Introduction
Endowing living organisms with new functions is difficult for several reasons. First, biological entities are complex open systems that operate far from thermodynamic equilibrium. Second, although we can control and define the components in a test tube, this is much harder to do in vivo. We rely on adding genetic material to the organism and controlling regulation of the new genes to alter the lifetime and expression of the protein or RNA components encoded. Third, whereas it is relatively straightforward to measure rates and equilibrium constants in vitro, these quantities are generally not easy to ascertain in vivo, making quantitative simulation and prediction of behavior difficult. Fourth, unlike in bulk in vitro reactions, the number of some types of molecules in individual cells is small and variable such that stochastic effects may dominate behavior. Fifth, though we can control the specificity of molecular processes in vitro by simply omitting or adding components, we do not know the cell-wide specificity of interactions for any molecule we might add to the cell. Furthermore, we cannot yet predict a priori the level of molecular specificity required to avoid the potential pleiotropic effects of added molecules. This is a particular concern given the variety and high concentrations of proteins, metabolites, RNA and DNA in the cell and the non-ideality of the cellular environment.
Despite these challenges, there have been recent notable successes in programming biology. Two recurring themes in these efforts are (i) the creation and use of modules to carry out discrete functions and (ii) the logical combination of modules with each other or with the cellular network to perform higher-order functions. The use of modules is inspired by the proposition that natural evolution may have arrived at cell networks that are, to some extent, modular1, 2, and also by the apparent similarity in construction principles between living matter and modular human constructions1. Modules are small collections of molecules whose function is discrete and separable from the rest of the cell. For example, a transcription factor and its DNA-binding site might constitute a module.
Efforts to reprogram biology fall into two broad categories. First, there have been efforts to explore the assembly of well-characterized naturally occurring modules into novel networks that may have unnatural and potentially emergent behavior. Second, there have been efforts to create unnatural modules and to use these modules to further expand the range of organismal function. Here we review recent work in each of these areas.
Combining natural modules to produce unnatural functions
New functions from repressor circuits. Several recent experiments have underscored the power of manipulating network topology within a cell to create new function. Researchers have used three well-characterized transcription repressor proteins (lac repressor, tet repressor and 
cI repressor) and their cognate DNA-binding (operator) sites as basic modules. Notably, each module is mutually orthogonal with the other modules such that there is little unintentional cross-talk. For example, the lac repressor acts on promoters containing the lac operator sequence but does not repress transcription from the tet or the cI operator sequence; a similar relation exists among other repressors and promoters.
Using two repressors (lacI and cI) arranged to mutually repress each other's transcription, Collins and co-workers created a genetic toggle switch3 (Fig. 1a) in which a protein's expression could be set to an on or off state by applying one of two transient stimuli: heat to inactivate cI or IPTG to remove lacI repression. Once set by the first stimulus, the protein expression level was stably inherited from cell to cell in the absence of the stimulus. The transient application of a second stimulus reset the protein's expression to its original level, and this state was then stably inherited in the absence of the second stimulus. The population of cells therefore 'remembered' (that is, epigenetically inherited) the state set by the last stimulus applied. A deterministic model using ordinary differential equations was used to predict the parameters and regimes necessary for bistability and stable switching, and the experimental components were chosen accordingly.
Figure 1: New functions from repressor circuits.
(a) Green fluorescent protein (GFP) expression controlled by a genetic toggle switch in response to heat and IPTG. (b) Interfacing of the toggle switch with the cell's network. Left, the sensor-toggle switch circuit. Right, biofilm formation with (bottom) and without (top) DNA damage. Ptrc and PL* are lacI- and cI-responsive promoters, respectively. (c) Oscillations in GFP expression in single cells as a function of the oscillator circuit. Top, fluorescence and bright-field images of cells containing the oscillator circuit. Bottom, fluorescence of a single cell as a function of time. The approximate timing of cell division events is indicated by the dashes next to the x axis. Adapted from refs. 3, 4 and 6 with permission.
