Negative feedback in the bone morphogenetic 4 (BMP4) synexpression group governs its dynamic signaling range and canalizes development

Malte Paulsena,1, Stefan Legewieb,c,1, Roland Eilsb,d, Emil Karaulanova, and Christof Niehrsa,c,2

aDivision of Molecular Embryology and bDepartment of Theoretical Bioinformatics, Deutsches Krebsforschungszentrum, D-69120 Heidelberg, Germany; dBioquant and Institute of Pharmacy and Molecular Biotechnology, Ruprecht-Karl-Universität Heidelberg, D-69120 Heidelberg, Germany; and cInstitute of Molecular Biology, D-55128 Mainz, Germany

Edited by Igor B. Dawid, National Institute of Child Health and Human, Bethesda, MD, and approved May 5, 2011 (received for review January 7, 2011) What makes embryogenesis a robust and canalized process is an may be enhanced in the presence of BAMBI negative feedback, important question in developmental biology. A bone morphoge- and that this may be essential for embryonic development. netic protein (BMP) morphogen gradient plays a key role in em- We provide evidence that synexpressed feedback inhibitors bryonic development, and we are beginning to understand how expand the dynamic BMP4 signaling range, and show the role of the self-regulating properties of its signaling circuitry ensure robust negative feedback in reducing phenotypic variability in verte- embryonic patterning. An unexplored question is why the BMP brate embryos. signaling circuit is organized as a modular synexpression group, with a prevalence of feedback inhibitors. Here, we provide evi- Results dence from direct experimentation and mathematical modeling Negative Feedback Expands the Dynamic BMP4 Signaling Range. We that the synexpressed feedback inhibitors BAMBI, SMAD6, and first addressed the role of BAMBI negative feedback on the dy- SMAD7 (i) expand the dynamic BMP signaling range essential for namic signaling range in HEK293 cells, which provide a simplified proper embryonic patterning and (ii) reduce interindividual pheno- model mimicking essential aspects of BMP4 signaling in Xenopus typic and molecular variability in Xenopus embryos. Thereby, neg- embryos: (i) HEK293 cells up-regulated BAMBI, SMAD6, and ative feedback linearizes signaling responses and confers robust to a lower degree SMAD7 expression in response to BMP4 (Fig. patterning, thus promoting canalized development. The presence 1A). This observation indicates that these mammalian cells con- of negative feedback inhibitors in other growth factor synexpres- tain all of the basic components of the BMP signaling cascade and sion groups suggests that these properties may constitute a gen- that the synexpression negative feedback loop is in place as it is in eral principle. embryos. By contrast, SMURF1, a nonsynexpressed inhibitor of BMP signaling (23), was expressed but not BMP4 inducible. (ii) systems biology | tail | eye | canalization HEK293 cells treated with recombinant human BMP4 activated reporter expression driven by the ventx2 (Xvent2) BMP- raded bone morphogenetic protein (BMP) signaling is an responsive element (24), which is active in Xenopus embryos, Gevolutionary conserved mechanism that regulates early em- indicating that these cells harbor transcriptional cofactors com- bryonic dorsoventral (DV) axial patterning (1). Pioneering work parable to those of embryos (Fig. 1B). in Drosophila and Xenopus embryos has shown that the BMP To assess BMP signaling in real-time, we generated stable morphogen gradient is self-regulating (2–7). These system prop- reporter HEK293 cell lines containing a multimerized ventx2 erties of BMP signaling canalize development and explain scal- BMP-responsive element (24), driving expression of a short-lived ing of patterning in embryos of different sizes. One unexplained luciferase enzyme (25). HEK293 reporter cells subjected to feature of the BMP signaling circuit is its organization as a syn- continuous BMP4 treatment for 13 h showed a rapid increase in expression group—i.e., genetic modules made up of , which reporter activity for over 8 h and then reached a plateau (Fig. 1B function in the same molecular process and display tight spatio- and SI Appendix, Fig. S2A). Similar profiles were obtained with temporal coexpression (8, 9). A particularly intriguing feature an independent cell clone (SI Appendix, Fig. S2B). Importantly, fi of the BMP4, FGF8, and Delta-Notch synexpression groups is this pro le was induced in a dose-dependent fashion over the prevalence of feedback inhibitors (8–15). The BMP4 synex- a nearly 100-fold BMP4 concentration range. We then tested the – pression group (SI Appendix, Fig. S1) (16) in vertebrates includes effect of negative feedback inhibition on the kinetics and dose as inhibitors the pseudo-BMP receptor BAMBI, the inhibitory response of BMP4 signaling. Knockdown of BAMBI by siRNA, SMAD6 and SMAD7, and Crossveinless2 (6, 10–13). During which did not adversely affect cell viability, reduced response vertebrate development, these genes are tightly coexpressed with differences at BMP4 concentrations between 5 and 50 ng/mL, the BMP4 growth factor, because they are themselves transcrip- such that they were hardly distinguishable (Fig. 1C and SI Ap- pendix, Fig. S2 A and B). Thus, down-regulation of BAMBI led tional targets of BMP signaling (6, 14, 16). What is the physio- to a ∼10-fold reduction in dynamic signaling range. Knockdown logical significance of this modular network organization and of of either SMAD6 or SMAD7 led to cell death upon continuous negative feedback? Signal transduction circuits are nonlinear systems with limited information transmission capacity, because their response satu- Author contributions: M.P., S.L., E.K., and C.N. designed research; M.P. and S.L. performed rates at high signal input. Negative feedbacks are known to lin- research; M.P., S.L., E.K., and C.N. analyzed data; and M.P., S.L., R.E., E.K., and C.N. wrote earize biochemical responses by sacrificing gain, thus allowing the paper. the network to respond gradually over a broad stimulus range The authors declare no conflict of interest. (17–21) (in analogy, negative feedback helps to convert the dy- This article is a PNAS Direct Submission. namic range of a telephone loudspeaker into that of a HiFi 1M.P. and S.L. contributed equally to this work. fi system). The BMP4 morphogen speci es multiple cell fates over 2To whom correspondence should be addressed. E-mail: [email protected]. an at least 30-fold concentration range (22). We therefore hy- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. pothesized that the dynamic range of BMP4 signaling responses 1073/pnas.1100179108/-/DCSupplemental.

