SKI is a protein coding gene that encompasses about 81.9 kb of genomic DNA and comprises 7 exons. It is located on the short arm of human chromosome 1 (1p36.33-p36.32).

SKI has 4 transcripts of which only one is protein coding (SKI-201, ENST00000378536.5), 237 orthologues, 3 paralogues ( SKIL, SKOR1, SKOR2) and is associated with 39 phenotypes. So far only a few factors are identified, which are involved in the regulation of SKI gene expression. In murine cardiac cells, serum-response factor ( SRF) is able to interact with CArG and CArG-like boxes in the 5UTR and induces Ski expression (Zhang et al., 2005). Moreover, PPARD induces Ski expression by binding to a direct repeat-1 (DR1) response element (-865~-853) in the Ski promoter region in rat skin fibroblasts (Li et al., 2012). Also as part of a negative feedback loop, retinoic acid seems to induce SKI gene expression in Xenopus embryos and in immortalised human keratinocytes (HaCaT) (Melling et al., 2013). In human leukemic cells, SKI expression is further dependent on MYB, which interacts with MYB binding sites in the SKI regulatory region (Frech et al., 2018). It was also reported that SKI expression is inhibited by hypermethylation of the SKI promoter region in human lung cancer cells (Xie et al., 2017). Laidlaw et al. (2020) showed in a mouse model that the transcription factor Hhex induces Ski expression in memory B cells. Post-transcriptionally SKI mRNA levels are also regulated by several microRNAs (miRs), like MIR155, MIR29A, MIR21, MIR-127-3p, MIR1908, MIR17-5p, MIR34A, MIR93, MIR339 all interacting with the 3UTR (s. Figure 1).

The SKI family consists of SKI, SKIL (alias SnoN, SnoN2, SnoI, SnoA), SKOR1 (alias CORL1, FUSSEL15, LBXCOR1), SKOR2 (alias CORL2, FUSSEL18) DACH1, DACH2.

SKI consists of 728 amino acids (aa) and contains three main domains. The DHD (dachshund homology domain, aa 91-192), also characterising the SKI family, is highly conserved and located near the N-terminus. It also contains Leu127, which is important for the stabilization of the DHD domain and thereby its interaction with NCOR and SMAD2 / SMAD3 (Ueki & Hayman, 2003, Wilson et al., 2004). The second conserved domain is the SAND-L (SAND-like, aa 219-312) domain. Both domains show homologies with DNA-interacting proteins but have lost the ability to bind DNA. They serve as domains for the interaction with transcription factors, transcriptional coregulators, kinases or other factors (s. Figure 2)(Bonnon & Atanasoski, 2012, Tecalco-Cruz et al., 2018). Via the interaction with other factors, SKI is recruited to the DNA and can act as co-regulator. The C-terminal part is more variable and comprises two coiled-coil (CC) regions mainly enabeling SKI homodimerization and heterodimerization with its paralog SnoN (Heyman & Stavnezer, 1994). The N-terminus further contains a proline-rich part (aa 61-89). Also, the region responsible for SKIs transformational activity lies in the N-terminal part (1-304) (Teclaco-Cruz, 2018). The degradation of SKI is regulated via PTM. For degradation via the ubiquitin-proteasome system (UPS) SKI is polyubiquitinylated at so far unidentified lysine residues by RNF111 (the E3 ligase ARKADIA). For the polyubiquitination SKI needs not only to interact with ARKADIA via aa 211-490 but also with SMAD2/3 (Nagano et al., 2007, Nagano et al., 2010). SKI is further a phosphoprotein (Struave et al., 1990). During mitosis SKI seems to be phosphorylated by CDK1 / CCNB1 (Cyclin B) and is localised at the centrososmes and mitotic spindle (Marcelain & Hayman, 2005). Furthermore, SKI phosphorylation and subsequent destabalization is mediated by the kinases AKT and AURKA. SKI phosphorylation by AKT at T458 is induced by insulin ( INS), insulin-like growth factor-1 ( IGF1R and IGF2R) and hepatocyte growth factor ( HGF) (Band et al., 2009). In contrast, AURKA phosphorylates SKI at S326 and S383 causing its degradation as well as centrosomes amplification and multipolar spindles formation (Mosquera et al, 2011, Rivas et al., 2016). S515 was described as another SKI phosphorylation site, which has no influence on SKI stability or activity. Also, the responsible kinase has not yet been identified (Nagata et al., 2010).