In addition to switching in response to IPTG or heat, the toggle switch could also be toggled by transiently increasing the activity of the transcription factor whose activity is repressed, or by transiently decreasing the activity of the transcription factor that is highly expressed. Collins and co-workers showed that these properties allow the toggle switch to be interfaced with the cell's natural networks4. In one example, they linked the SOS DNA repair pathway with traA-mediated biofilm formation via the toggle switch and observed cells permanently switching to a biofilm-forming state in response to the transient presence of a DNA-damaging agent (Fig. 1b).
A synthetic genetic toggle switch that is based on the same principles has recently been created in mammalian cells and mice5. Such epigenetic switches have potential applications in biotechnology and medicine. For example, they could control developmental fate through the transient introduction of small molecules. Moreover, if mammalian toggle switches can be interfaced with the cell's network, it should be possible to track the fates of individual cells (and their daughters) within multicellular organisms in response to diverse stimuli. This could allow correlation of cell fate with prior exposure to stimuli.
Elowitz and Leibler used three repressors (lacI, cI and tetR) to form a network in which each repressor exclusively represses the transcription of one other repressor, creating a cyclic negative feedback loop. Their goal was to create bacteria that expressed a protein whose concentration oscillated autonomously6. Using a continuous deterministic model in an approach similar to that used to design the toggle switch, the investigators predicted parameters that would support sustained limit-cycle oscillations rather than a stable steady state, and they tuned the modules accordingly. A substantial proportion of single cells that were transformed with the designed circuit showed periodic oscillations in the expression of fluorescent protein for up to 10 h (Fig. 1c), and the oscillations were uncorrelated to cell division events. These experiments show that an oscillator can be assembled in Escherichia coli from components that are not part of any biological clock and using a network topology that is not part of any known circadian oscillator. Other networks that produce toggle switch and oscillatory behaviors have recently been described7, and Liao and colleagues have created a metabolic-transcription oscillator that allows two metabolic pools to oscillate in response to excess glycolytic flux8.
Unlike natural biological clocks, the designed repressor-oscillator network was very noisy, and the average t1/2 for decorrelation of sibling cells' oscillations was approximately two cell divisions. The continuous deterministic model did not predict these properties. This highlights a limit of such models: they can predict the average properties of cellular ensembles or the time-averaged behavior of single cells, but not the behavior of members of a population or the time-dependent behavior of single cells9. A more realistic model—one that includes stochastic interactions and accounts for the discrete, rather than continuous, nature of molecules—does predict the decorrelation. The origins of stochasticity (or noise) and population heterogeneity have now been addressed experimentally10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and Xie and co-workers recently measured individual protein expression events in living cells20. An improved understanding of the way in which stochasticity affects emergent behavior and is exploited by biological systems21 will facilitate the assembly of genetic circuits that perform new functions. For example, to create circuits that function homogeneously across a population, it is clearly preferable to use modules, such as autorepressors (in which a transcription factor negatively regulates its own expression), that have minimal cell-to-cell heterogeneity10. For other types of synthetic circuits it might be advantageous to exploit probabilistic behavior21.
In contrast to the rational design of the toggle switch and oscillators, which were aimed at endowing cells with specific epigenetic and dynamic properties, Leibler and colleagues explored the range of logical outputs produced by the combinatorial rearrangement of the genes for three repressors (lacI, cI and tetR) and five promoters, each of which is regulated by only one of the repressors22. They created a library of 512 networks and evaluated the responses of individual networks to the inducers IPTG and the tetracycline analog aTc. A variety of logical behaviors were observed, including NAND, NOR and NOT IF. Notably, circuits with different logical properties can arise from one-step differences in connectivity, a fact that may speak to the 'evolvability' of networks to produce new phenotypes once a 'toolkit' of regulatory modules is in place. The results of Leibler and colleagues also show that networks having connectivity diagrams of the same structure can perform different logical functions, whereas networks with different connectivity diagram structures can produce similar logical outputs. Overall, these results demonstrate the difficulty of inferring logical function from connectivity diagrams, and they also re-emphasize the importance of understanding the molecular details of components when creating new logical functions.