10202–10207 | PNAS | June 21, 2011 | vol. 108 | no. 25 www.pnas.org/cgi/doi/10.1073/pnas.1100179108 Downloaded by guest on September 25, 2021 siControl factor acts as a morphogen (1). We therefore validated our find- A BAMBI D 4 10 SMAD6 siSMURF1 ings in animal cap cells isolated from Xenopus embryos. Dissoci- SMAD7 siBAMBI 8 SMURF1 3 ated animal cap cells respond to BMP4 with induction of the 6 immediate early target genes ventx1 (Xvent1) (26) and ventx3 2 (vex1) (27). Knockdown of bambi using a published antisense 4 morpholino oligonucleotide (MO) (4) did not significantly change 2 1 Relative expression target gene expression at low BMP4 levels, but increased gene 0 02468101214 Relative Luciferase0 Activity expression at higher BMP4 concentrations and led to premature time (h) 0 0.5 1 2.5 5 10 25 50 saturation at the highest BMP4 dose (Fig. 2 A and B). In contrast, BMP4 (ng/ml) using established MOs (28) to knockdown chordin and noggin, B siControl E 1600 BMP4 [ng/ml] encoding nonsynexpressed BMP inhibitors, increased the overall 0.5 1 siControl siBAMBI 5 25 signal without signal saturation (Fig. 2 C and D). We conclude that 1200 50 BMP4 BMP4 bambi negative feedback enhances the dynamic signaling range of pSMAD1 BMP4 signaling in Xenopus embryo animal caps. RLU 800 tSMAD1 siControl siSmurf1 Mathematical Modeling and Design Principles of the BMP4 400 BMP4 BMP4 pSMAD1 Synexpression Group. To gain insight into the system properties 0 tSMAD1 of the BMP4 synexpression group, we developed a mathematical 024681012 time (h) model of BMP signaling using ordinary differential equations

siBAMBI (Fig. 3A and SI Appendix, SI Methods). The model was calibrated C F by fitting to reporter gene assays in HEK293 cells. The best-fit 1600 BMP4 [ng/ml] 8 siControl 0.5 1 siBAMBI model reproduced the observed signaling compression at high 5 25 siSmurf1 1200 50 6 BMP4 doses in the absence of BAMBI (Fig. 3 B–D): The dynamic fi

RLU signaling range de ned as the fold BMP4 concentration change 800 4 required to shift reporter gene expression from 10% to 90% of 400 2 maximum was eightfold larger in the control model (160×) than in the BAMBI knockdown model (20×; Fig. 3D). Removing feed-

0 pSMAD1/5/8 (fold change) 0 024681012 0 0.5 1 5 25 50 back induction of BAMBI in the control model (Fig. 3D, green time (h) BMP4 (ng/ml) line) shifts the upper part of the dose–response curve to the left Fig. 1. BAMBI expands the dynamic BMP4 signaling range in HEK293 cells. (indicated by arrow in Fig. 3D), while leaving the lower part un- (A) Time course of inhibitor expression of HEK293 cells continuously stimu- affected, resulting in an overall steeper dose–response curve. lated with BMP4. Error bars indicate SD (n = 3); relative expression at t = 0 set Thus, modeling suggests that BAMBI feedback induction at high to 1. (B and C) Real-time luciferase reporter assays of a stable HEK293 BMP- BMP doses is essential for expanding the signaling range, whereas responsive luciferase reporter cell line transfected with control or siBAMBI constitutive BAMBI expression fares no better than control. To and induced with increasing amounts of BMP4. RLU, relative light units. The assay-to-assay variability was <100 RLU (n = 3). (D) Endpoint luciferase assay corroborate the important role of feedback regulation, a feed- of HEK293 cells transiently transfected with BMP-responsive luciferase re- back-less model, where BAMBI expression is unaffected by BMP porter and indicated siRNAs. Cells were continuously treated for 12 h with signaling, was fitted to the data (SI Appendix, SI Discussion and BMP4. Error bars indicate SD (n =4).(E) Western blot of endogenous phos- Figs. S3–S5). The feedback-less model shows a significantly worse phorylated and total SMAD1/5/8 in control, siBAMBI,orsiSMURF1 transfected fit to the experimental data than the feedback model with a dra- HEK293 cells induced for 14 h with BMP4 (0.5, 1, 5, 25, and 50 ng/ml). (F) Quantification of Western blots shown in E; relative phosphorylated SMAD1/ 5/8 level was normalized to total SMAD1/5/8 and set to 1 in nonstimulated samples. For siRNA knockdown efficiencies, see SI Appendix,Fig.S2F.

