Dynamics of TGF-β signaling reveal adaptive and pulsatile behaviors reflected in the nuclear localization of transcription factor Smad4

Abstract

The TGF-β pathway plays a vital role in development and disease and regulates transcription through a complex composed of receptor-regulated Smads (R-Smads) and Smad4. Extensive biochemical and genetic studies argue that the pathway is activated through R-Smad phosphorylation; however, the dynamics of signaling remain largely unexplored. We monitored signaling and transcriptional dynamics and found that although R-Smads stably translocate to the nucleus under continuous pathway stimulation, transcription of direct targets is transient. Surprisingly, Smad4 nuclear localization is confined to short pulses that coincide with transcriptional activity. Upon perturbation, the dynamics of transcription correlate with Smad4 nuclear localization rather than with R-Smad activity. In Xenopus embryos, Smad4 shows stereotyped, uncorrelated bursts of nuclear localization, but activated R-Smads are uniform. Thus, R-Smads relay graded information about ligand levels that is integrated with intrinsic temporal control reflected in Smad4 into the active signaling complex.

A small number of signaling pathways are used repeatedly throughout metazoan development, and their effects depend upon timing and context (1). Extensive biochemical characterization of these developmental signaling pathways has elucidated the sequence of events leading from ligand binding at the cell surface to regulation of transcription. Proper temporal control of pathway activity is crucial for normal development; however, the dynamic aspects of signaling are difficult to infer from population data and have lagged behind dissection of pathway components (2, 3). The few cases that have been examined have revealed rich dynamics that could not have been predicted from knowledge of the molecular interactions or from bulk measurements of protein modifications or mRNA levels (2–4).

The TGF-β pathway is essential for developmental processes including mesoderm specification and dorsal–ventral axis formation and is dysregulated in a variety of cancers. It also is an important model for pathway crosstalk and dynamics, because it has two branches that share several components including receptors and transcription factors (5, 6). Binding of ligands specific to each branch to receptor complexes leads to the phosphorylation of branch-specific transcription factors: TGF-β/activin/nodal ligands induce the phosphorylation of Smad2/3, whereas bone morphogenic proteins (BMPs) activate Smad1/5/8. Phosphorylation of the receptor-activated Smads (R-Smads) from either branch of the pathway results in complex formation with Smad4, nuclear accumulation, and transcriptional activation.

The prevailing model is that R-Smads carry pathway information with Smad4 mirroring their activity (7, 8). R-Smad phosphorylation is necessary for the nuclear accumulation and transcriptional activity of both the R-Smads and Smad4. Termination of signaling often is presumed to be caused by either degradation or dephosphorylation of activated R-Smads (9, 10), and it further is assumed that the continuous presence of activated R-Smads is synonymous with continuous transcriptional activity. We reexamine these issues here using dynamic measurements. We show that R-Smad phosphorylation and nuclear accumulation stably reflect the ligand level to which the cells are exposed; however, both nuclear Smad4 and transcriptional activity show an adaptive response, pulsing in response to changing ligand concentration and then returning to near-baseline levels. These results show that transcriptional dynamics are governed predominantly by feedback reflected in Smad4 but not receptor or R-Smad dynamics and force a substantial revision of the conventional view about a pathway ubiquitous in development and disease.

Results

Nuclear Smad2 Levels Stably Reflect the External Ligand Concentration.

To observe the dynamic response of Smad2 in single cells, we used the ePiggyBac transposable element system (11) to generate a clonal cell line of mouse myoblast C2C12 cells stably expressing an RFP-Smad2 fusion protein (Fig. 1A). These cells also stably express GFP-nuclear localization signal (NLS) to allow automated identification and tracking of cells using custom software. Western blot analysis revealed that RFP-Smad2 is expressed at levels close to that of endogenous Smad2 and is phosphorylated (denoted hereafter “pSmad2”) at its C terminus in response to signal (compare Fig. S1A and Fig. 1F), and quantitative RT-PCR (qRT-PCR) analysis showed similar induction of TGF-β target genes upon ligand stimulation in the modified cell line as compared with the parental cell line (compare Fig. S1B and Fig. 2B). RFP-Smad2 accumulated in the nucleus in response to stimulation with TGF-β1 (Fig. 1B). Continued nuclear localization of Smad2 was dependent on continuous signaling, because the addition of SB431542, a specific inhibitor of type I TGF-β receptors (12), 1 h after stimulation caused nuclear fluorescence to return to below the starting baseline within 4 h (Fig. 1C). Thus, RFP-Smad2 is a reliable reporter for the activation status of Smad2, accumulating in the nucleus when the pathway is stimulated with ligand and returning to the cytoplasm upon pathway inhibition.

Fig. 1.

