Research Article

Engineering Gene Networks to Emulate Drosophila Embryonic Pattern Formation

  • Mark Isalan mail,

    To whom correspondence should be addressed. E-mail:

    Affiliation: European Molecular Biology Laboratory, Structures and Biocomputing, Heidelberg, Germany

  • Caroline Lemerle,

    Affiliation: European Molecular Biology Laboratory, Structures and Biocomputing, Heidelberg, Germany

  • Luis Serrano

    Affiliation: European Molecular Biology Laboratory, Structures and Biocomputing, Heidelberg, Germany

  • Published: February 22, 2005
  • DOI: 10.1371/journal.pbio.0030064


Pattern formation is essential in the development of higher eukaryotes. For example, in the Drosophila embryo, maternal morphogen gradients establish gap gene expression domain patterning along the anterior-posterior axis, through linkage with an elaborate gene network. To understand the evolution and behaviour of such systems better, it is important to establish the minimal determinants required for patterning. We have therefore engineered artificial transcription-translation networks that generate simple patterns, crudely analogous to the Drosophila gap gene system. The Drosophila syncytium was modelled using DNA-coated paramagnetic beads fixed by magnets in an artificial chamber, forming a gene expression network. Transient expression domain patterns were generated using various levels of network connectivity. Generally, adding more transcription repression interactions increased the “sharpness” of the pattern while reducing overall expression levels. An accompanying computer model for our system allowed us to search for parameter sets compatible with patterning. While it is clear that the Drosophila embryo is far more complex than our simplified model, several features of interest emerge. For example, the model suggests that simple diffusion may be too rapid for Drosophila-scale patterning, implying that sublocalisation, or “trapping,” is required. Second, we find that for pattern formation to occur under the conditions of our in vitro reaction-diffusion system, the activator molecules must propagate faster than the inhibitors. Third, adding controlled protease degradation to the system stabilizes pattern formation over time. We have reconstituted transcriptional pattern formation from purified substances, including phage RNA polymerases, ribonucleotides, and an eukaryotic translation extract. We anticipate that the system described here will be generally applicable to the study of any biological network with a spatial component.


Engineering a system to emulate a particular behaviour can be an extremely informative approach to systems biology [1,2,3,4,5]. Even if the natural biochemical interactions are well characterized, it remains a considerable challenge to reconstruct a physical system with the appropriate behaviour. Step-by-step reconstruction allows theoretical assumptions and models to be refined. Not only is complexity reduced by removing the context of the whole organism, but by reconstitution of pattern formation from purified substances (in this case, RNA polymerases, ribonucleotides, and a translation extract) the sufficiency of a proposed mechanism of pattern formation can be demonstrated.

In this work, our primary aim was the development of an in vitro system that allows the careful buildup of complex networks under controlled conditions. To demonstrate the usefulness of such a system, we decided to reconstruct a developmental pattern-formation program based on the formation of a gradient of a transcription activator—a “morphogen”—and to link it to a network of transcription repressors. In a sense, we set out to design a patterning system similar to chemical reaction-diffusion systems (see below). However, the use of components such as transcription activators and repressors in an in vitro transcription-translation system made the system significantly closer to a biological system, albeit highly simplified when compared to the complexity of eukaryotic transcription [6].

Patterning systems can be thought of as belonging to one of two principal types: First, there are systems with homogeneous initial conditions that self-organise after early random symmetry-breaking events ([7,8]; for a biological example, Fucus, see [9]). Second, there are systems with initial localisation of the components, which can form concentration gradients of activities from their respective sources [10,11].

The first class of patterning system, involving reaction-diffusion from initially homogeneous conditions, was first proposed by Turing in 1952 [7] and was further developed by Meinhardt and Gierer in the 1970s, in their model of patterning with short-range autoactivator and long-range lateral inhibitor [8]. Although there are many likely biological candidates for such activator-inhibitor systems (reviewed in [12]), none has been reengineered from first principles. In contrast, significant progress has been made in reconstituting purely chemical reactions that self-organise, such as the Belousov-Zhabotinski reactions [13,14,15] and Turing-type, Meinhardt-Gierer (M-G), and oscillatory reactions [16,17,18,19,20]. In a more biological context, similar spatiotemporal patterns, consisting of propagating concentration waves, have been modelled for a glycolytic enzyme oscillator in yeast [21].

In the second class of patterning system, where the initial conditions are nonhomogeneous, patterning begins with an asymmetry: a localised source of morphogen (shape-defining molecules that form a concentration gradient and function in a concentration-dependent manner so as to determine positional information in a patterning field [22]; reviewed in [23,24,25]).

The question of how morphogen gradients are formed and maintained is still a matter of keen debate and study [26,27,28], with many proposed mechanisms (reviewed in [29,30]). In the simplest case, a stable gradient could be formed by passive diffusion [31] and uniform degradation, although it has been suggested that enhanced morphogen degradation near the source leads to increased robustness against morphogen fluctuations during patterning [32]. Also, “ligand trapping” by the receptor for a morphogen can have significant effects on the shape of a morphogen gradient [33], as in the case of Torso diffusing in the extracellular space surrounding the Drosophila oocyte [34].

In most cases in metazoa, morphogens define patterns over fields of many cells (reviewed in [30,35]), but there is one special case in embryonic development that has been particularly well studied, in which a morphogen gradient operates in a single-celled, multinuclear syncytium: the early Drosophila embryo [10,11]. In this system, early patterning is mediated by maternal morphogen factors, which are thought to diffuse and form gradients to guide patterning within a large multinuclear cell (reviewed in [36,37,38]). After egg deposition, an embryo forms a segmentation pattern within 3 h, under the influence of a hierarchical sequence of gene expression interactions involving gap genes, pair-rule genes, and segment polarity genes [39]. Maternal elements—in particular, the morphogen Bicoid—guide this process, setting distinct initial conditions.