BMP4 treatment, and we were unable to record data. To further rule out that effects were biased by locus-specific integration of the BMP reporter transgene, we used transient reporter trans- fections in HEK293 cells. Once again, BAMBI siRNA induced a ∼10-fold response compression at high BMP4 doses, while having only a minor effect at low BMP doses (Fig. 1D). In contrast to BAMBI depletion, siRNA knockdown of the nonfeedback BMP signaling inhibitor SMURF1 increased the overall signal without leading to signal saturation (Fig. 1D). We conclude that BAMBI is required to expand the dynamic BMP signaling range in HEK293 cells. Because BAMBI acts at the BMP receptor level, its effects should manifest also in SMAD1/5/8 phosphorylation. Indeed, in BMP4 titrations, SMAD1/5/8 phosphorylation plateaued at lower BMP4 doses in BAMBI siRNA than in control cells (Fig. 1 E and F; also, in HaCaT cell line, see SI Appendix, Fig. S2 C and D). In contrast, SMURF1 siRNA even enhanced the response at high Fig. 2. Bambi expands the dynamic BMP4 signaling range in Xenopus em- – fi BMP4 doses (Fig. 1 E and F), which corroborates that negative bryonic cells. (A D) qPCR quanti cation of ventx1 and ventx3 expression in BMP4-treated dissociated animal cap explants injected with the indicated feedback by BAMBI expands the dynamic signaling range of the antisense morpholinos (MO); error bars indicate SD (n = 3). The following

BMP circuit in HEK293 cells. doses of BMP4 were applied: 0.03, 0.1, 0.3, 0.9, and 2.7 μg/mL. Log scale BIOLOGY

The discrimination between different BMP4 doses is of par- interrupted as indicated to display basal expression in mock-treated cells. DEVELOPMENTAL ticular importance in early Xenopus embryos, where the growth Expression in control MO samples treated with 2.7 μg/mL BMP4 set to 1.

Paulsen et al. PNAS | June 21, 2011 | vol. 108 | no. 25 | 10203 Downloaded by guest on September 25, 2021 A B model fit vs. data D model fit vs. data F model analysis G experimental data siControl Dynamic signalling range BAMBI effect on dynamic 80 siControl Control 1 siBAMBI signaling range siBAMBI exp. data 20x - / + Control none expansion 60 BMP4 model 160x

4 40 exp. datapp mean model 2 fluorescence 1 20 Number of cells 0 Reporter expression (AU) 0 0 −2 01Time (h) 2 80 siControl 0.5 25 Control siBbS6S7 −4 1 50 siBAMBI Synchrony feedback-reporter induction feedback-reporter Synchrony 60 Reporter expression (AU) mean const. BAMBI 5 [ng/ml BMP4] −2 −1 0 1 2 fluorescence C 0 log10(feedback strength) siBAMBI -4 -2 log2 [BMP4] 6 8 40 1 model analysis exp. data H

Number of cells 20 model E model prediction vs. data siControl siBAMBI 1 siBAMBI/S6/S7 exp. data 0 model siControl 80 siSMURF1 siControl siSmurf1 low bambi 60 mean Reporter expression (AU) Control overexpr. 0 fb+ fluorescence 01Time (h) 2 40

low bambi Reporter expression (AU) -

overexpr.fb Number of cells 20 0 -4 -2 log2 [BMP4] 4 6 simulations of Frequency 0 0210 00 200 400 Reporter gene expression (AU) d2eGFP (FIU)

Fig. 3. Mathematical model of the BMP4 synexpression circuit. (A) Considered BMP target genes are BAMBI and SMAD6/7 feedback regulators and the destabilized luciferase reporter (R1 and R2: BMP receptor 1 and 2, respectively; S and pS: unphosphorylated and phosphorylated SMAD1/5/8, respectively; S4: SMAD4; see SI Appendix for details). (B and C) The best-fit model reproduces the dose-dependent behavior of BMP-induced reporter gene expression in siControl (B) and siBAMBI (C) cells. (D) Dose–response measured 12 h after stimulation (data points) and best-fit model simulations (lines). The dynamic signaling range defined as the fold increase in the BMP4 concentration required to shift reporter expression from 10% to 90% of maximum is indicated (Upper). Elimination of SMAD-dependent BAMBI induction in the feedback model (green line) specifically affects signaling at high BMP doses (indicated by arrow). (E) Dose–response measured 12 h after stimulation (data points) and best-fit model simulations (lines) in cells weakly overexpressing Xenopus bambi. The data agrees well with predictions of the best-fit feedback model (solid red line), but disagrees with predictions of the best-fit feedback-less model described in SI Appendix, Fig. S3 (dashed red line; SI Appendix, SI Discussion and Figs. S4 and S5). (F) Monte Carlo analysis of the dynamic signaling range. The effect of BAMBI depletion on the dynamic signaling range was quantified for randomly sampled parameter sets by dividing the dynamic range of the full model by that of the corresponding BAMBI-depleted model. Along the x axis, the simulations were sorted according to the strength of BAMBI feedback (i.e., the effect of BAMBI on its own induction; SI Appendix, SI Discussion). Along the y axis, the simulations were sorted based on the synchrony of reporter gene induction and BAMBI feedback: Synchrony values close to zero imply that reporter induction and feedback regulation occur in the same BMP concentration range. Increasing positive (negative) values imply that reporter gene induction occurs at lower (higher) BMP4 concentrations than feedback regulation (SI Appendix, SI Discussion). Note that parameters yielding maximal dynamic signaling range expansion (yellow to brown color) are found in the region of high feedback strength and high synchrony. (G) BAMBI and SMAD6/7 negative feedback reduce natural variability in gene expression in HEK293 cells. Stable HEK293 BMP-responsive desta- bilized eGFP (d2eGFP) reporter cells were transfected with siRNAs as indicated and treated for 12 h with 20 ng/mL BMP4. Destabilized eGFP fluorescence of 8,500 cells in G1 phase was measured by FACS analysis. Dashed lines indicate mean d2eGFP fluorescence of the corresponding population. FIU, fluorescence intensity units. (H) BAMBI and SMAD6/7 synergistically suppress cell-to-cell variability in the best-fit model. Reporter expression histograms were simulated by assuming correlated, log-normal–distributed fluctuations in all transcription and translation rates (extrinsic noise) in a deterministic model (SI Appendix, SI Discussion). The SMURF1 knockdown (Inset) was simulated by assuming that SMURF1 regulates the stability of SMAD , and thereby controls maximal SMAD activation. BMP receptor binding affinity was decreased relative to the best-fit model to account for BMP batch variability (SI Appendix, SI Discussion).

