Tissue-specific I-Smad mechanisms revealed by structure–function analysis in Drosophila

The Drosophila I-Smad Dad uses MH1-mediated transcriptional regulation in wing tissue but both MH1- and MH2-mediated mechanisms in neurons.

Abstract

Inhibitory Smads (I-Smads) regulate TGF-β/BMP signaling through multiple distinct mechanisms, but whether different tissues preferentially employ specific mechanisms remains unknown. To address this question, we performed structure–function analyses of the Drosophila I-Smad, Dad, and its vertebrate orthologs Smad6 and Smad7 in neural and wing tissues, measuring outputs of BMP signaling in vivo. We identified a 24–amino acid putative DNA-binding domain within the MH1 domain of Dad that is essential for inhibitory function in wing tissue but unessential in neural tissue. Structural analyses revealed that ΔDNA-binding domain disrupts a β-hairpin structure homologous to R-Smad DNA-binding regions. We also found that Dad requires an intact MH1 domain to disrupt wing development, whereas either MH1 or MH2 can independently disrupt BMP signaling in motor neurons. These findings support a model where Dad functions through MH1-mediated transcriptional regulation in wing primordium, but through multiple mechanisms in neurons. Comparative analysis revealed that vertebrate I-Smad orthologs also show tissue-specific activity patterns, with structural predictions suggesting that Smad6 retains ancestral DNA-binding capacity, whereas Smad7 has evolved enhanced MH2-mediated functions. These results reveal context-dependent mechanisms of I-Smads that further the understanding of TGF-β/BMP pathway regulation.

Introduction

TGF-β signaling is an ancient and evolutionarily conserved pathway for intercellular communication that controls diverse life processes such as axis formation, cell differentiation, and tissue homeostasis. Its pleiotropy necessitates precise intra- and extracellular modulation to elicit the appropriate effects in target cells (Massagué, 2012). Dysregulation of TGF-β signaling underlies the pathophysiology of numerous disease states and developmental disorders. For instance, as TGF-β inhibits mitosis and promotes apoptosis in many tissues, loss of TGF-β signaling often engenders tumorigenesis. In later oncogenic stages, however, cells become resistant to the apoptosis-inducing effects, and TGF-β promotes tumor growth and invasiveness via dedifferentiation, angiogenesis, and immune suppression (Derynck et al, 2001; Syed, 2016; Chan et al, 2022). Thus, the ability of cells to receive and interpret physiologically appropriate levels of TGF-β signaling is vital across metazoan life stages.

Inhibitory Smads (I-Smads) are intracellular proteins that ensure an appropriate response to TGF-β signaling. They are transcriptional targets of the signaling pathway, functioning in a negative-feedback mechanism (Miyazawa & Miyazono, 2017). Dysfunction of I-Smads is implicated in the pathogenesis of human diseases such as colorectal cancer, atrial fibrosis, and inflammatory bowel disease (Broderick et al, 2007; He et al, 2011; Fortini et al, 2014; Tang et al, 2019). Although mammalian genomes encode two I-Smads (Smad6 and Smad7), the Drosophila genome contains only a single I-Smad–encoding gene, Daughters against Decapentaplegic (Dad) (Tsuneizumi et al, 1997; Massagué et al, 2005). This lack of genetic redundancy, combined with powerful genetic tools and well-characterized phenotypes, makes Drosophila melanogaster an ideal system to study the mechanisms of I-Smad function in vivo.

I-Smads belong to the Smad family of proteins, which transduce signals from TGF-β receptors. The Smad family also includes R-Smads, which are the receptor-regulated Smads that become activated by type I receptors, and Co-Smads (common-mediator Smads), which form complexes with activated R-Smads to enable their nuclear translocation and transcriptional activity (Schmierer & Hill, 2007). Vertebrate R-Smads include Smad2 and Smad3, which respond to TGF-β and activin signals, and Smad1, Smad5, and Smad8/9, which respond to bone morphogenetic protein (BMP) signals (Massagué et al, 2005). Smads are characterized by the presence of two main functional domains: the MH1 domain, which mediates DNA binding and nuclear import, and the MH2 domain, which facilitates protein–protein interactions, including receptor binding, Smad oligomerization, and transcriptional activation. These two domains are connected by a variable linker region that contains regulatory sites for posttranslational modifications.

I-Smads share a similar domain organization with other Smads. The crystal structure of SMAD7 reveals that the MH2 domain shares a high degree of structural similarity with the MH2 domains of other Smads (Murayama et al, 2020). However, there is currently no crystal structure available for the I-Smad N-terminal region, and no consensus on whether it constitutes a true MH1 domain. The two mammalian I-Smad paralogs, Smad6 and Smad7, are divergent both in their structure and in their function. Smad7 inhibits both TGF-β and BMP signaling, whereas Smad6 specifically inhibits BMP signaling (Hanyu et al, 2001; Goto et al, 2007). Smad7 is also more potent than Smad6 when repression is quantified via the luciferase assay (Hanyu et al, 2001). Previous studies, largely performed in vitro, indicate that Dad specifically inhibits the BMP branch of TGF-β signaling, and this inhibition has been attributed to Dad preventing Thickveins, a BMP type I receptor, from phosphorylating the R-Smad Mothers against Decapentaplegic (Mad) (Inoue et al, 1998; Kamiya et al, 2008; Li et al, 2017).

Intriguingly, I-Smads have been shown to antagonize TGF-β signaling by several distinct mechanisms in addition to the abovementioned inhibition of type I receptors (Miyazawa & Miyazono, 2017). I-Smads can form complexes with R-Smads, thus preventing their association with co-Smads (Hata & Chen, 2016; Yan et al, 2016). I-Smads cooperate with E3 ubiquitin ligases to target R-Smads and other TGF-β components for destruction (Zhu et al, 1999; Kavsak et al, 2000; Ebisawa et al, 2001; Murakami et al, 2003; Morén et al, 2005). I-Smads can also accumulate in the nucleus and directly regulate transcription to counter the effects of R-Smads (Bai et al, 2000; Hanyu et al, 2001; Zhang et al, 2007). Given that I-Smads function through such diverse mechanisms and considering that many of the previous studies were performed in vitro, we sought to further explore the mechanisms of Dad and I-Smad orthologs by conducting structure–function studies in different in vivo cellular contexts.

Using a series of transgenic animals, here we provide evidence that the human orthologs of Dad, SMAD6 and SMAD7, have diverging functions in Drosophila wing and neural tissue. We further demonstrate that Dad lacking a specific, putative DNA-binding region of its MH1 domain is not sufficient to disrupt wing development yet is sufficient to inhibit BMP signaling in neural tissue. By comparing protein structure predictions, we extend these results to hypothesize that Smad6/Dad inhibit TGF-β signaling by a divergent mechanism from Smad7, and that this mechanism necessitates the MH1 domain, possibly for DNA binding. We also find that in both brain and wing tissues, Dad does not require palmitoylation at a C-terminal cysteine (C556) previously shown to be required for inhibition in ovarian tissue (Li et al, 2017). Together, these findings shed light on the molecular mechanisms of TGF-β regulation and provide compelling evidence for the evolution of distinct, tissue-specific functionality among I-Smad orthologs.