Cell-cell communication. Several groups have exploited natural cell-cell communication machinery to coordinate new behaviors in populations of cells. Most of this work has used a module derived from the natural quorum-sensing mechanisms of Vibrio fischeri. This module is composed of LuxI (which converts metabolites to N-3-oxohexanoyl-L-homoserine, AHL, which freely diffuses between cells), LuxR (a transcriptional activator that is dependent on AHL for activity) and the LuxR-responsive LuxI promoter. In response to high concentrations of AHL (for example, at high cell density), transcription of genes placed downstream of LuxI is activated.
Arnold and colleagues used this system to create a population-control circuit23. E. coli were endowed with constitutively expressed luxI, luxR and the LuxI promoter; the latter was used to drive the translation of a toxic gene fusion (lacZ
-ccdB). At low cell densities, the accumulation of AHL inside the cells is insufficient to activate transcription of lacZ-ccdB at levels necessary to kill cells. At high cell densities, high intracellular AHL concentrations lead to substantial activation of lacZ-ccdB, resulting in cell death. The circuit therefore allows cell growth up to, but not beyond, a certain cell density.
Arkin and co-workers have set out to create a strain of E. coli that specifically invades tumor cells. This work is based on the observation that upon intravenous injection in mammalian hosts, a variety of bacteria, including nonpathogens, localize to tumors and grow to densities of 109 cfu per g tissue, which is three to six orders of magnitude higher than in other host tissues24. Such tumor-invading bacteria could be used to deliver cytotoxic agents (which may themselves be produced by the bacteria). The lux module was used to create density-dependent invasion, and the LuxI promoter was used to drive expression of the invasin gene from Yersinia pseudotuberculosis. At high densities of E. coli, the invasin protein is produced at sufficient levels to cause adhesion and internalization of E. coli into mammalian cells expressing 
1-integrin.
Weiss and colleagues split the lux module to create two distinct types of cells—sender cells and receiver cells—and used these to explore synthetic cell pattern formation as a function of space and time (Fig. 2)25. The sender cells contained LuxI and synthesized AHL, which diffused into the surrounding medium. The receiver cells contained a feedforward loop with LuxR as the master regulator; in this loop, GFP expression was repressed either directly or indirectly by lacI. The receiver circuit allowed cells to produce a fluorescent protein in response to a narrow range of AHL concentrations. The investigators created three strains of receiver cells by using mutants of luxR with altered activity and by varying luxR copy number. Each strain responded to different ranges of AHL concentration. When a disk of sender cells is placed in a Petri dish containing a lawn composed of all three types of receiver cells, fluorescent bands form at discrete distances from the sender cells; the bands correspond to the AHL response range of each strain. The observed pattern shows that the lawn of genetically distinct but spatially undifferentiated cells has been spatially differentiated by the diffusive gradient generated by the single signal from the sender cells. Similar principles have been used to create transient responses to AHL signals26. Recently Collins and co-workers interfaced their genetic toggle switch with quorum sensing to produce cells that retain a memory of whether they have ever been grown at high density4. It should be possible to extend these approaches for use in the creation of artificial multicellular morphologies to be used for tissue engineering and for imprinting and fabricating materials. However, the greater sophistication and conditional signaling required for such applications will require additional orthogonal modes of cell-cell communication27, 28 as well as methods to spatially distinguish genetically identical cells.
Figure 2: A pattern-forming circuit.
(a) Sender cells broadcast AHL, and band detector cells respond to an AHL window. Band detector cells contain a feedforward loop that leads to the observed behavior by virtue of the differential repression thresholds for GFP expression in each branch of the loop. Blue stars represent tetracycline and purple circles AHL. Red lines and green arrows indicate functionality and activation, respectively. White-filled lines indicate ineffective activation or repression. White-filled shapes indicate that the corresponding protein is produced at decreased levels. (b) An example of a pattern formed by two different band-forming strains that initially form an undifferentiated lawn of E. coli on an agar plate. Adapted from ref. 25 with permission.