Work by Driever and Nüsslein-Volhard [10,11] demonstrated that Bicoid protein possessed three characteristics of a classic morphogen: (i) a localised source of cytoplasmic activity (through bicoid RNA transport to the anterior pole of the egg, involving microtubules and maternal genes [40,41,42,43,44]); (ii) formation of a concentration gradient from the source; and (iii) concentration-dependent activity that determines positional information within the gradient (reviewed in [38]). Bicoid has at least two functions that contribute to its function as a morphogen: transcription activation and translation inhibition. Acting as a transcription factor, Bicoid can activate a number of downstream gap genes, including hunchback, knirps, giant, and Krüppel, whose products cross-react in a complex and mainly repressive interaction network to modulate each other's expression (reviewed in [37]; modelled in [45,46]). However, Bicoid does not simply function independently as a morphogen at the top of the gap gene hierarchy. Although Bicoid is responsible for anterior expression of zygotic Hunchback [47], it actually requires maternally expressed Hunchback as a cofactor to function anteriorly [48]. Meanwhile, in the posterior, the maternal hunchback mRNA is initially translationally repressed by the posterior determinant Nanos [49]. Moreover, the terminal gap genes tailless and huckebein are activated independently but serve to repress zygotic gap gene expression at the poles of the embryo, thus influencing patterning [50,51,52]. In its other role, as a translation inhibitor of caudal mRNA, Bicoid initially inhibits the uniformly expressed mRNA to form a concentration gradient of the protein Caudal [53,54]. Thus, the Caudal gradient [55] is essentially the inverse of the Bicoid gradient, and Caudal functions as a transcription activator in the posterior of the egg, further influencing expression of the gap gene network. Some of the important interactions in the gap gene network are shown in Figure 1A, although it should be noted that this overview is an oversimplification and does not consider differences in maternal and zygotic factor expression over time, nor does it consider all of the factors involved.


Figure 1. Gene Circuits and Chambers

(A) Principal interactions in the Drosophila gap gene network, modelled after [37]. Relative levels and distributions of Hunchback (Hb), Giant (Gt), Krüppel (Kr), Knirps (Kni), Bicoid (Bcd), and Caudal (Cad) shown from anterior (left) to posterior (right). Green arrows indicate activation, red T-bars repression.

(B) Artificial gene network design, with transcription activators T7 and SP6 polymerases, and zinc finger repressors A, B, and C. Genes are immobilised on paramagnetic beads, and T7 forms a directional concentration gradient.

(C) Principal interactions in a simple designed network.

(D) Transcription-translation chamber. Genes for repressor A are localised at the “poles,” whereas B and C are ubiquitous. Gel slabs 4–6 have been excised, exposing the magnets below, illustrating gel dissection for Western blot analysis.

(E) Normalised Western data for four replicate chambers, showing mean levels of A, B, and C after 20 min (± One standard deviation).

(F) Sample Western blot from the four-replicate experiment.


From the outset, we chose to model our system around elements of the Drosophila gap gene network (Figure 1A), because we could study many elements of morphogenesis, such as gradient formation and the sufficiency of cross-repression for setting pattern boundaries, without the need for considering multiple cells, membrane-bound receptors, and cell-to-cell interactions. Using an in vitro model, we wanted to address the following questions about patterning. First, how easy is it to generate an expression pattern in a gradient, using a diffusing activator from a localised source? Second, one of the outstanding issues in the field is to what extent correct positioning of the gap protein domain boundaries is specified by maternal morphogen gradients and by cross-repression between gap genes; a recent model suggests that repression is crucial for patterning and that threshold-dependent interpretation of the maternal morphogen concentration is not sufficient [56]. We therefore wanted to test the effect of transcription repression on pattern formation directly, by progressively adding more repression interactions in a designed gene network. Third, uniform degradation of a diffusing morphogen is often assumed to account for steady-state gradient formation, so we set out to test the effects of adding controlled degradation to an in vitro patterning system. Finally, we wanted to use our model to see how the scale and pattern of the system are affected by the relative rates of diffusion of individual components, and whether nonuniform diffusion of activators and inhibitors are required to form a pattern.


Design of the Network and Development of the In Vitro Experimental Platform

We began by designing a simplified gene network to emulate elements of the Drosophila gap gene system (Figure 1). The aim was to develop a fully synthetic approach in which protein analogues completely unrelated to Drosophila would emulate some of the transcription activation and repression interactions thought to be important for patterning in the gap gene system. As activators, two sequence-specific polymerases were employed, T7 and SP6, that have been used successfully by others to engineer gene networks [57]. These two polymerases bind to their respective consensus DNA recognition sites to initiate transcription, and thus represent an extremely simplified mode of transcription when compared to the multifactor complexes required for eukaryotic transcription (reviewed in [6]).

T7 polymerase was chosen to be the “master activator” of the system and, by crude analogy, was expected to carry out some of the functions of Bicoid, namely transcription activation of downstream members in the gap gene hierarchy, in a concentration-dependent manner, from a localised source [10,11]. In Drosophila, the Bicoid morphogen gradient initially controls the shape of the Caudal protein gradient through translational repression of maternal mRNA [53,54], although later Caudal expression is under zygotic transcriptional control. To simplify this level of complexity, we decided to model Caudal transcription activation by a second gradient of T7 polymerase, from the opposite pole to our primary “Bicoid” gradient (compare Figure 1A and 1C). Residual activation between other members of the gap gene members (e.g., Figure 1A, Hunchback activating Krüppel or Krüppel activating knirps) was modelled nonexplicitly by having a homogeneous distribution of a second sequence-specific transcription activator, SP6 polymerase (Figure 1A and 1C).

Repression interactions between gap gene members were modelled by constructing three site-specific repressors to represent the repressor activities of Hunchback, Giant, and Krüppel (Figure 1B, repressors A, B, and C, respectively). The repressors were derived from artificial zinc finger DNA-binding domains that were engineered by phage display [58]. Variable gene-repression networks were therefore constructed by placing binding sites for the repressors in the appropriate gene expression constructs (see for comparison Figure 2). Repressor sites were either overlapping with the polymerase initiation sites (demonstrated to be effective using triplex-forming oligonucleotides [59]), or immediately downstream of the initiation sites (Figure 2). Therefore, by changing the identity of the repressor sites, the connectivity of the network could readily be modified to add or remove cross-repressive interactions.