Results

Creation of a series of transgenic animals to investigate tissue-specific I-Smad functionality

To investigate the structural basis for alternative I-Smad mechanisms, we created a series of transgenic Drosophila to express vertebrate I-Smad orthologs, Smad6 and Smad7, and targeted alterations of the Dad gene (Fig 1A). In addition to full-length Smad6 and Smad7 (Smad7L), a transgenic line expressing a shorter isoform of Smad7 lacking the MH1 domain (Smad7S) was developed. Dad truncations covering the MH1 and MH2 domains (Dad^ΔMH1 and Dad^ΔMH2) were constructed, along with deletions targeting specific residues implicated in DNA binding and trimerization (Dad^ΔDNABD and Dad^ΔTrimer) (Fig S1A–C). We also created a substitution replacing a C-terminal cysteine of Dad that is a reported target of palmitoylation (Dad^C556A).

Figure 1. Disruption of Drosophila wing development by I-Smad structure–function constructs and vertebrate orthologs.

(A) Protein domain organization of I-Smad constructs. The scale indicates length in amino acids. Locations of putative DNA-binding sites (blue) and a putative trimer interface (red) are indicated. (B) Quantification of wing area (mm²) in adult flies expressing I-Smad constructs in wing primordium with Ap-Gal4 and genetic background controls (Ap-Gal4/+). Data points for individual female or male wings are represented by ♀ or ♂ symbols, respectively. Color of bars/points indicates phenotypic class. Bars represent the mean ± SD. Statistical comparisons were performed using Dunnett’s test comparing individual genotypes with controls. ****P < 0.0001, **P < 0.01. Number of female wings in the order shown in the graph: 21, 28, 20, 19, 24, 25, 18, 23, 24, and 25. Numbers of male wings in the order shown in the graph: 20, 29, 22, 21, 21, 24, 18, 23, 24, and 17.

Figure S1. Residue alignments and precise locations of deletions.

(A) Diagram showing regions analyzed for alignment in a putative DNA-binding region (red boxes) or trimer interface sites (blue boxes). (B) Alignment of residues in the region corresponding to red boxes. The highlight shows residues deleted in ΔDNABD. Numbers show Clustal2.1 alignment scores. (C) Alignment of residues in the region corresponding to blue boxes. The highlight shows residues deleted in ΔTrimer. Numbers show Clustal2.1 alignment scores.

Phenotypic analysis of wing tissue after construct expression

To assess the potency of our gene products to inhibit BMP signaling in developing wing tissue, constructs were expressed with Ap-Gal4, which is expressed primarily in the dorsal compartment of the wing imaginal disk (Butterworth & King, 1965; O’Keefe et al, 1998). We found that phenotypes clustered into three groups based on severity. Group 1, which includes genetic background control, Dad^ΔMH1, and Dad^ΔDNABD, had wing phenotype indistinguishable from controls. The mean wing areas (in mm²) for males were 1.39 ± 0.146, 1.50 ± 0.089, and 1.53 ± 0.124, and for females were 1.96 ± 0.163, 1.90 ± 0.091, and 1.88 ± 0.147, respectively. Group 2 includes Dad^ΔMH2, Smad7S, Smad6, and Dad^ΔTrimer. This group had moderately severe phenotypes, with mean areas for males of 1.26 ± 0.085, 1.05 ± 0.143, 1.23 ± 0.183, and 0.749 ± 0.107. Group 2 female wings were 1.60 ± 0.076, 1.20 ± 0.198, 1.26 ± 0.167, and 0.710 ± 0.163, respectively. Finally, group 3, which includes Dad, Smad7L, and Dad^C556A, had the most severe phenotypes with mean areas for males of 0.214 ± 0.045, 0.219 ± 0.079, and 0.262 ± 0.047, and for females of 0.304 ± 0.076, 0.258 ± 0.112, and 0.280 ± 0.063, respectively (Fig 1B).

Based on these results, we conclude that Dad lacking its MH1 domain is insufficient to inhibit wing tissue growth. Dad with a 24 aa deletion in a putative DNA-binding site (Dad^ΔDNABD) within the MH1 domain is also insufficient. Conversely, Smad7 lacking the MH1 domain is sufficient to inhibit wing growth. A substitution converting the cysteine at position 556 to alanine (Dad^C556A) did not impair function. These results indicate that Dad must include MH1 domain to disrupt wing growth, whereas the vertebrate counterpart Smad7 can disrupt growth without an MH1 domain. Furthermore, Dad lacking an MH2 domain is still capable of producing some disruption in wing growth. As the Ap-Gal4 driver used for these experiments is expressed most strongly in the dorsal compartment of the developing wing, we sought to confirm our findings with an alternative wing-specific driver, rotund-Gal4 (Rn-Gal4), which is expressed most strongly in the central region and notum of the wing disk (St Pierre et al, 2002). When Dad^ΔMH1 was expressed using Rn-Gal4, the wing area remained similar to controls (1.69 ± 0.115 versus 1.53 ± 0.097 and 1.28 ± 0.144 versus 1.118 ± 0.114 for males and females, respectively) (Fig S2SA and B). Forcing expression of Dad with Rn-Gal4 produced a severe phenotype (0.108 ± 0.046 and 0.081 ± 0.036 for males and females, respectively). Thus, the Rn-Gal4 driver seems to represent a more potent condition than Ap-Gal4. Notably, we observed a loss of anterior crossveins in 19 of 27 (70.4%) and incomplete posterior crossveins in 4 of 27 (14.8%) Rn-Gal4>DadΔMH1 wings examined (Fig S2A). Thus, in more severe cases, MH2-mediated mechanisms can have some effect on wing development.

Figure S2. Expression of Dad^ΔMH1 in wing disks using an alternative Gal4 driver (Rn-Gal4) does not alter wing area.

(A) Representative images of wings from female genetic background control (Rn-Gal4/+) and animals overexpressing full-length Dad or Dad^ΔMH1 with rotund-Gal4 (Rn>Dad and Rn>ΔMH1). Notice missing anterior crossveins and incomplete posterior crossvein on Rn>ΔMH1. (B) Quantification of wing area (mm²) in adult flies. Data points for individual female or male wings are represented by ♀ or ♂ symbols, respectively. Color of bars/points indicates phenotypic class. Bars represent the mean ± SD. Statistical comparisons were performed using Dunnett’s test comparing individual genotypes with controls. ****P < 0.0001. Number of female wings in the order shown in the graph: 14, 13, and 15. Numbers of male wings in the order shown in the graph: 11, 14, and 12.

Structural analysis of MH1 domains

As the MH1 domain of Dad was sufficient to disrupt wing development, we sought to identify potential differences in protein structures among the MH1 domains of I-Smad orthologs. We first generated a model of the MH1 domain of the R-Smad, Smad3, using AlphaFold3 (Fig 2A). The model matched the previously published crystal structure (Shi et al, 1998). Notably, the β-hairpin structure and its interaction with DNA were correctly replicated. When we compared predictions of the MH1 structures of Dad, Smad6, and Smad7, we found that the β-hairpin structure is longer and more disorganized in Smad7 compared with Dad or Smad6, which had β-hairpin structures with antiparallel stretches that more closely resemble the DNA-binding structure in Smad3 (Fig 2B).