Full size image (77 KB)New cellular function from unnatural modules
Almost all of the work in unnatural circuit construction has relied on the rearrangement of natural transcription factor modules into new networks. This approach, in combination with modeling, has taught us much about building genetic circuits having topologies and functions not found in biology, and it is far from exhausted. However, it is clear that to fully realize the potential of living matter to be used for human purposes, it will be necessary to create new modules that perform functions not found in biology. Moreover, the creation of increasingly sophisticated circuits for programming cells with more complex behaviors will require an expansion in the number of well-characterized, programmable modules that function without cross-talk.
Reprogramming signal transduction. Cells respond to environmental cues by activating signal transduction cascades that control gene expression. Several strategies have recently been used to alter the input-output relationship for signal transduction modules. One way to alter the relationship between input and output in signal transduction is to alter the specificity of a natural receptor while leaving the signaling apparatus intact. Using a designed, chimeric photoreceptor that activates a two-component signaling pathway arranged to activate the transcription of lacZ, Voigt and co-workers created E. coli that could produce a black output in the dark but remained uncolored when exposed to light29. By projecting patterns of light onto a lawn of these bacteria grown on a solid medium containing S-gal (3,4-cyclohexenoesculetin--D-galactopyranoside), which is converted to a black precipitate by -galactosidase, they created a two-dimensional chemical 'photograph'. Goulian and co-workers recently isolated bacteria that show chemotaxis toward unnatural chemoattractants. They used a clever selection strategy in which they plated E. coli transformed with a library of chemotaxis receptor mutants at one end of a diffusive gradient of unnatural chemoattractant and then isolated those that moved toward the attractant30.
Hellinga and co-workers have pioneered a computational method that uses a dead-end elimination algorithm to alter the specificity of members of the E. coli periplasmic-binding protein family that form part of a two-component signal transduction module. The proteins are altered such that they bind unnatural molecules (Fig. 3a)31. Dead-end elimination algorithms32 allow an accelerated computational search of combinations of amino acid side chains that are found at different positions in the primary sequence by excluding from calculations side chain rotamers at each position that cannot contribute to the global energy minimum.
Figure 3: Examples of unnatural signal transduction.
(a) A TNT sensor. Left, the chemical structures of TNT and TNT analogs discriminated against by mutant RBPs. Right, the RBP variant bound to TNT (1) interacts with the Trg domain (thick black line) of a chimeric transmembrane histidine kinase, Trz, resulting in (2) autophosphorylation of the EnvZ domain (gray line) and (3) phosphate transfer to OmpR, which then (4) binds to the ompC promoter, upregulating lacZ transcription. OM, outer membrane; PP, periplasm; IM, inner membrane; CP, cytoplasm. (b) Left, rewiring scaffolds in MAP kinase pathways. The general MAP kinase cascade. Middle, natural scaffolds control the specificity of MAP kinase pathways. Right, diverter scaffolds allow the input-output relation of MAP kinase pathways to be rewired. Adapted from refs. 31 and 33 with permission.
Full size image (61 KB)In a matter of days, Hellinga and colleagues produced a ranked list of mutant sequences predicted to bind target ligands including trinitrotoluene (TNT), L-lactate and serotonin. They found a TNT-binding variant of E. coli ribose-binding protein (RBP) in the top three computational choices. The RBP variant bound TNT with a Kd of 2 nM and discriminated 50-fold against trinitrobenzene and more than 1,000-fold against 2,4-dinitrotoluene (2,4-DNT) and 2,6-DNT. By expressing this TNT-binding variant, the authors showed that cells can respond to TNT in the nanomolar range. Numerous potential applications of such sensing modules can be envisioned for the detection of environmental toxins and warfare agents.