Figure 2. Map of the Constructs Used in This Study

The repressor binding sites overlap with T7 or SP6 promoters and vary between constructs. In this way, it is possible to alter the connectivity of the repressive interactions by the products of genes A, B, and C. Repressive interactions are denoted by T-bars. The start codon of each gene is in Kozak context and is denoted by “GCC ATG G.”


Key to the strategy was the development of an experimental platform in which to model the volume of the Drosophila embryo and to carry out artificial gene network reactions. Plastic chambers were therefore developed, constructed over printed templates on petri dishes (Figure 1D; see also Materials and Methods and Protocol S1). The chambers were filled with a customised transcription-translation mixture, allowing gene network reactions to be carried out in situ. Additionally, small bar magnets were fixed under the chamber to create a spatially defined array, over which paramagnetic beads could be dispensed. By coating such streptavidin-linked beads with biotinylated PCR products, specific gene network constructs were tethered and sublocalised on the array.

Furthermore, ultra-low melting point agarose was added to the transcription-translation mixture, both to increase viscosity and to allow the reaction to be “fixed” in a gelling step at 4 °C. Through fixing, gel slices could be excised and assayed by Western blotting against FLAG-epitope tags on the expressed proteins. This design therefore enabled quantification of each output species present in the network (genes A, B, and C) for any given chamber position and time point.

The chambers were constructed such that the system components could be pipetted wherever desired, either homogeneously mixed with the transcription-translation mix or pipetted at defined loci, such as at the edges or “poles” of the chamber. As described above, these system components included the two soluble, purified transcription activators (T7 and SP6 polymerases) and three bead-tethered zinc finger transcription-repressor constructs, A, B, and C, which were themselves activated by the polymerases and could cross-repress each other (see Figure 1).

Positional information was therefore introduced into the artificial system in two ways. First, by injecting purified T7 polymerase at either pole of the chamber, the Bicoid activator distribution could be transiently modelled. Second, beads coated with different gene network constructs (genes A, B, and C), could be fixed at different positions on the magnetic array (see Figure 1D).

For example, repressor A genes were placed solely at the chamber edges (“poles”) to model, loosely, the distribution of embryonic Hunchback activity. This part of the model is a significant oversimplification: Although Hunchback is eventually expressed in two domains, one anterior and one posterior [48], it is only expressed anteriorly in the early embryo. Furthermore, while maternal hunchback mRNA is evenly distributed, the anterior domain of Hunchback protein forms through zygotic translation and transcription activation (under the control of Bicoid), while maternal RNA is translationally repressed posteriorly, under the influence of Nanos [49,60,61]. In the later phase of hunchback regulation, the posterior Hunchback domain forms through a combination of factors, including activation by Tailless [62,63] and hunchback autoactivation [64].

To complete the model, the genes for repressors B and C were distributed uniformly throughout the chamber on magnetic beads, to represent the ubiquitous distribution of genes in nuclei, throughout the embryo. Therefore, the spatial expression of genes B and C, who represent the downstream gap gene members giant and Krüppel (see Figure 1A and 1C), was dependent on differential activation by the T7 polymerase gradient and crossregulation between gene network members. However, it should be noted that the initial expression of these genes in Drosophila may not be achieved by crossregulation, because localised mRNA is seen before any protein is detectable (Krüppel and Giant are only detected unambiguously in early cycle 13 [46]).

Pattern Generation In Vitro from a Transcription Network

In our first experiments, we constructed a simple, minimal network with sequential transcription activation and repression (see Figure 1B and 1C). Although this basic system is far less complex than the Drosophila gap gene system, it was indeed sufficient to generate a crude target behaviour (see Figure 1E and 1F). Qualitatively, the pattern can be explained as follows, Gene A is activated by T7 polymerase from its source at either end of the chamber, and so is expressed most highly at these poles. Gene B is similarly activated, and so it is also less expressed in the middle of the cell. However, since gene B is repressed by protein A, its levels are also reduced at either pole. Finally, Gene C is activated by a ubiquitous SP6 polymerase, but is repressed by proteins A and B, and is consequently centrally distributed.

Progressing from the minimal network, we explored systems with a variety of connectivities (Figure 3), including a control network without repression interactions (Figure 3A), and one with extensive mutual or feedback interactions (Figure 3C). These were compared with the original network (Figure 3B) in a series of time-course experiments. Generally, we observed that the more repression interactions in a system, the lower the overall protein production but the “sharper” the pattern.


Figure 3. Alternative Gene Networks

At five set time points (15, 25, 35, 60, and 90 min), transcription-translation chambers were dissected into nine slabs for Western blot analysis.

(A) Control network with no repression sites between genes A, B, and C.

(B) Minimally repressed network (compare Figure 1).

(C) Mutual repression network with extensive negative interactions between species. Adding protease (“+ Degradation”) creates weak but time-stable patterns for both the “Repressed” and “Mutual” networks (35 versus 90 min). Quantitated graphs for the above data are available in Protocol S1.


All patterns degenerated to a significant degree by 60 min, indicating the transience of the system (Figure 3, 60 min). However, by adding Factor Xa protease, we were able approximately to match levels of production and degradation. Thus, the outputs became sharper, weaker, and more dynamically stable, hardly varying between 35 and 90 min (Figure 3B and 3C, “+ Degradation”).

Computer Exploration of Parameter Space

To study parameter sensitivity in our system more comprehensively, we constructed a computer model of the chamber and networks (Protocol S1). A series of coupled differential equations were simulated, yielding expression levels of gene products A, B, and C, for the three different levels of network connectivity coded by our gene network designs (Figure 4). The modelling was carried out at two scales, that of our experimental system (18 mm long) and that of a Drosophila embryo (500 μm long). As in the experimental system (Figures 3 and 4A), the simulations revealed a large difference of pattern between the unrepressed and repressed systems. The patterns are more similar, however, between the simple and mutually repressed networks (Figure 4B) but, as in our in vitro experiments (e.g., see Figure 3B and 3C, 15 min), adding feedback repression makes the peaks better resolved.