Figure 2. Human I-Smads inhibit BMP signaling in Drosophila tissues.

(A, B) AlphaFold3 structural predictions of MH1 domains. The β-hairpin region is shown in yellow. Remaining MH1 is shown in blue. (A) Smad3 MH1 domain showing DNA-binding region and predicted interaction. (B) Dad, Smad6, and Smad7 MH1 domains. (C) Representative wings from females expressing indicated construct in wing primordium. The scale bar represents 1 mm. (D, E) Transgenic control (TwitGFP, OK371-Gal4/+) compared with OK371-Gal4–driven overexpression of UAS-Dad, UAS-Smad6, UAS-Smad7L, and UAS-Smad7S. (D) Motor neurons in ventral nerve cord triple-labeled for Elav (blue), TwitGFP (green), and pMad (red). The scale bar represents 20 microns. (E) Quantification of pMad and TwitGFP intensity normalized to the genetic background control. Bars represent the mean ± SEM. Each dot represents mean data from motor neurons of one CNS. Statistical comparisons were performed using Dunn’s test with the Bonferroni correction. ****P < 0.0001, **P < 0.01. Number of brains analyzed for pMad in the order shown in the graph: 52, 24, 25, 20, and 28. Number of brains analyzed for TwitGFP in the order shown in the graph: 53, 27, 22, 19, and 26.

Effects of human I-Smad orthologs in neural tissue

To gauge the ability of specific I-Smad structural regions to inhibit BMP signaling in motor neurons, we expressed constructs using OK371-Gal4, a Gal4 enhancer trap in the proximity of the DVGLUT gene, which drives expression specifically in glutamatergic motor neurons (Mahr & Aberle, 2006; Sanyal, 2009). We began by determining the ability of the vertebrate I-Smads, Smad6 and Smad7, to inhibit BMP signaling in this neural tissue. BMP signaling was quantified by visualizing levels of phosphorylated Mad (pMad) in motor nuclei. Smad6, Smad7L, and Smad7S decreased nuclear pMad relative to the genetic background control (mean ± s.e.m.: 0.650 ± 0.098, 0.315 ± 0.040, and 0.495 ± 0.061, respectively) (Fig 2). To evaluate a downstream effector of BMP signaling, we used a GFP protein trap in the BMP target gene twit (Venken et al, 2011; Kim & Marqués, 2012; Sulkowski et al, 2016). Smad7L decreased TwitGFP levels to 0.519 ± 0.048 relative to the control. Interestingly, Smad6 and Smad7S did not cause a significant reduction in TwitGFP levels (0.897 ± 0.070 and 0.792 ± 0.072 relative to the control, respectively). Thus, vertebrate I-Smad orthologs were sufficient in inhibiting BMP signaling in neural (Fig 2D and E) and developing wing tissues (Fig 2C, quantified in Fig 1B). The wing phenotypes of these animals included changes to wing area, shape, venation, and blistering.

Regulation of noncanonical BMP signaling at neuromuscular synapses by I-Smads

BMP signaling at the neuromuscular junction (NMJ) encompasses both the canonical retrograde pathway, which culminates with pMad accumulation in the nucleus (Aberle et al, 2002; Marqués et al, 2002; McCabe et al, 2003), and a noncanonical pathway, characterized by pMad accumulation at synaptic active zones (Dudu et al, 2006; O’Connor-Giles et al, 2008; Higashi-Kovtun et al, 2010; Smith et al, 2012; Sulkowski et al, 2014, 2016). To test whether Dad affects the noncanonical BMP pathway, we examined synaptic pMad levels in animals overexpressing Dad in motor neurons (Fig 3A and B). We saw a significant reduction in pMad accumulation at NMJ synapses in animals overexpressing Dad in motor neurons relative to controls (0.313 ± 0.043; P < 0.005). We also observed defects in NMJ morphology reminiscent of canonical BMP pathway perturbations, in accordance with previous reports (Lee et al, 2016) (Fig 3C). We observed a reduction in type Ib boutons at muscle 4 from 57.5 ± 19.7 in controls to 27.7 ± 12.2, and a reduction in type Is from 27.6 ± 24.2 in controls to 13.3 ± 10.3. Therefore, I-Smads can inhibit both canonical and noncanonical BMP signaling at the NMJ.

Figure 3. Presynaptic Dad expression inhibits local BMP signaling at NMJ synapses.

Transgenic control (TwitGFP, OK371-Gal4/+) compared with OK371-Gal4–driven overexpression of UAS-Dad. (A) Muscle 4 NMJ triple-labeled for the presynaptic membrane marker HRP (blue), synaptic marker synaptotagmin (Syt) (green), and synaptic pMad (red). The scale bar represents 10 microns. (B) Quantification of synaptic pMad intensity. (C) Quantification of boutons present on abdominal muscle 4. Each dot represents one data point, which is the mean of one animal’s NMJ 4s from A2 and A3 normalized to control (4 NMJs per data point). Bars represent the mean ± SD. Statistical comparison was performed with a t test. ***P < 0.005, **P < 0.01. Eight animals were analyzed for each genotype.

I-Smad MH1 and MH2 domain truncations in neural tissue

To evaluate the necessity of I-Smad structural domains to inhibit BMP signaling in motor neurons, the expression of Dad constructs with truncations of the MH1 and MH2 domains (Dad^ΔMH1 and Dad^ΔMH2) was driven with OK371-Gal4 (Fig 4). Both Dad^ΔMH1 and Dad^ΔMH2 were sufficient to suppress nuclear pMad accumulation (0.201 ± 0.015 and 0.396 ± 0.058 relative to the control, respectively). Dad^ΔMH1 expression significantly reduced TwitGFP reporter gene expression to 0.525 ± 0.047 relative to the control. Dad^ΔMH2 reduced TwitGFP to 0.734 ± 0.083, but this reduction was not statistically significant. This analysis reveals that the functional domains of Dad can independently inhibit BMP signaling in neural tissue. Furthermore, the upstream effect on pMad accumulation can be achieved independently by either MH1 or MH2 domain, but the inability of Dad^ΔMH2 to affect TwitGFP levels suggests that the MH1 domain is required to affect gene expression. This contrasts with the results of expression in wing primordium tissue, where Dad^ΔMH1 showed no effect on wing development and Dad^ΔMH2 produced only a small reduction in area (Fig 4C, quantified in Fig 1B). Notably, we did not observe any qualitative differences (e.g., abnormal venation, wing blade borders, or blistering) in these wings besides the modest reduction in area in Dad^ΔMH2-expressing animals.

Figure 4. Dad can repress BMP signaling in motor neurons through multiple mechanisms.

Transgenic control (TwitGFP, OK371-Gal4/+) compared with OK371-Gal4–driven overexpression of UAS-Dad^ΔMH1 and UAS-Dad^ΔMH2. (A) Motor neurons in ventral nerve cord triple-labeled for Elav (blue), TwitGFP (green), and pMad (red). The scale bar represents 20 microns. (B) Quantification of pMad and TwitGFP intensity. Bars represent the mean ± SEM. (C) Representative images of wing phenotypes for females of indicated genotypes. The scale bar represents 1 mm. Statistical comparisons were performed using Dunn’s test with the Bonferroni correction. ****P < 0.0001, ***P < 0.001. Number of brains analyzed for pMad in the order shown in the graph: 24, 31, and 22. Number of brains analyzed for TwitGFP in the order shown in the graph: 23, 32, and 24.

