaction. The newly created toehold can be used by a free-in-solution T to bind again to A. Reproduced with permission from APS Physics, Diffusion modulation of DNA by toehold exchange, Rodjanapanyakul, T., Takabatake, F., Abe, K., Kawamata, I., Nomura, S. and Murata, S. Physical Review E, 97(5).

Fluorescence Recovery After Photobleaching (FRAP) was used to follow the diffusion speed variation of the modulated strand with varying concentrations of competitor and anchor. This technique involves labelling the molecule of interest (in this case the modulated strand) with fluorophores and exciting a region of the sample with a high laser intensity. This results in the bleaching of the fluorescence of the excited area. The diffusion of the labelled molecule of interest gradually restores the fluorescence in the area, and the recovery time is directly related to the diffusion speed.

Experimental data showed that by employing this diffusion modulation method, the effective average diffusion speed of a specific strand can be tuned along a sixfold variation in range. This method has even been proved to tune simultaneously the diffusion of two DNA strands without any crosstalk. The introduction of this diffusion tuning mechanism in solution opens the way for new dynamics and quantitative reaction-diffusion systems with complex behaviours and functions for DNA systems.

Read More:

-Rodjanapanyakul, T., Takabatake, F., Abe, K., Kawamata, I., Nomura, S. and Murata, S. (2018). Diffusion modulation of DNA by toehold exchange. Physical Review E, 97(5). Link:https://journals.aps.org/pre/abstract/10.1103/PhysRevE.97.052617

-Zadorin, A., Rondelez, Y., Galas, J. and Estevez-Torres, A. (2015). Synthesis of Programmable Reaction-Diffusion Fronts Using DNA Catalyzers. Physical Review Letters, 114(6). Link:https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.114.068301

-Kondo, S. and Miura, T. (2010). Reaction-Diffusion Model as a Framework for Understanding Biological Pattern Formation. Science, 329(5999), pp.1616-1620. Link:http://science.sciencemag.org/content/329/5999/1616.long

-Allen, P., Chen, X. and Ellington, A. (2012). Spatial Control of DNA Reaction Networks by DNA Sequence. Molecules, 17(11), pp.13390-13402. Link: http://www.mdpi.com/1420-3049/17/11/13390

-Padirac, A., Fujii, T., Estévez-Torres, A. and Rondelez, Y. (2013). Spatial Waves in Synthetic Biochemical Networks. Journal of the American Chemical Society, 135(39), pp.14586-14592. Link:https://pubs.acs.org/doi/abs/10.1021/ja403584p

Multidimensional biochemical information processing of dynamical patterns

Biological cells can send signals to each other by emitting signalling molecules. Cells can receive the signals by measuring the concentration of the signalling molecules using surface receptors. If the concentration does not change in time then this is just a single number, which does not transmit much information. Compare this with sending electrical signals down wires; if the voltage is maintained at a constant value, then very little information is sent. Instead, the voltage is changed very quickly and the information is encoded in the time varying pattern. i.e. you can transmit a series of bits by turning the voltage on and off. In a similar way, cells can transmit messages in the dynamical concentration of signalling molecules.

A recent paper ‘Multidimensional biochemical information processing of dynamical patterns’ by Yoshihiko Hasegawa from the University of Tokyo examines how a simple model of cell sensing can extract the most information from a time varying concentration. 

The model assumes that the cell sees is a combination of two separate signals encoded in the same molecule, one signal with a fast variation and one with a slow one (see figure). The challenge for the cell is to work out the strength of each individual component from the combined signal that it sees. The better it does, the more information it has. This challenge is complicated by the fact that the cell's readout is not 100% accurate. 

The signal is made by summing a fast signal and a slow signal

The author assumes that the cell has two decoding channels (cell-surface receptors with an associated downstream network) by which to process the signal. They find that when the cell's readout is accurate, the information is maximised when each decoder independently detects one component each. By contrast, when the cell's readout is inaccurate, it is better to have two identical decoders that are simply giving independent estimates of the overall signal strength, without trying to tease out the individual components.

The author then shows how to construct chemical reaction networks that implement the optimal decoders. They find that a possible network for the decoder is a large cascade network with additional feedback and feed-forward loops, which has been found in real biological signalling networks. They also show that the number of molecular species in these cascades can be reduced and the response function is still approximately correct.

 

Probabilistic switching circuits in DNA.

As seen in previous entries in this blog, DNA reaction networks are a substrate that enables the creation of computational systems. These reaction networks have been a research topic for quite a long time and they have their own set of advantages and operational challenges. One of the present challenges is that up to the present day, all biomolecular computing architectures have been conceived as digital deterministic ones, while actual molecular signals are analog in concentration and involve stochastic (or random) reaction events. These two signal characteristics, when controlled in space and time, allow for the emergency of complex phenomena (such as patterning or morphogenesis). Besides this, it is considered that the integration of the analog/stochastic architectures would expand the capabilities of DNA computing.