Smad2 activation is stable under continuous stimulation. (A) Constructs used to generate the RFP-Smad2 cell line. Triangles denote the ePiggyBac terminal repeats, arrows denote promoters, and gray boxes denote antibiotic resistance genes. (B) Images of live untreated cells (Upper) or cells treated with 5 ng/mL TGF-β1 for 1 h (Lower) showing accumulation of RFP-Smad2 in the nucleus upon TGF-β1 treatment. (C) Quantification of average nuclear RFP-Smad2 from live cell imaging in cells exposed first to 5 ng/mL TGF-β1 and then exposed to 10 uM SB431542 1 h later, showing that pathway inhibition results in relocalization of RFP-Smad2 to the cytoplasm (D) Time courses of nuclear RFP-Smad2 show stable nuclear accumulation in cells exposed to either high (5 ng/mL) or low (0.1 ng/mL) doses of TGF-β1. Black lines represent averages of all cells in the field of view (>50 cells). Colored lines represent selected single cells. Cell nuclei were identified automatically and tracked through time using the GFP-NLS nuclear marker. (E) Immunofluorescent staining for Smad2/3 at the indicated times after continuous application of 1 ng/mL TGF-β1 showing stable enrichment of endogenous Smad2/3 in the nucleus. (F) Western blots of total cell lysates from cells continuously exposed to 5 ng/mL TGF-β1 for the indicated times show stably elevated pSmad2/3 levels.

Fig. 2.

TGF-β–mediated transcriptional responses adapt under continuous stimulation. (A) Production of luciferase from TGF-β–sensitive reporters is limited to the first 3–4 h following stimulation. C2C12 Cells were transiently transfected with the CAGA12 or 3TP constructs. Forty-eight hours after transfection cells were treated with TGF-β and then with SB431542 at the indicated times after the addition of TGF-β. After 8 h, all samples were collected together and analyzed for luciferase activity. Activity of the TGF-β reporters was normalized using constitutive Renilla expression from the pRL-Tk plasmid. (B) Expression of TGF-β target genes by qRT-PCR as a function of time after TGF-β1 treatment showing a transient response to continuous TGF-β1 stimulation.

We used the RFP-Smad2 cell line to assess the response of single cells to varying doses of TGF-β1 ligand. In contrast to other signaling pathways such as NFκB that display binary on/off responses (3), essentially all cells responded to ligand at both high and low doses. The response was graded, with increasing TGF-β1 concentration resulting in increasing nuclear RFP-Smad2 (Fig. S1C). At the level of single cells, this response resulted in a rightward shift in the histogram of nuclear RFP-Smad2 values, but the distribution remained single peaked (Fig. S1D). At doses greater than 1 ng/mL, the response saturated, with RFP-Smad2 exclusively nuclear and all cells responding. We used time-lapse imaging to determine the dynamic response of Smad2 to stimulation with both high and low doses of ligand in individual cells (Fig. 1D and Movie S1). In both cases, response to ligand was detectable within 10 min and peaked at 1–2 h. Consistent with the results shown in Fig. S1D, nearly all cells responded at both high and low ligand doses. We found that nuclear enrichment of RFP-Smad2 was extremely long lived and lasted for the entire duration of the movie (at least 12 h) at both high and low ligand doses.

We confirmed these results by detecting endogenous Smad2/3 in the parental cell line by immunofluorescence (Fig. 1E). Nuclear accumulation of endogenous Smad2/3 occurred in a graded fashion in response to ligand stimulation (Fig. S1E) and was very long lived (Fig. 1E). Smad2/3 still was almost exclusively nuclear after 40 h of continuous stimulation with some reduction in intensity compared with 1.5 h, likely because of preferential degradation of phosphorylated Smad2 (13, 14). Consistent with the continuous presence of Smad2/3 in the nucleus, Western blotting revealed that cells continuously exposed to ligand had high levels of pSmad2/3 at all time points examined (up to 12 h) (Fig. 1F). We obtained similar results in the HaCaT cell line (Fig. S1 F and G). Taken together, our results imply that Smad2/3 activation stably reflects the level of ligand to which cells are exposed and will rise and fall with the levels of receptor activation.

Transcriptional Response to TGF-β Is Adaptive.

Given that Smad2 exhibits stable nuclear accumulation upon continuous TGF-β1 stimulation (Fig. 1 D and F), we asked whether the dynamics of TGF-β1–dependent transcription are sustained similarly. We first examined transcription from CAGA12 and 3TP synthetic luciferase reporters. CAGA12 is a reporter construct containing 12 copies of the Smad-binding element known as a “CAGA box” (15), whereas the 3TP reporter contains a 105-bp TGF-β–inducible element from the human Pai-1 promoter (16). To follow the kinetics of signaling, we stimulated cells with TGF-β1 and then added SB431542 at variable times after stimulation (17). Because luciferase is stable over the time scale of the experiment, the luciferase activity should continue to increase as long as the signaling pathway is active. The time point at which addition of the inhibitor no longer affects the final luciferase activity is also the time point at which the signaling pathway is no longer active. Surprisingly, and in contrast to our results for Smad2, we found that for both reporters most signaling-dependent transcription occurred within the first 2 h and that adding the inhibitor at 4 h had little or no effect (Fig. 2A). This outcome suggests that, despite the continuously elevated levels of nuclear, activated Smad2/3, transcriptional activity returns to baseline levels within 4 h.