Deletion of putative DNA-binding sites and trimerization interfaces

To determine whether specific regions of the I-Smad MH1 and MH2 domains are sufficient to inhibit BMP signaling in motor neurons, we drove the expression of Dad constructs with deletions in putative DNA-binding sites (Dad^ΔDNABD) or trimer interfaces (Dad^ΔTrimer) with OK371-Gal4 (Fig 5). Structural predictions show that the ΔDNABD mutation removes the β-hairpin structure implicated in DNA binding (Fig 5A). Both Dad^ΔDNABD and Dad^ΔTrimer significantly reduced the accumulation of nuclear pMad relative to the control (0.292 ± 0.0276 and 0.341 ± 0.0357, respectively) (Fig 5B and C). Although Dad^ΔTrimer caused a significant decrease in TwitGFP reporter gene levels (0.656 ± 0.062), Dad^ΔDNABD expression did not significantly change TwitGFP expression (0.849 ± 0.0457). Therefore, Dad lacking putative DNA-binding sites within the MH1 domain was not sufficient for repressing this target gene’s expression. However, this same construct was sufficient for decreasing pMad accumulation, indicating that these processes are separable and can be attributed to different structural motifs.

Figure 5. Dad lacking a putative DNA-binding region is sufficient for repressing pMad accumulation and TwitGFP levels in brains, but not for repressing wing development.

(A) AlphaFold3 structure prediction of Dad and Dad^ΔDNABD MH1 domains. The putative Dad DNA-binding domain is depicted in yellow (absent in Dad^ΔDNABD). (B, C) Transgenic control (TwitGFP, OK371-Gal4/+) compared with OK371-Gal4–driven overexpression of UAS-Dad^ΔTrimer and UAS-Dad^ΔDNABD. (B) Motor neurons in ventral nerve cord triple-labeled for Elav (blue), TwitGFP (green), and pMad (red). The scale bar represents 20 microns. (C) Quantification of pMad and TwitGFP intensity normalized to the genetic background control. Bars represent the mean ± SEM. Statistical comparisons were performed using Dunn’s test with the Bonferroni correction (pMad) or Tukey’s HSD test (TwitGFP). ***P < 0.001, **P < 0.01, *P < 0.05. (D) Representative images of wing phenotypes for females of indicated genotypes. The scale bar represents 1 mm. Number of brains analyzed for pMad in the order shown in the graph: 22, 13, 20, and 33. Number of brains analyzed for TwitGFP in the order shown in the graph: 21, 13, 20, and 32.

In wing tissue, Dad^ΔDNABD-expressing animals showed no alterations to area or morphology, with normal vein patterning and wing blade structure (Fig 5D, quantified in Fig 1B). In contrast, DadΔ^Trimer expression produced wings with defects in vein patterning and anterior border morphology, though without the blistering or severe structural disruptions observed with some other constructs. These findings indicate that different tissue types preferentially employ different I-Smad mechanisms, and suggest that the trimerization interface may play distinct roles in different cellular contexts.

Removal of C-terminal palmitoylation site

Previous studies identified a conserved C-terminal cysteine (C556) of Dad that serves as a palmitoylation site required for membrane association and BMP signaling inhibition (Li et al, 2017). We generated a Dad construct containing a substitution of C556 with alanine (Dad^C556A). This variant was expressed using Ap-Gal4 in the wing and OK371-Gal4 in the brain to assess tissue-specific effects on BMP signaling. In wing tissue, the expression of Dad^C556A produced severely reduced and malformed wings comparable to those of full-length Dad-expressing animals, placing it into the most severe phenotypic category (Fig 6C, quantified in Fig 1B). This finding suggests that C556 is not essential for I-Smad function in wing tissue, contrary to its reported function in ovarian tissue, where this same mutation reduced effectiveness to levels similar to the genetic background control (Li et al, 2017). The expression of Dad^C556A in motor neurons resulted in a dramatic reduction in both pMad and TwitGFP levels (0.275 ± 0.0319 and 0.395 ± 0.0469 relative to the control, respectively) (Fig 6A and B). These results indicate that C556-dependent palmitoylation is not universally required for Dad function and that this modification’s requirement may be tissue-specific, rather than universally applicable to I-Smad functionality.

Figure 6. Dad functions normally without cysteine 556 in wing and brain.

Transgenic control (TwitGFP, OK371-Gal4/+) compared with OK371-Gal4–driven overexpression of UAS-Dad^C556A. (A) Motor neurons in ventral nerve cord triple-labeled for Elav (blue), TwitGFP (green), and pMad (red). The scale bar represents 20 microns. (B) Quantification of pMad and TwitGFP intensity normalized to the genetic background control. Bars represent the mean ± SEM. Statistical comparisons were performed using Dunn’s test with the Bonferroni correction. ****P < 0.0001. (C) Representative images of wing phenotypes for females of indicated genotypes. The scale bar represents 1 mm. Number of brains analyzed for pMad in the order shown in the graph: 16 and 17. Number of brains analyzed for TwitGFP in the order shown in the graph: 15 and 20.

Structural predictions of physical interactions between Dad and target molecules

Our results support a model where I-Smads inhibit TGF-β/BMP signaling through several distinct mechanisms, and different tissues preferentially employ these different mechanisms. As we used structural prediction to compare putative DNA-binding motifs of R-Smads and I-Smads (Figs 2A and 5A), we wanted to test whether structural predictions can also visualize the physical interactions between I-Smads and the other target proteins, R-Smads and type I receptors. Indeed, AlphaFold3 models support a strong, complementary conformation between Dad and the R-Smad Mad and between Dad and the type I receptor Thickveins (Tkv) (Fig 7). The I-Smad/R-Smad conformation with the closest fit was between two I-Smads and one R-Smad (Fig 7A). There were five C-terminal residues of one of the Dad molecules that formed hydrogen bonds with the MH2 domain of Mad: M520, H523, G524, M528, and R568. The other Dad molecule used G402, E414, F415, P530, D547, and Y548 to bind with Mad (Fig S3). Structural predictions most strongly support an association between a single I-Smad and type I receptor (Fig 7B). We identified nine C-terminal residues of Dad that formed hydrogen bonds with the cytoplasmic region of Tkv: T406, W406, D487, L495, M528, P530, T544, I553, and E561. In agreement with prior knowledge, these residues are all found within the C-terminal MH2 domain of the I-Smad.

Figure 7. Structural predictions support multiple modes of I-Smad function.

(A) AlphaFold3 predicted protein interaction between two Dad molecules (I-Smad) and one Mad (R-Smad). Predicted hydrogen bonding residues are labeled. (B) AlphaFold3 predicted protein interaction between one Dad (I-Smad) and one Thickveins (type I receptor). Predicted hydrogen bonding residues are labeled. (C) I-Smads function through at least three distinct mechanisms: (1) preventing receptors from phosphorylating R-Smads, (2) binding to R-Smads and preventing formation of R-Smad/Co-Smad complex, and (3) interacting with DNA to counteract TGF-β/BMP transcriptional effects. In neural tissue, I-Smads function preferentially through either the first or second mechanisms. In wing tissue, I-Smads rely on the third mechanism.

