New Research In
Physical Sciences
Social Sciences
Featured Portals
Articles by Topic
Biological Sciences
Featured Portals
Articles by Topic
 Agricultural Sciences
 Anthropology
 Applied Biological Sciences
 Biochemistry
 Biophysics and Computational Biology
 Cell Biology
 Developmental Biology
 Ecology
 Environmental Sciences
 Evolution
 Genetics
 Immunology and Inflammation
 Medical Sciences
 Microbiology
 Neuroscience
 Pharmacology
 Physiology
 Plant Biology
 Population Biology
 Psychological and Cognitive Sciences
 Sustainability Science
 Systems Biology
Dynamics of the Dorsal morphogen gradient

Contributed by Jennifer LippincottSchwartz, October 28, 2009 (received for review September 6, 2009)
Abstract
The dorsoventral (DV) patterning of the Drosophila embryo depends on the nuclear localization gradient of Dorsal (Dl), a protein related to the mammalian NFκB transcription factors. Current understanding of how the Dl gradient works has been derived from studies of its transcriptional interpretation, but the gradient itself has not been quantified. In particular, it is not known whether the Dl gradient is stable or dynamic during the DV patterning of the embryo. To address this question, we developed a mathematical model of the Dl gradient and constrained its parameters by experimental data. Based on our computational analysis, we predict that the Dl gradient is dynamic and, to a first approximation, can be described as a concentration profile with increasing amplitude and constant shape. These timedependent properties of the Dl gradient are different from those of the Bicoid and MAPK phosphorylation gradients, which pattern the anterior and terminal regions of the embryo. Specifically, the gradient of the nuclear levels of Bicoid is stable, whereas the pattern of MAPK phosphorylation changes in both shape and amplitude. We attribute these striking differences in the dynamics of maternal morphogen gradients to the differences in the initial conditions and chemistries of the anterior, DV, and terminal systems.
A tissue patterned by morphogen gradients can change its transcriptional state, grow, or deform either in response to the gradients or independently of them (1–3). When these changes are much slower than the dynamics of the gradient, a tissue responds to a stable signal. Transcriptional interpretation of such signals can rely on differences in the expression thresholds of target genes with respect to the spatially distributed repressors or activators (2, 4). A different strategy for signal interpretation is required when the formation of positional information becomes intertwined with the dynamics of the patterned system (2, 5). Here, we suggest that the dorsoventral (DV) patterning of the Drosophila embryo operates in this regime.
The DV patterning of the Drosophila embryo depends on the nuclear localization gradient of Dorsal (Dl), a protein related to the NFκB family of transcription factors (6–10). Transcriptional interpretation of the Dl gradient depends on the differences in the affinities of the Dl binding sites in the Dltarget genes and several gene expression and signaling cascades initiated by Dl (6, 11, 12). A ventraltodorsal occupancy gradient of the Toll cell surface receptor provides the activating signal for the DV patterning (13). In the absence of this signal, Dl is sequestered in the cytoplasm, in complex with an inhibitory protein IκB, called Cactus (Cact) in Drosophila. In response to Toll signaling, the Dl–Cact complex dissociates, Cact is degraded, and Dl enters the nucleus to control gene expression. In the current model of DV patterning, positional information is established by the spatial pattern of Toll occupancy (13, 14).
The Dl gradient forms during the last five nuclear divisions in a syncytical blastoderm, a single cell with multiple nuclei (15). Because nuclei can be viewed as competing with Cact for Dl, an increase in the number of nuclei can influence the Dl gradient, but whether or not this happens is currently unknown. Dl undergoes rapid nucleocytoplasmic shuttling with a nuclear residence time of several minutes (16). Nuclei change in volume and undergo five synchronous divisions (15, 17). To explore how these processes contribute to the formation of the Dl gradient, we formulated a mathematical model that accounts for the nuclear import and export of Dl, its interaction with Cact, and the dynamics of nuclear density and volumes in the syncytial blastoderm. Based on the computational analysis of this model and a number of our modelbased experiments, we argue that the Dl gradient is dynamic and, to a first approximation, can be described as a spatial pattern with constant shape and increasing amplitude.