Under physiological conditions SKI is expressed at low levels in most tissues. Higher expression levels can be found in the brain, lung, male and female reproductive organs as well as bone marrow and lymphoid tissue (Human Protein Atlas v20.0 available from www.proteinatlas.org). As published in a study of Pearson-White et al. (1995), Ski is expressed in mature B and T lymphocytes as well as mature macrophages and mast cells. Further, Ski expression can be found in megakaryocyte-erythroid progenitor cells. Often mRNA levels do not correspond with protein levels (Nagano et al., 2010). Also, SKI expression was reported to be cell-cycle dependent showing the lowest protein levels in G0/G1 phase and an upregulation during mitosis (Marcelaine & Hayman, 2005). Furthermore, SKI was reported to be overexpressed in several tumor entities acting mostly oncogenic (Bonnon & Atanasoski 2012, Tecalco-Cruz et al., 2018, Liao et al., 2020).

SKI is predominantly localised in the nucleus (Colmenares et al., 1991). The NLS consists of the sequence PRKRKLT (s. Figure 2; aa 452-458) (Nagata et al., 2006). Aberrant cytoplasmic localisation of SKI was reported in primary invasive and metastatic melanomas preventing TGFB1 -induced SMAD3 nuclear translocation (Reed et al., 2001), primary esophageal squamous cell carcinoma (Fukuchi et al., 2004), colorectal cancer (Bravou et al., 2009), Barretts esophagus (Villanacci et al., 2008) and cervical cancer (Chen et al., 2013). In hepatocytes SKI also was reported to be localised in the cytoplasm, where it co-localises with CD63- and ALIX-positive vesicles. Here, SKI stability is dependent on actin dynamics (Vázquez-Victorio et al., 2015). Cellular function seems to differ between cytoplasmic and nuclear SKI (Nagata et al., 2006). In Schwann cells SKI was reported to partially co-localise with phosphorylated RB1 in the cytoplasm affecting TGFB-induced cell proliferation (Jacob et al., 2008).