Figure 4. Comparison of Experimental Data and Computer Simulations

Data are shown for the three gene networks described in Figure 3, showing outputs for proteins A (cyan), B (magenta) and C (dark blue).

(A) Quantitated Western blot data from Figure 3, after 25 min.

(B) Simulation data plotted as percentage of total output protein against chamber length, at the chamber (18-mm) or Drosophila (0.5-mm) scale. The model is described in full in Protocol S1.


Next, we explored the sensitivity of the simple repression network to diffusion parameters (Figure 5). Generally, we found that the A- and C-peaks were least sensitive to parameter variation, as there are no antagonistic forces against their formation (Figure 5A and 5C). By contrast, gene B is more sensitive: Twin-peak formation correlates with the relative diffusion ratios of activator (T7) and other mRNA/protein components (Figure 5B). To generate “target behaviour,” the activator must diffuse more rapidly than other species, within certain limits (approximately 5- to 50-fold faster for Figure 5B). However, the absolute values (and ratios) for diffusion merely alter the timing of the transient B-peak formation in a system of a given scale. For simplicity, only the 0.5-mm system is illustrated in Figure 5; similar conclusions were drawn from the 18-mm scale model.


Figure 5. Varying Diffusion and Degradation Parameters

Computer model of gene network, scaled to Drosophila length (0.5 mm). Diffusion parameters are varied for mRNA (Dm), protein (Dp), and T7 activator (DX). Data are plotted as percentage of total output protein (y-axes) against chamber position (x-axes), for 10-min simulations.

(A) Outputs for protein A.

(B) Output for protein B. Graphs with “target behaviour” are shaded grey, and the four asterisks mark the parameter sets used to generate outputs for proteins A and C.

(C) Outputs for protein C.

(D) Effect of adding protease degradation to B-output, shown at 15-min intervals, over a 2.5-h time course (parameters: DX = 0.43 μm2s−1; Dm = Dp = 0.02 μm2s−1; t1/2 = 770 s).


As in our chamber experiments, the computer model output became more “time-stable” by adding a degradation element (Figure 5D). Drosophila may exploit such mechanisms to some extent, since Bicoid protein degrades in vivo (t1/2 ≤ 1800 s [11]), although bicoid mRNA is unusually stable [65].


To develop a fully synthetic approach that will emulate elements of gap gene expression domain pattern formation, we created an in vitro transcription-translation system that allows flexible spatial gene network construction. The system is widely applicable, allowing control over factors such as localisation or diffusion, and the ability to add or remove components at will.

Repressive Interactions and Pattern Formation

A basic aim of our system was to see whether we could engineer a gradient of protein expression, using a diffusing activator from a localised source. We found this task straightforward in the transcription-translation chambers, using injected T7 polymerase, and this led us to try more complex expression-repression interactions. We constructed three types of gene network—unrepressed, simple repressed, and mutually repressed (see Figure 3)—representing different levels of network connectivity. Generally, in both our in vitro and computer models, we found that adding more connections resulted in better-resolved patterning, although the absolute levels of gene expression were reduced. Our in vitro results are essentially qualitative at this stage, but appear to agree with the observations of others—that crossrepression is crucial for the control of patterning boundaries [56]. It will be interesting to learn whether more sophisticated elements can be engineered into the system to begin to emulate the more complex features of gap gene expression domain patterning. For example, dynamic anterior shifts are seen in domain expression over time because of asymmetric gap-gap crossrepression [56]. Asymmetric repression and other circuits could, in the future, be engineered into our chambers by altering the repressor binding sites in the appropriate constructs. Such a system would require component turnover to achieve steady-state patterning. We have begun to tackle this project through our experiments with controlled protease degradation, but a further requirement would be to have autocatalytic production of T7 polymerase from a localised source, rather than the injected pulse of purified polymerase in our current model.

Interestingly, our experimental data showed a reproducible degree of patterning even in the unrepressed system (Figure 3A, 15 min, C output). Because gene C is activated by a separate polymerase (SP6), this patterning cannot result from competition for activator. Therefore, competition for other resources (such as ribosomes, nucleotides, and tRNAs) may allow A and B to “inhibit” C. Indeed, supplying extra components (particularly wheat germ extract and SP6 polymerase together) increases protein production under these conditions, including that of C (unpublished data). If competition can generate patterns, albeit less well defined ones than repression-connected networks, this could perhaps represent an evolutionary “network precursor” state: Weak patterns could be generated by localisation and competition between factors, and these could later be consolidated by evolution of a “true” negative network connection. However, this hypothesis may not be relevant to the situation inside an insect egg, as this has probably evolved to deliver nutrients very efficiently to the embryo, even at a very primitive evolutionary stage. It therefore remains to be seen whether competition effects would be as significant in vivo.

Since our models use minimal components to achieve spatial pattern formation, they demonstrate the ease with which very simple networks might evolve. Patterning may be achieved with only localisation, diffusion, and some kind of functional network connection, such as transcription activation, competition, or repression. The addition of extra layers of network properties, such as controlled degradation, could then fix and stabilize such patterns. In fact, since sublocalisation—followed by stepwise addition of network components—is sufficient to generate crude patterns, it might provide a plausible mechanism for early spatial network evolution inside a single cell. However, it should be noted that gap gene expression domain patterning probably evolved by a different mechanism, from an earlier multicellular state, where segments were added sequentially through polar growth. In fact, bicoid is absent in most other insects, and it has been proposed that Drosophila evolved bicoid by duplication of the homeodomain-encoding gene zerknüllt, found in lower Diptera [66].