Results
At the outset of this work, the only quantitative information about the spatial distribution of nuclear Dl could be found in the study by Zinzen et al. (18), who had characterized the DV pattern of nuclear Dl at a single time point. The domains of the Dltarget genes begin to form several nuclear cycles before cellularization, and it is important to determine whether these genes respond to a constant or timedependent signal. This question has been prompted by recent studies of the anteriorposterior (AP), DV, and terminal systems. First, the nuclear levels of Bcd, a morphogen that specifies the anterior structures of the embryo, are stable throughout the last five syncytial nuclear divisions. Bcd undergoes nucleocytoplasmic shuttling on the time scale of several minutes. After mitosis, the nuclear levels of Bcd drop to zero, but are then rapidly reestablished to the premitosis level. Thus, with the exception of a rapid transient associated with nuclear divisions, a particular point in the embryo is exposed to a constant level of nuclear Bcd, which is distributed in a spatial pattern with constant shape and amplitude. In contrast, the pattern of phosphorylated ERK/MAPK (dpERK), which specifies the terminal regions of the embryo, changes in both shape and amplitude. Over the five last nuclear divisions in the syncytial blastoderm, the nuclear levels of dpERK increase at the poles and decrease in the midbody of the embryo.
We asked whether the spatial pattern of nuclear localization of Dl is stable, similar to the pattern of Bcd, or dynamic, similar to the pattern of dpERK. Answering this question requires quantitative characterization of the nuclear levels of Dl along the DV axis and at multiple time points. Here, this goal is achieved using a combination of quantitative imaging and mathematical modeling of the biophysical processes associated with the DV patterning of the embryo. Our experimental and computational results reveal that the DV pattern of nuclear Dl behaves differently from both the Bcd and dpERK gradients. Similar to Bcd, the shape of the DV pattern of nuclear Dl has approximately constant shape, but the amplitude of this pattern increases with time.
To visualize the DV distribution of nuclear Dl, we used a transgenic line where one endogenous copy of dl was replaced by a dlgfp transgene (see Materials and Methods). Several control experiments were done to test whether the Dl gradient, quantified from the GFP signal in this line is close to the Dl gradient obtained by quantifying the wildtype αDl antibody stainings. First, fixed embryos were stained with αDl and αGFP antibodies and the fluorescent intensities of the nuclear αDl and αGFP stainings were compared with each other (Fig. 1 A–D). As shown in Fig. 1D, there is a strong linear correlation between the intensities of the αDl and αGFP antibody stainings. Thus, except for the differences in the background levels, the DV pattern of nuclear Dl obtained on the basis of the GFP staining is proportional to the pattern that is based on the αDl antibody staining. Because the αDl antibody detects both endogenous and GFPtagged Dl, whereas the αGFP antibody detects only GFPtagged Dl, the linear correlation suggests that the GFP tag does not significantly interfere with the normal processes of Dl transport and interactions. As an additional control experiment, we used the αDl antibody and stained the wildtype embryos together with embryos with one wildtype and one GFPtagged copy of dl (Fig. 1E). Comparison of the nuclear Dl gradients in the wildtype and transgenic embryos (Fig. 1F) indicates that they are very close to each other. Thus, the line with one wildtype and one GFPtagged dl can be used to monitor the dynamics of the Dl gradient in live imaging experiments.
To follow the dynamics of the DV pattern of nuclear Dl, we used the “end on” imaging technique (19), where embryos are mounted with their AP axis perpendicular to the horizontal surface, enabling the imaging along the DV axis of the embryo. The space–time plot of nuclear Dl extracted from a liveimaging experiment with >130 time points between cycles 11 and 14 is shown in Fig. 2A. The gradient of nuclear Dl is very dynamic. Concentration of nuclear Dl increases during interphase, which is followed by a drop to a low value during mitosis and a subsequent increase during the next cycle (Fig. 2 A and B).
