SMAD
Consensus: GTCT / AGAC (palindrome GTCTAGAC); also the 5-bp GC motif GGC(GC/CG)
SMAD proteins are the intracellular effectors of TGF-beta family signaling. Their MH1 domain inserts a beta-hairpin into the major groove and reads just four bases, GTCT or its reverse complement AGAC, often written as the CAGAC box or as the 8-bp palindrome GTCTAGAC. SMAD3, SMAD4 and the BMP-activated SMAD1/5/9 also bind a separate 5-bp GC-rich element, GGCGC or GGCCG. Both elements are far too short to be informative alone, and genome-wide SMAD occupancy is set mainly by whichever lineage-determining factor is present, so most SMAD ChIP peaks are better explained by a partner's motif than by an SBE. This is a family where a motif hit needs corroboration more than almost any other.
The factors
Eight human SMADs fall into three groups. The receptor-regulated R-SMADs are SMAD2 and SMAD3 (phosphorylated by TGF-beta and activin receptors) and SMAD1, SMAD5 and SMAD9/SMAD8 (phosphorylated by BMP receptors). SMAD4 is the shared co-SMAD. SMAD6 and SMAD7 are inhibitory and are not sequence-specific activators. DNA contact is made by the N-terminal MH1 domain, a compact domain with a bound zinc ion whose 11-residue beta-hairpin inserts into the major groove 9741623; the C-terminal MH2 domain mediates trimerization, so the functional unit on DNA is normally a trimer of two R-SMADs plus SMAD4, giving two or three DNA-contacting hairpins 29234012. One paralogue is a special case: wild-type SMAD2 carries a 30-residue insert encoded by exon 3 immediately before the hairpin that blocks direct DNA binding 9873005. Full-length SMAD2 reaches DNA through partners, most clearly the forkhead protein FOXH1/FAST1 9702197,9873005; the naturally occurring SMAD2 delta-exon-3 splice form does bind DNA, like SMAD3 9873005.
The motif
The classical Smad-binding element is 4 bp, GTCT, read on the other strand as AGAC and usually quoted with a flanking base as CAGAC ("CAGA box") or CAGACA. Zawel and colleagues showed that SMAD3 and SMAD4 both recognize an identical 8-bp palindrome, GTCTAGAC 9660945, and the Smad3 MH1 crystal structure explained why: the hairpin reads GTCT and two hairpins from a trimer can occupy the two halves of the palindrome 9741623. Real promoters usually contain tandem CAGA repeats rather than a perfect palindrome; the TGF-beta response element of the PAI-1/SERPINE1 promoter is the canonical example 9606191. Separately, SMADs bind a 5-bp GC-rich element with consensus GGC(GC) or GGC(CG). This was first described as a BMP-responsive element for SMAD1/5 10679014, but structural work showed that SMAD3 and SMAD4 bind the same 5GC motifs, and that these motifs occur in clusters in SMAD-bound regions genome-wide 29234012. So the family has two chemically distinct consensus sequences, not one; a database entry showing GTCTAGAC and one showing a GC-rich string can both be correct. The data for this family reflects that split, and its cohesion (0.76) is the lowest in this batch. The CAGAC-type and 5GC-type models do not interconvert, and paralogues are not separable from sequence within either type. The CIS-BP SMAD9 entry is a GC-rich model transferred from a Drosophila protein at 91% MH1 identity, not a human measurement.
Where it acts
SMAD sites are distributed across promoters and enhancers with no CpG-island preference for the CAGAC form, while 5GC clusters are by definition GC-rich and promoter-biased. What matters more is that SMAD occupancy is almost entirely conditional. Smad3 ChIP-seq in three different cell types recovers three largely non-overlapping site sets, each co-occupied by that cell type's lineage-determining factor: OCT4 in embryonic stem cells, MYOD1 in myotubes, PU.1 in pro-B cells 22036565. In other systems the recurring partners are FOXH1 and other forkhead proteins, RUNX1/2, AP-1 (JUN/FOS), TEAD, and the repressors TGIF1, SKI and SnoN. Named composite elements include the activin-responsive element of the Mix.2 promoter (a forkhead site next to a SMAD site) 9702197,9873005 and the BMP-responsive elements of the Id1 and Smad7 promoters. In practice a SMAD peak set looks like the partner's peak set with a weak SBE enrichment layered on top.