Diffusion Rates and Patterning

We were intrigued by our observations, both in vitro and in silico, that patterning required the activator molecule to diffuse or propagate more rapidly than the inhibitors. This is interesting because it is the opposite of the M-G system (described in the Introduction and reviewed in [12]). Long-range activation is not unknown in chemical patterning systems [67], although many biological models appear to require the M-G criterion for long-range inhibition (e.g., [68]). The other obvious differences between our system and the M-G model are the initial localisation of components and the lack of autocatalysis of the activator. It will be interesting to determine whether such M-G patterning systems can be recreated in our chambers, once further factors are considered, such as the avoidance of “autocatalytic explosions” or “global inhibitions.”

The computer model that we developed allowed us to test a broad range of parameters, such as diffusion and degradation rates (Figure 5), revealing differences between the requirements for patterning among the different species in the gene network. First, the more-connected member of the network (gene B) was much more sensitive to parameter variation than the less-connected members (genes A and C). This is perhaps to be expected, since protein B has two separate boundaries of expression (defined as a function of T7 distribution and both A and C expression), whereas proteins A and C have only single “edges” to be defined.

Another important feature of the system emerged when scaling the parameters for the model patterns to Drosophila scale (Figure 5, 0.5 mm “embryo,” 2.5 h). We found that, assuming simple diffusion, B-peak formation was compatible only with unphysiologically slow diffusion values (diffusion constants for mRNA [Dm] and protein [Dp] = 0.02 μm2s−1; for T7 activator [DX] = 0.43 μm2s−1). Since cellular proteins are expected to diffuse more rapidly (approximately 1–100 μm2s−1), this could be an artefact, reflecting the simplicity of our model. Nonetheless, simple diffusion still appears too rapid to account for Drosophila-scale patterning. It should be noted that a potential barrier to free diffusion is the active nuclear import of Bicoid and Hunchback [69]. In a separate example, diffusion of pair-rule transcripts is overridden by microtubule transport [70]. Controlled sublocalisation may therefore be crucial to limit apparent diffusion in vivo, allowing more precise patterning.


The understanding of how precision of patterning is achieved in Drosophila is still far from complete. In a recent study, it was shown that the Bicoid profile is far more variable between embryos than that of Hunchback, but the mechanism by which this noise is filtered remains unknown [71]. As more and more detailed experimental data are collected [72], and new mechanisms are proposed to account for patterning, it will be important to test the sufficiency of these mechanisms through experimental reconstitution. For such purposes, the chambers described here may be easily adapted to test different hypotheses. In vitro systems are a useful first step towards testing the sufficiency of a network—which might then be reengineered in the original target organism.

Combining simple reconstruction with theoretical modelling is a useful tool to discover and test general design principles in gene networks [73,74,75,76]. Until now, however, the spatial component essential in many biological processes has been ignored in these approaches. We anticipate that other networks, such as signalling cascades or metabolic networks, might also be studied using our system and that the spatial element, introduced through the beads, might provide new insights into complex systems.

Materials and Methods

Magnetic chamber construction.

A detailed, step-by-step description of the construction of the chamber can be found in Protocol S1. Briefly, nine stirring-bar magnets (1.5 mm × 8 mm; VWR International, Vienna, Austria; #4429025) were inserted vertically into a plasticine-filled standard petri dish, creating a magnetic array (see Figure 1D). Construction was guided with a grid template, laser-printed on a transparent acetate sheet, and fixed over the magnets and plasticine. The template was a 3 mm × 18 rectangle with nine subdivisions (“slabs”). A sterile cell culture dish (Nalge Nunc, Rochester, New York, United States; #150350) was fixed immediately above the magnetic array. Chamber borders (1 mm deep) were constructed on the base of this second dish, following the template, using strips cut from adhesive Hybriwell chambers (Sigma-Aldrich, St. Louis, Missouri, United States; #H1159–100EA).

Gene network constructs.

Maps of the constructs are illustrated in Figure 2. Repressors A, B, and C were derived from previously engineered zinc fingers [58]. Repressor A contained six zinc fingers, recognising the sequence 5′-AGGGAGGCGGACTGGGGA-3′, fused to the residues 11–55 of the Kox-1 repression domain [77] and a six-repeat FLAG epitope tag [78]. Repressor B contained six zinc fingers, recognising the sequence 5′-AGGGAGGCGGGAGCTTTC-3′ and fused to a three-repeat FLAG-tag. Repressor C contained three zinc fingers, recognising the sequence 5′-GGAGCTTTC-3′, fused to the Kox domain and a three-repeat FLAG-tag. The following polymerase consensus promoter regions were used: T7, 5′-TAATACGACTCACTATAG GGAG-3′; SP6, 5′-ATTTAGGTGACACTATAG AAGGG-3′. The gene network promoters were linked with neutral or repressor sites to the polymerase promoters.

In the following nucleotide sequences, zinc finger binding sites are indicated in lowercase, initiation nucleotides in bold, and promoter overlaps underlined. Unrepressed T7, 5′-TAATACGACTCACTATAG GGAGAAACACCATAG-3′ (see Figure 3A, constructs A and B, and Figure 3B, construct A). Unrepressed SP6, 5′-ATTTAGGTGACACTATAG AAGGGAAACACCATAG-3′ (see Figure 3A, construct C). T7 repressed by A (and weakly by B), 5′-TAATACGACTCACTATagggaggcggactgggga -3′ (see Figure 3B, construct B). SP6 repressed by A (and weakly by B), 5′-ATTTAGGTGACACTATAGAagggaggcggactgggga-3′ (see Figure 3B and 3C, construct C). T7 repressed by A and C (and weakly by B), 5′-TAATACGACTCACTATagggaggcggactggggaTggagctttc-3′ (see Figure 3C, construct B). T7 repressed by C (and weakly by B), 5′-TAATACGACTCACTATAGggagctttc-3′ (see Figure 3C, construct A). Constructs were cloned in pCaSpeR4, sequenced, and used to generate PCR DNA for in vitro transcription-translation.

Gene network reactions.