One of the main sources of variability in the endon imaging of the Dl gradient is introduced by the movement of nuclei in and out of the focal plane, both during interphase and between different nuclear division cycles. The inherent dynamics of the arrangement of the nuclei induces large variability in the profiles of the Dl gradients along the DV axis. This effect, which is particularly significant during the earlier nuclear division cycles, makes the quantitative analysis of the Dl gradients during these cycles extremely challenging. At the same time, the arrangement of nuclei is much more regular during cycle 14, when they are tightly packed under the plasma membrane. Thus, we use cycle14 data from four separate live imaging experiments to extract the DV pattern of nuclear Dl at a fixed time point: ≈15 min into cycle 14 (Fig. 2C). To characterize the dynamics of the DV pattern of nuclear Dl at the previous time points, we use the data in Fig. 2C as a quantitative constraint for the mathematical model that accounts for the dynamics of Dl/Cact interactions and nuclear divisions.
The objective of our model is to characterize the dynamics of the DV pattern of nuclear Dl during the last five syncytial cell cycles. We model the syncytium as a periodic arrangement of cuboidal compartments, each of which contains a single spherical nucleus and an associated cytoplasmic region (Fig. 3A). Dl, Cact, and Dl–Cact complex diffuse rapidly within each of the compartments and undergo slow exchange with the neighboring compartments. The kinetic part of our model is a subset of reactions included in the more detailed models of the mammalian NFκB system (Fig. 3B) (20). The association of Dl and Cact is modeled as a secondorder reaction with a spatially uniform rate constant. The dissociation of the Dl–Cact complex is modeled as a firstorder process with a rate constant that depends on the DV position, reflecting a DV pattern of Toll activation. We assume that this pattern remains constant during the entire patterning process. The rates of the nuclear import and export of Dl depend on the surface area of the nucleus. Finally, we assume free Cact is produced at a constant rate and degraded in a firstorder reaction.
Within each nuclear cycle nuclear radius increases (17). As a consequence of this change, nuclear and cytoplasmic concentrations are affected by both the chemical processes and volume changes. At specific time intervals that correspond to the detailed measurements of Foe and Alberts (15), nuclei divide. During mitoses, the content of each nucleus is instantaneously redistributed in the corresponding cytoplasmic region. After mitosis, the nuclear envelope reforms, and the number of syncytial compartments doubles. The model for doubling of compartments is shown in Fig. 3C; more details can be found in SI Appendix (Fig. S1).
The dynamics of the model depend on nine dimensionless parameters that characterize the spatial pattern of Toll activation, the rates of nuclear import and export of Dl, the relative amounts of total Dl and Cact, the rates at which the species move between the adjacent cytoplasmic compartments, and the formation and degradation rates of the Dl–Cact complex (see SI Appendix, Table S1). Given these parameters, numerical solution of the model predicts the spatial patterns of Dl, Cact, and Dl–Cact complex during nuclear cycles 10–14. Although the exact values of model parameters are largely unknown, they can be constrained by the available experimental data. Specifically, it is known that the Dl levels are mainly nuclear/cytoplasmic at the most ventral/dorsal points, respectively and that the levels of Cact at the dorsal side of the embryo are higher than those at the ventral side (2–4, 14). Based on the results of the live imaging studies, we required that the ventralmost level of nuclear Dl increases monotonically within the interphase of cycle 14 and that this level reaches 90% of its final value in <15 min (16). Finally, we used the spatial pattern of nuclear Dl at ≈15 min in cycle 14 (Fig. 2C), based on the quantification of the GFP autofluorescence from our live imaging experiments. This information does not define the parameters of our model uniquely. Rather, it constrains them to a “cloud,” or ensemble, in the ninedimensional space of parameters. Each point in this ensemble, i.e., a particular parameter vector, can be used to predict the properties of the DV system that cannot be readily measured experimentally. In particular, we are interested in the dynamics of nuclear Dl levels at all times from nuclear cycle 10 to cycle 14 and along the entire DV axis.