Another way to manipulate the specificity of signal transduction is to alter the specificity of the transduction cascade. Lim and co-workers rewired a mitogen-activated protein (MAP) kinase pathway in Saccharomyces cerevisiae such that the presence of alpha factor, which normally leads to the mating response, conferred osmotic resistance (Fig. 3b)33. Scaffold proteins contain binding sites for each of the kinases in a particular pathway, as well as for upstream input proteins and downstream output proteins. The mating and osmotic-response pathways use different scaffold proteins but share a common kinase, Ste11. Lim and colleagues created an unnatural 'diverter' scaffold by fusing the mating-response scaffold Ste5 with the osmotic-response scaffold Pbs2 and mutating the chimera to destroy interactions with the downstream mating output (Ste7) and upstream osmolarity sensors (Sho1). The two pathways were connected by virtue of their common node, Ste11, allowing a wild-type osmotic response to be triggered by alpha factor. This result shows that, to a first approximation34, chimeric scaffolds are sufficient to confer specificity on signal transduction pathways, suggesting a route to creating new scaffold modules for reprogramming signal transduction.
Genetic code expansion. Nowhere is modularity more important than in maintaining the fidelity of the genetic code. Each of the 20 amino acids is loaded onto a unique set of isoacceptor tRNAs by 1 of 20 aminoacyl-tRNA synthetase enzymes and incorporated into peptide chains on the ribosome, and any cross-talk in these interactions is catastrophic (Fig. 4a). Schultz and co-workers have site-specifically incorporated unnatural amino acids into proteins in living prokaryotic and eukaryotic cells35, 36. They have approached this problem in two steps. In the first step, a new module composed of (i) an orthogonal tRNA that decodes a blank codon (in most cases the amber stop codon) and is not a substrate for the endogenous aminoacyl-tRNA synthetases and (ii) an orthogonal aminoacyl-tRNA synthetase that exclusively aminoacylates the orthogonal tRNA (but no other cellular tRNA) with a natural amino acid is added to the cell36, 37. These orthogonal tRNA synthetase pairs are derived from heterologous organisms, exploiting the fortuitous divergence of sequence and specificity in the evolution of different species38. With the orthogonal synthetase and tRNA in hand, Schultz and colleagues altered the specificity of the synthetase enzyme so that it exclusively aminoacylated its cognate tRNA (but no other tRNAs) with an unnatural amino acid. To achieve this, they used one of three variations of a double-sieve selection (or screen), in which suppression of the amber stop codon is either positively or negatively selected for in the presence or the absence, respectively, of an unnatural amino acid (Fig. 4b)35, 37, 39, 40. Numerous useful unnatural amino acids, including photo-cross-linkers, chemical handles, heavy atoms, redox sensors, post-translational modifications and fluorophores, have been added to the genetic code of both prokaryotic and eukaryotic organisms41. Yokoyama and co-workers developed tRNAs that allow this strategy to be applied to mammalian cells42, 43. The strategy has recently been extended to the simultaneous site-specific incorporation of two unnatural amino acids in response to a four-base codon and an amber codon44. Schultz and coworkers have also created an autonomous 21-amino-acid bacterium45 that biosynthesizes p-amino-L-phenylalanine (p-AF)from simple carbon sources using a heterologous biosynthetic pathway (Fig. 4c) and site-specifically incorporates the unnatural amino acid into proteins.
Figure 4: Expanding the genetic code.
(a) Modularity in maintaining and expanding the genetic code. Each natural amino acid (yellow circle) functions with (black line) one cognate synthetase (blue circle), which aminoacylates (black line) only isoacceptor tRNAs (green circles). Each aminoacylated tRNA functions at the ribosome. The addition of a new amino acid (orange circle), synthetase enzyme (red circle) and tRNA (purple circle) to the cell requires that none of the numerous potential interactions denoted by gray lines take place and that the three new components function with each other and with the ribosome. (b) An example of a double-sieve selection used to expand the genetic code. By placing selector codons in GAL4 in S. cerevisiae, expression of a variety of reporter genes is linked to amber suppression. To select mutant synthetase–tRNA pairs able to incorporate amino acids in response to a selector codon, cells are first grown in the presence of unnatural amino acids, and then clones expressing an essential reporter gene URA3 or HIS3 are selected. To select clones that exclusively incorporate the desired unnatural amino acid, rather than one of the 20 natural amino acids, the unnatural amino acid is omitted from the medium, and cells that do not express a conditionally toxic gene (for example, URA3 in the presence of 5-FOA (5-fluoroorotic acid) are selected. (c) An autonomous 21-amino-acid bacterium. PapA, papB and papC convert chorismate, 2, to p-aminophenylpyruvic acid, then the E. coli aromatic aminotransferase completes the biosynthesis to afford p-AF, 1. A p-AF–specific aminoacyl-tRNA synthetase incorporates p-AF into proteins in response to the amber codon. Adapted from ref. 37 with permission.