Paramagnetic beads were coated with PCR DNA (with one primer biotinylated) using a Dynabeads Kilobase Binder Kit (Dynal, Oslo, Norway; #601.01). Typically, gene A was used at 800 fmol per 10 μl of beads, resuspended in 8 μl of water; 200 fmol of gene B and 140 fmol of gene C were combined with 20 μl of beads, and resuspended in 20 μl of water.

Transcription-translation mixture was prepared that included 2.5 μl of water; 28 μl of ultra-low melting point agarose (Sigma; #A2576) solution (prepared as 1.5% [w/v] in boiling water and cooled to 30 °C); and TNT Coupled Wheat Germ Extract System (Promega, Madison, Wisconsin, United States; #L4130 and #L4140), which comprised 20 μl of TNT wheat germ extract, 1.2 μl of TNT reaction buffer, 0.6 μl of amino acid mixture (1 mM), 1.2 μl of RNasin (not included in TNT kit), and 0.5 μl of SP6 polymerase. 54 μl of this mixture was dispensed per chamber.

For degradation experiments, 2.25 units of Factor Xa (Amersham Biosciences, Little Chalfont, United Kingdom) were added per chamber. Coated Dynabeads were injected at appropriate positions over the magnetic array: typically, 100 fmol of gene A (1 μl), 5 fmol of gene B, and 3.5 fmol of gene C (0.5 μl). T7 polymerase (0.5 μl; from Promega TNT kit) was immediately injected at the chamber edges. After timed incubations at 25 °C, chambers were transferred to 4 °C for 35 min, to form a gel. Gel slices were cut with a razor blade (guided by the printed template) and aspirated with a P10 Gilson pipette. Samples were mixed with 10 μl of SDS-loading buffer and analysed by SDS-PAGE, Western blotting, and ECL, with anti-M2 FLAG antibody (Sigma; #F3165). Further details on this step can be found in Protocol S1.

Computer modelling.

A Perl script was written to simulate the diffusion-coupled expression of genes A, B, and C, by T7 and SP6 phage polymerases, in a translation extract. The program parameters and script are fully described in Protocols S1–S3. 18 mm-scale chamber model: Parameters included separate diffusion (and degradation) rates for RNA and protein; a separate apparent diffusion for injected T7, modelled from experimental observations (rapid initial diffusion with exponential decay; Section 5 of Protocol S1); estimated binding constants for all interacting species (zinc finger dissociation constants were estimated from previous work on related three- and six-finger constructs [58,79,80]); and estimated transcription-translation rates. For adapting the model to the 0.5-mm Drosophila scale, chamber size was scaled down, and only simple diffusion was allowed for all components; for simplicity, transcription-translation rates were not varied (Section 3 of Protocol S1).

Supporting Information

Protocol S1. Detailed Description of Model


(1.2 MB PDF).

Protocol S2. Parameter File for Simulations

This file contains the default parameters for the computer model in a format that can be read by the Perl script.


(5 KB DOC).

Protocol S3. Computer Program Script for Simulations

This text file is a Perl script to run the computer simulations described in the manuscript.


(22 KB DOC).

Accession Numbers

The Locuslink (, or GeneID (​b=gene, accession numbers of the genes and proteins discussed in this paper are Bicoid (40830), caudal (35341), giant (31227), huckebein (40549), hunchback (41032), knirps (40287), Krüppel (38012), Nanos (42297), tailless (43656), Torso (35717), and zerknüllt (40828).


We would like to thank K. Michalodimitrakis, C. Gonzalez, and B. Schönwetter for helpful discussions. MI is supported by an International Research Fellowship from the Wellcome Trust, United Kingdom.

Author Contributions

MI and LS conceived and designed the experiments. MI performed the experiments. CL analyzed the data and contributed reagents/materials/analysis tools. MI, CL, and LS wrote the paper.