Thus, we start with a relatively small amount of experimental data on the DV pattern of nuclear Dl at a specific time point during the cycle 14, identify an ensemble of parameter vectors that satisfy these constraints, and then use this ensemble to predict the Dl gradient at all times. A similar approach has been successfully used to make modelbased statistical predictions about the dynamics of other cell signaling systems (21). To identify parameter sets that satisfy the experimentally based constraints for our model, we used a stochastic evolutionary optimization technique (22). The Dl gradient dynamics predicted for one particular parameter vector is presented in Fig. 4A that shows a surface plot that represents the dynamics of nuclear Dl at all times and all positions along the DV axis. Fig. 4B shows a comparison of the nuclear Dl gradients reached at the end of all nuclear cycles, and Fig. 4C shows the dynamics of the nuclear Dl level at the ventralmost point. After obtaining a large set of acceptable parameters (≈40,000), we used the resulting ensemble as the basis for statistical analysis of the Dl gradients predicted by the model.
As a first step in the statistical analysis of model predictions, we used the identified ensemble to calculate a distribution function for the ratio of the amplitudes of the Dl gradients at the end of nuclear cycles 14 and 10 (A_{14}/A_{10}; Fig. 5Ai). We chose this metric based on the previous analysis of the gradients of dpERK, which patterns the terminal regions of the embryo, and the Bcd gradient, which patterns the AP axis (17, 23). For the nuclear Bcd gradient, which is stable during cycles 10 and 14, A_{14}/A_{10} ≈ 1. For the dpERK gradient, however, A_{14}/A_{10} > 1, which reflects a pattern that is amplified at the poles. The distribution function for the ratio of the maxima of the spatial patterns of nuclear Dl at cycles 14 and 10 has a single peak at ≈1.5 (Fig. 5Aii). Upon further examination, we found that approximately twothirds of the parameter sets predict a continuous increase in the ventral levels of nuclear Dl at the end of each cycle (from cycle 10 to 14; Fig. 5Aii Lower Inset). The remaining parameter sets predict multiple combinations of increase and decrease in nuclear Dl level from one cycle to another (Fig. 5Aii Upper Inset).
To test whether the amplitudes of the Dl gradients change monotonically between subsequent nuclear cycles, embryos expressing the histoneGFP transgene, which marks the nuclei, were stained using the αDl antibody. In this case, we used lateral images of embryos and located the peak of the gradient in each image (see Materials and Methods). Using a previously developed image processing approach (23), we assigned each embryo to one of the four temporal classes, corresponding to the nuclear cycles 11, 12, 13, and 14. For each group, we measured the amplitude of the gradient at the ventralmost point in the embryo (Fig. 5B). We analyzed the resulting dataset using a linear regression model and found a strong correlation (P < 0.001) between the age of the embryo (i.e., the nuclear division cycle) and the amplitude of the gradient (Fig. 5B). Based on these results, we conclude that the amplitude of the Dl gradient is an increasing function of the number of nuclei. Based on this, we restricted the further statistical analysis to those members of the identified ensemble of parameter sets that satisfied this additional constraint (≈67% of the parameter vectors in the original ensemble).
Focusing on these remaining twothirds of parameter sets, we examined how the shape of the Dl gradient changes from one nuclear cycle to another. We constructed a distribution function for the change in the halfwidth of the gradients from cycles 10 to 14 (Fig. 5C). Strikingly, we found that this distribution function has a peak at 0.05 and corresponds to the halfwidth change of at most 15%, which implies that the shape of the gradient is only weakly affected by nuclear divisions. We have also confirmed that the same is true for other nuclear cycles. In fact, the Dl gradients at the end of the five nuclear cycles can be collapsed into one shape by simple rescaling (Fig. 5D). To a first approximation, the DV distribution of nuclear Dl can be viewed as a pattern with constant shape and increasing amplitude. In addition to characterizing how the amplitude of the gradient changes from one nuclear cycle to another, we found that all points along the DV axes are exposed to a monotonically increasing level of Dl within each nuclear cycle. Thus, based on the statistical analysis of the gradients predicted by the model, we conclude that the nuclear levels of Dl are monotonically increasing in time during the interphase of syncytial nuclear divisions; after mitosis, they are reestablished, reaching a value that it is higher than in the previous cycle.