SKI is a well described inhibitor of TGFβ signalling but has also influence on the associated BMP signalling and other cellular pathways relevant for cancer development including nuclear hormone receptor signalling, Sonic hedgehog ( SHH) signalling, Hippo signalling, PI3K/Akt signalling and Wnt/β-catenin signalling (Bonnon & Atanasoski, 2012, Liao et al., 2020). SKI acts as a corepressor for SMAD proteins thereby inhibiting TGFβ signalling. Via the interaction with SMAD complexes SKI blocks the interaction with co-activator molecules like EP300 (CBP/P300) and additionally recruits co-repressor complexes to SMAD target genes (Luo et al., 1999, Akiyoshi et al., 1999). SKI is also able to suppress BMP signalling via weak interactions with BMP-specific SMAD complexes and inhibition of BMP target gene expression by interacting with homologous domain interaction protein kinase 2 ( HIPK2) (Wang et al., 2000, Takeda et al., 2004, Harada et al., 2003). The presence of SKI is also needed for the inhibition of nuclear hormone receptor signalling. SKI is an essential part of the NCOR1 / NCOR2, SIN3A and HDAC-recruiting co-repressor complex and contributes to the inhibition of thyroid hormone receptors ( THRA and THRB), RARA and VDR signalling (Nomura et al., 1999, Dahl et al., 1998, Ritter et al., 2006, Ueki & Hayman, 2003). SKI further is involved in the inhibition of SHH signalling by binding to GLI3 protein and recruiting HDAC1 to SHH-induced genes like GLI1 (Dai et al., 2002). In contrast, in pancreatic cancer cells SKI was reported to augment pluripotency of pancreatic cancer stem cells via induction of SHH signalling (Song et al., 2016). SKI can also act as a tumor suppressor and was reported to inhibit breast and lung cancer progression via blocking Hippo/TAZ signalling by increasing TAZ phosphorylation or recruiting co-repressors to TAZ (Rashidian et al., 2015, Xie et al., 2017). SKI is not only a repressor but also activates certain pathways. It was published by Zhao et al. (2020) that SKI knockdown in osteosarcoma (OS) cells causes dephosphorylation of PI3K and AKT leading to decreased OS cell proliferation and migration. Furthermore, in melanoma cells SKI together with FHL2 promotes Wnt/β-catenin signalling and induces the expression of the tumorigenic genes MITF and NRCAM (Chen et al., 2003). In colorectal cancer SKI also significantly correlates with Wnt/β-catenin signalling (Bravou et al., 2009).
SKI interacts with several PU.1 different factors (s. Figure 2) and functions as a co-regulator. SKI was reported to inhibit i.a. GATA1 (Ueki et al., 2004), SPI1 (PU.1) (Ueki et al., 2008), TP53, together with SIRT1 or via MDM2 sumoylation, (Inoue et al., 2011, Ding et al., 2012) and RUNX1 (Feld et al., 2018). It also acts as a co-activator for factors like nuclear factor I (NFI) (Tarapore et al., 1997) or MYOD1, SIX1 and EYA3 during myogenesis (Kobayashi et al., 2007, Zhang & Stavnezer, 2009).
As in mature tissues, also during murine embryonic development low levels of SKI are expressed in most tissues. Highest expression levels can be found in brain and lung (Lyons et al., 1994, Namciu et al., 1995). SKI is crucial for the neuronal development, myogenesis and limb formation. In murine models, SKI downregulation causes aberrant neural tube closure leading to exencephaly, aberrant craniofacial, ocular, skeletal and muscle development (Berk et al., 1997, McGannon et al., 2006, Shinagawa and Ishii, 2003, Colmenares et al., 2002). SKI is further involved in the myelination of Schwann cells (Atanasoski et al., 2004). Even though SKI is relevant for myogenesis, it was reported that ectopic expression of SKI causes hypertrophy of fast skeletal muscle fibers (Lana et al., 1996, Leferovich et al., 1995, Sutrave et al., 1990, 2000). In humans SKI also may be relevant for neuronal and craniofacial development, since patients with the 1p36 deletion syndrome show similar symptoms like mice with a Ski deletion including developmental delay, orofacial clefting and congenital heart defects (Colmenares et al., 2002, Jordan et al., 2015). Moreover, SKI plays a role in hematopoietic differentiation (Namciu et al., 1994, Pearson-White et al., 1995, Dahl et al., 1998, Ueki et al., 2004, 2008, Singbrandt et al., 2014 Zhang et al., 2017). SKI is overexpressed in memory B cells (MBCs) and induces cell proliferation and differentiation (Bhattacharya et al., 2007, Laidlaw et al., 2020). Moreover, SKI is higher expressed in actively maintained quiescent tumour-infiltrating CD8+ T cells from liver tumours compared to natural quiescent T cells from the peripheral blood (Zhang et al., 2009). SKI is upregulated via T cell integrin leukocyte function-associated antigen-1(LFA-1) interacting with ICAM1 in T cells. Consequentially, SKI contributes to the inhibition of the TGFβ-mediated T cell quiescent state (Verma et al., 2012). In a mouse model, Ski inhibits CD103 (Itgae) expression in CD8+ T cells via histone hypoacetylation. This mechanism is dependent on SMAD4. The suppression of CD103+CD8+ T cell generation further leads to a higher susceptibility to secondary viral infections (Wu et al., 2020). In T cells, SKI and SMAD inhibit IL21 -induced differentiation into T helper 17 cells (Th17) via suppressing retinoic acid receptor-related orphan receptor γt (RORγt) expression (Zhang et al., 2017, 2019). SKI affects fibroblastic proliferation and development (Jinnin et al., 2007, Liu et al., 2008, Cunnington et al., 2011) and induces chondrocytic differentiation via inhibition of TGFβ signalling (Kim et al., 2012). The expression of SKI alters under many pathological conditions including demyelination or peripheral nerve damage (Atanasoski et al., 2004), wound healing (Liu et al., 2006, 2010, Li et al., 2011), liver regeneration (Macias-Silva et al., 2002), skeletal muscle regeneration (Soeta et al., 2001) and obstructive nephropathy (Fukasawa et al., 2006).