  1. 1. Elowitz MB, Leibler S (2000) A synthetic oscillatory network of transcriptional regulators. Nature 403: 335–338.
  2. 2. Gardner TS, Cantor CR, Collins JJ (2000) Construction of a genetic toggle switch in Escherichia coli. Nature 403: 339–342.
  3. 3. Becskei A, Serrano L (2000) Engineering stability in gene networks by autoregulation. Nature 405: 590–593.
  4. 4. Guet CC, Elowitz MB, Hsing W, Leibler S (2002) Combinatorial synthesis of genetic networks. Science 296: 1466–1470.
  5. 5. Hasty J, McMillen D, Collins JJ (2002) Engineered gene circuits. Nature 420: 224–230.
  6. 6. Lee TI, Young RA (2000) Transcription of eukaryotic protein-coding genes. Annu Rev Genet 34: 77–137.
  7. 7. Turing AM (1952) The chemical basis of morphogenesis. Philos Trans R Soc Lond B Biol Sci 237: 37–72.
  8. 8. Gierer A, Meinhardt H (1972) A theory of biological pattern formation. Kybernetik 12: 30–39.
  9. 9. Goodner B, Quatrano RS (1993) Fucus embryogenesis: A model to study the establishment of polarity. Plant Cell 5: 1471–1481.
  10. 10. Driever W, Nusslein-Volhard C (1988) The bicoid protein determines position in the Drosophila embryo in a concentration-dependent manner. Cell 54: 95–104.
  11. 11. Driever W, Nusslein-Volhard C (1988) A gradient of bicoid protein in Drosophila embryos. Cell 54: 83–93.
  12. 12. Meinhardt H, Gierer A (2000) Pattern formation by local self-activation and lateral inhibition. Bioessays 22: 753–760.
  13. 13. Belousov BP (1959) A periodic chemical reaction and its mechanism. Sb Ref Radiats Med Medgiz Moscow 1959: 145–147.
  14. 14. Zaikin A, Zhabotinski A (1970) Concentration wave propagation in two-dimensional liquid-phase self-oscillating system. Nature 225: 535–537.
  15. 15. Winfree A (1972) Spiral waves of chemical activity. Science 175: 634–636.
  16. 16. Castets VV, Dulos E, Boissonade J, De Kepper P (1990) Experimental evidence of a sustained standing Turing-type nonequilibrium chemical pattern. Phys Rev Lett 64: 2953–2956.
  17. 17. Dulos E, Boissonade J, Perraud JJ, Rudovics B, De Kepper P (1996) Chemical morphogenesis: Turing patterns in an experimental chemical system. Acta Biotheor 44: 249–261.
  18. 18. Ouyang Q, Swinney HL (1991) Transition to chemical turbulence. Chaos 1: 411–420.
  19. 19. Lin AL, Hagberg A, Ardelea A, Bertram M, Swinney HL, et al. (2000) Four-phase patterns in forced oscillatory systems. Phys Rev E Stat Phys Plasmas Fluids Relat Interdiscip Topics 62: 3790–3798.
  20. 20. Lin AL, Bertram M, Martinez K, Swinney HL, Ardelea A, et al. (2000) Resonant phase patterns in a reaction-diffusion system. Phys Rev Lett 84: 4240–4243.
  21. 21. Goldbeter A (1973) Patterns of spatiotemporal organization in an allosteric enzyme model. Proc Natl Acad Sci U S A 70: 3255–3259.
  22. 22. Wolpert L (1969) Positional information and the spatial pattern of cellular differentiation. J Theor Biol 25: 1–47.
  23. 23. Lawrence PA (1992) The making of a fly: The genetics of animal design. Oxford: Blackwell Scientific Publications. 228 p.
  24. 24. Lawrence PA, Struhl G (1996) Morphogens, compartments, and pattern: Lessons from Drosophila? Cell 85: 951–961.
  25. 25. Lawrence PA (2001) Morphogens: How big is the big picture? Nat Cell Biol 3: E151–E154.
  26. 26. Kerszberg M, Wolpert L (1998) Mechanisms for positional signalling by morphogen transport: A theoretical study. J Theor Biol 191: 103–114.
  27. 27. Entchev EV, Schwabedissen A, Gonzalez-Gaitan M (2000) Gradient formation of the TGF-beta homolog Dpp. Cell 103: 981–991.
  28. 28. Lander AD, Nie Q, Wan FY (2002) Do morphogen gradients arise by diffusion? Dev Cell 2: 785–796.
  29. 29. Vincent JP, Dubois L (2002) Morphogen transport along epithelia, an integrated trafficking problem. Dev Cell 3: 615–623.
  30. 30. Teleman AA, Strigini M, Cohen SM (2001) Shaping morphogen gradients. Cell 105: 559–562.
  31. 31. Crick F (1970) Diffusion in embryogenesis. Nature 225: 420–422.
  32. 32. Eldar A, Rosin D, Shilo BZ, Barkai N (2003) Self-enhanced ligand degradation underlies robustness of morphogen gradients. Dev Cell 5: 635–646.
  33. 33. Berezhkovskii AM, Batsilas L, Shvartsman SY (2004) Ligand trapping in epithelial layers and cell cultures. Biophys Chem 107: 221–227.
  34. 34. Casanova J, Struhl G (1993) The torso receptor localizes as well as transduces the spatial signal specifying terminal body pattern in Drosophila. Nature 362: 152–155.
  35. 35. Gurdon JB, Bourillot PY (2001) Morphogen gradient interpretation. Nature 413: 797–803.
  36. 36. St Johnston D, Nusslein-Volhard C (1992) The origin of pattern and polarity in the Drosophila embryo. Cell 68: 201–219.
  37. 37. Rivera-Pomar R, Jackle H (1996) From gradients to stripes in Drosophila embryogenesis: Filling in the gaps. Trends Genet 12: 478–483.
  38. 38. Ephrussi A, St Johnston D (2004) Seeing is believing: The bicoid morphogen gradient matures. Cell 116: 143–152.
  39. 39. Nusslein-Volhard C, Wieschaus E (1980) Mutations affecting segment number and polarity in Drosophila. Nature 287: 795–801.
  40. 40. Pokrywka NJ, Stephenson EC (1991) Microtubules mediate the localization of bicoid RNA during Drosophila oogenesis. Development 113: 55–66.
  41. 41. Lane ME, Kalderon D (1994) RNA localization along the anteroposterior axis of the Drosophila oocyte requires PKA-mediated signal transduction to direct normal microtubule organization. Genes Dev 8: 2986–2995.
  42. 42. Roth S, Neuman-Silberberg FS, Barcelo G, Schupbach T (1995) Cornichon and the EGF receptor signaling process are necessary for both anterior-posterior and dorsal-ventral pattern formation in Drosophila. Cell 81: 967–978.
  43. 43. Gonzalez-Reyes A, Elliott H, St Johnston D (1995) Polarization of both major body axes in Drosophila by gurken-torpedo signalling. Nature 375: 654–658.
  44. 44. Berleth T, Burri M, Thoma G, Bopp D, Richstein S, et al. (1988) The role of localization of bicoid RNA in organizing the anterior pattern of the Drosophila embryo. Embo J 7: 1749–1756.