Discussion
The DV patterning of the Drosophila embryo by the Dl gradient is arguably the beststudied morphogenetic patterning event (6, 11, 24, 25). Multiple genes controlled by Dl were identified, and sequencespecific patterns of their transcriptional regulation have become progressively understood. At the same time, the spatial distribution of Dl and its dynamics have not been directly characterized. Both of these pieces of information are essential for quantitative understanding of the DV patterning. For instance, the relative arrangement of the expression domains of the Dl target genes has been interpreted within the framework of thermodynamic models (18, 25). A key assumption of such models is that the input “seen” by the regulatory regions of the Dl target genes is stable, but whether or not this is the case is unknown. Given the fact that the Dl undergoes nucleocytoplasmic shuttling in a medium with increasing number of nuclei, the answer to the question about the stability of the nuclear levels of Dl is far from obvious. Here, we answer this question based on the imaging experiments with fixed and live embryos, mathematical modeling, and computational analysis.
Each of these approaches has its relative advantages and limitations. Experiments with fixed embryos have limited temporal resolution and cannot follow the dynamics of the gradient within the same embryo. We have used imaging of fixed embryos to test the applicability of the DlGFP transgene and characterize how the ventralmost levels of nuclear Dl change from one nuclear cycle to the next. In contrast to experiments with fixed embryos, live imaging provides a dynamic view of the Dl pattern as a function of time. At the same time, highresolution images are typically collected from a single optical section, and quantitative analysis of the Dl dynamics is compounded by the dynamics of nuclear rearrangements. We have used live imaging to quantify the spatial pattern of Dl at a specific time point when the arrangement of nuclei is stable. This information provides one of the inputs for our mathematical model, which is clearly an approximation of processes in the real embryo. We assumed that the spatial pattern of Toll occupancy and the total amount of Dl in the embryo remain constant and used a simple geometric model to describe the syncytial embryo. As more data become available, our model can be extended to describe these additional effects. For example, including the Dldependent synthesis of Cact can explain why the shape of the Dl gradient is dynamic during the DV patterning in other insects (26, 27). Nevertheless, by capturing what we believe are the essential features of the Dl gradient and its interaction with the dynamics of the syncytium, our model provides a basis for more complex mathematical models that are essential for understanding the DV patterning in the wildtype and mutant backgrounds and exploring the evolvability of the DV patterning system (24, 28).
Based on the computational analysis of this model, we argue that the Dl morphogen is distributed in a dynamic pattern with increasing amplitude and constant shape. These dynamics are different from the dynamics of Bcd and MAPK phosphorylation gradients, which are formed in the same medium and at the same time as the Dl gradient (29). Specifically, neither the shape nor the amplitude of the nuclear levels of Bcd are affected by changes in the number of nuclei. The twopeaked pattern of dpERK, however, changes in both shape and amplitude.
What is the origin of these striking differences between the dynamics of the three gradients? The terminal and DV gradients are initiated only when nuclei have reached the plasma membrane and respond to signals from the activated cell surface receptors. At the same time, the functional Bcd gradient is already established at this stage (30). Thus, the stability of the nuclear Bcd gradient during the last five nuclear divisions can be due to the fact that this gradient starts to form earlier and is not affected by nuclei, which sample only a small fraction of total Bcd at any given point along the AP axis (17, 31). Another key factor is the difference in the “chemistries” of Bcd, Dl, and dpERK molecules. We have previously proposed that Bcd is not degraded on the time scale relevant for the formation of the gradient (31). In contrast, the molecules that pattern the DV and terminal regions are reactive: Dl interacts with Cact and dpERK is dephosphorylated. Both of these processes can be affected by changes in nuclear density. Why, then, does the same change in nuclear density lead to different changes in the Dl and dpERK gradients?