SKI homologues have been identified in vertebrates from fish to human.

Several coding and non-coding sequence variants of SKI were identified, also containing one coding non-synonymous variant, rs28384811 (White et al., 2008).
In patients with Shprintzen-Goldberg syndrome (SGS) mutations in exon 1 of SKI are responsible for ~90 % of the cases. The mutations can be found in the SMAD interaction domain and DHD domain resulting in the substitution or deletion of amino acids and leading to the induction of TGFβ signalling (Doyle et al., 2012, Carmignac et al., 2012, Au et al., 2014, Schepers et al., 2015, Polinska et al., 2016, Saito et al., 2017). However, it was proposed that during embryogenesis SKI mutations in the SMAD interaction domain or DHD domain effect TGFβ signalling differently, since a patient carrying a mutation in the SKI DHD domain developed lipomeningomyelocele, tethered cord and spina bifida but no SGS characteristics like intellectual disability, craniofacial or cardiovascular abnormalities (Zhang et al., 2019). Also, another mutation in the SKI DHD domain affecting the amino acid Thr180 was reported to be associated with marfanoid syndrome with thoracic aortic aneurysm but no intellectual disability (Arnaud et al., 2020).
A small deletion (576 kb) of 1p36.33-p36.32 encompassing SKI was found in a patient with limb malformations, congenital heart disease (CHD), epilepsy and mild development delay (Zhu et al., 2013).
In a paediatric case of de novo acute myeloid leukaemia with del(5q) a SKI out-of-frame fusion transcript with PRDM16, PRDM16(exon 1)/SKI(exon 2) t(1;1)(p36;p36) PRDM16/SKI, was identified together with another translocation, RUNX1(exon 6)/ USP42 (exon 3) t(7;21)(p22;q22) RUNX1/USP42 (Masetti et al., 2014).

SKI mutation	AA change	domain	White et al. (2008)	Doyle et al. (2012)	Carmignac et al. (2012)	Au et al. (2014)	Schepers et al. (2014)	Poninska et al. (2016)	Saito et al. (2017)	Zhang et al. (2019)	Arnaud et al. (2020)	Total
c.59C>G	p.Thr20Arg	SMAD							1			1
c.59C>A	p.Thr20Lys	SMAD						1				1
c.62T>G	p.Leu21Arg	SMAD		1								1
c.82T>A	p.Ser28Thr	SMAD					1					1
c.92C>T	p.Ser31Leu	SMAD			1		1					2
c.94C>G	p.Leu32Val	SMAD		2*	3*		1					5	*
c.95T>C	p.Leu32Pro	SMAD			1							1
c.100G>A	p.Gly34Ser	SMAD		1	1		1					3
c.100G>T	p.Gly34Cys	SMAD		1	1							2
c.101G>A	p.Gly34Asp	SMAD		1			1					2
c.101G>T	p.Gly34Val	SMAD			1		1					2
c.101G>C	p.Gly34Ala	SMAD					1					1
c.103C>T	p.Pro35Ser	SMAD		1	1	1	2					5
c.104C>A	p.Pro35Gln	SMAD			1							1
c.185C>G	p.Ala62Gly	SMAD	3									3
c.280_291del	p.Ser94_Ser97del	DHD			1							1
c.283_291del	p.Asp95_Ser97del	DHD		1	1							2
c.289_300del	p.Ser97_Arg100del	DHD					1					1
c.336C>G	p.Cys112Trp	DHD								1		1
c.347G>A	p.Gly116Glu	DHD		1		1						2
c.349G>C	p.Gly117Arg	DHD		1								1
c.539C>T	p.Thr180Met	DHD									5	5
c.539C>A	p.Thr180Lys	DHD									2	2
c.539C>G	p.Thr180Arg	DHD									2	2
			3	8	9	2	10	1	1	1	9	48

*one common patient
Table1. Overview of the published SKI mutations leading to AA changes or deletions (adapted from Schepers et al., 2015, Arnaud et al., 2020).