Qian’s group present the first step towards the integration of stochastic/analog systems in DNA circuits. The implementation is done through irreversible strand displacement reactions in which a given input can bind to 2 different gate molecules with a probability that depends on the gate concentration. The reaction with one gate or the other leads to two different outputs. This architecture, implemented in switches and signal splitters, works stochastically in the single molecule level. But, when observed in bulk, the stochastic elements in the circuit allow the transformation of a digital signal of n bits into an analog signal able to take 2ⁿ values.

Figure 1. Example of a circuit with 3 stochastic elements (2 splitters and 1 gate), and the different analog signal values that it can output.

This architecture has been proved to be able to allow the implementation of feedback loops that allow adjustment of the intensity of the output signal. Despite some limitations (in larger circuits, fine tuning of the probability becomes harder thus requiring correction of the initial input signal), the proposed probabilistic circuits are a bridge that allows the interconnection of digital and analog computational circuits.

References:

Wilhelm, D., Bruck, J., & Qian, L. (2018). Probabilistic switching circuits in DNA. Proceedings of the National Academy of Sciences, 201715926.

Paper Summary: Thermodynamics with continuous information flow

Paper Summary:
Thermodynamics with continuous information flow (2014) J M Horowitz and M Esposito

This paper discusses a way of quantifying the flows of information in the same way as flows of energy. Consider two random processes which hop between discrete states. Two independent systems of this type can be coupled together in a way so that the state of the first system, x, affects the transition propensities of the second system, y, and vice-versa. At any given instant, either x or y could change, but crucially not both. This property defines the system as bipartite. The diagram below (taken from the paper) illustrates the creation of a bipartite system from x and y.

Examples of systems of this type could include a ligand bound to a receptor which catalyses the creation of an activated complex. It is not possible for a ligand to bind to a receptor, and an activated complex to be created in a single step: they have to happen sequentially. However it is true that whether or not a ligand is bound to the receptor affects the rate with which activated complexes are formed.

If a system is bipartite all properties of the system to be decomposed into an x-subsystem component and the y-subsystem component, separating out the contributions from transitions involving the two subsystems. There are many useful properties of a system such as this, and for a rigorous thermodynamic analysis, a key one is the entropy. The second law of thermodynamics states that the entropy of the universe must increase with time. However this is only for the universe as a whole. On a more microscopic level it is possible for some parts of the system to decrease entropy, as long as other parts make up the shortfall. Traditionally we break the change in entropy down into the change in entropy of the system itself and the change of entropy in the environment due to the system. If we further decompose into separate contributions for our two subsystems an extra term appears for each,  quantifying the information flow around the system.

To the x component a term is added which takes into account how much information x learns about y during a single step. Similarly, to the y component a term is added which takes into account how much information y learns about x during a step. Splitting the entropy in this way gives a sense of how the information moves around the system; it flows from x->y or from y->x. Incorporating the information term allows us to apply the second law individually to each subsystem, rather than only at the level of the whole system. It also allows us to understand how a decrease in entropy in one subsystem can be caused by information flowing from one subsystem to the other, in a manner that is consistent with the second law. 

Recent Developments in DNA Origami

Javier Cabello – PhD student

Using both technique and imagination, origami allows the production of many shapes with only a folded sheet of paper. After the birth of DNA nanotechnology, the potential of this folding art inspired Paul Rothermund to develop a technique that allows the creation of complex nanoscale objects out of DNA. A single DNA strand is a flexible material that can be folded by hundreds of smaller strands (“staples”). These staples bind to specific non-consecutive regions of the larger strand (“template”) by Watson-Crick base-pairing, folding the template into a DNA structure with a particular shape.

DNA origami structures have a wide variety of potential uses that greatly differ from the original biological role of the molecule. These range from drug delivery to the production of electric circuits, with the biocompatible nature of DNA suggesting application in living organisms. However, the structures created are still limited by the size of the template, which becomes harder to fold when longer. Another problem with scaling up the technology is the cost of synthesising large numbers of staple strands.

Last November, four publications in Nature focused on overcoming the limitations of DNA origami. Three of the publications proposed new approaches for the assembly of bigger DNA structures. All three involved combining blocks of different sizes and complexity to form bigger structures.

Gigadalton-scale shape-programmable DNA assemblies

The first approach, suggested by Wagenbauer, Sigl and Dietz, involves making origami structures that can then self-polymerize. The resultant structures depend on the geometry of the individual origami building blocks. This approach, heavily inspired by viral capsid assembly, was used to form DNA rings from V-shaped DNA components. Afterwards, these rings were assembled to form DNA tubes. The method was generalised to produce different figures, as dodecahedra, by varying V-block angles and adding blocks with more than two binding sites.