matically decreased and less robust effect of BAMBI on the dy- levels. Increasing basal BAMBI expression attenuates this feed- namic signaling range. back term and thereby compresses the signaling range, while To confirm our numerical results, we derived a simplified having no effect on the signaling range of a feedback-less system version of our model and obtained an analytical approximation (kfeedback = 0); this leads to the prediction that feedback and for BMP-dependent reporter expression under steady-state feedback-less model variants can be discriminated by experi- conditions mentally overexpressing BAMBI. In line, systematic numerical analyses also predict that BAMBI overexpression compresses the ½BMPn ½ ≈ · pffiffiffiffiffiffiffiffiffiffiffiffiffi Á signaling range in the full feedback model, but not in the feed- Reporter kmax n n ½BMP þ kfeedback· ½BMP þ kbasal back-less variant (SI Appendix, SI Discussion and Figs. S4 and S5). To validate this prediction, experimental dose–response data This sigmoidal equation contains constants relating to transcrip- were gathered in HEK293T cells moderately overexpressing tional induction of the reporter (kmax, n), the strength of BAMBI bambi from a constitutive promoter (Fig. 3E). In line with pre- feedback induction (kfeedback) and basal BAMBI expression levels dictions of the best-fit feedback model, bambi overexpression (kbasal). As expected from our numerical studies, feedback ex- inhibited signaling at low BMP doses, whereas it had only a mi- pands the dynamic range by introducing a slowly growing square nor impact at high BMP levels, thus resulting in an overall root term that selectively reduces signaling at intermediate BMP compression of the signaling range. We propose that BAMBI