We attribute the fact that amplification of the dpERK levels is restricted to the poles, whereas the nuclear Dl levels increase everywhere, to the different spatial extents of the sources that activate the DV and terminal systems. Our recent experiments suggest that the occupancy of the Torso receptor, which activates MAPK, is sharply localized to the poles and that the formation of the dpERK gradient relies on the diffusion of activated MAPK to the midbody regions (23, 32). By decreasing the distance to which the dpERK molecule can diffuse before being trapped or dephosphorylated, an increase in the nuclear density can prevent the lateral transport of dpERK to the midbody regions and amplify its level at the poles. However, we argue that a significant fraction of nuclei along the DV axis are exposed to appreciable levels of Toll signaling. As a consequence, lateral movement of the intracellular signaling components is less important for the DV patterning than for patterning of the terminal system, which agrees with the imaging experiments that revealed a slow Dl exchange between the adjacent cytoplasmic regions (16).
Understanding how a dynamic Dl gradient specifies multiple gene expression boundaries along the DV axis requires quantitative studies of the dynamics of other patterning signals in the early embryo (18, 33–37). As an example, we discuss the expression of genes that are expressed in the ventral and lateral regions of the embryo. Consider the regulation of sog, a gene repressed in the prospective mesoderm and expressed in more lateral regions. It is possible that the ventral boundary of this pattern is stabilized by the feedforward loop that is induced by Dl and relies on the positive autoregulation of Twist, a high threshold target of Dl (18). If this network operates in a bistable regime, then its output, and hence the ventral boundary of sog, can be stable even with increasing levels of Dl. The dorsal boundary, however, depends on the joint regulation of sog by both Dl and Zelda, a maternally provided activator of early zygotic transcription (38, 39). We speculate that the dorsal boundary of the sog pattern is stabilized by the combined effect of the temporally increasing levels of spatially distributed Dl and decreasing levels of spatially uniform Zelda. Thus, increasing levels of nuclear Dl can be important for the dynamic expression of Dltarget genes. Direct tests of this prediction can rely on the combined experimental, modeling, and computational strategy described in this work.
Materials and Methods
Fly Lines.
Dorsal mutants dl^{1} and dl^{4} were obtained from the Bloomington Stock Center at the Indiana University and crossed with the dlGFP/TM3 line. Live imaging was performed with embryos of the genotype dl^{1}/+;p[w+dlGFP]/+.
Antibody Staining and Live Imaging.
Embryos were dechorionated in bleach for 1 min and then washed in 0.7% NaCl containing 0.05% Triton X100. They were then treated with heptane for 30 s and fixed in a 1:1 mixture of heptane/4% paraformadehyde in PBS for 20 min at room temperature. After fresh heptane was added, embryos were devitelinized in equal volumes of methanol and stored in methanol at −20 °C. For immunostaining, embryos were rehydrated for 10 min in PBS with 0.3% Triton X100 (PBT) and blocked for 1 h with 2% BSA in PBT at room temperature; this was followed by a 2h incubation with the primary antibody. The primary antibodies used were mouse antiDorsal 7A4 (Developmental Studies Hybridoma Bank at the University of Iowa) at 1:50 and rabbit antiGFP (Molecular Probes) at 1:1,500. Embryos were washed in PBT and subsequently stained with fluorescently coupled secondary antibodies (Molecular Probes). For endon imaging, embryos were mounted on their posterior pole and imaged as described (19).
Embryos were imaged at ≈70 μm from the posterior pole, at a position where the diameter of the embryo was ≈140–150 μm. Control embryos for secondary background, wildtype embryos, and dl1;GFPdorsal embryos were immunostained on the same day and imaged under the same settings of the confocal microscope. For quantification, small regions of interest were drawn in the nuclei marked by Hoechst dye (Molecular Probes) using the software Image J. Average fluorescence intensity with corresponding positions of nuclei was obtained for antiDl and antiGFP. The gradients were normalized between zero and one by subtracting the minimum intensity measurement within a nucleus and then dividing the entire gradient by the maximum intensity.