Figure S3. Structural prediction of association between Dad and Mad.

AlphaFold3 predicted protein interaction between two Dad molecules (I-Smad) and one Mad (R-Smad). Inset shows a zoomed-in view of the association of one of the Dad molecules with Mad.

Discussion

Structural basis of I-Smad tissue-specific mechanisms

The modes by which I-Smads have been shown to inhibit TGF-β/BMP signaling can be classified into three distinct mechanisms (Fig 7C): (1) interfering with type I receptors, either by direct physical interaction, or by recruitment of effector molecules such as BAMBI, E3 ubiquitin ligases (Smurf1 and Smurf2), or GADD34; (2) preventing R-Smad/Co-Smad complex formation both by direct physical interaction and by promoting ubiquitination; and (3) directly regulating transcription to counteract expression changes of TGF-β/BMP target genes (Miyazawa & Miyazono, 2017). Our current study supports the hypothesis that certain tissue types use specific I-Smad mechanisms preferentially over others.

Confirming previous results, we found that the expression of the Drosophila I-Smad, Dad, in wing primordium results in abnormally small, shriveled wings (Figs 1B and 2C) (Inoue et al, 1998). Seeking to determine which parts of Dad are sufficient for this phenotype, we identified a 24 aa deletion in a putative DNA-binding region within the MH1 domain (ΔDNABD) (Fig 1A). Protein structure prediction reveals a complete disruption of a β-hairpin structure resembling the R-Smad motif, which closely contacts the DNA molecule (Figs 2A and 4A). The expression of this Dad^ΔDNABD construct in wing primordium did not affect wing growth (Figs 1B and 5D), but expression in motor neurons disrupted BMP signaling in this neural tissue (Fig 5), indicating that an alternative mechanism is used in this tissue compared with wing disk. Indeed, Dad with only an MH1 or an MH2 domain (Dad^ΔMH2 and Dad^ΔMH1) disrupted BMP signaling in neural tissue (Fig 4). However, Dad with only MH2 domain (Dad^ΔMH1) was not sufficient to disrupt wing growth (Figs 1B and 4C). Based on these results, we propose that Dad functions primarily through an MH1-mediated mechanism in this wing tissue, possibly direct transcriptional regulation, but through multiple mechanisms in neural tissue (Fig 7C). This assertion is supported by the inability of Dad^ΔMH1 or Dad^ΔDNABD to affect gene expression output (TwitGFP) in motor neurons (Figs 4 and 5), whereas both retain the ability to reduce nuclear pMad accumulation. Future experiments testing whether Dad^ΔMH1 can inhibit noncanonical synaptic pMad accumulation would further distinguish MH1-mediated transcriptional mechanisms from MH2-mediated cytoplasmic mechanisms.

These alternative mechanisms may hold the key to maintaining appropriate levels of TGF-β/BMP signaling in different tissue types. For example, we found that Dad with a cysteine-to-alanine substitution at position 556 (Dad^C556A) was fully effective at disrupting both wing development and BMP signaling outputs in motor neurons (Figs 1B and 6). Previous work shows that this mutation abolishes the ability of Dad to cause loss of germ cells in ovarian tissue (Li et al, 2017). As this residue is reported to serve as a palmitoylation site that promotes association with receptors on the plasma membrane, it appears that Dad functions independently of this membrane association in wing tissue and motor neurons, but relies upon it in germline tissue, providing another example of tissue-specific mechanism usage. Further experiments such as genetic epistasis will be required to determine the exact mechanism(s) used in these tissue types. As Dad overexpression inhibited both canonical and noncanonical BMP pathways in motor neurons, further work can also determine which regions are required for the noncanonical pathway.

Important caveats and limitations

Our approach relies on transgenic overexpression of I-Smad constructs, which has both strengths and limitations. Although overexpression provides experimental tractability and has been widely used to study BMP pathway components (Tsuneizumi et al, 1997; Khalsa et al, 1998; Lee et al, 2016; Galeone et al, 2017), it may not fully recapitulate physiological I-Smad function and could potentially lead to nonphysiological interactions or artifacts. I-Smads function within complex feedback loops that regulate BMP signaling, and supraphysiological expression levels could saturate or bypass regulatory mechanisms that operate under endogenous conditions. Future studies employing CRISPR/Cas9-mediated knock-in approaches (Akiyama et al, 2018) to express mutant I-Smads from endogenous loci would provide more physiologically relevant insights and help validate our findings.

In addition, although the use of a single integration site (VK00033) controls for transcriptional position effects and ensures comparable mRNA levels across constructs, differential protein stability could influence steady-state protein levels and contribute to apparent differences in activity. This is a particular consideration when interpreting constructs that show reduced or absent activity, as we cannot definitively distinguish between loss of function because of removal of critical domains versus reduced protein stability. Despite these caveats, several observations support the validity of our conclusions: (1) constructs showing weak or no activity in one tissue (e.g., Dad^ΔMH1 in wing) produce robust, reproducible phenotypes in other tissues (motor neurons), arguing against simple inadequate expression; (2) the consistent tissue-specific patterns we observe across multiple independent assays (wing morphology, nuclear pMad, TwitGFP, and synaptic pMad) and multiple I-Smads (Dad, Smad6, and Smad7) strongly support context-dependent mechanisms; and (3) our experiments with an alternative wing driver (Rn-Gal4; Fig S2) corroborate our initial findings while revealing additional phenotypic nuance.

Although direct measurement of pMad in wing imaginal disks would provide additional mechanistic insight, the domain-specific differences in wing morphology, coupled with our direct BMP signaling measurements in neural tissue, support tissue-specific mechanisms. To test this model, we analyzed predicted protein structures and observed three characteristics: (1) the DNA-binding loop of Dad shares a similar structure with that of R-Smads (Fig 2A); (2) the structure of I-Smad associating with the type I receptor reveals binding between specific residues; and (3) the structure of I-Smads associating with R-Smads also reveals interaction sites (Fig 7A and B).

Multiple functions performed by distinct structural regions in other proteins

The fact that a single protein can use different structural regions to work through such disparate interactions is remarkable but is not without precedent. Trio, a large, multi-domain protein, exemplifies this versatility. The N-terminal GEF1, central GEF2, and C-terminal protein kinase domains each catalyze distinct reactions that affect different aspects of cellular physiology (Bircher & Koleske, 2021). Although GEF1 activates Rac1 to promote membrane ruffling and cell migration, GEF2 stimulates RhoA to induce stress fiber formation, with the protein kinase domain providing additional signaling capabilities. Another example is the extracellular protein Cerberus, which uses distinct mechanisms to inhibit three different pathways: directly binding to Wnt proteins in the extracellular space, preventing Nodal from interacting with its receptors, and antagonizing BMP signaling through direct protein interaction (Piccolo et al, 1999; Kawano & Kypta, 2003; Tavares et al, 2007; Aykul et al, 2015). Thus, proteins can use different structural motifs to carry out distinct interactions, although further research is necessary to determine whether these different mechanisms are used preferentially depending on the cellular context.