10204 | www.pnas.org/cgi/doi/10.1073/pnas.1100179108 Paulsen et al. Downloaded by guest on September 25, 2021 overexpression compresses the signaling range by compromising motes robustness by buffering against noise through concerted the feedback circuitry. variation in target genes and their inhibitors. A systematic Monte Carlo model analysis approach charac- terizing the model behavior for randomly chosen parameter Bambi and Smad6/7 Negative Feedbacks Canalize Xenopus Devel- values was used to further understand the requirements for opment. Extending from our in vitro results, we next addressed BAMBI to be effective in expanding the signaling range. This the important and unexplored question: What is the significance analysis revealed that strong BAMBI feedback alone is not suf- of negative feedback on endogenous developmental fluctuations ficient for signaling range expansion. The plot of Fig. 3F contains and intrinsic noise in embryos? To evaluate the role of bambi and a region where the effect of BAMBI on the dynamic signaling smad6/7 negative feedback on canalization, we focused on tail range is maximal (yellow to brown color) when two aspects co- development, because it is amenable to morphometric analysis incide: strong BAMBI feedback regulation (x axis) and maximal and because it is a continuation of the gastrula patterning process synchrony of feedback regulation and target gene induction (y (30) regulated by BMP and Wnt signaling (31). We measured axis). The latter suggests that for BAMBI to be effective in interindividual variation of tail length in Xenopus tadpole-stage expanding the signaling range, its own BMP4 dose–response embryos (Fig. 4 A and B) and calculated the coefficient of vari- curve should fall into the same concentration range as that of ation (Fig. 4B). Experimental manipulation of embryos may itself typical BMP target genes (SI Appendix, SI Discussion and induce variability, e.g., by variable microinjection. To control for Fig. S6). this, we first analyzed variability in embryos in which we knocked We validated this prediction by a dose–response qPCR ex- down nonsynexpressed regulators of BMP and Wnt pathways. pression analysis of BAMBI, SMAD6, SMAD7, and the BMP lu- Embryos were injected with control MO or MOs targeting the ciferase reporter, which confirmed that these synexpressed target BMP antagonists chordin/noggin. This reduced the average tail genes are indeed synchronously induced by BMP4 (SI Appendix, length by about 20% without affecting body length. Importantly, Fig. S7). This synchrony requirement may be the key reason for the the variability of tail length between individual embryos was un- near identical expression patterns within the BMP4 synexpression affected in chordin/noggin morphants; the coefficient of variation group. We therefore propose that synchronous coregulation and was 5% in control and experimental tadpoles. Similarly, knock- thus functional organization in synexpression groups is essential down of the Wnt coreceptor lrp6 using an established MO (32) for expansion of the BMP4 dynamic signaling range. Of note, this also reduced average tail length without affecting interindividual demand for synchronous BMP4 responsiveness places an addi- tail length variability (Fig. 4B). Thus, the phenotypic variation tional constraint on the coevolution of BMP4 synexpression group induced by manipulating Wnt or BMP signaling via antisense MO promoters (16). injection is negligible. Knockdown of the synexpressed BMP4 Negative feedback can generally be important for signaling feedback inhibitors bambi, smad6,orsmad7, either individually homeostasis by promoting robustness under conditions of fluc- or combined, all led to a reduction in tail length similar to that in tuating environment and intrinsic noise (18, 19, 21). To test the lrp6 or chordin/noggin morphants. Importantly, however, this was effect of BMP4 synexpression on response variability, we used accompanied by an up-to-threefold increase in tail length vari- stable HEK293 BMP reporter cells driving the expression of ability compared with controls. The developing eye is another destabilized eGFP (d2eGFP) to analyze the distribution of re- organ exhibiting BMP signaling (33) and prominent BMP4 syn- porter expression at the single-cell level by flow cytometry. The expression (SI Appendix, Fig. S1B) (16). Like for the tail, we ob- histogram of d2eGFP expression in G1 phase cells showed a served a significant increase in eye size variation in bambi/smad6/ unimodal distribution, which shifted to the right following smad7 MO-injected embryos compared with control or chordin/ BMP4 treatment, reflecting induced d2eGFP expression (Fig. noggin morphants (Fig. 4 C and D). 3G, Top, Inset). We then analyzed BMP4 stimulated cells under A greater variability of morphology may stem from greater various knockdown conditions and compared (i) the population variability in gene expression. We therefore asked if this in- mean of the fluorescence and (ii) the d2eGFP distribution pro- creased morphological variability would manifest in variable file (Fig. 3G). siBAMBI, triple siBAMBI/SMAD6/SMAD7, and spatial gene expression. We performed in situ hybridization and siSMURF1 treatment all led to an increase of the mean fluo- morphometrically analyzed the expression domain of myf5 in rescence. This is expected, because knockdown of these inhib- gastrulae, which is induced by intermediate doses of the BMP4 itors should enhance BMP signaling. Importantly, siBAMBI, and morphogen (22). Bambi knockdown reduced the myf5 expression more pronounced triple siBAMBI/SMAD6/SMAD7 treatment, domain and increased interindividual variability of the domain broadened the distribution profile. This manifested itself in size (SI Appendix, Fig. S8 A and B). In contrast, chordin/noggin lower peak values (y axis) and broader curve shape. In contrast, knockdown reduced the size of the myf5 domain even more, but even though siSMURF1 also increased the mean fluorescence, it without affecting domain size variability. did so without changing the width of the d2eGFP distribution. Finally, we quantitatively monitored the expression of the di- These results are in agreement with the notion that synexpressed rect BMP target genes ventx1, ventx2, and ventx3 in single embryos BMP inhibitors enhance uniformity of the BMP signaling re- at the gastrula stage. Knockdown of chordin/noggin enhanced sponse, thereby promoting robustness. BMP target gene expression but did not affect interindividual Consistent with the experimental data, in our best-fit model, expression variability (Fig. 4 E and F). In contrast, combined BAMBI and SMAD6/7 feedback also act synergistically to sup- knockdown of bambi, smad6, and smad7 only moderately induced press cell-to-cell variability of target gene expression (Fig. 3H). the BMP target gene expression but markedly increased expres- In contrast, down-regulation of a nonfeedback inhibitor such as sion variability. The expression of the FGF/Nodal target gene bra SMURF1 has little effect on simulated cell-to-cell variability (Fig. was unaffected by all treatments. We conclude that synexpressed 3H, Inset), supporting the notion that feedback induction is re- BMP inhibitors promote developmental robustness and ensure quired for noise suppression. Notably, noise suppression by canalized development during early gene expression and sub- BAMBI and SMAD6/7 feedback required the assumption that sequent tail formation in Xenopus. fluctuations in gene expression are mainly due to correlated ex- trinsic noise (SI Appendix, SI Discussion). Correlated extrinsic Discussion noise affecting many genes simultaneously arises from fluctua- Synexpression is an intriguing phenomenon, and a number of

tions in global regulatory factors (e.g., the basal transcription growth factor pathways, including BMP4, FGF8, and Notch-Delta, BIOLOGY fi

machinery), and it dominates in eukaryotes (29). Under these show this modular conserved organization (8, 9). Our ndings DEVELOPMENTAL conditions, functional organization in synexpression groups pro- indicate that this design principle promotes enhanced dynamic

Paulsen et al. PNAS | June 21, 2011 | vol. 108 | no. 25 | 10205 Downloaded by guest on September 25, 2021 2.0 A E Co MO Chor/Nogg MO Bb/S6/S7 MO

Control MO Chor/Nogg MO Smad6 MO Smad7 MO BAMBI MO Bb/S6/S7 MO 1.0 0.35 B body C Control MO Chor/Nogg MO BbS6S7 MO tail 0.30 Relative expression

0.25 0 ventx1 ventx2 ventx3 bra

Length (cm) 0.20 100 F Co MO 0.15 200 D * Chor/Nogg MO 80 ** ** body * 20 Bb/S6/S7 MO 16 ** ** tail ** ns 60 12 100 10 8 40

4 Average eye size 20 0 Coefficient of variation (%)

0 of variation (%) Coefficient 0 Coefficient of variation (%) Coefficient Co MO Co MO ventx1 ventx2 ventx3 bra Co MO Lrp6 MO uninjected BAMBI MO Bb/S6/S7 MO Bb/S6/S7 MO Smad6 MOSmad7 MO Chor/Nogg MO Bb/S6/S7 MO Chor/Nogg MO Chor/Nogg MO