We also examined the kinetics of endogenous target gene (18) expression by qRT-PCR in unmodified C2C12 cells. The response of all three target genes we studied was transient, rising over the first 2 h after stimulation and returning to baseline within 4–6 h (Fig. 2B). The dynamics of transcription in the RFP-Smad2 cells are very similar to those of the unmodified cell line (Fig. S1B). Thus, both luciferase reporter and qRT-PCR assays support a window of ∼4 h for TGF-β–dependent transcription under continuous ligand stimulation. The discrepancy between the dynamics of nuclear activated Smad2 and transcription was surprising, because activated R-Smads in the nucleus typically are assumed to indicate that the pathway is active. We reasoned that the short-lived transcription may result from the transient activity of other pathway components and sought to determine the status of the R-Smad binding partner Smad4, which is required for TGF-β–mediated transcriptional responses.

Smad4 Response Is Transient and Coincides with Transcription.

To monitor the dynamics of Smad4 in single living cells, we created a clonal C2C12 cell line expressing a GFP-Smad4 fusion protein as well as unlabeled Smad2 to enhance the Smad4 response (8, 19) and an RFP-H2B fusion protein for automated cell identification and tracking (Fig. 3A). Smad2 expression was similar to that in the RFP-Smad2 cell line (Fig. S2 B and C), and Smad2 phosphorylation (Fig. S2A) and nuclear accumulation (Fig. S2D) were similar to that in the parental cell line. Importantly, the dynamics of TGF-β target gene induction were nearly identical to those in unmodified C2C12 cells (compare Fig. 2B and Fig. S2E). Under continuous TGF-β1 stimulation, Smad4 accumulated rapidly in the nucleus of nearly every cell (Fig. 3B and Movie S2). Surprisingly, however, we found that, in contrast to Smad2, after the initial burst of nuclear localization, Smad4 relocalized to the cytoplasm after 4 h of continuous TGF-β1 stimulation (Fig. 3 B–D and Movie S2). Quantification of nuclear fluorescence in single cells confirmed that all cells had returned to near-baseline values within 4 h (Fig. 3C). Thus, the timing of the transcriptional response coincides with Smad4 but not Smad2 nuclear accumulation (Fig. 3D).

Fig. 3.

Smad4 adapts to a step in ligand concentration, and Smad4 nuclear localization coincides with transcription. (A) Constructs used in generating the Smad4 cell line. Notation is as in Fig. 1A. (B) Snapshots from a time-lapse movie of cells exposed to 5 ng/mL TGF-β1 at time t = 0. Smad4 is enriched in the nucleus after 1 h but becomes mostly cytoplasmic by 4 h. (C) Quantification of single-cell trajectories (colored lines) and average trajectory (black line). (D) Comparison of average RFP-Smad2 and GFP-Smad4 fluorescence in the nucleus as a function of the duration of continuous TGF-β1 treatment. Smad2 nuclear accumulation is sustained, whereas Smad4 is transient. (E) Smad4 amplitude but not kinetics is dependent on the dose of TGF-β1 ligand. Comparison of average nuclear to cytoplasmic GFP-Smad4 in cells treated with either 0.1 or 5 ng/mL TGF-β1. (F) Smad4 kinetics does not require continuous signaling. C2C12 GFP-Smad4 Cells were treated with 5 ng/mL TGF-β1 only or with 5 ng/mL TGF-β1 and then with 10 μM SB431542 1 h later. Traces show average nuclear fluorescence as a function of time.

The kinetics of the Smad4 pulse were not dependent on ligand dose, because cells treated with high (5 ng/mL) and low (0.1 ng/mL) doses of TGF-β1 produced pulses of the same duration but different amplitudes (Fig. 3E). These kinetics also were not dependent on continuous signaling from the receptor. Withdrawal of the ligand or treatment of cells with SB431542 1 h after TGF-β1 stimulation had little effect on the kinetics of the pulse and affected only the baseline level to which signaling returned (Fig. 3F). The dynamics of the Smad4 pulse are not particular to the levels of Smad expression in our cell line, because the dynamics of all responding cells in the polyclonal cell line from which the clonal line was selected were nearly identical. These results show that Smad4 undergoes stereotyped pulses of nuclear localization whose timing does not depend on ligand dose or on the duration of stimulation.