Switching I-Smad mechanisms based on intracellular environment

It is likely that the complement of molecules within a cell determines which mechanism an I-Smad will work through in that cell. Recent studies have begun to elucidate the identities of I-Smad partner molecules. For example, Liu et al (2023) recently identified a gene encoding part of the Integrator complex in a screen for genes required for female germline maintenance (Liu et al, 2023). In germline tissue and S2 cells, the Integrator complex is required for Dad to degrade Tkv. It is possible that certain Integrator complex components are absent from wing disks, and thus, Dad functions through an alternative mechanism. Testing this hypothesis may prove challenging, however, as there are eight genes encoding Integrator complex components (Hao et al, 2008).

Under the assumption that I-Smads evolved from an ancestral Smad gene that encoded a protein with DNA-binding properties, we favor a model wherein Smad6 and Smad7 began differentiating after their duplication in the vertebrate lineage. Considering the higher percent sequence identity between Dad and Smad6, as well as the specificity of both molecules for the BMP branch of TGF-β signaling (Goto et al, 2007), it can be inferred that Smad6 retains more of the ancestral function shared with Dad. Therefore, Smad7 may have evolved additional mechanisms, for example, an increased effectiveness of the MH2-mediated mechanisms (Fig 1B) (Hanyu et al, 2001) and a reduction of the MH1-mediated mechanisms. The structural similarity of the DNA-binding loop between R-Smads, Smad6, and Dad, but not Smad7 (Figs 2A and S1B), supports this hypothesis.

Applications and future directions

These findings indicate that any potential diagnosis or treatment involving I-Smads must consider the mechanism(s) employed by the target tissue type. We demonstrate that to inhibit wing tissue growth, the Drosophila I-Smad functions through an MH1-mediated modality, most likely involving DNA binding. A similar phenomenon may occur during the pathogenesis of diseases where I-Smad dysfunction is implicated. For instance, in colorectal cancer, where SMAD7 variants are associated with increased disease risk, the therapeutic efficacy of I-Smad modulators may depend on whether the target tissue relies primarily on the MH1-mediated transcriptional regulation or the MH2-mediated protein interactions.

The tissue-specific nature of I-Smad mechanisms has profound implications for drug development. Current therapeutic approaches targeting TGF-β/BMP signaling often assume universal mechanisms across tissues. Our findings suggest that interventions may require tissue-specific strategies—enhancing MH1-mediated DNA binding in some contexts while targeting MH2-mediated protein interactions in others. This could explain variable clinical outcomes observed with TGF-β pathway modulators (Colak & Ten Dijke, 2017; Melisi et al, 2021; Yap et al, 2021) and suggest a need to reassess existing therapeutic strategies.

Future work must systematically map which mechanisms operate in clinically relevant tissues. It will be critical to determine whether intestinal epithelium, cardiac tissue, and other disease-relevant cell types preferentially employ specific I-Smad mechanisms, and whether these mechanisms shift during pathological states. In addition, the evolutionary conservation of tissue-specific mechanism usage across species requires investigation to validate the translational relevance of our findings. Our versatile in vivo platform provides a powerful tool for further mechanistic dissection. Domain swap experiments between Dad, Smad6, and Smad7 could reveal the molecular determinants of tissue specificity, whereas systematic analysis of I-Smad partner molecules in different tissues may identify cellular factors that dictate choice of mechanisms. Ultimately, extending these structure–function principles to vertebrate systems will be essential for developing the next generation of therapeutics targeting TGF-β/BMP signaling disorders.

Materials and Methods

Molecular biology and transgenesis

Plasmids were created using standard molecular biology methods. All plasmid sequences were verified using whole plasmid nanopore sequencing (Eurofins Genomics). Injections were performed by Rainbow Transgenics, Inc. All constructs were inserted into the VK00033 locus on the third chromosome. Insertions were verified by sequencing. Complete lists of plasmids, primers, and oligonucleotides used in this study are included below.

Complete list of plasmids used for this study.

Plasmid	Creation	Source	Notes
pUAST-attB	N/A	Stock 1419; https://dgrc.bio.indiana.edu//stock/1419; RRID:DGRC_1419; DGRC	Destination vector for transformation
pBSPPC	N/A	pBSPPc was a gift from Cuiqing Ma (plasmid #116918; https://n2t.net/addgene:116918; RRID:Addgene_116918; Addgene) (Xu et al, 2013)	Intermediate vector
CS2 Smad6	N/A	CS2 Smad6 was a gift from Joan Massague (plasmid #14960; https://n2t.net/addgene:14960; RRID:Addgene_14960; Addgene) (Hata et al, 1998)	Source of Smad6 CDS
TFORF0111	N/A	TFORF0111 was a gift from Feng Zhang (plasmid #141518; https://n2t.net/addgene:141518; RRID:Addgene_141518; Addgene) (Joung et al, 2023)	Source of Smad7S CDS
pBSPPc-Smad6	Insert from CS2 Smad6 cloned into pBSPPc using BamHI/XbaI sites	Current study	​
pUAST-attB-Smad6	Insert from pBSPPc-Smad6 cloned into pUAST-attB using XhoI/XbaI sites	Current study	Transformation vector for Smad6
pBSPPc-NMC	Additional cloning sites added to pBSPPc by inserting oligonucleotide into PstI/BamHI sites	Current study	To accommodate Smad7S insert
pBSPPc-NMC-Smad7S	Insert from TFORF0111 cloned into pBSPPc-NMC using BmtI/SpeI sites	Current study	​
pUAST-attB-Smad7S	Insert from pBSPPc-NMC-Smad7S (BamHI/XbaI) cloned into pUAST-attB using XhoI/XbaI sites	Current study	Transformation vector for Smad7S
LD47465	N/A	LD47465 (Stock 1333567; https://dgrc.bio.indiana.edu//stock/1333567; RRID:DGRC_1333567; DGRC)	Source for Dad CDS
pUAST-attB-Dad	PCR product from LD47465 with XhoI/XbaI sites added cloned into pUAST-attB	​	Transformation vector for UAS-Dad
PCMV5-Smad7-HA	N/A	pCMV5-Smad7-HA was a gift from Jeff Wrana (plasmid #11733; https://n2t.net/addgene:11733; RRID:Addgene_11733; Addgene) (Hayashi et al, 1997)	Source for Smad7L CDS
PCMV5-Smad7	Deleted HA tag from pCMV5-Smad7-HA with PCR	Current study	​
pUAST-attB-Smad7L	Insert from pCMV5-Smad7 cloned into pUAST-attB using EcoRI/XbaI sites	Current study	Transformation vector for UAS-Smad7L
pUAST-attB-Dad-ΔMH1	Deleted	Current study	Transformation vector for ΔMH1
pUAST-attB-Dad-ΔDNA-BD	​	Current study	Transformation vector for ΔDNABD
pUAST-attB-Dad-ΔMH2	​	Current study	Transformation vector for ΔMH2
pUAST-attB-Dad-ΔTrimer	​	Current study	Transformation vector for ΔTrimer

Complete list of oligonucleotides used for this study.