Fig. 4. Bambi and smad6/7 negative feedback canalize development. (A) Phenotypes of stage-36/7 X. laevis embryos injected with the indicated MOs. (B)(Upper) Representative body and tail length of stage-36/7 X. laevis embryos injected with MOs as in A.(Lower) Average coefficient of variation of body and tail length of stage 36/7 X. laevis embryos injected as indicated in A of (n = 3) biological replicates with 15–30 embryos per replicate. (C) Eye morphology of stage-36/7 tadpoles from embryos injected as indicated. (D) Quantification of the relative eye size. Graph shows average eye area (Left) and coefficient of variation (Right) of three biological replicates with 15–30 embryos per replicate. (E) qPCR expression analysis of the indicated genes from individual X. laevis embryos of stage 10.5, which were injected at the four-cell stage with the indicated MOs (n =9–10 embryos per sample). (F) Coefficient of variation of early BMP4 marker gene expression from individual stage 10.5 X. laevis embryos as in C (n = 2). Error bars, SD; *P < 0.05 and **P < 0.01 (unpaired t test).

signaling range and signaling robustness and suggests that both was placed into the 37 °C prewarmed luminometer (Fluoroscan Ascent FL; phenomena are coupled (34). The importance of negative feed- Thermo Scientific). The luciferase activity of each well was measured every back in conferring robustness for embryonic development was 6 min for 13 h. Data analysis was performed in Excel, and the background previously shown in embryos challenged by experimental pertur- activity of the respective noninduced samples was subtracted from the bations (7, 35–37). Here we demonstrate that negative feedback other samples. canalizes development also in nonperturbed embryos. Our results FACS Analysis. A total of 75,000 BREx4-SV40-d2eGFP cells were reverse further suggest that the tight spatiotemporal coexpression observed fl transfected with 12.5 pmol control, BAMBI,orSMURF1 siRNAs and 7.5 pmol in synexpression groups re ects the need for synchronous expres- BAMBI + 5 pmol SMAD6/SMAD7 siRNAs using DharmaFECT1 transfection sion of feedback inhibitors with BMP target genes for improved reagent and seeded onto TurboFECT-coated 24-well plates. Media was ex- performance of feedback inhibition. Feedback gain and noise changed 24 h later to 3% FCS, 2% BSA DMEM without phenol red supple- suppression may be further enhanced if synexpressed feedback mented with 5 ng/mL rc-mNoggin for 18 h. Cells were washed 2× with regulators act cooperatively at different levels of the signaling cas- Noggin-free 3% FCS and 2% BSA DMEM for 15 min and then induced with cade, explaining why synexpression groups typically contain multi- 20 ng/mL rc-hBMP4 for 12 h. Cells were harvested by short trypsinization ple feedback inhibitors (18). Indeed, our analytical and numerical buffered in 10 mM Hepes (Lonza). Fifty percent of the cells, or maximally studies support the benefit of multiple negative feedback regulators 75,000 cells/mL, were stained with 5 μM DRAQ5 (Axxora) at 37 °C in order for acting in parallel (SI Appendix, SI Discussion and Fig. S9). cell-cycle gating. A total of 8,500 G1 cells of each sample were measured using a FACSCalibur Flow Cytometer system measuring d2eGFP fluorescence. Single-mutant phenotypes of synexpressed feedback inhibitors Nonliving and disrupted cells were excluded by forward scatter/side scatter are often mild (38–43); rather than regulating patterning, mani- fl fi gating combined with a trigger function set to 25% less d2eGFP uores- festing in a speci c malformation in mutants, they may promote cence than noninduced control cells. From the other half, knockdown effi- phenotypic fidelity and reduce variability due to noise. In conclu- ciencies were checked after total RNA purification using Qiagen RNeasy sion, this study exemplifies the significance of negative feedback in columns (Qiagen), and reverse transcription was performed as described and canalizing natural phenotypic variability in a vertebrate embryo, analyzed by qPCR. a parameter at the interface of evolution and development (35, 36, 44, 45). X. laevis in Vitro Fertilization, Microinjection, Staging, and in Situ Hybridization. In vitro fertilization, embryo culture, microinjection, culture of embryos, and Methods in situ hybridization were performed as previously described (22). If not – × Real-Time Luciferase Assays. Real-time luciferase assays were performed in 96- otherwise indicated, embryos were cultured at 20 22 °C in 0.1 Barth (1/ well Falcon plates, white plastic. A total of 10,000 cells/well were seeded onto 1,000 penicillin/streptomycin) and explants or dissociated animal cap cells in × fi poly-L-lysine (Sigma)-treated wells and cultured for 18 h. Five picomoles of 1.0 modi ed Barth buffer supplemented with 2% BSA (1/1,000 penicillin/ a nontargeting control siRNA or BAMBI siRNAs (SI Appendix) were trans- streptomycin; New England BioLabs) as previously described (46). fected in duplicates using DharmaFECT 1 transfection reagent according to the manufacturer’s instructions. At 24 h later, media was exchanged and Animal Cap BMP4 Dose–Response. Four-cell–stage X. laevis embryos were replaced by media containing 3% FCS, 2% BSA (New England BioLabs), 1% injected with 15 ng/blastomere of control and BAMBI into the animal P/S, 1% L-Gln, 20 mM Hepes (pH 7.4) (Lonza), 0.5 mM D-luciferin (Promega) in hemisphere or, in case of Chordin-Noggin MO, at the border of the animal/ DMEM without phenol red (Lonza). Cells were cultured for 4.5 h at 37 °C, 5% equatorial region. Animal cap explants were then excised at stage 8.5 from – CO2, and then received 4× concentrated rc-hBMP4 (R&D Systems) dissolved 35 40 embryos and transferred to a HEMA (Sigma)-treated test tube and in assay media, yielding a final concentration of 0.0, 0.5, 1.0, 5.0, 25.0, or dissociated in 1× Ca-free Barth buffer (pH 7.6) for 5–8 min. Dissociated an- 50.0 ng/mL rc-hBMP4 and a final volume of 200 μL media. The plate bottom imal caps were pelleted for 1 min at 100 × g and resuspended in 120 μLof