  45. 45. Sanchez L, Thieffry D (2001) A logical analysis of the Drosophila gap-gene system. J Theor Biol 211: 115–141.
  46. 46. Jaeger J, Blagov M, Kosman D, Kozlov KN, Manu , et al. (2004) Dynamical analysis of regulatory interactions in the gap gene system of Drosophila melanogaster. Genetics 167: 1721–1737.
  47. 47. Driever W, Nusslein-Volhard C (1989) The bicoid protein is a positive regulator of hunchback transcription in the early Drosophila embryo. Nature 337: 138–143.
  48. 48. Schroder C, Tautz D, Seifert E, Jackle H (1988) Differential regulation of the two transcripts from the Drosophila gap segmentation gene hunchback. EMBO J 7: 2881–2887.
  49. 49. Wharton RP, Struhl G (1991) RNA regulatory elements mediate control of Drosophila body pattern by the posterior morphogen nanos. Cell 67: 955–967.
  50. 50. Weigel D, Jurgens G, Klingler M, Jackle H (1990) Two gap genes mediate maternal terminal pattern information in Drosophila. Science 248: 495–498.
  51. 51. Hoch M, Gerwin N, Taubert H, Jackle H (1992) Competition for overlapping sites in the regulatory region of the Drosophila gene Kruppel. Science 256: 94–97.
  52. 52. Pankratz MJ, Busch M, Hoch M, Seifert E, Jackle H (1992) Spatial control of the gap gene knirps in the Drosophila embryo by posterior morphogen system. Science 255: 986–989.
  53. 53. Rivera-Pomar R, Niessing D, Schmidt-Ott U, Gehring WJ, Jackle H (1996) RNA binding and translational suppression by bicoid. Nature 379: 746–749.
  54. 54. Dubnau J, Struhl G (1996) RNA recognition and translational regulation by a homeodomain protein. Nature 379: 694–699.
  55. 55. Macdonald PM, Struhl G (1986) A molecular gradient in early Drosophila embryos and its role in specifying the body pattern. Nature 324: 537–545.
  56. 56. Jaeger J, Surkova S, Blagov M, Janssens H, Kosman D, et al. (2004) Dynamic control of positional information in the early Drosophila embryo. Nature 430: 368–371.
  57. 57. Noireaux V, Bar-Ziv R, Libchaber A (2003) Principles of cell-free genetic circuit assembly. Proc Natl Acad Sci U S A 100: 12672–12677.
  58. 58. Isalan M, Klug A, Choo Y (2001) A rapid, generally applicable method to engineer zinc fingers illustrated by targeting the HIV-1 promoter. Nat Biotechnol 19: 656–660.
  59. 59. Maher LJ (1992) Inhibition of T7 RNA polymerase initiation by triple-helical DNA complexes: A model for artificial gene repression. Biochemistry 31: 7587–7594.
  60. 60. Hulskamp M, Schroder C, Pfeifle C, Jackle H, Tautz D (1989) Posterior segmentation of the Drosophila embryo in the absence of a maternal posterior organizer gene. Nature 338: 629–632.
  61. 61. Hulskamp M, Pfeifle C, Tautz D (1990) A morphogenetic gradient of hunchback protein organizes the expression of the gap genes Kruppel and knirps in the early Drosophila embryo. Nature 346: 577–580.
  62. 62. Casanova J (1990) Pattern formation under the control of the terminal system in the Drosophila embryo. Development 110: 621–628.
  63. 63. Reinitz J, Levine M (1990) Control of the initiation of homeotic gene expression by the gap genes giant and tailless in Drosophila. Dev Biol 140: 57–72.
  64. 64. Simpson-Brose M, Treisman J, Desplan C (1994) Synergy between the hunchback and bicoid morphogens is required for anterior patterning in Drosophila. Cell 78: 855–865.
  65. 65. Surdej P, Jacobs-Lorena M (1998) Developmental regulation of bicoid mRNA stability is mediated by the first 43 nucleotides of the 3′ untranslated region. Mol Cell Biol 18: 2892–2900.
  66. 66. Stauber M, Jackle H, Schmidt-Ott U (1999) The anterior determinant bicoid of Drosophila is a derived Hox class 3 gene. Proc Natl Acad Sci U S A 96: 3786–3789.
  67. 67. Fuentes M, Kuperman MN, Boissonade J, Dulos E, Gauffre F, et al. (2002) Dynamical effects induced by long range activation in a nonequilibrium reaction-diffusion system. Phys Rev E Stat Nonlin Soft Matter Phys 66: 56205.
  68. 68. Shvartsman SY, Muratov CB, Lauffenburger DA (2002) Modeling and computational analysis of EGF receptor-mediated cell communication in Drosophila oogenesis. Development 129: 2577–2589.
  69. 69. Epps JL, Tanda S (1998) The Drosophila semushi mutation blocks nuclear import of bicoid during embryogenesis. Curr Biol 8: 1277–1280.
  70. 70. Wilkie GS, Davis I (2001) Drosophila wingless and pair-rule transcripts localize apically by dynein-mediated transport of RNA particles. Cell 105: 209–219.
  71. 71. Houchmandzadeh B, Wieschaus E, Leibler S (2002) Establishment of developmental precision and proportions in the early Drosophila embryo. Nature 415: 798–802.
  72. 72. Poustelnikova E, Pisarev A, Blagov M, Samsonova M, Reinitz J (2004) FlyEx database. Available: Accessed 08 January 2005.
  73. 73. von Dassow G, Meir E, Munro EM, Odell GM (2000) The segment polarity network is a robust developmental module. Nature 406: 188–192.
  74. 74. Barkai N, Leibler S (1997) Robustness in simple biochemical networks. Nature 387: 913–917.
  75. 75. Elowitz MB, Levine AJ, Siggia ED, Swain PS (2002) Stochastic gene expression in a single cell. Science 297: 1183–1186.
  76. 76. Blake WJ, M KA, Cantor CR, Collins JJ (2003) Noise in eukaryotic gene expression. Nature 422: 633–637.
  77. 77. Thiesen HJ, Meyer W (1993) Krab domains analyzed in human Cys/His-type zinc-finger proteins KOX 1, KOX 8, and KOX 19. Ann N Y Acad Sci 684: 243–245.
  78. 78. Brizzard BL, Chubet RG, Vizard DL (1994) Immunoaffinity purification of FLAG epitope-tagged bacterial alkaline phosphatase using a novel monoclonal antibody and peptide elution. Biotechniques 16: 730–735.
  79. 79. Moore M, Choo Y, Klug A (2001) Design of polyzinc finger peptides with structured linkers. Proc Natl Acad Sci U S A 98: 1432–1436.
  80. 80. Moore M, Klug A, Choo Y (2001) Improved DNA binding specificity from polyzinc finger peptides by using strings of two-finger units. Proc Natl Acad Sci U S A 98: 1437–1441.