Primer/oligonucleotide	Sequence (5′-3′)	Notes
pBSPPc-NMC-sense	GCATGCTAGCTTAAGATCTG	Hybridized to create PstI/BamHI overhangs to clone into pBSPPc
pBSPPc-NMC-antisense	GATCCAGATCTTAAGCTAGCATGCTGCA
pOT2-Dad-XhoI-F	GCTCGAGATGATATTCCCAAGAGAAAAGAAGGT	To obtain Dad CDS from LD47465 and add XhoI/XbaI sites
pOT2-Dad-XbaI-R	CTCTAGATCACCGCAGATGACTAAAGTGA
Smad7-HA-Del-F	TAATCTAGAGGATCCCG	To delete HA tag from pCMV5-Smad7-HA
Smad7-HA-Del-R	CCGGCTGTTGAAGATG
Δ MH1-F	GAAACAGAATCCCCAAC	To delete MH1 domain from pUAST-attB-Dad
Δ MH1-R	TGGAATCGTAAGTGTGG
Δ DNA-BD-F	AAGGAACTGAAGCGAC	To delete a putative DNA-binding region from pUAST-attB-Dad
Δ DNA-R	GCACGGTATCAGGATG
Δ MH2-F	CATCTGCGGTGATCTAG	To delete MH2 domain from pUAST-attB-Dad
Δ MH2-R	CACCTGACTGTTGATG
Δ Trimer-F	GTGGCCAGTGAGGTG	To delete a putative trimer interface region from pUAST-attB-Dad
Δ Trimer-R	GATATTGACTGCATTCGTTTTG
C556A-F1	TATCATGGGCGCGCCCTGCTGGC	To create pUAST-attB-DadC556A by combining two PCR products using the Gibson assembly
C556A-R1	GTTGACCAGCTGCCGCCATC
C556A-F2	GATGGCGGCAGCTGGTCAAC
C556A-R2	GCCAGCAGGGCGCGCCCATGATA
pUAST-attB-Dad-1F	GGAATTCGTTAACAGATCTGCGG	To verify full-length Dad construct
pUAST-attB-Dad-1R	CAGTCAGCACGTCCAGAGC
pUAST-attB-Dad-2F	TTTCGGAAATGCTGCGGCG
pUAST-attB-Dad-2R	AACACCGCCGTACCAACG
pUAST-attB-Dad-3F	ATGTGCTGCAATCCGCTGC
pUAST-attB-Dad-3R	CCAAAAGCGCGTCATACGGT
pUAST-attB-Dad-4F	GGACAGCATGTGCCTACGTG
pUAST-attB-Dad-4R	TGAAGGAACCTTACTTCTGTGGTG
pUAST-attB-Smad6-1F	TATCGAATTCCTGCAGCCCG	To verify if Smad6 insertion was successful
pUAST-attB-Smad6-1R	CCGCCGGCGCGCCCCGGGACGCCAG
pUAST-attB-Smad6-2F	GAGCTCCCTGCTGGACGTG
pUAST-attB-Smad6-2R	CCGACGGCCCTACCGTGTGCTGCAA
pUAST-attB-Smad6-3F	CGCCGACCTCCGCCT
pUAST-attB-Smad6-3R	AGCAGCGCAGCGAGTCGGTGCGGCG
pUAST-attB-Smad6-4F	CAGACGCCAGCATGTCTCC
pUAST-attB-Smad6-4R	TGGTCTAGAGGATCTTTGTGAAGGA
pUAST-attB-Smad7-1F	GGGAATTGGGAATTCGCTGC	To verify if Smad7 insertion was successful
pUAST-attB-Smad7-1R	GAAGATCAACCCCGAGCTGG
pUAST-attB-Smad7-2F	GGCCGGATCTCAGGCATT
pUAST-attB-Smad7-2R	TTTGTGAAGGAACCTTACTTCTGTG
pUAST-attB-Dad-ΔMH1-1F	GGAATTCGTTAACAGATCTGCGG	To verify if MH1 domain deletion was successful
pUAST-attB-Dad-ΔMH1-1R	CGACGGAAGCGATCTGGAT
pUAST-attB-Dad-ΔMH1-2F	GTGCGACCGCTGCTGTA
pUAST-attB-Dad-ΔMH1-2R	CAGTCAGGTGTGGTGCCAAA
pUAST-attB-Dad-ΔMH1-3F	CATTTCGAACATCTACAAGCCG
pUAST-attB-Dad-ΔMH1-3R	AGGGCTGAACTTCTGAGCAT
pUAST-attB-Dad-ΔMH1-4F	TCTGGACCGAGTGTGCAAAG
pUAST-attB-Dad-ΔMH1-4R	TTGTTGAAGGAACCTTACTTCTGTGG
pUAST-attB-Dad-ΔDNABD-L-1F	ATTGGGAATTCGTTAACAGATCTGCGGC	To verify if DNA-binding domain deletion was successful
pUAST-attB-Dad-ΔDNABD-L-1R	CTGGTTACAGCTGCGCAAGCTCCTG
pUAST-attB-Dad-ΔDNABD-L-2F	CCAAATCACATGGATGTGTTGCCGCC
pUAST-attB-Dad-ΔDNABD-L-2R	GCGATGTCGTGGAACCCACCACCAC
pUAST-attB-Dad-ΔDNABD-L-3F	CTCAAGCGCAAGCAGAGAAACGAAC
pUAST-attB-Dad-ΔDNABD-L-3R	TAACATCAACAGTCAGGTGTGGTGC
pUAST-attB-Dad-ΔDNABD-L-4F	ACCAGCATTTCGAACATCTACAAGC
pUAST-attB-Dad-ΔDNABD-L-4R	TTGTGAAGGAACCTTACTTCTGTGG
pUAST-attB-Dad-ΔMH2-1F	AATTCGTTAACAGATCTGCGGC	To verify if MH2 domain deletion was successful
pUAST-attB-Dad-ΔMH2-1R	GGAAGCGATCTGGATCAGGATC
pUAST-attB-Dad-ΔMH2-2F	GCTGTACCGCTCCTGGTTAC
pUAST-attB-Dad-ΔMH2-2R	GTGCATCCTGATACCGTGCA
pUAST-attB-Dad-ΔMH2-3F	CGCCGACCAAGACGCA
pUAST-attB-Dad-ΔMH2-3R	AGAGGATCTTTGTGAAGGAACCT
pUAST-attB-Dad-C556A-1F	GAATTCGTTAACAGATCTGCGGC	To verify if C556A substitution deletion was successful
pUAST-attB-Dad-C556A-1R	TACAGCTGCGCAAGCTCCTG
pUAST-attB-Dad-C556A-2F	CACATGGATGTGTTGCCGC
pUAST-attB-Dad-C556A-2R	CCTACCTGCAGTGCATCCTG
pUAST-attB-Dad-C556A-3F	AAGAGTCGCCTAGATCCGCC
pUAST-attB-Dad-C556A-3R	TCTCTACGAGTCGGTTACCACT
pUAST-attB-Dad-C556A-4F	CAGCGCTCAGCACGTGGA
pUAST-attB-Dad-C556A-4R	TTGTGAAGGAACCTTACTTCTGTGG

Drosophila stocks and husbandry

Animals were reared at 25°C on standard cornmeal molasses medium. To control for crowding, crosses were set up with eight females crossed with six males, and adults were passed to a new vial every 3–4 d. For expression in motor neurons, OK371-Gal4 (Mahr & Aberle, 2006) was used. For expression in wing primordium, Ap-Gal4 or Rn-Gal4 was used. A complete list of the strains used for this study is included below.