10206 | www.pnas.org/cgi/doi/10.1073/pnas.1100179108 Paulsen et al. Downloaded by guest on September 25, 2021 2% BSA, 1× modified Barth buffer (pH 7.6.). A total of 6× 20 μL of the cell Intrapopulation Variability of BMP4 Target Gene Expression. Four-cell–stage suspension were placed into 0.5 mg/mL fibronectin-treated 96-well plates X. laevis embryos were injected with 15 ng/blastomere of control, 5 ng/blas- and induced with 20 μLof2× concentrated rc-hBMP4, yielding doses of 0, tomere of each BAMBI/Smad6/Smad7, or 2.5 ng/blastomere of each Chordin/ 0.03, 0.1, 0.3, 0.9, and 2.7 μg/mL rc-hBMP4. Total RNA was prepared when Noggin + 10 ng/blastomere control MO into the equatorial region and cul- sibling embryos reached stage 12.5 by phenol/chloroform extraction, and RT- tured in 0.3× Barth (pH 7.6) until stage 10.5. Ten single embryos of each qPCR assays were performed (SI Appendix). experimental setup and of the same developmental stage were individually harvested. Total RNA was extracted by phenol/chloroform extraction, and fi Morphometry Analysis of Average X. laevis Embryo Body and Tail Length. RT-qPCR assays were performed. The coef cient of variation was calculated for each experimental subset using the following formula: (SD/mean ex- X. laevis two- to four-cell–stage embryos of the same fertilization were pression) × 100%. injected with either 60 ng control; 10 ng Chordin + 10 ng Noggin + 40 ng control; 40 ng Smad6 + 20 ng control; 40 ng Smad7 + 20 ng control; 40 ng Dual Luciferase Assays, Morpholinos, qRT-PCR and Primers, siRNAs, Western BAMBI + 20 ng control; or 20 ng Smad6 + 20 ng Smad7 + 20 ng BAMBI MO Blot, and Other BMP4 Dose–Response Assays and Mathematical Modeling. in a total volume of 20 nL/embryo. Embryos were cultured at room tem- Please refer to the detailed methods and reagents in SI Appendix. perature until they reached stage 37/38, and were then fixed in a buffer containing 4% PFA. The absolute body and tail length, or the relative eye ACKNOWLEDGMENTS. We thank P. Stannek and T. Maiwald for expert size of individual embryos, were measured using ImageJ. The corresponding technical help and U. Klingmüller for modeling suggestions. This work was coefficient of variation was calculated with the following equation: (SD/ supported by the Deutsche Forschungsgemeinschaft (German Research mean) × 100%. Foundation) and the Bundesministerium für Bildung und Forschung.