Biology and disease
TGF-beta/BMP signaling through SMADs controls epithelial growth arrest, extracellular matrix production, epithelial-mesenchymal transition, regulatory T cell induction, bone and cartilage formation, and left-right and dorsoventral patterning. Smad3-null mice die within months from impaired mucosal immunity with defective T cell responses to TGF-beta 10064594. SMAD4 was identified as DPC4, a candidate tumor suppressor at 18q21.1 that was homozygously deleted in 25 of 84 pancreatic carcinomas (about 30%), with potentially inactivating mutations in 6 of 27 of the others 8553070; germline SMAD4 mutations cause juvenile polyposis and, with an overlapping phenotype, hereditary hemorrhagic telangiectasia 9582123,15031030. SMAD4 and SMAD2/SMAD3 are also recurrently mutated in colorectal cancer, and loss of SMAD signaling converts TGF-beta from a growth suppressor to a driver of invasion and immune evasion 37714133. SMAD9 mutations cause a form of pulmonary arterial hypertension, consistent with the BMP9/BMPR2 axis 19211612.
Reading a hit
Treat an isolated SBE match as essentially no evidence. GTCT or AGAC occurs about once every 128 bases on either strand, so a 4-bp core appears tens of millions of times in the human genome; even the 8-bp palindrome GTCTAGAC is expected tens of thousands of times by chance. A SMAD hit becomes meaningful only in one of three forms: several CAGA repeats within a short window, a cluster of 5GC motifs, or an SBE within a few base pairs of a validated partner site (a forkhead, RUNX, AP-1 or TEAD motif). The second problem is that ChIP-derived SMAD motifs in these collections are contaminated by indirect binding almost by construction, since SMAD complexes are recruited by lineage factors. Every HOCOMOCO SMAD2 model here rests on ChIP-seq only (evidence code P), and wild-type SMAD2 cannot bind DNA directly 9873005, so those matrices describe the composition of SMAD2 peaks rather than SMAD2's own specificity. Finally, do not use a hit to name a paralogue. CAGAC does not mean TGF-beta rather than BMP, because SMAD1/5 bind CAGAC-type sites too and SMAD3/4 bind GC-type sites; the original "TGF-beta uses CAGAC, BMP uses GC" division does not survive the structural and genome-wide data 29234012,33510867.
How this was worked out
- 1998mediumBinding-site selection from random oligonucleotides, reporter assays, SMAD4-null cells9660945
SMAD3 and SMAD4 each recognize the same 8-bp palindrome, GTCTAGAC, and tandem copies confer TGF-beta responsiveness that is lost in SMAD4-deleted cells, establishing SMADs as sequence-specific DNA-binding factors.
- 1998mediumPromoter mutagenesis, EMSA with recombinant proteins and nuclear extracts, reporter assays9606191
Three CAGA boxes in the PAI-1/SERPINE1 promoter are bound directly by SMAD3 and SMAD4, are required for TGF-beta induction, and respond to TGF-beta and activin but not BMP; recombinant SMAD1 and SMAD2 do not bind them.
Still open then: What, if anything, the BMP-activated SMADs bind directly.
The 2.8 A structure of the SMAD3 MH1 domain on DNA shows that base-specific recognition is made entirely by a conserved 11-residue beta-hairpin in the major groove.
Wild-type SMAD2 does not bind DNA directly because of the exon 3 insert before the hairpin; it reaches target promoters with SMAD4 through the forkhead partner FAST/FOXH1, whereas the delta-exon-3 isoform and SMAD3 bind DNA themselves.
BMP-activated SMAD1 with SMAD4 binds the GC-rich GCCGnCGC element first defined for Drosophila Mad, weakly and only with multiple repeats, and GCCG reporters respond to BMP but not TGF-beta, supporting a split between GC-type BMP sites and CAGA-type TGF-beta sites.
Still open then: Whether the GC versus CAGA division reflected intrinsic specificity or the reporter contexts tested.