Full size image (73 KB)Post-transcriptional control of gene expression. Methods for creating additional gene regulation modules that function without cross-talk are required for the synthesis of more complex logical and dynamic networks. Recent work has focused primarily on creating modules for this purpose that make use of post-transcriptional control of gene expression.
Bayer and Smolke created ligand-dependent riboregulators that act in trans to repress gene expression in S. cerevisiae (Fig. 5a)46. Their strategy relies on combining, in a single small RNA, known ligand-binding aptamers derived from in vitro selection experiments with an antisense sequence. In the absence of ligand, the small RNA is predicted to preferentially adopt a structure in which the antisense sequence is sequestered by base pairing. In the presence of the small-molecule ligand, the small RNA undergoes a conformational switch to a form that can bind to a specific mRNA sequence and function as an antisense agent. In the presence of high concentrations of ligand (1 mM), gene expression can be completely repressed to background levels. Moreover, Bayer and Smolke demonstrated the modularity of the strategy by creating two distinct ligand-binding aptamers that targeted two distinct mRNAs; they then showed that the aptamers could independently modulate gene expression in the same cell.
Figure 5: Synthetic post-transcriptional gene regulation.
(a) The design of antiswitches. Small RNAs show poor binding to their cognate mRNAs. On addition of specific small molecules (theophyline or tetracycline) to the cell, the cognate antiswitch is activated and represses gene expression. (b) The design of a synthetic riboregulator. RBS, ribosome binding site; cr, cis repressor sequence; taRNA, transactivating RNA. Adapted from refs. 46 and 47 with permission.
Full size image (51 KB)Collins and co-workers have explored the potential of synthetic riboregulators as post-transcriptional gene regulation modules (Fig. 5b)47. Their strategy involved creating repressor sequences that are appended to the 5' end of a natural transcript and are designed to form hairpins with the natural sequence containing the ribosome-binding site in E. coli. These additional sequences repress gene expression by occluding ribosome binding. Gene expression is reactivated by the expression, in trans, of a second RNA that binds to the cis repression sequence and frees the natural ribosome-binding site sequence for translation. Cis repressor sequences that base-pair with the natural sequence (but are also interspersed with mismatches) provide basal repression in excess of 95%, which compares favorably to the basal repression achieved with antisense and transribozyme systems. Furthermore, mRNA translation can be increased by approximately an order of magnitude by transactivating RNA, and, importantly, in the two cases examined no transactivation of noncognate cis-repressed sequence was observed. This technology shows promise as a scalable route to gene regulation. Because the method does not, in principle, require a specific promoter or target a specific coding sequence, it should be quite versatile. It might be possible to regulate gene expression from native promoters within the genome, and this may allow detailed studies of gene regulation networks.
Rather than targeting post-transcriptional accessibility of mRNA, Rackham and Chin targeted the translational machine itself: the ribosome48. In E. coli and other bacteria, the ribosome is directed to mRNA via interactions between rRNA and mRNA sequences 5' to the start codon (Fig. 6a). From the outset, the aim was to create new versions of the ribosome (orthogonal ribosomes) that exclusively translate mRNAs containing altered mRNA-binding sites that are not substrates for the endogenous ribosome (orthogonal mRNAs). To achieve this, Rackham and Chin created a general two-step selection method that allows either positive or negative selection of cis or trans effectors of gene expression in response to one of two small molecules. The selection uses a cat-upp gene fusion. In the presence of chloramphenicol, clones expressing the fused gene survive by virtue of its chloramphenicol acetyl-transferase activity. In the presence of 5-fluorouracil (5-FU), the cells expressing the fusion are poisoned by virtue of uracil phosphoribosyl transferase–mediated conversion of 5-FU to toxic products.