1. De Robertis EM, Kuroda H (2004) Dorsal-ventral patterning and neural induction in 24. Hata A, et al. (2000) OAZ uses distinct DNA- and protein-binding zinc fingers in Xenopus embryos. Annu Rev Cell Dev Biol 20:285–308. separate BMP-Smad and Olf signaling pathways. Cell 100:229–240. 2. Oelgeschläger M, Larraín J, Geissert D, De Robertis EM (2000) The evolutionarily 25. Ueda HR, et al. (2005) System-level identification of transcriptional circuits underlying conserved BMP-binding protein Twisted gastrulation promotes BMP signalling. mammalian circadian clocks. Nat Genet 37:187–192. Nature 405:757–763. 26. Gawantka V, Delius H, Hirschfeld K, Blumenstock C, Niehrs C (1995) Antagonizing the 3. Eldar A, et al. (2002) Robustness of the BMP morphogen gradient in Drosophila Spemann organizer: Role of the homeobox gene Xvent-1. EMBO J 14:6268–6279. embryonic patterning. Nature 419:304–308. 27. Shapira E, Marom K, Yelin R, Levy A, Fainsod A (1999) A role for the homeobox gene 4. Reversade B, De Robertis EM (2005) Regulation of ADMP and BMP2/4/7 at opposite Xvex-1 as part of the BMP-4 ventral signaling pathway. Mech Dev 86:99–111. embryonic poles generates a self-regulating morphogenetic field. Cell 123: 28. Kuroda H, Wessely O, De Robertis EM (2004) Neural induction in Xenopus: 1147–1160. Requirement for ectodermal and endomesodermal signals via Chordin, Noggin, beta- 5. Lee HX, Ambrosio AL, Reversade B, De Robertis EM (2006) Embryonic dorsal-ventral Catenin, and Cerberus. PLoS Biol 2:e92. signaling: Secreted frizzled-related proteins as inhibitors of tolloid proteinases. Cell 29. Sigal A, et al. (2006) Variability and memory of protein levels in human cells. Nature – 124:147 159. 444:643–646. 6. Ambrosio AL, et al. (2008) Crossveinless-2 is a BMP feedback inhibitor that binds 30. Gont LK, Steinbeisser H, Blumberg B, de Robertis EM (1993) Tail formation as – Chordin/BMP to regulate Xenopus embryonic patterning. Dev Cell 15:248 260. a continuation of gastrulation: The multiple cell populations of the Xenopus tailbud 7. Ben-Zvi D, Shilo BZ, Fainsod A, Barkai N (2008) Scaling of the BMP activation gradient derive from the late blastopore lip. Development 119:991–1004. – in Xenopus embryos. Nature 453:1205 1211. 31. Beck CW, Whitman M, Slack JM (2001) The role of BMP signaling in outgrowth and fi 8. Gawantka V, et al. (1998) Gene expression screening in Xenopus identi es molecular patterning of the Xenopus tail bud. Dev Biol 238:303–314. pathways, predicts gene function and provides a global view of embryonic 32. Hassler C, et al. (2007) Kremen is required for neural crest induction in Xenopus and patterning. Mech Dev 77:95–141. promotes LRP6-mediated Wnt signaling. Development 134:4255–4263. 9. Niehrs C, Pollet N (1999) Synexpression groups in eukaryotes. Nature 402:483–487. 33. Goldman DC, Donley N, Christian JL (2009) Genetic interaction between Bmp2 and 10. Imamura T, et al. (1997) Smad6 inhibits signalling by the TGF-beta superfamily. Bmp4 reveals shared functions during multiple aspects of mouse organogenesis. Nature 389:622–626. Mech Dev 126:117–127. 11. Nakao A, et al. (1997) Identification of Smad7, a TGFbeta-inducible antagonist of TGF- 34. Hornung G, Barkai N (2008) Noise propagation and signaling sensitivity in biological beta signalling. Nature 389:631–635. networks: A role for positive feedback. PLOS Comput Biol 4:e8. 12. Casellas R, Brivanlou AH (1998) Xenopus Smad7 inhibits both the activin and BMP 35. Eldar A, Rosin D, Shilo BZ, Barkai N (2003) Self-enhanced ligand degradation underlies pathways and acts as a neural inducer. Dev Biol 198:1–12. robustness of morphogen gradients. Dev Cell 5:635–646. 13. Onichtchouk D, et al. (1999) Silencing of TGF-beta signalling by the pseudoreceptor 36. Bénazet JD, et al. (2009) A self-regulatory system of interlinked signaling feedback BAMBI. Nature 401:480–485. loops controls mouse limb patterning. Science 323:1050–1053. 14. Ishida W, et al. (2000) Smad6 is a Smad1/5-induced Smad inhibitor. Characterization 37. Li X, Cassidy JJ, Reinke CA, Fischboeck S, Carthew RW (2009) A microRNA imparts of bone morphogenetic protein-responsive element in the mouse Smad6 promoter. robustness against environmental fluctuation during development. Cell 137:273–282. J Biol Chem 275:6075–6079. 38. Galvin KM, et al. (2000) A role for smad6 in development and homeostasis of the 15. Legewie S, Herzel H, Westerhoff HV, Blüthgen N (2008) Recurrent design patterns in – the feedback regulation of the mammalian signalling network. Mol Syst Biol 4:190. cardiovascular system. Nat Genet 24:171 174. 16. Karaulanov E, Knöchel W, Niehrs C (2004) Transcriptional regulation of BMP4 39. Lin W, et al. (2005) Synergistic activity of Sef and Sprouty proteins in regulating the – synexpression in transgenic Xenopus. EMBO J 23:844–856. expression of Gbx2 in the mid-hindbrain region. Genesis 41:110 115. 17. Becskei A, Serrano L (2000) Engineering stability in gene networks by autoregulation. 40. Li R, et al. (2006) Deletion of exon I of SMAD7 in mice results in altered B cell – Nature 405:590–593. responses. J Immunol 176:6777 6784. 18. Stelling J, Sauer U, Szallasi Z, Doyle FJ, III, Doyle J (2004) Robustness of cellular 41. Chen J, Bush JO, Ovitt CE, Lan Y, Jiang R (2007) The TGF-beta pseudoreceptor gene functions. Cell 118:675–685. Bambi is dispensable for mouse embryonic development and postnatal survival. – 19. Barkai N, Shilo BZ (2007) Variability and robustness in biomolecular systems. Mol Cell Genesis 45:482 486. 28:755–760. 42. Phng LK, et al. (2009) Nrarp coordinates endothelial Notch and Wnt signaling to 20. Yu RC, et al. (2008) Negative feedback that improves information transmission in control vessel density in angiogenesis. Dev Cell 16:70–82. yeast signalling. Nature 456:755–761. 43. Tramullas M, et al. (2010) BAMBI (bone morphogenetic protein and activin 21. Kiel C, Yus E, Serrano L (2010) Engineering signal transduction pathways. Cell 140: membrane-bound inhibitor) reveals the involvement of the transforming growth 33–47. factor-beta family in pain modulation. J Neurosci 30:1502–1511. 22. Dosch R, Gawantka V, Delius H, Blumenstock C, Niehrs C (1997) Bmp-4 acts as 44. Houchmandzadeh B, Wieschaus E, Leibler S (2002) Establishment of developmental a morphogen in dorsoventral mesoderm patterning in Xenopus. Development 124: precision and proportions in the early Drosophila embryo. Nature 415:798–802. 2325–2334. 45. Gregor T, Tank DW, Wieschaus EF, Bialek W (2007) Probing the limits to positional 23. Zhu H, Kavsak P, Abdollah S, Wrana JL, Thomsen GH (1999) A SMAD information. Cell 130:153–164. targets the BMP pathway and affects embryonic pattern formation. Nature 400: 46. Gurdon JB, et al. (1999) Single cells can sense their position in a morphogen gradient. 687–693. Development 126:5309–5317. BIOLOGY DEVELOPMENTAL

Paulsen et al. PNAS | June 21, 2011 | vol. 108 | no. 25 | 10207 Downloaded by guest on September 25, 2021