SMAD3 ChIP-seq in embryonic stem cells, myotubes and pro-B cells gives largely distinct site sets, each co-occupied with and dependent on that cell's lineage-determining factor (OCT4, MYOD1, PU.1); inducing MYOD1 redirects SMAD3.
Structures of SMAD3 and SMAD4 on goosecoid promoter DNA define the 5-bp GC motif GGC(GC)|(CG) as a site for TGF-beta-activated, BMP-activated and common SMADs alike, and 5GC clusters are frequent in SMAD-bound regions genome-wide. This overturns the view that CAGAC is the TGF-beta site and GC-rich elements the BMP site.
SMAD5 and SMAD8 MH1 structures show all MH1 domains bind 5GC motifs similarly, but SMAD2/3/4 MH1 domains bind as monomers while SMAD1/5/8 form helix-swapped dimers, and site clustering and spacing in ChIP-seq peaks differ between BMP- and TGF-beta-responsive genes.
Still open then: Whether monomer-versus-dimer propensity actually drives pathway-specific target selection in cells.
Still unresolved
- How TGF-beta and BMP SMADs reach different targets is unresolved now that both bind CAGAC-type and 5GC sites: one proposal is that MH1 monomer (SMAD2/3/4) versus helix-swapped dimer (SMAD1/5/8) propensity sets site spacing and clustering, the alternative is that partner factors such as FOXH1 or lineage-determining factors choose the sites and the SMAD motif contributes little.
- The ChIP-derived HOCOMOCO SMAD2, SMAD3 and SMAD4 models (for example tGTCTGdCACc and AGGTGncAGACA) do not match the in vitro palindrome GTCTAGAC, and it is not settled whether they describe composite SMAD-partner sites or simply peak composition shaped by recruiting factors.
The lowest-confidence family in this batch. A 4-bp SBE is not a finding; even the 8-bp palindrome is common by chance. The family has two unrelated consensus sequences (CAGAC-type and 5GC-type), so conflicting database entries may both be right, and paralogues are not separable within either. Most SMAD ChIP peaks are explained by a partner's motif, not by an SBE, and all HOCOMOCO SMAD2 models here are ChIP-only while full-length SMAD2 cannot bind DNA directly. The CIS-BP SMAD9 model is transferred from Drosophila. Use SMAD hits only as supporting evidence alongside a partner site or experimental occupancy.
References
- Shi Y, Wang YF, Jayaraman L, et al. Crystal structure of a Smad MH1 domain bound to DNA: insights on DNA binding in TGF-beta signaling. Cell 1998;94:585-94. PMID 9741623
- Zawel L, Dai JL, Buckhaults P, et al. Human Smad3 and Smad4 are sequence-specific transcription activators. Mol Cell 1998;1:611-7. PMID 9660945
- Dennler S, Itoh S, Vivien D, et al. Direct binding of Smad3 and Smad4 to critical TGF beta-inducible elements in the promoter of human plasminogen activator inhibitor-type 1 gene. EMBO J 1998;17:3091-100. PMID 9606191
- Martin-Malpartida P, Batet M, Kaczmarska Z, et al. Structural basis for genome wide recognition of 5-bp GC motifs by SMAD transcription factors. Nat Commun 2017;8:2070. PMID 29234012
- Kusanagi K, Inoue H, Ishidou Y, et al. Characterization of a bone morphogenetic protein-responsive Smad-binding element. Mol Biol Cell 2000;11:555-65. PMID 10679014
- Ruiz L, Kaczmarska Z, Gomes T, et al. Unveiling the dimer/monomer propensities of Smad MH1-DNA complexes. Comput Struct Biotechnol J 2021;19:632-646. PMID 33510867