Live imaging was performed with dl1 heterozygous embryos expressing GFPdorsal (dl^{1}/+;p[w+GFPdorsal]/+) to match the wildtype dl copy number. Flies were allowed to lay embryos for 1 h at 25 °C on grape juice agar plates. Embryos were dechorionated and mounted on their posterior pole in Lab Tek chambers coated with silane (40), immersed in PBS, and imaged on an LSM 510 confocal microscope. Imaging was performed at ≈70 μm from the posterior pole. Quantification was performed with Image J.
For lateral imaging of fixed embryos, they were dechorionated for 1 min in 100% bleach, fixed for 20 min by gentle shaking on a nutator, and devitellinized by vigorous 1min shaking in a mixture of heptane and methanol. Next, embryos were quickly rehydrated and transferred to the blocking and antibody steps of the protocol. All further processing was done with 0.02% PBSTriton X100 as the diluting solution. Dl antibody (Developmental Hybridoma Bank) was used in 1:100 dilution, and goat antimouse Alexa Fluor 546 antibody (Invitrogen; 1:500 dilution) was used as a secondary antibody. Imaging was done on a Zeiss LSM510 confocal microscope, with a Zeiss 20× (NA 0.6) Aplan objective.
Acknowledgments
We thank Tsuyoshi Hirashima for help with image processing; Dmitry Papatsenko, Christine Rushlow, Ruth Steward, and Mike Levine for helpful discussions; and Christine Sample, Miriam Osterfield, Alistair Boettiger, and Lily Cheung Chang for critical reading of the manuscript.
Footnotes
 ^{1}To whom correspondence may be addressed. Email: jlippin{at}helix.nih.gov or stas{at}princeton.edu

Author contributions: J.S.K., R.R., Y.K., R.D., J.L.S., and S.Y.S. designed research; J.S.K., R.R., Y.K., and S.Y.S. performed research; J.S.K., R.R., Y.K., V.K.L., J.L.S., and S.Y.S. analyzed data; and J.S.K., J.L.S., and S.Y.S. wrote the paper.

The authors declare no conflict of interest.

This article contains supporting information online at www.pnas.org/cgi/content/full/0912395106/DCSupplemental.
References
 ↵
 ↵
 Ibanes M,
 Belmonte JCI
 ↵
 Jaeger J,
 Irons D,
 Monk N
 ↵
 Ashe HL,
 Briscoe J
 ↵
 ↵
 Hong JW,
 Hendrix DA,
 Papatsenko D,
 Levine MS
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 Chopra VS,
 Levine M
 ↵
 ↵
 ↵
 Foe VE,
 Alberts BM
 ↵
 DeLotto R,
 DeLotto Y,
 Steward R,
 LippincottSchwartz J
 ↵
 ↵
 ↵
 ↵
 Cheong R,
 Hoffmann A,
 Levchenko A
 ↵
 ↵
 Coley DA
 ↵
 ↵
 Markstein M,
 et al.
 ↵
 Papatsenko D,
 Levine M
 ↵
 ↵
 ↵
 Crocker J,
 Tamori Y,
 Erives A
 ↵
 ↵
 ↵
 ↵
 Berezhkovskii AM,
 Coppey M,
 Shvartsman SY
 ↵
 Umulis DM,
 Serpe M,
 O'Connor MB,
 Othmer HG
 ↵
 ↵
 Mizutani CM,
 Meyer N,
 Roelink H,
 Bier E
 ↵
 ↵
 ↵
 ↵
 ↵
 Mavrakis M,
 Rikhy R,
 Lilly M,
 LippincottSchwartz J
Citation Manager Formats
Sign up for Article Alerts
Article Classifications
 Biological Sciences
 Developmental Biology