We sought to confirm these results by detecting endogenous Smad4 by immunofluorescence in the parental C2C12 cell line; however, all antibodies tested showed either weak or nonspecific staining. Immunofluorescent analysis of Smad4 in HaCaT cells revealed that in dense cultures Smad4 uniformly shifted to the nucleus in response to TGF-β1 treatment and then returned to baseline within 4 h, paralleling the time course in C2C12 cells (Fig. S2 F and H). Fractionating the nuclear and cytoplasmic compartments in dense cultures of HaCaT cells also revealed transient responses for Smad4 and sustained responses for Smad2 (Fig. S2I). Interestingly, in low-density cultures, nuclear localization was heterogeneous, and treatment with TGF-β1 did not synchronize the cells; Smad4 remained heterogeneous (Fig. S2G). Smad4 nuclear localization in both cases was caused by TGF-β signaling, because it was prevented by treatment with SB431542. These results should be contrasted with those for Smad2 in HaCaT cells, which showed long-lived, uniform nuclear accumulation in response to TGF-β1 in both low- and high-density cultures (Fig. S1). Based on the heterogeneity of Smad4 localization in sparse cultures of HaCaT cells, we speculate that under these conditions Smad4 may show repeated asynchronous pulses of nuclear localization similar to those observed in the Xenopus embryo (see below).

Relocalization of Smad4 to the Cytoplasm Is Dependent on New Protein Synthesis.

We next investigated whether negative feedback dependent on new protein synthesis is required for the removal of Smad4 from the nucleus in C2C12 cells. Consistent with this hypothesis, treatment with the protein synthesis inhibitor cycloheximide greatly prolonged the time that GFP-Smad4 spends in the nucleus under continuous TGF-β1 stimulation (Fig. 4 A and B and Movie S3).

Fig. 4.

Smad4 adaptation requires new protein synthesis, and transcriptional dynamics correlate with Smad4 but not Smad2 nuclear accumulation. (A) Snapshots from time-lapse imaging of GFP-Smad4 cells treated with TGF-β1 and cycloheximide showing sustained nuclear accumulation of Smad4. (B) Quantification of the average ratio of nuclear to cytoplasmic GFP-Smad4 in cells treated with TGF-β1 and either cycloheximide or MG132. Smad4 adaptation requires new protein synthesis but not protein degradation. (C) Quantification of nuclear RFP-Smad2 in cells treated with TGF-β1 and either MG132 or cycloheximide. Inhibition of protein synthesis leads to loss of Smad2 from the cell nucleus, whereas inhibition of the proteasome results in continued accumulation of Smad2 in the cell nucleus. (D and E) Expression of TGF-β target genes by qRT-PCR as a function of duration of treatment in the GFP-Smad4 C2C12 cell line treated with TGF-β1 and either cycloheximide (chx) (D) or MG132 (E). Inhibiting the proteasome has little effect on dynamics, whereas inhibiting protein synthesis converts the transient response to a sustained one. The increase in Smad7 at long treatment with MG132 may be a homeostatic response to high levels of Smad2 that accumulate after prolonged MG132 treatment (see C).

We used the RFP-Smad2 cell line to determine the effects of cycloheximide on Smad2 dynamics. In agreement with previous results (17), the amount of Smad2 in the nucleus decreased more in cells treated with TGF-β1 and cycloheximide than in cells treated with TGF-β1 alone (Fig. 4C). When cells were treated with cycloheximide and TGF-β1, RFP-Smad2 accumulated normally in the cell nucleus but began to decline after 6 h and returned to baseline levels after 16 h. Thus, the effects on Smad2 nuclear localization were confined to times relatively late in the response, and both Smad2 and Smad4 show sustained nuclear accumulation in the presence of cycloheximide (Fig. 4 B and C).

Following individual cells shows a more nuanced picture of the Smad4 response when protein synthesis is inhibited. The initial response of the C2C12 GFP-Smad4 cell line to treatment with TGF-β1 and cycloheximide was uniform, but later the population resolved into high- and low-response groups (Fig. S3 A and B). Measuring Smad2/3 by immunofluorescence in GFP-Smad4 cells showed a correlation between their levels at both early and late times (Fig. S3 C and D), but only Smad4 was binary at late times, whereas Smad2/3 was always graded (compare Fig. S3 B and E). The bimodal distribution of Smad4 and the correlation between Smad2 and 4 at long times suggests that without new protein synthesis cells make a binary decision whether to maintain activation of Smad4 based on the level of activated Smad2/3. A possible mechanism for the binary Smad4 response is presented in the mathematical modeling sections in the SI Text.

Transcriptional Dynamics Correlate with Smad4 but Not Smad2 Nuclear Localization.