- Mullen AC, Orlando DA, Newman JJ, et al. Master transcription factors determine cell-type-specific responses to TGF-β signaling. Cell 2011;147:565-76. PMID 22036565
- Labbé E, Silvestri C, Hoodless PA, et al. Smad2 and Smad3 positively and negatively regulate TGF beta-dependent transcription through the forkhead DNA-binding protein FAST2. Mol Cell 1998;2:109-20. PMID 9702197
- Yagi K, Goto D, Hamamoto T, et al. Alternatively spliced variant of Smad2 lacking exon 3. Comparison with wild-type Smad2 and Smad3. J Biol Chem 1999;274:703-9. PMID 9873005
- Yang X, Letterio JJ, Lechleider RJ, et al. Targeted disruption of SMAD3 results in impaired mucosal immunity and diminished T cell responsiveness to TGF-beta. EMBO J 1999;18:1280-91. PMID 10064594
- Hahn SA, Schutte M, Hoque AT, et al. DPC4, a candidate tumor suppressor gene at human chromosome 18q21.1. Science 1996;271:350-3. PMID 8553070
- Massagué J, Sheppard D. TGF-β signaling in health and disease. Cell 2023;186:4007-4037. PMID 37714133
- Howe JR, Roth S, Ringold JC, et al. Mutations in the SMAD4/DPC4 gene in juvenile polyposis. Science 1998;280:1086-8. PMID 9582123
- Gallione CJ, Repetto GM, Legius E, et al. A combined syndrome of juvenile polyposis and hereditary haemorrhagic telangiectasia associated with mutations in MADH4 (SMAD4). Lancet 2004;363:852-9. PMID 15031030
- Shintani M, Yagi H, Nakayama T, et al. A new nonsense mutation of SMAD8 associated with pulmonary arterial hypertension. J Med Genet 2009;46:331-7. PMID 19211612
20 motifs in this family — JASPAR 4 · HOCOMOCO 8 · CIS-BP 4 · Vierstra 4
| motif | database | consensus | bits |
|---|---|---|---|
| SMAD3 (MA0795.1) | JASPAR | cGTCTAGACA | 15.6 |
| M03482_2.00 | CIS-BP | cGTCTAGACA | 15.6 |
| SMAD3.H14CORE.0.S.BSB | HOCOMOCO | GTCTAGACA | 13.9 |
| SMAD5 (MA1557.1) | JASPAR | GTCTAGACA | 13.5 |
| M05743_2.00 | CIS-BP | GTCTAGACA | 13.5 |
| AC0323:SMAD:SMAD | Vierstra | GTCTAGAC | 13.3 |
| M05889_2.00Mus musculus 1 | CIS-BP | CtAGACa | 11.7 |
| Smad4 (MA1153.2) | JASPAR | tGTCTaG | 11.5 |
| SMAD4.H14CORE.1.P.BPB | HOCOMOCO | tGTCTGdCACc | 11.1 |
| SMAD2 (MA1964.2) | JASPAR | CCAGAC | 10.6 |
| AC0597:SMAD:SMAD | Vierstra | gGTGhcAGACa | 10.6 |
| SMAD2.H14CORE.1.P.BPB | HOCOMOCO | CTGTCTG | 10.4 |
| SMAD3.H14CORE.1.P.BPB | HOCOMOCO | CCAGACAg | 10.4 |
| SMAD2.H14CORE.0.P.BPB | HOCOMOCO | tGTCTGdCACc | 10.1 |
| SMAD3.H14CORE.2.P.CPC | HOCOMOCO | AGGTGncAGACA | 9.9 |
| AC0371:SMAD:['SMAD/NF-1_DNA-binding_domain_factors'] | Vierstra | GTCTGG | 9.9 |
| SMAD5.H14CORE.0.M.CMC | HOCOMOCO | CGTCT | 9.6 |
| SMAD4.H14CORE.0.P.BPB | HOCOMOCO | tGTCTGg | 8.6 |
| M06446_2.00Drosophila melanogaster 0.91 | CIS-BP | GCGhCGcCmG | 8.6 |
| AC0598:SMAD:SMAD | Vierstra | CTghCacCt | 7 |
HOCOMOCO tags read as assay + quality grade: P ChIP-seq, S HT-SELEX, M methyl-HT-SELEX, G GHT-SELEX, I SMiLE-seq, B PBM; grade A means two or more assay types agreed. A P-only matrix is a record of where the protein was crosslinked, which need not be a sequence it binds itself.
The factors, elsewhere
These entries are written from the primary literature and every citation is checked against PubMed, but this one can still be out of date, too confident, or simply wrong — especially where a motif’s published site and its database matrices disagree. Corrections and additions are welcome, and they are the fastest way to make this page better for everyone.