Figure 6: Orthogonal translation.
(a) Evolving orthogonal ribosomes by duplication and selection. Duplication followed by positive and negative selection leads to three sets of ribosome mRNA pairs that are orthogonal to the endogenous pair. Several of the unnatural ribosome-mRNA pairs are orthogonal with respect to each other. Gray lines correspond to weak or unmeasurable interactions; black lines correspond to strong functional interactions. (b) An example of a translational logic gate. Production of ribosomes A and B is required to reconstitute -galactosidase from the 
and 
fragments, as measured by the conversion of fluorescein di--D-galactopyranoside to fluorescein. Adapted from refs. 48 and 49 with permission.
Rackham and Chin placed a library of mutant ribosome-binding sites upstream of cat-upp and selected orthogonal mRNA sequences on 5-FU. They then selected mutant 16S rRNA sequences that assemble into mutant ribosomes and drive efficient translation from the cognate orthogonal mRNAs, and they showed that these new ribosomes do not appreciably translate endogenous transcripts. Three mutant ribosome-mRNA modules were found that are orthogonal to the endogenous ribosome and mRNAs. Two of these pairs were mutually orthogonal with respect to each other, and these pairs have been used to create translational AND and OR gates (Fig. 6b)49. Moreover, orthogonal ribosome-mRNA pairs have been a key tool in large-scale mapping of functionally important nucleotides in the 6,000-Å2 ribosome subunit interface50.
The orthogonal ribosomes and binding sites provide portable post-transcriptional gene regulation modules for regulating any gene in E. coli. These modules are free of the requirement to translate the proteome, and this should allow directed divergence of the ribosome to perform new functions. Indeed, efforts are underway to further expand and reprogram key aspects of ribosome function in the context of orthogonal ribosomes. The selection methods developed to create orthogonal ribosome-mRNA modules can be used to create other cis or trans gene-regulatory modules and by extension signal transduction modules. Furthermore, through temporally controlled application of selection agents, it should be possible to directly select for gene networks displaying particular switching, dynamic or logical behavior.
Future directions
The assembly of large DNAs encoding any potential unnatural circuit is likely to become routine as a result of the decreasing cost of both large-scale oligonucleotide synthesis51 and methods for template-independent assembly52, 53, 54, 55 and error correction51, 56. More complex circuits allow the creation of more complex function but also bring a new set of challenges. A central challenge will be the creation of new modules having prescribed specificity with respect to every molecule in the cell's network and also with respect to other modules. In general, orthogonal modules will be useful, but modules that specifically interface with the cell will be crucial for interfacing synthetic networks with existing biology. Poor impedance matching (the mismatch in the strength of a signal output from one module and the input required by another module) is a problem in synthetic networks, but engineering principles such as standardization of modules57, in combination with evolutionary approaches24, 58, may greatly facilitate progress in this regard.
For more complex circuits it may be necessary to actively manage the cell's resources to deal with the increased metabolic load imposed on the cell by the circuit. One potential class of solution is suggested by efforts to improve the yield of lycopene in a particular biosynthetic pathway. Lycopene biosynthesis has been coupled to systems that detect the cell's metabolic state and adjust the resources available for production accordingly, leading to an increase in lycopene yield59. Another class of solution might be the use of cells that are able to divert a larger fraction of their activities to synthetic circuits by virtue of accessing a quiescent state60, 61.
Designing increasingly complex circuits will require a better understanding of design principles in biology and models that deal with the stochastic nature of biological processes62. Self replication and evolution to perform higher-order functions differentiates living matter from nonliving matter and creates challenges in redirecting the purpose of living matter. These challenges have not been faced in engineering nonliving matter and may require new approaches for creating molecules, modules, network motifs and networks that perform new functions. Finally, if we can endow nonliving matter with purpose and the ability to self replicate63, it may be possible to apply the modular principles of network assembly to whole new classes of molecules and to assemble complex and dynamic systems from arbitrary self-replicating molecules constrained only by the limitations of physical law.
mRNA pairs