As noted above, inhibiting protein synthesis decreased the duration of Smad2 nuclear accumulation but prolonged Smad4 nuclear accumulation (Fig. 4 B and C). Thus, the differing effects on Smad2 and Smad4 provided another opportunity to test whether Smad2 or Smad4 controls transcriptional dynamics. Consistent with the retention of Smad4 in the nucleus, target gene expression was prolonged, peaking at 4–6 h after stimulation compared with 2 h without cycloheximide, and target genes still were significantly up-regulated 12 h after stimulation (Fig. 4D; compare with Fig. 2B). The effects of cycloheximide treatment support our contention that signaling is limited by Smad4 rather than Smad2 nuclear localization, because cycloheximide shortens the duration of Smad2 nuclear localization, lengthens the duration of Smad4 nuclear localization, and lengthens the time for TGF-β–dependent transcription.

Because ubiquitin-dependent degradation has been shown to modulate the extent of Smad2 retention in the nucleus (13), we investigated the effects of proteasome inhibition on Smad2 and Smad4 nuclear accumulation and TGF-β–dependent transcription. Inhibition of the proteasome with MG132 caused the amount of Smad2 in the cell nucleus to increase throughout the period of observation (15 h) (Fig. 4C). However, it had very little effect on the dynamics of Smad4 nuclear localization (Fig. 4B) and did not prolong the time of expression of TGF-β target genes. (Fig. 4E). The dynamics of transcription in cells treated with TGF-β1 and either cycloheximide or MG132 were very similar in the parental and Smad-overexpressing lines (Fig. S3 F and G). Thus, taken together, the results reported in this section suggest that Smad4 adaptation involves new protein synthesis but not protein degradation and that transcriptional dynamics are affected by perturbations to Smad4 but not Smad2 dynamics.

Either BMP or Activin Signaling Induces Transient Bursts of Nuclear Smad4 in the Early Frog Embryo.

To examine TGF-β dynamics in vivo, we turned to the Xenopus embryo. During development, TGF-β ligands act as morphogens and play crucial roles in mesoderm induction and dorsal ventral patterning (20–22). We chose to study the animal pole (future ectoderm) of the embryo, because it is patterned by endogenous BMP signaling (23, 24), is responsive to exogenous activin/nodal signaling, and is a convenient tissue for imaging. We began by studying BMP signaling, because BMP signaling is the endogenous signal in this tissue; the animal cap displays no activin/nodal activity.

To study BMP-dependent R-Smad dynamics in the animal cap, we injected mRNA encoding GFP-Smad1 and mCherry-H2B fusion proteins into the embryo at the two-cell stage and dissected the animal pole of the embryo for imaging at the late blastula stage (stage 9). In agreement with a previous study (25), we found that GFP-Smad1 was predominantly nuclear in this tissue. Surprisingly, GFP-Smad1 remained predominantly nuclear even when BMP signaling was inhibited by coexpression of its inhibitors Cerberus, Chordin, or DNAlk3 (Fig. 5A and Fig. S4). Strangely, all the BMP inhibitors we tested reduced fluorescence from both GFP-Smad1 and mCherry-H2B; however, the ratio of nuclear to cytoplasmic GFP-Smad1 was reduced only modestly (Fig. 5B), suggesting that pathway activity has little effect on the localization of Smad1. Despite the presence of Smad1 in the nucleus, BMP-mediated transcription of an XVent2-luc reporter was repressed strongly by all of the BMP inhibitors tested (Fig. S4A). Immunofluorescence for the endogenous protein confirmed nuclear localization of Smad1/5/8 both in untreated animal caps and in the presence of Cerberus (Fig. S4B). Thus, nuclear localization of Smad1 is not a reliable indicator of pathway activity in this system, and we turned to assay the status of C-terminally phosphorylated Smad1/5/8. C-terminally phosphorylated Smad1/5/8 was distributed homogeneously throughout animal cap explants (Fig. 5A). The staining was specific, because it was inhibited by Cerberus injection (Fig. 5A), and injection of the dominant-negative type I BMP receptor Alk3 (DNAlk3) into one cell at the two-cell stage inhibited pSmad1/5/8 specifically in the half-embryo receiving the injection (Fig. S4C).

Fig. 5.

Smad1 is homogeneously activated, whereas Smad4 is heterogeneous and exhibits brief bursts of nuclear localization in Xenopus animal cap explants. (A) Smad1 phosphorylation but not Smad1 localization responds to inhibition of BMP signaling. Images are of animal pole tissue from embryos injected at the two-cell stage with 100 pg GFP-Smad1 mRNA and 100 pg mCherry-H2B mRNA in each cell. At stage 9 embryos were fixed and stained by immunofluorescence for pSmad1/5/8. (B) Quantification of average nuclear-to-cytoplasmic ratios from the images in A. GFP-Smad1 localization does not depend on BMP signaling. (C) Smad4 localization is heterogeneous. Images of a stage-9 animal cap explant from an embryo injected at the two-cell stage with 50 pg per cell of Venus-Smad4 mRNA and 100 pg per cell of mCherry-H2B mRNA. (D) Zoomed-in image showing the pulses in single cells. (Upper) A cell that pulses without dividing. (Lower) A cell that divides during the pulse.