Fractal assembly of micrometre-scale DNA origami arrays with arbitrary patterns

Tikhomirov, Petersen and Qian’s work revolves around the design of a robust algorithm to produce ordered assembly of complex patterns in 2D. In this work, individual origami tiles were first folded in isolation. Tiles were then combined in groups of four to produce larger “super-tiles”. Successively, four of these assembled larger tiles were employed to produce even bigger tiles. To achieve the necessary control of this process, the authors used three basic design rules, which they incorporated into a computational tool that they named the “FracTile Compiler”. This compiler allows the systematic design of complex, puzzle-like patterns. The resultant array can show an arbitrary picture, displayed by unique modifications to the surface of each tile, as they do with “La Gioconda” (ed.: the Mona Lisa for English readers) among others (see fig. 1).

Fig. 1 Complex shapes created using the method of Tikhomorov et al., reproduced with permission from Springer Nature, Fractal assembly of micrometre-scale DNA origami arrays with arbitrary patterns, G Tikhomirov, P Petersen, L Qian, Nature 552 (7683), 67.

Programmable self-assembly of three-dimensional nanostructures from 10,000 unique components

In this case, Ong et al improved a previously published variant related to DNA origami: DNA bricks. Instead of using large DNA strands as in previous approaches, the authors design structures built from smaller building blocks, of only 56 nucleotides each. Each of these bricks can bind to three others, forming a dense “lego”-like structure. By using 10,000 different sequences for these DNA bricks, the authors achieve the production of cubes of more than 1 GDa (3000 tunes the mass of the largest proteins). These cubes were proposed to serve as a canvas to produce different structures by simply removing bricks from the design, like chiselling a rock. This approach even allows the production of negative structures by carving cavities inside the DNA cube. To help in the design task, the authors created the software “NanoBricks” and proved its utility by producing different patterns. such as a helix or a teddy bear inside the DNA cube.

Biotechnological mass production of DNA origami

Despite these significant improvements in the production of larger DNA origami structures, all their possible applications are being hindered by the pecuniary costs involved in synthesising the component strands. However, new methods inspired by the current biotechnological industry are being developed to mass produce DNA origami designs.

Praetorius et al. have demonstrated the possibility of biotechnological mass production of a DNA origami design. Their approach involves encoding all the components necessary for the folding of an origami (ie. template and staples) in a single genetic sequence, and incorporating it into bacteria. Of course, this isn’t all. Between the different components they added DNA sequences with enzymatic activity: DNAzymes. The whole DNA strand is then produced in a culture of bacteria, and the DNAzymes, in the presence of zinc, cut themselves out of the sequence, separating it into the desired components. The result is that the staples and template, now separate, can assemble themselves to form the intended DNA origami structure.

This new approach, if escalated, would drastically reduce the production costs of DNA origami designs. However, the production of the initial genetic material would still require an expensive synthesis of the desired sequences for each new design produced. In order to make this solution even more affordable the produced DNA origami should be a very general or standardized one which can be customized with minor changes. Who knows if the answer lies in any of the previous papers? My humble guess is that DNA bricks are good candidates.

Read more:

- Rothemund, P. (2006). Folding DNA to create nanoscale shapes and patterns. Nature, 440(7082), pp.297-302. Link: https://www.nature.com/articles/nature04586

- Wagenbauer, K., Sigl, C. and Dietz, H. (2017). Gigadalton-scale shape-programmable DNA assemblies. Nature, 552(7683), pp.78-83. Link: https://www.nature.com/articles/nature24651

- Tikhomirov, G., Petersen, P. and Qian, L. (2017). Fractal assembly of micrometre-scale DNA origami arrays with arbitrary patterns. Nature, 552(7683), pp.67-71. Link: https://www.nature.com/articles/nature24655

- Ong, L., Hanikel, N., Yaghi, O., Grun, C., Strauss, M., Bron, P., Lai-Kee-Him, J., Schueder, F., Wang, B., Wang, P., Kishi, J., Myhrvold, C., Zhu, A., Jungmann, R., Bellot, G., Ke, Y. and Yin, P. (2017). Programmable self-assembly of three-dimensional nanostructures from 10,000 unique components. Nature, 552(7683), pp.72-77. Link: https://www.nature.com/articles/nature24648

- Praetorius, F., Kick, B., Behler, K., Honemann, M., Weuster-Botz, D. and Dietz, H. (2017). Biotechnological mass production of DNA origami. Nature, 552(7683). pp.84-87. Link: https://www.nature.com/articles/nature24650