Drosophila stocks used for this study.

Abbreviation	Genotype	Source
w1118	w[1118]	BDSC: 2376
OK371	w[1118]; P{w[+mW.hs] = GawB}VGlut1[OK371]	BDSC: 26160
Ap-Gal4	w[1118]; P{y[+t7.7] w[+mC] = GMR42A06-GAL4}attP2	BDSC: 41425
DadRNAi	y[1] v[1]; P{y[+t7.7] v[+t1.8] = TRiP.JF02133}attP2	BDSC: 26235
TwitGFP	w[1118]; Mi{y[+mDint2] = MIC}twit[MI06552]	BDSC: 41449
Rn-Gal4	w[*]; P{w[+mW.hs] = GawB}rn[GAL4-5] P{w[+mC] = UAS-GFP.U}3/TM3, Sb[1] Ser[1]	BDSC: 78345
UAS-Dad	w[1118]; +; P{UAS-Dad}VK00037	Current study
UAS-Smad6	w[1118]; +; P{UAS-SMAD6}VK00037	Current study
UAS-Smad7	w[1118]; +; P{UAS-SMAD7.1}VK00037	Current study
UAS-Smad7S	w[1118]; +; P{UAS-SMAD7.3}VK00037	Current study
ΔMH1	w[1118]; +; P{UAS-DadΔMH1}VK00037	Current study
ΔMH2	w[1118]; +; P{UAS-DadΔMH2}VK00037	Current study
ΔDNABD	w[1118]; +; P{UAS-DadΔDNABD}VK00037	Current study
ΔTrimer	w[1118]; +; P{UAS-DadΔTrimer}VK00037	Current study
DadC556A	w[1118]; +; P{UAS-DadC556A}VK00037	Current study

Immunostaining and microscopy

Nervous system imaging and immunostaining were performed as described previously (Brown et al, 2019). Briefly, wandering third instar larvae were dissected in hemolymph-like-3.1 (HL3.1) saline (Feng et al, 2004), fixed in 4% PFA diluted with 1XPBS for 20 min, and washed several times in wash buffer (1XPBS with 0.3% TX-100). Primary antibodies were used at the following concentrations: rabbit anti-pSmad1/5 (Cat# 9516, RRID:AB_491015; Cell Signaling Technology) at 1:500, rabbit anti-pMad PS1 (Cat# pMad, RRID:AB_2617125; Carl-Henrik Heldin Göteborg University) at 1:500, chicken anti-GFP (Cat# ab13960, RRID: AB_300798; Abcam) at 1:1,000, mouse anti-synaptotagmin (Cat# 3H2 2D7, RRID:AB_528483; DSHB) at 1:500, and rat anti-Elav (Cat# 7E8A10, RRID: AB_528218; DSHB) at 1:500. Secondary antibodies were used at the following concentrations: goat anti-chicken Alexa Fluor 488 (Cat# A-11039, RRID:AB_2534096; Thermo Fisher Scientific) at 1:500, goat anti-rabbit Alexa Fluor 568 (Cat# A-11011, RRID: AB_143157; Thermo Fisher Scientific) at 1:500, goat anti-rat Alexa Fluor 647 (Cat# A-21247, RRID: AB_141778; Thermo Fisher Scientific), and goat anti-HRP Alexa Fluor 488 (Cat# 123-545-021, RRID:AB_2338965; Jackson ImmunoResearch Labs). Samples were processed in parallel such that all incubation times were identical. Brains were mounted in SlowFade Diamond (Cat# S36972; Thermo Fisher Scientific), and filets were mounted in ProLong Diamond (Cat# P36961; Thermo Fisher Scientific). Images were acquired with a Zeiss LSM 880 or Nikon Ti-2 confocal microscope using a 20X or 63X objective. Images were acquired and analyzed while blinded to genotype.

Wing dissection, imaging, and analysis

Newly eclosed (between 8 and 24 h) adults were immediately fixed in 100% ethanol. Wings were dissected and transferred to 70% ethanol to rehydrate. Wings were then arranged on a microscope slide and covered with ProLong Diamond mounting medium (Cat# P36965; Thermo Fisher Scientific). A coverslip was placed on top, and the slide was allowed to cure overnight and covered with a small weight to flatten the preparation. Edges were sealed with nail polish. Slides were imaged with a compound microscope using a 20X objective. Area was determined by manually tracing the border of the wing image, including the hinge, but not the surrounding tissue, using the polygon tool in ImageJ, and quantifying the area of the resulting polygon. Images were acquired and analyzed while blinded to genotype. All representative images show female wings. Representative images shown in all figures except Fig S1 are from data quantified in Fig 1B.

Structural prediction and analyses

Full-length protein sequences were obtained from the National Center for Biotechnology Information (NCBI) protein database (https://www.ncbi.nlm.nih.gov/protein/) and verified for completeness and accuracy. Protein domains and putative trimerization/DNA-binding sites were identified using InterPro (https://www.ebi.ac.uk/interpro/) (Blum et al, 2025). Domain boundaries for deletion constructs were defined based on InterPro domain predictions and structural alignments with characterized Smad proteins. Complete amino acid sequences were submitted to AlphaFold3 server (https://alphafoldserver.com/) using default parameters (Abramson et al, 2024). Predicted structures were downloaded from AlphaFold3 in PDB format. Structural visualization was performed using UCSF ChimeraX (Meng et al, 2023). Confidence scores provided by AlphaFold3 were used to assess the reliability of predicted structural regions, with high-confidence regions (pLDDT > 70) considered reliable for structural interpretation. To determine I-Smad aligned residues, CLUSTALW was used (https://www.genome.jp/tools-bin/clustalw) with default parameters for multiple alignments.

Data analysis and statistics

Pixel intensity was measured from maximum projection confocal micrographs using a custom ImageJ script (AIABS) as previously described (Brown et al, 2019). Intensity measurements were performed while blinded to genotype. All data were subjected to outlier analysis with the Z-test method with threshold ≥ 2 STDEV from the mean. Normality of data points for each genotype was tested using a Shapiro–Wilk test. Similarity of variances was tested using Levene’s test. For datasets with multiple genotypes where measurements from each genotype were normally distributed and variances were similar, differences across genotypes were tested by one-way ANOVA. As a post hoc test, when multiple samples were compared only with control, Dunnett’s test was performed (Fig 1B). For data that were not normally distributed or had unequal variances, differences across genotypes were tested by the Kruskal–Wallis test. Pairwise differences were tested by post hoc Dunn’s test with the Bonferroni correction (Figs 2E, 4B, 5C, and 6B). For comparing two groups with normal distributions, a t test was used (Fig 3B). Data were analyzed with IBM SPSS Statistics (version 27) and R (version 4.3.2; R Core Team, 2023). Graphs were created with the ggplot2 R package (Wickham, 2016). Scripts used for statistical analysis and plot creation are available upon request. Cutoff for statistical significance was P < 0.05. All statistically significant differences were indicated by * as follows: ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05.