The homogeneous presence of pSmad1 in all cells suggested relatively simple dynamics. To see whether Smad4 localization was similarly homogeneous, we injected mRNAs encoding Venus-Smad4 and mCherry-H2B fusion proteins into both cells of the animal pole of the embryo at the two-cell stage. In blastula-stage animal cap explants, Smad4 showed a striking heterogeneous pattern, strongly accumulated in some nuclei but absent from others (Fig. 5C). Heterogeneity was not an imaging artifact, because the nuclear marker mCherry-H2B was clearly detectable in every cell. Time-lapse imaging revealed that the heterogeneity resulted from transient bursts of Smad4 nuclear localization. In nearly every cell, Smad4 entered the nucleus in pulses lasting ∼30 min each (Fig. 5D and Movie S4). The bursts were uncorrelated spatially with no apparent pattern in timing. Pulses were not periodic, because the time between pulses in the same cell was variable.

Because animal pole cells divide approximately every 30 min during the blastula stage, we examined the time-lapse movies to determine whether cells divided during the pulses of nuclear Smad4. Indeed, cell divisions during the pulses were common; however, examples of pulses both with and without cell divisions could be found readily (Fig. 5D). When the cells divided during the pulse, the daughter cells typically showed enhanced nuclear Smad4 after division. Thus, the pulses are not particular to any cell-cycle phase. However, we cannot exclude a role for the cell cycle in the initiation of pulses.

We hypothesized that the bursts of nuclear Smad4 were dependent on the endogenous BMP signaling in the animal pole. Indeed, coinjection of DNAlk3 led to a drastic reduction in the frequency of bursts of nuclear Smad4 (Fig. 6 A–C and Movie S5), suggesting that BMP signaling is responsible for the pulses of nuclear Smad4. Quantifying nuclear Smad4 as a function of time in single cells, we found that inhibiting BMP signaling led to a large reduction in the number of bursts without affecting the amplitude or duration of the remaining ones (Fig. 6B and Fig. S5A). The experiments described above showing uniform activation of Smad1/5/8 across the animal cap suggest that the bursting dynamics are limited to Smad4. Quantitatively, we observed no correlation between levels of pSmad1/5/8 and nuclear localization of Smad4 (Fig. S5 B and C).

Fig. 6.

Pulses of nuclear Smad4 can be induced by either BMP or activin/nodal signaling in Xenopus animal cap explants. (A) Pulses are dependent on endogenous BMP signaling. Image of a stage-9 animal cap explant in which 50 pg Venus Smad4 and 100 pg mCherry-H2B mRNA were coinjected with 1 ng of the BMP inhibitor DNAlk3 mRNA. The animal cap explant shown is from a sibling embryo of the one used to make the explants shown in Fig. 5 C and D with injections and imaging performed at the same times. (B) Quantification of number of pulses and pulse height and duration. (C) Quantification of single-cell traces from the time-lapse movies. (D) Response of Smad4 to activin treatment. Embryos were injected in the animal pole with Venus-Smad4, mCherry-H2B, and DNAlk3 mRNAs to inhibit BMP signaling. Animal cap explants were dissected at stage 9 and treated with activin. Pictures are shown before (Left) or 45 m after (Right) activin treatment. (E) Quantification of number of bursts in animal cap explants injected with Venus-Smad4 and DNAlk3 and either left untreated or treated with activin (10 ng/mL).

Because our mammalian cell-culture studies focused on the TGF-β/activin/nodal branch of the pathway, we sought to address whether signaling through this branch could induce similar pulses of activity in Xenopus embryos. We prevented the BMP-dependent bursts of nuclear Smad4 by injecting DNAlk3 and then treated the animal cap explants with 10 ng/mL activin. Bursts similar in frequency to those observed in response to endogenous BMP signaling were observed (Fig. 6 D and E and Movie S6). Immunofluorescent analysis revealed that Smad2/3 accumulated homogeneously in every cell in the animal cap treated with activin (Fig. S5D). Similarly, we found that cell dissociation abrogated the BMP-dependent bursts of nuclear Smad4 and that bursts could be restored by treating the dissociated cells with activin (Fig. S5E). Taken together, our experiments in the early Xenopus embryo show that Smad4, but not the R-Smads, shows repeated pulses of transient activity in response to either BMP or activin signaling and suggest that the temporal duration of signaling is limited by Smad4 as in mammalian cell culture.

Our studies in cell culture showed transient induction of Smad4, whereas those in Xenopus embryos showed repeated bursting. To understand the connection between these behaviors, we created a simple mathematical model (SI Text, SI Materials and Methods and Fig. S6). The result show that transient induction and bursting can coexist readily in the same model and that additional feedback mechanisms could be responsible for creating the bursting dynamics in the Xenopus embryo.

Discussion

We have shown that the previously accepted model in which activated, nuclear-localized R-Smads are synonymous with pathway activation need to be refined. Under continuous ligand stimulation, R-Smads remain active, but transcription returns to baseline levels. This basic discrepancy was established using both traditional assays in unmodified cells and live-cell imaging in cells expressing an RFP-Smad2 reporter. In our revision, the R-Smads respond in a uniform, graded manner to the level of agonist. Smad4 enters the nucleus with the R-Smads but then mediates the temporal adaptation of the pathway that terminates transcription. Adaptation requires protein synthesis, but its dynamics do not depend significantly on the agonist level. The heterodimer between the two factors then imposes both digital time control and graded activity on the pathway output.

During the embryonic development, TGF-β ligands act as morphogens with the cell fate outcome dictated not only by the concentration of ligand but also by the timing and duration of stimulation (21, 22). It will be important to reexamine the temporal aspects of TGF-β–mediated fate specification in light of our finding that TGF-β transcription likely is bursting during development. Future work will seek to connect bursts of Smad4 to fate decisions and to determine whether the number of Smad4 bursts plays a role in specifying cell fate. In addition to its role in development, TGF-β is dysregulated in a variety of cancers (26), and the timing of signaling also plays an important role. For example, TGF-β is transiently up-regulated in metastasizing breast cancer cells (27). It seems likely that dysregulation of the timing of signaling will play a role in disease.

Because Smad4 is shared between the two branches of the TGF-β pathway, our model raises further questions about branch interactions (28). If temporal control of the pathway is mediated by Smad4, this control naturally would induce correlations between the timing of TGF-β and BMP-dependent transcription, with the relative amplitudes determined by the activation of branch-specific R-Smads. It will be of interest to determine if adaptation induced by TGF-β will truncate the response to BMP presented a short time later, and conversely. Of equal interest is the response to temporal pulses of agonist. Is there a refractory period defined by the adaptation?

Our model postulates that the duration of Smad4 nuclear localization is independent of that for R-Smads and coincides with the timing of transcription. Although there is abundant evidence that nuclear-localized, activated R-Smads are necessary for transcription, it is worth reviewing the experiments leading to the previous model that temporal control is regulated primarily through the R-Smads. Because Smad4 is not phosphorylated by the receptors or inactivated by linker phosphorylation, it appeared as the passive cofactor to the R-Smads. The nuclear import of Smad4 in response to agonist requires the R-Smads (8), and inhibiting the pathway at the receptor level with SB431542 or Noggin causes export of both R-Smads and Smad4 (17). Also the timing of nuclear Smad4 and transcription of pathway targets was examined in several experiments (17, 29) but always in the presence of cycloheximide (SI Text) and thus do not contradict our conclusions. In both Xenopus animal caps and sparsely plated HaCaT cells we found Smad4 localization is intermittent but clearly is signaling dependent. Thus, population assays for nuclear Smad4 (29) should show that it varies with the activated nuclear R-Smads. Discrepancies among other experiments (SI Text) involving the duration of the transcriptional response in HaCaT cells may be explained by cell density, because in dense cultures Smad4 localization follows the same dynamics as in C2C12 cells (Fig. S2).

Experiments with transgenic mice challenge the notion that Smad4 localization is mechanistically responsible for the termination of transcription. Mice homozygous for a constitutively nuclear Smad4 appear normal (30). Presumably the duration of TGF-β signaling must be close to that in wild-type embryos to permit normal development. Thus, we favor the model that Smad4 nuclear export is downstream of pathway inactivation but is not causal to termination of transcription. Indeed the fractional changes in nuclear Smad4 that we observe would be inadequate to explain the differences in transcriptional output in a simple physical–chemical model of Smad heterodimer formation. Multiple biochemical events may be collectively responsible for the adaptation we found in the transcriptional response of the TGF-β pathway to a step stimulus. One possibility is SnoN repression of Smad2/3 transcriptional targets (31). This factor is degraded by TGF-β signaling but also is a transcriptional target of TGF-β, thus explaining the effect we see with cycloheximide. Second, Smad4 monoubiquitination disrupts the complex with R-Smads and inhibits pathway activity (32, 33). Similarly, Smad4 sumoylation also can inhibit the pathway (34, 35). The fact that the adaption is reflected only in Smad4 excludes negative feedback mechanisms, such as induction of the inhibitory Smad7 or degradation of activated R-Smads, that primarily target R-Smads or receptors.

There is a large literature on time-dependent signaling (reviewed in ref. 36) motivated by the realization that the same pathway can link different stimuli to distinct transcriptional targets. A logical source for the information governing specificity is the time course of the signal or the temporal activity of pathway intermediates. A system that merits comparison with the present work is NFκB (2, 3). The single-cell data (3) following nuclear localization of NFκB in response to a step in TNF-α show a largely digital and adaptive response (but see ref. 37). The intensity of the step defines the fraction of cells that respond, and the nuclear-localized NFκB returns to near base line in 90 min. In contrast, although we observe adaptation in response to ligand stimulation, we do not find a binary response for intermediate ligand levels; rather, all cells respond. An earlier study on NFκB dynamics observed less adaptation and more pronounced oscillations (38), and it may be that the microfluidic chambers in the more recent study allow cell-secreted factors or ligand depletion to modulate the response (3).

Signaling pathways commonly exhibit multiple levels of negative feedback operating on the receptors, factor localization by posttranscriptional modifications, and transcription of inhibitory factors. All potentially can generate bursts of activity, as in the p53 system (4, 39). In analogy to our observations in Xenopus experiments, where the bursts changed in number but not shape in response to pathway inhibition, the p53 bursts are present in normally cycling cells and become more frequent with DNA damage (4). Adaptation is another consequence of pathway negative feedback and can depend on secondary interactions among the principal pathway components. A notable example is in PC-12 cells, where the Erk response is adaptive in response to EGF but sustained in response to NGF (40).

As our experiments in mammalian culture cells and Xenopus embryos reveal, the behavior of Smad4 can be either adaptive or pulsatile, depending on the context. Our mathematical modeling suggests that these behaviors are closely connected and that the same model can encompass both. Additional feedbacks or external inputs such as those from cell-to-cell contacts, cross-talk with other pathways, or the cell cycle could serve to generate pulsing or oscillatory behaviors from a core adaptive circuit. Distinguishing these possibilities will require further dissection of the mechanisms controlling temporal control of Smad4 nuclear localization and transcription.

Materials and Methods

Constructs.

All constructs for cell-culture studies were cloned into ePiggyBac vectors, and all constructs for Xenopus experiments were cloned in the pCS2+ vector. Details can be found in SI Materials and Methods.

Cell Culture, Transfections, and Selection of Stable Lines.

Both C2C12 and HaCaT cells were maintained in DMEM containing 10% (vol/vol) FBS. All transfections were performed with Lipofectamine 2000 (Invitrogen) according to the manufacturer’s instructions. Following the transfections, cells were selected with blasticidin (10 μg/mL), puromycin (4 μg/mL), or Zeocin (500 μg/mL). Clonal lines were established by plating as single cells and selecting clones manually.

Reagents.

Cycloheximide was purchased from Sigma-Aldrich and was used at a concentration of 50 μg/mL to inhibit protein synthesis (29, 41). MG132 was purchased from Sigma-Aldrich and used at a concentration of 10 μM to inhibit proteasomal degradation (42). SB431542 was purchased from Tocris Bioscience and used at a concentration of 10 μM to inhibit Smad2/3 phosphorylation (12). TGF-β1 and activin A were purchased from R&D Systems and used at the concentrations indicated in the text.

Xenopus Embryo Manipulation.

Eggs were collected, fertilized in vitro, dejellied, and injected with mRNA using standard techniques (SI Materials and Methods). For imaging of animal caps, animal caps were dissected at stage 9 and grown on fibronectin-coated glass in a low-calcium medium that does not dissociate cells but prevents rounding of animal caps, which occurs rapidly in normal explant medium.

Live Cell Imaging and Image Analysis.

C2C12 and HaCaT cells were plated in glass-bottomed dishes (MatTek) at least 1 d before imaging. Imaging was performed in L15 medium containing 10% (vol/vol) FBS using an IX71 Olympus inverted microscope with a 20×, 0.75 Na lens. Xenopus animal caps were placed on glass-bottomed dishes that had been coated with fibronectin (20 μg/mL) (Sigma) and imaged using an inverted Zeiss confocal microscope (LSM 510) with a 20×, 0.7 Na lens and collecting a Z stack containing four to six slices every 3–5 min. Image analysis was performed with custom software written in MATLAB as described in SI Materials and Methods.

Immunofluorescence.

Immunofluorescence on culture cells was performed using standard techniques. Immunofluorescence in Xenopus animal caps was performed according to the procedure described by Larabell et al. (43). Details and antibodies can be found in SI Materials and Methods.

Western Blotting and qRT-PCR.

Western blotting and qRT-PCR were performed using standard methods. Antibodies and primers are listed in SI Materials and Methods.

Luciferase Assays.

At the two- or four-cell stage, 15 pg of XVent-luc reporter plasmid was injected into each cell. At stage 11, four embryos were lysed together in 50 μL of passive lysis buffer (Promega), and the luciferase assay was performed using the Promega luciferase assay system according to the manufacturer’s instructions. CAGA12 and 3TP experiments in C2C12 cells were performed using the dual luciferase assay kit (Promega). Details can be found in SI Materials and Methods.